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_sctp.h> 39 #include <rte_prefetch.h> 40 #include <rte_string_fns.h> 41 #include <rte_flow.h> 42 #include <rte_gro.h> 43 #include <rte_gso.h> 44 45 #include "testpmd.h" 46 47 #define IP_DEFTTL 64 /* from RFC 1340. */ 48 #define IP_VERSION 0x40 49 #define IP_HDRLEN 0x05 /* default IP header length == five 32-bits words. */ 50 #define IP_VHL_DEF (IP_VERSION | IP_HDRLEN) 51 52 #define GRE_CHECKSUM_PRESENT 0x8000 53 #define GRE_KEY_PRESENT 0x2000 54 #define GRE_SEQUENCE_PRESENT 0x1000 55 #define GRE_EXT_LEN 4 56 #define GRE_SUPPORTED_FIELDS (GRE_CHECKSUM_PRESENT | GRE_KEY_PRESENT |\ 57 GRE_SEQUENCE_PRESENT) 58 59 /* We cannot use rte_cpu_to_be_16() on a constant in a switch/case */ 60 #if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN 61 #define _htons(x) ((uint16_t)((((x) & 0x00ffU) << 8) | (((x) & 0xff00U) >> 8))) 62 #else 63 #define _htons(x) (x) 64 #endif 65 66 uint16_t vxlan_gpe_udp_port = 4790; 67 68 /* structure that caches offload info for the current packet */ 69 struct testpmd_offload_info { 70 uint16_t ethertype; 71 uint8_t gso_enable; 72 uint16_t l2_len; 73 uint16_t l3_len; 74 uint16_t l4_len; 75 uint8_t l4_proto; 76 uint8_t is_tunnel; 77 uint16_t outer_ethertype; 78 uint16_t outer_l2_len; 79 uint16_t outer_l3_len; 80 uint8_t outer_l4_proto; 81 uint16_t tso_segsz; 82 uint16_t tunnel_tso_segsz; 83 uint32_t pkt_len; 84 }; 85 86 /* simplified GRE header */ 87 struct simple_gre_hdr { 88 uint16_t flags; 89 uint16_t proto; 90 } __attribute__((__packed__)); 91 92 static uint16_t 93 get_udptcp_checksum(void *l3_hdr, void *l4_hdr, uint16_t ethertype) 94 { 95 if (ethertype == _htons(ETHER_TYPE_IPv4)) 96 return rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr); 97 else /* assume ethertype == ETHER_TYPE_IPv6 */ 98 return rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr); 99 } 100 101 /* Parse an IPv4 header to fill l3_len, l4_len, and l4_proto */ 102 static void 103 parse_ipv4(struct ipv4_hdr *ipv4_hdr, struct testpmd_offload_info *info) 104 { 105 struct tcp_hdr *tcp_hdr; 106 107 info->l3_len = (ipv4_hdr->version_ihl & 0x0f) * 4; 108 info->l4_proto = ipv4_hdr->next_proto_id; 109 110 /* only fill l4_len for TCP, it's useful for TSO */ 111 if (info->l4_proto == IPPROTO_TCP) { 112 tcp_hdr = (struct tcp_hdr *)((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 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 ipv6_hdr *ipv6_hdr, struct testpmd_offload_info *info) 123 { 124 struct tcp_hdr *tcp_hdr; 125 126 info->l3_len = sizeof(struct 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 tcp_hdr *)((char *)ipv6_hdr + info->l3_len); 132 info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2; 133 } else if (info->l4_proto == IPPROTO_UDP) 134 info->l4_len = sizeof(struct udp_hdr); 135 else 136 info->l4_len = 0; 137 } 138 139 /* 140 * Parse an ethernet header to fill the ethertype, l2_len, l3_len and 141 * ipproto. This function is able to recognize IPv4/IPv6 with one optional vlan 142 * header. The l4_len argument is only set in case of TCP (useful for TSO). 143 */ 144 static void 145 parse_ethernet(struct ether_hdr *eth_hdr, struct testpmd_offload_info *info) 146 { 147 struct ipv4_hdr *ipv4_hdr; 148 struct ipv6_hdr *ipv6_hdr; 149 150 info->l2_len = sizeof(struct ether_hdr); 151 info->ethertype = eth_hdr->ether_type; 152 153 if (info->ethertype == _htons(ETHER_TYPE_VLAN)) { 154 struct vlan_hdr *vlan_hdr = (struct vlan_hdr *)(eth_hdr + 1); 155 156 info->l2_len += sizeof(struct vlan_hdr); 157 info->ethertype = vlan_hdr->eth_proto; 158 } 159 160 switch (info->ethertype) { 161 case _htons(ETHER_TYPE_IPv4): 162 ipv4_hdr = (struct ipv4_hdr *) ((char *)eth_hdr + info->l2_len); 163 