1 /* SPDX-License-Identifier: BSD-3-Clause 2 * Copyright(c) 2010-2014 Intel Corporation. 3 * Copyright 2014 6WIND S.A. 4 */ 5 6 #include <stdarg.h> 7 #include <stdio.h> 8 #include <errno.h> 9 #include <stdint.h> 10 #include <unistd.h> 11 #include <inttypes.h> 12 13 #include <sys/queue.h> 14 #include <sys/stat.h> 15 16 #include <rte_common.h> 17 #include <rte_byteorder.h> 18 #include <rte_log.h> 19 #include <rte_debug.h> 20 #include <rte_cycles.h> 21 #include <rte_memory.h> 22 #include <rte_memcpy.h> 23 #include <rte_launch.h> 24 #include <rte_eal.h> 25 #include <rte_per_lcore.h> 26 #include <rte_lcore.h> 27 #include <rte_atomic.h> 28 #include <rte_branch_prediction.h> 29 #include <rte_mempool.h> 30 #include <rte_mbuf.h> 31 #include <rte_interrupts.h> 32 #include <rte_pci.h> 33 #include <rte_ether.h> 34 #include <rte_ethdev.h> 35 #include <rte_ip.h> 36 #include <rte_tcp.h> 37 #include <rte_udp.h> 38 #include <rte_vxlan.h> 39 #include <rte_sctp.h> 40 #include <rte_gtp.h> 41 #include <rte_prefetch.h> 42 #include <rte_string_fns.h> 43 #include <rte_flow.h> 44 #include <rte_gro.h> 45 #include <rte_gso.h> 46 47 #include "testpmd.h" 48 49 #define IP_DEFTTL 64 /* from RFC 1340. */ 50 51 #define GRE_CHECKSUM_PRESENT 0x8000 52 #define GRE_KEY_PRESENT 0x2000 53 #define GRE_SEQUENCE_PRESENT 0x1000 54 #define GRE_EXT_LEN 4 55 #define GRE_SUPPORTED_FIELDS (GRE_CHECKSUM_PRESENT | GRE_KEY_PRESENT |\ 56 GRE_SEQUENCE_PRESENT) 57 58 /* We cannot use rte_cpu_to_be_16() on a constant in a switch/case */ 59 #if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN 60 #define _htons(x) ((uint16_t)((((x) & 0x00ffU) << 8) | (((x) & 0xff00U) >> 8))) 61 #else 62 #define _htons(x) (x) 63 #endif 64 65 uint16_t vxlan_gpe_udp_port = 4790; 66 67 /* structure that caches offload info for the current packet */ 68 struct testpmd_offload_info { 69 uint16_t ethertype; 70 uint8_t gso_enable; 71 uint16_t l2_len; 72 uint16_t l3_len; 73 uint16_t l4_len; 74 uint8_t l4_proto; 75 uint8_t is_tunnel; 76 uint16_t outer_ethertype; 77 uint16_t outer_l2_len; 78 uint16_t outer_l3_len; 79 uint8_t outer_l4_proto; 80 uint16_t tso_segsz; 81 uint16_t tunnel_tso_segsz; 82 uint32_t pkt_len; 83 }; 84 85 /* simplified GRE header */ 86 struct simple_gre_hdr { 87 uint16_t flags; 88 uint16_t proto; 89 } __rte_packed; 90 91 static uint16_t 92 get_udptcp_checksum(void *l3_hdr, void *l4_hdr, uint16_t ethertype) 93 { 94 if (ethertype == _htons(RTE_ETHER_TYPE_IPV4)) 95 return rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr); 96 else /* assume ethertype == RTE_ETHER_TYPE_IPV6 */ 97 return rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr); 98 } 99 100 /* Parse an IPv4 header to fill l3_len, l4_len, and l4_proto */ 101 static void 102 parse_ipv4(struct rte_ipv4_hdr *ipv4_hdr, struct testpmd_offload_info *info) 103 { 104 struct rte_tcp_hdr *tcp_hdr; 105 106 info->l3_len = (ipv4_hdr->version_ihl & 0x0f) * 4; 107 info->l4_proto = ipv4_hdr->next_proto_id; 108 109 /* only fill l4_len for TCP, it's useful for TSO */ 110 if (info->l4_proto == IPPROTO_TCP) { 111 tcp_hdr = (struct rte_tcp_hdr *) 112 ((char *)ipv4_hdr + info->l3_len); 113 info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2; 114 } else if (info->l4_proto == IPPROTO_UDP) 115 info->l4_len = sizeof(struct rte_udp_hdr); 116 else 117 info->l4_len = 0; 118 } 119 120 /* Parse an IPv6 header to fill l3_len, l4_len, and l4_proto */ 121 static void 122 parse_ipv6(struct rte_ipv6_hdr *ipv6_hdr, struct testpmd_offload_info *info) 123 { 124 struct rte_tcp_hdr *tcp_hdr; 125 126 info->l3_len = sizeof(struct rte_ipv6_hdr); 127 info->l4_proto = ipv6_hdr->proto; 128 129 /* only fill l4_len for TCP, it's useful for TSO */ 130 if (info->l4_proto == IPPROTO_TCP) { 131 tcp_hdr = (struct rte_tcp_hdr *) 132 ((char *)ipv6_hdr + info->l3_len); 133 info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2; 134 } else if (info->l4_proto == IPPROTO_UDP) 135 info->l4_len = sizeof(struct rte_udp_hdr); 136 else 137 info->l4_len = 0; 138 } 139 140 /* 141 * Parse an ethernet header to fill the ethertype, l2_len, l3_len and 142 * ipproto. This function is able to recognize IPv4/IPv6 with optional VLAN 143 * headers. The l4_len argument is only set in case of TCP (useful for TSO). 