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 #if RTE_LOG_DP_LEVEL >= RTE_LOG_DEBUG
43 #define L3FWDACL_DEBUG
45 #define DO_RFC_1812_CHECKS
47 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
49 #define MAX_JUMBO_PKT_LEN 9600
51 #define MEMPOOL_CACHE_SIZE 256
54 * This expression is used to calculate the number of mbufs needed
55 * depending on user input, taking into account memory for rx and tx hardware
56 * rings, cache per lcore and mtable per port per lcore.
57 * RTE_MAX is used to ensure that NB_MBUF never goes below a
58 * minimum value of 8192
61 #define NB_MBUF RTE_MAX(\
62 (nb_ports * nb_rx_queue * nb_rxd + \
63 nb_ports * nb_lcores * MAX_PKT_BURST + \
64 nb_ports * n_tx_queue * nb_txd + \
65 nb_lcores * MEMPOOL_CACHE_SIZE), \
68 #define MAX_PKT_BURST 32
69 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
73 /* Configure how many packets ahead to prefetch, when reading packets */
74 #define PREFETCH_OFFSET 3
77 * Configurable number of RX/TX ring descriptors
79 #define RTE_TEST_RX_DESC_DEFAULT 1024
80 #define RTE_TEST_TX_DESC_DEFAULT 1024
81 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
82 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
84 /* ethernet addresses of ports */
85 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
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 = ETHER_MAX_LEN,
130 .offloads = (DEV_RX_OFFLOAD_CRC_STRIP |
131 DEV_RX_OFFLOAD_CHECKSUM),
136 .rss_hf = ETH_RSS_IP | ETH_RSS_UDP |
137 ETH_RSS_TCP | ETH_RSS_SCTP,
141 .mq_mode = ETH_MQ_TX_NONE,
145 static struct rte_mempool *pktmbuf_pool[NB_SOCKETS];
147 /***********************start of ACL part******************************/
148 #ifdef DO_RFC_1812_CHECKS
150 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len);
153 send_single_packet(struct rte_mbuf *m, uint16_t port);
155 #define MAX_ACL_RULE_NUM 100000
156 #define DEFAULT_MAX_CATEGORIES 1
157 #define L3FWD_ACL_IPV4_NAME "l3fwd-acl-ipv4"
158 #define L3FWD_ACL_IPV6_NAME "l3fwd-acl-ipv6"
159 #define ACL_LEAD_CHAR ('@')
160 #define ROUTE_LEAD_CHAR ('R')
161 #define COMMENT_LEAD_CHAR ('#')
162 #define OPTION_CONFIG "config"
163 #define OPTION_NONUMA "no-numa"
164 #define OPTION_ENBJMO "enable-jumbo"
165 #define OPTION_RULE_IPV4 "rule_ipv4"
166 #define OPTION_RULE_IPV6 "rule_ipv6"
167 #define OPTION_SCALAR "scalar"
168 #define ACL_DENY_SIGNATURE 0xf0000000
169 #define RTE_LOGTYPE_L3FWDACL RTE_LOGTYPE_USER3
170 #define acl_log(format, ...) RTE_LOG(ERR, L3FWDACL, format, ##__VA_ARGS__)
171 #define uint32_t_to_char(ip, a, b, c, d) do {\
172 *a = (unsigned char)(ip >> 24 & 0xff);\
173 *b = (unsigned char)(ip >> 16 & 0xff);\
174 *c = (unsigned char)(ip >> 8 & 0xff);\
175 *d = (unsigned char)(ip & 0xff);\
177 #define OFF_ETHHEAD (sizeof(struct ether_hdr))
178 #define OFF_IPV42PROTO (offsetof(struct ipv4_hdr, next_proto_id))
179 #define OFF_IPV62PROTO (offsetof(struct ipv6_hdr, proto))
180 #define MBUF_IPV4_2PROTO(m) \
181 rte_pktmbuf_mtod_offset((m), uint8_t *, OFF_ETHHEAD + OFF_IPV42PROTO)
182 #define MBUF_IPV6_2PROTO(m) \
183 rte_pktmbuf_mtod_offset((m), uint8_t *, OFF_ETHHEAD + OFF_IPV62PROTO)
185 #define GET_CB_FIELD(in, fd, base, lim, dlm) do { \
189 val = strtoul((in), &end, (base)); \
190 if (errno != 0 || end[0] != (dlm) || val > (lim)) \
192 (fd) = (typeof(fd))val; \
197 * ACL rules should have higher priorities than route ones to ensure ACL rule
198 * always be found when input packets have multi-matches in the database.
199 * A exception case is performance measure, which can define route rules with
200 * higher priority and route rules will always be returned in each lookup.
201 * Reserve range from ACL_RULE_PRIORITY_MAX + 1 to
202 * RTE_ACL_MAX_PRIORITY for route entries in performance measure
204 #define ACL_RULE_PRIORITY_MAX 0x10000000
207 * Forward port info save in ACL lib starts from 1
208 * since ACL assume 0 is invalid.
209 * So, need add 1 when saving and minus 1 when forwarding packets.
211 #define FWD_PORT_SHIFT 1
214 * Rule and trace formats definitions.
227 * That effectively defines order of IPV4VLAN classifications:
229 * - VLAN (TAG and DOMAIN)
232 * - PORTS (SRC and DST)
235 RTE_ACL_IPV4VLAN_PROTO,
236 RTE_ACL_IPV4VLAN_VLAN,
237 RTE_ACL_IPV4VLAN_SRC,
238 RTE_ACL_IPV4VLAN_DST,
239 RTE_ACL_IPV4VLAN_PORTS,
243 struct rte_acl_field_def ipv4_defs[NUM_FIELDS_IPV4] = {
245 .type = RTE_ACL_FIELD_TYPE_BITMASK,
246 .size = sizeof(uint8_t),
247 .field_index = PROTO_FIELD_IPV4,
248 .input_index = RTE_ACL_IPV4VLAN_PROTO,
252 .type = RTE_ACL_FIELD_TYPE_MASK,
253 .size = sizeof(uint32_t),
254 .field_index = SRC_FIELD_IPV4,
255 .input_index = RTE_ACL_IPV4VLAN_SRC,
256 .offset = offsetof(struct ipv4_hdr, src_addr) -
257 offsetof(struct ipv4_hdr, next_proto_id),
260 .type = RTE_ACL_FIELD_TYPE_MASK,
261 .size = sizeof(uint32_t),
262 .field_index = DST_FIELD_IPV4,
263 .input_index = RTE_ACL_IPV4VLAN_DST,
264 .offset = offsetof(struct ipv4_hdr, dst_addr) -
265 offsetof(struct ipv4_hdr, next_proto_id),
268 .type = RTE_ACL_FIELD_TYPE_RANGE,
269 .size = sizeof(uint16_t),
270 .field_index = SRCP_FIELD_IPV4,
271 .input_index = RTE_ACL_IPV4VLAN_PORTS,
272 .offset = sizeof(struct ipv4_hdr) -
273 offsetof(struct ipv4_hdr, next_proto_id),
276 .type = RTE_ACL_FIELD_TYPE_RANGE,
277 .size = sizeof(uint16_t),
278 .field_index = DSTP_FIELD_IPV4,
279 .input_index = RTE_ACL_IPV4VLAN_PORTS,
280 .offset = sizeof(struct ipv4_hdr) -
281 offsetof(struct ipv4_hdr, next_proto_id) +
286 #define IPV6_ADDR_LEN 16
287 #define IPV6_ADDR_U16 (IPV6_ADDR_LEN / sizeof(uint16_t))
