1 /*- 2 * BSD LICENSE 3 * 4 * Copyright(c) 2010-2014 Intel Corporation. All rights reserved. 5 * Copyright 2014 6WIND S.A. 6 * All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 12 * * Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * * Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in 16 * the documentation and/or other materials provided with the 17 * distribution. 18 * * Neither the name of Intel Corporation nor the names of its 19 * contributors may be used to endorse or promote products derived 20 * from this software without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 23 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 24 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 25 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 26 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 27 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 28 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 29 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 30 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 31 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 32 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 33 */ 34 35 #include <stdarg.h> 36 #include <stdio.h> 37 #include <errno.h> 38 #include <stdint.h> 39 #include <unistd.h> 40 #include <inttypes.h> 41 42 #include <sys/queue.h> 43 #include <sys/stat.h> 44 45 #include <rte_common.h> 46 #include <rte_byteorder.h> 47 #include <rte_log.h> 48 #include <rte_debug.h> 49 #include <rte_cycles.h> 50 #include <rte_memory.h> 51 #include <rte_memcpy.h> 52 #include <rte_launch.h> 53 #include <rte_eal.h> 54 #include <rte_per_lcore.h> 55 #include <rte_lcore.h> 56 #include <rte_atomic.h> 57 #include <rte_branch_prediction.h> 58 #include <rte_mempool.h> 59 #include <rte_mbuf.h> 60 #include <rte_interrupts.h> 61 #include <rte_pci.h> 62 #include <rte_ether.h> 63 #include <rte_ethdev.h> 64 #include <rte_ip.h> 65 #include <rte_tcp.h> 66 #include <rte_udp.h> 67 #include <rte_sctp.h> 68 #include <rte_prefetch.h> 69 #include <rte_string_fns.h> 70 #include <rte_flow.h> 71 #include <rte_gro.h> 72 #include <rte_gso.h> 73 74 #include "testpmd.h" 75 76 #define IP_DEFTTL 64 /* from RFC 1340. */ 77 #define IP_VERSION 0x40 78 #define IP_HDRLEN 0x05 /* default IP header length == five 32-bits words. */ 79 #define IP_VHL_DEF (IP_VERSION | IP_HDRLEN) 80 81 #define GRE_KEY_PRESENT 0x2000 82 #define GRE_KEY_LEN 4 83 #define GRE_SUPPORTED_FIELDS GRE_KEY_PRESENT 84 85 /* We cannot use rte_cpu_to_be_16() on a constant in a switch/case */ 86 #if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN 87 #define _htons(x) ((uint16_t)((((x) & 0x00ffU) << 8) | (((x) & 0xff00U) >> 8))) 88 #else 89 #define _htons(x) (x) 90 #endif 91 92 /* structure that caches offload info for the current packet */ 93 struct testpmd_offload_info { 94 uint16_t ethertype; 95 uint8_t gso_enable; 96 uint16_t l2_len; 97 uint16_t l3_len; 98 uint16_t l4_len; 99 uint8_t l4_proto; 100 uint8_t is_tunnel; 101 uint16_t outer_ethertype; 102 uint16_t outer_l2_len; 103 uint16_t outer_l3_len; 104 uint8_t outer_l4_proto; 105 uint16_t tso_segsz; 106 uint16_t tunnel_tso_segsz; 107 uint32_t pkt_len; 108 }; 109 110 /* simplified GRE header */ 111 struct simple_gre_hdr { 112 uint16_t flags; 113 uint16_t proto; 114 } __attribute__((__packed__)); 115 116 static uint16_t 117 get_udptcp_checksum(void *l3_hdr, void *l4_hdr, uint16_t ethertype) 118 { 119 if (ethertype == _htons(ETHER_TYPE_IPv4)) 120 return rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr); 121 else /* assume ethertype == ETHER_TYPE_IPv6 */ 122 return rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr); 123 } 124 125 /* Parse an IPv4 header to fill l3_len, l4_len, and l4_proto */ 126 static void 127 parse_ipv4(struct ipv4_hdr *ipv4_hdr, struct testpmd_offload_info *info) 128 { 129 struct tcp_hdr *tcp_hdr; 130 131 info->l3_len = (ipv4_hdr->version_ihl & 0x0f) * 4; 132 info->l4_proto = ipv4_hdr->next_proto_id; 133 134 /* only fill l4_len for TCP, it's useful for TSO */ 135 if (info->l4_proto == IPPROTO_TCP) { 136 tcp_hdr = (struct tcp_hdr *)((char *)ipv4_hdr + info->l3_len); 137 info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2; 138 } else 139 info->l4_len = 0; 140 } 141 142 /* Parse an IPv6 header to fill l3_len, l4_len, and l4_proto */ 143 static void 144 parse_ipv6(struct ipv6_hdr *ipv6_hdr, struct testpmd_offload_info *info) 145 { 146 struct tcp_hdr *tcp_hdr; 147 148 info->l3_len = sizeof(struct ipv6_hdr); 149 info->l4_proto = ipv6_hdr->proto; 150 151 /* only fill l4_len for TCP, it's useful for TSO */ 152 if (info->l4_proto == IPPROTO_TCP) { 153 tcp_hdr = (struct tcp_hdr *)((char *)ipv6_hdr + info->l3_len); 154 info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2; 155 } else 156 info->l4_len = 0; 157 } 158 159 /* 160 * Parse an ethernet header to fill the ethertype, l2_len, l3_len and 161 * ipproto. This function is able to recognize IPv4/IPv6 with one optional vlan 162 * header. The l4_len argument is only set in case of TCP (useful for TSO). 163 */ 164 static void 165 parse_ethernet(struct ether_hdr *eth_hdr, struct testpmd_offload_info *info) 166 { 167 struct ipv4_hdr *ipv4_hdr; 168 struct ipv6_hdr *ipv6_hdr; 169 170 info->l2_len = sizeof(struct ether_hdr); 171 info->ethertype = eth_hdr->ether_type; 172 173 if (info->ethertype == _htons(ETHER_TYPE_VLAN)) { 174 struct vlan_hdr *vlan_hdr = (struct vlan_hdr *)(eth_hdr + 1); 175 176 info->l2_len += sizeof(struct vlan_hdr); 177 info->ethertype = vlan_hdr->eth_proto; 178 } 179 180 switch (info->ethertype) { 181 case _htons(ETHER_TYPE_IPv4): 182 ipv4_hdr = (struct ipv4_hdr *) ((char *)eth_hdr + info->l2_len); 183 parse_ipv4(ipv4_hdr, info); 184 break; 185 case _htons(ETHER_TYPE_IPv6): 186 ipv6_hdr = (struct ipv6_hdr *) ((char *)eth_hdr + info->l2_len); 187 parse_ipv6(ipv6_hdr, info); 188 break; 189 default: 190 info->l4_len = 0; 191 info->l3_len = 0; 192 info->l4_proto = 0; 193 break; 194 } 195 } 196 197 /* Parse a vxlan header */ 198 static void 199 parse_vxlan(struct udp_hdr *udp_hdr, 200 struct testpmd_offload_info *info, 201 uint32_t pkt_type) 202 { 203 struct ether_hdr *eth_hdr; 204 205 /* check udp destination port, 4789 is the default vxlan port 206 * (rfc7348) or that the rx offload flag is set (i40e only 207 * currently) */ 208 if (udp_hdr->dst_port != _htons(4789) && 209 RTE_ETH_IS_TUNNEL_PKT(pkt_type) == 0) 210 return; 211 212 info->is_tunnel = 1; 213 info->outer_ethertype = info->ethertype; 214 info->outer_l2_len = info->l2_len; 215 info->outer_l3_len = info->l3_len; 216 info->outer_l4_proto = info->l4_proto; 217 218 eth_hdr = (struct ether_hdr *)((char *)udp_hdr + 219 sizeof(struct udp_hdr) + 220 sizeof(struct vxlan_hdr)); 221 222 parse_ethernet(eth_hdr, info); 223 info->l2_len += ETHER_VXLAN_HLEN; /* add udp + vxlan */ 224 } 225 226 /* Parse a gre header */ 227 static void 228 parse_gre(struct simple_gre_hdr *gre_hdr, struct testpmd_offload_info *info) 229 { 230 struct ether_hdr *eth_hdr; 231 struct ipv4_hdr *ipv4_hdr; 232 struct ipv6_hdr *ipv6_hdr; 233 uint8_t gre_len = 0; 234 235 /* check which fields are supported */ 236 if ((gre_hdr->flags & _htons(~GRE_SUPPORTED_FIELDS)) != 