144 */ 145 static void 146 parse_ethernet(struct rte_ether_hdr *eth_hdr, struct testpmd_offload_info *info) 147 { 148 struct rte_ipv4_hdr *ipv4_hdr; 149 struct rte_ipv6_hdr *ipv6_hdr; 150 struct rte_vlan_hdr *vlan_hdr; 151 152 info->l2_len = sizeof(struct rte_ether_hdr); 153 info->ethertype = eth_hdr->ether_type; 154 155 while (info->ethertype == _htons(RTE_ETHER_TYPE_VLAN) || 156 info->ethertype == _htons(RTE_ETHER_TYPE_QINQ)) { 157 vlan_hdr = (struct rte_vlan_hdr *) 158 ((char *)eth_hdr + info->l2_len); 159 info->l2_len += sizeof(struct rte_vlan_hdr); 160 info->ethertype = vlan_hdr->eth_proto; 161 } 162 163 switch (info->ethertype) { 164 case _htons(RTE_ETHER_TYPE_IPV4): 165 ipv4_hdr = (struct rte_ipv4_hdr *) 166 ((char *)eth_hdr + info->l2_len); 167 parse_ipv4(ipv4_hdr, info); 168 break; 169 case _htons(RTE_ETHER_TYPE_IPV6): 170 ipv6_hdr = (struct rte_ipv6_hdr *) 171 ((char *)eth_hdr + info->l2_len); 172 parse_ipv6(ipv6_hdr, info); 173 break; 174 default: 175 info->l4_len = 0; 176 info->l3_len = 0; 177 info->l4_proto = 0; 178 break; 179 } 180 } 181 182 /* 183 * Parse a GTP protocol header. 184 * No optional fields and next extension header type. 185 */ 186 static void 187 parse_gtp(struct rte_udp_hdr *udp_hdr, 188 struct testpmd_offload_info *info) 189 { 190 struct rte_ipv4_hdr *ipv4_hdr; 191 struct rte_ipv6_hdr *ipv6_hdr; 192 struct rte_gtp_hdr *gtp_hdr; 193 uint8_t gtp_len = sizeof(*gtp_hdr); 194 uint8_t ip_ver; 195 196 /* Check udp destination port. */ 197 if (udp_hdr->dst_port != _htons(RTE_GTPC_UDP_PORT) && 198 udp_hdr->src_port != _htons(RTE_GTPC_UDP_PORT) && 199 udp_hdr->dst_port != _htons(RTE_GTPU_UDP_PORT)) 200 return; 201 202 info->is_tunnel = 1; 203 info->outer_ethertype = info->ethertype; 204 info->outer_l2_len = info->l2_len; 205 info->outer_l3_len = info->l3_len; 206 info->outer_l4_proto = info->l4_proto; 207 info->l2_len = 0; 208 209 gtp_hdr = (struct rte_gtp_hdr *)((char *)udp_hdr + 210 sizeof(struct rte_udp_hdr)); 211 212 /* 213 * Check message type. If message type is 0xff, it is 214 * a GTP data packet. If not, it is a GTP control packet 215 */ 216 if (gtp_hdr->msg_type == 0xff) { 217 ip_ver = *(uint8_t *)((char *)udp_hdr + 218 sizeof(struct rte_udp_hdr) + 219 sizeof(struct rte_gtp_hdr)); 220 ip_ver = (ip_ver) & 0xf0; 221 222 if (ip_ver == RTE_GTP_TYPE_IPV4) { 223 ipv4_hdr = (struct rte_ipv4_hdr *)((char *)gtp_hdr + 224 gtp_len); 225 info->ethertype = _htons(RTE_ETHER_TYPE_IPV4); 226 parse_ipv4(ipv4_hdr, info); 227 } else if (ip_ver == RTE_GTP_TYPE_IPV6) { 228 ipv6_hdr = (struct rte_ipv6_hdr *)((char *)gtp_hdr + 229 gtp_len); 230 info->ethertype = _htons(RTE_ETHER_TYPE_IPV6); 231 parse_ipv6(ipv6_hdr, info); 232 } 233 } else { 234 info->ethertype = 0; 235 info->l4_len = 0; 236 info->l3_len = 0; 237 info->l4_proto = 0; 238 } 239 240 info->l2_len += RTE_ETHER_GTP_HLEN; 241 } 242 243 /* Parse a vxlan header */ 244 static void 245 parse_vxlan(struct rte_udp_hdr *udp_hdr, 246 struct testpmd_offload_info *info, 247 uint32_t pkt_type) 248 { 249 struct rte_ether_hdr *eth_hdr; 250 251 /* check udp destination port, 4789 is the default vxlan port 252 * (rfc7348) or that the rx offload flag is set (i40e only 253 * currently) */ 254 if (udp_hdr->dst_port != _htons(4789) && 255 RTE_ETH_IS_TUNNEL_PKT(pkt_type) == 0) 256 return; 257 258 info->is_tunnel = 1; 259 info->outer_ethertype = info->ethertype; 260 info->outer_l2_len = info->l2_len; 261 info->outer_l3_len = info->l3_len; 262 info->outer_l4_proto = info->l4_proto; 263 264 eth_hdr = (struct rte_ether_hdr *)((char *)udp_hdr + 265 sizeof(struct rte_udp_hdr) + 266 sizeof(struct rte_vxlan_hdr)); 267 268 parse_ethernet(eth_hdr, info); 269 info->l2_len += RTE_ETHER_VXLAN_HLEN; /* add udp + vxlan */ 270 } 271 272 /* Parse a vxlan-gpe header */ 273 static void 274 parse_vxlan_gpe(struct rte_udp_hdr *udp_hdr, 275 struct testpmd_offload_info *info) 276 { 277 struct rte_ether_hdr *eth_hdr; 278 struct rte_ipv4_hdr *ipv4_hdr; 279 struct rte_ipv6_hdr *ipv6_hdr; 280 struct rte_vxlan_gpe_hdr *vxlan_gpe_hdr; 281 uint8_t