1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2016 Intel Corporation
11 #include <sys/queue.h>
16 #include <rte_common.h>
17 #include <rte_byteorder.h>
19 #include <rte_memory.h>
20 #include <rte_memcpy.h>
22 #include <rte_launch.h>
23 #include <rte_atomic.h>
24 #include <rte_cycles.h>
25 #include <rte_prefetch.h>
26 #include <rte_lcore.h>
27 #include <rte_per_lcore.h>
28 #include <rte_branch_prediction.h>
29 #include <rte_interrupts.h>
30 #include <rte_random.h>
31 #include <rte_debug.h>
32 #include <rte_ether.h>
33 #include <rte_ethdev.h>
34 #include <rte_mempool.h>
39 #include <rte_string_fns.h>
42 #include <cmdline_parse.h>
43 #include <cmdline_parse_etheraddr.h>
45 #if RTE_LOG_DP_LEVEL >= RTE_LOG_DEBUG
46 #define L3FWDACL_DEBUG
48 #define DO_RFC_1812_CHECKS
50 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
52 #define MAX_JUMBO_PKT_LEN 9600
54 #define MEMPOOL_CACHE_SIZE 256
57 * This expression is used to calculate the number of mbufs needed
58 * depending on user input, taking into account memory for rx and tx hardware
59 * rings, cache per lcore and mtable per port per lcore.
60 * RTE_MAX is used to ensure that NB_MBUF never goes below a
61 * minimum value of 8192
64 #define NB_MBUF RTE_MAX(\
65 (nb_ports * nb_rx_queue * nb_rxd + \
66 nb_ports * nb_lcores * MAX_PKT_BURST + \
67 nb_ports * n_tx_queue * nb_txd + \
68 nb_lcores * MEMPOOL_CACHE_SIZE), \
71 #define MAX_PKT_BURST 32
72 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
76 /* Configure how many packets ahead to prefetch, when reading packets */
77 #define PREFETCH_OFFSET 3
80 * Configurable number of RX/TX ring descriptors
82 #define RTE_TEST_RX_DESC_DEFAULT 1024
83 #define RTE_TEST_TX_DESC_DEFAULT 1024
84 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
85 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
87 /* mask of enabled ports */
88 static uint32_t enabled_port_mask;
89 static int promiscuous_on; /**< Ports set in promiscuous mode off by default. */
90 static int numa_on = 1; /**< NUMA is enabled by default. */
92 struct lcore_rx_queue {
95 } __rte_cache_aligned;
97 #define MAX_RX_QUEUE_PER_LCORE 16
98 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
99 #define MAX_RX_QUEUE_PER_PORT 128
101 #define MAX_LCORE_PARAMS 1024
102 struct lcore_params {
106 } __rte_cache_aligned;
108 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
109 static struct lcore_params lcore_params_array_default[] = {
121 static struct lcore_params *lcore_params = lcore_params_array_default;
122 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
123 sizeof(lcore_params_array_default[0]);
125 static struct rte_eth_conf port_conf = {
127 .mq_mode = ETH_MQ_RX_RSS,
128 .max_rx_pkt_len = RTE_ETHER_MAX_LEN,
130 .offloads = DEV_RX_OFFLOAD_CHECKSUM,
135 .rss_hf = ETH_RSS_IP | ETH_RSS_UDP |
136 ETH_RSS_TCP | ETH_RSS_SCTP,
140 .mq_mode = ETH_MQ_TX_NONE,
144 static struct rte_mempool *pktmbuf_pool[NB_SOCKETS];
146 /* ethernet addresses of ports */
147 static struct rte_ether_hdr port_l2hdr[RTE_MAX_ETHPORTS];
149 static const struct {
151 enum rte_acl_classify_alg alg;
155 .alg = RTE_ACL_CLASSIFY_SCALAR,
159 .alg = RTE_ACL_CLASSIFY_SSE,
163 .alg = RTE_ACL_CLASSIFY_AVX2,
167 .alg = RTE_ACL_CLASSIFY_NEON,
171 .alg = RTE_ACL_CLASSIFY_ALTIVEC,
175 .alg = RTE_ACL_CLASSIFY_AVX512X16,
179 .alg = RTE_ACL_CLASSIFY_AVX512X32,
183 /***********************start of ACL part******************************/
184 #ifdef DO_RFC_1812_CHECKS
186 is_valid_ipv4_pkt(struct rte_ipv4_hdr *pkt, uint32_t link_len);
189 send_single_packet(struct rte_mbuf *m, uint16_t port);
191 #define MAX_ACL_RULE_NUM 100000
192 #define DEFAULT_MAX_CATEGORIES 1
193 #define L3FWD_ACL_IPV4_NAME "l3fwd-acl-ipv4"
194 #define L3FWD_ACL_IPV6_NAME "l3fwd-acl-ipv6"
195 #define ACL_LEAD_CHAR ('@')
196 #define ROUTE_LEAD_CHAR ('R')
197 #define COMMENT_LEAD_CHAR ('#')
198 #define OPTION_CONFIG "config"
199 #define OPTION_NONUMA "no-numa"
200 #define OPTION_ENBJMO "enable-jumbo"
201 #define OPTION_RULE_IPV4 "rule_ipv4"
202 #define OPTION_RULE_IPV6 "rule_ipv6"
203 #define OPTION_ALG "alg"
204 #define OPTION_ETH_DEST "eth-dest"
205 #define ACL_DENY_SIGNATURE 0xf0000000
206 #define RTE_LOGTYPE_L3FWDACL RTE_LOGTYPE_USER3
207 #define acl_log(format, ...) RTE_LOG(ERR, L3FWDACL, format, ##__VA_ARGS__)
208 #define uint32_t_to_char(ip, a, b, c, d) do {\
209 *a = (unsigned char)(ip >> 24 & 0xff);\
210 *b = (unsigned char)(ip >> 16 & 0xff);\
211 *c = (unsigned char)(ip >> 8 & 0xff);\
212 *d = (unsigned char)(ip & 0xff);\
214 #define OFF_ETHHEAD (sizeof(struct rte_ether_hdr))
215 #define OFF_IPV42PROTO (offsetof(struct rte_ipv4_hdr, next_proto_id))
216 #define OFF_IPV62PROTO (offsetof(struct rte_ipv6_hdr, proto))
217 #define MBUF_IPV4_2PROTO(m) \
218 rte_pktmbuf_mtod_offset((m), uint8_t *, OFF_ETHHEAD + OFF_IPV42PROTO)
219 #define MBUF_IPV6_2PROTO(m) \
220 rte_pktmbuf_mtod_offset((m), uint8_t *, OFF_ETHHEAD + OFF_IPV62PROTO)
222 #define GET_CB_FIELD(in, fd, base, lim, dlm) do { \
226 val = strtoul((in), &end, (base)); \
227 if (errno != 0 || end[0] != (dlm) || val > (lim)) \
229 (fd) = (typeof(fd))val; \
234 * ACL rules should have higher priorities than route ones to ensure ACL rule
235 * always be found when input packets have multi-matches in the database.
236 * A exception case is performance measure, which can define route rules with
237 * higher priority and route rules will always be returned in each lookup.
238 * Reserve range from ACL_RULE_PRIORITY_MAX + 1 to
239 * RTE_ACL_MAX_PRIORITY for route entries in performance measure
241 #define ACL_RULE_PRIORITY_MAX 0x10000000
244 * Forward port info save in ACL lib starts from 1
245 * since ACL assume 0 is invalid.
246 * So, need add 1 when saving and minus 1 when forwarding packets.
248 #define FWD_PORT_SHIFT 1
251 * Rule and trace formats definitions.
