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_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 /***********************start of ACL part******************************/
147 #ifdef DO_RFC_1812_CHECKS
149 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len);
152 send_single_packet(struct rte_mbuf *m, uint16_t port);
154 #define MAX_ACL_RULE_NUM 100000
155 #define DEFAULT_MAX_CATEGORIES 1
156 #define L3FWD_ACL_IPV4_NAME "l3fwd-acl-ipv4"
157 #define L3FWD_ACL_IPV6_NAME "l3fwd-acl-ipv6"
158 #define ACL_LEAD_CHAR ('@')
159 #define ROUTE_LEAD_CHAR ('R')
160 #define COMMENT_LEAD_CHAR ('#')
161 #define OPTION_CONFIG "config"
162 #define OPTION_NONUMA "no-numa"
163 #define OPTION_ENBJMO "enable-jumbo"
164 #define OPTION_RULE_IPV4 "rule_ipv4"
165 #define OPTION_RULE_IPV6 "rule_ipv6"
166 #define OPTION_SCALAR "scalar"
167 #define ACL_DENY_SIGNATURE 0xf0000000
168 #define RTE_LOGTYPE_L3FWDACL RTE_LOGTYPE_USER3
169 #define acl_log(format, ...) RTE_LOG(ERR, L3FWDACL, format, ##__VA_ARGS__)
170 #define uint32_t_to_char(ip, a, b, c, d) do {\
171 *a = (unsigned char)(ip >> 24 & 0xff);\
172 *b = (unsigned char)(ip >> 16 & 0xff);\
173 *c = (unsigned char)(ip >> 8 & 0xff);\
174 *d = (unsigned char)(ip & 0xff);\
176 #define OFF_ETHHEAD (sizeof(struct ether_hdr))
177 #define OFF_IPV42PROTO (offsetof(struct ipv4_hdr, next_proto_id))
178 #define OFF_IPV62PROTO (offsetof(struct ipv6_hdr, proto))
179 #define MBUF_IPV4_2PROTO(m) \
180 rte_pktmbuf_mtod_offset((m), uint8_t *, OFF_ETHHEAD + OFF_IPV42PROTO)
181 #define MBUF_IPV6_2PROTO(m) \
182 rte_pktmbuf_mtod_offset((m), uint8_t *, OFF_ETHHEAD + OFF_IPV62PROTO)
184 #define GET_CB_FIELD(in, fd, base, lim, dlm) do { \
188 val = strtoul((in), &end, (base)); \
189 if (errno != 0 || end[0] != (dlm) || val > (lim)) \
191 (fd) = (typeof(fd))val; \
196 * ACL rules should have higher priorities than route ones to ensure ACL rule
197 * always be found when input packets have multi-matches in the database.
198 * A exception case is performance measure, which can define route rules with
199 * higher priority and route rules will always be returned in each lookup.
200 * Reserve range from ACL_RULE_PRIORITY_MAX + 1 to
201 * RTE_ACL_MAX_PRIORITY for route entries in performance measure
203 #define ACL_RULE_PRIORITY_MAX 0x10000000
206 * Forward port info save in ACL lib starts from 1
207 * since ACL assume 0 is invalid.
208 * So, need add 1 when saving and minus 1 when forwarding packets.
210 #define FWD_PORT_SHIFT 1
213 * Rule and trace formats definitions.
226 * That effectively defines order of IPV4VLAN classifications:
228 * - VLAN (TAG and DOMAIN)
231 * - PORTS (SRC and DST)
234 RTE_ACL_IPV4VLAN_PROTO,
235 RTE_ACL_IPV4VLAN_VLAN,
236 RTE_ACL_IPV4VLAN_SRC,
237 RTE_ACL_IPV4VLAN_DST,
238 RTE_ACL_IPV4VLAN_PORTS,
242 struct rte_acl_field_def ipv4_defs[NUM_FIELDS_IPV4] = {
244 .type = RTE_ACL_FIELD_TYPE_BITMASK,
245 .size = sizeof(uint8_t),
246 .field_index = PROTO_FIELD_IPV4,
247 .input_index = RTE_ACL_IPV4VLAN_PROTO,
251 .type = RTE_ACL_FIELD_TYPE_MASK,
252 .size = sizeof(uint32_t),
253 .field_index = SRC_FIELD_IPV4,
254 .input_index = RTE_ACL_IPV4VLAN_SRC,
255 .offset = offsetof(struct ipv4_hdr, src_addr) -
256 offsetof(struct ipv4_hdr, next_proto_id),
259 .type = RTE_ACL_FIELD_TYPE_MASK,
260 .size = sizeof(uint32_t),
261 .field_index = DST_FIELD_IPV4,
262 .input_index = RTE_ACL_IPV4VLAN_DST,
263 .offset = offsetof(struct ipv4_hdr, dst_addr) -
264 offsetof(struct ipv4_hdr, next_proto_id),
267 .type = RTE_ACL_FIELD_TYPE_RANGE,
268 .size = sizeof(uint16_t),
269 .field_index = SRCP_FIELD_IPV4,
270 .input_index = RTE_ACL_IPV4VLAN_PORTS,
271 .offset = sizeof(struct ipv4_hdr) -
272 offsetof(struct ipv4_hdr, next_proto_id),
275 .type = RTE_ACL_FIELD_TYPE_RANGE,
276 .size = sizeof(uint16_t),
277 .field_index = DSTP_FIELD_IPV4,
278 .input_index = RTE_ACL_IPV4VLAN_PORTS,
279 .offset = sizeof(struct ipv4_hdr) -
280 offsetof(struct ipv4_hdr, next_proto_id) +
285 #define IPV6_ADDR_LEN 16
286 #define IPV6_ADDR_U16 (IPV6_ADDR_LEN / sizeof(uint16_t))
287 #define IPV6_ADDR_U32 (IPV6_ADDR_LEN / sizeof(uint32_t))
304 struct rte_acl_field_def ipv6_defs[NUM_FIELDS_IPV6] = {
306 .type = RTE_ACL_FIELD_TYPE_BITMASK,
307 .size = sizeof(uint8_t),
308 .field_index = PROTO_FIELD_IPV6,
309 .input_index = PROTO_FIELD_IPV6,
313 .type = RTE_ACL_FIELD_TYPE_MASK,
314 .size = sizeof(uint32_t),
315 .field_index = SRC1_FIELD_IPV6,
316 .input_index = SRC1_FIELD_IPV6,