vxlan_gpe_len = sizeof(*vxlan_gpe_hdr); 282 283 /* Check udp destination port. */ 284 if (udp_hdr->dst_port != _htons(vxlan_gpe_udp_port)) 285 return; 286 287 vxlan_gpe_hdr = (struct rte_vxlan_gpe_hdr *)((char *)udp_hdr + 288 sizeof(struct rte_udp_hdr)); 289 290 if (!vxlan_gpe_hdr->proto || vxlan_gpe_hdr->proto == 291 RTE_VXLAN_GPE_TYPE_IPV4) { 292 info->is_tunnel = 1; 293 info->outer_ethertype = info->ethertype; 294 info->outer_l2_len = info->l2_len; 295 info->outer_l3_len = info->l3_len; 296 info->outer_l4_proto = info->l4_proto; 297 298 ipv4_hdr = (struct rte_ipv4_hdr *)((char *)vxlan_gpe_hdr + 299 vxlan_gpe_len); 300 301 parse_ipv4(ipv4_hdr, info); 302 info->ethertype = _htons(RTE_ETHER_TYPE_IPV4); 303 info->l2_len = 0; 304 305 } else if (vxlan_gpe_hdr->proto == RTE_VXLAN_GPE_TYPE_IPV6) { 306 info->is_tunnel = 1; 307 info->outer_ethertype = info->ethertype; 308 info->outer_l2_len = info->l2_len; 309 info->outer_l3_len = info->l3_len; 310 info->outer_l4_proto = info->l4_proto; 311 312 ipv6_hdr = (struct rte_ipv6_hdr *)((char *)vxlan_gpe_hdr + 313 vxlan_gpe_len); 314 315 info->ethertype = _htons(RTE_ETHER_TYPE_IPV6); 316 parse_ipv6(ipv6_hdr, info); 317 info->l2_len = 0; 318 319 } else if (vxlan_gpe_hdr->proto == RTE_VXLAN_GPE_TYPE_ETH) { 320 info->is_tunnel = 1; 321 info->outer_ethertype = info->ethertype; 322 info->outer_l2_len = info->l2_len; 323 info->outer_l3_len = info->l3_len; 324 info->outer_l4_proto = info->l4_proto; 325 326 eth_hdr = (struct rte_ether_hdr *)((char *)vxlan_gpe_hdr + 327 vxlan_gpe_len); 328 329 parse_ethernet(eth_hdr, info); 330 } else 331 return; 332 333 info->l2_len += RTE_ETHER_VXLAN_GPE_HLEN; 334 } 335 336 /* Parse a gre header */ 337 static void 338 parse_gre(struct simple_gre_hdr *gre_hdr, struct testpmd_offload_info *info) 339 { 340 struct rte_ether_hdr *eth_hdr; 341 struct rte_ipv4_hdr *ipv4_hdr; 342 struct rte_ipv6_hdr *ipv6_hdr; 343 uint8_t gre_len = 0; 344 345 gre_len += sizeof(struct simple_gre_hdr); 346 347 if (gre_hdr->flags & _htons(GRE_KEY_PRESENT)) 348 gre_len += GRE_EXT_LEN; 349 if (gre_hdr->flags & _htons(GRE_SEQUENCE_PRESENT)) 350 gre_len += GRE_EXT_LEN; 351 if (gre_hdr->flags & _htons(GRE_CHECKSUM_PRESENT)) 352 gre_len += GRE_EXT_LEN; 353 354 if (gre_hdr->proto == _htons(RTE_ETHER_TYPE_IPV4)) { 355 info->is_tunnel = 1; 356 info->outer_ethertype = info->ethertype; 357 info->outer_l2_len = info->l2_len; 358 info->outer_l3_len = info->l3_len; 359 info->outer_l4_proto = info->l4_proto; 360 361 ipv4_hdr = (struct rte_ipv4_hdr *)((char *)gre_hdr + gre_len); 362 363 parse_ipv4(ipv4_hdr, info); 364 info->ethertype = _htons(RTE_ETHER_TYPE_IPV4); 365 info->l2_len = 0; 366 367 } else if (gre_hdr->proto == _htons(RTE_ETHER_TYPE_IPV6)) { 368 info->is_tunnel = 1; 369 info->outer_ethertype = info->ethertype; 370 info->outer_l2_len = info->l2_len; 371 info->outer_l3_len = info->l3_len; 372 info->outer_l4_proto = info->l4_proto; 373 374 ipv6_hdr = (struct rte_ipv6_hdr *)((char *)gre_hdr + gre_len); 375 376 info->ethertype = _htons(RTE_ETHER_TYPE_IPV6); 377 parse_ipv6(ipv6_hdr, info); 378 info->l2_len = 0; 379 380 } else if (gre_hdr->proto == _htons(RTE_ETHER_TYPE_TEB)) { 381 info->is_tunnel = 1; 382 info->outer_ethertype = info->ethertype; 383 info->outer_l2_len = info->l2_len; 384 info->outer_l3_len = info->l3_len; 385 info->outer_l4_proto = info->l4_proto; 386 387 eth_hdr = (struct rte_ether_hdr *)((char *)gre_hdr + gre_len); 388 389 parse_ethernet(eth_hdr, info); 390 } else 391 return; 392 393 info->l2_len += gre_len; 394 } 395 396 397 /* Parse an encapsulated ip or ipv6 header */ 398 static void 399 parse_encap_ip(void *encap_ip, struct testpmd_offload_info *info) 400 { 401 struct rte_ipv4_hdr *ipv4_hdr = encap_ip; 402 struct rte_ipv6_hdr *ipv6_hdr = encap_ip; 403 uint8_t ip_version; 404 405 ip_version = (ipv4_hdr->version_ihl & 0xf0) >> 4; 406 407 if (ip_version != 4 && ip_version != 6) 408 return; 409 410 info->is_tunnel = 1; 411 info->outer_ethertype = info->ethertype; 412 info->outer_l2_len = info->l2_len; 413 info->outer_l3_len = info->l3_len; 414 415 if (ip_version == 4) { 416 parse_ipv4(ipv4_hdr, info); 417 info->ethertype = _htons(RTE_ETHER_TYPE_IPV4); 418 } else { 419 parse_ipv6(ipv6_hdr, info); 420 info->ethertype = _htons(RTE_ETHER_TYPE_IPV6); 421 } 422 info->l2_len = 0; 423 } 424 425 /* if possible, calculate the checksum of a packet in hw or sw, 426 * depending on the testpmd command line configuration */ 427 static uint64_t 428 process_inner_cksums(void *l3_hdr, const struct testpmd_offload_info *info, 429 uint64_t tx_offloads) 430 { 431 struct rte_ipv4_hdr *ipv4_hdr = l3_hdr; 432 struct rte_udp_hdr *udp_hdr; 433 struct rte_tcp_hdr *tcp_hdr; 434 struct rte_sctp_hdr *sctp_hdr; 435 uint64_t ol_flags = 0; 436 uint32_t max_pkt_len, tso_segsz = 0; 437 438 /* ensure packet is large enough to require tso */ 439 if (!info->is_tunnel) { 440 max_pkt_len = info->l2_len + info->l3_len + info->l4_len + 441 info->tso_segsz; 442 if (info->tso_segsz != 0 && info->pkt_len > max_pkt_len) 443 tso_segsz = info->tso_segsz; 444 } else { 445 max_pkt_len = info->outer_l2_len + info->outer_l3_len + 446 info->l2_len + info->l3_len + info->l4_len + 447 info->tunnel_tso_segsz; 448 if (info->tunnel_tso_segsz != 0 && info->pkt_len > max_pkt_len) 449 tso_segsz = info->tunnel_tso_segsz; 450 } 451 452 if (info->ethertype == _htons(RTE_ETHER_TYPE_IPV4)) { 453 ipv4_hdr = l3_hdr; 454 ipv4_hdr->hdr_checksum = 0; 455 456 ol_flags |= PKT_TX_IPV4; 457 if (info->l4_proto == IPPROTO_TCP && tso_segsz) { 458 ol_flags |= PKT_TX_IP_CKSUM; 459 } else { 460 if (tx_offloads & DEV_TX_OFFLOAD_IPV4_CKSUM) 461 ol_flags |= PKT_TX_IP_CKSUM; 462 else 463 ipv4_hdr->hdr_checksum = 464 rte_ipv4_cksum(ipv4_hdr); 465 } 466 } else if (info->ethertype == _htons(RTE_ETHER_TYPE_IPV6)) 467 ol_flags |= PKT_TX_IPV6; 468 else 469 return 0; /* packet type not supported, nothing to do */ 470 471 if (info->l4_proto == IPPROTO_UDP) { 472 udp_hdr = (struct rte_udp_hdr *)((char *)l3_hdr + info->l3_len); 473 /* do not recalculate udp cksum if it was 0 */ 474 if (udp_hdr->dgram_cksum != 0) { 475 udp_hdr->dgram_cksum = 0; 476 if (tx_offloads & DEV_TX_OFFLOAD_UDP_CKSUM) 477 ol_flags |= PKT_TX_UDP_CKSUM; 478 else { 479 udp_hdr->dgram_cksum = 480 get_udptcp_checksum(l3_hdr, udp_hdr, 481 info->ethertype); 482 } 483 } 484 if (info->gso_enable) 485 ol_flags |= PKT_TX_UDP_SEG; 486 } else if (info->l4_proto == IPPROTO_TCP) { 487 tcp_hdr = (struct rte_tcp_hdr *)((char *)l3_hdr + info->l3_len); 488 tcp_hdr->cksum = 0; 489 if (tso_segsz) 490 ol_flags |= PKT_TX_TCP_SEG; 491 else if (tx_offloads & DEV_TX_OFFLOAD_TCP_CKSUM) 492 ol_flags |= PKT_TX_TCP_CKSUM; 493 else { 494 tcp_hdr->cksum = 495 get_udptcp_checksum(l3_hdr, tcp_hdr, 496 info->ethertype); 497 } 498 if (info->gso_enable) 499 ol_flags |= PKT_TX_TCP_SEG; 500 } else if (info->l4_proto == IPPROTO_SCTP) { 501 sctp_hdr = (struct rte_sctp_hdr *) 502 ((char *)l3_hdr + info->l3_len); 503 sctp_hdr->cksum = 0; 504 /* sctp payload must be a multiple of 4 to be 505 * offloaded */ 506 if ((tx_offloads & DEV_TX_OFFLOAD_SCTP_CKSUM) && 507 ((ipv4_hdr->total_length & 0x3) == 0)) { 508 ol_flags |= PKT_TX_SCTP_CKSUM; 509 } else { 510 /* XXX implement CRC32c, example available in 511 * RFC3309 */ 512 } 513 } 514 515 return ol_flags; 516 } 517 518 /* Calculate the checksum of outer header */ 519 static uint64_t 520 process_outer_cksums(void *outer_l3_hdr, struct testpmd_offload_info *info, 521 uint64_t tx_offloads, int tso_enabled) 522 { 523 struct rte_ipv4_hdr *ipv4_hdr = outer_l3_hdr; 524 struct rte_ipv6_hdr *ipv6_hdr = outer_l3_hdr; 525 struct rte_udp_hdr *udp_hdr; 526 uint64_t ol_flags = 0; 527 528 if (info->outer_ethertype == _htons(RTE_ETHER_TYPE_IPV4)) { 529 ipv4_hdr->hdr_checksum = 0; 530 ol_flags |= PKT_TX_OUTER_IPV4; 531 532 if (tx_offloads & DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM) 533 ol_flags |= PKT_TX_OUTER_IP_CKSUM; 534 else 535 ipv4_hdr->hdr_checksum = rte_ipv4_cksum(ipv4_hdr); 536 } else 537 ol_flags |= PKT_TX_OUTER_IPV6; 538 539 if (info->outer_l4_proto != IPPROTO_UDP) 540 return ol_flags; 541 542 udp_hdr = (struct rte_udp_hdr *) 543 ((char *)outer_l3_hdr + info->outer_l3_len); 544 545 if (tso_enabled) 546 ol_flags |= PKT_TX_TCP_SEG; 547 548 /* Skip SW outer UDP checksum generation if HW supports