288 #define IPV6_ADDR_U32 (IPV6_ADDR_LEN / sizeof(uint32_t))
305 struct rte_acl_field_def ipv6_defs[NUM_FIELDS_IPV6] = {
307 .type = RTE_ACL_FIELD_TYPE_BITMASK,
308 .size = sizeof(uint8_t),
309 .field_index = PROTO_FIELD_IPV6,
310 .input_index = PROTO_FIELD_IPV6,
314 .type = RTE_ACL_FIELD_TYPE_MASK,
315 .size = sizeof(uint32_t),
316 .field_index = SRC1_FIELD_IPV6,
317 .input_index = SRC1_FIELD_IPV6,
318 .offset = offsetof(struct ipv6_hdr, src_addr) -
319 offsetof(struct ipv6_hdr, proto),
322 .type = RTE_ACL_FIELD_TYPE_MASK,
323 .size = sizeof(uint32_t),
324 .field_index = SRC2_FIELD_IPV6,
325 .input_index = SRC2_FIELD_IPV6,
326 .offset = offsetof(struct ipv6_hdr, src_addr) -
327 offsetof(struct ipv6_hdr, proto) + sizeof(uint32_t),
330 .type = RTE_ACL_FIELD_TYPE_MASK,
331 .size = sizeof(uint32_t),
332 .field_index = SRC3_FIELD_IPV6,
333 .input_index = SRC3_FIELD_IPV6,
334 .offset = offsetof(struct ipv6_hdr, src_addr) -
335 offsetof(struct ipv6_hdr, proto) + 2 * sizeof(uint32_t),
338 .type = RTE_ACL_FIELD_TYPE_MASK,
339 .size = sizeof(uint32_t),
340 .field_index = SRC4_FIELD_IPV6,
341 .input_index = SRC4_FIELD_IPV6,
342 .offset = offsetof(struct ipv6_hdr, src_addr) -
343 offsetof(struct ipv6_hdr, proto) + 3 * sizeof(uint32_t),
346 .type = RTE_ACL_FIELD_TYPE_MASK,
347 .size = sizeof(uint32_t),
348 .field_index = DST1_FIELD_IPV6,
349 .input_index = DST1_FIELD_IPV6,
350 .offset = offsetof(struct ipv6_hdr, dst_addr)
351 - offsetof(struct ipv6_hdr, proto),
354 .type = RTE_ACL_FIELD_TYPE_MASK,
355 .size = sizeof(uint32_t),
356 .field_index = DST2_FIELD_IPV6,
357 .input_index = DST2_FIELD_IPV6,
358 .offset = offsetof(struct ipv6_hdr, dst_addr) -
359 offsetof(struct ipv6_hdr, proto) + sizeof(uint32_t),
362 .type = RTE_ACL_FIELD_TYPE_MASK,
363 .size = sizeof(uint32_t),
364 .field_index = DST3_FIELD_IPV6,
365 .input_index = DST3_FIELD_IPV6,
366 .offset = offsetof(struct ipv6_hdr, dst_addr) -
367 offsetof(struct ipv6_hdr, proto) + 2 * sizeof(uint32_t),
370 .type = RTE_ACL_FIELD_TYPE_MASK,
371 .size = sizeof(uint32_t),
372 .field_index = DST4_FIELD_IPV6,
373 .input_index = DST4_FIELD_IPV6,
374 .offset = offsetof(struct ipv6_hdr, dst_addr) -
375 offsetof(struct ipv6_hdr, proto) + 3 * sizeof(uint32_t),
378 .type = RTE_ACL_FIELD_TYPE_RANGE,
379 .size = sizeof(uint16_t),
380 .field_index = SRCP_FIELD_IPV6,
381 .input_index = SRCP_FIELD_IPV6,
382 .offset = sizeof(struct ipv6_hdr) -
383 offsetof(struct ipv6_hdr, proto),
386 .type = RTE_ACL_FIELD_TYPE_RANGE,
387 .size = sizeof(uint16_t),
388 .field_index = DSTP_FIELD_IPV6,
389 .input_index = SRCP_FIELD_IPV6,
390 .offset = sizeof(struct ipv6_hdr) -
391 offsetof(struct ipv6_hdr, proto) + sizeof(uint16_t),
400 CB_FLD_SRC_PORT_HIGH,
403 CB_FLD_DST_PORT_HIGH,
409 RTE_ACL_RULE_DEF(acl4_rule, RTE_DIM(ipv4_defs));
410 RTE_ACL_RULE_DEF(acl6_rule, RTE_DIM(ipv6_defs));
412 struct acl_search_t {
413 const uint8_t *data_ipv4[MAX_PKT_BURST];
414 struct rte_mbuf *m_ipv4[MAX_PKT_BURST];
415 uint32_t res_ipv4[MAX_PKT_BURST];
418 const uint8_t *data_ipv6[MAX_PKT_BURST];
419 struct rte_mbuf *m_ipv6[MAX_PKT_BURST];
420 uint32_t res_ipv6[MAX_PKT_BURST];
425 char mapped[NB_SOCKETS];
426 struct rte_acl_ctx *acx_ipv4[NB_SOCKETS];
427 struct rte_acl_ctx *acx_ipv6[NB_SOCKETS];
428 #ifdef L3FWDACL_DEBUG
429 struct acl4_rule *rule_ipv4;
430 struct acl6_rule *rule_ipv6;
435 const char *rule_ipv4_name;
436 const char *rule_ipv6_name;
440 const char cb_port_delim[] = ":";
443 print_one_ipv4_rule(struct acl4_rule *rule, int extra)
445 unsigned char a, b, c, d;
447 uint32_t_to_char(rule->field[SRC_FIELD_IPV4].value.u32,
449 printf("%hhu.%hhu.%hhu.%hhu/%u ", a, b, c, d,
450 rule->field[SRC_FIELD_IPV4].mask_range.u32);
451 uint32_t_to_char(rule->field[DST_FIELD_IPV4].value.u32,
453 printf("%hhu.%hhu.%hhu.%hhu/%u ", a, b, c, d,
454 rule->field[DST_FIELD_IPV4].mask_range.u32);
455 printf("%hu : %hu %hu : %hu 0x%hhx/0x%hhx ",
456 rule->field[SRCP_FIELD_IPV4].value.u16,
457 rule->field[SRCP_FIELD_IPV4].mask_range.u16,
458 rule->field[DSTP_FIELD_IPV4].value.u16,
459 rule->field[DSTP_FIELD_IPV4].mask_range.u16,
460 rule->field[PROTO_FIELD_IPV4].value.u8,
461 rule->field[PROTO_FIELD_IPV4].mask_range.u8);
463 printf("0x%x-0x%x-0x%x ",
464 rule->data.category_mask,
466 rule->data.userdata);
470 print_one_ipv6_rule(struct acl6_rule *rule, int extra)
472 unsigned char a, b, c, d;
474 uint32_t_to_char(rule->field[SRC1_FIELD_IPV6].value.u32,
476 printf("%.2x%.2x:%.2x%.2x", a, b, c, d);
477 uint32_t_to_char(rule->field[SRC2_FIELD_IPV6].value.u32,
479 printf(":%.2x%.2x:%.2x%.2x", a, b, c, d);
480 uint32_t_to_char(rule->field[SRC3_FIELD_IPV6].value.u32,
482 printf(":%.2x%.2x:%.2x%.2x", a, b, c, d);
483 uint32_t_to_char(rule->field[SRC4_FIELD_IPV6].value.u32,
485 printf(":%.2x%.2x:%.2x%.2x/%u ", a, b, c, d,
486 rule->field[SRC1_FIELD_IPV6].mask_range.u32
487 + rule->field[SRC2_FIELD_IPV6].mask_range.u32
488 + rule->field[SRC3_FIELD_IPV6].mask_range.u32
489 + rule->field[SRC4_FIELD_IPV6].mask_range.u32);
491 uint32_t_to_char(rule->field[DST1_FIELD_IPV6].value.u32,
493 printf("%.2x%.2x:%.2x%.2x", a, b, c, d);
494 uint32_t_to_char(rule->field[DST2_FIELD_IPV6].value.u32,
496 printf(":%.2x%.2x:%.2x%.2x", a, b, c, d);
497 uint32_t_to_char(rule->field[DST3_FIELD_IPV6].value.u32,
499 printf(":%.2x%.2x:%.2x%.2x", a, b, c, d);
500 uint32_t_to_char(rule->field[DST4_FIELD_IPV6].value.u32,
502 printf(":%.2x%.2x:%.2x%.2x/%u ", a, b, c, d,
503 rule->field[DST1_FIELD_IPV6].mask_range.u32
504 + rule->field[DST2_FIELD_IPV6].mask_range.u32
505 + rule->field[DST3_FIELD_IPV6].mask_range.u32