0) 237 return; 238 239 gre_len += sizeof(struct simple_gre_hdr); 240 241 if (gre_hdr->flags & _htons(GRE_KEY_PRESENT)) 242 gre_len += GRE_KEY_LEN; 243 244 if (gre_hdr->proto == _htons(ETHER_TYPE_IPv4)) { 245 info->is_tunnel = 1; 246 info->outer_ethertype = info->ethertype; 247 info->outer_l2_len = info->l2_len; 248 info->outer_l3_len = info->l3_len; 249 info->outer_l4_proto = info->l4_proto; 250 251 ipv4_hdr = (struct ipv4_hdr *)((char *)gre_hdr + gre_len); 252 253 parse_ipv4(ipv4_hdr, info); 254 info->ethertype = _htons(ETHER_TYPE_IPv4); 255 info->l2_len = 0; 256 257 } else if (gre_hdr->proto == _htons(ETHER_TYPE_IPv6)) { 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 ipv6_hdr = (struct ipv6_hdr *)((char *)gre_hdr + gre_len); 265 266 info->ethertype = _htons(ETHER_TYPE_IPv6); 267 parse_ipv6(ipv6_hdr, info); 268 info->l2_len = 0; 269 270 } else if (gre_hdr->proto == _htons(ETHER_TYPE_TEB)) { 271 info->is_tunnel = 1; 272 info->outer_ethertype = info->ethertype; 273 info->outer_l2_len = info->l2_len; 274 info->outer_l3_len = info->l3_len; 275 info->outer_l4_proto = info->l4_proto; 276 277 eth_hdr = (struct ether_hdr *)((char *)gre_hdr + gre_len); 278 279 parse_ethernet(eth_hdr, info); 280 } else 281 return; 282 283 info->l2_len += gre_len; 284 } 285 286 287 /* Parse an encapsulated ip or ipv6 header */ 288 static void 289 parse_encap_ip(void *encap_ip, struct testpmd_offload_info *info) 290 { 291 struct ipv4_hdr *ipv4_hdr = encap_ip; 292 struct ipv6_hdr *ipv6_hdr = encap_ip; 293 uint8_t ip_version; 294 295 ip_version = (ipv4_hdr->version_ihl & 0xf0) >> 4; 296 297 if (ip_version != 4 && ip_version != 6) 298 return; 299 300 info->is_tunnel = 1; 301 info->outer_ethertype = info->ethertype; 302 info->outer_l2_len = info->l2_len; 303 info->outer_l3_len = info->l3_len; 304 305 if (ip_version == 4) { 306 parse_ipv4(ipv4_hdr, info); 307 info->ethertype = _htons(ETHER_TYPE_IPv4); 308 } else { 309 parse_ipv6(ipv6_hdr, info); 310 info->ethertype = _htons(ETHER_TYPE_IPv6); 311 } 312 info->l2_len = 0; 313 } 314 315 /* if possible, calculate the checksum of a packet in hw or sw, 316 * depending on the testpmd command line configuration */ 317 static uint64_t 318 process_inner_cksums(void *l3_hdr, const struct testpmd_offload_info *info, 319 uint64_t tx_offloads) 320 { 321 struct ipv4_hdr *ipv4_hdr = l3_hdr; 322 struct udp_hdr *udp_hdr; 323 struct tcp_hdr *tcp_hdr; 324 struct sctp_hdr *sctp_hdr; 325 uint64_t ol_flags = 0; 326 uint32_t max_pkt_len, tso_segsz = 0; 327 328 /* ensure packet is large enough to require tso */ 329 if (!info->is_tunnel) { 330 max_pkt_len = info->l2_len + info->l3_len + info->l4_len + 331 info->tso_segsz; 332 if (info->tso_segsz != 0 && info->pkt_len > max_pkt_len) 333 tso_segsz = info->tso_segsz; 334 } else { 335 max_pkt_len = info->outer_l2_len + info->outer_l3_len + 336 info->l2_len + info->l3_len + info->l4_len + 337 info->tunnel_tso_segsz; 338 if (info->tunnel_tso_segsz != 0 && info->pkt_len > max_pkt_len) 339 tso_segsz = info->tunnel_tso_segsz; 340 } 341 342 if (info->ethertype == _htons(ETHER_TYPE_IPv4)) { 343 ipv4_hdr = l3_hdr; 344 ipv4_hdr->hdr_checksum = 0; 345 346 ol_flags |= PKT_TX_IPV4; 347 if (info->l4_proto == IPPROTO_TCP && tso_segsz) { 348 ol_flags |= PKT_TX_IP_CKSUM; 349 } else { 350 if (tx_offloads & DEV_TX_OFFLOAD_IPV4_CKSUM) 351 ol_flags |= PKT_TX_IP_CKSUM; 352 else 353 ipv4_hdr->hdr_checksum = 354 rte_ipv4_cksum(ipv4_hdr); 355 } 356 } else if (info->ethertype == _htons(ETHER_TYPE_IPv6)) 357 ol_flags |= PKT_TX_IPV6; 358 else 359 return 0; /* packet type not supported, nothing