264 * That effectively defines order of IPV4VLAN classifications:
266 * - VLAN (TAG and DOMAIN)
269 * - PORTS (SRC and DST)
272 RTE_ACL_IPV4VLAN_PROTO,
273 RTE_ACL_IPV4VLAN_VLAN,
274 RTE_ACL_IPV4VLAN_SRC,
275 RTE_ACL_IPV4VLAN_DST,
276 RTE_ACL_IPV4VLAN_PORTS,
280 struct rte_acl_field_def ipv4_defs[NUM_FIELDS_IPV4] = {
282 .type = RTE_ACL_FIELD_TYPE_BITMASK,
283 .size = sizeof(uint8_t),
284 .field_index = PROTO_FIELD_IPV4,
285 .input_index = RTE_ACL_IPV4VLAN_PROTO,
289 .type = RTE_ACL_FIELD_TYPE_MASK,
290 .size = sizeof(uint32_t),
291 .field_index = SRC_FIELD_IPV4,
292 .input_index = RTE_ACL_IPV4VLAN_SRC,
293 .offset = offsetof(struct rte_ipv4_hdr, src_addr) -
294 offsetof(struct rte_ipv4_hdr, next_proto_id),
297 .type = RTE_ACL_FIELD_TYPE_MASK,
298 .size = sizeof(uint32_t),
299 .field_index = DST_FIELD_IPV4,
300 .input_index = RTE_ACL_IPV4VLAN_DST,
301 .offset = offsetof(struct rte_ipv4_hdr, dst_addr) -
302 offsetof(struct rte_ipv4_hdr, next_proto_id),
305 .type = RTE_ACL_FIELD_TYPE_RANGE,
306 .size = sizeof(uint16_t),
307 .field_index = SRCP_FIELD_IPV4,
308 .input_index = RTE_ACL_IPV4VLAN_PORTS,
309 .offset = sizeof(struct rte_ipv4_hdr) -
310 offsetof(struct rte_ipv4_hdr, next_proto_id),
313 .type = RTE_ACL_FIELD_TYPE_RANGE,
314 .size = sizeof(uint16_t),
315 .field_index = DSTP_FIELD_IPV4,
316 .input_index = RTE_ACL_IPV4VLAN_PORTS,
317 .offset = sizeof(struct rte_ipv4_hdr) -
318 offsetof(struct rte_ipv4_hdr, next_proto_id) +
323 #define IPV6_ADDR_LEN 16
324 #define IPV6_ADDR_U16 (IPV6_ADDR_LEN / sizeof(uint16_t))
325 #define IPV6_ADDR_U32 (IPV6_ADDR_LEN / sizeof(uint32_t))
342 struct rte_acl_field_def ipv6_defs[NUM_FIELDS_IPV6] = {
344 .type = RTE_ACL_FIELD_TYPE_BITMASK,
345 .size = sizeof(uint8_t),
346 .field_index = PROTO_FIELD_IPV6,
347 .input_index = PROTO_FIELD_IPV6,
351 .type = RTE_ACL_FIELD_TYPE_MASK,
352 .size = sizeof(uint32_t),
353 .field_index = SRC1_FIELD_IPV6,
354 .input_index = SRC1_FIELD_IPV6,
355 .offset = offsetof(struct rte_ipv6_hdr, src_addr) -
356 offsetof(struct rte_ipv6_hdr, proto),
359 .type = RTE_ACL_FIELD_TYPE_MASK,
360 .size = sizeof(uint32_t),
361 .field_index = SRC2_FIELD_IPV6,
362 .input_index = SRC2_FIELD_IPV6,
363 .offset = offsetof(struct rte_ipv6_hdr, src_addr) -
364 offsetof(struct rte_ipv6_hdr, proto) + sizeof(uint32_t),
367 .type = RTE_ACL_FIELD_TYPE_MASK,
368 .size = sizeof(uint32_t),
369 .field_index = SRC3_FIELD_IPV6,
370 .input_index = SRC3_FIELD_IPV6,
371 .offset = offsetof(struct rte_ipv6_hdr, src_addr) -
372 offsetof(struct rte_ipv6_hdr, proto) +
373 2 * sizeof(uint32_t),
376 .type = RTE_ACL_FIELD_TYPE_MASK,
377 .size = sizeof(uint32_t),
378 .field_index = SRC4_FIELD_IPV6,
379 .input_index = SRC4_FIELD_IPV6,
380 .offset = offsetof(struct rte_ipv6_hdr, src_addr) -
381 offsetof(struct rte_ipv6_hdr, proto) +
382 3 * sizeof(uint32_t),
385 .type = RTE_ACL_FIELD_TYPE_MASK,
386 .size = sizeof(uint32_t),
387 .field_index = DST1_FIELD_IPV6,
388 .input_index = DST1_FIELD_IPV6,
389 .offset = offsetof(struct rte_ipv6_hdr, dst_addr)
390 - offsetof(struct rte_ipv6_hdr, proto),
393 .type = RTE_ACL_FIELD_TYPE_MASK,
394 .size = sizeof(uint32_t),
395 .field_index = DST2_FIELD_IPV6,
396 .input_index = DST2_FIELD_IPV6,
397 .offset = offsetof(struct rte_ipv6_hdr, dst_addr) -
398 offsetof(struct rte_ipv6_hdr, proto) + sizeof(uint32_t),
401 .type = RTE_ACL_FIELD_TYPE_MASK,
402 .size = sizeof(uint32_t),
403 .field_index = DST3_FIELD_IPV6,
404 .input_index = DST3_FIELD_IPV6,
405 .offset = offsetof(struct rte_ipv6_hdr, dst_addr) -
406 offsetof(struct rte_ipv6_hdr, proto) +
407 2 * sizeof(uint32_t),
410 .type = RTE_ACL_FIELD_TYPE_MASK,
411 .size = sizeof(uint32_t),
412 .field_index = DST4_FIELD_IPV6,
413 .input_index = DST4_FIELD_IPV6,
414 .offset = offsetof(struct rte_ipv6_hdr, dst_addr) -
415 offsetof(struct rte_ipv6_hdr, proto) +
416 3 * sizeof(uint32_t),
419 .type = RTE_ACL_FIELD_TYPE_RANGE,
420 .size = sizeof(uint16_t),
421 .field_index = SRCP_FIELD_IPV6,
422 .input_index = SRCP_FIELD_IPV6,
423 .offset = sizeof(struct rte_ipv6_hdr) -
424 offsetof(struct rte_ipv6_hdr, proto),
427 .type = RTE_ACL_FIELD_TYPE_RANGE,
428 .size = sizeof(uint16_t),
429 .field_index = DSTP_FIELD_IPV6,
430 .input_index = SRCP_FIELD_IPV6,
431 .offset = sizeof(struct rte_ipv6_hdr) -
432 offsetof(struct rte_ipv6_hdr, proto) + sizeof(uint16_t),
441 CB_FLD_SRC_PORT_HIGH,
444 CB_FLD_DST_PORT_HIGH,
450 RTE_ACL_RULE_DEF(acl4_rule, RTE_DIM(ipv4_defs));
451 RTE_ACL_RULE_DEF(acl6_rule, RTE_DIM(ipv6_defs));
453 struct acl_search_t {
454 const uint8_t *data_ipv4[MAX_PKT_BURST];
455 struct rte_mbuf *m_ipv4[MAX_PKT_BURST];
456 uint32_t res_ipv4[MAX_PKT_BURST];
459 const uint8_t *data_ipv6[MAX_PKT_BURST];
460 struct rte_mbuf *m_ipv6[MAX_PKT_BURST];
461 uint32_t res_ipv6[MAX_PKT_BURST];
466 char mapped[NB_SOCKETS];
467 struct rte_acl_ctx *acx_ipv4[NB_SOCKETS];
468 struct rte_acl_ctx *acx_ipv6[NB_SOCKETS];
469 #ifdef L3FWDACL_DEBUG
470 struct acl4_rule *rule_ipv4;
471 struct acl6_rule *rule_ipv6;
476 const char *rule_ipv4_name;
477 const char *rule_ipv6_name;
478 enum rte_acl_classify_alg alg;
481 const char cb_port_delim[] = ":";
484 print_one_ipv4_rule(struct acl4_rule *rule, int extra)
486 unsigned char a, b, c, d;
488 uint32_t_to_char(rule->field[SRC_FIELD_IPV4].value.u32,
490 printf("%hhu.%hhu.%hhu.%hhu/%u ", a, b, c, d,
491 rule->field[SRC_FIELD_IPV4].mask_range.u32);
492 uint32_t_to_char(rule->field[DST_FIELD_IPV4].value.u32,
494 printf("%hhu.%hhu.%hhu.%hhu/%u ", a, b, c, d,
495 rule->field[DST_FIELD_IPV4].mask_range.u32);
496 printf("%hu : %hu %hu : %hu 0x%hhx/0x%hhx ",
497 rule->field[SRCP_FIELD_IPV4].value.u16,
498 rule->field[SRCP_FIELD_IPV4].mask_range.u16,
499 rule->field[DSTP_FIELD_IPV4].value.u16,
500 rule->field[DSTP_FIELD_IPV4].mask_range.u16,
501 rule->field[PROTO_FIELD_IPV4].value.u8,
502 rule->field[PROTO_FIELD_IPV4].mask_range.u8);
504 printf("0x%x-0x%x-0x%x ",
505 rule->data.category_mask,
507 rule->data.userdata);
511 print_one_ipv6_rule(struct acl6_rule *rule, int extra)
513 unsigned char a, b, c, d;
515 uint32_t_to_char(rule->field[SRC1_FIELD_IPV6].value.u32,
517 printf("%.2x%.2x:%.2x%.2x", a, b, c, d);
518 uint32_t_to_char(rule->field[SRC2_FIELD_IPV6].value.u32,
520 printf(":%.2x%.2x:%.2x%.2x", a, b, c, d);
521 uint32_t_to_char(rule->field[SRC3_FIELD_IPV6].value.u32,
523 printf(":%.2x%.2x:%.2x%.2x", a, b, c, d);
524 uint32_t_to_char(rule->field[SRC4_FIELD_IPV6].value.u32,
526 printf(":%.2x%.2x:%.2x%.2x/%u ", a, b, c, d,