317 .offset = offsetof(struct ipv6_hdr, src_addr) -
318 offsetof(struct ipv6_hdr, proto),
321 .type = RTE_ACL_FIELD_TYPE_MASK,
322 .size = sizeof(uint32_t),
323 .field_index = SRC2_FIELD_IPV6,
324 .input_index = SRC2_FIELD_IPV6,
325 .offset = offsetof(struct ipv6_hdr, src_addr) -
326 offsetof(struct ipv6_hdr, proto) + sizeof(uint32_t),
329 .type = RTE_ACL_FIELD_TYPE_MASK,
330 .size = sizeof(uint32_t),
331 .field_index = SRC3_FIELD_IPV6,
332 .input_index = SRC3_FIELD_IPV6,
333 .offset = offsetof(struct ipv6_hdr, src_addr) -
334 offsetof(struct ipv6_hdr, proto) + 2 * sizeof(uint32_t),
337 .type = RTE_ACL_FIELD_TYPE_MASK,
338 .size = sizeof(uint32_t),
339 .field_index = SRC4_FIELD_IPV6,
340 .input_index = SRC4_FIELD_IPV6,
341 .offset = offsetof(struct ipv6_hdr, src_addr) -
342 offsetof(struct ipv6_hdr, proto) + 3 * sizeof(uint32_t),
345 .type = RTE_ACL_FIELD_TYPE_MASK,
346 .size = sizeof(uint32_t),
347 .field_index = DST1_FIELD_IPV6,
348 .input_index = DST1_FIELD_IPV6,
349 .offset = offsetof(struct ipv6_hdr, dst_addr)
350 - offsetof(struct ipv6_hdr, proto),
353 .type = RTE_ACL_FIELD_TYPE_MASK,
354 .size = sizeof(uint32_t),
355 .field_index = DST2_FIELD_IPV6,
356 .input_index = DST2_FIELD_IPV6,
357 .offset = offsetof(struct ipv6_hdr, dst_addr) -
358 offsetof(struct ipv6_hdr, proto) + sizeof(uint32_t),
361 .type = RTE_ACL_FIELD_TYPE_MASK,
362 .size = sizeof(uint32_t),
363 .field_index = DST3_FIELD_IPV6,
364 .input_index = DST3_FIELD_IPV6,
365 .offset = offsetof(struct ipv6_hdr, dst_addr) -
366 offsetof(struct ipv6_hdr, proto) + 2 * sizeof(uint32_t),
369 .type = RTE_ACL_FIELD_TYPE_MASK,
370 .size = sizeof(uint32_t),
371 .field_index = DST4_FIELD_IPV6,
372 .input_index = DST4_FIELD_IPV6,
373 .offset = offsetof(struct ipv6_hdr, dst_addr) -
374 offsetof(struct ipv6_hdr, proto) + 3 * sizeof(uint32_t),
377 .type = RTE_ACL_FIELD_TYPE_RANGE,
378 .size = sizeof(uint16_t),
379 .field_index = SRCP_FIELD_IPV6,
380 .input_index = SRCP_FIELD_IPV6,
381 .offset = sizeof(struct ipv6_hdr) -
382 offsetof(struct ipv6_hdr, proto),
385 .type = RTE_ACL_FIELD_TYPE_RANGE,
386 .size = sizeof(uint16_t),
387 .field_index = DSTP_FIELD_IPV6,
388 .input_index = SRCP_FIELD_IPV6,
389 .offset = sizeof(struct ipv6_hdr) -
390 offsetof(struct ipv6_hdr, proto) + sizeof(uint16_t),
399 CB_FLD_SRC_PORT_HIGH,
402 CB_FLD_DST_PORT_HIGH,
408 RTE_ACL_RULE_DEF(acl4_rule, RTE_DIM(ipv4_defs));
409 RTE_ACL_RULE_DEF(acl6_rule, RTE_DIM(ipv6_defs));
411 struct acl_search_t {
412 const uint8_t *data_ipv4[MAX_PKT_BURST];
413 struct rte_mbuf *m_ipv4[MAX_PKT_BURST];
414 uint32_t res_ipv4[MAX_PKT_BURST];
417 const uint8_t *data_ipv6[MAX_PKT_BURST];
418 struct rte_mbuf *m_ipv6[MAX_PKT_BURST];
419 uint32_t res_ipv6[MAX_PKT_BURST];
424 char mapped[NB_SOCKETS];
425 struct rte_acl_ctx *acx_ipv4[NB_SOCKETS];
426 struct rte_acl_ctx *acx_ipv6[NB_SOCKETS];
427 #ifdef L3FWDACL_DEBUG
428 struct acl4_rule *rule_ipv4;
429 struct acl6_rule *rule_ipv6;
434 const char *rule_ipv4_name;
435 const char *rule_ipv6_name;
439 const char cb_port_delim[] = ":";
442 print_one_ipv4_rule(struct acl4_rule *rule, int extra)
444 unsigned char a, b, c, d;
446 uint32_t_to_char(rule->field[SRC_FIELD_IPV4].value.u32,
448 printf("%hhu.%hhu.%hhu.%hhu/%u ", a, b, c, d,
449 rule->field[SRC_FIELD_IPV4].mask_range.u32);
450 uint32_t_to_char(rule->field[DST_FIELD_IPV4].value.u32,
452 printf("%hhu.%hhu.%hhu.%hhu/%u ", a, b, c, d,
453 rule->field[DST_FIELD_IPV4].mask_range.u32);
454 printf("%hu : %hu %hu : %hu 0x%hhx/0x%hhx ",
455 rule->field[SRCP_FIELD_IPV4].value.u16,
456 rule->field[SRCP_FIELD_IPV4].mask_range.u16,
457 rule->field[DSTP_FIELD_IPV4].value.u16,
458 rule->field[DSTP_FIELD_IPV4].mask_range.u16,
459 rule->field[PROTO_FIELD_IPV4].value.u8,
460 rule->field[PROTO_FIELD_IPV4].mask_range.u8);
462 printf("0x%x-0x%x-0x%x ",
463 rule->data.category_mask,
465 rule->data.userdata);
469 print_one_ipv6_rule(struct acl6_rule *rule, int extra)
471 unsigned char a, b, c, d;
473 uint32_t_to_char(rule->field[SRC1_FIELD_IPV6].value.u32,
475 printf("%.2x%.2x:%.2x%.2x", a, b, c, d);
476 uint32_t_to_char(rule->field[SRC2_FIELD_IPV6].value.u32,
478 printf(":%.2x%.2x:%.2x%.2x", a, b, c, d);
479 uint32_t_to_char(rule->field[SRC3_FIELD_IPV6].value.u32,
481 printf(":%.2x%.2x:%.2x%.2x", a, b, c, d);
482 uint32_t_to_char(rule->field[SRC4_FIELD_IPV6].value.u32,
484 printf(":%.2x%.2x:%.2x%.2x/%u ", a, b, c, d,
485 rule->field[SRC1_FIELD_IPV6].mask_range.u32
486 + rule->field[SRC2_FIELD_IPV6].mask_range.u32