it */ 549 if (tx_offloads & DEV_TX_OFFLOAD_OUTER_UDP_CKSUM) { 550 if (info->outer_ethertype == _htons(RTE_ETHER_TYPE_IPV4)) 551 udp_hdr->dgram_cksum 552 = rte_ipv4_phdr_cksum(ipv4_hdr, ol_flags); 553 else 554 udp_hdr->dgram_cksum 555 = rte_ipv6_phdr_cksum(ipv6_hdr, ol_flags); 556 557 ol_flags |= PKT_TX_OUTER_UDP_CKSUM; 558 return ol_flags; 559 } 560 561 /* outer UDP checksum is done in software. In the other side, for 562 * UDP tunneling, like VXLAN or Geneve, outer UDP checksum can be 563 * set to zero. 564 * 565 * If a packet will be TSOed into small packets by NIC, we cannot 566 * set/calculate a non-zero checksum, because it will be a wrong 567 * value after the packet be split into several small packets. 568 */ 569 if (tso_enabled) 570 udp_hdr->dgram_cksum = 0; 571 572 /* do not recalculate udp cksum if it was 0 */ 573 if (udp_hdr->dgram_cksum != 0) { 574 udp_hdr->dgram_cksum = 0; 575 if (info->outer_ethertype == _htons(RTE_ETHER_TYPE_IPV4)) 576 udp_hdr->dgram_cksum = 577 rte_ipv4_udptcp_cksum(ipv4_hdr, udp_hdr); 578 else 579 udp_hdr->dgram_cksum = 580 rte_ipv6_udptcp_cksum(ipv6_hdr, udp_hdr); 581 } 582 583 return ol_flags; 584 } 585 586 /* 587 * Helper function. 588 * Performs actual copying. 589 * Returns number of segments in the destination mbuf on success, 590 * or negative error code on failure. 591 */ 592 static int 593 mbuf_copy_split(const struct rte_mbuf *ms, struct rte_mbuf *md[], 594 uint16_t seglen[], uint8_t nb_seg) 595 { 596 uint32_t dlen, slen, tlen; 597 uint32_t i, len; 598 const struct rte_mbuf *m; 599 const uint8_t *src; 600 uint8_t *dst; 601 602 dlen = 0; 603 slen = 0; 604 tlen = 0; 605 606 dst = NULL; 607 src = NULL; 608 609 m = ms; 610 i = 0; 611 while (ms != NULL && i != nb_seg) { 612 613 if (slen == 0) { 614 slen = rte_pktmbuf_data_len(ms); 615 src = rte_pktmbuf_mtod(ms, const uint8_t *); 616 } 617 618 if (dlen == 0) { 619 dlen = RTE_MIN(seglen[i], slen); 620 md[i]->data_len = dlen; 621 md[i]->next = (i + 1 == nb_seg) ? NULL : md[i + 1]; 622 dst = rte_pktmbuf_mtod(md[i], uint8_t *); 623 } 624 625 len = RTE_MIN(slen, dlen); 626 memcpy(dst, src, len); 627 tlen += len; 628 slen -= len; 629 dlen -= len; 630 src += len; 631 dst += len; 632 633 if (slen == 0) 634 ms = ms->next; 635 if (dlen == 0) 636 i++; 637 } 638 639 if (ms != NULL) 640 return -ENOBUFS; 641 else if (tlen != m->pkt_len) 642 return -EINVAL; 643 644 md[0]->nb_segs = nb_seg; 645 md[0]->pkt_len = tlen; 646 md[0]->vlan_tci = m->vlan_tci; 647 md[0]->vlan_tci_outer = m->vlan_tci_outer; 648 md[0]->ol_flags = m->ol_flags; 649 md[0]->tx_offload = m->tx_offload; 650 651 return nb_seg; 652 } 653 654 /* 655 * Allocate a new mbuf with up to tx_pkt_nb_segs segments. 656 * Copy packet contents and offload information into the new segmented mbuf. 657 */ 658 static struct rte_mbuf * 659 pkt_copy_split(const struct rte_mbuf *pkt) 660 { 661 int32_t n, rc; 662 uint32_t i, len, nb_seg; 663 struct rte_mempool *mp; 664 uint16_t seglen[RTE_MAX_SEGS_PER_PKT]; 665 struct rte_mbuf *p, *md[RTE_MAX_SEGS_PER_PKT]; 666 667 mp = current_fwd_lcore()->mbp; 668 669 if (tx_pkt_split == TX_PKT_SPLIT_RND) 670 nb_seg = random() % tx_pkt_nb_segs + 1; 671 else 672 nb_seg = tx_pkt_nb_segs; 673 674 memcpy(seglen, tx_pkt_seg_lengths, nb_seg * sizeof(seglen[0])); 675 676 /* calculate number of segments to use and their length. */ 677 len = 0; 678 for (i = 0; i != nb_seg && len < pkt->pkt_len; i++) { 679 len += seglen[i]; 680 md[i] = NULL; 681 } 682 683 n = pkt->pkt_len - len; 684 685 /* update size of the last segment to fit rest of the packet */ 686 if (n >= 0) { 687 seglen[i - 1] += n; 688 len += n; 689 } 690 691 nb_seg = i; 692 while (i != 0) { 693 p = rte_pktmbuf_alloc(mp); 694 if (p == NULL) { 695 TESTPMD_LOG(ERR, 696 "failed to allocate %u-th of %u mbuf " 697 "from mempool: %s\n", 698 nb_seg - i, nb_seg, mp->name); 699 break; 700 } 701 702 md[--i] = p; 703 if (rte_pktmbuf_tailroom(md[i]) < seglen[i]) { 704 TESTPMD_LOG(ERR, "mempool %s, %u-th segment: " 705 "expected seglen: %u, " 706 "actual mbuf tailroom: %u\n", 707 mp->name, i, seglen[i], 708 rte_pktmbuf_tailroom(md[i])); 709 break; 710 } 711 } 712 713 /* all