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 #include "main.h"
  78
  79 #define APP_LOOKUP_EXACT_MATCH          0
  80 #define APP_LOOKUP_LPM                  1
  81 #define DO_RFC_1812_CHECKS
  82
  83 #ifndef APP_LOOKUP_METHOD
  84 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
  85 #endif
  86
  87 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
  88 #include <rte_hash.h>
  89 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
  90 #include <rte_lpm.h>
  91 #include <rte_lpm6.h>
  92 #else
  93 #error "APP_LOOKUP_METHOD set to incorrect value"
  94 #endif
  95
  96 #define MAX_PKT_BURST 32
  97
  98 #include "ipv4_rsmbl.h"
  99
 100 #ifndef IPv6_BYTES
 101 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
 102                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
 103 #define IPv6_BYTES(addr) \
 104         addr[0],  addr[1], addr[2],  addr[3], \
 105         addr[4],  addr[5], addr[6],  addr[7], \
 106         addr[8],  addr[9], addr[10], addr[11],\
 107         addr[12], addr[13],addr[14], addr[15]
 108 #endif
 109
 110
 111 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
 112
 113 #define MAX_PORTS       RTE_MAX_ETHPORTS
 114
 115 #define MAX_JUMBO_PKT_LEN  9600
 116
 117 #define IPV6_ADDR_LEN 16
 118
 119 #define MEMPOOL_CACHE_SIZE 256
 120
 121 #define BUF_SIZE        2048
 122 #define MBUF_SIZE       \
 123         (BUF_SIZE + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
 124
 125 #define MAX_FLOW_NUM    UINT16_MAX
 126 #define MIN_FLOW_NUM    1
 127 #define DEF_FLOW_NUM    0x1000
 128
 129 /* TTL numbers are in ms. */
 130 #define MAX_FLOW_TTL    (3600 * MS_PER_S)
 131 #define MIN_FLOW_TTL    1
 132 #define DEF_FLOW_TTL    MS_PER_S
 133
 134 #define DEF_MBUF_NUM    0x400
 135
 136 /* Should be power of two. */
 137 #define IPV4_FRAG_TBL_BUCKET_ENTRIES    2
 138
 139 static uint32_t max_flow_num = DEF_FLOW_NUM;
 140 static uint32_t max_flow_ttl = DEF_FLOW_TTL;
 141
 142 /*
 143  * RX and TX Prefetch, Host, and Write-back threshold values should be
 144  * carefully set for optimal performance. Consult the network
 145  * controller's datasheet and supporting DPDK documentation for guidance
 146  * on how these parameters should be set.
 147  */
 148 #define RX_PTHRESH 8 /**< Default values of RX prefetch threshold reg. */
 149 #define RX_HTHRESH 8 /**< Default values of RX host threshold reg. */
 150 #define RX_WTHRESH 4 /**< Default values of RX write-back threshold reg. */
 151
 152 /*
 153  * These default values are optimized for use with the Intel(R) 82599 10 GbE
 154  * Controller and the DPDK ixgbe PMD. Consider using other values for other
 155  * network controllers and/or network drivers.
 156  */
 157 #define TX_PTHRESH 36 /**< Default values of TX prefetch threshold reg. */
 158 #define TX_HTHRESH 0  /**< Default values of TX host threshold reg. */
 159 #define TX_WTHRESH 0  /**< Default values of TX write-back threshold reg. */
 160
 161 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 162
 163 #define NB_SOCKETS 8
 164
 165 /* Configure how many packets ahead to prefetch, when reading packets */
 166 #define PREFETCH_OFFSET 3
 167
 168 /*
 169  * Configurable number of RX/TX ring descriptors
 170  */
 171 #define RTE_TEST_RX_DESC_DEFAULT 128
 172 #define RTE_TEST_TX_DESC_DEFAULT 512
 173
 174 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 175 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 176
 177 /* ethernet addresses of ports */
 178 static struct ether_addr ports_eth_addr[MAX_PORTS];
 179
 180 /* mask of enabled ports */
 181 static uint32_t enabled_port_mask = 0;
 182 static int promiscuous_on = 0; /**< Ports set in promiscuous mode off by default. */
 183 static int numa_on = 1; /**< NUMA is enabled by default. */
 184
 185 struct mbuf_table {
 186         uint32_t len;
 187         uint32_t head;
 188         uint32_t tail;
 189         struct rte_mbuf *m_table[0];
 190 };
 191
 192 struct lcore_rx_queue {
 193         uint8_t port_id;
 194         uint8_t queue_id;
 195 } __rte_cache_aligned;
 196
 197 #define MAX_RX_QUEUE_PER_LCORE 16
 198 #define MAX_TX_QUEUE_PER_PORT MAX_PORTS
 199 #define MAX_RX_QUEUE_PER_PORT 128
 200
 201 #define MAX_LCORE_PARAMS 1024
 202 struct lcore_params {
 203         uint8_t port_id;
 204         uint8_t queue_id;
 205         uint8_t lcore_id;
 206 } __rte_cache_aligned;
 207
 208 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 209 static struct lcore_params lcore_params_array_default[] = {
 210         {0, 0, 2},
 211         {0, 1, 2},
 212         {0, 2, 2},
 213         {1, 0, 2},
 214         {1, 1, 2},
 215         {1, 2, 2},
 216         {2, 0, 2},
 217         {3, 0, 3},
 218         {3, 1, 3},
 219 };
 220
 221 static struct lcore_params * lcore_params = lcore_params_array_default;
 222 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
 223                                 sizeof(lcore_params_array_default[0]);
 224
 225 static struct rte_eth_conf port_conf = {
 226         .rxmode = {
 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 ipv4_frag_tbl *frag_tbl[MAX_RX_QUEUE_PER_LCORE];
 411         struct rte_mempool *pool[MAX_RX_QUEUE_PER_LCORE];
 412         struct ipv4_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         uint16_t flag_offset, ip_flag, ip_ofs;
 649
 650         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 651
 652         if (m->ol_flags & PKT_RX_IPV4_HDR) {
 653                 /* Handle IPv4 headers.*/
 654                 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
 655
 656 #ifdef DO_RFC_1812_CHECKS
 657                 /* Check to make sure the packet is valid (RFC1812) */
 658                 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt.pkt_len) < 0) {
 659                         rte_pktmbuf_free(m);
 660                         return;
 661                 }
 662
 663                 /* Update time to live and header checksum */
 664                 --(ipv4_hdr->time_to_live);
 665                 ++(ipv4_hdr->hdr_checksum);
 666 #endif
 667
 668                 flag_offset = rte_be_to_cpu_16(ipv4_hdr->fragment_offset);
 669                 ip_ofs = (uint16_t)(flag_offset & IPV4_HDR_OFFSET_MASK);