to do */ 360 361 if (info->l4_proto == IPPROTO_UDP) { 362 udp_hdr = (struct udp_hdr *)((char *)l3_hdr + info->l3_len); 363 /* do not recalculate udp cksum if it was 0 */ 364 if (udp_hdr->dgram_cksum != 0) { 365 udp_hdr->dgram_cksum = 0; 366 if (tx_offloads & DEV_TX_OFFLOAD_UDP_CKSUM) 367 ol_flags |= PKT_TX_UDP_CKSUM; 368 else { 369 udp_hdr->dgram_cksum = 370 get_udptcp_checksum(l3_hdr, udp_hdr, 371 info->ethertype); 372 } 373 } 374 } else if (info->l4_proto == IPPROTO_TCP) { 375 tcp_hdr = (struct tcp_hdr *)((char *)l3_hdr + info->l3_len); 376 tcp_hdr->cksum = 0; 377 if (tso_segsz) 378 ol_flags |= PKT_TX_TCP_SEG; 379 else if (tx_offloads & DEV_TX_OFFLOAD_TCP_CKSUM) 380 ol_flags |= PKT_TX_TCP_CKSUM; 381 else { 382 tcp_hdr->cksum = 383 get_udptcp_checksum(l3_hdr, tcp_hdr, 384 info->ethertype); 385 } 386 if (info->gso_enable) 387 ol_flags |= PKT_TX_TCP_SEG; 388 } else if (info->l4_proto == IPPROTO_SCTP) { 389 sctp_hdr = (struct sctp_hdr *)((char *)l3_hdr + info->l3_len); 390 sctp_hdr->cksum = 0; 391 /* sctp payload must be a multiple of 4 to be 392 * offloaded */ 393 if ((tx_offloads & DEV_TX_OFFLOAD_SCTP_CKSUM) && 394 ((ipv4_hdr->total_length & 0x3) == 0)) { 395 ol_flags |= PKT_TX_SCTP_CKSUM; 396 } else { 397 /* XXX implement CRC32c, example available in 398 * RFC3309 */ 399 } 400 } 401 402 return ol_flags; 403 } 404 405 /* Calculate the checksum of outer header */ 406 static uint64_t 407 process_outer_cksums(void *outer_l3_hdr, struct testpmd_offload_info *info, 408 uint64_t tx_offloads, int tso_enabled) 409 { 410 struct ipv4_hdr *ipv4_hdr = outer_l3_hdr; 411 struct ipv6_hdr *ipv6_hdr = outer_l3_hdr; 412 struct udp_hdr *udp_hdr; 413 uint64_t ol_flags = 0; 414 415 if (info->outer_ethertype == _htons(ETHER_TYPE_IPv4)) { 416 ipv4_hdr->hdr_checksum = 0; 417 ol_flags |= PKT_TX_OUTER_IPV4; 418 419 if (tx_offloads & DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM) 420 ol_flags |= PKT_TX_OUTER_IP_CKSUM; 421 else 422 ipv4_hdr->hdr_checksum = rte_ipv4_cksum(ipv4_hdr); 423 } else 424 ol_flags |= PKT_TX_OUTER_IPV6; 425 426 if (info->outer_l4_proto != IPPROTO_UDP) 427 return ol_flags; 428 429 udp_hdr = (struct udp_hdr *)((char *)outer_l3_hdr + info->outer_l3_len); 430 431 /* outer UDP checksum is done in software as we have no hardware 432 * supporting it today, and no API for it. In the other side, for 433 * UDP tunneling, like VXLAN or Geneve, outer UDP checksum can be 434 * set to zero. 435 * 436 * If a packet will be TSOed into small packets by NIC, we cannot 437 * set/calculate a non-zero checksum, because it will be a wrong 438 * value after the packet be split into several small packets. 439 */ 440 if (tso_enabled) 441 udp_hdr->dgram_cksum = 0; 442 443 /* do not recalculate udp cksum if it was 0 */ 444 if (udp_hdr->dgram_cksum != 0) { 445 udp_hdr->dgram_cksum = 0; 446 if (info->outer_ethertype == _htons(ETHER_TYPE_IPv4)) 447 udp_hdr->dgram_cksum = 448 rte_ipv4_udptcp_cksum(ipv4_hdr, udp_hdr); 449 else 450 udp_hdr->dgram_cksum = 451 rte_ipv6_udptcp_cksum(ipv6_hdr, udp_hdr); 452 } 453 454 return ol_flags; 455 } 456 457 /* 458 * Helper function. 459 * Performs actual copying. 460 * Returns number of segments in the destination mbuf on success, 461 * or negative error code on failure. 