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 "ipv4_rsmbl.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                 .max_rx_pkt_len = ETHER_MAX_LEN,
 227                 .split_hdr_size = 0,
 228                 .header_split   = 0, /**< Header Split disabled */
 229                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 230                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 231                 .jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
 232                 .hw_strip_crc   = 0, /**< CRC stripped by hardware */
 233         },
 234         .rx_adv_conf = {
 235                 .rss_conf = {
 236                         .rss_key = NULL,
 237                         .rss_hf = ETH_RSS_IPV4 | ETH_RSS_IPV6,
 238                 },
 239         },
 240         .txmode = {
 241                 .mq_mode = ETH_MQ_TX_NONE,
 242         },
 243 };
 244
 245 static const struct rte_eth_rxconf rx_conf = {
 246         .rx_thresh = {
 247                 .pthresh = RX_PTHRESH,
 248                 .hthresh = RX_HTHRESH,
 249                 .wthresh = RX_WTHRESH,
 250         },
 251         .rx_free_thresh = 32,
 252 };
 253
 254 static const struct rte_eth_txconf tx_conf = {
 255         .tx_thresh = {
 256                 .pthresh = TX_PTHRESH,
 257                 .hthresh = TX_HTHRESH,
 258                 .wthresh = TX_WTHRESH,
 259         },
 260         .tx_free_thresh = 0, /* Use PMD default values */
 261         .tx_rs_thresh = 0, /* Use PMD default values */
 262         .txq_flags = 0x0,
 263 };
 264
 265 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 266
 267 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 268 #include <rte_hash_crc.h>
 269 #define DEFAULT_HASH_FUNC       rte_hash_crc
 270 #else
 271 #include <rte_jhash.h>
 272 #define DEFAULT_HASH_FUNC       rte_jhash
 273 #endif
 274
 275 struct ipv4_5tuple {
 276         uint32_t ip_dst;
 277         uint32_t ip_src;
 278         uint16_t port_dst;
 279         uint16_t port_src;
 280         uint8_t  proto;
 281 } __attribute__((__packed__));
 282
 283 struct ipv6_5tuple {
 284         uint8_t  ip_dst[IPV6_ADDR_LEN];
 285         uint8_t  ip_src[IPV6_ADDR_LEN];
 286         uint16_t port_dst;
 287         uint16_t port_src;
 288         uint8_t  proto;
 289 } __attribute__((__packed__));
 290
 291 struct ipv4_l3fwd_route {
 292         struct ipv4_5tuple key;
 293         uint8_t if_out;
 294 };
 295
 296 struct ipv6_l3fwd_route {
 297         struct ipv6_5tuple key;
 298         uint8_t if_out;
 299 };
 300
 301 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 302         {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
 303         {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
 304         {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
 305         {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
 306 };
 307
 308 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 309         {
 310                 {
 311                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 312                          0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 313                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 314                          0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
 315                          1, 10, IPPROTO_UDP
 316                 }, 4
 317         },
 318 };
 319
 320 typedef struct rte_hash lookup_struct_t;
 321 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 322 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 323
 324 #define L3FWD_HASH_ENTRIES      1024
 325
 326 #define IPV4_L3FWD_NUM_ROUTES \
 327         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 328
 329 #define IPV6_L3FWD_NUM_ROUTES \
 330         (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
 331
 332 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 333 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 334 #endif
 335
 336 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 337 struct ipv4_l3fwd_route {
 338         uint32_t ip;
 339         uint8_t  depth;
 340         uint8_t  if_out;
 341 };
 342
 343 struct ipv6_l3fwd_route {
 344         uint8_t ip[16];
 345         uint8_t  depth;
 346         uint8_t  if_out;
 347 };
 348
 349 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 350         {IPv4(1,1,1,0), 24, 0},
 351         {IPv4(2,1,1,0), 24, 1},
 352         {IPv4(3,1,1,0), 24, 2},
 353         {IPv4(4,1,1,0), 24, 3},
 354         {IPv4(5,1,1,0), 24, 4},
 355         {IPv4(6,1,1,0), 24, 5},
 356         {IPv4(7,1,1,0), 24, 6},
 357         {IPv4(8,1,1,0), 24, 7},
 358 };
 359
 360 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 361         {{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
 362         {{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
 363         {{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
 364         {{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
 365         {{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
 366         {{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
 367         {{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
 368         {{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
 369 };
 370
 371 #define IPV4_L3FWD_NUM_ROUTES \
 372         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 373 #define IPV6_L3FWD_NUM_ROUTES \
 374         (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
 375
 376 #define IPV4_L3FWD_LPM_MAX_RULES         1024
 377 #define IPV6_L3FWD_LPM_MAX_RULES         1024
 378 #define IPV6_L3FWD_LPM_NUMBER_TBL8S (1 << 16)
 379
 380 typedef struct rte_lpm lookup_struct_t;
 381 typedef struct rte_lpm6 lookup6_struct_t;
 382 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 383 static lookup6_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 384 #endif
 385
 386 struct tx_lcore_stat {
 387         uint64_t call;
 388         uint64_t drop;
 389         uint64_t queue;
 390         uint64_t send;
 391 };
 392
 393 #ifdef IPV4_FRAG_TBL_STAT
 394 #define TX_LCORE_STAT_UPDATE(s, f, v)   ((s)->f += (v))
 395 #else
 396 #define TX_LCORE_STAT_UPDATE(s, f, v)   do {} while (0)
 397 #endif /* IPV4_FRAG_TBL_STAT */
 398
 399 struct lcore_conf {
 400         uint16_t n_rx_queue;
 401         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 402         uint16_t tx_queue_id[MAX_PORTS];
 403         lookup_struct_t * ipv4_lookup_struct;
 404 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 405         lookup6_struct_t * ipv6_lookup_struct;
 406 #else
 407         lookup_struct_t * ipv6_lookup_struct;
 408 #endif
 409         struct ipv4_frag_tbl *frag_tbl[MAX_RX_QUEUE_PER_LCORE];
 410         struct rte_mempool *pool[MAX_RX_QUEUE_PER_LCORE];
 411         struct ipv4_frag_death_row death_row;
 412         struct mbuf_table *tx_mbufs[MAX_PORTS];
 413         struct tx_lcore_stat tx_stat;
 414 } __rte_cache_aligned;
 415
 416 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 417
 418 /*
 419  * If number of queued packets reached given threahold, then
 420  * send burst of packets on an output interface.
