examples/l3fwd/l3fwd_lpm.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2016 Intel Corporation
   3  */
   4
   5 #include <stdio.h>
   6 #include <stdlib.h>
   7 #include <stdint.h>
   8 #include <inttypes.h>
   9 #include <sys/types.h>
  10 #include <string.h>
  11 #include <sys/queue.h>
  12 #include <stdarg.h>
  13 #include <errno.h>
  14 #include <getopt.h>
  15 #include <stdbool.h>
  16
  17 #include <rte_debug.h>
  18 #include <rte_ether.h>
  19 #include <rte_ethdev.h>
  20 #include <rte_cycles.h>
  21 #include <rte_mbuf.h>
  22 #include <rte_ip.h>
  23 #include <rte_tcp.h>
  24 #include <rte_udp.h>
  25 #include <rte_lpm.h>
  26 #include <rte_lpm6.h>
  27
  28 #include "l3fwd.h"
  29
  30 struct ipv4_l3fwd_lpm_route {
  31         uint32_t ip;
  32         uint8_t  depth;
  33         uint8_t  if_out;
  34 };
  35
  36 struct ipv6_l3fwd_lpm_route {
  37         uint8_t ip[16];
  38         uint8_t  depth;
  39         uint8_t  if_out;
  40 };
  41
  42 /* 192.18.0.0/16 are set aside for RFC2544 benchmarking. */
  43 static struct ipv4_l3fwd_lpm_route ipv4_l3fwd_lpm_route_array[] = {
  44         {IPv4(192, 18, 0, 0), 24, 0},
  45         {IPv4(192, 18, 1, 0), 24, 1},
  46         {IPv4(192, 18, 2, 0), 24, 2},
  47         {IPv4(192, 18, 3, 0), 24, 3},
  48         {IPv4(192, 18, 4, 0), 24, 4},
  49         {IPv4(192, 18, 5, 0), 24, 5},
  50         {IPv4(192, 18, 6, 0), 24, 6},
  51         {IPv4(192, 18, 7, 0), 24, 7},
  52 };
  53
  54 /* 2001:0200::/48 is IANA reserved range for IPv6 benchmarking (RFC5180) */
  55 static struct ipv6_l3fwd_lpm_route ipv6_l3fwd_lpm_route_array[] = {
  56         {{32, 1, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}, 48, 0},
  57         {{32, 1, 2, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0}, 48, 1},
  58         {{32, 1, 2, 0, 0, 0, 0, 0, 0, 2, 0, 0, 0, 0, 0, 0}, 48, 2},
  59         {{32, 1, 2, 0, 0, 0, 0, 0, 0, 3, 0, 0, 0, 0, 0, 0}, 48, 3},
  60         {{32, 1, 2, 0, 0, 0, 0, 0, 0, 4, 0, 0, 0, 0, 0, 0}, 48, 4},
  61         {{32, 1, 2, 0, 0, 0, 0, 0, 0, 5, 0, 0, 0, 0, 0, 0}, 48, 5},
  62         {{32, 1, 2, 0, 0, 0, 0, 0, 0, 6, 0, 0, 0, 0, 0, 0}, 48, 6},
  63         {{32, 1, 2, 0, 0, 0, 0, 0, 0, 7, 0, 0, 0, 0, 0, 0}, 48, 7},
  64 };
  65
  66 #define IPV4_L3FWD_LPM_NUM_ROUTES \
  67         (sizeof(ipv4_l3fwd_lpm_route_array) / sizeof(ipv4_l3fwd_lpm_route_array[0]))
  68 #define IPV6_L3FWD_LPM_NUM_ROUTES \
  69         (sizeof(ipv6_l3fwd_lpm_route_array) / sizeof(ipv6_l3fwd_lpm_route_array[0]))
  70
  71 #define IPV4_L3FWD_LPM_MAX_RULES         1024
  72 #define IPV4_L3FWD_LPM_NUMBER_TBL8S (1 << 8)
  73 #define IPV6_L3FWD_LPM_MAX_RULES         1024
  74 #define IPV6_L3FWD_LPM_NUMBER_TBL8S (1 << 16)
  75
  76 struct rte_lpm *ipv4_l3fwd_lpm_lookup_struct[NB_SOCKETS];
  77 struct rte_lpm6 *ipv6_l3fwd_lpm_lookup_struct[NB_SOCKETS];
  78
  79 static inline uint16_t
  80 lpm_get_ipv4_dst_port(void *ipv4_hdr, uint16_t portid, void *lookup_struct)
  81 {
  82         uint32_t next_hop;
  83         struct rte_lpm *ipv4_l3fwd_lookup_struct =
  84                 (struct rte_lpm *)lookup_struct;
  85
  86         return (uint16_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
  87                 rte_be_to_cpu_32(((struct ipv4_hdr *)ipv4_hdr)->dst_addr),
  88                 &next_hop) == 0) ? next_hop : portid);
  89 }
  90
  91 static inline uint16_t
  92 lpm_get_ipv6_dst_port(void *ipv6_hdr, uint16_t portid, void *lookup_struct)
  93 {
  94         uint32_t next_hop;
  95         struct rte_lpm6 *ipv6_l3fwd_lookup_struct =
  96                 (struct rte_lpm6 *)lookup_struct;
  97
  98         return (uint16_t) ((rte_lpm6_lookup(ipv6_l3fwd_lookup_struct,
  99                         ((struct ipv6_hdr *)ipv6_hdr)->dst_addr,
 100                         &next_hop) == 0) ?  next_hop : portid);
 101 }
 102
 103 static __rte_always_inline uint16_t
 104 lpm_get_dst_port(const struct lcore_conf *qconf, struct rte_mbuf *pkt,
 105                 uint16_t portid)
 106 {
 107         struct ipv6_hdr *ipv6_hdr;