487 + rule->field[SRC3_FIELD_IPV6].mask_range.u32
488 + rule->field[SRC4_FIELD_IPV6].mask_range.u32);
490 uint32_t_to_char(rule->field[DST1_FIELD_IPV6].value.u32,
492 printf("%.2x%.2x:%.2x%.2x", a, b, c, d);
493 uint32_t_to_char(rule->field[DST2_FIELD_IPV6].value.u32,
495 printf(":%.2x%.2x:%.2x%.2x", a, b, c, d);
496 uint32_t_to_char(rule->field[DST3_FIELD_IPV6].value.u32,
498 printf(":%.2x%.2x:%.2x%.2x", a, b, c, d);
499 uint32_t_to_char(rule->field[DST4_FIELD_IPV6].value.u32,
501 printf(":%.2x%.2x:%.2x%.2x/%u ", a, b, c, d,
502 rule->field[DST1_FIELD_IPV6].mask_range.u32
503 + rule->field[DST2_FIELD_IPV6].mask_range.u32
504 + rule->field[DST3_FIELD_IPV6].mask_range.u32
505 + rule->field[DST4_FIELD_IPV6].mask_range.u32);
507 printf("%hu : %hu %hu : %hu 0x%hhx/0x%hhx ",
508 rule->field[SRCP_FIELD_IPV6].value.u16,
509 rule->field[SRCP_FIELD_IPV6].mask_range.u16,
510 rule->field[DSTP_FIELD_IPV6].value.u16,
511 rule->field[DSTP_FIELD_IPV6].mask_range.u16,
512 rule->field[PROTO_FIELD_IPV6].value.u8,
513 rule->field[PROTO_FIELD_IPV6].mask_range.u8);
515 printf("0x%x-0x%x-0x%x ",
516 rule->data.category_mask,
518 rule->data.userdata);
521 /* Bypass comment and empty lines */
523 is_bypass_line(char *buff)
528 if (buff[0] == COMMENT_LEAD_CHAR)
531 while (buff[i] != '\0') {
532 if (!isspace(buff[i]))
539 #ifdef L3FWDACL_DEBUG
541 dump_acl4_rule(struct rte_mbuf *m, uint32_t sig)
543 uint32_t offset = sig & ~ACL_DENY_SIGNATURE;
544 unsigned char a, b, c, d;
545 struct ipv4_hdr *ipv4_hdr = rte_pktmbuf_mtod_offset(m,
547 sizeof(struct ether_hdr));
549 uint32_t_to_char(rte_bswap32(ipv4_hdr->src_addr), &a, &b, &c, &d);
550 printf("Packet Src:%hhu.%hhu.%hhu.%hhu ", a, b, c, d);
551 uint32_t_to_char(rte_bswap32(ipv4_hdr->dst_addr), &a, &b, &c, &d);
552 printf("Dst:%hhu.%hhu.%hhu.%hhu ", a, b, c, d);
554 printf("Src port:%hu,Dst port:%hu ",
555 rte_bswap16(*(uint16_t *)(ipv4_hdr + 1)),
556 rte_bswap16(*((uint16_t *)(ipv4_hdr + 1) + 1)));
557 printf("hit ACL %d - ", offset);
559 print_one_ipv4_rule(acl_config.rule_ipv4 + offset, 1);
565 dump_acl6_rule(struct rte_mbuf *m, uint32_t sig)
568 uint32_t offset = sig & ~ACL_DENY_SIGNATURE;
569 struct ipv6_hdr *ipv6_hdr = rte_pktmbuf_mtod_offset(m,
571 sizeof(struct ether_hdr));
573 printf("Packet Src");
574 for (i = 0; i < RTE_DIM(ipv6_hdr->src_addr); i += sizeof(uint16_t))
576 ipv6_hdr->src_addr[i], ipv6_hdr->src_addr[i + 1]);
579 for (i = 0; i < RTE_DIM(ipv6_hdr->dst_addr); i += sizeof(uint16_t))
581 ipv6_hdr->dst_addr[i], ipv6_hdr->dst_addr[i + 1]);
583 printf("\nSrc port:%hu,Dst port:%hu ",
584 rte_bswap16(*(uint16_t *)(ipv6_hdr + 1)),
585 rte_bswap16(*((uint16_t *)(ipv6_hdr + 1) + 1)));
586 printf("hit ACL %d - ", offset);
588 print_one_ipv6_rule(acl_config.rule_ipv6 + offset, 1);
592 #endif /* L3FWDACL_DEBUG */
595 dump_ipv4_rules(struct acl4_rule *rule, int num, int extra)
599 for (i = 0; i < num; i++, rule++) {
600 printf("\t%d:", i + 1);
601 print_one_ipv4_rule(rule, extra);
607 dump_ipv6_rules(struct acl6_rule *rule, int num, int extra)
611 for (i = 0; i < num; i++, rule++) {
612 printf("\t%d:", i + 1);
613 print_one_ipv6_rule(rule, extra);
618 #ifdef DO_RFC_1812_CHECKS
620 prepare_one_packet(struct rte_mbuf **pkts_in, struct acl_search_t *acl,
623 struct ipv4_hdr *ipv4_hdr;
624 struct rte_mbuf *pkt = pkts_in[index];
626 if (RTE_ETH_IS_IPV4_HDR(pkt->packet_type)) {
627 ipv4_hdr = rte_pktmbuf_mtod_offset(pkt, struct ipv4_hdr *,
628 sizeof(struct ether_hdr));
630 /* Check to make sure the packet is valid (RFC1812) */
631 if (is_valid_ipv4_pkt(ipv4_hdr, pkt->pkt_len) >= 0) {
633 /* Update time to live and header checksum */
634 --(ipv4_hdr->time_to_live);
635 ++(ipv4_hdr->hdr_checksum);
637 /* Fill acl structure */
638 acl->data_ipv4[acl->num_ipv4] = MBUF_IPV4_2PROTO(pkt);
639 acl->m_ipv4[(acl->num_ipv4)++] = pkt;
642 /* Not a valid IPv4 packet */
643 rte_pktmbuf_free(pkt);
645 } else if (RTE_ETH_IS_IPV6_HDR(pkt->packet_type)) {
646 /* Fill acl structure */
647 acl->data_ipv6[acl->num_ipv6] = MBUF_IPV6_2PROTO(pkt);
648 acl->m_ipv6[(acl->num_ipv6)++] = pkt;
651 /* Unknown type, drop the packet */
652 rte_pktmbuf_free(pkt);
658 prepare_one_packet(struct rte_mbuf **pkts_in, struct acl_search_t *acl,
661 struct rte_mbuf *pkt = pkts_in[index];
663 if (RTE_ETH_IS_IPV4_HDR(pkt->packet_type)) {
664 /* Fill acl structure */
665 acl->data_ipv4[acl->num_ipv4] = MBUF_IPV4_2PROTO(pkt);
666 acl->m_ipv4[(acl->num_ipv4)++] = pkt;
668 } else if (RTE_ETH_IS_IPV6_HDR(pkt->packet_type)) {