 421  */
 422 static inline uint32_t
 423 send_burst(struct lcore_conf *qconf, uint32_t thresh, uint8_t port)
 424 {
 425         uint32_t fill, len, k, n;
 426         struct mbuf_table *txmb;
 427
 428         txmb = qconf->tx_mbufs[port];
 429         len = txmb->len;
 430
 431         if ((int32_t)(fill = txmb->head - txmb->tail) < 0)
 432                 fill += len;
 433
 434         if (fill >= thresh) {
 435                 n = RTE_MIN(len - txmb->tail, fill);
 436
 437                 k = rte_eth_tx_burst(port, qconf->tx_queue_id[port],
 438                         txmb->m_table + txmb->tail, (uint16_t)n);
 439
 440                 TX_LCORE_STAT_UPDATE(&qconf->tx_stat, call, 1);
 441                 TX_LCORE_STAT_UPDATE(&qconf->tx_stat, send, k);
 442
 443                 fill -= k;
 444                 if ((txmb->tail += k) == len)
 445                         txmb->tail = 0;
 446         }
 447
 448         return (fill);
 449 }
 450
 451 /* Enqueue a single packet, and send burst if queue is filled */
 452 static inline int
 453 send_single_packet(struct rte_mbuf *m, uint8_t port)
 454 {
 455         uint32_t fill, lcore_id, len;
 456         struct lcore_conf *qconf;
 457         struct mbuf_table *txmb;
 458
 459         lcore_id = rte_lcore_id();
 460         qconf = &lcore_conf[lcore_id];
 461
 462         txmb = qconf->tx_mbufs[port];
 463         len = txmb->len;
 464
 465         fill = send_burst(qconf, MAX_PKT_BURST, port);
 466
 467         if (fill == len - 1) {
 468                 TX_LCORE_STAT_UPDATE(&qconf->tx_stat, drop, 1);
 469                 rte_pktmbuf_free(txmb->m_table[txmb->tail]);
 470                 if (++txmb->tail == len)
 471                         txmb->tail = 0;
 472         }
 473
 474         TX_LCORE_STAT_UPDATE(&qconf->tx_stat, queue, 1);
 475         txmb->m_table[txmb->head] = m;
 476         if(++txmb->head == len)
 477                 txmb->head = 0;
 478
 479         return (0);
 480 }
 481
 482 #ifdef DO_RFC_1812_CHECKS
 483 static inline int
 484 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
 485 {
 486         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 487         /*
 488          * 1. The packet length reported by the Link Layer must be large
 489          * enough to hold the minimum length legal IP datagram (20 bytes).
 490          */
 491         if (link_len < sizeof(struct ipv4_hdr))
 492                 return -1;
 493
 494         /* 2. The IP checksum must be correct. */
 495         /* this is checked in H/W */
 496
 497         /*
 498          * 3. The IP version number must be 4. If the version number is not 4
 499          * then the packet may be another version of IP, such as IPng or
 500          * ST-II.
 501          */
 502         if (((pkt->version_ihl) >> 4) != 4)
 503                 return -3;
 504         /*
 505          * 4. The IP header length field must be large enough to hold the
 506          * minimum length legal IP datagram (20 bytes = 5 words).
 507          */
 508         if ((pkt->version_ihl & 0xf) < 5)
 509                 return -4;
 510
 511         /*
 512          * 5. The IP total length field must be large enough to hold the IP
 513          * datagram header, whose length is specified in the IP header length
 514          * field.
 515          */
 516         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
 517                 return -5;
 518
 519         return 0;
 520 }
 521 #endif
 522
 523 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 524 static void
 525 print_ipv4_key(struct ipv4_5tuple key)
 526 {
 527         printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, proto = %d\n",
 528                         (unsigned)key.ip_dst, (unsigned)key.ip_src, key.port_dst, key.port_src, key.proto);
 529 }
 530 static void
 531 print_ipv6_key(struct ipv6_5tuple key)
 532 {
 533         printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
 534                 "port dst = %d, port src = %d, proto = %d\n",
 535                 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
 536                 key.port_dst, key.port_src, key.proto);
 537 }
 538
 539 static inline uint8_t
 540 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr,  uint8_t portid, lookup_struct_t * ipv4_l3fwd_lookup_struct)
 541 {
 542         struct ipv4_5tuple key;
 543         struct tcp_hdr *tcp;
 544         struct udp_hdr *udp;
 545         int ret = 0;
 546
 547         key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
 548         key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
 549         key.proto = ipv4_hdr->next_proto_id;
 550
 551         switch (ipv4_hdr->next_proto_id) {
 552         case IPPROTO_TCP:
 553                 tcp = (struct tcp_hdr *)((unsigned char *) ipv4_hdr +
 554                                         sizeof(struct ipv4_hdr));
 555                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 556                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 557                 break;
 558
 559         case IPPROTO_UDP:
 560                 udp = (struct udp_hdr *)((unsigned char *) ipv4_hdr +
 561                                         sizeof(struct ipv4_hdr));
 562                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 563                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 564                 break;
 565
 566         default:
 567                 key.port_dst = 0;
 568                 key.port_src = 0;
 569                 break;
 570         }
 571
 572         /* Find destination port */
 573         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 574         return (uint8_t)((ret < 0)? portid : ipv4_l3fwd_out_if[ret]);
 575 }
 576
 577 static inline uint8_t
 578 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr,  uint8_t portid, lookup_struct_t * ipv6_l3fwd_lookup_struct)
 579 {
 580         struct ipv6_5tuple key;
 581         struct tcp_hdr *tcp;
 582         struct udp_hdr *udp;
 583         int ret = 0;
 584
 585         memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
 586         memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
 587
 588         key.proto = ipv6_hdr->proto;
 589
 590         switch (ipv6_hdr->proto) {
 591         case IPPROTO_TCP:
 592                 tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr +
 593                                         sizeof(struct ipv6_hdr));
 594                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 595                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 596                 break;
 597
 598         case IPPROTO_UDP:
 599                 udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr +
 600                                         sizeof(struct ipv6_hdr));
 601                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 602                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 603                 break;
 604
 605         default:
 606                 key.port_dst = 0;
 607                 key.port_src = 0;
 608                 break;
 609         }
 610
 611         /* Find destination port */
 612         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 613         return (uint8_t)((ret < 0)? portid : ipv6_l3fwd_out_if[ret]);
 614 }
 615 #endif
 616
 617 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 618 static inline uint8_t
 619 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr,  uint8_t portid, lookup_struct_t * ipv4_l3fwd_lookup_struct)
 620 {
 621         uint8_t next_hop;
 622
 623         return (uint8_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
 624                         rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
 625                         next_hop : portid);
 626 }
 627
 628 static inline uint8_t
 629 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr,  uint8_t portid, lookup6_struct_t * ipv6_l3fwd_lookup_struct)
 630 {
 631         uint8_t next_hop;
 632
 633         return (uint8_t) ((rte_lpm6_lookup(ipv6_l3fwd_lookup_struct,
 634                         ipv6_hdr->dst_addr, &next_hop) == 0)?