 108         struct ipv4_hdr *ipv4_hdr;
 109         struct ether_hdr *eth_hdr;
 110
 111         if (RTE_ETH_IS_IPV4_HDR(pkt->packet_type)) {
 112
 113                 eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
 114                 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
 115
 116                 return lpm_get_ipv4_dst_port(ipv4_hdr, portid,
 117                                              qconf->ipv4_lookup_struct);
 118         } else if (RTE_ETH_IS_IPV6_HDR(pkt->packet_type)) {
 119
 120                 eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
 121                 ipv6_hdr = (struct ipv6_hdr *)(eth_hdr + 1);
 122
 123                 return lpm_get_ipv6_dst_port(ipv6_hdr, portid,
 124                                              qconf->ipv6_lookup_struct);
 125         }
 126
 127         return portid;
 128 }
 129
 130 /*
 131  * lpm_get_dst_port optimized routine for packets where dst_ipv4 is already
 132  * precalculated. If packet is ipv6 dst_addr is taken directly from packet
 133  * header and dst_ipv4 value is not used.
 134  */
 135 static __rte_always_inline uint16_t
 136 lpm_get_dst_port_with_ipv4(const struct lcore_conf *qconf, struct rte_mbuf *pkt,
 137         uint32_t dst_ipv4, uint16_t portid)
 138 {
 139         uint32_t next_hop;
 140         struct ipv6_hdr *ipv6_hdr;
 141         struct ether_hdr *eth_hdr;
 142
 143         if (RTE_ETH_IS_IPV4_HDR(pkt->packet_type)) {
 144                 return (uint16_t) ((rte_lpm_lookup(qconf->ipv4_lookup_struct,
 145                                                    dst_ipv4, &next_hop) == 0)
 146                                    ? next_hop : portid);
 147
 148         } else if (RTE_ETH_IS_IPV6_HDR(pkt->packet_type)) {
 149
 150                 eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
 151                 ipv6_hdr = (struct ipv6_hdr *)(eth_hdr + 1);
 152
 153                 return (uint16_t) ((rte_lpm6_lookup(qconf->ipv6_lookup_struct,
 154                                 ipv6_hdr->dst_addr, &next_hop) == 0)
 155                                 ? next_hop : portid);
 156
 157         }
 158
 159         return portid;
 160 }
 161
 162 #if defined(RTE_ARCH_X86)
 163 #include "l3fwd_lpm_sse.h"
 164 #elif defined RTE_MACHINE_CPUFLAG_NEON
 165 #include "l3fwd_lpm_neon.h"
 166 #elif defined(RTE_ARCH_PPC_64)
 167 #include "l3fwd_lpm_altivec.h"
 168 #else
 169 #include "l3fwd_lpm.h"
 170 #endif
 171
 172 /* main processing loop */
 173 int
 174 lpm_main_loop(__attribute__((unused)) void *dummy)
 175 {
 176         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 177         unsigned lcore_id;
 178         uint64_t prev_tsc, diff_tsc, cur_tsc;
 179         int i, nb_rx;
 180         uint16_t portid;
 181         uint8_t queueid;
 182         struct lcore_conf *qconf;
 183         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
 184                 US_PER_S * BURST_TX_DRAIN_US;
 185
 186         prev_tsc = 0;
 187
 188         lcore_id = rte_lcore_id();
 189         qconf = &lcore_conf[lcore_id];
 190
 191         if (qconf->n_rx_queue == 0) {
 192                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
 193                 return 0;
 194         }
 195
 196         RTE_LOG(INFO, L3FWD, "entering main loop on lcore %u\n", lcore_id);
 197
 198         for (i = 0; i < qconf->n_rx_queue; i++) {
 199
 200                 portid = qconf->rx_queue_list[i].port_id;
 201                 queueid = qconf->rx_queue_list[i].queue_id;
 202                 RTE_LOG(INFO, L3FWD,
 203                         " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
 204                         lcore_id, portid, queueid);
 205         }
 206
 207         while (!force_quit) {
 208
 209                 cur_tsc = rte_rdtsc();
 210
 211                 /*
 212                  * TX burst queue drain
 213                  */
 214                 diff_tsc = cur_tsc - prev_tsc;
 215                 if (unlikely(diff_tsc > drain_tsc)) {
 216