527 rule->field[SRC1_FIELD_IPV6].mask_range.u32
528 + rule->field[SRC2_FIELD_IPV6].mask_range.u32
529 + rule->field[SRC3_FIELD_IPV6].mask_range.u32
530 + rule->field[SRC4_FIELD_IPV6].mask_range.u32);
532 uint32_t_to_char(rule->field[DST1_FIELD_IPV6].value.u32,
534 printf("%.2x%.2x:%.2x%.2x", a, b, c, d);
535 uint32_t_to_char(rule->field[DST2_FIELD_IPV6].value.u32,
537 printf(":%.2x%.2x:%.2x%.2x", a, b, c, d);
538 uint32_t_to_char(rule->field[DST3_FIELD_IPV6].value.u32,
540 printf(":%.2x%.2x:%.2x%.2x", a, b, c, d);
541 uint32_t_to_char(rule->field[DST4_FIELD_IPV6].value.u32,
543 printf(":%.2x%.2x:%.2x%.2x/%u ", a, b, c, d,
544 rule->field[DST1_FIELD_IPV6].mask_range.u32
545 + rule->field[DST2_FIELD_IPV6].mask_range.u32
546 + rule->field[DST3_FIELD_IPV6].mask_range.u32
547 + rule->field[DST4_FIELD_IPV6].mask_range.u32);
549 printf("%hu : %hu %hu : %hu 0x%hhx/0x%hhx ",
550 rule->field[SRCP_FIELD_IPV6].value.u16,
551 rule->field[SRCP_FIELD_IPV6].mask_range.u16,
552 rule->field[DSTP_FIELD_IPV6].value.u16,
553 rule->field[DSTP_FIELD_IPV6].mask_range.u16,
554 rule->field[PROTO_FIELD_IPV6].value.u8,
555 rule->field[PROTO_FIELD_IPV6].mask_range.u8);
557 printf("0x%x-0x%x-0x%x ",
558 rule->data.category_mask,
560 rule->data.userdata);
563 /* Bypass comment and empty lines */
565 is_bypass_line(char *buff)
570 if (buff[0] == COMMENT_LEAD_CHAR)
573 while (buff[i] != '\0') {
574 if (!isspace(buff[i]))
581 #ifdef L3FWDACL_DEBUG
583 dump_acl4_rule(struct rte_mbuf *m, uint32_t sig)
585 uint32_t offset = sig & ~ACL_DENY_SIGNATURE;
586 unsigned char a, b, c, d;
587 struct rte_ipv4_hdr *ipv4_hdr =
588 rte_pktmbuf_mtod_offset(m, struct rte_ipv4_hdr *,
589 sizeof(struct rte_ether_hdr));
591 uint32_t_to_char(rte_bswap32(ipv4_hdr->src_addr), &a, &b, &c, &d);
592 printf("Packet Src:%hhu.%hhu.%hhu.%hhu ", a, b, c, d);
593 uint32_t_to_char(rte_bswap32(ipv4_hdr->dst_addr), &a, &b, &c, &d);
594 printf("Dst:%hhu.%hhu.%hhu.%hhu ", a, b, c, d);
596 printf("Src port:%hu,Dst port:%hu ",
597 rte_bswap16(*(uint16_t *)(ipv4_hdr + 1)),
598 rte_bswap16(*((uint16_t *)(ipv4_hdr + 1) + 1)));
599 printf("hit ACL %d - ", offset);
601 print_one_ipv4_rule(acl_config.rule_ipv4 + offset, 1);
607 dump_acl6_rule(struct rte_mbuf *m, uint32_t sig)
610 uint32_t offset = sig & ~ACL_DENY_SIGNATURE;
611 struct rte_ipv6_hdr *ipv6_hdr =
612 rte_pktmbuf_mtod_offset(m, struct rte_ipv6_hdr *,
613 sizeof(struct rte_ether_hdr));
615 printf("Packet Src");
616 for (i = 0; i < RTE_DIM(ipv6_hdr->src_addr); i += sizeof(uint16_t))
618 ipv6_hdr->src_addr[i], ipv6_hdr->src_addr[i + 1]);
621 for (i = 0; i < RTE_DIM(ipv6_hdr->dst_addr); i += sizeof(uint16_t))
623 ipv6_hdr->dst_addr[i], ipv6_hdr->dst_addr[i + 1]);
625 printf("\nSrc port:%hu,Dst port:%hu ",
626 rte_bswap16(*(uint16_t *)(ipv6_hdr + 1)),
627 rte_bswap16(*((uint16_t *)(ipv6_hdr + 1) + 1)));
628 printf("hit ACL %d - ", offset);
630 print_one_ipv6_rule(acl_config.rule_ipv6 + offset, 1);
634 #endif /* L3FWDACL_DEBUG */
637 dump_ipv4_rules(struct acl4_rule *rule, int num, int extra)
641 for (i = 0; i < num; i++, rule++) {
642 printf("\t%d:", i + 1);
643 print_one_ipv4_rule(rule, extra);
649 dump_ipv6_rules(struct acl6_rule *rule, int num, int extra)
653 for (i = 0; i < num; i++, rule++) {
654 printf("\t%d:", i + 1);
655 print_one_ipv6_rule(rule, extra);
660 #ifdef DO_RFC_1812_CHECKS
662 prepare_one_packet(struct rte_mbuf **pkts_in, struct acl_search_t *acl,
665 struct rte_ipv4_hdr *ipv4_hdr;
666 struct rte_mbuf *pkt = pkts_in[index];
668 if (RTE_ETH_IS_IPV4_HDR(pkt->packet_type)) {
669 ipv4_hdr = rte_pktmbuf_mtod_offset(pkt, struct rte_ipv4_hdr *,
670 sizeof(struct rte_ether_hdr));
672 /* Check to make sure the packet is valid (RFC1812) */
673 if (is_valid_ipv4_pkt(ipv4_hdr, pkt->pkt_len) >= 0) {
675 /* Update time to live and header checksum */
676 --(ipv4_hdr->time_to_live);
677 ++(ipv4_hdr->hdr_checksum);
679 /* Fill acl structure */
680 acl->data_ipv4[acl->num_ipv4] = MBUF_IPV4_2PROTO(pkt);
681 acl->m_ipv4[(acl->num_ipv4)++] = pkt;
684 /* Not a valid IPv4 packet */
685 rte_pktmbuf_free(pkt);
687 } else if (RTE_ETH_IS_IPV6_HDR(pkt->packet_type)) {
688 /* Fill acl structure */
689 acl->data_ipv6[acl->num_ipv6] = MBUF_IPV6_2PROTO(pkt);
690 acl->m_ipv6[(acl->num_ipv6)++] = pkt;
693 /* Unknown type, drop the packet */
694 rte_pktmbuf_free(pkt);
700 prepare_one_packet(struct rte_mbuf **pkts_in, struct acl_search_t *acl,
703 struct rte_mbuf *pkt = pkts_in[index];
705 if (RTE_ETH_IS_IPV4_HDR(pkt->packet_type)) {
706 /* Fill acl structure */
707 acl->data_ipv4[acl->num_ipv4] = MBUF_IPV4_2PROTO(pkt);
708 acl->m_ipv4[(acl->num_ipv4)++] = pkt;
710 } else if (RTE_ETH_IS_IPV6_HDR(pkt->packet_type)) {
711 /* Fill acl structure */
712 acl->data_ipv6[acl->num_ipv6] = MBUF_IPV6_2PROTO(pkt);
713 acl->m_ipv6[(acl->num_ipv6)++] = pkt;
715 /* Unknown type, drop the packet */
716 rte_pktmbuf_free(pkt);
719 #endif /* DO_RFC_1812_CHECKS */
722 prepare_acl_parameter(struct rte_mbuf **pkts_in, struct acl_search_t *acl,
730 /* Prefetch first packets */
731 for (i = 0; i < PREFETCH_OFFSET && i < nb_rx; i++) {
732 rte_prefetch0(rte_pktmbuf_mtod(
733 pkts_in[i], void *));
736 for (i = 0; i < (nb_rx - PREFETCH_OFFSET); i++) {
737 rte_prefetch0(rte_pktmbuf_mtod(pkts_in[
738 i + PREFETCH_OFFSET], void *));
739 prepare_one_packet(pkts_in, acl, i);
742 /* Process left packets */
743 for (; i < nb_rx; i++)
744 prepare_one_packet(pkts_in, acl, i);
748 send_one_packet(struct rte_mbuf *m, uint32_t res)
750 if (likely((res & ACL_DENY_SIGNATURE) == 0 && res != 0)) {
751 /* forward packets */
752 send_single_packet(m,
753 (uint8_t)(res - FWD_PORT_SHIFT));
755 /* in the ACL list, drop it */
756 #ifdef L3FWDACL_DEBUG
757 if ((res & ACL_DENY_SIGNATURE) != 0) {
758 if (RTE_ETH_IS_IPV4_HDR(m->packet_type))
759 dump_acl4_rule(m, res);
760 else if (RTE_ETH_IS_IPV6_HDR(m->packet_type))
761 dump_acl6_rule(m, res);
771 send_packets(struct rte_mbuf **m, uint32_t *res, int num)
775 /* Prefetch first packets */
776 for (i = 0; i < PREFETCH_OFFSET && i < num; i++) {
777 rte_prefetch0(rte_pktmbuf_mtod(
781 for (i = 0; i < (num - PREFETCH_OFFSET); i++) {
782 rte_prefetch0(rte_pktmbuf_mtod(m[
783 i + PREFETCH_OFFSET], void *));
784 send_one_packet(m[i], res[i]);
787 /* Process left packets */
789 send_one_packet(m[i], res[i]);
793 * Parses IPV6 address, exepcts the following format:
794 * XXXX:XXXX:XXXX:XXXX:XXXX:XXXX:XXXX:XXXX (where X - is a hexedecimal digit).