 635                         next_hop : portid);
 636 }
 637 #endif
 638
 639 static inline void
 640 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid, uint32_t queue,
 641         struct lcore_conf *qconf, uint64_t tms)
 642 {
 643         struct ether_hdr *eth_hdr;
 644         struct ipv4_hdr *ipv4_hdr;
 645         void *d_addr_bytes;
 646         uint8_t dst_port;
 647         uint16_t flag_offset, ip_flag, ip_ofs;
 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                 flag_offset = rte_be_to_cpu_16(ipv4_hdr->fragment_offset);
 668                 ip_ofs = (uint16_t)(flag_offset & IPV4_HDR_OFFSET_MASK);
 669                 ip_flag = (uint16_t)(flag_offset & IPV4_HDR_MF_FLAG);
 670
 671                  /* if it is a fragmented packet, then try to reassemble. */
 672                 if (ip_flag != 0 || ip_ofs  != 0) {
 673
 674                         struct rte_mbuf *mo;
 675                         struct ipv4_frag_tbl *tbl;
 676                         struct ipv4_frag_death_row *dr;
 677
 678                         tbl = qconf->frag_tbl[queue];
 679                         dr = &qconf->death_row;
 680
 681                         /* prepare mbuf: setup l2_len/l3_len. */
 682                         m->pkt.vlan_macip.f.l2_len = sizeof(*eth_hdr);
 683                         m->pkt.vlan_macip.f.l3_len = sizeof(*ipv4_hdr);
 684
 685                         /* process this fragment. */
 686                         if ((mo = ipv4_frag_mbuf(tbl, dr, m, tms, ipv4_hdr,
 687                                         ip_ofs, ip_flag)) == NULL)
 688                                 /* no packet to send out. */
 689                                 return;
 690
 691                         /* we have our packet reassembled. */
 692                         if (mo != m) {
 693                                 m = mo;
 694                                 eth_hdr = rte_pktmbuf_mtod(m,
 695                                         struct ether_hdr *);
 696                                 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
 697                         }
 698                 }
 699
 700                 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
 701                         qconf->ipv4_lookup_struct);
 702                 if (dst_port >= MAX_PORTS ||
 703                                 (enabled_port_mask & 1 << dst_port) == 0)
 704                         dst_port = portid;
 705
 706                 /* 02:00:00:00:00:xx */
 707                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 708                 *((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)dst_port << 40);
 709
 710                 /* src addr */
 711                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 712
 713                 send_single_packet(m, dst_port);
 714         }
 715         else {
 716                 /* Handle IPv6 headers.*/
 717                 struct ipv6_hdr *ipv6_hdr;
 718
 719                 ipv6_hdr = (struct ipv6_hdr *)(rte_pktmbuf_mtod(m, unsigned char *) +
 720                                 sizeof(struct ether_hdr));
 721
 722                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid, qconf->ipv6_lookup_struct);
 723
 724                 if (dst_port >= MAX_PORTS || (enabled_port_mask & 1 << dst_port) == 0)
 725                         dst_port = portid;
 726
 727                 /* 02:00:00:00:00:xx */
 728                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 729                 *((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)dst_port << 40);
 730
 731                 /* src addr */
 732                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 733
 734                 send_single_packet(m, dst_port);
 735         }
 736
 737 }
 738
 739 /* main processing loop */
 740 static int
 741 main_loop(__attribute__((unused)) void *dummy)
 742 {
 743         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 744         unsigned lcore_id;
 745         uint64_t diff_tsc, cur_tsc, prev_tsc;
 746         int i, j, nb_rx;
 747         uint8_t portid, queueid;
 748         struct lcore_conf *qconf;
 749         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 750
 751         prev_tsc = 0;
 752
 753         lcore_id = rte_lcore_id();
 754         qconf = &lcore_conf[lcore_id];
 755
 756         if (qconf->n_rx_queue == 0) {
 757                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
 758                 return 0;
 759         }
 760
 761         RTE_LOG(INFO, L3FWD, "entering main loop on lcore %u\n", lcore_id);
 762
 763         for (i = 0; i < qconf->n_rx_queue; i++) {
 764
 765                 portid = qconf->rx_queue_list[i].port_id;
 766                 queueid = qconf->rx_queue_list[i].queue_id;
 767                 RTE_LOG(INFO, L3FWD, " -- lcoreid=%u portid=%hhu rxqueueid=%hhu\n", lcore_id,
 768                         portid, queueid);
 769         }
 770
 771         while (1) {
 772
 773                 cur_tsc = rte_rdtsc();
 774
 775                 /*
 776                  * TX burst queue drain
 777                  */
 778                 diff_tsc = cur_tsc - prev_tsc;
 779                 if (unlikely(diff_tsc > drain_tsc)) {
 780
 781                         /*
 782                          * This could be optimized (use queueid instead of
 783                          * portid), but it is not called so often
 784                          */
 785                         for (portid = 0; portid < MAX_PORTS; portid++) {
 786                                 if ((enabled_port_mask & (1 << portid)) != 0)
 787                                         send_burst(qconf, 1, portid);
 788                         }
 789
 790                         prev_tsc = cur_tsc;
 791                 }
 792
 793                 /*
 794                  * Read packet from RX queues
 795                  */
 796                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 797
 798                         portid = qconf->rx_queue_list[i].port_id;
 799                         queueid = qconf->rx_queue_list[i].queue_id;
 800
 801                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
 802                                 MAX_PKT_BURST);
 803
 804                         /* Prefetch first packets */
 805                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 806                                 rte_prefetch0(rte_pktmbuf_mtod(
 807                                                 pkts_burst[j], void *));
 808                         }
 809
 810                         /* Prefetch and forward already prefetched packets */
 811                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 812                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 813                                         j + PREFETCH_OFFSET], void *));
 814                                 l3fwd_simple_forward(pkts_burst[j], portid,
 815                                         i, qconf, cur_tsc);
 816                         }
 817
 818                         /* Forward remaining prefetched packets */
 819                         for (; j < nb_rx; j++) {
 820                                 l3fwd_simple_forward(pkts_burst[j], portid,
 821                                         i, qconf, cur_tsc);