462 */ 463 static int 464 mbuf_copy_split(const struct rte_mbuf *ms, struct rte_mbuf *md[], 465 uint16_t seglen[], uint8_t nb_seg) 466 { 467 uint32_t dlen, slen, tlen; 468 uint32_t i, len; 469 const struct rte_mbuf *m; 470 const uint8_t *src; 471 uint8_t *dst; 472 473 dlen = 0; 474 slen = 0; 475 tlen = 0; 476 477 dst = NULL; 478 src = NULL; 479 480 m = ms; 481 i = 0; 482 while (ms != NULL && i != nb_seg) { 483 484 if (slen == 0) { 485 slen = rte_pktmbuf_data_len(ms); 486 src = rte_pktmbuf_mtod(ms, const uint8_t *); 487 } 488 489 if (dlen == 0) { 490 dlen = RTE_MIN(seglen[i], slen); 491 md[i]->data_len = dlen; 492 md[i]->next = (i + 1 == nb_seg) ? NULL : md[i + 1]; 493 dst = rte_pktmbuf_mtod(md[i], uint8_t *); 494 } 495 496 len = RTE_MIN(slen, dlen); 497 memcpy(dst, src, len); 498 tlen += len; 499 slen -= len; 500 dlen -= len; 501 src += len; 502 dst += len; 503 504 if (slen == 0) 505 ms = ms->next; 506 if (dlen == 0) 507 i++; 508 } 509 510 if (ms != NULL) 511 return -ENOBUFS; 512 else if (tlen != m->pkt_len) 513 return -EINVAL; 514 515 md[0]->nb_segs = nb_seg; 516 md[0]->pkt_len = tlen; 517 md[0]->vlan_tci = m->vlan_tci; 518 md[0]->vlan_tci_outer = m->vlan_tci_outer; 519 md[0]->ol_flags = m->ol_flags; 520 md[0]->tx_offload = m->tx_offload; 521 522 return nb_seg; 523 } 524 525 /* 526 * Allocate a new mbuf with up to tx_pkt_nb_segs segments. 527 * Copy packet contents and offload information into then new segmented mbuf. 528 */ 529 static struct rte_mbuf * 530 pkt_copy_split(const struct rte_mbuf *pkt) 531 { 532 int32_t n, rc; 533 uint32_t i, len, nb_seg; 534 struct rte_mempool *mp; 535 uint16_t seglen[RTE_MAX_SEGS_PER_PKT]; 536 struct rte_mbuf *p, *md[RTE_MAX_SEGS_PER_PKT]; 537 538 mp = current_fwd_lcore()->mbp; 539 540 if (tx_pkt_split == TX_PKT_SPLIT_RND) 541 nb_seg = random() % tx_pkt_nb_segs + 1; 542 else 543 nb_seg = tx_pkt_nb_segs; 544 545 memcpy(seglen, tx_pkt_seg_lengths, nb_seg * sizeof(seglen[0])); 546 547 /* calculate number of segments to use and their length. */ 548 len = 0; 549 for (i = 0; i != nb_seg && len < pkt->pkt_len; i++) { 550 len += seglen[i]; 551 md[i] = NULL; 552 } 553 554 n = pkt->pkt_len - len; 555 556 /* update size of the last segment to fit rest of the packet */ 557 if (n >= 0) { 558 seglen[i - 1] += n; 559 len += n; 560 } 561 562 nb_seg = i; 563 while (i != 0) { 564 p = rte_pktmbuf_alloc(mp); 565 if (p == NULL) { 566 TESTPMD_LOG(ERR, 567 "failed to allocate %u-th of %u mbuf " 568 "from mempool: %s\n", 569 nb_seg - i, nb_seg, mp->name); 570 break; 571 } 572 573 md[--i] = p; 574 if (rte_pktmbuf_tailroom(md[i]) < seglen[i]) { 575 TESTPMD_LOG(ERR, "mempool %s, %u-th segment: " 576 "expected seglen: %u, " 577 "actual mbuf tailroom: %u\n", 578 mp->name, i, seglen[i], 579 rte_pktmbuf_tailroom(md[i])); 580 break; 581 } 582 } 583 584 /* all mbufs successfully allocated, do copy */ 585 if (i == 0) { 586 rc = mbuf_copy_split(pkt, md, seglen, nb_seg); 587 if (rc < 0) 588 TESTPMD_LOG(ERR, 589 "mbuf_copy_split for %p(len=%u, nb_seg=%u) " 590 "into %u segments failed with error code: %d\n", 591 pkt, pkt->pkt_len, pkt->nb_segs, nb_seg, rc); 592 593 /* figure out how many mbufs to free. */ 594 i = RTE_MAX(rc, 0); 595 } 596 597 /* free unused mbufs */ 598 for (; i != nb_seg; i++) { 599 rte_pktmbuf_free_seg(md[i]); 600 md[i] = NULL; 601 } 602 603 return md[0]; 604 } 605 606 /* 607 * Receive a burst of packets, and for each packet: 608 * - parse packet, and try to recognize a supported packet type (1) 609 * - if it's not a supported packet type, don't touch the packet, else: 610 * - reprocess the checksum of all supported layers. This is done in SW 611 * or HW, depending on testpmd command line configuration 612 * - if TSO is enabled in testpmd command line, also flag the mbuf for TCP 613 * segmentation offload (this implies HW TCP checksum) 614 * Then transmit packets on the output port. 615 * 616 * (1) Supported packets are: 617 * Ether / (vlan) / IP|IP6 / UDP|TCP|SCTP . 