797 parse_ipv6_addr(const char *in, const char **end, uint32_t v[IPV6_ADDR_U32],
800 uint32_t addr[IPV6_ADDR_U16];
802 GET_CB_FIELD(in, addr[0], 16, UINT16_MAX, ':');
803 GET_CB_FIELD(in, addr[1], 16, UINT16_MAX, ':');
804 GET_CB_FIELD(in, addr[2], 16, UINT16_MAX, ':');
805 GET_CB_FIELD(in, addr[3], 16, UINT16_MAX, ':');
806 GET_CB_FIELD(in, addr[4], 16, UINT16_MAX, ':');
807 GET_CB_FIELD(in, addr[5], 16, UINT16_MAX, ':');
808 GET_CB_FIELD(in, addr[6], 16, UINT16_MAX, ':');
809 GET_CB_FIELD(in, addr[7], 16, UINT16_MAX, dlm);
813 v[0] = (addr[0] << 16) + addr[1];
814 v[1] = (addr[2] << 16) + addr[3];
815 v[2] = (addr[4] << 16) + addr[5];
816 v[3] = (addr[6] << 16) + addr[7];
822 parse_ipv6_net(const char *in, struct rte_acl_field field[4])
827 const uint32_t nbu32 = sizeof(uint32_t) * CHAR_BIT;
830 rc = parse_ipv6_addr(in, &mp, v, '/');
835 GET_CB_FIELD(mp, m, 0, CHAR_BIT * sizeof(v), 0);
837 /* put all together. */
838 for (i = 0; i != RTE_DIM(v); i++) {
839 if (m >= (i + 1) * nbu32)
840 field[i].mask_range.u32 = nbu32;
842 field[i].mask_range.u32 = m > (i * nbu32) ?
845 field[i].value.u32 = v[i];
852 parse_cb_ipv6_rule(char *str, struct rte_acl_rule *v, int has_userdata)
855 char *s, *sp, *in[CB_FLD_NUM];
856 static const char *dlm = " \t\n";
857 int dim = has_userdata ? CB_FLD_NUM : CB_FLD_USERDATA;
860 for (i = 0; i != dim; i++, s = NULL) {
861 in[i] = strtok_r(s, dlm, &sp);
866 rc = parse_ipv6_net(in[CB_FLD_SRC_ADDR], v->field + SRC1_FIELD_IPV6);
868 acl_log("failed to read source address/mask: %s\n",
869 in[CB_FLD_SRC_ADDR]);
873 rc = parse_ipv6_net(in[CB_FLD_DST_ADDR], v->field + DST1_FIELD_IPV6);
875 acl_log("failed to read destination address/mask: %s\n",
876 in[CB_FLD_DST_ADDR]);
881 GET_CB_FIELD(in[CB_FLD_SRC_PORT_LOW],
882 v->field[SRCP_FIELD_IPV6].value.u16,
884 GET_CB_FIELD(in[CB_FLD_SRC_PORT_HIGH],
885 v->field[SRCP_FIELD_IPV6].mask_range.u16,
888 if (strncmp(in[CB_FLD_SRC_PORT_DLM], cb_port_delim,
889 sizeof(cb_port_delim)) != 0)
892 /* destination port. */
893 GET_CB_FIELD(in[CB_FLD_DST_PORT_LOW],
894 v->field[DSTP_FIELD_IPV6].value.u16,
896 GET_CB_FIELD(in[CB_FLD_DST_PORT_HIGH],
897 v->field[DSTP_FIELD_IPV6].mask_range.u16,
900 if (strncmp(in[CB_FLD_DST_PORT_DLM], cb_port_delim,
901 sizeof(cb_port_delim)) != 0)
904 if (v->field[SRCP_FIELD_IPV6].mask_range.u16
905 < v->field[SRCP_FIELD_IPV6].value.u16
906 || v->field[DSTP_FIELD_IPV6].mask_range.u16
907 < v->field[DSTP_FIELD_IPV6].value.u16)
910 GET_CB_FIELD(in[CB_FLD_PROTO], v->field[PROTO_FIELD_IPV6].value.u8,
912 GET_CB_FIELD(in[CB_FLD_PROTO], v->field[PROTO_FIELD_IPV6].mask_range.u8,
916 GET_CB_FIELD(in[CB_FLD_USERDATA], v->data.userdata,
923 * Parse ClassBench rules file.
925 * '@'<src_ipv4_addr>'/'<masklen> <space> \
926 * <dst_ipv4_addr>'/'<masklen> <space> \
927 * <src_port_low> <space> ":" <src_port_high> <space> \
928 * <dst_port_low> <space> ":" <dst_port_high> <space> \
932 parse_ipv4_net(const char *in, uint32_t *addr, uint32_t *mask_len)
934 uint8_t a, b, c, d, m;
936 GET_CB_FIELD(in, a, 0, UINT8_MAX, '.');
937 GET_CB_FIELD(in, b, 0, UINT8_MAX, '.');
938 GET_CB_FIELD(in, c, 0, UINT8_MAX, '.');
939 GET_CB_FIELD(in, d, 0, UINT8_MAX, '/');
940 GET_CB_FIELD(in, m, 0, sizeof(uint32_t) * CHAR_BIT, 0);
942 addr[0] = RTE_IPV4(a, b, c, d);
949 parse_cb_ipv4vlan_rule(char *str, struct rte_acl_rule *v, int has_userdata)
952 char *s, *sp, *in[CB_FLD_NUM];
953 static const char *dlm = " \t\n";
954 int dim = has_userdata ? CB_FLD_NUM : CB_FLD_USERDATA;
957 for (i = 0; i != dim; i++, s = NULL) {
958 in[i] = strtok_r(s, dlm, &sp);
963 rc = parse_ipv4_net(in[CB_FLD_SRC_ADDR],
964 &v->field[SRC_FIELD_IPV4].value.u32,
965 &v->field[SRC_FIELD_IPV4].mask_range.u32);
967 acl_log("failed to read source address/mask: %s\n",
968 in[CB_FLD_SRC_ADDR]);
972 rc = parse_ipv4_net(in[CB_FLD_DST_ADDR],
973 &v->field[DST_FIELD_IPV4].value.u32,
974 &v->field[DST_FIELD_IPV4].mask_range.u32);
976 acl_log("failed to read destination address/mask: %s\n",
977 in[CB_FLD_DST_ADDR]);
981 GET_CB_FIELD(in[CB_FLD_SRC_PORT_LOW],
982 v->field[SRCP_FIELD_IPV4].value.u16,
984 GET_CB_FIELD(in[CB_FLD_SRC_PORT_HIGH],
985 v->field[SRCP_FIELD_IPV4].mask_range.u16,
988 if (strncmp(in[CB_FLD_SRC_PORT_DLM], cb_port_delim,
989 sizeof(cb_port_delim)) != 0)
992 GET_CB_FIELD(in[CB_FLD_DST_PORT_LOW],
993 v->field[DSTP_FIELD_IPV4].value.u16,
995 GET_CB_FIELD(in[CB_FLD_DST_PORT_HIGH],
996 v->field[DSTP_FIELD_IPV4].mask_range.u16,
999 if (strncmp(in[CB_FLD_DST_PORT_DLM], cb_port_delim,
1000 sizeof(cb_port_delim)) != 0)
1003 if (v->field[SRCP_FIELD_IPV4].mask_range.u16
1004 < v->field[SRCP_FIELD_IPV4].value.u16
1005 || v->field[DSTP_FIELD_IPV4].mask_range.u16
1006 < v->field[DSTP_FIELD_IPV4].value.u16)
1009 GET_CB_FIELD(in[CB_FLD_PROTO], v->field[PROTO_FIELD_IPV4].value.u8,
1011 GET_CB_FIELD(in[CB_FLD_PROTO], v->field[PROTO_FIELD_IPV4].mask_range.u8,
1015 GET_CB_FIELD(in[CB_FLD_USERDATA], v->data.userdata, 0,
1022 add_rules(const char *rule_path,
1023 struct rte_acl_rule **proute_base,
1024 unsigned int *proute_num,
1025 struct rte_acl_rule **pacl_base,
1026 unsigned int *pacl_num, uint32_t rule_size,
1027 int (*parser)(char *, struct rte_acl_rule*, int))
1029 uint8_t *acl_rules, *route_rules;
1030 struct rte_acl_rule *next;
1031 unsigned int acl_num = 0, route_num = 0, total_num = 0;
1032 unsigned int acl_cnt = 0, route_cnt = 0;
1033 char buff[LINE_MAX];
1034 FILE *fh = fopen(rule_path, "rb");
1039 rte_exit(EXIT_FAILURE, "%s: Open %s failed\n", __func__,
1042 while ((fgets(buff, LINE_MAX, fh) != NULL)) {
1043 if (buff[0] == ROUTE_LEAD_CHAR)
1045 else if (buff[0] == ACL_LEAD_CHAR)
1050 rte_exit(EXIT_FAILURE, "Not find any route entries in %s!\n",
1053 val = fseek(fh, 0, SEEK_SET);
1055 rte_exit(EXIT_FAILURE, "%s: File seek operation failed\n",
1059 acl_rules = calloc(acl_num, rule_size);
1061 if (NULL == acl_rules)
1062 rte_exit(EXIT_FAILURE, "%s: failed to malloc memory\n",
1065 route_rules = calloc(route_num, rule_size);
1067 if (NULL == route_rules)
1068 rte_exit(EXIT_FAILURE, "%s: failed to malloc memory\n",
1072 while (fgets(buff, LINE_MAX, fh) != NULL) {
1075 if (is_bypass_line(buff))
1081 if (s == ROUTE_LEAD_CHAR)
1082 next = (struct rte_acl_rule *)(route_rules +
1083 route_cnt * rule_size);
1086 else if (s == ACL_LEAD_CHAR)
1087 next = (struct rte_acl_rule *)(acl_rules +
1088 acl_cnt * rule_size);
1092 rte_exit(EXIT_FAILURE,
1093 "%s Line %u: should start with leading "
1095 rule_path, i, ROUTE_LEAD_CHAR, ACL_LEAD_CHAR);