506 + rule->field[DST4_FIELD_IPV6].mask_range.u32);
508 printf("%hu : %hu %hu : %hu 0x%hhx/0x%hhx ",
509 rule->field[SRCP_FIELD_IPV6].value.u16,
510 rule->field[SRCP_FIELD_IPV6].mask_range.u16,
511 rule->field[DSTP_FIELD_IPV6].value.u16,
512 rule->field[DSTP_FIELD_IPV6].mask_range.u16,
513 rule->field[PROTO_FIELD_IPV6].value.u8,
514 rule->field[PROTO_FIELD_IPV6].mask_range.u8);
516 printf("0x%x-0x%x-0x%x ",
517 rule->data.category_mask,
519 rule->data.userdata);
522 /* Bypass comment and empty lines */
524 is_bypass_line(char *buff)
529 if (buff[0] == COMMENT_LEAD_CHAR)
532 while (buff[i] != '\0') {
533 if (!isspace(buff[i]))
540 #ifdef L3FWDACL_DEBUG
542 dump_acl4_rule(struct rte_mbuf *m, uint32_t sig)
544 uint32_t offset = sig & ~ACL_DENY_SIGNATURE;
545 unsigned char a, b, c, d;
546 struct ipv4_hdr *ipv4_hdr = rte_pktmbuf_mtod_offset(m,
548 sizeof(struct ether_hdr));
550 uint32_t_to_char(rte_bswap32(ipv4_hdr->src_addr), &a, &b, &c, &d);
551 printf("Packet Src:%hhu.%hhu.%hhu.%hhu ", a, b, c, d);
552 uint32_t_to_char(rte_bswap32(ipv4_hdr->dst_addr), &a, &b, &c, &d);
553 printf("Dst:%hhu.%hhu.%hhu.%hhu ", a, b, c, d);
555 printf("Src port:%hu,Dst port:%hu ",
556 rte_bswap16(*(uint16_t *)(ipv4_hdr + 1)),
557 rte_bswap16(*((uint16_t *)(ipv4_hdr + 1) + 1)));
558 printf("hit ACL %d - ", offset);
560 print_one_ipv4_rule(acl_config.rule_ipv4 + offset, 1);
566 dump_acl6_rule(struct rte_mbuf *m, uint32_t sig)
569 uint32_t offset = sig & ~ACL_DENY_SIGNATURE;
570 struct ipv6_hdr *ipv6_hdr = rte_pktmbuf_mtod_offset(m,
572 sizeof(struct ether_hdr));
574 printf("Packet Src");
575 for (i = 0; i < RTE_DIM(ipv6_hdr->src_addr); i += sizeof(uint16_t))
577 ipv6_hdr->src_addr[i], ipv6_hdr->src_addr[i + 1]);
580 for (i = 0; i < RTE_DIM(ipv6_hdr->dst_addr); i += sizeof(uint16_t))
582 ipv6_hdr->dst_addr[i], ipv6_hdr->dst_addr[i + 1]);
584 printf("\nSrc port:%hu,Dst port:%hu ",
585 rte_bswap16(*(uint16_t *)(ipv6_hdr + 1)),
586 rte_bswap16(*((uint16_t *)(ipv6_hdr + 1) + 1)));
587 printf("hit ACL %d - ", offset);
589 print_one_ipv6_rule(acl_config.rule_ipv6 + offset, 1);
593 #endif /* L3FWDACL_DEBUG */
596 dump_ipv4_rules(struct acl4_rule *rule, int num, int extra)
600 for (i = 0; i < num; i++, rule++) {
601 printf("\t%d:", i + 1);
602 print_one_ipv4_rule(rule, extra);
608 dump_ipv6_rules(struct acl6_rule *rule, int num, int extra)
612 for (i = 0; i < num; i++, rule++) {
613 printf("\t%d:", i + 1);
614 print_one_ipv6_rule(rule, extra);
619 #ifdef DO_RFC_1812_CHECKS
621 prepare_one_packet(struct rte_mbuf **pkts_in, struct acl_search_t *acl,
624 struct ipv4_hdr *ipv4_hdr;
625 struct rte_mbuf *pkt = pkts_in[index];
627 if (RTE_ETH_IS_IPV4_HDR(pkt->packet_type)) {
628 ipv4_hdr = rte_pktmbuf_mtod_offset(pkt, struct ipv4_hdr *,
629 sizeof(struct ether_hdr));
631 /* Check to make sure the packet is valid (RFC1812) */
632 if (is_valid_ipv4_pkt(ipv4_hdr, pkt->pkt_len) >= 0) {
634 /* Update time to live and header checksum */
635 --(ipv4_hdr->time_to_live);
636 ++(ipv4_hdr->hdr_checksum);
638 /* Fill acl structure */
639 acl->data_ipv4[acl->num_ipv4] = MBUF_IPV4_2PROTO(pkt);
640 acl->m_ipv4[(acl->num_ipv4)++] = pkt;
643 /* Not a valid IPv4 packet */
644 rte_pktmbuf_free(pkt);
646 } else if (RTE_ETH_IS_IPV6_HDR(pkt->packet_type)) {
647 /* Fill acl structure */
648 acl->data_ipv6[acl->num_ipv6] = MBUF_IPV6_2PROTO(pkt);
649 acl->m_ipv6[(acl->num_ipv6)++] = pkt;
652 /* Unknown type, drop the packet */
653 rte_pktmbuf_free(pkt);
659 prepare_one_packet(struct rte_mbuf **pkts_in, struct acl_search_t *acl,
662 struct rte_mbuf *pkt = pkts_in[index];
664 if (RTE_ETH_IS_IPV4_HDR(pkt->packet_type)) {
665 /* Fill acl structure */
666 acl->data_ipv4[acl->num_ipv4] = MBUF_IPV4_2PROTO(pkt);
667 acl->m_ipv4[(acl->num_ipv4)++] = pkt;
669 } else if (RTE_ETH_IS_IPV6_HDR(pkt->packet_type)) {
670 /* Fill acl structure */
671 acl->data_ipv6[acl->num_ipv6] = MBUF_IPV6_2PROTO(pkt);
672 acl->m_ipv6[(acl->num_ipv6)++] = pkt;
674 /* Unknown type, drop the packet */
675 rte_pktmbuf_free(pkt);
678 #endif /* DO_RFC_1812_CHECKS */
681 prepare_acl_parameter(struct rte_mbuf **pkts_in, struct acl_search_t *acl,
689 /* Prefetch first packets */
690 for (i = 0; i < PREFETCH_OFFSET && i < nb_rx; i++) {
691 rte_prefetch0(rte_pktmbuf_mtod(
692 pkts_in[i], void *));
695 for (i = 0; i < (nb_rx - PREFETCH_OFFSET); i++) {
696 rte_prefetch0(rte_pktmbuf_mtod(pkts_in[
697 i + PREFETCH_OFFSET], void *));
698 prepare_one_packet(pkts_in, acl, i);
701 /* Process left packets */
702 for (; i < nb_rx; i++)
703 prepare_one_packet(pkts_in, acl, i);
707 send_one_packet(struct rte_mbuf *m, uint32_t res)
709 if (likely((res & ACL_DENY_SIGNATURE) == 0 && res != 0)) {
710 /* forward packets */
711 send_single_packet(m,
712 (uint8_t)(res - FWD_PORT_SHIFT));
714 /* in the ACL list, drop it */
715 #ifdef L3FWDACL_DEBUG
716 if ((res & ACL_DENY_SIGNATURE) != 0) {
717 if (RTE_ETH_IS_IPV4_HDR(m->packet_type))
718 dump_acl4_rule(m, res);
719 else if (RTE_ETH_IS_IPV6_HDR(m->packet_type))
720 dump_acl6_rule(m, res);
730 send_packets(struct rte_mbuf **m, uint32_t *res, int num)
734 /* Prefetch first packets */
735 for (i = 0; i < PREFETCH_OFFSET && i < num; i++) {
736 rte_prefetch0(rte_pktmbuf_mtod(
740 for (i = 0; i < (num - PREFETCH_OFFSET); i++) {
741 rte_prefetch0(rte_pktmbuf_mtod(m[
742 i + PREFETCH_OFFSET], void *));
743 send_one_packet(m[i], res[i]);
746 /* Process left packets */
748 send_one_packet(m[i], res[i]);
752 * Parses IPV6 address, exepcts the following format:
753 * XXXX:XXXX:XXXX:XXXX:XXXX:XXXX:XXXX:XXXX (where X - is a hexedecimal digit).