 822                         }
 823
 824                         ipv4_frag_free_death_row(&qconf->death_row,
 825                                 PREFETCH_OFFSET);
 826                 }
 827         }
 828 }
 829
 830 static int
 831 check_lcore_params(void)
 832 {
 833         uint8_t queue, lcore;
 834         uint16_t i;
 835         int socketid;
 836
 837         for (i = 0; i < nb_lcore_params; ++i) {
 838                 queue = lcore_params[i].queue_id;
 839                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
 840                         printf("invalid queue number: %hhu\n", queue);
 841                         return -1;
 842                 }
 843                 lcore = lcore_params[i].lcore_id;
 844                 if (!rte_lcore_is_enabled(lcore)) {
 845                         printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
 846                         return -1;
 847                 }
 848                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
 849                         (numa_on == 0)) {
 850                         printf("warning: lcore %hhu is on socket %d with numa off \n",
 851                                 lcore, socketid);
 852                 }
 853         }
 854         return 0;
 855 }
 856
 857 static int
 858 check_port_config(const unsigned nb_ports)
 859 {
 860         unsigned portid;
 861         uint16_t i;
 862
 863         for (i = 0; i < nb_lcore_params; ++i) {
 864                 portid = lcore_params[i].port_id;
 865                 if ((enabled_port_mask & (1 << portid)) == 0) {
 866                         printf("port %u is not enabled in port mask\n", portid);
 867                         return -1;
 868                 }
 869                 if (portid >= nb_ports) {
 870                         printf("port %u is not present on the board\n", portid);
 871                         return -1;
 872                 }
 873         }
 874         return 0;
 875 }
 876
 877 static uint8_t
 878 get_port_n_rx_queues(const uint8_t port)
 879 {
 880         int queue = -1;
 881         uint16_t i;
 882
 883         for (i = 0; i < nb_lcore_params; ++i) {
 884                 if (lcore_params[i].port_id == port && lcore_params[i].queue_id > queue)
 885                         queue = lcore_params[i].queue_id;
 886         }
 887         return (uint8_t)(++queue);
 888 }
 889
 890 static int
 891 init_lcore_rx_queues(void)
 892 {
 893         uint16_t i, nb_rx_queue;
 894         uint8_t lcore;
 895
 896         for (i = 0; i < nb_lcore_params; ++i) {
 897                 lcore = lcore_params[i].lcore_id;
 898                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
 899                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
 900                         printf("error: too many queues (%u) for lcore: %u\n",
 901                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
 902                         return -1;
 903                 } else {
 904                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
 905                                 lcore_params[i].port_id;
 906                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
 907                                 lcore_params[i].queue_id;
 908                         lcore_conf[lcore].n_rx_queue++;
 909                 }
 910         }
 911         return 0;
 912 }
 913
 914 /* display usage */
 915 static void
 916 print_usage(const char *prgname)
 917 {
 918         printf ("%s [EAL options] -- -p PORTMASK -P"
 919                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]"
 920                 "  [--enable-jumbo [--max-pkt-len PKTLEN]]"
 921                 "  [--maxflows=<flows>]  [--flowttl=<ttl>[(s|ms)]]\n"
 922                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
 923                 "  -P : enable promiscuous mode\n"
 924                 "  --config (port,queue,lcore): rx queues configuration\n"
 925                 "  --no-numa: optional, disable numa awareness\n"
 926                 "  --enable-jumbo: enable jumbo frame"
 927                 " which max packet len is PKTLEN in decimal (64-9600)\n"
 928                 "  --maxflows=<flows>: optional, maximum number of flows "
 929                 "supported\n"
 930                 "  --flowttl=<ttl>[(s|ms)]: optional, maximum TTL for each "
 931                 "flow\n",
 932                 prgname);
 933 }
 934
 935 static uint32_t
 936 parse_flow_num(const char *str, uint32_t min, uint32_t max, uint32_t *val)
 937 {
 938         char *end;
 939         uint64_t v;
 940
 941         /* parse decimal string */
 942         errno = 0;
 943         v = strtoul(str, &end, 10);
 944         if (errno != 0 || *end != '\0')
 945                 return (-EINVAL);
 946
 947         if (v < min || v > max)
 948                 return (-EINVAL);
 949
 950         *val = (uint32_t)v;
 951         return (0);
 952 }
 953
 954 static int
 955 parse_flow_ttl(const char *str, uint32_t min, uint32_t max, uint32_t *val)
 956 {
 957         char *end;
 958         uint64_t v;
 959
 960         static const char frmt_sec[] = "s";
 961         static const char frmt_msec[] = "ms";
 962
 963         /* parse decimal string */
 964         errno = 0;
 965         v = strtoul(str, &end, 10);
 966         if (errno != 0)
 967                 return (-EINVAL);
 968
 969         if (*end != '\0') {
 970                 if (strncmp(frmt_sec, end, sizeof(frmt_sec)) == 0)
 971                         v *= MS_PER_S;
 972                 else if (strncmp(frmt_msec, end, sizeof (frmt_msec)) != 0)
 973                         return (-EINVAL);
 974         }
 975
 976         if (v < min || v > max)
 977                 return (-EINVAL);
 978
 979         *val = (uint32_t)v;
 980         return (0);
 981 }
 982
 983
 984 static int parse_max_pkt_len(const char *pktlen)
 985 {
 986         char *end = NULL;
 987         unsigned long len;
 988
 989         /* parse decimal string */
 990         len = strtoul(pktlen, &end, 10);
 991         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
 992                 return -1;
 993
 994         if (len == 0)
 995                 return -1;
 996
 997         return len;
 998 }
 999
1000 static int
1001 parse_portmask(const char *portmask)
1002 {
1003         char *end = NULL;
1004         unsigned long pm;
1005
1006         /* parse hexadecimal string */
1007         pm = strtoul(portmask, &end, 16);
1008         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1009                 return -1;
1010
1011         if (pm == 0)
1012                 return -1;
1013
1014         return pm;
1015 }
1016
1017 static int
1018 parse_config(const char *q_arg)
1019 {
1020         char s[256];
1021         const char *p, *p0 = q_arg;
1022         char *end;
1023         enum fieldnames {
1024                 FLD_PORT = 0,
1025                 FLD_QUEUE,
1026                 FLD_LCORE,
1027                 _NUM_FLD
1028         };
1029         unsigned long int_fld[_NUM_FLD];
1030         char *str_fld[_NUM_FLD];
1031         int i;
1032         unsigned size;
1033
1034         nb_lcore_params = 0;
1035
1036         while ((p = strchr(p0,'(')) != NULL) {
1037                 ++p;
1038                 if((p0 = strchr(p,')')) == NULL)