1097 if (parser(buff + 1, next, s == ROUTE_LEAD_CHAR) != 0)
1098 rte_exit(EXIT_FAILURE,
1099 "%s Line %u: parse rules error\n",
1102 if (s == ROUTE_LEAD_CHAR) {
1103 /* Check the forwarding port number */
1104 if ((enabled_port_mask & (1 << next->data.userdata)) ==
1106 rte_exit(EXIT_FAILURE,
1107 "%s Line %u: fwd number illegal:%u\n",
1108 rule_path, i, next->data.userdata);
1109 next->data.userdata += FWD_PORT_SHIFT;
1112 next->data.userdata = ACL_DENY_SIGNATURE + acl_cnt;
1116 next->data.priority = RTE_ACL_MAX_PRIORITY - total_num;
1117 next->data.category_mask = -1;
1123 *pacl_base = (struct rte_acl_rule *)acl_rules;
1124 *pacl_num = acl_num;
1125 *proute_base = (struct rte_acl_rule *)route_rules;
1126 *proute_num = route_cnt;
1132 usage_acl_alg(char *buf, size_t sz)
1134 uint32_t i, n, rc, tn;
1138 for (i = 0; i < RTE_DIM(acl_alg); i++) {
1139 rc = snprintf(buf + n, sz - n,
1140 i == RTE_DIM(acl_alg) - 1 ? "%s" : "%s|",
1151 str_acl_alg(enum rte_acl_classify_alg alg)
1155 for (i = 0; i != RTE_DIM(acl_alg); i++) {
1156 if (alg == acl_alg[i].alg)
1157 return acl_alg[i].name;
1163 static enum rte_acl_classify_alg
1164 parse_acl_alg(const char *alg)
1168 for (i = 0; i != RTE_DIM(acl_alg); i++) {
1169 if (strcmp(alg, acl_alg[i].name) == 0)
1170 return acl_alg[i].alg;
1173 return RTE_ACL_CLASSIFY_DEFAULT;
1177 dump_acl_config(void)
1179 printf("ACL option are:\n");
1180 printf(OPTION_RULE_IPV4": %s\n", parm_config.rule_ipv4_name);
1181 printf(OPTION_RULE_IPV6": %s\n", parm_config.rule_ipv6_name);
1182 printf(OPTION_ALG": %s\n", str_acl_alg(parm_config.alg));
1186 check_acl_config(void)
1188 if (parm_config.rule_ipv4_name == NULL) {
1189 acl_log("ACL IPv4 rule file not specified\n");
1191 } else if (parm_config.rule_ipv6_name == NULL) {
1192 acl_log("ACL IPv6 rule file not specified\n");
1199 static struct rte_acl_ctx*
1200 setup_acl(struct rte_acl_rule *route_base,
1201 struct rte_acl_rule *acl_base, unsigned int route_num,
1202 unsigned int acl_num, int ipv6, int socketid)
1204 char name[PATH_MAX];
1205 struct rte_acl_param acl_param;
1206 struct rte_acl_config acl_build_param;
1207 struct rte_acl_ctx *context;
1208 int dim = ipv6 ? RTE_DIM(ipv6_defs) : RTE_DIM(ipv4_defs);
1210 /* Create ACL contexts */
1211 snprintf(name, sizeof(name), "%s%d",
1212 ipv6 ? L3FWD_ACL_IPV6_NAME : L3FWD_ACL_IPV4_NAME,
1215 acl_param.name = name;
1216 acl_param.socket_id = socketid;
1217 acl_param.rule_size = RTE_ACL_RULE_SZ(dim);
1218 acl_param.max_rule_num = MAX_ACL_RULE_NUM;
1220 if ((context = rte_acl_create(&acl_param)) == NULL)
1221 rte_exit(EXIT_FAILURE, "Failed to create ACL context\n");
1223 if (parm_config.alg != RTE_ACL_CLASSIFY_DEFAULT &&
1224 rte_acl_set_ctx_classify(context, parm_config.alg) != 0)
1225 rte_exit(EXIT_FAILURE,
1226 "Failed to setup classify method for ACL context\n");
1228 if (rte_acl_add_rules(context, route_base, route_num) < 0)
1229 rte_exit(EXIT_FAILURE, "add rules failed\n");
1231 if (rte_acl_add_rules(context, acl_base, acl_num) < 0)
1232 rte_exit(EXIT_FAILURE, "add rules failed\n");
1234 /* Perform builds */
1235 memset(&acl_build_param, 0, sizeof(acl_build_param));
1237 acl_build_param.num_categories = DEFAULT_MAX_CATEGORIES;
1238 acl_build_param.num_fields = dim;
1239 memcpy(&acl_build_param.defs, ipv6 ? ipv6_defs : ipv4_defs,
1240 ipv6 ? sizeof(ipv6_defs) : sizeof(ipv4_defs));
1242 if (rte_acl_build(context, &acl_build_param) != 0)
1243 rte_exit(EXIT_FAILURE, "Failed to build ACL trie\n");
1245 rte_acl_dump(context);
1256 struct rte_acl_rule *acl_base_ipv4, *route_base_ipv4,
1257 *acl_base_ipv6, *route_base_ipv6;
1258 unsigned int acl_num_ipv4 = 0, route_num_ipv4 = 0,
1259 acl_num_ipv6 = 0, route_num_ipv6 = 0;
1261 if (check_acl_config() != 0)
1262 rte_exit(EXIT_FAILURE, "Failed to get valid ACL options\n");
1266 /* Load rules from the input file */
1267 if (add_rules(parm_config.rule_ipv4_name, &route_base_ipv4,
1268 &route_num_ipv4, &acl_base_ipv4, &acl_num_ipv4,
1269 sizeof(struct acl4_rule), &parse_cb_ipv4vlan_rule) < 0)
1270 rte_exit(EXIT_FAILURE, "Failed to add rules\n");
1272 acl_log("IPv4 Route entries %u:\n", route_num_ipv4);
1273 dump_ipv4_rules((struct acl4_rule *)route_base_ipv4, route_num_ipv4, 1);
1275 acl_log("IPv4 ACL entries %u:\n", acl_num_ipv4);
1276 dump_ipv4_rules((struct acl4_rule *)acl_base_ipv4, acl_num_ipv4, 1);
1278 if (add_rules(parm_config.rule_ipv6_name, &route_base_ipv6,
1280 &acl_base_ipv6, &acl_num_ipv6,
1281 sizeof(struct acl6_rule), &parse_cb_ipv6_rule) < 0)
1282 rte_exit(EXIT_FAILURE, "Failed to add rules\n");
1284 acl_log("IPv6 Route entries %u:\n", route_num_ipv6);
1285 dump_ipv6_rules((struct acl6_rule *)route_base_ipv6, route_num_ipv6, 1);
1287 acl_log("IPv6 ACL entries %u:\n", acl_num_ipv6);
1288 dump_ipv6_rules((struct acl6_rule *)acl_base_ipv6, acl_num_ipv6, 1);
1290 memset(&acl_config, 0, sizeof(acl_config));
1292 /* Check sockets a context should be created on */
1294 acl_config.mapped[0] = 1;
1296 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1297 if (rte_lcore_is_enabled(lcore_id) == 0)
1300 socketid = rte_lcore_to_socket_id(lcore_id);
1301 if (socketid >= NB_SOCKETS) {
1302 acl_log("Socket %d of lcore %u is out "
1304 socketid, lcore_id, NB_SOCKETS);
1305 free(route_base_ipv4);
1306 free(route_base_ipv6);
1307 free(acl_base_ipv4);
1308 free(acl_base_ipv6);
1312 acl_config.mapped[socketid] = 1;
1316 for (i = 0; i < NB_SOCKETS; i++) {
1317 if (acl_config.mapped[i]) {
1318 acl_config.acx_ipv4[i] = setup_acl(route_base_ipv4,
1319 acl_base_ipv4, route_num_ipv4, acl_num_ipv4,
1322 acl_config.acx_ipv6[i] = setup_acl(route_base_ipv6,
1323 acl_base_ipv6, route_num_ipv6, acl_num_ipv6,
1328 free(route_base_ipv4);
1329 free(route_base_ipv6);
1331 #ifdef L3FWDACL_DEBUG
1332 acl_config.rule_ipv4 = (struct acl4_rule *)acl_base_ipv4;
1333 acl_config.rule_ipv6 = (struct acl6_rule *)acl_base_ipv6;
1335 free(acl_base_ipv4);
1336 free(acl_base_ipv6);
1342 /***********************end of ACL part******************************/
1345 uint16_t n_rx_queue;
1346 struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
1348 uint16_t tx_port_id[RTE_MAX_ETHPORTS];
1349 uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
1350 struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
1351 } __rte_cache_aligned;
1353 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
1355 /* Enqueue a single packet, and send burst if queue is filled */