669 /* Fill acl structure */
670 acl->data_ipv6[acl->num_ipv6] = MBUF_IPV6_2PROTO(pkt);
671 acl->m_ipv6[(acl->num_ipv6)++] = pkt;
673 /* Unknown type, drop the packet */
674 rte_pktmbuf_free(pkt);
677 #endif /* DO_RFC_1812_CHECKS */
680 prepare_acl_parameter(struct rte_mbuf **pkts_in, struct acl_search_t *acl,
688 /* Prefetch first packets */
689 for (i = 0; i < PREFETCH_OFFSET && i < nb_rx; i++) {
690 rte_prefetch0(rte_pktmbuf_mtod(
691 pkts_in[i], void *));
694 for (i = 0; i < (nb_rx - PREFETCH_OFFSET); i++) {
695 rte_prefetch0(rte_pktmbuf_mtod(pkts_in[
696 i + PREFETCH_OFFSET], void *));
697 prepare_one_packet(pkts_in, acl, i);
700 /* Process left packets */
701 for (; i < nb_rx; i++)
702 prepare_one_packet(pkts_in, acl, i);
706 send_one_packet(struct rte_mbuf *m, uint32_t res)
708 if (likely((res & ACL_DENY_SIGNATURE) == 0 && res != 0)) {
709 /* forward packets */
710 send_single_packet(m,
711 (uint8_t)(res - FWD_PORT_SHIFT));
713 /* in the ACL list, drop it */
714 #ifdef L3FWDACL_DEBUG
715 if ((res & ACL_DENY_SIGNATURE) != 0) {
716 if (RTE_ETH_IS_IPV4_HDR(m->packet_type))
717 dump_acl4_rule(m, res);
718 else if (RTE_ETH_IS_IPV6_HDR(m->packet_type))
719 dump_acl6_rule(m, res);
729 send_packets(struct rte_mbuf **m, uint32_t *res, int num)
733 /* Prefetch first packets */
734 for (i = 0; i < PREFETCH_OFFSET && i < num; i++) {
735 rte_prefetch0(rte_pktmbuf_mtod(
739 for (i = 0; i < (num - PREFETCH_OFFSET); i++) {
740 rte_prefetch0(rte_pktmbuf_mtod(m[
741 i + PREFETCH_OFFSET], void *));
742 send_one_packet(m[i], res[i]);
745 /* Process left packets */
747 send_one_packet(m[i], res[i]);
751 * Parses IPV6 address, exepcts the following format:
752 * XXXX:XXXX:XXXX:XXXX:XXXX:XXXX:XXXX:XXXX (where X - is a hexedecimal digit).
755 parse_ipv6_addr(const char *in, const char **end, uint32_t v[IPV6_ADDR_U32],
758 uint32_t addr[IPV6_ADDR_U16];
760 GET_CB_FIELD(in, addr[0], 16, UINT16_MAX, ':');
761 GET_CB_FIELD(in, addr[1], 16, UINT16_MAX, ':');
762 GET_CB_FIELD(in, addr[2], 16, UINT16_MAX, ':');
763 GET_CB_FIELD(in, addr[3], 16, UINT16_MAX, ':');
764 GET_CB_FIELD(in, addr[4], 16, UINT16_MAX, ':');
765 GET_CB_FIELD(in, addr[5], 16, UINT16_MAX, ':');
766 GET_CB_FIELD(in, addr[6], 16, UINT16_MAX, ':');
767 GET_CB_FIELD(in, addr[7], 16, UINT16_MAX, dlm);
771 v[0] = (addr[0] << 16) + addr[1];
772 v[1] = (addr[2] << 16) + addr[3];
773 v[2] = (addr[4] << 16) + addr[5];
774 v[3] = (addr[6] << 16) + addr[7];
780 parse_ipv6_net(const char *in, struct rte_acl_field field[4])
785 const uint32_t nbu32 = sizeof(uint32_t) * CHAR_BIT;
788 rc = parse_ipv6_addr(in, &mp, v, '/');
793 GET_CB_FIELD(mp, m, 0, CHAR_BIT * sizeof(v), 0);
795 /* put all together. */
796 for (i = 0; i != RTE_DIM(v); i++) {
797 if (m >= (i + 1) * nbu32)
798 field[i].mask_range.u32 = nbu32;
800 field[i].mask_range.u32 = m > (i * nbu32) ?
803 field[i].value.u32 = v[i];
810 parse_cb_ipv6_rule(char *str, struct rte_acl_rule *v, int has_userdata)
813 char *s, *sp, *in[CB_FLD_NUM];
814 static const char *dlm = " \t\n";
815 int dim = has_userdata ? CB_FLD_NUM : CB_FLD_USERDATA;
818 for (i = 0; i != dim; i++, s = NULL) {
819 in[i] = strtok_r(s, dlm, &sp);
824 rc = parse_ipv6_net(in[CB_FLD_SRC_ADDR], v->field + SRC1_FIELD_IPV6);
826 acl_log("failed to read source address/mask: %s\n",
827 in[CB_FLD_SRC_ADDR]);
831 rc = parse_ipv6_net(in[CB_FLD_DST_ADDR], v->field + DST1_FIELD_IPV6);
833 acl_log("failed to read destination address/mask: %s\n",
834 in[CB_FLD_DST_ADDR]);
839 GET_CB_FIELD(in[CB_FLD_SRC_PORT_LOW],
840 v->field[SRCP_FIELD_IPV6].value.u16,
842 GET_CB_FIELD(in[CB_FLD_SRC_PORT_HIGH],
843 v->field[SRCP_FIELD_IPV6].mask_range.u16,
846 if (strncmp(in[CB_FLD_SRC_PORT_DLM], cb_port_delim,
847 sizeof(cb_port_delim)) != 0)
850 /* destination port. */
851 GET_CB_FIELD(in[CB_FLD_DST_PORT_LOW],
852 v->field[DSTP_FIELD_IPV6].value.u16,
854 GET_CB_FIELD(in[CB_FLD_DST_PORT_HIGH],
855 v->field[DSTP_FIELD_IPV6].mask_range.u16,
858 if (strncmp(in[CB_FLD_DST_PORT_DLM], cb_port_delim,
859 sizeof(cb_port_delim)) != 0)
862 if (v->field[SRCP_FIELD_IPV6].mask_range.u16
863 < v->field[SRCP_FIELD_IPV6].value.u16
864 || v->field[DSTP_FIELD_IPV6].mask_range.u16
865 < v->field[DSTP_FIELD_IPV6].value.u16)
868 GET_CB_FIELD(in[CB_FLD_PROTO], v->field[PROTO_FIELD_IPV6].value.u8,
870 GET_CB_FIELD(in[CB_FLD_PROTO], v->field[PROTO_FIELD_IPV6].mask_range.u8,
874 GET_CB_FIELD(in[CB_FLD_USERDATA], v->data.userdata,
881 * Parse ClassBench rules file.