 670                 ip_flag = (uint16_t)(flag_offset & IPV4_HDR_MF_FLAG);
 671
 672                  /* if it is a fragmented packet, then try to reassemble. */
 673                 if (ip_flag != 0 || ip_ofs  != 0) {
 674
 675                         struct rte_mbuf *mo;
 676                         struct ipv4_frag_tbl *tbl;
 677                         struct ipv4_frag_death_row *dr;
 678
 679                         tbl = qconf->frag_tbl[queue];
 680                         dr = &qconf->death_row;
 681
 682                         /* prepare mbuf: setup l2_len/l3_len. */
 683                         m->pkt.vlan_macip.f.l2_len = sizeof(*eth_hdr);
 684                         m->pkt.vlan_macip.f.l3_len = sizeof(*ipv4_hdr);
 685
 686                         /* process this fragment. */
 687                         if ((mo = ipv4_frag_mbuf(tbl, dr, m, tms, ipv4_hdr,
 688                                         ip_ofs, ip_flag)) == NULL)
 689                                 /* no packet to send out. */
 690                                 return;
 691
 692                         /* we have our packet reassembled. */
 693                         if (mo != m) {
 694                                 m = mo;
 695                                 eth_hdr = rte_pktmbuf_mtod(m,
 696                                         struct ether_hdr *);
 697                                 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
 698                         }
 699                 }
 700
 701                 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
 702                         qconf->ipv4_lookup_struct);
 703                 if (dst_port >= MAX_PORTS ||
 704                                 (enabled_port_mask & 1 << dst_port) == 0)
 705                         dst_port = portid;
 706
 707                 /* 02:00:00:00:00:xx */
 708                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 709                 *((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)dst_port << 40);
 710
 711                 /* src addr */
 712                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 713
 714                 send_single_packet(m, dst_port);
 715         }
 716         else {
 717                 /* Handle IPv6 headers.*/
 718                 struct ipv6_hdr *ipv6_hdr;
 719
 720                 ipv6_hdr = (struct ipv6_hdr *)(rte_pktmbuf_mtod(m, unsigned char *) +
 721                                 sizeof(struct ether_hdr));
 722
 723                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid, qconf->ipv6_lookup_struct);
 724
 725                 if (dst_port >= MAX_PORTS || (enabled_port_mask & 1 << dst_port) == 0)
 726                         dst_port = portid;
 727
 728                 /* 02:00:00:00:00:xx */
 729                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 730                 *((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)dst_port << 40);
 731
 732                 /* src addr */
 733                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 734
 735                 send_single_packet(m, dst_port);
 736         }
 737
 738 }
 739
 740 /* main processing loop */
 741 static int
 742 main_loop(__attribute__((unused)) void *dummy)
 743 {
 744         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 745         unsigned lcore_id;
 746         uint64_t diff_tsc, cur_tsc, prev_tsc;
 747         int i, j, nb_rx;
 748         uint8_t portid, queueid;
 749         struct lcore_conf *qconf;
 750         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 751
 752         prev_tsc = 0;
 753
 754         lcore_id = rte_lcore_id();
 755         qconf = &lcore_conf[lcore_id];
 756
 757         if (qconf->n_rx_queue == 0) {
 758                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
 759                 return 0;
 760         }
 761
 762         RTE_LOG(INFO, L3FWD, "entering main loop on lcore %u\n", lcore_id);
 763
 764         for (i = 0; i < qconf->n_rx_queue; i++) {
 765
 766                 portid = qconf->rx_queue_list[i].port_id;
 767                 queueid = qconf->rx_queue_list[i].queue_id;
 768                 RTE_LOG(INFO, L3FWD, " -- lcoreid=%u portid=%hhu rxqueueid=%hhu\n", lcore_id,
 769                         portid, queueid);
 770         }
 771
 772         while (1) {
 773
 774                 cur_tsc = rte_rdtsc();
 775
 776                 /*
 777                  * TX burst queue drain
 778                  */
 779                 diff_tsc = cur_tsc - prev_tsc;
 780                 if (unlikely(diff_tsc > drain_tsc)) {
 781
 782                         /*
 783                          * This could be optimized (use queueid instead of
 784                          * portid), but it is not called so often
 785                          */
 786                         for (portid = 0; portid < MAX_PORTS; portid++) {
 787                                 if ((enabled_port_mask & (1 << portid)) != 0)
 788                                         send_burst(qconf, 1, portid);
 789                         }
 790
 791                         prev_tsc = cur_tsc;
 792                 }
 793
 794                 /*
 795                  * Read packet from RX queues
 796                  */
 797                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 798
 799                         portid = qconf->rx_queue_list[i].port_id;
 800                         queueid = qconf->rx_queue_list[i].queue_id;
 801
 802                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
 803                                 MAX_PKT_BURST);
 804
 805                         /* Prefetch first packets */
 806                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 807                                 rte_prefetch0(rte_pktmbuf_mtod(
 808                                                 pkts_burst[j], void *));
 809                         }
 810
 811                         /* Prefetch and forward already prefetched packets */
 812                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 813                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 814                                         j + PREFETCH_OFFSET], void *));