1357 send_single_packet(struct rte_mbuf *m, uint16_t port)
1360 struct lcore_conf *qconf;
1361 struct rte_ether_hdr *eh;
1363 lcore_id = rte_lcore_id();
1365 /* update src and dst mac*/
1366 eh = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
1367 memcpy(eh, &port_l2hdr[port], sizeof(eh->d_addr) + sizeof(eh->s_addr));
1369 qconf = &lcore_conf[lcore_id];
1370 rte_eth_tx_buffer(port, qconf->tx_queue_id[port],
1371 qconf->tx_buffer[port], m);
1374 #ifdef DO_RFC_1812_CHECKS
1376 is_valid_ipv4_pkt(struct rte_ipv4_hdr *pkt, uint32_t link_len)
1378 /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
1380 * 1. The packet length reported by the Link Layer must be large
1381 * enough to hold the minimum length legal IP datagram (20 bytes).
1383 if (link_len < sizeof(struct rte_ipv4_hdr))
1386 /* 2. The IP checksum must be correct. */
1387 /* this is checked in H/W */
1390 * 3. The IP version number must be 4. If the version number is not 4
1391 * then the packet may be another version of IP, such as IPng or
1394 if (((pkt->version_ihl) >> 4) != 4)
1397 * 4. The IP header length field must be large enough to hold the
1398 * minimum length legal IP datagram (20 bytes = 5 words).
1400 if ((pkt->version_ihl & 0xf) < 5)
1404 * 5. The IP total length field must be large enough to hold the IP
1405 * datagram header, whose length is specified in the IP header length
1408 if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct rte_ipv4_hdr))
1415 /* main processing loop */
1417 main_loop(__rte_unused void *dummy)
1419 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1421 uint64_t prev_tsc, diff_tsc, cur_tsc;
1425 struct lcore_conf *qconf;
1427 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
1428 / US_PER_S * BURST_TX_DRAIN_US;
1431 lcore_id = rte_lcore_id();
1432 qconf = &lcore_conf[lcore_id];
1433 socketid = rte_lcore_to_socket_id(lcore_id);
1435 if (qconf->n_rx_queue == 0) {
1436 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
1440 RTE_LOG(INFO, L3FWD, "entering main loop on lcore %u\n", lcore_id);
1442 for (i = 0; i < qconf->n_rx_queue; i++) {
1444 portid = qconf->rx_queue_list[i].port_id;
1445 queueid = qconf->rx_queue_list[i].queue_id;
1446 RTE_LOG(INFO, L3FWD,
1447 " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
1448 lcore_id, portid, queueid);
1453 cur_tsc = rte_rdtsc();
1456 * TX burst queue drain
1458 diff_tsc = cur_tsc - prev_tsc;
1459 if (unlikely(diff_tsc > drain_tsc)) {
1460 for (i = 0; i < qconf->n_tx_port; ++i) {
1461 portid = qconf->tx_port_id[i];
1462 rte_eth_tx_buffer_flush(portid,
1463 qconf->tx_queue_id[portid],
1464 qconf->tx_buffer[portid]);
1470 * Read packet from RX queues
1472 for (i = 0; i < qconf->n_rx_queue; ++i) {
1474 portid = qconf->rx_queue_list[i].port_id;
1475 queueid = qconf->rx_queue_list[i].queue_id;
1476 nb_rx = rte_eth_rx_burst(portid, queueid,
1477 pkts_burst, MAX_PKT_BURST);
1480 struct acl_search_t acl_search;
1482 prepare_acl_parameter(pkts_burst, &acl_search,
1485 if (acl_search.num_ipv4) {
1487 acl_config.acx_ipv4[socketid],
1488 acl_search.data_ipv4,
1489 acl_search.res_ipv4,
1490 acl_search.num_ipv4,
1491 DEFAULT_MAX_CATEGORIES);
1493 send_packets(acl_search.m_ipv4,
1494 acl_search.res_ipv4,
1495 acl_search.num_ipv4);
1498 if (acl_search.num_ipv6) {
1500 acl_config.acx_ipv6[socketid],
1501 acl_search.data_ipv6,
1502 acl_search.res_ipv6,
1503 acl_search.num_ipv6,
1504 DEFAULT_MAX_CATEGORIES);
1506 send_packets(acl_search.m_ipv6,
1507 acl_search.res_ipv6,
1508 acl_search.num_ipv6);
1516 check_lcore_params(void)
1518 uint8_t queue, lcore;
1522 for (i = 0; i < nb_lcore_params; ++i) {
1523 queue = lcore_params[i].queue_id;
1524 if (queue >= MAX_RX_QUEUE_PER_PORT) {
1525 printf("invalid queue number: %hhu\n", queue);
1528 lcore = lcore_params[i].lcore_id;
1529 if (!rte_lcore_is_enabled(lcore)) {
1530 printf("error: lcore %hhu is not enabled in "
1531 "lcore mask\n", lcore);
1534 socketid = rte_lcore_to_socket_id(lcore);
1535 if (socketid != 0 && numa_on == 0) {
1536 printf("warning: lcore %hhu is on socket %d "
1545 check_port_config(void)
1550 for (i = 0; i < nb_lcore_params; ++i) {
1551 portid = lcore_params[i].port_id;
1553 if ((enabled_port_mask & (1 << portid)) == 0) {
1554 printf("port %u is not enabled in port mask\n", portid);
1557 if (!rte_eth_dev_is_valid_port(portid)) {
1558 printf("port %u is not present on the board\n", portid);
1566 get_port_n_rx_queues(const uint16_t port)
1571 for (i = 0; i < nb_lcore_params; ++i) {
1572 if (lcore_params[i].port_id == port &&
1573 lcore_params[i].queue_id > queue)
1574 queue = lcore_params[i].queue_id;
1576 return (uint8_t)(++queue);
1580 init_lcore_rx_queues(void)
1582 uint16_t i, nb_rx_queue;
1585 for (i = 0; i < nb_lcore_params; ++i) {
1586 lcore = lcore_params[i].lcore_id;
1587 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1588 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1589 printf("error: too many queues (%u) for lcore: %u\n",
1590 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1593 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1594 lcore_params[i].port_id;
1595 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1596 lcore_params[i].queue_id;
1597 lcore_conf[lcore].n_rx_queue++;
1605 print_usage(const char *prgname)
1609 usage_acl_alg(alg, sizeof(alg));
1610 printf("%s [EAL options] -- -p PORTMASK -P"
1611 "--"OPTION_RULE_IPV4"=FILE"
1612 "--"OPTION_RULE_IPV6"=FILE"
1613 " [--"OPTION_CONFIG" (port,queue,lcore)[,(port,queue,lcore]]"
1614 " [--"OPTION_ENBJMO" [--max-pkt-len PKTLEN]]\n"
1615 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1616 " -P : enable promiscuous mode\n"
1617 " --"OPTION_CONFIG": (port,queue,lcore): "
1618 "rx queues configuration\n"
1619 " --"OPTION_NONUMA": optional, disable numa awareness\n"
1620 " --"OPTION_ENBJMO": enable jumbo frame"
1621 " which max packet len is PKTLEN in decimal (64-9600)\n"
1622 " --"OPTION_RULE_IPV4"=FILE: specify the ipv4 rules entries "
1624 "Each rule occupy one line. "
1625 "2 kinds of rules are supported. "
1626 "One is ACL entry at while line leads with character '%c', "
1627 "another is route entry at while line leads with "
1629 " --"OPTION_RULE_IPV6"=FILE: specify the ipv6 rules "
1631 " --"OPTION_ALG": ACL classify method to use, one of: %s\n",
1632 prgname, ACL_LEAD_CHAR, ROUTE_LEAD_CHAR, alg);
1636 parse_max_pkt_len(const char *pktlen)
1641 /* parse decimal string */
1642 len = strtoul(pktlen, &end, 10);