883 * '@'<src_ipv4_addr>'/'<masklen> <space> \
884 * <dst_ipv4_addr>'/'<masklen> <space> \
885 * <src_port_low> <space> ":" <src_port_high> <space> \
886 * <dst_port_low> <space> ":" <dst_port_high> <space> \
890 parse_ipv4_net(const char *in, uint32_t *addr, uint32_t *mask_len)
892 uint8_t a, b, c, d, m;
894 GET_CB_FIELD(in, a, 0, UINT8_MAX, '.');
895 GET_CB_FIELD(in, b, 0, UINT8_MAX, '.');
896 GET_CB_FIELD(in, c, 0, UINT8_MAX, '.');
897 GET_CB_FIELD(in, d, 0, UINT8_MAX, '/');
898 GET_CB_FIELD(in, m, 0, sizeof(uint32_t) * CHAR_BIT, 0);
900 addr[0] = IPv4(a, b, c, d);
907 parse_cb_ipv4vlan_rule(char *str, struct rte_acl_rule *v, int has_userdata)
910 char *s, *sp, *in[CB_FLD_NUM];
911 static const char *dlm = " \t\n";
912 int dim = has_userdata ? CB_FLD_NUM : CB_FLD_USERDATA;
915 for (i = 0; i != dim; i++, s = NULL) {
916 in[i] = strtok_r(s, dlm, &sp);
921 rc = parse_ipv4_net(in[CB_FLD_SRC_ADDR],
922 &v->field[SRC_FIELD_IPV4].value.u32,
923 &v->field[SRC_FIELD_IPV4].mask_range.u32);
925 acl_log("failed to read source address/mask: %s\n",
926 in[CB_FLD_SRC_ADDR]);
930 rc = parse_ipv4_net(in[CB_FLD_DST_ADDR],
931 &v->field[DST_FIELD_IPV4].value.u32,
932 &v->field[DST_FIELD_IPV4].mask_range.u32);
934 acl_log("failed to read destination address/mask: %s\n",
935 in[CB_FLD_DST_ADDR]);
939 GET_CB_FIELD(in[CB_FLD_SRC_PORT_LOW],
940 v->field[SRCP_FIELD_IPV4].value.u16,
942 GET_CB_FIELD(in[CB_FLD_SRC_PORT_HIGH],
943 v->field[SRCP_FIELD_IPV4].mask_range.u16,
946 if (strncmp(in[CB_FLD_SRC_PORT_DLM], cb_port_delim,
947 sizeof(cb_port_delim)) != 0)
950 GET_CB_FIELD(in[CB_FLD_DST_PORT_LOW],
951 v->field[DSTP_FIELD_IPV4].value.u16,
953 GET_CB_FIELD(in[CB_FLD_DST_PORT_HIGH],
954 v->field[DSTP_FIELD_IPV4].mask_range.u16,
957 if (strncmp(in[CB_FLD_DST_PORT_DLM], cb_port_delim,
958 sizeof(cb_port_delim)) != 0)
961 if (v->field[SRCP_FIELD_IPV4].mask_range.u16
962 < v->field[SRCP_FIELD_IPV4].value.u16
963 || v->field[DSTP_FIELD_IPV4].mask_range.u16
964 < v->field[DSTP_FIELD_IPV4].value.u16)
967 GET_CB_FIELD(in[CB_FLD_PROTO], v->field[PROTO_FIELD_IPV4].value.u8,
969 GET_CB_FIELD(in[CB_FLD_PROTO], v->field[PROTO_FIELD_IPV4].mask_range.u8,
973 GET_CB_FIELD(in[CB_FLD_USERDATA], v->data.userdata, 0,
980 add_rules(const char *rule_path,
981 struct rte_acl_rule **proute_base,
982 unsigned int *proute_num,
983 struct rte_acl_rule **pacl_base,
984 unsigned int *pacl_num, uint32_t rule_size,
985 int (*parser)(char *, struct rte_acl_rule*, int))
987 uint8_t *acl_rules, *route_rules;
988 struct rte_acl_rule *next;
989 unsigned int acl_num = 0, route_num = 0, total_num = 0;
990 unsigned int acl_cnt = 0, route_cnt = 0;
992 FILE *fh = fopen(rule_path, "rb");
997 rte_exit(EXIT_FAILURE, "%s: Open %s failed\n", __func__,
1000 while ((fgets(buff, LINE_MAX, fh) != NULL)) {
1001 if (buff[0] == ROUTE_LEAD_CHAR)
1003 else if (buff[0] == ACL_LEAD_CHAR)
1008 rte_exit(EXIT_FAILURE, "Not find any route entries in %s!\n",
1011 val = fseek(fh, 0, SEEK_SET);
1013 rte_exit(EXIT_FAILURE, "%s: File seek operation failed\n",
1017 acl_rules = calloc(acl_num, rule_size);
1019 if (NULL == acl_rules)
1020 rte_exit(EXIT_FAILURE, "%s: failed to malloc memory\n",
1023 route_rules = calloc(route_num, rule_size);
1025 if (NULL == route_rules)
1026 rte_exit(EXIT_FAILURE, "%s: failed to malloc memory\n",
1030 while (fgets(buff, LINE_MAX, fh) != NULL) {
1033 if (is_bypass_line(buff))
1039 if (s == ROUTE_LEAD_CHAR)
1040 next = (struct rte_acl_rule *)(route_rules +
1041 route_cnt * rule_size);
1044 else if (s == ACL_LEAD_CHAR)
1045 next = (struct rte_acl_rule *)(acl_rules +
1046 acl_cnt * rule_size);
1050 rte_exit(EXIT_FAILURE,
1051 "%s Line %u: should start with leading "
1053 rule_path, i, ROUTE_LEAD_CHAR, ACL_LEAD_CHAR);
1055 if (parser(buff + 1, next, s == ROUTE_LEAD_CHAR) != 0)
1056 rte_exit(EXIT_FAILURE,
1057 "%s Line %u: parse rules error\n",
1060 if (s == ROUTE_LEAD_CHAR) {
1061 /* Check the forwarding port number */
1062 if ((enabled_port_mask & (1 << next->data.userdata)) ==
1064 rte_exit(EXIT_FAILURE,
1065 "%s Line %u: fwd number illegal:%u\n",
1066 rule_path, i, next->data.userdata);
1067 next->data.userdata += FWD_PORT_SHIFT;
1070 next->data.userdata = ACL_DENY_SIGNATURE + acl_cnt;
1074 next->data.priority = RTE_ACL_MAX_PRIORITY - total_num;
1075 next->data.category_mask = -1;
1081 *pacl_base = (struct rte_acl_rule *)acl_rules;
1082 *pacl_num = acl_num;
1083 *proute_base = (struct rte_acl_rule *)route_rules;
1084 *proute_num = route_cnt;
1090 dump_acl_config(void)
1092 printf("ACL option are:\n");
1093 printf(OPTION_RULE_IPV4": %s\n", parm_config.rule_ipv4_name);
1094 printf(OPTION_RULE_IPV6": %s\n", parm_config.rule_ipv6_name);
1095 printf(OPTION_SCALAR": %d\n", parm_config.scalar);
1099 check_acl_config(void)
1101 if (parm_config.rule_ipv4_name == NULL) {
1102 acl_log("ACL IPv4 rule file not specified\n");
1104 } else if (parm_config.rule_ipv6_name == NULL) {
1105 acl_log("ACL IPv6 rule file not specified\n");
1112 static struct rte_acl_ctx*
1113 setup_acl(struct rte_acl_rule *route_base,