parse_ipv4(ipv4_hdr, info); 164 break; 165 case _htons(ETHER_TYPE_IPv6): 166 ipv6_hdr = (struct ipv6_hdr *) ((char *)eth_hdr + info->l2_len); 167 parse_ipv6(ipv6_hdr, info); 168 break; 169 default: 170 info->l4_len = 0; 171 info->l3_len = 0; 172 info->l4_proto = 0; 173 break; 174 } 175 } 176 177 /* Parse a vxlan header */ 178 static void 179 parse_vxlan(struct udp_hdr *udp_hdr, 180 struct testpmd_offload_info *info, 181 uint32_t pkt_type) 182 { 183 struct ether_hdr *eth_hdr; 184 185 /* check udp destination port, 4789 is the default vxlan port 186 * (rfc7348) or that the rx offload flag is set (i40e only 187 * currently) */ 188 if (udp_hdr->dst_port != _htons(4789) && 189 RTE_ETH_IS_TUNNEL_PKT(pkt_type) == 0) 190 return; 191 192 info->is_tunnel = 1; 193 info->outer_ethertype = info->ethertype; 194 info->outer_l2_len = info->l2_len; 195 info->outer_l3_len = info->l3_len; 196 info->outer_l4_proto = info->l4_proto; 197 198 eth_hdr = (struct ether_hdr *)((char *)udp_hdr + 199 sizeof(struct udp_hdr) + 200 sizeof(struct vxlan_hdr)); 201 202 parse_ethernet(eth_hdr, info); 203 info->l2_len += ETHER_VXLAN_HLEN; /* add udp + vxlan */ 204 } 205 206 /* Parse a vxlan-gpe header */ 207 static void 208 parse_vxlan_gpe(struct udp_hdr *udp_hdr, 209 struct testpmd_offload_info *info) 210 { 211 struct ether_hdr *eth_hdr; 212 struct ipv4_hdr *ipv4_hdr; 213 struct ipv6_hdr *ipv6_hdr; 214 struct vxlan_gpe_hdr *vxlan_gpe_hdr; 215 uint8_t vxlan_gpe_len = sizeof(*vxlan_gpe_hdr); 216 217 /* Check udp destination port. */ 218 if (udp_hdr->dst_port != _htons(vxlan_gpe_udp_port)) 219 return; 220 221 vxlan_gpe_hdr = (struct vxlan_gpe_hdr *)((char *)udp_hdr + 222 sizeof(struct udp_hdr)); 223 224 if (!vxlan_gpe_hdr->proto || vxlan_gpe_hdr->proto == 225 VXLAN_GPE_TYPE_IPV4) { 226 info->is_tunnel = 1; 227 info->outer_ethertype = info->ethertype; 228 info->outer_l2_len = info->l2_len; 229 info->outer_l3_len = info->l3_len; 230 info->outer_l4_proto = info->l4_proto; 231 232 ipv4_hdr = (struct ipv4_hdr *)((char *)vxlan_gpe_hdr + 233 vxlan_gpe_len); 234 235 parse_ipv4(ipv4_hdr, info); 236 info->ethertype = _htons(ETHER_TYPE_IPv4); 237 info->l2_len = 0; 238 239 } else if (vxlan_gpe_hdr->proto == VXLAN_GPE_TYPE_IPV6) { 240 info->is_tunnel = 1; 241 info->outer_ethertype = info->ethertype; 242 info->outer_l2_len = info->l2_len; 243 info->outer_l3_len = info->l3_len; 244 info->outer_l4_proto = info->l4_proto; 245 246 ipv6_hdr = (struct ipv6_hdr *)((char *)vxlan_gpe_hdr + 247 vxlan_gpe_len); 248 249 info->ethertype = _htons(ETHER_TYPE_IPv6); 250 parse_ipv6(ipv6_hdr, info); 251 info->l2_len = 0; 252 253 } else if (vxlan_gpe_hdr->proto == VXLAN_GPE_TYPE_ETH) { 254 info->is_tunnel = 1; 255 info->outer_ethertype = info->ethertype; 256 info->outer_l2_len = info->l2_len; 257 info->outer_l3_len = info->l3_len; 258 info->outer_l4_proto = info->l4_proto; 259 260 eth_hdr = (struct ether_hdr *)((char *)vxlan_gpe_hdr + 261 vxlan_gpe_len); 262 263 parse_ethernet(eth_hdr, info); 264 } else 265 return; 266 267 info->l2_len += ETHER_VXLAN_GPE_HLEN; 268 } 269 270 /* Parse a gre header */ 271 static void 272 parse_gre(struct simple_gre_hdr *gre_hdr, struct testpmd_offload_info *info) 273 { 274 struct ether_hdr *eth_hdr; 275 struct ipv4_hdr *ipv4_hdr; 276 struct ipv6_hdr *ipv6_hdr; 277 uint8_t gre_len = 0; 278 279 gre_len += sizeof(struct simple_gre_hdr); 280 281 if (gre_hdr->flags & _htons(GRE_KEY_PRESENT)) 282 gre_len += GRE_EXT_LEN; 283 if (gre_hdr->flags & _htons(GRE_SEQUENCE_PRESENT)) 284 gre_len += GRE_EXT_LEN; 285 if (gre_hdr->flags & _htons(GRE_CHECKSUM_PRESENT)) 286 gre_len += GRE_EXT_LEN; 287 288 if (gre_hdr->proto == _htons(ETHER_TYPE_IPv4)) { 289 info->is_tunnel = 1; 290 info->outer_ethertype = info->ethertype; 291 info->outer_l2_len = info->l2_len; 292 info->outer_l3_len = info->l3_len; 293 info->outer_l4_proto = info->l4_proto; 294 295 ipv4_hdr = (struct ipv4_hdr *)((char *)gre_hdr + gre_len); 296 297 parse_ipv4(ipv4_hdr, info); 298 info->ethertype = _htons(ETHER_TYPE_IPv4); 299 info->l2_len = 0; 300 301 } else if (gre_hdr->proto == _htons(ETHER_TYPE_IPv6)) { 302 info->is_tunnel = 1; 303 info->outer_ethertype = info->ethertype; 304 info->outer_l2_len = info->l2_len; 305 info->outer_l3_len = info->l3_len; 306 info->outer_l4_proto = info->l4_proto; 307 308 ipv6_hdr = (struct ipv6_hdr *)((char *)gre_hdr + gre_len); 309 310 info->ethertype = _htons(ETHER_TYPE_IPv6); 311 parse_ipv6(ipv6_hdr, info); 312 info->l2_len = 0; 313 314 } else if (gre_hdr->proto == _htons(ETHER_TYPE_TEB)) { 315 info->is_tunnel = 1; 316 info->outer_ethertype = info->ethertype; 317 info->outer_l2_len = info->l2_len; 318 info->outer_l3_len = info->l3_len; 319 info->outer_l4_proto = info->l4_proto; 320 321 eth_hdr = (struct ether_hdr *)((char *)gre_hdr + gre_len); 322 323 parse_ethernet(eth_hdr, info); 324 } else 325 return; 326 327 info->l2_len += gre_len; 328 } 329 330 331 /* Parse an encapsulated ip or ipv6 header */ 332 static void 333 parse_encap_ip(void *encap_ip, struct testpmd_offload_info *info) 334 { 335 struct ipv4_hdr *ipv4_hdr = encap_ip; 336 struct ipv6_hdr *ipv6_hdr = encap_ip; 337 uint8_t ip_version; 338 339 ip_version = (ipv4_hdr->version_ihl & 0xf0) >> 4; 340 341 if (ip_version != 4 && ip_version != 6) 342 return; 343 344 info->is_tunnel = 1; 345 info->outer_ethertype = info->ethertype; 346 info->outer_l2_len = info->l2_len; 347 info->outer_l3_len = info->l3_len; 348 349 if (ip_version == 4) { 350 parse_ipv4(ipv4_hdr, info); 351 info->ethertype = _htons(ETHER_TYPE_IPv4); 352 } else { 353 parse_ipv6(ipv6_hdr, info); 354 info->ethertype = _htons(ETHER_TYPE_IPv6); 355 } 356 info->l2_len = 0; 357 } 358 359 /* if possible, calculate the checksum of a packet in hw or sw, 360 * depending on the testpmd command line configuration */ 361 static uint64_t 362 process_inner_cksums(void *l3_hdr, const struct testpmd_offload_info *info, 363 uint64_t tx_offloads) 364 { 365 struct ipv4_hdr *ipv4_hdr = l3_hdr; 366 struct udp_hdr *udp_hdr; 367 struct tcp_hdr *tcp_hdr; 368 struct sctp_hdr *sctp_hdr; 369 uint64_t ol_flags = 0; 370 uint32_t max_pkt_len, tso_segsz = 0; 371 372 /* ensure packet is large enough to require tso */ 373 if (!info->is_tunnel) { 374 max_pkt_len = info->l2_len + info->l3_len + info->l4_len + 375 info->tso_segsz; 376 if (info->tso_segsz != 0 && info->pkt_len > max_pkt_len) 377 tso_segsz = info->tso_segsz; 378 } else { 379 max_pkt_len = info->outer_l2_len + info->outer_l3_len + 380 info->l2_len + info->l3_len + info->l4_len + 381 info->tunnel_tso_segsz; 382 if (info->tunnel_tso_segsz != 0 && info->pkt_len > max_pkt_len) 383 tso_segsz = info->tunnel_tso_segsz; 384 } 385 386 if (info->ethertype == _htons(ETHER_TYPE_IPv4)) { 387 ipv4_hdr = l3_hdr; 388 ipv4_hdr->hdr_checksum = 0; 389 390 ol_flags |= PKT_TX_IPV4; 391 if (info->l4_proto == IPPROTO_TCP && tso_segsz) { 392 ol_flags |= PKT_TX_IP_CKSUM; 393 } else { 394 if (tx_offloads & DEV_TX_OFFLOAD_IPV4_CKSUM) 395 ol_flags |= PKT_TX_IP_CKSUM; 396 else 397 ipv4_hdr->hdr_checksum = 398 rte_ipv4_cksum(ipv4_hdr); 399 } 400 } else if (info->ethertype == _htons(ETHER_TYPE_IPv6)) 401 ol_flags |= PKT_TX_IPV6; 402 else 403 return 0; /* packet type not supported, nothing to do */ 404 405 if (info->l4_proto == IPPROTO_UDP) { 406 udp_hdr = (struct udp_hdr *)((char *)l3_hdr + info->l3_len); 407 /* do not recalculate udp cksum if it was 0 */ 408 if (udp_hdr->dgram_cksum != 0) { 409 udp_hdr->dgram_cksum = 0; 410 if (tx_offloads & DEV_TX_OFFLOAD_UDP_CKSUM) 411 ol_flags |= PKT_TX_UDP_CKSUM; 412 else { 413 udp_hdr->dgram_cksum = 414 get_udptcp_checksum(l3_hdr, udp_hdr, 415 info->ethertype); 416 } 417 } 418 if (info->gso_enable) 419 ol_flags |= PKT_TX_UDP_SEG; 420 } else if (info->l4_proto == IPPROTO_TCP) { 421 tcp_hdr = (struct tcp_hdr *)((char *)l3_hdr + info->l3_len); 422 tcp_hdr->cksum = 0; 423 if (tso_segsz) 424 ol_flags |= PKT_TX_TCP_SEG; 425 else if (tx_offloads & DEV_TX_OFFLOAD_TCP_CKSUM) 426 ol_flags |= PKT_TX_TCP_CKSUM; 427 else { 428 tcp_hdr->cksum = 429 get_udptcp_checksum(l3_hdr, tcp_hdr, 430 info->ethertype); 431 } 432 if (info->gso_enable) 433 ol_flags |= PKT_TX_TCP_SEG; 434 } else if (info->l4_proto == IPPROTO_SCTP) { 435 sctp_hdr = (struct sctp_hdr *)((char *)l3_hdr + info->l3_len); 436 sctp_hdr->cksum = 0; 437 /* sctp payload must be a multiple of 4 to be 438 * offloaded */ 439 if ((tx_offloads & DEV_TX_OFFLOAD_SCTP_CKSUM) && 440 ((ipv4_hdr->total_length & 0x3) == 0)) { 441 ol_flags |= PKT_TX_SCTP_CKSUM; 442 } else { 443 /* XXX implement CRC32c, example available in 444 * RFC3309 */ 445 } 446 } 447 448 return ol_flags; 449 } 450 451 /* Calculate the checksum of outer header */ 452 static uint64_t 453 process_outer_cksums(void *outer_l3_hdr, struct testpmd_offload_info *info, 454 uint64_t tx_offloads, int tso_enabled) 455 { 456 struct ipv4_hdr *ipv4_hdr = outer_l3_hdr; 457 struct ipv6_hdr *ipv6_hdr = outer_l3_hdr; 458 struct udp_hdr *udp_hdr; 459 uint64_t ol_flags = 0; 460 461 if (info->outer_ethertype == _htons(ETHER_TYPE_IPv4)) { 462 ipv4_hdr->hdr_checksum = 0; 463 ol_flags |= PKT_TX_OUTER_IPV4; 464 465 if (tx_offloads & DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM) 466 ol_flags |= PKT_TX_OUTER_IP_CKSUM; 467 else 468 ipv4_hdr->hdr_checksum = rte_ipv4_cksum(ipv4_hdr); 469 } else 470 ol_flags |= PKT_TX_OUTER_IPV6; 471 472 if (info->outer_l4_proto != IPPROTO_UDP) 473 return ol_flags; 474 475 /* Skip SW outer UDP checksum generation if HW supports it */ 476 if (tx_offloads & DEV_TX_OFFLOAD_OUTER_UDP_CKSUM) { 477 ol_flags |= PKT_TX_OUTER_UDP_CKSUM; 478 return ol_flags; 479 } 480 481 udp_hdr = (struct udp_hdr *)((char *)outer_l3_hdr + info->outer_l3_len); 482 483 /* outer UDP checksum is done in software. In the other side, for 484 * UDP tunneling, like VXLAN or Geneve, outer UDP checksum can be 485 * set to zero. 486 * 487 * If a packet will be TSOed into small packets by NIC, we cannot 488 * set/calculate a non-zero checksum, because it will be a wrong 489 * value after the packet be split into several small packets. 490 */ 491 if (tso_enabled) 492 udp_hdr->dgram_cksum = 0; 493 494 /* do not recalculate udp cksum if it was 0 */ 495 if (udp_hdr->dgram_cksum != 0) { 496 udp_hdr->dgram_cksum = 0; 497 if (info->outer_ethertype == _htons(ETHER_TYPE_IPv4)) 498 udp_hdr->dgram_cksum = 499 rte_ipv4_udptcp_cksum(ipv4_hdr, udp_hdr); 500 else 501 udp_hdr->dgram_cksum = 502 rte_ipv6_udptcp_cksum(ipv6_hdr, udp_hdr); 503 } 504 505 return ol_flags; 506 } 507 508 /* 509 * Helper function. 510 * Performs actual copying. 511 * Returns number of segments in the destination mbuf on success, 512 * or negative error code on failure. 