1643 if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1653 parse_portmask(const char *portmask)
1658 /* parse hexadecimal string */
1659 pm = strtoul(portmask, &end, 16);
1660 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1667 parse_config(const char *q_arg)
1670 const char *p, *p0 = q_arg;
1678 unsigned long int_fld[_NUM_FLD];
1679 char *str_fld[_NUM_FLD];
1683 nb_lcore_params = 0;
1685 while ((p = strchr(p0, '(')) != NULL) {
1687 if ((p0 = strchr(p, ')')) == NULL)
1691 if (size >= sizeof(s))
1694 snprintf(s, sizeof(s), "%.*s", size, p);
1695 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1698 for (i = 0; i < _NUM_FLD; i++) {
1700 int_fld[i] = strtoul(str_fld[i], &end, 0);
1701 if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
1704 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1705 printf("exceeded max number of lcore params: %hu\n",
1709 lcore_params_array[nb_lcore_params].port_id =
1710 (uint8_t)int_fld[FLD_PORT];
1711 lcore_params_array[nb_lcore_params].queue_id =
1712 (uint8_t)int_fld[FLD_QUEUE];
1713 lcore_params_array[nb_lcore_params].lcore_id =
1714 (uint8_t)int_fld[FLD_LCORE];
1717 lcore_params = lcore_params_array;
1722 parse_eth_dest(const char *optarg)
1724 unsigned long portid;
1728 portid = strtoul(optarg, &port_end, 0);
1729 if (errno != 0 || port_end == optarg || *port_end++ != ',')
1730 return "Invalid format";
1731 else if (portid >= RTE_MAX_ETHPORTS)
1732 return "port value exceeds RTE_MAX_ETHPORTS("
1733 RTE_STR(RTE_MAX_ETHPORTS) ")";
1735 if (cmdline_parse_etheraddr(NULL, port_end, &port_l2hdr[portid].d_addr,
1736 sizeof(port_l2hdr[portid].d_addr)) < 0)
1737 return "Invalid ethernet address";
1741 /* Parse the argument given in the command line of the application */
1743 parse_args(int argc, char **argv)
1748 char *prgname = argv[0];
1749 static struct option lgopts[] = {
1750 {OPTION_CONFIG, 1, 0, 0},
1751 {OPTION_NONUMA, 0, 0, 0},
1752 {OPTION_ENBJMO, 0, 0, 0},
1753 {OPTION_RULE_IPV4, 1, 0, 0},
1754 {OPTION_RULE_IPV6, 1, 0, 0},
1755 {OPTION_ALG, 1, 0, 0},
1756 {OPTION_ETH_DEST, 1, 0, 0},
1762 while ((opt = getopt_long(argc, argvopt, "p:P",
1763 lgopts, &option_index)) != EOF) {
1768 enabled_port_mask = parse_portmask(optarg);
1769 if (enabled_port_mask == 0) {
1770 printf("invalid portmask\n");
1771 print_usage(prgname);
1776 printf("Promiscuous mode selected\n");
1782 if (!strncmp(lgopts[option_index].name,
1784 sizeof(OPTION_CONFIG))) {
1785 ret = parse_config(optarg);
1787 printf("invalid config\n");
1788 print_usage(prgname);
1793 if (!strncmp(lgopts[option_index].name,
1795 sizeof(OPTION_NONUMA))) {
1796 printf("numa is disabled\n");
1800 if (!strncmp(lgopts[option_index].name,
1801 OPTION_ENBJMO, sizeof(OPTION_ENBJMO))) {
1802 struct option lenopts = {
1809 printf("jumbo frame is enabled\n");
1810 port_conf.rxmode.offloads |=
1811 DEV_RX_OFFLOAD_JUMBO_FRAME;
1812 port_conf.txmode.offloads |=
1813 DEV_TX_OFFLOAD_MULTI_SEGS;
1816 * if no max-pkt-len set, then use the
1817 * default value RTE_ETHER_MAX_LEN
1819 if (0 == getopt_long(argc, argvopt, "",
1820 &lenopts, &option_index)) {
1821 ret = parse_max_pkt_len(optarg);
1823 (ret > MAX_JUMBO_PKT_LEN)) {
1824 printf("invalid packet "
1826 print_usage(prgname);
1829 port_conf.rxmode.max_rx_pkt_len = ret;
1831 printf("set jumbo frame max packet length "
1834 port_conf.rxmode.max_rx_pkt_len);
1837 if (!strncmp(lgopts[option_index].name,
1839 sizeof(OPTION_RULE_IPV4)))
1840 parm_config.rule_ipv4_name = optarg;
1842 if (!strncmp(lgopts[option_index].name,
1844 sizeof(OPTION_RULE_IPV6))) {
1845 parm_config.rule_ipv6_name = optarg;
1848 if (!strncmp(lgopts[option_index].name,
1849 OPTION_ALG, sizeof(OPTION_ALG))) {
1850 parm_config.alg = parse_acl_alg(optarg);
1851 if (parm_config.alg ==
1852 RTE_ACL_CLASSIFY_DEFAULT) {
1853 printf("unknown %s value:\"%s\"\n",
1854 OPTION_ALG, optarg);
1855 print_usage(prgname);
1860 if (!strncmp(lgopts[option_index].name, OPTION_ETH_DEST,
1861 sizeof(OPTION_ETH_DEST))) {
1862 const char *serr = parse_eth_dest(optarg);
1864 printf("invalid %s value:\"%s\": %s\n",
1865 OPTION_ETH_DEST, optarg, serr);
1866 print_usage(prgname);
1874 print_usage(prgname);
1880 argv[optind-1] = prgname;
1883 optind = 1; /* reset getopt lib */
1888 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
1890 char buf[RTE_ETHER_ADDR_FMT_SIZE];
1891 rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
1892 printf("%s%s", name, buf);
1896 init_mem(unsigned nb_mbuf)
1902 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1903 if (rte_lcore_is_enabled(lcore_id) == 0)
1907 socketid = rte_lcore_to_socket_id(lcore_id);
1911 if (socketid >= NB_SOCKETS) {
1912 rte_exit(EXIT_FAILURE,
1913 "Socket %d of lcore %u is out of range %d\n",
1914 socketid, lcore_id, NB_SOCKETS);
1916 if (pktmbuf_pool[socketid] == NULL) {
1917 snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1918 pktmbuf_pool[socketid] =
1919 rte_pktmbuf_pool_create(s, nb_mbuf,
1920 MEMPOOL_CACHE_SIZE, 0,
1921 RTE_MBUF_DEFAULT_BUF_SIZE,
1923 if (pktmbuf_pool[socketid] == NULL)
1924 rte_exit(EXIT_FAILURE,
1925 "Cannot init mbuf pool on socket %d\n",
1928 printf("Allocated mbuf pool on socket %d\n",
1935 /* Check the link status of all ports in up to 9s, and print them finally */
1937 check_all_ports_link_status(uint32_t port_mask)
1939 #define CHECK_INTERVAL 100 /* 100ms */
1940 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1942 uint8_t count, all_ports_up, print_flag = 0;
1943 struct rte_eth_link link;
1945 char link_status_text[RTE_ETH_LINK_MAX_STR_LEN];
1947 printf("\nChecking link status");
1949 for (count = 0; count <= MAX_CHECK_TIME; count++) {
1951 RTE_ETH_FOREACH_DEV(portid) {
1952 if ((port_mask & (1 << portid)) == 0)
1954 memset(&link, 0, sizeof(link));
1955 ret = rte_eth_link_get_nowait(portid, &link);
1958 if (print_flag == 1)
1959 printf("Port %u link get failed: %s\n",
1960 portid, rte_strerror(-ret));
1963 /* print link status if flag set */
1964 if (print_flag == 1) {
1965 rte_eth_link_to_str(link_status_text,
1966 sizeof(link_status_text), &link);
1967 printf("Port %d %s\n", portid,
1971 /* clear all_ports_up flag if any link down */
1972 if (link.link_status == ETH_LINK_DOWN) {
1977 /* after finally printing all link status, get out */
1978 if (print_flag == 1)
1981 if (all_ports_up == 0) {
1984 rte_delay_ms(CHECK_INTERVAL);
1987 /* set the print_flag if all ports up or timeout */
1988 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1996 * build-up default vaues for dest MACs.