1114 struct rte_acl_rule *acl_base, unsigned int route_num,
1115 unsigned int acl_num, int ipv6, int socketid)
1117 char name[PATH_MAX];
1118 struct rte_acl_param acl_param;
1119 struct rte_acl_config acl_build_param;
1120 struct rte_acl_ctx *context;
1121 int dim = ipv6 ? RTE_DIM(ipv6_defs) : RTE_DIM(ipv4_defs);
1123 /* Create ACL contexts */
1124 snprintf(name, sizeof(name), "%s%d",
1125 ipv6 ? L3FWD_ACL_IPV6_NAME : L3FWD_ACL_IPV4_NAME,
1128 acl_param.name = name;
1129 acl_param.socket_id = socketid;
1130 acl_param.rule_size = RTE_ACL_RULE_SZ(dim);
1131 acl_param.max_rule_num = MAX_ACL_RULE_NUM;
1133 if ((context = rte_acl_create(&acl_param)) == NULL)
1134 rte_exit(EXIT_FAILURE, "Failed to create ACL context\n");
1136 if (parm_config.scalar && rte_acl_set_ctx_classify(context,
1137 RTE_ACL_CLASSIFY_SCALAR) != 0)
1138 rte_exit(EXIT_FAILURE,
1139 "Failed to setup classify method for ACL context\n");
1141 if (rte_acl_add_rules(context, route_base, route_num) < 0)
1142 rte_exit(EXIT_FAILURE, "add rules failed\n");
1144 if (rte_acl_add_rules(context, acl_base, acl_num) < 0)
1145 rte_exit(EXIT_FAILURE, "add rules failed\n");
1147 /* Perform builds */
1148 memset(&acl_build_param, 0, sizeof(acl_build_param));
1150 acl_build_param.num_categories = DEFAULT_MAX_CATEGORIES;
1151 acl_build_param.num_fields = dim;
1152 memcpy(&acl_build_param.defs, ipv6 ? ipv6_defs : ipv4_defs,
1153 ipv6 ? sizeof(ipv6_defs) : sizeof(ipv4_defs));
1155 if (rte_acl_build(context, &acl_build_param) != 0)
1156 rte_exit(EXIT_FAILURE, "Failed to build ACL trie\n");
1158 rte_acl_dump(context);
1169 struct rte_acl_rule *acl_base_ipv4, *route_base_ipv4,
1170 *acl_base_ipv6, *route_base_ipv6;
1171 unsigned int acl_num_ipv4 = 0, route_num_ipv4 = 0,
1172 acl_num_ipv6 = 0, route_num_ipv6 = 0;
1174 if (check_acl_config() != 0)
1175 rte_exit(EXIT_FAILURE, "Failed to get valid ACL options\n");
1179 /* Load rules from the input file */
1180 if (add_rules(parm_config.rule_ipv4_name, &route_base_ipv4,
1181 &route_num_ipv4, &acl_base_ipv4, &acl_num_ipv4,
1182 sizeof(struct acl4_rule), &parse_cb_ipv4vlan_rule) < 0)
1183 rte_exit(EXIT_FAILURE, "Failed to add rules\n");
1185 acl_log("IPv4 Route entries %u:\n", route_num_ipv4);
1186 dump_ipv4_rules((struct acl4_rule *)route_base_ipv4, route_num_ipv4, 1);
1188 acl_log("IPv4 ACL entries %u:\n", acl_num_ipv4);
1189 dump_ipv4_rules((struct acl4_rule *)acl_base_ipv4, acl_num_ipv4, 1);
1191 if (add_rules(parm_config.rule_ipv6_name, &route_base_ipv6,
1193 &acl_base_ipv6, &acl_num_ipv6,
1194 sizeof(struct acl6_rule), &parse_cb_ipv6_rule) < 0)
1195 rte_exit(EXIT_FAILURE, "Failed to add rules\n");
1197 acl_log("IPv6 Route entries %u:\n", route_num_ipv6);
1198 dump_ipv6_rules((struct acl6_rule *)route_base_ipv6, route_num_ipv6, 1);
1200 acl_log("IPv6 ACL entries %u:\n", acl_num_ipv6);
1201 dump_ipv6_rules((struct acl6_rule *)acl_base_ipv6, acl_num_ipv6, 1);
1203 memset(&acl_config, 0, sizeof(acl_config));
1205 /* Check sockets a context should be created on */
1207 acl_config.mapped[0] = 1;
1209 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1210 if (rte_lcore_is_enabled(lcore_id) == 0)
1213 socketid = rte_lcore_to_socket_id(lcore_id);
1214 if (socketid >= NB_SOCKETS) {
1215 acl_log("Socket %d of lcore %u is out "
1217 socketid, lcore_id, NB_SOCKETS);
1218 free(route_base_ipv4);
1219 free(route_base_ipv6);
1220 free(acl_base_ipv4);
1221 free(acl_base_ipv6);
1225 acl_config.mapped[socketid] = 1;
1229 for (i = 0; i < NB_SOCKETS; i++) {
1230 if (acl_config.mapped[i]) {
1231 acl_config.acx_ipv4[i] = setup_acl(route_base_ipv4,
1232 acl_base_ipv4, route_num_ipv4, acl_num_ipv4,
1235 acl_config.acx_ipv6[i] = setup_acl(route_base_ipv6,
1236 acl_base_ipv6, route_num_ipv6, acl_num_ipv6,
1241 free(route_base_ipv4);
1242 free(route_base_ipv6);
1244 #ifdef L3FWDACL_DEBUG
1245 acl_config.rule_ipv4 = (struct acl4_rule *)acl_base_ipv4;
1246 acl_config.rule_ipv6 = (struct acl6_rule *)acl_base_ipv6;
1248 free(acl_base_ipv4);
1249 free(acl_base_ipv6);
1255 /***********************end of ACL part******************************/
1258 uint16_t n_rx_queue;
1259 struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
1261 uint16_t tx_port_id[RTE_MAX_ETHPORTS];
1262 uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
1263 struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
1264 } __rte_cache_aligned;
1266 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
1268 /* Enqueue a single packet, and send burst if queue is filled */
1270 send_single_packet(struct rte_mbuf *m, uint16_t port)
1273 struct lcore_conf *qconf;
1275 lcore_id = rte_lcore_id();
1277 qconf = &lcore_conf[lcore_id];
1278 rte_eth_tx_buffer(port, qconf->tx_queue_id[port],
1279 qconf->tx_buffer[port], m);
1282 #ifdef DO_RFC_1812_CHECKS
1284 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
1286 /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
1288 * 1. The packet length reported by the Link Layer must be large
1289 * enough to hold the minimum length legal IP datagram (20 bytes).