 815                                 l3fwd_simple_forward(pkts_burst[j], portid,
 816                                         i, qconf, cur_tsc);
 817                         }
 818
 819                         /* Forward remaining prefetched packets */
 820                         for (; j < nb_rx; j++) {
 821                                 l3fwd_simple_forward(pkts_burst[j], portid,
 822                                         i, qconf, cur_tsc);
 823                         }
 824
 825                         ipv4_frag_free_death_row(&qconf->death_row,
 826                                 PREFETCH_OFFSET);
 827                 }
 828         }
 829 }
 830
 831 static int
 832 check_lcore_params(void)
 833 {
 834         uint8_t queue, lcore;
 835         uint16_t i;
 836         int socketid;
 837
 838         for (i = 0; i < nb_lcore_params; ++i) {
 839                 queue = lcore_params[i].queue_id;
 840                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
 841                         printf("invalid queue number: %hhu\n", queue);
 842                         return -1;
 843                 }
 844                 lcore = lcore_params[i].lcore_id;
 845                 if (!rte_lcore_is_enabled(lcore)) {
 846                         printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
 847                         return -1;
 848                 }
 849                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
 850                         (numa_on == 0)) {
 851                         printf("warning: lcore %hhu is on socket %d with numa off \n",
 852                                 lcore, socketid);
 853                 }
 854         }
 855         return 0;
 856 }
 857
 858 static int
 859 check_port_config(const unsigned nb_ports)
 860 {
 861         unsigned portid;
 862         uint16_t i;
 863
 864         for (i = 0; i < nb_lcore_params; ++i) {
 865                 portid = lcore_params[i].port_id;
 866                 if ((enabled_port_mask & (1 << portid)) == 0) {
 867                         printf("port %u is not enabled in port mask\n", portid);
 868                         return -1;
 869                 }
 870                 if (portid >= nb_ports) {
 871                         printf("port %u is not present on the board\n", portid);
 872                         return -1;
 873                 }
 874         }
 875         return 0;
 876 }
 877
 878 static uint8_t
 879 get_port_n_rx_queues(const uint8_t port)
 880 {
 881         int queue = -1;
 882         uint16_t i;
 883
 884         for (i = 0; i < nb_lcore_params; ++i) {
 885                 if (lcore_params[i].port_id == port && lcore_params[i].queue_id > queue)
 886                         queue = lcore_params[i].queue_id;
 887         }
 888         return (uint8_t)(++queue);
 889 }
 890
 891 static int
 892 init_lcore_rx_queues(void)
 893 {
 894         uint16_t i, nb_rx_queue;
 895         uint8_t lcore;
 896
 897         for (i = 0; i < nb_lcore_params; ++i) {
 898                 lcore = lcore_params[i].lcore_id;
 899                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
 900                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
 901                         printf("error: too many queues (%u) for lcore: %u\n",
 902                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
 903                         return -1;
 904                 } else {
 905                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
 906                                 lcore_params[i].port_id;
 907                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
 908                                 lcore_params[i].queue_id;
 909                         lcore_conf[lcore].n_rx_queue++;
 910                 }
 911         }
 912         return 0;
 913 }
 914
 915 /* display usage */
 916 static void
 917 print_usage(const char *prgname)
 918 {
 919         printf ("%s [EAL options] -- -p PORTMASK -P"
 920                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]"
 921                 "  [--enable-jumbo [--max-pkt-len PKTLEN]]"
 922                 "  [--maxflows=<flows>]  [--flowttl=<ttl>[(s|ms)]]\n"
 923                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
 924                 "  -P : enable promiscuous mode\n"
 925                 "  --config (port,queue,lcore): rx queues configuration\n"
 926                 "  --no-numa: optional, disable numa awareness\n"
 927                 "  --enable-jumbo: enable jumbo frame"
 928                 " which max packet len is PKTLEN in decimal (64-9600)\n"
 929                 "  --maxflows=<flows>: optional, maximum number of flows "
 930                 "supported\n"
 931                 "  --flowttl=<ttl>[(s|ms)]: optional, maximum TTL for each "
 932                 "flow\n",
 933                 prgname);
 934 }
 935
 936 static uint32_t
 937 parse_flow_num(const char *str, uint32_t min, uint32_t max, uint32_t *val)
 938 {
 939         char *end;
 940         uint64_t v;
 941
 942         /* parse decimal string */
 943         errno = 0;
 944         v = strtoul(str, &end, 10);
 945         if (errno != 0 || *end != '\0')
 946                 return (-EINVAL);
 947
 948         if (v < min || v > max)
 949                 return (-EINVAL);
 950
 951         *val = (uint32_t)v;
 952         return (0);
 953 }
 954
 955 static int
 956 parse_flow_ttl(const char *str, uint32_t min, uint32_t max, uint32_t *val)
 957 {
 958         char *end;
 959         uint64_t v;
 960
 961         static const char frmt_sec[] = "s";
 962         static const char frmt_msec[] = "ms";
 963
 964         /* parse decimal string */
 965         errno = 0;
 966         v = strtoul(str, &end, 10);
 967         if (errno != 0)
 968                 return (-EINVAL);
 969
 970         if (*end != '\0') {
 971                 if (strncmp(frmt_sec, end, sizeof(frmt_sec)) == 0)
 972                         v *= MS_PER_S;
 973                 else if (strncmp(frmt_msec, end, sizeof (frmt_msec)) != 0)
 974                         return (-EINVAL);
 975         }