1999 set_default_dest_mac(void)
2003 for (i = 0; i != RTE_DIM(port_l2hdr); i++) {
2004 port_l2hdr[i].d_addr.addr_bytes[0] = RTE_ETHER_LOCAL_ADMIN_ADDR;
2005 port_l2hdr[i].d_addr.addr_bytes[5] = i;
2010 main(int argc, char **argv)
2012 struct lcore_conf *qconf;
2013 struct rte_eth_dev_info dev_info;
2014 struct rte_eth_txconf *txconf;
2019 uint32_t n_tx_queue, nb_lcores;
2021 uint8_t nb_rx_queue, queue, socketid;
2024 ret = rte_eal_init(argc, argv);
2026 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
2030 set_default_dest_mac();
2032 /* parse application arguments (after the EAL ones) */
2033 ret = parse_args(argc, argv);
2035 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
2037 if (check_lcore_params() < 0)
2038 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
2040 ret = init_lcore_rx_queues();
2042 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
2044 nb_ports = rte_eth_dev_count_avail();
2046 if (check_port_config() < 0)
2047 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
2049 /* Add ACL rules and route entries, build trie */
2050 if (app_acl_init() < 0)
2051 rte_exit(EXIT_FAILURE, "app_acl_init failed\n");
2053 nb_lcores = rte_lcore_count();
2055 /* initialize all ports */
2056 RTE_ETH_FOREACH_DEV(portid) {
2057 struct rte_eth_conf local_port_conf = port_conf;
2059 /* skip ports that are not enabled */
2060 if ((enabled_port_mask & (1 << portid)) == 0) {
2061 printf("\nSkipping disabled port %d\n", portid);
2066 printf("Initializing port %d ... ", portid);
2069 nb_rx_queue = get_port_n_rx_queues(portid);
2070 n_tx_queue = nb_lcores;
2071 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
2072 n_tx_queue = MAX_TX_QUEUE_PER_PORT;
2073 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
2074 nb_rx_queue, (unsigned)n_tx_queue);
2076 ret = rte_eth_dev_info_get(portid, &dev_info);
2078 rte_exit(EXIT_FAILURE,
2079 "Error during getting device (port %u) info: %s\n",
2080 portid, strerror(-ret));
2082 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
2083 local_port_conf.txmode.offloads |=
2084 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
2086 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
2087 dev_info.flow_type_rss_offloads;
2088 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
2089 port_conf.rx_adv_conf.rss_conf.rss_hf) {
2090 printf("Port %u modified RSS hash function based on hardware support,"
2091 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
2093 port_conf.rx_adv_conf.rss_conf.rss_hf,
2094 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
2097 ret = rte_eth_dev_configure(portid, nb_rx_queue,
2098 (uint16_t)n_tx_queue, &local_port_conf);
2100 rte_exit(EXIT_FAILURE,
2101 "Cannot configure device: err=%d, port=%d\n",
2104 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
2107 rte_exit(EXIT_FAILURE,
2108 "rte_eth_dev_adjust_nb_rx_tx_desc: err=%d, port=%d\n",
2111 ret = rte_eth_macaddr_get(portid, &port_l2hdr[portid].s_addr);
2113 rte_exit(EXIT_FAILURE,
2114 "rte_eth_macaddr_get: err=%d, port=%d\n",
2117 print_ethaddr("Dst MAC:", &port_l2hdr[portid].d_addr);
2118 print_ethaddr(", Src MAC:", &port_l2hdr[portid].s_addr);
2122 ret = init_mem(NB_MBUF);
2124 rte_exit(EXIT_FAILURE, "init_mem failed\n");
2126 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2127 if (rte_lcore_is_enabled(lcore_id) == 0)
2130 /* Initialize TX buffers */
2131 qconf = &lcore_conf[lcore_id];
2132 qconf->tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
2133 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
2134 rte_eth_dev_socket_id(portid));
2135 if (qconf->tx_buffer[portid] == NULL)
2136 rte_exit(EXIT_FAILURE, "Can't allocate tx buffer for port %u\n",
2139 rte_eth_tx_buffer_init(qconf->tx_buffer[portid], MAX_PKT_BURST);
2142 /* init one TX queue per couple (lcore,port) */
2144 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2145 if (rte_lcore_is_enabled(lcore_id) == 0)
2149 socketid = (uint8_t)
2150 rte_lcore_to_socket_id(lcore_id);
2154 printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
2157 ret = rte_eth_dev_info_get(portid, &dev_info);
2159 rte_exit(EXIT_FAILURE,
2160 "Error during getting device (port %u) info: %s\n",
2161 portid, strerror(-ret));
2163 txconf = &dev_info.default_txconf;
2164 txconf->offloads = local_port_conf.txmode.offloads;
2165 ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
2168 rte_exit(EXIT_FAILURE,
2169 "rte_eth_tx_queue_setup: err=%d, "
2170 "port=%d\n", ret, portid);
2172 qconf = &lcore_conf[lcore_id];
2173 qconf->tx_queue_id[portid] = queueid;
2176 qconf->tx_port_id[qconf->n_tx_port] = portid;
2182 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2183 if (rte_lcore_is_enabled(lcore_id) == 0)
2185 qconf = &lcore_conf[lcore_id];
2186 printf("\nInitializing rx queues on lcore %u ... ", lcore_id);
2188 /* init RX queues */
2189 for (queue = 0; queue < qconf->n_rx_queue; ++queue) {
2190 struct rte_eth_rxconf rxq_conf;
2192 portid = qconf->rx_queue_list[queue].port_id;
2193 queueid = qconf->rx_queue_list[queue].queue_id;
2196 socketid = (uint8_t)
2197 rte_lcore_to_socket_id(lcore_id);
2201 printf("rxq=%d,%d,%d ", portid, queueid, socketid);
2204 ret = rte_eth_dev_info_get(portid, &dev_info);
2206 rte_exit(EXIT_FAILURE,
2207 "Error during getting device (port %u) info: %s\n",
2208 portid, strerror(-ret));
2210 rxq_conf = dev_info.default_rxconf;
2211 rxq_conf.offloads = port_conf.rxmode.offloads;
2212 ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
2213 socketid, &rxq_conf,
2214 pktmbuf_pool[socketid]);
2216 rte_exit(EXIT_FAILURE,
2217 "rte_eth_rx_queue_setup: err=%d,"
2218 "port=%d\n", ret, portid);
2225 RTE_ETH_FOREACH_DEV(portid) {
2226 if ((enabled_port_mask & (1 << portid)) == 0)
2230 ret = rte_eth_dev_start(portid);
2232 rte_exit(EXIT_FAILURE,
2233 "rte_eth_dev_start: err=%d, port=%d\n",
2237 * If enabled, put device in promiscuous mode.
2238 * This allows IO forwarding mode to forward packets
2239 * to itself through 2 cross-connected ports of the
2242 if (promiscuous_on) {
2243 ret = rte_eth_promiscuous_enable(portid);
2245 rte_exit(EXIT_FAILURE,
2246 "rte_eth_promiscuous_enable: err=%s, port=%u\n",
2247 rte_strerror(-ret), portid);
2251 check_all_ports_link_status(enabled_port_mask);
2253 /* launch per-lcore init on every lcore */
2254 rte_eal_mp_remote_launch(main_loop, NULL, CALL_MAIN);
2255 RTE_LCORE_FOREACH_WORKER(lcore_id) {
2256 if (rte_eal_wait_lcore(lcore_id) < 0)