513 */ 514 static int 515 mbuf_copy_split(const struct rte_mbuf *ms, struct rte_mbuf *md[], 516 uint16_t seglen[], uint8_t nb_seg) 517 { 518 uint32_t dlen, slen, tlen; 519 uint32_t i, len; 520 const struct rte_mbuf *m; 521 const uint8_t *src; 522 uint8_t *dst; 523 524 dlen = 0; 525 slen = 0; 526 tlen = 0; 527 528 dst = NULL; 529 src = NULL; 530 531 m = ms; 532 i = 0; 533 while (ms != NULL && i != nb_seg) { 534 535 if (slen == 0) { 536 slen = rte_pktmbuf_data_len(ms); 537 src = rte_pktmbuf_mtod(ms, const uint8_t *); 538 } 539 540 if (dlen == 0) { 541 dlen = RTE_MIN(seglen[i], slen); 542 md[i]->data_len = dlen; 543 md[i]->next = (i + 1 == nb_seg) ? NULL : md[i + 1]; 544 dst = rte_pktmbuf_mtod(md[i], uint8_t *); 545 } 546 547 len = RTE_MIN(slen, dlen); 548 memcpy(dst, src, len); 549 tlen += len; 550 slen -= len; 551 dlen -= len; 552 src += len; 553 dst += len; 554 555 if (slen == 0) 556 ms = ms->next; 557 if (dlen == 0) 558 i++; 559 } 560 561 if (ms != NULL) 562 return -ENOBUFS; 563 else if (tlen != m->pkt_len) 564 return -EINVAL; 565 566 md[0]->nb_segs = nb_seg; 567 md[0]->pkt_len = tlen; 568 md[0]->vlan_tci = m->vlan_tci; 569 md[0]->vlan_tci_outer = m->vlan_tci_outer; 570 md[0]->ol_flags = m->ol_flags; 571 md[0]->tx_offload = m->tx_offload; 572 573 return nb_seg; 574 } 575 576 /* 577 * Allocate a new mbuf with up to tx_pkt_nb_segs segments. 578 * Copy packet contents and offload information into the new segmented mbuf. 579 */ 580 static struct rte_mbuf * 581 pkt_copy_split(const struct rte_mbuf *pkt) 582 { 583 int32_t n, rc; 584 uint32_t i, len, nb_seg; 585 struct rte_mempool *mp; 586 uint16_t seglen[RTE_MAX_SEGS_PER_PKT]; 587 struct rte_mbuf *p, *md[RTE_MAX_SEGS_PER_PKT]; 588 589 mp = current_fwd_lcore()->mbp; 590 591 if (tx_pkt_split == TX_PKT_SPLIT_RND) 592 nb_seg = random() % tx_pkt_nb_segs + 1; 593 else 594 nb_seg = tx_pkt_nb_segs; 595 596 memcpy(seglen, tx_pkt_seg_lengths, nb_seg * sizeof(seglen[0])); 597 598 /* calculate number of segments to use and their length. */ 599 len = 0; 600 for (i = 0; i != nb_seg && len < pkt->pkt_len; i++) { 601 len += seglen[i]; 602 md[i] = NULL; 603 } 604 605 n = pkt->pkt_len - len; 606 607 /* update size of the last segment to fit rest of the packet */ 608 if (n >= 0) { 609 seglen[i - 1] += n; 610 len += n; 611 } 612 613 nb_seg = i; 614 while (i != 0) { 615 p = rte_pktmbuf_alloc(mp); 616 if (p == NULL) { 617 TESTPMD_LOG(ERR, 618 "failed to allocate %u-th of %u mbuf " 619 "from mempool: %s\n", 620 nb_seg - i, nb_seg, mp->name); 621 break; 622 } 623 624 md[--i] = p; 625 if (rte_pktmbuf_tailroom(md[i]) < seglen[i]) { 626 TESTPMD_LOG(ERR, "mempool %s, %u-th segment: " 627 "expected seglen: %u, " 628 "actual mbuf tailroom: %u\n", 629 mp->name, i, seglen[i], 630 rte_pktmbuf_tailroom(md[i])); 631 break; 632 } 633 } 634 635 /* all mbufs successfully allocated, do copy */ 636 if (i == 0) { 637 rc = mbuf_copy_split(pkt, md, seglen, nb_seg); 638 if (rc < 0) 639 TESTPMD_LOG(ERR, 640 "mbuf_copy_split for %p(len=%u, nb_seg=%u) " 641 "into %u segments failed with error code: %d\n", 642 pkt, pkt->pkt_len, pkt->nb_segs, nb_seg, rc); 643 644 /* figure out how many mbufs to free. */ 645 i = RTE_MAX(rc, 0); 646 } 647 648 /* free unused mbufs */ 649 for (; i != nb_seg; i++) { 650 rte_pktmbuf_free_seg(md[i]); 651 md[i] = NULL; 652 } 653 654 return md[0]; 655 } 656 657 /* 658 * Receive a burst of packets, and for each packet: 659 * - parse packet, and try to recognize a supported packet type (1) 660 * - if it's not a supported packet type, don't touch the packet, else: 661 * - reprocess the checksum of all supported layers. This is done in SW 662 * or HW, depending on testpmd command line configuration 663 * - if TSO is enabled in testpmd command line, also flag the mbuf for TCP 664 * segmentation offload (this implies HW TCP checksum) 665 * Then transmit packets on the output port. 666 * 667 * (1) Supported packets are: 668 * Ether / (vlan) / IP|IP6 / UDP|TCP|SCTP . 