618 * Ether / (vlan) / outer IP|IP6 / outer UDP / VxLAN / Ether / IP|IP6 / 619 * UDP|TCP|SCTP 620 * Ether / (vlan) / outer IP|IP6 / GRE / Ether / IP|IP6 / UDP|TCP|SCTP 621 * Ether / (vlan) / outer IP|IP6 / GRE / IP|IP6 / UDP|TCP|SCTP 622 * Ether / (vlan) / outer IP|IP6 / IP|IP6 / UDP|TCP|SCTP 623 * 624 * The testpmd command line for this forward engine sets the flags 625 * TESTPMD_TX_OFFLOAD_* in ports[tx_port].tx_ol_flags. They control 626 * wether a checksum must be calculated in software or in hardware. The 627 * IP, UDP, TCP and SCTP flags always concern the inner layer. The 628 * OUTER_IP is only useful for tunnel packets. 629 */ 630 static void 631 pkt_burst_checksum_forward(struct fwd_stream *fs) 632 { 633 struct rte_mbuf *pkts_burst[MAX_PKT_BURST]; 634 struct rte_mbuf *gso_segments[GSO_MAX_PKT_BURST]; 635 struct rte_gso_ctx *gso_ctx; 636 struct rte_mbuf **tx_pkts_burst; 637 struct rte_port *txp; 638 struct rte_mbuf *m, *p; 639 struct ether_hdr *eth_hdr; 640 void *l3_hdr = NULL, *outer_l3_hdr = NULL; /* can be IPv4 or IPv6 */ 641 void **gro_ctx; 642 uint16_t gro_pkts_num; 643 uint8_t gro_enable; 644 uint16_t nb_rx; 645 uint16_t nb_tx; 646 uint16_t nb_prep; 647 uint16_t i; 648 uint64_t rx_ol_flags, tx_ol_flags; 649 uint64_t tx_offloads; 650 uint32_t retry; 651 uint32_t rx_bad_ip_csum; 652 uint32_t rx_bad_l4_csum; 653 struct testpmd_offload_info info; 654 uint16_t nb_segments = 0; 655 int ret; 656 657 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES 658 uint64_t start_tsc; 659 uint64_t end_tsc; 660 uint64_t core_cycles; 661 #endif 662 663 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES 664 start_tsc = rte_rdtsc(); 665 #endif 666 667 /* receive a burst of packet */ 668 nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue, pkts_burst, 669 nb_pkt_per_burst); 670 if (unlikely(nb_rx == 0)) 671 return; 672 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS 673 fs->rx_burst_stats.pkt_burst_spread[nb_rx]++; 674 #endif 675 fs->rx_packets += nb_rx; 676 rx_bad_ip_csum = 0; 677 rx_bad_l4_csum = 0; 678 gro_enable = gro_ports[fs->rx_port].enable; 679 680 txp = &ports[fs->tx_port]; 681 tx_offloads = txp->dev_conf.txmode.offloads; 682 memset(&info, 0, sizeof(info)); 683 info.tso_segsz = txp->tso_segsz; 684 info.tunnel_tso_segsz = txp->tunnel_tso_segsz; 685 if (gso_ports[fs->tx_port].enable) 686 info.gso_enable = 1; 687 688 for (i = 0; i < nb_rx; i++) { 689 if (likely(i < nb_rx - 1)) 690 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[i + 1], 691 void *)); 692 693 m = pkts_burst[i]; 694 info.is_tunnel = 0; 695 info.pkt_len = rte_pktmbuf_pkt_len(m); 696 tx_ol_flags = 0; 697 rx_ol_flags = m->ol_flags; 698 699 /* Update the L3/L4 checksum error packet statistics */ 700 if ((rx_ol_flags & PKT_RX_IP_CKSUM_MASK) == PKT_RX_IP_CKSUM_BAD) 701 rx_bad_ip_csum += 1; 702 if ((rx_ol_flags & PKT_RX_L4_CKSUM_MASK) == PKT_RX_L4_CKSUM_BAD) 703 rx_bad_l4_csum += 1; 704 705 /* step 1: dissect packet, parsing optional vlan, ip4/ip6, vxlan 706 * and inner headers */ 707 708 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *); 709 ether_addr_copy(&peer_eth_addrs[fs->peer_addr], 710 ð_hdr->d_addr); 711 ether_addr_copy(&ports[fs->tx_port].eth_addr, 712 ð_hdr->s_addr); 713 parse_ethernet(eth_hdr, &info); 714 l3_hdr = (char *)eth_hdr + info.l2_len; 715 716 /* check if it's a supported tunnel */ 717 if (txp->parse_tunnel) { 718 if (info.l4_proto == IPPROTO_UDP) { 719 struct udp_hdr *udp_hdr; 720 721 udp_hdr = (struct udp_hdr *)((char *)l3_hdr + 722 info.l3_len); 723 parse_vxlan(udp_hdr, &info, m->packet_type); 