 217                         for (i = 0; i < qconf->n_tx_port; ++i) {
 218                                 portid = qconf->tx_port_id[i];
 219                                 if (qconf->tx_mbufs[portid].len == 0)
 220                                         continue;
 221                                 send_burst(qconf,
 222                                         qconf->tx_mbufs[portid].len,
 223                                         portid);
 224                                 qconf->tx_mbufs[portid].len = 0;
 225                         }
 226
 227                         prev_tsc = cur_tsc;
 228                 }
 229
 230                 /*
 231                  * Read packet from RX queues
 232                  */
 233                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 234                         portid = qconf->rx_queue_list[i].port_id;
 235                         queueid = qconf->rx_queue_list[i].queue_id;
 236                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
 237                                 MAX_PKT_BURST);
 238                         if (nb_rx == 0)
 239                                 continue;
 240
 241 #if defined RTE_ARCH_X86 || defined RTE_MACHINE_CPUFLAG_NEON \
 242                          || defined RTE_ARCH_PPC_64
 243                         l3fwd_lpm_send_packets(nb_rx, pkts_burst,
 244                                                 portid, qconf);
 245 #else
 246                         l3fwd_lpm_no_opt_send_packets(nb_rx, pkts_burst,
 247                                                         portid, qconf);
 248 #endif /* X86 */
 249                 }
 250         }
 251
 252         return 0;
 253 }
 254
 255 void
 256 setup_lpm(const int socketid)
 257 {
 258         struct rte_lpm6_config config;
 259         struct rte_lpm_config config_ipv4;
 260         unsigned i;
 261         int ret;
 262         char s[64];
 263
 264         /* create the LPM table */
 265         config_ipv4.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
 266         config_ipv4.number_tbl8s = IPV4_L3FWD_LPM_NUMBER_TBL8S;
 267         config_ipv4.flags = 0;
 268         snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
 269         ipv4_l3fwd_lpm_lookup_struct[socketid] =
 270                         rte_lpm_create(s, socketid, &config_ipv4);
 271         if (ipv4_l3fwd_lpm_lookup_struct[socketid] == NULL)
 272                 rte_exit(EXIT_FAILURE,
 273                         "Unable to create the l3fwd LPM table on socket %d\n",
 274                         socketid);
 275
 276         /* populate the LPM table */
 277         for (i = 0; i < IPV4_L3FWD_LPM_NUM_ROUTES; i++) {
 278
 279                 /* skip unused ports */
 280                 if ((1 << ipv4_l3fwd_lpm_route_array[i].if_out &
 281                                 enabled_port_mask) == 0)
 282                         continue;
 283
 284                 ret = rte_lpm_add(ipv4_l3fwd_lpm_lookup_struct[socketid],
 285                         ipv4_l3fwd_lpm_route_array[i].ip,
 286                         ipv4_l3fwd_lpm_route_array[i].depth,
 287                         ipv4_l3fwd_lpm_route_array[i].if_out);
 288
 289                 if (ret < 0) {
 290                         rte_exit(EXIT_FAILURE,
 291                                 "Unable to add entry %u to the l3fwd LPM table on socket %d\n",
 292                                 i, socketid);
 293                 }
 294
 295                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
 296                         (unsigned)ipv4_l3fwd_lpm_route_array[i].ip,
 297                         ipv4_l3fwd_lpm_route_array[i].depth,
 298                         ipv4_l3fwd_lpm_route_array[i].if_out);
 299         }
 300
 301         /* create the LPM6 table */
 302         snprintf(s, sizeof(s), "IPV6_L3FWD_LPM_%d", socketid);
 303
 304         config.max_rules = IPV6_L3FWD_LPM_MAX_RULES;
 305         config.number_tbl8s = IPV6_L3FWD_LPM_NUMBER_TBL8S;
 306         config.flags = 0;