 976
 977         if (v < min || v > max)
 978                 return (-EINVAL);
 979
 980         *val = (uint32_t)v;
 981         return (0);
 982 }
 983
 984
 985 static int parse_max_pkt_len(const char *pktlen)
 986 {
 987         char *end = NULL;
 988         unsigned long len;
 989
 990         /* parse decimal string */
 991         len = strtoul(pktlen, &end, 10);
 992         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
 993                 return -1;
 994
 995         if (len == 0)
 996                 return -1;
 997
 998         return len;
 999 }
1000
1001 static int
1002 parse_portmask(const char *portmask)
1003 {
1004         char *end = NULL;
1005         unsigned long pm;
1006
1007         /* parse hexadecimal string */
1008         pm = strtoul(portmask, &end, 16);
1009         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1010                 return -1;
1011
1012         if (pm == 0)
1013                 return -1;
1014
1015         return pm;
1016 }
1017
1018 static int
1019 parse_config(const char *q_arg)
1020 {
1021         char s[256];
1022         const char *p, *p0 = q_arg;
1023         char *end;
1024         enum fieldnames {
1025                 FLD_PORT = 0,
1026                 FLD_QUEUE,
1027                 FLD_LCORE,
1028                 _NUM_FLD
1029         };
1030         unsigned long int_fld[_NUM_FLD];
1031         char *str_fld[_NUM_FLD];
1032         int i;
1033         unsigned size;
1034
1035         nb_lcore_params = 0;
1036
1037         while ((p = strchr(p0,'(')) != NULL) {
1038                 ++p;
1039                 if((p0 = strchr(p,')')) == NULL)
1040                         return -1;
1041
1042                 size = p0 - p;
1043                 if(size >= sizeof(s))
1044                         return -1;
1045
1046                 rte_snprintf(s, sizeof(s), "%.*s", size, p);
1047                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
1048                         return -1;
1049                 for (i = 0; i < _NUM_FLD; i++){
1050                         errno = 0;
1051                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1052                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
1053                                 return -1;
1054                 }
1055                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1056                         printf("exceeded max number of lcore params: %hu\n",
1057                                 nb_lcore_params);
1058                         return -1;
1059                 }
1060                 lcore_params_array[nb_lcore_params].port_id = (uint8_t)int_fld[FLD_PORT];
1061                 lcore_params_array[nb_lcore_params].queue_id = (uint8_t)int_fld[FLD_QUEUE];
1062                 lcore_params_array[nb_lcore_params].lcore_id = (uint8_t)int_fld[FLD_LCORE];
1063                 ++nb_lcore_params;
1064         }
1065         lcore_params = lcore_params_array;
1066         return 0;
1067 }
1068
1069 /* Parse the argument given in the command line of the application */
1070 static int
1071 parse_args(int argc, char **argv)
1072 {
1073         int opt, ret;
1074         char **argvopt;
1075         int option_index;
1076         char *prgname = argv[0];
1077         static struct option lgopts[] = {
1078                 {"config", 1, 0, 0},
1079                 {"no-numa", 0, 0, 0},
1080                 {"enable-jumbo", 0, 0, 0},
1081                 {"maxflows", 1, 0, 0},
1082                 {"flowttl", 1, 0, 0},
1083                 {NULL, 0, 0, 0}
1084         };
1085
1086         argvopt = argv;
1087
1088         while ((opt = getopt_long(argc, argvopt, "p:P",
1089                                 lgopts, &option_index)) != EOF) {
1090
1091                 switch (opt) {
1092                 /* portmask */
1093                 case 'p':
1094                         enabled_port_mask = parse_portmask(optarg);
1095                         if (enabled_port_mask == 0) {
1096                                 printf("invalid portmask\n");
1097                                 print_usage(prgname);
1098                                 return -1;
1099                         }
1100                         break;
1101                 case 'P':
1102                         printf("Promiscuous mode selected\n");
1103                         promiscuous_on = 1;
1104                         break;
1105
1106                 /* long options */
1107                 case 0:
1108                         if (!strncmp(lgopts[option_index].name, "config", 6)) {
1109                                 ret = parse_config(optarg);
1110                                 if (ret) {
1111                                         printf("invalid config\n");
1112                                         print_usage(prgname);
1113                                         return -1;
1114                                 }
1115                         }
1116
1117                         if (!strncmp(lgopts[option_index].name, "no-numa", 7)) {
1118                                 printf("numa is disabled \n");
1119                                 numa_on = 0;
1120                         }
1121
1122                         if (!strncmp(lgopts[option_index].name,
1123                                         "maxflows", 8)) {
1124                                 if ((ret = parse_flow_num(optarg, MIN_FLOW_NUM,
1125                                                 MAX_FLOW_NUM,
1126                                                 &max_flow_num)) != 0) {
1127                                         printf("invalid value: \"%s\" for "
1128                                                 "parameter %s\n",
1129                                                 optarg,
1130                                                 lgopts[option_index].name);
1131                                         print_usage(prgname);
1132                                         return (ret);
1133                                 }
1134                         }
1135
1136                         if (!strncmp(lgopts[option_index].name, "flowttl", 7)) {
1137                                 if ((ret = parse_flow_ttl(optarg, MIN_FLOW_TTL,