724 if (info.is_tunnel) 725 tx_ol_flags |= PKT_TX_TUNNEL_VXLAN; 726 } else if (info.l4_proto == IPPROTO_GRE) { 727 struct simple_gre_hdr *gre_hdr; 728 729 gre_hdr = (struct simple_gre_hdr *) 730 ((char *)l3_hdr + info.l3_len); 731 parse_gre(gre_hdr, &info); 732 if (info.is_tunnel) 733 tx_ol_flags |= PKT_TX_TUNNEL_GRE; 734 } else if (info.l4_proto == IPPROTO_IPIP) { 735 void *encap_ip_hdr; 736 737 encap_ip_hdr = (char *)l3_hdr + info.l3_len; 738 parse_encap_ip(encap_ip_hdr, &info); 739 if (info.is_tunnel) 740 tx_ol_flags |= PKT_TX_TUNNEL_IPIP; 741 } 742 } 743 744 /* update l3_hdr and outer_l3_hdr if a tunnel was parsed */ 745 if (info.is_tunnel) { 746 outer_l3_hdr = l3_hdr; 747 l3_hdr = (char *)l3_hdr + info.outer_l3_len + info.l2_len; 748 } 749 750 /* step 2: depending on user command line configuration, 751 * recompute checksum either in software or flag the 752 * mbuf to offload the calculation to the NIC. If TSO 753 * is configured, prepare the mbuf for TCP segmentation. */ 754 755 /* process checksums of inner headers first */ 756 tx_ol_flags |= process_inner_cksums(l3_hdr, &info, 757 tx_offloads); 758 759 /* Then process outer headers if any. Note that the software 760 * checksum will be wrong if one of the inner checksums is 761 * processed in hardware. */ 762 if (info.is_tunnel == 1) { 763 tx_ol_flags |= process_outer_cksums(outer_l3_hdr, &info, 764 tx_offloads, 765 !!(tx_ol_flags & PKT_TX_TCP_SEG)); 766 } 767 768 /* step 3: fill the mbuf meta data (flags and header lengths) */ 769 770 if (info.is_tunnel == 1) { 771 if (info.tunnel_tso_segsz || 772 (tx_offloads & 773 DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM) || 774 (tx_ol_flags & PKT_TX_OUTER_IPV6)) { 775 m->outer_l2_len = info.outer_l2_len; 776 m->outer_l3_len = info.outer_l3_len; 777 m->l2_len = info.l2_len; 778 m->l3_len = info.l3_len; 779 m->l4_len = info.l4_len; 780 m->tso_segsz = info.tunnel_tso_segsz; 781 } 782 else { 783 /* if there is a outer UDP cksum 784 processed in sw and the inner in hw, 785 the outer checksum will be wrong as 786 the payload will be modified by the 787 hardware */ 788 m->l2_len = info.outer_l2_len + 789 info.outer_l3_len + info.l2_len; 790 m->l3_len = info.l3_len; 791 m->l4_len = info.l4_len; 792 } 793 } else { 794 /* this is only useful if an offload flag is 795 * set, but it does not hurt to fill it in any 796 * case */ 797 m->l2_len = info.l2_len; 798 m->l3_len = info.l3_len; 799 m->l4_len = info.l4_len; 800 m->tso_segsz = info.tso_segsz; 801 } 802 m->ol_flags = tx_ol_flags; 803 804 /* Do split & copy for the packet. */ 805 if (tx_pkt_split != TX_PKT_SPLIT_OFF) { 806 p = pkt_copy_split(m); 807 if (p != NULL) { 808 rte_pktmbuf_free(m); 809 m = p; 810 pkts_burst[i] = m; 811 } 812 } 813 814 /* if verbose mode is enabled, dump debug info */ 815 if (verbose_level > 0) { 816 char buf[256]; 817 818 printf("-----------------\n"); 819 printf("port=%u, mbuf=%p, pkt_len=%u, nb_segs=%u:\n", 820 fs->rx_port, m, m->pkt_len, m->nb_segs); 821 /* dump rx parsed packet info */ 822 rte_get_rx_ol_flag_list(rx_ol_flags, buf, sizeof(buf)); 823 printf("rx: l2_len=%d ethertype=%x l3_len=%d " 824 "l4_proto=%d l4_len=%d flags=%s\n", 825 info.l2_len, rte_be_to_cpu_16(info.ethertype), 826 info.l3_len, info.l4_proto, info.l4_len, buf); 827 if (rx_ol_flags & PKT_RX_LRO) 828 printf("rx: m->lro_segsz=%u\n", m->tso_segsz); 829 if (info.is_tunnel == 1) 830 printf("rx: outer_l2_len=%d outer_ethertype=%x " 831 "outer_l3_len=%d\n", info.outer_l2_len, 832 rte_be_to_cpu_16(info.outer_ethertype), 833 info.outer_l3_len); 834 /* dump