 307         ipv6_l3fwd_lpm_lookup_struct[socketid] = rte_lpm6_create(s, socketid,
 308                                 &config);
 309         if (ipv6_l3fwd_lpm_lookup_struct[socketid] == NULL)
 310                 rte_exit(EXIT_FAILURE,
 311                         "Unable to create the l3fwd LPM table on socket %d\n",
 312                         socketid);
 313
 314         /* populate the LPM table */
 315         for (i = 0; i < IPV6_L3FWD_LPM_NUM_ROUTES; i++) {
 316
 317                 /* skip unused ports */
 318                 if ((1 << ipv6_l3fwd_lpm_route_array[i].if_out &
 319                                 enabled_port_mask) == 0)
 320                         continue;
 321
 322                 ret = rte_lpm6_add(ipv6_l3fwd_lpm_lookup_struct[socketid],
 323                         ipv6_l3fwd_lpm_route_array[i].ip,
 324                         ipv6_l3fwd_lpm_route_array[i].depth,
 325                         ipv6_l3fwd_lpm_route_array[i].if_out);
 326
 327                 if (ret < 0) {
 328                         rte_exit(EXIT_FAILURE,
 329                                 "Unable to add entry %u to the l3fwd LPM table on socket %d\n",
 330                                 i, socketid);
 331                 }
 332
 333                 printf("LPM: Adding route %s / %d (%d)\n",
 334                         "IPV6",
 335                         ipv6_l3fwd_lpm_route_array[i].depth,
 336                         ipv6_l3fwd_lpm_route_array[i].if_out);
 337         }
 338 }
 339
 340 int
 341 lpm_check_ptype(int portid)
 342 {
 343         int i, ret;
 344         int ptype_l3_ipv4 = 0, ptype_l3_ipv6 = 0;
 345         uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
 346
 347         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
 348         if (ret <= 0)
 349                 return 0;
 350
 351         uint32_t ptypes[ret];
 352
 353         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
 354         for (i = 0; i < ret; ++i) {
 355                 if (ptypes[i] & RTE_PTYPE_L3_IPV4)
 356                         ptype_l3_ipv4 = 1;
 357                 if (ptypes[i] & RTE_PTYPE_L3_IPV6)
 358                         ptype_l3_ipv6 = 1;
 359         }
 360
 361         if (ptype_l3_ipv4 == 0)
 362                 printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
 363
 364         if (ptype_l3_ipv6 == 0)
 365                 printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
 366
 367         if (ptype_l3_ipv4 && ptype_l3_ipv6)
 368                 return 1;
 369
 370         return 0;
 371
 372 }
 373
 374 static inline void
 375 lpm_parse_ptype(struct rte_mbuf *m)
 376 {
 377         struct ether_hdr *eth_hdr;
 378         uint32_t packet_type = RTE_PTYPE_UNKNOWN;
 379         uint16_t ether_type;
 380
 381         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 382         ether_type = eth_hdr->ether_type;
 383         if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4))
 384                 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
 385         else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6))
 386                 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
 387
 388         m->packet_type = packet_type;
 389 }
 390
 391 uint16_t
 392 lpm_cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
 393                    struct rte_mbuf *pkts[], uint16_t nb_pkts,
 394                    uint16_t max_pkts __rte_unused,
 395                    void *user_param __rte_unused)
 396 {
 397         unsigned i;
 398
 399         for (i = 0; i < nb_pkts; ++i)
 400                 lpm_parse_ptype(pkts[i]);
 401
 402         return nb_pkts;
 403 }
 404
 405 /* Return ipv4/ipv6 lpm fwd lookup struct. */
 406 void *
 407 lpm_get_ipv4_l3fwd_lookup_struct(const int socketid)
 408 {
 409         return ipv4_l3fwd_lpm_lookup_struct[socketid];
 410 }
 411
 412 void *
 413 lpm_get_ipv6_l3fwd_lookup_struct(const int socketid)
 414 {
 415         return ipv6_l3fwd_lpm_lookup_struct[socketid];
 416 }