756 parse_ipv6_addr(const char *in, const char **end, uint32_t v[IPV6_ADDR_U32],
759 uint32_t addr[IPV6_ADDR_U16];
761 GET_CB_FIELD(in, addr[0], 16, UINT16_MAX, ':');
762 GET_CB_FIELD(in, addr[1], 16, UINT16_MAX, ':');
763 GET_CB_FIELD(in, addr[2], 16, UINT16_MAX, ':');
764 GET_CB_FIELD(in, addr[3], 16, UINT16_MAX, ':');
765 GET_CB_FIELD(in, addr[4], 16, UINT16_MAX, ':');
766 GET_CB_FIELD(in, addr[5], 16, UINT16_MAX, ':');
767 GET_CB_FIELD(in, addr[6], 16, UINT16_MAX, ':');
768 GET_CB_FIELD(in, addr[7], 16, UINT16_MAX, dlm);
772 v[0] = (addr[0] << 16) + addr[1];
773 v[1] = (addr[2] << 16) + addr[3];
774 v[2] = (addr[4] << 16) + addr[5];
775 v[3] = (addr[6] << 16) + addr[7];
781 parse_ipv6_net(const char *in, struct rte_acl_field field[4])
786 const uint32_t nbu32 = sizeof(uint32_t) * CHAR_BIT;
789 rc = parse_ipv6_addr(in, &mp, v, '/');
794 GET_CB_FIELD(mp, m, 0, CHAR_BIT * sizeof(v), 0);
796 /* put all together. */
797 for (i = 0; i != RTE_DIM(v); i++) {
798 if (m >= (i + 1) * nbu32)
799 field[i].mask_range.u32 = nbu32;
801 field[i].mask_range.u32 = m > (i * nbu32) ?
804 field[i].value.u32 = v[i];
811 parse_cb_ipv6_rule(char *str, struct rte_acl_rule *v, int has_userdata)
814 char *s, *sp, *in[CB_FLD_NUM];
815 static const char *dlm = " \t\n";
816 int dim = has_userdata ? CB_FLD_NUM : CB_FLD_USERDATA;
819 for (i = 0; i != dim; i++, s = NULL) {
820 in[i] = strtok_r(s, dlm, &sp);
825 rc = parse_ipv6_net(in[CB_FLD_SRC_ADDR], v->field + SRC1_FIELD_IPV6);
827 acl_log("failed to read source address/mask: %s\n",
828 in[CB_FLD_SRC_ADDR]);
832 rc = parse_ipv6_net(in[CB_FLD_DST_ADDR], v->field + DST1_FIELD_IPV6);
834 acl_log("failed to read destination address/mask: %s\n",
835 in[CB_FLD_DST_ADDR]);
840 GET_CB_FIELD(in[CB_FLD_SRC_PORT_LOW],
841 v->field[SRCP_FIELD_IPV6].value.u16,
843 GET_CB_FIELD(in[CB_FLD_SRC_PORT_HIGH],
844 v->field[SRCP_FIELD_IPV6].mask_range.u16,
847 if (strncmp(in[CB_FLD_SRC_PORT_DLM], cb_port_delim,
848 sizeof(cb_port_delim)) != 0)
851 /* destination port. */
852 GET_CB_FIELD(in[CB_FLD_DST_PORT_LOW],
853 v->field[DSTP_FIELD_IPV6].value.u16,
855 GET_CB_FIELD(in[CB_FLD_DST_PORT_HIGH],
856 v->field[DSTP_FIELD_IPV6].mask_range.u16,
859 if (strncmp(in[CB_FLD_DST_PORT_DLM], cb_port_delim,
860 sizeof(cb_port_delim)) != 0)
863 if (v->field[SRCP_FIELD_IPV6].mask_range.u16
864 < v->field[SRCP_FIELD_IPV6].value.u16
865 || v->field[DSTP_FIELD_IPV6].mask_range.u16
866 < v->field[DSTP_FIELD_IPV6].value.u16)
869 GET_CB_FIELD(in[CB_FLD_PROTO], v->field[PROTO_FIELD_IPV6].value.u8,
871 GET_CB_FIELD(in[CB_FLD_PROTO], v->field[PROTO_FIELD_IPV6].mask_range.u8,
875 GET_CB_FIELD(in[CB_FLD_USERDATA], v->data.userdata,
882 * Parse ClassBench rules file.