tx packet info */ 835 if ((tx_offloads & (DEV_TX_OFFLOAD_IPV4_CKSUM | 836 DEV_TX_OFFLOAD_UDP_CKSUM | 837 DEV_TX_OFFLOAD_TCP_CKSUM | 838 DEV_TX_OFFLOAD_SCTP_CKSUM)) || 839 info.tso_segsz != 0) 840 printf("tx: m->l2_len=%d m->l3_len=%d " 841 "m->l4_len=%d\n", 842 m->l2_len, m->l3_len, m->l4_len); 843 if (info.is_tunnel == 1) { 844 if ((tx_offloads & 845 DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM) || 846 (tx_ol_flags & PKT_TX_OUTER_IPV6)) 847 printf("tx: m->outer_l2_len=%d " 848 "m->outer_l3_len=%d\n", 849 m->outer_l2_len, 850 m->outer_l3_len); 851 if (info.tunnel_tso_segsz != 0 && 852 (m->ol_flags & PKT_TX_TCP_SEG)) 853 printf("tx: m->tso_segsz=%d\n", 854 m->tso_segsz); 855 } else if (info.tso_segsz != 0 && 856 (m->ol_flags & PKT_TX_TCP_SEG)) 857 printf("tx: m->tso_segsz=%d\n", m->tso_segsz); 858 rte_get_tx_ol_flag_list(m->ol_flags, buf, sizeof(buf)); 859 printf("tx: flags=%s", buf); 860 printf("\n"); 861 } 862 } 863 864 if (unlikely(gro_enable)) { 865 if (gro_flush_cycles == GRO_DEFAULT_FLUSH_CYCLES) { 866 nb_rx = rte_gro_reassemble_burst(pkts_burst, nb_rx, 867 &(gro_ports[fs->rx_port].param)); 868 } else { 869 gro_ctx = current_fwd_lcore()->gro_ctx; 870 nb_rx = rte_gro_reassemble(pkts_burst, nb_rx, gro_ctx); 871 872 if (++fs->gro_times >= gro_flush_cycles) { 873 gro_pkts_num = rte_gro_get_pkt_count(gro_ctx); 874 if (gro_pkts_num > MAX_PKT_BURST - nb_rx) 875 gro_pkts_num = MAX_PKT_BURST - nb_rx; 876 877 nb_rx += rte_gro_timeout_flush(gro_ctx, 0, 878 RTE_GRO_TCP_IPV4, 879 &pkts_burst[nb_rx], 880 gro_pkts_num); 881 fs->gro_times = 0; 882 } 883 } 884 } 885 886 if (gso_ports[fs->tx_port].enable == 0) 887 tx_pkts_burst = pkts_burst; 888 else { 889 gso_ctx = &(current_fwd_lcore()->gso_ctx); 890 gso_ctx->gso_size = gso_max_segment_size; 891 for (i = 0; i < nb_rx; i++) { 892 ret = rte_gso_segment(pkts_burst[i], gso_ctx, 893 &gso_segments[nb_segments], 894 GSO_MAX_PKT_BURST - nb_segments); 895 if (ret >= 0) 896 nb_segments += ret; 897 else { 898 TESTPMD_LOG(DEBUG, "Unable to segment packet"); 899 rte_pktmbuf_free(pkts_burst[i]); 900 } 901 } 902 903 tx_pkts_burst = gso_segments; 904 nb_rx = nb_segments; 905 } 906 907 nb_prep = rte_eth_tx_prepare(fs->tx_port, fs->tx_queue, 908 tx_pkts_burst, nb_rx); 909 if (nb_prep != nb_rx) 910 printf("Preparing packet burst to transmit failed: %s\n", 911 rte_strerror(rte_errno)); 912 913 nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue, tx_pkts_burst, 914 nb_prep); 915 916 /* 917 * Retry if necessary 918 */ 919 if (unlikely(nb_tx < nb_rx) && fs->retry_enabled) { 920 retry = 0; 921 while (nb_tx < nb_rx && retry++ < burst_tx_retry_num) { 922 rte_delay_us(burst_tx_delay_time); 923 nb_tx += rte_eth_tx_burst(fs->tx_port, fs->tx_queue, 924 &tx_pkts_burst[nb_tx], nb_rx - nb_tx); 925 } 926 } 927 fs->tx_packets += nb_tx; 928 fs->rx_bad_ip_csum += rx_bad_ip_csum; 929 fs->rx_bad_l4_csum += rx_bad_l4_csum; 930 931 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS 932 fs->tx_burst_stats.pkt_burst_spread[nb_tx]++; 933 #endif 934 if (unlikely(nb_tx < nb_rx)) { 935 fs->fwd_dropped += (nb_rx - nb_tx); 936 do { 937 rte_pktmbuf_free(tx_pkts_burst[nb_tx]); 938 } while (++nb_tx < nb_rx); 939 } 940 941 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES 942 end_tsc = rte_rdtsc(); 943 core_cycles = (end_tsc - start_tsc); 944 fs->core_cycles = (uint64_t) (fs->core_cycles + core_cycles); 945 #endif 946 } 947 948 struct fwd_engine csum_fwd_engine = { 949 .fwd_mode_name = "csum", 950 .port_fwd_begin = NULL, 951 .port_fwd_end = NULL, 952 .packet_fwd = pkt_burst_checksum_forward, 953 }; 954