669 * Ether / (vlan) / outer IP|IP6 / outer UDP / VxLAN / Ether / IP|IP6 / 670 * UDP|TCP|SCTP 671 * Ether / (vlan) / outer IP|IP6 / outer UDP / VXLAN-GPE / Ether / IP|IP6 / 672 * UDP|TCP|SCTP 673 * Ether / (vlan) / outer IP|IP6 / outer UDP / VXLAN-GPE / IP|IP6 / 674 * UDP|TCP|SCTP 675 * Ether / (vlan) / outer IP|IP6 / GRE / Ether / IP|IP6 / UDP|TCP|SCTP 676 * Ether / (vlan) / outer IP|IP6 / GRE / IP|IP6 / UDP|TCP|SCTP 677 * Ether / (vlan) / outer IP|IP6 / IP|IP6 / UDP|TCP|SCTP 678 * 679 * The testpmd command line for this forward engine sets the flags 680 * TESTPMD_TX_OFFLOAD_* in ports[tx_port].tx_ol_flags. They control 681 * wether a checksum must be calculated in software or in hardware. The 682 * IP, UDP, TCP and SCTP flags always concern the inner layer. The 683 * OUTER_IP is only useful for tunnel packets. 684 */ 685 static void 686 pkt_burst_checksum_forward(struct fwd_stream *fs) 687 { 688 struct rte_mbuf *pkts_burst[MAX_PKT_BURST]; 689 struct rte_mbuf *gso_segments[GSO_MAX_PKT_BURST]; 690 struct rte_gso_ctx *gso_ctx; 691 struct rte_mbuf **tx_pkts_burst; 692 struct rte_port *txp; 693 struct rte_mbuf *m, *p; 694 struct ether_hdr *eth_hdr; 695 void *l3_hdr = NULL, *outer_l3_hdr = NULL; /* can be IPv4 or IPv6 */ 696 void **gro_ctx; 697 uint16_t gro_pkts_num; 698 uint8_t gro_enable; 699 uint16_t nb_rx; 700 uint16_t nb_tx; 701 uint16_t nb_prep; 702 uint16_t i; 703 uint64_t rx_ol_flags, tx_ol_flags; 704 uint64_t tx_offloads; 705 uint32_t retry; 706 uint32_t rx_bad_ip_csum; 707 uint32_t rx_bad_l4_csum; 708 uint32_t rx_bad_outer_l4_csum; 709 struct testpmd_offload_info info; 710 uint16_t nb_segments = 0; 711 int ret; 712 713 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES 714 uint64_t start_tsc; 715 uint64_t end_tsc; 716 uint64_t core_cycles; 717 #endif 718 719 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES 720 start_tsc = rte_rdtsc(); 721 #endif 722 723 /* receive a burst of packet */ 724 nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue, pkts_burst, 725 nb_pkt_per_burst); 726 if (unlikely(nb_rx == 0)) 727 return; 728 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS 729 fs->rx_burst_stats.pkt_burst_spread[nb_rx]++; 730 #endif 731 fs->rx_packets += nb_rx; 732 rx_bad_ip_csum = 0; 733 rx_bad_l4_csum = 0; 734 rx_bad_outer_l4_csum = 0; 735 gro_enable = gro_ports[fs->rx_port].enable; 736 737 txp = &ports[fs->tx_port]; 738 tx_offloads = txp->dev_conf.txmode.offloads; 739 memset(&info, 0, sizeof(info)); 740 info.tso_segsz = txp->tso_segsz; 741 info.tunnel_tso_segsz = txp->tunnel_tso_segsz; 742 if (gso_ports[fs->tx_port].enable) 743 info.gso_enable = 1; 744 745 for (i = 0; i < nb_rx; i++) { 746 if (likely(i < nb_rx - 1)) 747 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[i + 1], 748 void *)); 749 750 m = pkts_burst[i]; 751 info.is_tunnel = 0; 752 info.pkt_len = rte_pktmbuf_pkt_len(m); 753 tx_ol_flags = m->ol_flags & 754 (IND_ATTACHED_MBUF | EXT_ATTACHED_MBUF); 755 rx_ol_flags = m->ol_flags; 756 757 /* Update the L3/L4 checksum error packet statistics */ 758 if ((rx_ol_flags & PKT_RX_IP_CKSUM_MASK) == PKT_RX_IP_CKSUM_BAD) 759 rx_bad_ip_csum += 1; 760 if ((rx_ol_flags & PKT_RX_L4_CKSUM_MASK) == PKT_RX_L4_CKSUM_BAD) 761 rx_bad_l4_csum += 1; 762 if (rx_ol_flags & PKT_RX_OUTER_L4_CKSUM_BAD) 763 rx_bad_outer_l4_csum += 1; 764 765 /* step 1: dissect packet, parsing optional vlan, ip4/ip6, vxlan 766 * and inner headers */ 767 768 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *); 769 ether_addr_copy(&peer_eth_addrs[fs->peer_addr], 770 ð_hdr->d_addr); 771 ether_addr_copy(&ports[fs->tx_port].eth_addr, 772 ð_hdr->s_addr); 773 parse_ethernet(eth_hdr, &info); 774 l3_hdr = (char *)eth_hdr + info.l2_len; 775 776 /* check if it's a supported tunnel */ 777 if (txp->parse_tunnel) { 778 if (info.l4_proto == IPPROTO_UDP) { 779 struct udp_hdr *udp_hdr; 780 781 udp_hdr = (struct udp_hdr *)((char *)l3_hdr + 782 info.l3_len); 783 parse_vxlan_gpe(udp_hdr, &info); 784 if (info.is_tunnel) { 785 tx_ol_flags |= PKT_TX_TUNNEL_VXLAN_GPE; 786 } else { 787 parse_vxlan(udp_hdr, &info, 788 m->packet_type); 789 if (info.is_tunnel) 790 tx_ol_flags |= 791 PKT_TX_TUNNEL_VXLAN; 792 } 793 } else if (info.l4_proto == IPPROTO_GRE) { 794 struct simple_gre_hdr *gre_hdr; 795 796 gre_hdr = (struct simple_gre_hdr *) 797 ((char *)l3_hdr + info.l3_len); 798 parse_gre(gre_hdr, &info); 799 if (info.is_tunnel) 800 tx_ol_flags |= PKT_TX_TUNNEL_GRE; 801 } else if (info.l4_proto == IPPROTO_IPIP) { 802 void *encap_ip_hdr; 803 804 encap_ip_hdr = (char *)l3_hdr + info.l3_len; 805 parse_encap_ip(encap_ip_hdr, &info); 806 if (info.is_tunnel) 807 tx_ol_flags |= PKT_TX_TUNNEL_IPIP; 808 } 809 } 810 811 /* update l3_hdr and outer_l3_hdr if a tunnel was parsed */ 812 if (info.is_tunnel) { 813 outer_l3_hdr = l3_hdr; 814 l3_hdr = (char *)l3_hdr + info.outer_l3_len + info.l2_len; 815 } 816 817 /* step 2: depending on user command line configuration, 818 * recompute checksum either in software or flag the 819 * mbuf to offload the calculation to the NIC. If TSO 820 * is configured, prepare the mbuf for TCP segmentation. */ 821 822 /* process checksums of inner headers first */ 823 tx_ol_flags |= process_inner_cksums(l3_hdr, &info, 824 tx_offloads); 825 826 /* Then process outer headers if any. Note that the software 827 * checksum will be wrong if one of the inner checksums is 828 * processed in hardware. */ 829 if (info.is_tunnel == 1) { 830 tx_ol_flags |= process_outer_cksums(outer_l3_hdr, &info, 831 tx_offloads, 832 !!(tx_ol_flags & PKT_TX_TCP_SEG)); 833 } 834 835 /* step 3: fill the mbuf meta data (flags and header lengths) */ 836 837 m->tx_offload = 0; 838 if (info.is_tunnel == 1) { 839 if (info.tunnel_tso_segsz || 840 (tx_offloads & 841 DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM) || 842 (tx_offloads & 843 DEV_TX_OFFLOAD_OUTER_UDP_CKSUM) || 844 (tx_ol_flags & PKT_TX_OUTER_IPV6)) { 845 m->outer_l2_len = info.outer_l2_len; 846 m->outer_l3_len = info.outer_l3_len; 847 m->l2_len = info.l2_len; 848 m->l3_len = info.l3_len; 849 m->l4_len = info.l4_len; 850 m->tso_segsz = info.tunnel_tso_segsz; 851 } 852 else { 853 /* if there is a outer UDP cksum 854 processed in sw and the inner in hw, 855 the outer checksum will be wrong as 856 the payload will be modified by the 857 hardware */ 858 m->l2_len = info.outer_l2_len + 859 info.outer_l3_len + info.l2_len; 860 m->l3_len = info.l3_len; 861 m->l4_len = info.l4_len; 862 } 863 } else { 864 /* this is only useful if an offload flag is 865 * set, but it does not hurt to fill it in any 866 * case */ 867 m->l2_len = info.l2_len; 868 m->l3_len = info.l3_len; 869 m->l4_len = info.l4_len; 870 m->tso_segsz = info.tso_segsz; 871 } 872 m->ol_flags = tx_ol_flags; 873 874 /* Do split & copy for the packet. */ 875 if (tx_pkt_split != TX_PKT_SPLIT_OFF) { 876 p = pkt_copy_split(m); 877 if (p != NULL) { 878 rte_pktmbuf_free(m); 879 m = p; 880 pkts_burst[i] = m; 881 } 882 } 883 884 /* if verbose mode is enabled, dump debug info */ 885 if (verbose_level > 0) { 886 char buf[256]; 887 888 printf("-----------------\n"); 889 printf("port=%u, mbuf=%p, pkt_len=%u, nb_segs=%u:\n", 890 fs->rx_port, m, m->pkt_len, m->nb_segs); 891 /* dump rx parsed packet info */ 892 rte_get_rx_ol_flag_list(rx_ol_flags, buf, sizeof(buf)); 893 printf("rx: l2_len=%d ethertype=%x l3_len=%d " 894 "l4_proto=%d l4_len=%d flags=%s\n", 895 info.l2_len, rte_be_to_cpu_16(info.ethertype), 896 info.l3_len, info.l4_proto, info.l4_len, buf); 897 if (rx_ol_flags & PKT_RX_LRO) 898 printf("rx: m->lro_segsz=%u\n", m->tso_segsz); 899 if (info.is_tunnel == 1) 900 printf("rx: outer_l2_len=%d outer_ethertype=%x " 901 "outer_l3_len=%d\n", info.outer_l2_len, 902 rte_be_to_cpu_16(info.outer_ethertype), 903 info.outer_l3_len); 904 /* dump tx