884 * '@'<src_ipv4_addr>'/'<masklen> <space> \
885 * <dst_ipv4_addr>'/'<masklen> <space> \
886 * <src_port_low> <space> ":" <src_port_high> <space> \
887 * <dst_port_low> <space> ":" <dst_port_high> <space> \
891 parse_ipv4_net(const char *in, uint32_t *addr, uint32_t *mask_len)
893 uint8_t a, b, c, d, m;
895 GET_CB_FIELD(in, a, 0, UINT8_MAX, '.');
896 GET_CB_FIELD(in, b, 0, UINT8_MAX, '.');
897 GET_CB_FIELD(in, c, 0, UINT8_MAX, '.');
898 GET_CB_FIELD(in, d, 0, UINT8_MAX, '/');
899 GET_CB_FIELD(in, m, 0, sizeof(uint32_t) * CHAR_BIT, 0);
901 addr[0] = IPv4(a, b, c, d);
908 parse_cb_ipv4vlan_rule(char *str, struct rte_acl_rule *v, int has_userdata)
911 char *s, *sp, *in[CB_FLD_NUM];
912 static const char *dlm = " \t\n";
913 int dim = has_userdata ? CB_FLD_NUM : CB_FLD_USERDATA;
916 for (i = 0; i != dim; i++, s = NULL) {
917 in[i] = strtok_r(s, dlm, &sp);
922 rc = parse_ipv4_net(in[CB_FLD_SRC_ADDR],
923 &v->field[SRC_FIELD_IPV4].value.u32,
924 &v->field[SRC_FIELD_IPV4].mask_range.u32);
926 acl_log("failed to read source address/mask: %s\n",
927 in[CB_FLD_SRC_ADDR]);
931 rc = parse_ipv4_net(in[CB_FLD_DST_ADDR],
932 &v->field[DST_FIELD_IPV4].value.u32,
933 &v->field[DST_FIELD_IPV4].mask_range.u32);
935 acl_log("failed to read destination address/mask: %s\n",
936 in[CB_FLD_DST_ADDR]);
940 GET_CB_FIELD(in[CB_FLD_SRC_PORT_LOW],
941 v->field[SRCP_FIELD_IPV4].value.u16,
943 GET_CB_FIELD(in[CB_FLD_SRC_PORT_HIGH],
944 v->field[SRCP_FIELD_IPV4].mask_range.u16,
947 if (strncmp(in[CB_FLD_SRC_PORT_DLM], cb_port_delim,
948 sizeof(cb_port_delim)) != 0)
951 GET_CB_FIELD(in[CB_FLD_DST_PORT_LOW],
952 v->field[DSTP_FIELD_IPV4].value.u16,
954 GET_CB_FIELD(in[CB_FLD_DST_PORT_HIGH],
955 v->field[DSTP_FIELD_IPV4].mask_range.u16,
958 if (strncmp(in[CB_FLD_DST_PORT_DLM], cb_port_delim,
959 sizeof(cb_port_delim)) != 0)
962 if (v->field[SRCP_FIELD_IPV4].mask_range.u16
963 < v->field[SRCP_FIELD_IPV4].value.u16
964 || v->field[DSTP_FIELD_IPV4].mask_range.u16
965 < v->field[DSTP_FIELD_IPV4].value.u16)
968 GET_CB_FIELD(in[CB_FLD_PROTO], v->field[PROTO_FIELD_IPV4].value.u8,
970 GET_CB_FIELD(in[CB_FLD_PROTO], v->field[PROTO_FIELD_IPV4].mask_range.u8,
974 GET_CB_FIELD(in[CB_FLD_USERDATA], v->data.userdata, 0,
981 add_rules(const char *rule_path,
982 struct rte_acl_rule **proute_base,
983 unsigned int *proute_num,
984 struct rte_acl_rule **pacl_base,
985 unsigned int *pacl_num, uint32_t rule_size,
986 int (*parser)(char *, struct rte_acl_rule*, int))
988 uint8_t *acl_rules, *route_rules;
989 struct rte_acl_rule *next;
990 unsigned int acl_num = 0, route_num = 0, total_num = 0;
991 unsigned int acl_cnt = 0, route_cnt = 0;
993 FILE *fh = fopen(rule_path, "rb");
998 rte_exit(EXIT_FAILURE, "%s: Open %s failed\n", __func__,
1001 while ((fgets(buff, LINE_MAX, fh) != NULL)) {
1002 if (buff[0] == ROUTE_LEAD_CHAR)
1004 else if (buff[0] == ACL_LEAD_CHAR)
1009 rte_exit(EXIT_FAILURE, "Not find any route entries in %s!\n",
1012 val = fseek(fh, 0, SEEK_SET);
1014 rte_exit(EXIT_FAILURE, "%s: File seek operation failed\n",
1018 acl_rules = calloc(acl_num, rule_size);
1020 if (NULL == acl_rules)
1021 rte_exit(EXIT_FAILURE, "%s: failed to malloc memory\n",
1024 route_rules = calloc(route_num, rule_size);
1026 if (NULL == route_rules)
1027 rte_exit(EXIT_FAILURE, "%s: failed to malloc memory\n",
1031 while (fgets(buff, LINE_MAX, fh) != NULL) {
1034 if (is_bypass_line(buff))
1040 if (s == ROUTE_LEAD_CHAR)
1041 next = (struct rte_acl_rule *)(route_rules +
1042 route_cnt * rule_size);
1045 else if (s == ACL_LEAD_CHAR)
1046 next = (struct rte_acl_rule *)(acl_rules +
1047 acl_cnt * rule_size);
1051 rte_exit(EXIT_FAILURE,
1052 "%s Line %u: should start with leading "
1054 rule_path, i, ROUTE_LEAD_CHAR, ACL_LEAD_CHAR);
1056 if (parser(buff + 1, next, s == ROUTE_LEAD_CHAR) != 0)
1057 rte_exit(EXIT_FAILURE,
1058 "%s Line %u: parse rules error\n",
1061 if (s == ROUTE_LEAD_CHAR) {
1062 /* Check the forwarding port number */
1063 if ((enabled_port_mask & (1 << next->data.userdata)) ==
1065 rte_exit(EXIT_FAILURE,
1066 "%s Line %u: fwd number illegal:%u\n",
1067 rule_path, i, next->data.userdata);
1068 next->data.userdata += FWD_PORT_SHIFT;
1071 next->data.userdata = ACL_DENY_SIGNATURE + acl_cnt;
1075 next->data.priority = RTE_ACL_MAX_PRIORITY - total_num;
1076 next->data.category_mask = -1;
1082 *pacl_base = (struct rte_acl_rule *)acl_rules;
1083 *pacl_num = acl_num;
1084 *proute_base = (struct rte_acl_rule *)route_rules;
1085 *proute_num = route_cnt;
1091 dump_acl_config(void)
1093 printf("ACL option are:\n");
1094 printf(OPTION_RULE_IPV4": %s\n", parm_config.rule_ipv4_name);
1095 printf(OPTION_RULE_IPV6": %s\n", parm_config.rule_ipv6_name);
1096 printf(OPTION_SCALAR": %d\n", parm_config.scalar);
1100 check_acl_config(void)
1102 if (parm_config.rule_ipv4_name == NULL) {
1103 acl_log("ACL IPv4 rule file not specified\n");
1105 } else if (parm_config.rule_ipv6_name == NULL) {
1106 acl_log("ACL IPv6 rule file not specified\n");
1113 static struct rte_acl_ctx*
1114 setup_acl(struct rte_acl_rule *route_base,
1115 struct rte_acl_rule *acl_base, unsigned int route_num,
1116 unsigned int acl_num, int ipv6, int socketid)
1118 char name[PATH_MAX];
1119 struct rte_acl_param acl_param;
1120 struct rte_acl_config acl_build_param;
1121 struct rte_acl_ctx *context;
1122 int dim = ipv6 ? RTE_DIM(ipv6_defs) : RTE_DIM(ipv4_defs);
1124 /* Create ACL contexts */
1125 snprintf(name, sizeof(name), "%s%d",
1126 ipv6 ? L3FWD_ACL_IPV6_NAME : L3FWD_ACL_IPV4_NAME,
1129 acl_param.name = name;
1130 acl_param.socket_id = socketid;
1131 acl_param.rule_size = RTE_ACL_RULE_SZ(dim);
1132 acl_param.max_rule_num = MAX_ACL_RULE_NUM;
1134 if ((context = rte_acl_create(&acl_param)) == NULL)
1135 rte_exit(EXIT_FAILURE, "Failed to create ACL context\n");
1137 if (parm_config.scalar && rte_acl_set_ctx_classify(context,
1138 RTE_ACL_CLASSIFY_SCALAR) != 0)
1139 rte_exit(EXIT_FAILURE,
1140 "Failed to setup classify method for ACL context\n");
1142 if (rte_acl_add_rules(context, route_base, route_num) < 0)
1143 rte_exit(EXIT_FAILURE, "add rules failed\n");
1145 if (rte_acl_add_rules(context, acl_base, acl_num) < 0)
1146 rte_exit(EXIT_FAILURE, "add rules failed\n");
1148 /* Perform builds */
1149 memset(&acl_build_param, 0, sizeof(acl_build_param));
1151 acl_build_param.num_categories = DEFAULT_MAX_CATEGORIES;
1152 acl_build_param.num_fields = dim;
1153 memcpy(&acl_build_param.defs, ipv6 ? ipv6_defs : ipv4_defs,
1154 ipv6 ? sizeof(ipv6_defs) : sizeof(ipv4_defs));
1156 if (rte_acl_build(context, &acl_build_param) != 0)
1157 rte_exit(EXIT_FAILURE, "Failed to build ACL trie\n");
1159 rte_acl_dump(context);
1170 struct rte_acl_rule *acl_base_ipv4, *route_base_ipv4,
1171 *acl_base_ipv6, *route_base_ipv6;