1291 if (link_len < sizeof(struct ipv4_hdr))
1294 /* 2. The IP checksum must be correct. */
1295 /* this is checked in H/W */
1298 * 3. The IP version number must be 4. If the version number is not 4
1299 * then the packet may be another version of IP, such as IPng or
1302 if (((pkt->version_ihl) >> 4) != 4)
1305 * 4. The IP header length field must be large enough to hold the
1306 * minimum length legal IP datagram (20 bytes = 5 words).
1308 if ((pkt->version_ihl & 0xf) < 5)
1312 * 5. The IP total length field must be large enough to hold the IP
1313 * datagram header, whose length is specified in the IP header length
1316 if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
1323 /* main processing loop */
1325 main_loop(__attribute__((unused)) void *dummy)
1327 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1329 uint64_t prev_tsc, diff_tsc, cur_tsc;
1333 struct lcore_conf *qconf;
1335 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
1336 / US_PER_S * BURST_TX_DRAIN_US;
1339 lcore_id = rte_lcore_id();
1340 qconf = &lcore_conf[lcore_id];
1341 socketid = rte_lcore_to_socket_id(lcore_id);
1343 if (qconf->n_rx_queue == 0) {
1344 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
1348 RTE_LOG(INFO, L3FWD, "entering main loop on lcore %u\n", lcore_id);
1350 for (i = 0; i < qconf->n_rx_queue; i++) {
1352 portid = qconf->rx_queue_list[i].port_id;
1353 queueid = qconf->rx_queue_list[i].queue_id;
1354 RTE_LOG(INFO, L3FWD,
1355 " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
1356 lcore_id, portid, queueid);
1361 cur_tsc = rte_rdtsc();
1364 * TX burst queue drain
1366 diff_tsc = cur_tsc - prev_tsc;
1367 if (unlikely(diff_tsc > drain_tsc)) {
1368 for (i = 0; i < qconf->n_tx_port; ++i) {
1369 portid = qconf->tx_port_id[i];
1370 rte_eth_tx_buffer_flush(portid,
1371 qconf->tx_queue_id[portid],
1372 qconf->tx_buffer[portid]);
1378 * Read packet from RX queues
1380 for (i = 0; i < qconf->n_rx_queue; ++i) {
1382 portid = qconf->rx_queue_list[i].port_id;
1383 queueid = qconf->rx_queue_list[i].queue_id;
1384 nb_rx = rte_eth_rx_burst(portid, queueid,
1385 pkts_burst, MAX_PKT_BURST);
1388 struct acl_search_t acl_search;
1390 prepare_acl_parameter(pkts_burst, &acl_search,
1393 if (acl_search.num_ipv4) {
1395 acl_config.acx_ipv4[socketid],
1396 acl_search.data_ipv4,
1397 acl_search.res_ipv4,
1398 acl_search.num_ipv4,
1399 DEFAULT_MAX_CATEGORIES);
1401 send_packets(acl_search.m_ipv4,
1402 acl_search.res_ipv4,
1403 acl_search.num_ipv4);
1406 if (acl_search.num_ipv6) {
1408 acl_config.acx_ipv6[socketid],
1409 acl_search.data_ipv6,
1410 acl_search.res_ipv6,
1411 acl_search.num_ipv6,
1412 DEFAULT_MAX_CATEGORIES);
1414 send_packets(acl_search.m_ipv6,
1415 acl_search.res_ipv6,
1416 acl_search.num_ipv6);
1424 check_lcore_params(void)
1426 uint8_t queue, lcore;
1430 for (i = 0; i < nb_lcore_params; ++i) {
1431 queue = lcore_params[i].queue_id;
1432 if (queue >= MAX_RX_QUEUE_PER_PORT) {
1433 printf("invalid queue number: %hhu\n", queue);
1436 lcore = lcore_params[i].lcore_id;
1437 if (!rte_lcore_is_enabled(lcore)) {
1438 printf("error: lcore %hhu is not enabled in "
1439 "lcore mask\n", lcore);
1442 socketid = rte_lcore_to_socket_id(lcore);
1443 if (socketid != 0 && numa_on == 0) {
1444 printf("warning: lcore %hhu is on socket %d "
1453 check_port_config(void)
1458 for (i = 0; i < nb_lcore_params; ++i) {
1459 portid = lcore_params[i].port_id;
1461 if ((enabled_port_mask & (1 << portid)) == 0) {
1462 printf("port %u is not enabled in port mask\n", portid);
1465 if (!rte_eth_dev_is_valid_port(portid)) {
1466 printf("port %u is not present on the board\n", portid);
1474 get_port_n_rx_queues(const uint16_t port)
1479 for (i = 0; i < nb_lcore_params; ++i) {
1480 if (lcore_params[i].port_id == port &&
1481 lcore_params[i].queue_id > queue)
1482 queue = lcore_params[i].queue_id;
1484 return (uint8_t)(++queue);
1488 init_lcore_rx_queues(void)
1490 uint16_t i, nb_rx_queue;
1493 for (i = 0; i < nb_lcore_params; ++i) {
1494 lcore = lcore_params[i].lcore_id;
1495 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1496 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1497 printf("error: too many queues (%u) for lcore: %u\n",
1498 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1501 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1502 lcore_params[i].port_id;
1503 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1504 lcore_params[i].queue_id;
1505 lcore_conf[lcore].n_rx_queue++;
1513 print_usage(const char *prgname)
1515 printf("%s [EAL options] -- -p PORTMASK -P"
1516 "--"OPTION_RULE_IPV4"=FILE"
1517 "--"OPTION_RULE_IPV6"=FILE"
1518 " [--"OPTION_CONFIG" (port,queue,lcore)[,(port,queue,lcore]]"
1519 " [--"OPTION_ENBJMO" [--max-pkt-len PKTLEN]]\n"
1520 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1521 " -P : enable promiscuous mode\n"
1522 " --"OPTION_CONFIG": (port,queue,lcore): "
1523 "rx queues configuration\n"
1524 " --"OPTION_NONUMA": optional, disable numa awareness\n"
1525 " --"OPTION_ENBJMO": enable jumbo frame"
1526 " which max packet len is PKTLEN in decimal (64-9600)\n"
1527 " --"OPTION_RULE_IPV4"=FILE: specify the ipv4 rules entries "
1529 "Each rule occupy one line. "
1530 "2 kinds of rules are supported. "
1531 "One is ACL entry at while line leads with character '%c', "
1532 "another is route entry at while line leads with "