1138                                                 MAX_FLOW_TTL,
1139                                                 &max_flow_ttl)) != 0) {
1140                                         printf("invalid value: \"%s\" for "
1141                                                 "parameter %s\n",
1142                                                 optarg,
1143                                                 lgopts[option_index].name);
1144                                         print_usage(prgname);
1145                                         return (ret);
1146                                 }
1147                         }
1148
1149                         if (!strncmp(lgopts[option_index].name, "enable-jumbo", 12)) {
1150                                 struct option lenopts = {"max-pkt-len", required_argument, 0, 0};
1151
1152                                 printf("jumbo frame is enabled \n");
1153                                 port_conf.rxmode.jumbo_frame = 1;
1154
1155                                 /* if no max-pkt-len set, use the default value ETHER_MAX_LEN */
1156                                 if (0 == getopt_long(argc, argvopt, "", &lenopts, &option_index)) {
1157                                         ret = parse_max_pkt_len(optarg);
1158                                         if ((ret < 64) || (ret > MAX_JUMBO_PKT_LEN)){
1159                                                 printf("invalid packet length\n");
1160                                                 print_usage(prgname);
1161                                                 return -1;
1162                                         }
1163                                         port_conf.rxmode.max_rx_pkt_len = ret;
1164                                 }
1165                                 printf("set jumbo frame max packet length to %u\n",
1166                                                 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1167                         }
1168
1169                         break;
1170
1171                 default:
1172                         print_usage(prgname);
1173                         return -1;
1174                 }
1175         }
1176
1177         if (optind >= 0)
1178                 argv[optind-1] = prgname;
1179
1180         ret = optind-1;
1181         optind = 0; /* reset getopt lib */
1182         return ret;
1183 }
1184
1185 static void
1186 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1187 {
1188         printf ("%s%02X:%02X:%02X:%02X:%02X:%02X", name,
1189                 eth_addr->addr_bytes[0],
1190                 eth_addr->addr_bytes[1],
1191                 eth_addr->addr_bytes[2],
1192                 eth_addr->addr_bytes[3],
1193                 eth_addr->addr_bytes[4],
1194                 eth_addr->addr_bytes[5]);
1195 }
1196
1197 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1198 static void
1199 setup_hash(int socketid)
1200 {
1201         struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1202                 .name = NULL,
1203                 .entries = L3FWD_HASH_ENTRIES,
1204                 .bucket_entries = 4,
1205                 .key_len = sizeof(struct ipv4_5tuple),
1206                 .hash_func = DEFAULT_HASH_FUNC,
1207                 .hash_func_init_val = 0,
1208         };
1209
1210         struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1211                 .name = NULL,
1212                 .entries = L3FWD_HASH_ENTRIES,
1213                 .bucket_entries = 4,
1214                 .key_len = sizeof(struct ipv6_5tuple),
1215                 .hash_func = DEFAULT_HASH_FUNC,
1216                 .hash_func_init_val = 0,
1217         };
1218
1219         unsigned i;
1220         int ret;
1221         char s[64];
1222
1223         /* create ipv4 hash */
1224         rte_snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1225         ipv4_l3fwd_hash_params.name = s;
1226         ipv4_l3fwd_hash_params.socket_id = socketid;
1227         ipv4_l3fwd_lookup_struct[socketid] = rte_hash_create(&ipv4_l3fwd_hash_params);
1228         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1229                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1230                                 "socket %d\n", socketid);
1231
1232         /* create ipv6 hash */
1233         rte_snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1234         ipv6_l3fwd_hash_params.name = s;
1235         ipv6_l3fwd_hash_params.socket_id = socketid;
1236         ipv6_l3fwd_lookup_struct[socketid] = rte_hash_create(&ipv6_l3fwd_hash_params);
1237         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1238                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1239                                 "socket %d\n", socketid);
1240
1241
1242         /* populate the ipv4 hash */
1243         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1244                 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1245                                 (void *) &ipv4_l3fwd_route_array[i].key);
1246                 if (ret < 0) {
1247                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1248                                 "l3fwd hash on socket %d\n", i, socketid);
1249                 }
1250                 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1251                 printf("Hash: Adding key\n");
1252                 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1253         }
1254
1255         /* populate the ipv6 hash */
1256         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1257                 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1258                                 (void *) &ipv6_l3fwd_route_array[i].key);
1259                 if (ret < 0) {
1260                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1261                                 "l3fwd hash on socket %d\n", i, socketid);
1262                 }
1263                 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1264                 printf("Hash: Adding key\n");
1265                 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1266         }
1267 }
1268 #endif
1269