mbufs successfully allocated, do copy */ 714 if (i == 0) { 715 rc = mbuf_copy_split(pkt, md, seglen, nb_seg); 716 if (rc < 0) 717 TESTPMD_LOG(ERR, 718 "mbuf_copy_split for %p(len=%u, nb_seg=%u) " 719 "into %u segments failed with error code: %d\n", 720 pkt, pkt->pkt_len, pkt->nb_segs, nb_seg, rc); 721 722 /* figure out how many mbufs to free. */ 723 i = RTE_MAX(rc, 0); 724 } 725 726 /* free unused mbufs */ 727 for (; i != nb_seg; i++) { 728 rte_pktmbuf_free_seg(md[i]); 729 md[i] = NULL; 730 } 731 732 return md[0]; 733 } 734 735 /* 736 * Receive a burst of packets, and for each packet: 737 * - parse packet, and try to recognize a supported packet type (1) 738 * - if it's not a supported packet type, don't touch the packet, else: 739 * - reprocess the checksum of all supported layers. This is done in SW 740 * or HW, depending on testpmd command line configuration 741 * - if TSO is enabled in testpmd command line, also flag the mbuf for TCP 742 * segmentation offload (this implies HW TCP checksum) 743 * Then transmit packets on the output port. 744 * 745 * (1) Supported packets are: 746 * Ether / (vlan) / IP|IP6 / UDP|TCP|SCTP . 747 * Ether / (vlan) / outer IP|IP6 / outer UDP / VxLAN / Ether / IP|IP6 / 748 * UDP|TCP|SCTP 749 * Ether / (vlan) / outer IP|IP6 / outer UDP / VXLAN-GPE / Ether / IP|IP6 / 750 * UDP|TCP|SCTP 751 * Ether / (vlan) / outer IP|IP6 / outer UDP / VXLAN-GPE / IP|IP6 / 752 * UDP|TCP|SCTP 753 * Ether / (vlan) / outer IP / outer UDP / GTP / IP|IP6 / UDP|TCP|SCTP 754 * Ether / (vlan) / outer IP|IP6 / GRE / Ether / IP|IP6 / UDP|TCP|SCTP 755 * Ether / (vlan) / outer IP|IP6 / GRE / IP|IP6 / UDP|TCP|SCTP 756 * Ether / (vlan) / outer IP|IP6 / IP|IP6 / UDP|TCP|SCTP 757 * 758 * The testpmd command line for this forward engine sets the flags 759 * TESTPMD_TX_OFFLOAD_* in ports[tx_port].tx_ol_flags. They control 760 * wether a checksum must be calculated in software or in hardware. The 761 * IP, UDP, TCP and SCTP flags always concern the inner layer. The 762 * OUTER_IP is only useful for tunnel packets. 763 */ 764 static void 765 pkt_burst_checksum_forward(struct fwd_stream *fs) 766 { 767 struct rte_mbuf *pkts_burst[MAX_PKT_BURST]; 768 struct rte_mbuf *gso_segments[GSO_MAX_PKT_BURST]; 769 struct rte_gso_ctx *gso_ctx; 770 struct rte_mbuf **tx_pkts_burst; 771 struct rte_port *txp; 772 struct rte_mbuf *m, *p; 773 struct rte_ether_hdr *eth_hdr; 774 void *l3_hdr = NULL, *outer_l3_hdr = NULL; /* can be IPv4 or IPv6 */ 775 void **gro_ctx; 776 uint16_t gro_pkts_num; 777 uint8_t gro_enable; 778 uint16_t nb_rx; 779 uint16_t nb_tx; 780 uint16_t nb_prep; 781 uint16_t i; 782 uint64_t rx_ol_flags, tx_ol_flags; 783 uint64_t tx_offloads; 784 uint32_t retry; 785 uint32_t rx_bad_ip_csum; 786 uint32_t rx_bad_l4_csum; 787 uint32_t rx_bad_outer_l4_csum; 788 struct testpmd_offload_info info; 789 uint16_t nb_segments = 0; 790 int ret; 791 792 uint64_t start_tsc = 0; 793 794 get_start_cycles(&start_tsc); 795 796 /* receive a burst of packet */ 797 nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue, pkts_burst, 798 nb_pkt_per_burst); 799 inc_rx_burst_stats(fs, nb_rx); 800 if (unlikely(nb_rx == 0)) 801 return; 802 803 fs->rx_packets += nb_rx; 804 rx_bad_ip_csum = 0; 805 rx_bad_l4_csum = 0; 806 rx_bad_outer_l4_csum = 0; 807 gro_enable = gro_ports[fs->rx_port].enable; 808 809 txp = &ports[fs->tx_port]; 810 tx_offloads = txp->dev_conf.txmode.offloads; 811 memset(&info, 0, sizeof(info)); 812 info.tso_segsz = txp->tso_segsz; 813 info.tunnel_tso_segsz = txp->tunnel_tso_segsz; 814 if (gso_ports[fs->tx_port].enable) 815 info.gso_enable = 1; 816 817 for (i = 0; i < nb_rx; i++) { 818 if (likely(i < nb_rx - 1)) 819 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[i + 1], 820 void *)); 821 822 m = pkts_burst[i]; 823 info.is_tunnel = 0; 824 info.pkt_len = rte_pktmbuf_pkt_len(m); 825 tx_ol_flags = m->ol_flags & 826 (IND_ATTACHED_MBUF | EXT_ATTACHED_MBUF); 827 rx_ol_flags = m->ol_flags; 828 829 /* Update the L3/L4 checksum error packet statistics */ 830 if ((rx_ol_flags & PKT_RX_IP_CKSUM_MASK) == PKT_RX_IP_CKSUM_BAD) 831 rx_bad_ip_csum += 1; 832 if ((rx_ol_flags & PKT_RX_L4_CKSUM_MASK) == PKT_RX_L4_CKSUM_BAD) 833 rx_bad_l4_csum += 1; 834 if (rx_ol_flags & PKT_RX_OUTER_L4_CKSUM_BAD) 835 rx_bad_outer_l4_csum += 1; 836 837 /* step 1: dissect packet, parsing optional vlan, ip4/ip6, vxlan 838 * and inner headers */ 839 840 eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *); 841 rte_ether_addr_copy(&peer_eth_addrs[fs->peer_addr], 842 ð_hdr->d_addr); 843 rte_ether_addr_copy(&ports[fs->tx_port].eth_addr, 844 ð_hdr->s_addr); 845 parse_ethernet(eth_hdr, &info); 846 l3_hdr = (char *)eth_hdr + info.l2_len; 847 848 /* check if it's a supported tunnel */ 849 if (txp->parse_tunnel) { 850 if (info.l4_proto == IPPROTO_UDP) { 851 struct rte_udp_hdr *udp_hdr; 852 853 udp_hdr = (struct rte_udp_hdr *) 854 ((char *)l3_hdr + info.l3_len); 855 parse_gtp(udp_hdr, &info); 856 if (info.is_tunnel) { 857 tx_ol_flags |= PKT_TX_TUNNEL_GTP; 858 goto tunnel_update; 859 } 860 parse_vxlan_gpe(udp_hdr, &info); 861 if (info.is_tunnel) { 862 tx_ol_flags |= 863 PKT_TX_TUNNEL_VXLAN_GPE; 864 goto tunnel_update; 865 } 866 parse_vxlan(udp_hdr, &info, 867 m->packet_type); 868 if (info.is_tunnel) 869 tx_ol_flags |= 870 PKT_TX_TUNNEL_VXLAN; 871 } else if (info.l4_proto == IPPROTO_GRE) { 872 struct simple_gre_hdr *gre_hdr; 873 874 gre_hdr = (struct simple_gre_hdr *) 875 ((char *)l3_hdr + info.l3_len); 876 parse_gre(gre_hdr, &info); 877 if (info.is_tunnel) 878 tx_ol_flags |= PKT_TX_TUNNEL_GRE; 879 } else if (info.l4_proto == IPPROTO_IPIP) { 880 void *encap_ip_hdr; 881 882 encap_ip_hdr = (char *)l3_hdr + info.l3_len; 883 parse_encap_ip(encap_ip_hdr, &info); 884 if (info.is_tunnel) 885 tx_ol_flags |= PKT_TX_TUNNEL_IPIP; 886 } 887 } 888 889 tunnel_update: 890 /* update l3_hdr and outer_l3_hdr if a tunnel was parsed */ 891 if (info.is_tunnel) { 892 outer_l3_hdr = l3_hdr; 893 l3_hdr = (char *)l3_hdr + info.outer_l3_len + info.l2_len; 894 } 895 896 /* step 2: depending on user command line configuration, 897 * recompute checksum either in software or flag the 898 * mbuf to offload the calculation to the NIC. If TSO 899 * is configured, prepare the mbuf for TCP segmentation. */ 900 901 /* process checksums of inner headers first */ 902 tx_ol_flags |= process_inner_cksums(l3_hdr, &info, 903 tx_offloads); 904 905 /* Then process outer headers if any. Note that the software 906 * checksum will be wrong if one of the inner checksums is 907 * processed in hardware. */ 908 if (info.is_tunnel == 1) { 909 tx_ol_flags |= process_outer_cksums(outer_l3_hdr, &info, 910 tx_offloads, 911 !!(tx_ol_flags & PKT_TX_TCP_SEG)); 912 } 913 914 /* step 3: fill the mbuf meta data (flags and header lengths) */ 915 916 m->tx_offload = 0; 917 if (info.is_tunnel == 1) { 918 if (info.tunnel_tso_segsz || 919 (tx_offloads & 920 DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM) || 921 (tx_offloads & 922 DEV_TX_OFFLOAD_OUTER_UDP_CKSUM) || 923 (tx_ol_flags & PKT_TX_OUTER_IPV6)) { 924 m->outer_l2_len = info.outer_l2_len; 925 m->outer_l3_len = info.outer_l3_len; 926 m->l2_len = info.l2_len; 927 m->l3_len = info.l3_len; 928 m->l4_len = info.l4_len; 929 m->tso_segsz = info.tunnel_tso_segsz; 930 } 931 else { 932 /* if there is a outer UDP cksum 933 processed in sw and the inner in hw, 934 the outer checksum will be wrong as 935 the payload will be modified by the 936 hardware */ 937 m->l2_len = info.outer_l2_len + 938 info.outer_l3_len + info.l2_len; 939 m->l3_len = info.l3_len; 940 m->l4_len = info.l4_len; 941 } 942 } else { 943 /* this is only useful if an offload flag is 944 * set, but it does not hurt to fill it in any 945 * case */ 946 m->l2_len = info.l2_len; 947 m->l3_len = info.l3_len; 948 m->l4_len = info.l4_len; 949 m->tso_segsz = info.tso_segsz; 950 } 951 m->ol_flags = tx_ol_flags; 952 953 /* Do split & copy for the packet. */ 954 if (tx_pkt_split != TX_PKT_SPLIT_OFF) { 955 p = pkt_copy_split(m); 956 if (p != NULL) { 957 rte_pktmbuf_free(m); 958 m = p; 959 pkts_burst[i] = m; 960 } 961 } 962 963 /* if verbose mode is enabled, dump debug info */ 964 if (verbose_level > 0) { 965 char buf[256]; 966 967 printf("-----------------\n"); 968 printf("port=%u, mbuf=%p, pkt_len=%u, nb_segs=%u:\n", 969 fs->rx_port, m, m->pkt_len, m->nb_segs); 