1039                         return -1;
1040
1041                 size = p0 - p;
1042                 if(size >= sizeof(s))
1043                         return -1;
1044
1045                 rte_snprintf(s, sizeof(s), "%.*s", size, p);
1046                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
1047                         return -1;
1048                 for (i = 0; i < _NUM_FLD; i++){
1049                         errno = 0;
1050                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1051                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
1052                                 return -1;
1053                 }
1054                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1055                         printf("exceeded max number of lcore params: %hu\n",
1056                                 nb_lcore_params);
1057                         return -1;
1058                 }
1059                 lcore_params_array[nb_lcore_params].port_id = (uint8_t)int_fld[FLD_PORT];
1060                 lcore_params_array[nb_lcore_params].queue_id = (uint8_t)int_fld[FLD_QUEUE];
1061                 lcore_params_array[nb_lcore_params].lcore_id = (uint8_t)int_fld[FLD_LCORE];
1062                 ++nb_lcore_params;
1063         }
1064         lcore_params = lcore_params_array;
1065         return 0;
1066 }
1067
1068 /* Parse the argument given in the command line of the application */
1069 static int
1070 parse_args(int argc, char **argv)
1071 {
1072         int opt, ret;
1073         char **argvopt;
1074         int option_index;
1075         char *prgname = argv[0];
1076         static struct option lgopts[] = {
1077                 {"config", 1, 0, 0},
1078                 {"no-numa", 0, 0, 0},
1079                 {"enable-jumbo", 0, 0, 0},
1080                 {"maxflows", 1, 0, 0},
1081                 {"flowttl", 1, 0, 0},
1082                 {NULL, 0, 0, 0}
1083         };
1084
1085         argvopt = argv;
1086
1087         while ((opt = getopt_long(argc, argvopt, "p:P",
1088                                 lgopts, &option_index)) != EOF) {
1089
1090                 switch (opt) {
1091                 /* portmask */
1092                 case 'p':
1093                         enabled_port_mask = parse_portmask(optarg);
1094                         if (enabled_port_mask == 0) {
1095                                 printf("invalid portmask\n");
1096                                 print_usage(prgname);
1097                                 return -1;
1098                         }
1099                         break;
1100                 case 'P':
1101                         printf("Promiscuous mode selected\n");
1102                         promiscuous_on = 1;
1103                         break;
1104
1105                 /* long options */
1106                 case 0:
1107                         if (!strncmp(lgopts[option_index].name, "config", 6)) {
1108                                 ret = parse_config(optarg);
1109                                 if (ret) {
1110                                         printf("invalid config\n");
1111                                         print_usage(prgname);
1112                                         return -1;
1113                                 }
1114                         }
1115
1116                         if (!strncmp(lgopts[option_index].name, "no-numa", 7)) {
1117                                 printf("numa is disabled \n");
1118                                 numa_on = 0;
1119                         }
1120
1121                         if (!strncmp(lgopts[option_index].name,
1122                                         "maxflows", 8)) {
1123                                 if ((ret = parse_flow_num(optarg, MIN_FLOW_NUM,
1124                                                 MAX_FLOW_NUM,
1125                                                 &max_flow_num)) != 0) {
1126                                         printf("invalid value: \"%s\" for "
1127                                                 "parameter %s\n",
1128                                                 optarg,
1129                                                 lgopts[option_index].name);
1130                                         print_usage(prgname);
1131                                         return (ret);
1132                                 }
1133                         }
1134
1135                         if (!strncmp(lgopts[option_index].name, "flowttl", 7)) {
1136                                 if ((ret = parse_flow_ttl(optarg, MIN_FLOW_TTL,
1137                                                 MAX_FLOW_TTL,
1138                                                 &max_flow_ttl)) != 0) {
1139                                         printf("invalid value: \"%s\" for "
1140                                                 "parameter %s\n",
1141                                                 optarg,
1142                                                 lgopts[option_index].name);
1143                                         print_usage(prgname);
1144                                         return (ret);
1145                                 }
1146                         }
1147
1148                         if (!strncmp(lgopts[option_index].name, "enable-jumbo", 12)) {
1149                                 struct option lenopts = {"max-pkt-len", required_argument, 0, 0};
1150
1151                                 printf("jumbo frame is enabled \n");
1152                                 port_conf.rxmode.jumbo_frame = 1;
1153
1154                                 /* if no max-pkt-len set, use the default value ETHER_MAX_LEN */
1155                                 if (0 == getopt_long(argc, argvopt, "", &lenopts, &option_index)) {
1156                                         ret = parse_max_pkt_len(optarg);
1157                                         if ((ret < 64) || (ret > MAX_JUMBO_PKT_LEN)){
1158                                                 printf("invalid packet length\n");
1159                                                 print_usage(prgname);
1160                                                 return -1;
1161                                         }
1162                                         port_conf.rxmode.max_rx_pkt_len = ret;
1163                                 }
1164                                 printf("set jumbo frame max packet length to %u\n",
1165                                                 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1166                         }
1167
1168                         break;
1169
1170                 default:
1171                         print_usage(prgname);
1172                         return -1;
1173                 }
1174         }
1175
1176         if (optind >= 0)
1177                 argv[optind-1] = prgname;
1178
1179         ret = optind-1;
1180         optind = 0; /* reset getopt lib */
1181         return ret;
1182 }
1183
1184 static void
1185 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1186 {
1187         printf ("%s%02X:%02X:%02X:%02X:%02X:%02X", name,
1188                 eth_addr->addr_bytes[0],
1189                 eth_addr->addr_bytes[1],
1190                 eth_addr->addr_bytes[2],
1191                 eth_addr->addr_bytes[3],
1192                 eth_addr->addr_bytes[4],