1172 unsigned int acl_num_ipv4 = 0, route_num_ipv4 = 0,
1173 acl_num_ipv6 = 0, route_num_ipv6 = 0;
1175 if (check_acl_config() != 0)
1176 rte_exit(EXIT_FAILURE, "Failed to get valid ACL options\n");
1180 /* Load rules from the input file */
1181 if (add_rules(parm_config.rule_ipv4_name, &route_base_ipv4,
1182 &route_num_ipv4, &acl_base_ipv4, &acl_num_ipv4,
1183 sizeof(struct acl4_rule), &parse_cb_ipv4vlan_rule) < 0)
1184 rte_exit(EXIT_FAILURE, "Failed to add rules\n");
1186 acl_log("IPv4 Route entries %u:\n", route_num_ipv4);
1187 dump_ipv4_rules((struct acl4_rule *)route_base_ipv4, route_num_ipv4, 1);
1189 acl_log("IPv4 ACL entries %u:\n", acl_num_ipv4);
1190 dump_ipv4_rules((struct acl4_rule *)acl_base_ipv4, acl_num_ipv4, 1);
1192 if (add_rules(parm_config.rule_ipv6_name, &route_base_ipv6,
1194 &acl_base_ipv6, &acl_num_ipv6,
1195 sizeof(struct acl6_rule), &parse_cb_ipv6_rule) < 0)
1196 rte_exit(EXIT_FAILURE, "Failed to add rules\n");
1198 acl_log("IPv6 Route entries %u:\n", route_num_ipv6);
1199 dump_ipv6_rules((struct acl6_rule *)route_base_ipv6, route_num_ipv6, 1);
1201 acl_log("IPv6 ACL entries %u:\n", acl_num_ipv6);
1202 dump_ipv6_rules((struct acl6_rule *)acl_base_ipv6, acl_num_ipv6, 1);
1204 memset(&acl_config, 0, sizeof(acl_config));
1206 /* Check sockets a context should be created on */
1208 acl_config.mapped[0] = 1;
1210 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1211 if (rte_lcore_is_enabled(lcore_id) == 0)
1214 socketid = rte_lcore_to_socket_id(lcore_id);
1215 if (socketid >= NB_SOCKETS) {
1216 acl_log("Socket %d of lcore %u is out "
1218 socketid, lcore_id, NB_SOCKETS);
1219 free(route_base_ipv4);
1220 free(route_base_ipv6);
1221 free(acl_base_ipv4);
1222 free(acl_base_ipv6);
1226 acl_config.mapped[socketid] = 1;
1230 for (i = 0; i < NB_SOCKETS; i++) {
1231 if (acl_config.mapped[i]) {
1232 acl_config.acx_ipv4[i] = setup_acl(route_base_ipv4,
1233 acl_base_ipv4, route_num_ipv4, acl_num_ipv4,
1236 acl_config.acx_ipv6[i] = setup_acl(route_base_ipv6,
1237 acl_base_ipv6, route_num_ipv6, acl_num_ipv6,
1242 free(route_base_ipv4);
1243 free(route_base_ipv6);
1245 #ifdef L3FWDACL_DEBUG
1246 acl_config.rule_ipv4 = (struct acl4_rule *)acl_base_ipv4;
1247 acl_config.rule_ipv6 = (struct acl6_rule *)acl_base_ipv6;
1249 free(acl_base_ipv4);
1250 free(acl_base_ipv6);
1256 /***********************end of ACL part******************************/
1259 uint16_t n_rx_queue;
1260 struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
1262 uint16_t tx_port_id[RTE_MAX_ETHPORTS];
1263 uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
1264 struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
1265 } __rte_cache_aligned;
1267 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
1269 /* Enqueue a single packet, and send burst if queue is filled */
1271 send_single_packet(struct rte_mbuf *m, uint16_t port)
1274 struct lcore_conf *qconf;
1276 lcore_id = rte_lcore_id();
1278 qconf = &lcore_conf[lcore_id];
1279 rte_eth_tx_buffer(port, qconf->tx_queue_id[port],
1280 qconf->tx_buffer[port], m);
1283 #ifdef DO_RFC_1812_CHECKS
1285 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
1287 /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
1289 * 1. The packet length reported by the Link Layer must be large
1290 * enough to hold the minimum length legal IP datagram (20 bytes).
1292 if (link_len < sizeof(struct ipv4_hdr))
1295 /* 2. The IP checksum must be correct. */
1296 /* this is checked in H/W */
1299 * 3. The IP version number must be 4. If the version number is not 4
1300 * then the packet may be another version of IP, such as IPng or
1303 if (((pkt->version_ihl) >> 4) != 4)
1306 * 4. The IP header length field must be large enough to hold the
1307 * minimum length legal IP datagram (20 bytes = 5 words).
1309 if ((pkt->version_ihl & 0xf) < 5)
1313 * 5. The IP total length field must be large enough to hold the IP
1314 * datagram header, whose length is specified in the IP header length
1317 if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
1324 /* main processing loop */
1326 main_loop(__attribute__((unused)) void *dummy)
1328 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1330 uint64_t prev_tsc, diff_tsc, cur_tsc;
1334 struct lcore_conf *qconf;
1336 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
1337 / US_PER_S * BURST_TX_DRAIN_US;
1340 lcore_id = rte_lcore_id();
1341 qconf = &lcore_conf[lcore_id];
1342 socketid = rte_lcore_to_socket_id(lcore_id);
1344 if (qconf->n_rx_queue == 0) {
1345 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
1349 RTE_LOG(INFO, L3FWD, "entering main loop on lcore %u\n", lcore_id);
1351 for (i = 0; i < qconf->n_rx_queue; i++) {
1353 portid = qconf->rx_queue_list[i].port_id;
1354 queueid = qconf->rx_queue_list[i].queue_id;
1355 RTE_LOG(INFO, L3FWD,
1356 " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
1357 lcore_id, portid, queueid);
1362 cur_tsc = rte_rdtsc();
1365 * TX burst queue drain
1367 diff_tsc = cur_tsc - prev_tsc;
1368 if (unlikely(diff_tsc > drain_tsc)) {
1369 for (i = 0; i < qconf->n_tx_port; ++i) {
1370 portid = qconf->tx_port_id[i];
1371 rte_eth_tx_buffer_flush(portid,
1372 qconf->tx_queue_id[portid],
1373 qconf->tx_buffer[portid]);
1379 * Read packet from RX queues
1381 for (i = 0; i < qconf->n_rx_queue; ++i) {
1383 portid = qconf->rx_queue_list[i].port_id;
1384 queueid = qconf->rx_queue_list[i].queue_id;
1385 nb_rx = rte_eth_rx_burst(portid, queueid,
1386 pkts_burst, MAX_PKT_BURST);
1389 struct acl_search_t acl_search;
1391 prepare_acl_parameter(pkts_burst, &acl_search,
1394 if (acl_search.num_ipv4) {
1396 acl_config.acx_ipv4[socketid],
1397 acl_search.data_ipv4,
1398 acl_search.res_ipv4,
1399 acl_search.num_ipv4,
1400 DEFAULT_MAX_CATEGORIES);
1402 send_packets(acl_search.m_ipv4,
1403 acl_search.res_ipv4,
1404 acl_search.num_ipv4);
1407 if (acl_search.num_ipv6) {
1409 acl_config.acx_ipv6[socketid],
1410 acl_search.data_ipv6,
1411 acl_search.res_ipv6,
1412 acl_search.num_ipv6,
1413 DEFAULT_MAX_CATEGORIES);
1415 send_packets(acl_search.m_ipv6,
1416 acl_search.res_ipv6,
1417 acl_search.num_ipv6);
1425 check_lcore_params(void)
1427 uint8_t queue, lcore;
1431 for (i = 0; i < nb_lcore_params; ++i) {
1432 queue = lcore_params[i].queue_id;
1433 if (queue >= MAX_RX_QUEUE_PER_PORT) {
1434 printf("invalid queue number: %hhu\n", queue);
1437 lcore = lcore_params[i].lcore_id;
1438 if (!rte_lcore_is_enabled(lcore)) {
1439 printf("error: lcore %hhu is not enabled in "
1440 "lcore mask\n", lcore);
1443 socketid = rte_lcore_to_socket_id(lcore);
1444 if (socketid != 0 && numa_on == 0) {
1445 printf("warning: lcore %hhu is on socket %d "
1454 check_port_config(void)
1459 for (i = 0; i < nb_lcore_params; ++i) {