1270 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1271 static void
1272 setup_lpm(int socketid)
1273 {
1274         struct rte_lpm6_config config;
1275         unsigned i;
1276         int ret;
1277         char s[64];
1278
1279         /* create the LPM table */
1280         rte_snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1281         ipv4_l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
1282                                 IPV4_L3FWD_LPM_MAX_RULES, 0);
1283         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1284                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1285                                 " on socket %d\n", socketid);
1286
1287         /* populate the LPM table */
1288         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1289                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1290                         ipv4_l3fwd_route_array[i].ip,
1291                         ipv4_l3fwd_route_array[i].depth,
1292                         ipv4_l3fwd_route_array[i].if_out);
1293
1294                 if (ret < 0) {
1295                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1296                                 "l3fwd LPM table on socket %d\n",
1297                                 i, socketid);
1298                 }
1299
1300                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
1301                         (unsigned)ipv4_l3fwd_route_array[i].ip,
1302                         ipv4_l3fwd_route_array[i].depth,
1303                         ipv4_l3fwd_route_array[i].if_out);
1304         }
1305
1306         /* create the LPM6 table */
1307         rte_snprintf(s, sizeof(s), "IPV6_L3FWD_LPM_%d", socketid);
1308
1309         config.max_rules = IPV6_L3FWD_LPM_MAX_RULES;
1310         config.number_tbl8s = IPV6_L3FWD_LPM_NUMBER_TBL8S;
1311         config.flags = 0;
1312         ipv6_l3fwd_lookup_struct[socketid] = rte_lpm6_create(s, socketid,
1313                                 &config);
1314         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1315                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1316                                 " on socket %d\n", socketid);
1317
1318         /* populate the LPM table */
1319         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1320                 ret = rte_lpm6_add(ipv6_l3fwd_lookup_struct[socketid],
1321                         ipv6_l3fwd_route_array[i].ip,
1322                         ipv6_l3fwd_route_array[i].depth,
1323                         ipv6_l3fwd_route_array[i].if_out);
1324
1325                 if (ret < 0) {
1326                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1327                                 "l3fwd LPM table on socket %d\n",
1328                                 i, socketid);
1329                 }
1330
1331                 printf("LPM: Adding route %s / %d (%d)\n",
1332                         "IPV6",
1333                         ipv6_l3fwd_route_array[i].depth,
1334                         ipv6_l3fwd_route_array[i].if_out);
1335         }
1336 }
1337 #endif
1338
1339 static int
1340 init_mem(void)
1341 {
1342         struct lcore_conf *qconf;
1343         int socketid;
1344         unsigned lcore_id;
1345
1346         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1347                 if (rte_lcore_is_enabled(lcore_id) == 0)
1348                         continue;
1349
1350                 if (numa_on)
1351                         socketid = rte_lcore_to_socket_id(lcore_id);
1352                 else
1353                         socketid = 0;
1354
1355                 if (socketid >= NB_SOCKETS) {
1356                         rte_exit(EXIT_FAILURE,
1357                                 "Socket %d of lcore %u is out of range %d\n",
1358                                 socketid, lcore_id, NB_SOCKETS);
1359                 }
1360
1361 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1362                         setup_lpm(socketid);
1363 #else
1364                         setup_hash(socketid);
1365 #endif
1366                 qconf = &lcore_conf[lcore_id];
1367                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1368                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1369         }
1370         return 0;
1371 }
1372
1373 /* Check the link status of all ports in up to 9s, and print them finally */
1374 static void
1375 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
1376 {
1377 #define CHECK_INTERVAL 100 /* 100ms */
1378 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1379         uint8_t portid, count, all_ports_up, print_flag = 0;
1380         struct rte_eth_link link;
1381
1382         printf("\nChecking link status");
1383         fflush(stdout);
1384         for (count = 0; count <= MAX_CHECK_TIME; count++) {
1385                 all_ports_up = 1;
1386                 for (portid = 0; portid < port_num; portid++) {
1387                         if ((port_mask & (1 << portid)) == 0)
1388                                 continue;
1389                         memset(&link, 0, sizeof(link));
1390                         rte_eth_link_get_nowait(portid, &link);
1391                         /* print link status if flag set */
1392                         if (print_flag == 1) {
1393                                 if (link.link_status)
1394                                         printf("Port %d Link Up - speed %u "
1395                                                 "Mbps - %s\n", (uint8_t)portid,
1396                                                 (unsigned)link.link_speed,
1397                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1398                                         ("full-duplex") : ("half-duplex\n"));
1399                                 else
1400                                         printf("Port %d Link Down\n",
1401                                                 (uint8_t)portid);
1402                                 continue;
1403                         }
1404                         /* clear all_ports_up flag if any link down */
1405                         if (link.link_status == 0) {
1406                                 all_ports_up = 0;