1193                 eth_addr->addr_bytes[5]);
1194 }
1195
1196 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1197 static void
1198 setup_hash(int socketid)
1199 {
1200         struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1201                 .name = NULL,
1202                 .entries = L3FWD_HASH_ENTRIES,
1203                 .bucket_entries = 4,
1204                 .key_len = sizeof(struct ipv4_5tuple),
1205                 .hash_func = DEFAULT_HASH_FUNC,
1206                 .hash_func_init_val = 0,
1207         };
1208
1209         struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1210                 .name = NULL,
1211                 .entries = L3FWD_HASH_ENTRIES,
1212                 .bucket_entries = 4,
1213                 .key_len = sizeof(struct ipv6_5tuple),
1214                 .hash_func = DEFAULT_HASH_FUNC,
1215                 .hash_func_init_val = 0,
1216         };
1217
1218         unsigned i;
1219         int ret;
1220         char s[64];
1221
1222         /* create ipv4 hash */
1223         rte_snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1224         ipv4_l3fwd_hash_params.name = s;
1225         ipv4_l3fwd_hash_params.socket_id = socketid;
1226         ipv4_l3fwd_lookup_struct[socketid] = rte_hash_create(&ipv4_l3fwd_hash_params);
1227         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1228                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1229                                 "socket %d\n", socketid);
1230
1231         /* create ipv6 hash */
1232         rte_snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1233         ipv6_l3fwd_hash_params.name = s;
1234         ipv6_l3fwd_hash_params.socket_id = socketid;
1235         ipv6_l3fwd_lookup_struct[socketid] = rte_hash_create(&ipv6_l3fwd_hash_params);
1236         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1237                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1238                                 "socket %d\n", socketid);
1239
1240
1241         /* populate the ipv4 hash */
1242         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1243                 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1244                                 (void *) &ipv4_l3fwd_route_array[i].key);
1245                 if (ret < 0) {
1246                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1247                                 "l3fwd hash on socket %d\n", i, socketid);
1248                 }
1249                 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1250                 printf("Hash: Adding key\n");
1251                 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1252         }
1253
1254         /* populate the ipv6 hash */
1255         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1256                 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1257                                 (void *) &ipv6_l3fwd_route_array[i].key);
1258                 if (ret < 0) {
1259                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1260                                 "l3fwd hash on socket %d\n", i, socketid);
1261                 }
1262                 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1263                 printf("Hash: Adding key\n");
1264                 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1265         }
1266 }
1267 #endif
1268
1269 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1270 static void
1271 setup_lpm(int socketid)
1272 {
1273         struct rte_lpm6_config config;
1274         unsigned i;
1275         int ret;
1276         char s[64];
1277
1278         /* create the LPM table */
1279         rte_snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1280         ipv4_l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
1281                                 IPV4_L3FWD_LPM_MAX_RULES, 0);
1282         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1283                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1284                                 " on socket %d\n", socketid);
1285
1286         /* populate the LPM table */
1287         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1288                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1289                         ipv4_l3fwd_route_array[i].ip,
1290                         ipv4_l3fwd_route_array[i].depth,
1291                         ipv4_l3fwd_route_array[i].if_out);
1292
1293                 if (ret < 0) {
1294                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1295                                 "l3fwd LPM table on socket %d\n",
1296                                 i, socketid);
1297                 }
1298
1299                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
1300                         (unsigned)ipv4_l3fwd_route_array[i].ip,
1301                         ipv4_l3fwd_route_array[i].depth,
1302                         ipv4_l3fwd_route_array[i].if_out);
1303         }
1304
1305         /* create the LPM6 table */
1306         rte_snprintf(s, sizeof(s), "IPV6_L3FWD_LPM_%d", socketid);
1307
1308         config.max_rules = IPV6_L3FWD_LPM_MAX_RULES;
1309         config.number_tbl8s = IPV6_L3FWD_LPM_NUMBER_TBL8S;
1310         config.flags = 0;
1311         ipv6_l3fwd_lookup_struct[socketid] = rte_lpm6_create(s, socketid,
1312                                 &config);
1313         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1314                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1315                                 " on socket %d\n", socketid);
1316
1317         /* populate the LPM table */
1318         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1319                 ret = rte_lpm6_add(ipv6_l3fwd_lookup_struct[socketid],
1320                         ipv6_l3fwd_route_array[i].ip,
1321                         ipv6_l3fwd_route_array[i].depth,
1322                         ipv6_l3fwd_route_array[i].if_out);
1323
1324                 if (ret < 0) {
1325                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1326                                 "l3fwd LPM table on socket %d\n",
1327                                 i, socketid);
1328                 }
1329
1330                 printf("LPM: Adding route %s / %d (%d)\n",
1331                         "IPV6",
1332                         ipv6_l3fwd_route_array[i].depth,
1333                         ipv6_l3fwd_route_array[i].if_out);
1334         }
1335 }
1336 #endif
1337
1338 static int
1339 init_mem(void)
1340 {
1341         struct lcore_conf *qconf;
1342         int socketid;
1343         unsigned lcore_id;
1344
1345         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1346                 if (rte_lcore_is_enabled(lcore_id) == 0)
1347                         continue;
1348
1349                 if (numa_on)
1350                         socketid = rte_lcore_to_socket_id(lcore_id);
1351                 else
1352                         socketid = 0;
1353
1354                 if (socketid >= NB_SOCKETS) {
1355                         rte_exit(EXIT_FAILURE,
1356                                 "Socket %d of lcore %u is out of range %d\n",
1357                                 socketid, lcore_id, NB_SOCKETS);
1358                 }
1359