1460 portid = lcore_params[i].port_id;
1462 if ((enabled_port_mask & (1 << portid)) == 0) {
1463 printf("port %u is not enabled in port mask\n", portid);
1466 if (!rte_eth_dev_is_valid_port(portid)) {
1467 printf("port %u is not present on the board\n", portid);
1475 get_port_n_rx_queues(const uint16_t port)
1480 for (i = 0; i < nb_lcore_params; ++i) {
1481 if (lcore_params[i].port_id == port &&
1482 lcore_params[i].queue_id > queue)
1483 queue = lcore_params[i].queue_id;
1485 return (uint8_t)(++queue);
1489 init_lcore_rx_queues(void)
1491 uint16_t i, nb_rx_queue;
1494 for (i = 0; i < nb_lcore_params; ++i) {
1495 lcore = lcore_params[i].lcore_id;
1496 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1497 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1498 printf("error: too many queues (%u) for lcore: %u\n",
1499 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1502 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1503 lcore_params[i].port_id;
1504 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1505 lcore_params[i].queue_id;
1506 lcore_conf[lcore].n_rx_queue++;
1514 print_usage(const char *prgname)
1516 printf("%s [EAL options] -- -p PORTMASK -P"
1517 "--"OPTION_RULE_IPV4"=FILE"
1518 "--"OPTION_RULE_IPV6"=FILE"
1519 " [--"OPTION_CONFIG" (port,queue,lcore)[,(port,queue,lcore]]"
1520 " [--"OPTION_ENBJMO" [--max-pkt-len PKTLEN]]\n"
1521 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1522 " -P : enable promiscuous mode\n"
1523 " --"OPTION_CONFIG": (port,queue,lcore): "
1524 "rx queues configuration\n"
1525 " --"OPTION_NONUMA": optional, disable numa awareness\n"
1526 " --"OPTION_ENBJMO": enable jumbo frame"
1527 " which max packet len is PKTLEN in decimal (64-9600)\n"
1528 " --"OPTION_RULE_IPV4"=FILE: specify the ipv4 rules entries "
1530 "Each rule occupy one line. "
1531 "2 kinds of rules are supported. "
1532 "One is ACL entry at while line leads with character '%c', "
1533 "another is route entry at while line leads with "
1535 " --"OPTION_RULE_IPV6"=FILE: specify the ipv6 rules "
1537 " --"OPTION_SCALAR": Use scalar function to do lookup\n",
1538 prgname, ACL_LEAD_CHAR, ROUTE_LEAD_CHAR);
1542 parse_max_pkt_len(const char *pktlen)
1547 /* parse decimal string */
1548 len = strtoul(pktlen, &end, 10);
1549 if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1559 parse_portmask(const char *portmask)
1564 /* parse hexadecimal string */
1565 pm = strtoul(portmask, &end, 16);
1566 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1576 parse_config(const char *q_arg)
1579 const char *p, *p0 = q_arg;
1587 unsigned long int_fld[_NUM_FLD];
1588 char *str_fld[_NUM_FLD];
1592 nb_lcore_params = 0;
1594 while ((p = strchr(p0, '(')) != NULL) {
1596 if ((p0 = strchr(p, ')')) == NULL)
1600 if (size >= sizeof(s))
1603 snprintf(s, sizeof(s), "%.*s", size, p);
1604 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1607 for (i = 0; i < _NUM_FLD; i++) {
1609 int_fld[i] = strtoul(str_fld[i], &end, 0);
1610 if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
1613 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1614 printf("exceeded max number of lcore params: %hu\n",
1618 lcore_params_array[nb_lcore_params].port_id =
1619 (uint8_t)int_fld[FLD_PORT];
1620 lcore_params_array[nb_lcore_params].queue_id =
1621 (uint8_t)int_fld[FLD_QUEUE];
1622 lcore_params_array[nb_lcore_params].lcore_id =
1623 (uint8_t)int_fld[FLD_LCORE];
1626 lcore_params = lcore_params_array;
1630 /* Parse the argument given in the command line of the application */
1632 parse_args(int argc, char **argv)
1637 char *prgname = argv[0];
1638 static struct option lgopts[] = {
1639 {OPTION_CONFIG, 1, 0, 0},
1640 {OPTION_NONUMA, 0, 0, 0},
1641 {OPTION_ENBJMO, 0, 0, 0},
1642 {OPTION_RULE_IPV4, 1, 0, 0},
1643 {OPTION_RULE_IPV6, 1, 0, 0},
1644 {OPTION_SCALAR, 0, 0, 0},
1650 while ((opt = getopt_long(argc, argvopt, "p:P",
1651 lgopts, &option_index)) != EOF) {
1656 enabled_port_mask = parse_portmask(optarg);
1657 if (enabled_port_mask == 0) {
1658 printf("invalid portmask\n");
1659 print_usage(prgname);
1664 printf("Promiscuous mode selected\n");
1670 if (!strncmp(lgopts[option_index].name,
1672 sizeof(OPTION_CONFIG))) {
1673 ret = parse_config(optarg);
1675 printf("invalid config\n");
1676 print_usage(prgname);
1681 if (!strncmp(lgopts[option_index].name,
1683 sizeof(OPTION_NONUMA))) {
1684 printf("numa is disabled\n");
1688 if (!strncmp(lgopts[option_index].name,
1689 OPTION_ENBJMO, sizeof(OPTION_ENBJMO))) {
1690 struct option lenopts = {
1697 printf("jumbo frame is enabled\n");
1698 port_conf.rxmode.offloads |=
1699 DEV_RX_OFFLOAD_JUMBO_FRAME;
1700 port_conf.txmode.offloads |=
1701 DEV_TX_OFFLOAD_MULTI_SEGS;
1704 * if no max-pkt-len set, then use the
1705 * default value ETHER_MAX_LEN
1707 if (0 == getopt_long(argc, argvopt, "",
1708 &lenopts, &option_index)) {
1709 ret = parse_max_pkt_len(optarg);
1711 (ret > MAX_JUMBO_PKT_LEN)) {
1712 printf("invalid packet "
1714 print_usage(prgname);
1717 port_conf.rxmode.max_rx_pkt_len = ret;
1719 printf("set jumbo frame max packet length "
1722 port_conf.rxmode.max_rx_pkt_len);
1725 if (!strncmp(lgopts[option_index].name,
1727 sizeof(OPTION_RULE_IPV4)))
1728 parm_config.rule_ipv4_name = optarg;
1730 if (!strncmp(lgopts[option_index].name,
1732 sizeof(OPTION_RULE_IPV6))) {
1733 parm_config.rule_ipv6_name = optarg;
1736 if (!strncmp(lgopts[option_index].name,
1737 OPTION_SCALAR, sizeof(OPTION_SCALAR)))
1738 parm_config.scalar = 1;
1744 print_usage(prgname);
1750 argv[optind-1] = prgname;
1753 optind = 1; /* reset getopt lib */
1758 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1760 char buf[ETHER_ADDR_FMT_SIZE];
1761 ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
1762 printf("%s%s", name, buf);
1766 init_mem(unsigned nb_mbuf)
1772 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1773 if (rte_lcore_is_enabled(lcore_id) == 0)
1777 socketid = rte_lcore_to_socket_id(lcore_id);
1781 if (socketid >= NB_SOCKETS) {
1782 rte_exit(EXIT_FAILURE,
1783 "Socket %d of lcore %u is out of range %d\n",
1784 socketid, lcore_id, NB_SOCKETS);
1786 if (pktmbuf_pool[socketid] == NULL) {
1787 snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1788 pktmbuf_pool[socketid] =
1789 rte_pktmbuf_pool_create(s, nb_mbuf,
1790 MEMPOOL_CACHE_SIZE, 0,
1791 RTE_MBUF_DEFAULT_BUF_SIZE,
1793 if (pktmbuf_pool[socketid] == NULL)
1794 rte_exit(EXIT_FAILURE,
1795 "Cannot init mbuf pool on socket %d\n",
1798 printf("Allocated mbuf pool on socket %d\n",
1805 /* Check the link status of all ports in up to 9s, and print them finally */
1807 check_all_ports_link_status(uint32_t port_mask)
1809 #define CHECK_INTERVAL 100 /* 100ms */