packet info */ 905 if ((tx_offloads & (DEV_TX_OFFLOAD_IPV4_CKSUM | 906 DEV_TX_OFFLOAD_UDP_CKSUM | 907 DEV_TX_OFFLOAD_TCP_CKSUM | 908 DEV_TX_OFFLOAD_SCTP_CKSUM)) || 909 info.tso_segsz != 0) 910 printf("tx: m->l2_len=%d m->l3_len=%d " 911 "m->l4_len=%d\n", 912 m->l2_len, m->l3_len, m->l4_len); 913 if (info.is_tunnel == 1) { 914 if ((tx_offloads & 915 DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM) || 916 (tx_offloads & 917 DEV_TX_OFFLOAD_OUTER_UDP_CKSUM) || 918 (tx_ol_flags & PKT_TX_OUTER_IPV6)) 919 printf("tx: m->outer_l2_len=%d " 920 "m->outer_l3_len=%d\n", 921 m->outer_l2_len, 922 m->outer_l3_len); 923 if (info.tunnel_tso_segsz != 0 && 924 (m->ol_flags & PKT_TX_TCP_SEG)) 925 printf("tx: m->tso_segsz=%d\n", 926 m->tso_segsz); 927 } else if (info.tso_segsz != 0 && 928 (m->ol_flags & PKT_TX_TCP_SEG)) 929 printf("tx: m->tso_segsz=%d\n", m->tso_segsz); 930 rte_get_tx_ol_flag_list(m->ol_flags, buf, sizeof(buf)); 931 printf("tx: flags=%s", buf); 932 printf("\n"); 933 } 934 } 935 936 if (unlikely(gro_enable)) { 937 if (gro_flush_cycles == GRO_DEFAULT_FLUSH_CYCLES) { 938 nb_rx = rte_gro_reassemble_burst(pkts_burst, nb_rx, 939 &(gro_ports[fs->rx_port].param)); 940 } else { 941 gro_ctx = current_fwd_lcore()->gro_ctx; 942 nb_rx = rte_gro_reassemble(pkts_burst, nb_rx, gro_ctx); 943 944 if (++fs->gro_times >= gro_flush_cycles) { 945 gro_pkts_num = rte_gro_get_pkt_count(gro_ctx); 946 if (gro_pkts_num > MAX_PKT_BURST - nb_rx) 947 gro_pkts_num = MAX_PKT_BURST - nb_rx; 948 949 nb_rx += rte_gro_timeout_flush(gro_ctx, 0, 950 RTE_GRO_TCP_IPV4, 951 &pkts_burst[nb_rx], 952 gro_pkts_num); 953 fs->gro_times = 0; 954 } 955 } 956 } 957 958 if (gso_ports[fs->tx_port].enable == 0) 959 tx_pkts_burst = pkts_burst; 960 else { 961 gso_ctx = &(current_fwd_lcore()->gso_ctx); 962 gso_ctx->gso_size = gso_max_segment_size; 963 for (i = 0; i < nb_rx; i++) { 964 ret = rte_gso_segment(pkts_burst[i], gso_ctx, 965 &gso_segments[nb_segments], 966 GSO_MAX_PKT_BURST - nb_segments); 967 if (ret >= 0) 968 nb_segments += ret; 969 else { 970 TESTPMD_LOG(DEBUG, "Unable to segment packet"); 971 rte_pktmbuf_free(pkts_burst[i]); 972 } 973 } 974 975 tx_pkts_burst = gso_segments; 976 nb_rx = nb_segments; 977 } 978 979 nb_prep = rte_eth_tx_prepare(fs->tx_port, fs->tx_queue, 980 tx_pkts_burst, nb_rx); 981 if (nb_prep != nb_rx) 982 printf("Preparing packet burst to transmit failed: %s\n", 983 rte_strerror(rte_errno)); 984 985 nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue, tx_pkts_burst, 986 nb_prep); 987 988 /* 989 * Retry if necessary 990 */ 991 if (unlikely(nb_tx < nb_rx) && fs->retry_enabled) { 992 retry = 0; 993 while (nb_tx < nb_rx && retry++ < burst_tx_retry_num) { 994 rte_delay_us(burst_tx_delay_time); 995 nb_tx += rte_eth_tx_burst(fs->tx_port, fs->tx_queue, 996 &tx_pkts_burst[nb_tx], nb_rx - nb_tx); 997 } 998 } 999 fs->tx_packets += nb_tx; 1000 fs->rx_bad_ip_csum += rx_bad_ip_csum; 1001 fs->rx_bad_l4_csum += rx_bad_l4_csum; 1002 fs->rx_bad_outer_l4_csum += rx_bad_outer_l4_csum; 1003 1004 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS 1005 fs->tx_burst_stats.pkt_burst_spread[nb_tx]++; 1006 #endif 1007 if (unlikely(nb_tx < nb_rx)) { 1008 fs->fwd_dropped += (nb_rx - nb_tx); 1009 do { 1010 rte_pktmbuf_free(tx_pkts_burst[nb_tx]); 1011 } while (++nb_tx < nb_rx); 1012 } 1013 1014 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES 1015 end_tsc = rte_rdtsc(); 1016 core_cycles = (end_tsc - start_tsc); 1017 fs->core_cycles = (uint64_t) (fs->core_cycles + core_cycles); 1018 #endif 1019 } 1020 1021 struct fwd_engine csum_fwd_engine = { 1022 .fwd_mode_name = "csum", 1023 .port_fwd_begin = NULL, 1024 .port_fwd_end = NULL, 1025 .packet_fwd = pkt_burst_checksum_forward, 1026 }; 1027