1360 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1361                         setup_lpm(socketid);
1362 #else
1363                         setup_hash(socketid);
1364 #endif
1365                 qconf = &lcore_conf[lcore_id];
1366                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1367                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1368         }
1369         return 0;
1370 }
1371
1372 /* Check the link status of all ports in up to 9s, and print them finally */
1373 static void
1374 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
1375 {
1376 #define CHECK_INTERVAL 100 /* 100ms */
1377 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1378         uint8_t portid, count, all_ports_up, print_flag = 0;
1379         struct rte_eth_link link;
1380
1381         printf("\nChecking link status");
1382         fflush(stdout);
1383         for (count = 0; count <= MAX_CHECK_TIME; count++) {
1384                 all_ports_up = 1;
1385                 for (portid = 0; portid < port_num; portid++) {
1386                         if ((port_mask & (1 << portid)) == 0)
1387                                 continue;
1388                         memset(&link, 0, sizeof(link));
1389                         rte_eth_link_get_nowait(portid, &link);
1390                         /* print link status if flag set */
1391                         if (print_flag == 1) {
1392                                 if (link.link_status)
1393                                         printf("Port %d Link Up - speed %u "
1394                                                 "Mbps - %s\n", (uint8_t)portid,
1395                                                 (unsigned)link.link_speed,
1396                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1397                                         ("full-duplex") : ("half-duplex\n"));
1398                                 else
1399                                         printf("Port %d Link Down\n",
1400                                                 (uint8_t)portid);
1401                                 continue;
1402                         }
1403                         /* clear all_ports_up flag if any link down */
1404                         if (link.link_status == 0) {
1405                                 all_ports_up = 0;
1406                                 break;
1407                         }
1408                 }
1409                 /* after finally printing all link status, get out */
1410                 if (print_flag == 1)
1411                         break;
1412
1413                 if (all_ports_up == 0) {
1414                         printf(".");
1415                         fflush(stdout);
1416                         rte_delay_ms(CHECK_INTERVAL);
1417                 }
1418
1419                 /* set the print_flag if all ports up or timeout */
1420                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1421                         print_flag = 1;
1422                         printf("done\n");
1423                 }
1424         }
1425 }
1426 static void
1427 setup_port_tbl(struct lcore_conf *qconf, uint32_t lcore, int socket,
1428         uint32_t port)
1429 {
1430         struct mbuf_table *mtb;
1431         uint32_t n;
1432         size_t sz;
1433
1434         n = RTE_MAX(max_flow_num, 2UL * MAX_PKT_BURST);
1435         sz = sizeof (*mtb) + sizeof (mtb->m_table[0]) *  n;
1436
1437         if ((mtb = rte_zmalloc_socket(__func__, sz, CACHE_LINE_SIZE,
1438                         socket)) == NULL)
1439                 rte_exit(EXIT_FAILURE, "%s() for lcore: %u, port: %u "
1440                         "failed to allocate %zu bytes\n",
1441                         __func__, lcore, port, sz);
1442
1443         mtb->len = n;
1444         qconf->tx_mbufs[port] = mtb;
1445 }
1446
1447 static void
1448 setup_queue_tbl(struct lcore_conf *qconf, uint32_t lcore, int socket,
1449         uint32_t queue)
1450 {
1451         uint32_t nb_mbuf;
1452         uint64_t frag_cycles;
1453         char buf[RTE_MEMPOOL_NAMESIZE];
1454
1455         frag_cycles = (rte_get_tsc_hz() + MS_PER_S - 1) / MS_PER_S *
1456                 max_flow_ttl;
1457
1458         if ((qconf->frag_tbl[queue] = ipv4_frag_tbl_create(max_flow_num,
1459                         IPV4_FRAG_TBL_BUCKET_ENTRIES, max_flow_num, frag_cycles,
1460                         socket)) == NULL)
1461                 rte_exit(EXIT_FAILURE, "ipv4_frag_tbl_create(%u) on "
1462                         "lcore: %u for queue: %u failed\n",
1463                         max_flow_num, lcore, queue);
1464
1465         /*
1466          * At any given moment up to <max_flow_num * (MAX_FRAG_NUM - 1)>
1467          * mbufs could be stored int the fragment table.
1468          * Plus, each TX queue can hold up to <max_flow_num> packets.
1469          */
1470
1471         nb_mbuf = 2 * RTE_MAX(max_flow_num, 2UL * MAX_PKT_BURST) * MAX_FRAG_NUM;
1472         nb_mbuf *= (port_conf.rxmode.max_rx_pkt_len + BUF_SIZE - 1) / BUF_SIZE;
1473         nb_mbuf += RTE_TEST_RX_DESC_DEFAULT + RTE_TEST_TX_DESC_DEFAULT;
1474
1475         nb_mbuf = RTE_MAX(nb_mbuf, (uint32_t)DEF_MBUF_NUM);
1476
1477         rte_snprintf(buf, sizeof(buf), "mbuf_pool_%u_%u", lcore, queue);
1478
1479         if ((qconf->pool[queue] = rte_mempool_create(buf, nb_mbuf, MBUF_SIZE, 0,
1480                         sizeof(struct rte_pktmbuf_pool_private),
1481                         rte_pktmbuf_pool_init, NULL, rte_pktmbuf_init, NULL,
1482                         socket, MEMPOOL_F_SP_PUT | MEMPOOL_F_SC_GET)) == NULL)
1483                 rte_exit(EXIT_FAILURE, "mempool_create(%s) failed", buf);
1484 }
1485
1486 static void
1487 queue_dump_stat(void)
1488 {
1489         uint32_t i, lcore;
1490         const struct lcore_conf *qconf;
1491
1492         for (lcore = 0; lcore < RTE_MAX_LCORE; lcore++) {
1493                 if (rte_lcore_is_enabled(lcore) == 0)
1494                         continue;
1495
1496                 qconf = lcore_conf + lcore;
1497                 for (i = 0; i < qconf->n_rx_queue; i++) {
1498
1499                         fprintf(stdout, " -- lcoreid=%u portid=%hhu "
1500                                 "rxqueueid=%hhu frag tbl stat:\n",
1501                                 lcore,  qconf->rx_queue_list[i].port_id,
1502                                 qconf->rx_queue_list[i].queue_id);
1503                         ipv4_frag_tbl_dump_stat(stdout, qconf->frag_tbl[i]);
1504                         fprintf(stdout, "TX bursts:\t%" PRIu64 "\n"
1505                                 "TX packets _queued:\t%" PRIu64 "\n"
1506                                 "TX packets dropped:\t%" PRIu64 "\n"
1507                                 "TX packets send:\t%" PRIu64 "\n",
1508                                 qconf->tx_stat.call,
1509                                 qconf->tx_stat.queue,
1510                                 qconf->tx_stat.drop,
1511                                 qconf->tx_stat.send);
1512                 }
1513         }
1514 }
1515
1516 static void
1517 signal_handler(int signum)
1518 {
1519         queue_dump_stat();
1520         if (signum != SIGUSR1)
1521                 rte_exit(0, "received signal: %d, exiting\n", signum);
1522 }
1523
1524 int
1525 MAIN(int argc, char **argv)
1526 {
1527         struct lcore_conf *qconf;
1528         int ret;