1810 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1812 uint8_t count, all_ports_up, print_flag = 0;
1813 struct rte_eth_link link;
1815 printf("\nChecking link status");
1817 for (count = 0; count <= MAX_CHECK_TIME; count++) {
1819 RTE_ETH_FOREACH_DEV(portid) {
1820 if ((port_mask & (1 << portid)) == 0)
1822 memset(&link, 0, sizeof(link));
1823 rte_eth_link_get_nowait(portid, &link);
1824 /* print link status if flag set */
1825 if (print_flag == 1) {
1826 if (link.link_status)
1828 "Port%d Link Up. Speed %u Mbps %s\n",
1829 portid, link.link_speed,
1830 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1831 ("full-duplex") : ("half-duplex\n"));
1833 printf("Port %d Link Down\n", portid);
1836 /* clear all_ports_up flag if any link down */
1837 if (link.link_status == ETH_LINK_DOWN) {
1842 /* after finally printing all link status, get out */
1843 if (print_flag == 1)
1846 if (all_ports_up == 0) {
1849 rte_delay_ms(CHECK_INTERVAL);
1852 /* set the print_flag if all ports up or timeout */
1853 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1861 main(int argc, char **argv)
1863 struct lcore_conf *qconf;
1864 struct rte_eth_dev_info dev_info;
1865 struct rte_eth_txconf *txconf;
1870 uint32_t n_tx_queue, nb_lcores;
1872 uint8_t nb_rx_queue, queue, socketid;
1875 ret = rte_eal_init(argc, argv);
1877 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1881 /* parse application arguments (after the EAL ones) */
1882 ret = parse_args(argc, argv);
1884 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1886 if (check_lcore_params() < 0)
1887 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1889 ret = init_lcore_rx_queues();
1891 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1893 nb_ports = rte_eth_dev_count_avail();
1895 if (check_port_config() < 0)
1896 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1898 /* Add ACL rules and route entries, build trie */
1899 if (app_acl_init() < 0)
1900 rte_exit(EXIT_FAILURE, "app_acl_init failed\n");
1902 nb_lcores = rte_lcore_count();
1904 /* initialize all ports */
1905 RTE_ETH_FOREACH_DEV(portid) {
1906 struct rte_eth_conf local_port_conf = port_conf;
1908 /* skip ports that are not enabled */
1909 if ((enabled_port_mask & (1 << portid)) == 0) {
1910 printf("\nSkipping disabled port %d\n", portid);
1915 printf("Initializing port %d ... ", portid);
1918 nb_rx_queue = get_port_n_rx_queues(portid);
1919 n_tx_queue = nb_lcores;
1920 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
1921 n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1922 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1923 nb_rx_queue, (unsigned)n_tx_queue);
1924 rte_eth_dev_info_get(portid, &dev_info);
1925 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
1926 local_port_conf.txmode.offloads |=
1927 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
1929 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
1930 dev_info.flow_type_rss_offloads;
1931 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
1932 port_conf.rx_adv_conf.rss_conf.rss_hf) {
1933 printf("Port %u modified RSS hash function based on hardware support,"
1934 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
1936 port_conf.rx_adv_conf.rss_conf.rss_hf,
1937 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
1940 ret = rte_eth_dev_configure(portid, nb_rx_queue,
1941 (uint16_t)n_tx_queue, &local_port_conf);
1943 rte_exit(EXIT_FAILURE,
1944 "Cannot configure device: err=%d, port=%d\n",
1947 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
1950 rte_exit(EXIT_FAILURE,
1951 "rte_eth_dev_adjust_nb_rx_tx_desc: err=%d, port=%d\n",
1954 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1955 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1959 ret = init_mem(NB_MBUF);
1961 rte_exit(EXIT_FAILURE, "init_mem failed\n");
1963 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1964 if (rte_lcore_is_enabled(lcore_id) == 0)
1967 /* Initialize TX buffers */
1968 qconf = &lcore_conf[lcore_id];
1969 qconf->tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
1970 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
1971 rte_eth_dev_socket_id(portid));
1972 if (qconf->tx_buffer[portid] == NULL)
1973 rte_exit(EXIT_FAILURE, "Can't allocate tx buffer for port %u\n",
1976 rte_eth_tx_buffer_init(qconf->tx_buffer[portid], MAX_PKT_BURST);
1979 /* init one TX queue per couple (lcore,port) */
1981 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1982 if (rte_lcore_is_enabled(lcore_id) == 0)
1986 socketid = (uint8_t)
1987 rte_lcore_to_socket_id(lcore_id);
1991 printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1994 rte_eth_dev_info_get(portid, &dev_info);
1995 txconf = &dev_info.default_txconf;
1996 txconf->offloads = local_port_conf.txmode.offloads;
1997 ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
2000 rte_exit(EXIT_FAILURE,
2001 "rte_eth_tx_queue_setup: err=%d, "
2002 "port=%d\n", ret, portid);
2004 qconf = &lcore_conf[lcore_id];
2005 qconf->tx_queue_id[portid] = queueid;
2008 qconf->tx_port_id[qconf->n_tx_port] = portid;
2014 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2015 if (rte_lcore_is_enabled(lcore_id) == 0)
2017 qconf = &lcore_conf[lcore_id];
2018 printf("\nInitializing rx queues on lcore %u ... ", lcore_id);
2020 /* init RX queues */
2021 for (queue = 0; queue < qconf->n_rx_queue; ++queue) {
2022 struct rte_eth_dev *dev;
2023 struct rte_eth_conf *conf;
2024 struct rte_eth_rxconf rxq_conf;
2026 portid = qconf->rx_queue_list[queue].port_id;
2027 queueid = qconf->rx_queue_list[queue].queue_id;
2028 dev = &rte_eth_devices[portid];
2029 conf = &dev->data->dev_conf;
2032 socketid = (uint8_t)
2033 rte_lcore_to_socket_id(lcore_id);
2037 printf("rxq=%d,%d,%d ", portid, queueid, socketid);
2040 rte_eth_dev_info_get(portid, &dev_info);
2041 rxq_conf = dev_info.default_rxconf;
2042 rxq_conf.offloads = conf->rxmode.offloads;
2043 ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
2044 socketid, &rxq_conf,
2045 pktmbuf_pool[socketid]);
2047 rte_exit(EXIT_FAILURE,
2048 "rte_eth_rx_queue_setup: err=%d,"
2049 "port=%d\n", ret, portid);
2056 RTE_ETH_FOREACH_DEV(portid) {
2057 if ((enabled_port_mask & (1 << portid)) == 0)
2061 ret = rte_eth_dev_start(portid);
2063 rte_exit(EXIT_FAILURE,
2064 "rte_eth_dev_start: err=%d, port=%d\n",
2068 * If enabled, put device in promiscuous mode.
2069 * This allows IO forwarding mode to forward packets
2070 * to itself through 2 cross-connected ports of the
2074 rte_eth_promiscuous_enable(portid);
2077 check_all_ports_link_status(enabled_port_mask);
2079 /* launch per-lcore init on every lcore */
2080 rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
2081 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
2082 if (rte_eal_wait_lcore(lcore_id) < 0)