1407                                 break;
1408                         }
1409                 }
1410                 /* after finally printing all link status, get out */
1411                 if (print_flag == 1)
1412                         break;
1413
1414                 if (all_ports_up == 0) {
1415                         printf(".");
1416                         fflush(stdout);
1417                         rte_delay_ms(CHECK_INTERVAL);
1418                 }
1419
1420                 /* set the print_flag if all ports up or timeout */
1421                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1422                         print_flag = 1;
1423                         printf("done\n");
1424                 }
1425         }
1426 }
1427 static void
1428 setup_port_tbl(struct lcore_conf *qconf, uint32_t lcore, int socket,
1429         uint32_t port)
1430 {
1431         struct mbuf_table *mtb;
1432         uint32_t n;
1433         size_t sz;
1434
1435         n = RTE_MAX(max_flow_num, 2UL * MAX_PKT_BURST);
1436         sz = sizeof (*mtb) + sizeof (mtb->m_table[0]) *  n;
1437
1438         if ((mtb = rte_zmalloc_socket(__func__, sz, CACHE_LINE_SIZE,
1439                         socket)) == NULL)
1440                 rte_exit(EXIT_FAILURE, "%s() for lcore: %u, port: %u "
1441                         "failed to allocate %zu bytes\n",
1442                         __func__, lcore, port, sz);
1443
1444         mtb->len = n;
1445         qconf->tx_mbufs[port] = mtb;
1446 }
1447
1448 static void
1449 setup_queue_tbl(struct lcore_conf *qconf, uint32_t lcore, int socket,
1450         uint32_t queue)
1451 {
1452         uint32_t nb_mbuf;
1453         uint64_t frag_cycles;
1454         char buf[RTE_MEMPOOL_NAMESIZE];
1455
1456         frag_cycles = (rte_get_tsc_hz() + MS_PER_S - 1) / MS_PER_S *
1457                 max_flow_ttl;
1458
1459         if ((qconf->frag_tbl[queue] = ipv4_frag_tbl_create(max_flow_num,
1460                         IPV4_FRAG_TBL_BUCKET_ENTRIES, max_flow_num, frag_cycles,
1461                         socket)) == NULL)
1462                 rte_exit(EXIT_FAILURE, "ipv4_frag_tbl_create(%u) on "
1463                         "lcore: %u for queue: %u failed\n",
1464                         max_flow_num, lcore, queue);
1465
1466         /*
1467          * At any given moment up to <max_flow_num * (MAX_FRAG_NUM - 1)>
1468          * mbufs could be stored int the fragment table.
1469          * Plus, each TX queue can hold up to <max_flow_num> packets.
1470          */
1471
1472         nb_mbuf = 2 * RTE_MAX(max_flow_num, 2UL * MAX_PKT_BURST) * MAX_FRAG_NUM;
1473         nb_mbuf *= (port_conf.rxmode.max_rx_pkt_len + BUF_SIZE - 1) / BUF_SIZE;
1474         nb_mbuf += RTE_TEST_RX_DESC_DEFAULT + RTE_TEST_TX_DESC_DEFAULT;
1475
1476         nb_mbuf = RTE_MAX(nb_mbuf, (uint32_t)DEF_MBUF_NUM);
1477
1478         rte_snprintf(buf, sizeof(buf), "mbuf_pool_%u_%u", lcore, queue);
1479
1480         if ((qconf->pool[queue] = rte_mempool_create(buf, nb_mbuf, MBUF_SIZE, 0,
1481                         sizeof(struct rte_pktmbuf_pool_private),
1482                         rte_pktmbuf_pool_init, NULL, rte_pktmbuf_init, NULL,
1483                         socket, MEMPOOL_F_SP_PUT | MEMPOOL_F_SC_GET)) == NULL)
1484                 rte_exit(EXIT_FAILURE, "mempool_create(%s) failed", buf);
1485 }
1486
1487 static void
1488 queue_dump_stat(void)
1489 {
1490         uint32_t i, lcore;
1491         const struct lcore_conf *qconf;
1492
1493         for (lcore = 0; lcore < RTE_MAX_LCORE; lcore++) {
1494                 if (rte_lcore_is_enabled(lcore) == 0)
1495                         continue;
1496
1497                 qconf = lcore_conf + lcore;
1498                 for (i = 0; i < qconf->n_rx_queue; i++) {
1499
1500                         fprintf(stdout, " -- lcoreid=%u portid=%hhu "
1501                                 "rxqueueid=%hhu frag tbl stat:\n",
1502                                 lcore,  qconf->rx_queue_list[i].port_id,
1503                                 qconf->rx_queue_list[i].queue_id);
1504                         ipv4_frag_tbl_dump_stat(stdout, qconf->frag_tbl[i]);
1505                         fprintf(stdout, "TX bursts:\t%" PRIu64 "\n"
1506                                 "TX packets _queued:\t%" PRIu64 "\n"
1507                                 "TX packets dropped:\t%" PRIu64 "\n"
1508                                 "TX packets send:\t%" PRIu64 "\n",
1509                                 qconf->tx_stat.call,
1510                                 qconf->tx_stat.queue,
1511                                 qconf->tx_stat.drop,
1512                                 qconf->tx_stat.send);
1513                 }
1514         }
1515 }
1516
1517 static void
1518 signal_handler(int signum)
1519 {
1520         queue_dump_stat();
1521         if (signum != SIGUSR1)
1522                 rte_exit(0, "received signal: %d, exiting\n", signum);
1523 }
1524
1525 int
1526 MAIN(int argc, char **argv)
1527 {
1528         struct lcore_conf *qconf;
1529         int ret;
1530         unsigned nb_ports;
1531         uint16_t queueid;
1532         unsigned lcore_id;
1533         uint32_t n_tx_queue, nb_lcores;
1534         uint8_t portid, nb_rx_queue, queue, socketid;
1535
1536         /* init EAL */
1537         ret = rte_eal_init(argc, argv);
1538         if (ret < 0)
1539                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1540         argc -= ret;
1541         argv += ret;
1542
1543         /* parse application arguments (after the EAL ones) */
1544         ret = parse_args(argc, argv);
1545         if (ret < 0)
1546                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1547
1548         if (check_lcore_params() < 0)
1549                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1550
1551         ret = init_lcore_rx_queues();
1552         if (ret < 0)
1553                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1554
1555
1556         /* init driver(s) */
1557         if (rte_pmd_init_all() < 0)
1558                 rte_exit(EXIT_FAILURE, "Cannot init pmd\n");
1559
1560         if (rte_eal_pci_probe() < 0)
1561                 rte_exit(EXIT_FAILURE, "Cannot probe PCI\n");