1529         unsigned nb_ports;
1530         uint16_t queueid;
1531         unsigned lcore_id;
1532         uint32_t n_tx_queue, nb_lcores;
1533         uint8_t portid, nb_rx_queue, queue, socketid;
1534
1535         /* init EAL */
1536         ret = rte_eal_init(argc, argv);
1537         if (ret < 0)
1538                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1539         argc -= ret;
1540         argv += ret;
1541
1542         /* parse application arguments (after the EAL ones) */
1543         ret = parse_args(argc, argv);
1544         if (ret < 0)
1545                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1546
1547         if (check_lcore_params() < 0)
1548                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1549
1550         ret = init_lcore_rx_queues();
1551         if (ret < 0)
1552                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1553
1554
1555         /* init driver(s) */
1556         if (rte_pmd_init_all() < 0)
1557                 rte_exit(EXIT_FAILURE, "Cannot init pmd\n");
1558
1559         if (rte_eal_pci_probe() < 0)
1560                 rte_exit(EXIT_FAILURE, "Cannot probe PCI\n");
1561
1562         nb_ports = rte_eth_dev_count();
1563         if (nb_ports > MAX_PORTS)
1564                 nb_ports = MAX_PORTS;
1565
1566         if (check_port_config(nb_ports) < 0)
1567                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1568
1569         nb_lcores = rte_lcore_count();
1570
1571         /* initialize all ports */
1572         for (portid = 0; portid < nb_ports; portid++) {
1573                 /* skip ports that are not enabled */
1574                 if ((enabled_port_mask & (1 << portid)) == 0) {
1575                         printf("\nSkipping disabled port %d\n", portid);
1576                         continue;
1577                 }
1578
1579                 /* init port */
1580                 printf("Initializing port %d ... ", portid );
1581                 fflush(stdout);
1582
1583                 nb_rx_queue = get_port_n_rx_queues(portid);
1584                 n_tx_queue = nb_lcores;
1585                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
1586                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1587                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1588                         nb_rx_queue, (unsigned)n_tx_queue );
1589                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
1590                                         (uint16_t)n_tx_queue, &port_conf);
1591                 if (ret < 0)
1592                         rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
1593                                 ret, portid);
1594
1595                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1596                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1597                 printf(", ");
1598
1599                 /* init memory */
1600                 ret = init_mem();
1601                 if (ret < 0)
1602                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
1603
1604                 /* init one TX queue per couple (lcore,port) */
1605                 queueid = 0;
1606                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1607                         if (rte_lcore_is_enabled(lcore_id) == 0)
1608                                 continue;
1609
1610                         if (numa_on)
1611                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1612                         else
1613                                 socketid = 0;
1614
1615                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1616                         fflush(stdout);
1617                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1618                                                      socketid, &tx_conf);
1619                         if (ret < 0)
1620                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
1621                                         "port=%d\n", ret, portid);
1622
1623                         qconf = &lcore_conf[lcore_id];
1624                         qconf->tx_queue_id[portid] = queueid;
1625                         setup_port_tbl(qconf, lcore_id, socketid, portid);
1626                         queueid++;
1627                 }
1628                 printf("\n");
1629         }
1630
1631         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1632                 if (rte_lcore_is_enabled(lcore_id) == 0)
1633                         continue;
1634                 qconf = &lcore_conf[lcore_id];
1635                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1636                 fflush(stdout);
1637                 /* init RX queues */
1638                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1639                         portid = qconf->rx_queue_list[queue].port_id;
1640                         queueid = qconf->rx_queue_list[queue].queue_id;
1641
1642                         if (numa_on)
1643                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1644                         else
1645                                 socketid = 0;
1646
1647                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1648                         fflush(stdout);
1649
1650                         setup_queue_tbl(qconf, lcore_id, socketid, queue);
1651
1652                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1653                                         socketid, &rx_conf, qconf->pool[queue]);
1654                         if (ret < 0)
1655                                 rte_exit(EXIT_FAILURE,
1656                                         "rte_eth_rx_queue_setup: err=%d,"
1657                                         "port=%d\n", ret, portid);
1658                 }
1659         }
1660
1661         printf("\n");
1662
1663         /* start ports */
1664         for (portid = 0; portid < nb_ports; portid++) {
1665                 if ((enabled_port_mask & (1 << portid)) == 0) {
1666                         continue;
1667                 }
1668                 /* Start device */
1669                 ret = rte_eth_dev_start(portid);
1670                 if (ret < 0)
1671                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1672                                 ret, portid);
1673
1674                 /*
1675                  * If enabled, put device in promiscuous mode.
1676                  * This allows IO forwarding mode to forward packets
1677                  * to itself through 2 cross-connected  ports of the
1678                  * target machine.
1679                  */
1680                 if (promiscuous_on)
1681                         rte_eth_promiscuous_enable(portid);
1682         }
1683
1684         check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
1685
1686         signal(SIGUSR1, signal_handler);
1687         signal(SIGTERM, signal_handler);
1688         signal(SIGINT, signal_handler);
1689
1690         /* launch per-lcore init on every lcore */
1691         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1692         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1693                 if (rte_eal_wait_lcore(lcore_id) < 0)
1694                         return -1;
1695         }
1696
1697         return 0;
1698 }