970 /* dump rx parsed packet info */ 971 rte_get_rx_ol_flag_list(rx_ol_flags, buf, sizeof(buf)); 972 printf("rx: l2_len=%d ethertype=%x l3_len=%d " 973 "l4_proto=%d l4_len=%d flags=%s\n", 974 info.l2_len, rte_be_to_cpu_16(info.ethertype), 975 info.l3_len, info.l4_proto, info.l4_len, buf); 976 if (rx_ol_flags & PKT_RX_LRO) 977 printf("rx: m->lro_segsz=%u\n", m->tso_segsz); 978 if (info.is_tunnel == 1) 979 printf("rx: outer_l2_len=%d outer_ethertype=%x " 980 "outer_l3_len=%d\n", info.outer_l2_len, 981 rte_be_to_cpu_16(info.outer_ethertype), 982 info.outer_l3_len); 983 /* dump tx packet info */ 984 if ((tx_offloads & (DEV_TX_OFFLOAD_IPV4_CKSUM | 985 DEV_TX_OFFLOAD_UDP_CKSUM | 986 DEV_TX_OFFLOAD_TCP_CKSUM | 987 DEV_TX_OFFLOAD_SCTP_CKSUM)) || 988 info.tso_segsz != 0) 989 printf("tx: m->l2_len=%d m->l3_len=%d " 990 "m->l4_len=%d\n", 991 m->l2_len, m->l3_len, m->l4_len); 992 if (info.is_tunnel == 1) { 993 if ((tx_offloads & 994 DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM) || 995 (tx_offloads & 996 DEV_TX_OFFLOAD_OUTER_UDP_CKSUM) || 997 (tx_ol_flags & PKT_TX_OUTER_IPV6)) 998 printf("tx: m->outer_l2_len=%d " 999 "m->outer_l3_len=%d\n", 1000 m->outer_l2_len, 1001 m->outer_l3_len); 1002 if (info.tunnel_tso_segsz != 0 && 1003 (m->ol_flags & PKT_TX_TCP_SEG)) 1004 printf("tx: m->tso_segsz=%d\n", 1005 m->tso_segsz); 1006 } else if (info.tso_segsz != 0 && 1007 (m->ol_flags & PKT_TX_TCP_SEG)) 1008 printf("tx: m->tso_segsz=%d\n", m->tso_segsz); 1009 rte_get_tx_ol_flag_list(m->ol_flags, buf, sizeof(buf)); 1010 printf("tx: flags=%s", buf); 1011 printf("\n"); 1012 } 1013 } 1014 1015 if (unlikely(gro_enable)) { 1016 if (gro_flush_cycles == GRO_DEFAULT_FLUSH_CYCLES) { 1017 nb_rx = rte_gro_reassemble_burst(pkts_burst, nb_rx, 1018 &(gro_ports[fs->rx_port].param)); 1019 } else { 1020 gro_ctx = current_fwd_lcore()->gro_ctx; 1021 nb_rx = rte_gro_reassemble(pkts_burst, nb_rx, gro_ctx); 1022 1023 if (++fs->gro_times >= gro_flush_cycles) { 1024 gro_pkts_num = rte_gro_get_pkt_count(gro_ctx); 1025 if (gro_pkts_num > MAX_PKT_BURST - nb_rx) 1026 gro_pkts_num = MAX_PKT_BURST - nb_rx; 1027 1028 nb_rx += rte_gro_timeout_flush(gro_ctx, 0, 1029 RTE_GRO_TCP_IPV4, 1030 &pkts_burst[nb_rx], 1031 gro_pkts_num); 1032 fs->gro_times = 0; 1033 } 1034 } 1035 } 1036 1037 if (gso_ports[fs->tx_port].enable == 0) 1038 tx_pkts_burst = pkts_burst; 1039 else { 1040 gso_ctx = &(current_fwd_lcore()->gso_ctx); 1041 gso_ctx->gso_size = gso_max_segment_size; 1042 for (i = 0; i < nb_rx; i++) { 1043 ret = rte_gso_segment(pkts_burst[i], gso_ctx, 1044 &gso_segments[nb_segments], 1045 GSO_MAX_PKT_BURST - nb_segments); 1046 if (ret >= 0) 1047 nb_segments += ret; 1048 else { 1049 TESTPMD_LOG(DEBUG, "Unable to segment packet"); 1050 rte_pktmbuf_free(pkts_burst[i]); 1051 } 1052 } 1053 1054 tx_pkts_burst = gso_segments; 1055 nb_rx = nb_segments; 1056 } 1057 1058 nb_prep = rte_eth_tx_prepare(fs->tx_port, fs->tx_queue, 1059 tx_pkts_burst, nb_rx); 1060 if (nb_prep != nb_rx) 1061 printf("Preparing packet burst to transmit failed: %s\n", 1062 rte_strerror(rte_errno)); 1063 1064 nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue, tx_pkts_burst, 1065 nb_prep); 1066 1067 /* 1068 * Retry if necessary 1069 */ 1070 if (unlikely(nb_tx < nb_rx) && fs->retry_enabled) { 1071 retry = 0; 1072 while (nb_tx < nb_rx && retry++ < burst_tx_retry_num) { 1073 rte_delay_us(burst_tx_delay_time); 1074 nb_tx += rte_eth_tx_burst(fs->tx_port, fs->tx_queue, 1075 &tx_pkts_burst[nb_tx], nb_rx - nb_tx); 1076 } 1077 } 1078 fs->tx_packets += nb_tx; 1079 fs->rx_bad_ip_csum += rx_bad_ip_csum; 1080 fs->rx_bad_l4_csum += rx_bad_l4_csum; 1081 fs->rx_bad_outer_l4_csum += rx_bad_outer_l4_csum; 1082 1083 inc_tx_burst_stats(fs, nb_tx); 1084 if (unlikely(nb_tx < nb_rx)) { 1085 fs->fwd_dropped += (nb_rx - nb_tx); 1086 do { 1087 rte_pktmbuf_free(tx_pkts_burst[nb_tx]); 1088 } while (++nb_tx < nb_rx); 1089 } 1090 1091 get_end_cycles(fs, start_tsc); 1092 } 1093 1094 struct fwd_engine csum_fwd_engine = { 1095 .fwd_mode_name = "csum", 1096 .port_fwd_begin = NULL, 1097 .port_fwd_end = NULL, 1098 .packet_fwd = pkt_burst_checksum_forward, 1099 }; 1100