1562
1563         nb_ports = rte_eth_dev_count();
1564         if (nb_ports > MAX_PORTS)
1565                 nb_ports = MAX_PORTS;
1566
1567         if (check_port_config(nb_ports) < 0)
1568                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1569
1570         nb_lcores = rte_lcore_count();
1571
1572         /* initialize all ports */
1573         for (portid = 0; portid < nb_ports; portid++) {
1574                 /* skip ports that are not enabled */
1575                 if ((enabled_port_mask & (1 << portid)) == 0) {
1576                         printf("\nSkipping disabled port %d\n", portid);
1577                         continue;
1578                 }
1579
1580                 /* init port */
1581                 printf("Initializing port %d ... ", portid );
1582                 fflush(stdout);
1583
1584                 nb_rx_queue = get_port_n_rx_queues(portid);
1585                 n_tx_queue = nb_lcores;
1586                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
1587                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1588                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1589                         nb_rx_queue, (unsigned)n_tx_queue );
1590                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
1591                                         (uint16_t)n_tx_queue, &port_conf);
1592                 if (ret < 0)
1593                         rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
1594                                 ret, portid);
1595
1596                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1597                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1598                 printf(", ");
1599
1600                 /* init memory */
1601                 ret = init_mem();
1602                 if (ret < 0)
1603                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
1604
1605                 /* init one TX queue per couple (lcore,port) */
1606                 queueid = 0;
1607                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1608                         if (rte_lcore_is_enabled(lcore_id) == 0)
1609                                 continue;
1610
1611                         if (numa_on)
1612                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1613                         else
1614                                 socketid = 0;
1615
1616                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1617                         fflush(stdout);
1618                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1619                                                      socketid, &tx_conf);
1620                         if (ret < 0)
1621                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
1622                                         "port=%d\n", ret, portid);
1623
1624                         qconf = &lcore_conf[lcore_id];
1625                         qconf->tx_queue_id[portid] = queueid;
1626                         setup_port_tbl(qconf, lcore_id, socketid, portid);
1627                         queueid++;
1628                 }
1629                 printf("\n");
1630         }
1631
1632         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1633                 if (rte_lcore_is_enabled(lcore_id) == 0)
1634                         continue;
1635                 qconf = &lcore_conf[lcore_id];
1636                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1637                 fflush(stdout);
1638                 /* init RX queues */
1639                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1640                         portid = qconf->rx_queue_list[queue].port_id;
1641                         queueid = qconf->rx_queue_list[queue].queue_id;
1642
1643                         if (numa_on)
1644                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1645                         else
1646                                 socketid = 0;
1647
1648                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1649                         fflush(stdout);
1650
1651                         setup_queue_tbl(qconf, lcore_id, socketid, queue);
1652
1653                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1654                                         socketid, &rx_conf, qconf->pool[queue]);
1655                         if (ret < 0)
1656                                 rte_exit(EXIT_FAILURE,
1657                                         "rte_eth_rx_queue_setup: err=%d,"
1658                                         "port=%d\n", ret, portid);
1659                 }
1660         }
1661
1662         printf("\n");
1663
1664         /* start ports */
1665         for (portid = 0; portid < nb_ports; portid++) {
1666                 if ((enabled_port_mask & (1 << portid)) == 0) {
1667                         continue;
1668                 }
1669                 /* Start device */
1670                 ret = rte_eth_dev_start(portid);
1671                 if (ret < 0)
1672                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1673                                 ret, portid);
1674
1675                 /*
1676                  * If enabled, put device in promiscuous mode.
1677                  * This allows IO forwarding mode to forward packets
1678                  * to itself through 2 cross-connected  ports of the
1679                  * target machine.
1680                  */
1681                 if (promiscuous_on)
1682                         rte_eth_promiscuous_enable(portid);
1683         }
1684
1685         check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
1686
1687         signal(SIGUSR1, signal_handler);
1688         signal(SIGTERM, signal_handler);
1689         signal(SIGINT, signal_handler);
1690
1691         /* launch per-lcore init on every lcore */
1692         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1693         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1694                 if (rte_eal_wait_lcore(lcore_id) < 0)
1695                         return -1;
1696         }
1697
1698         return 0;
1699 }