1534 " --"OPTION_RULE_IPV6"=FILE: specify the ipv6 rules "
1536 " --"OPTION_SCALAR": Use scalar function to do lookup\n",
1537 prgname, ACL_LEAD_CHAR, ROUTE_LEAD_CHAR);
1541 parse_max_pkt_len(const char *pktlen)
1546 /* parse decimal string */
1547 len = strtoul(pktlen, &end, 10);
1548 if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1558 parse_portmask(const char *portmask)
1563 /* parse hexadecimal string */
1564 pm = strtoul(portmask, &end, 16);
1565 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1575 parse_config(const char *q_arg)
1578 const char *p, *p0 = q_arg;
1586 unsigned long int_fld[_NUM_FLD];
1587 char *str_fld[_NUM_FLD];
1591 nb_lcore_params = 0;
1593 while ((p = strchr(p0, '(')) != NULL) {
1595 if ((p0 = strchr(p, ')')) == NULL)
1599 if (size >= sizeof(s))
1602 snprintf(s, sizeof(s), "%.*s", size, p);
1603 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1606 for (i = 0; i < _NUM_FLD; i++) {
1608 int_fld[i] = strtoul(str_fld[i], &end, 0);
1609 if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
1612 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1613 printf("exceeded max number of lcore params: %hu\n",
1617 lcore_params_array[nb_lcore_params].port_id =
1618 (uint8_t)int_fld[FLD_PORT];
1619 lcore_params_array[nb_lcore_params].queue_id =
1620 (uint8_t)int_fld[FLD_QUEUE];
1621 lcore_params_array[nb_lcore_params].lcore_id =
1622 (uint8_t)int_fld[FLD_LCORE];
1625 lcore_params = lcore_params_array;
1629 /* Parse the argument given in the command line of the application */
1631 parse_args(int argc, char **argv)
1636 char *prgname = argv[0];
1637 static struct option lgopts[] = {
1638 {OPTION_CONFIG, 1, 0, 0},
1639 {OPTION_NONUMA, 0, 0, 0},
1640 {OPTION_ENBJMO, 0, 0, 0},
1641 {OPTION_RULE_IPV4, 1, 0, 0},
1642 {OPTION_RULE_IPV6, 1, 0, 0},
1643 {OPTION_SCALAR, 0, 0, 0},
1649 while ((opt = getopt_long(argc, argvopt, "p:P",
1650 lgopts, &option_index)) != EOF) {
1655 enabled_port_mask = parse_portmask(optarg);
1656 if (enabled_port_mask == 0) {
1657 printf("invalid portmask\n");
1658 print_usage(prgname);
1663 printf("Promiscuous mode selected\n");
1669 if (!strncmp(lgopts[option_index].name,
1671 sizeof(OPTION_CONFIG))) {
1672 ret = parse_config(optarg);
1674 printf("invalid config\n");
1675 print_usage(prgname);
1680 if (!strncmp(lgopts[option_index].name,
1682 sizeof(OPTION_NONUMA))) {
1683 printf("numa is disabled\n");
1687 if (!strncmp(lgopts[option_index].name,
1688 OPTION_ENBJMO, sizeof(OPTION_ENBJMO))) {
1689 struct option lenopts = {
1696 printf("jumbo frame is enabled\n");
1697 port_conf.rxmode.offloads |=
1698 DEV_RX_OFFLOAD_JUMBO_FRAME;
1699 port_conf.txmode.offloads |=
1700 DEV_TX_OFFLOAD_MULTI_SEGS;
1703 * if no max-pkt-len set, then use the
1704 * default value ETHER_MAX_LEN
1706 if (0 == getopt_long(argc, argvopt, "",
1707 &lenopts, &option_index)) {
1708 ret = parse_max_pkt_len(optarg);
1710 (ret > MAX_JUMBO_PKT_LEN)) {
1711 printf("invalid packet "
1713 print_usage(prgname);
1716 port_conf.rxmode.max_rx_pkt_len = ret;
1718 printf("set jumbo frame max packet length "
1721 port_conf.rxmode.max_rx_pkt_len);
1724 if (!strncmp(lgopts[option_index].name,
1726 sizeof(OPTION_RULE_IPV4)))
1727 parm_config.rule_ipv4_name = optarg;
1729 if (!strncmp(lgopts[option_index].name,
1731 sizeof(OPTION_RULE_IPV6))) {
1732 parm_config.rule_ipv6_name = optarg;
1735 if (!strncmp(lgopts[option_index].name,
1736 OPTION_SCALAR, sizeof(OPTION_SCALAR)))
1737 parm_config.scalar = 1;
1743 print_usage(prgname);
1749 argv[optind-1] = prgname;
1752 optind = 1; /* reset getopt lib */
1757 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1759 char buf[ETHER_ADDR_FMT_SIZE];
1760 ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
1761 printf("%s%s", name, buf);
1765 init_mem(unsigned nb_mbuf)
1771 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1772 if (rte_lcore_is_enabled(lcore_id) == 0)
1776 socketid = rte_lcore_to_socket_id(lcore_id);
1780 if (socketid >= NB_SOCKETS) {
1781 rte_exit(EXIT_FAILURE,
1782 "Socket %d of lcore %u is out of range %d\n",
1783 socketid, lcore_id, NB_SOCKETS);
1785 if (pktmbuf_pool[socketid] == NULL) {
1786 snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1787 pktmbuf_pool[socketid] =
1788 rte_pktmbuf_pool_create(s, nb_mbuf,
1789 MEMPOOL_CACHE_SIZE, 0,
1790 RTE_MBUF_DEFAULT_BUF_SIZE,
1792 if (pktmbuf_pool[socketid] == NULL)
1793 rte_exit(EXIT_FAILURE,
1794 "Cannot init mbuf pool on socket %d\n",
1797 printf("Allocated mbuf pool on socket %d\n",
1804 /* Check the link status of all ports in up to 9s, and print them finally */
1806 check_all_ports_link_status(uint32_t port_mask)
1808 #define CHECK_INTERVAL 100 /* 100ms */
1809 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1811 uint8_t count, all_ports_up, print_flag = 0;
1812 struct rte_eth_link link;
1814 printf("\nChecking link status");
1816 for (count = 0; count <= MAX_CHECK_TIME; count++) {
1818 RTE_ETH_FOREACH_DEV(portid) {
1819 if ((port_mask & (1 << portid)) == 0)
1821 memset(&link, 0, sizeof(link));
1822 rte_eth_link_get_nowait(portid, &link);
1823 /* print link status if flag set */
1824 if (print_flag == 1) {
1825 if (link.link_status)
1827 "Port%d Link Up. Speed %u Mbps %s\n",
1828 portid, link.link_speed,
1829 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1830 ("full-duplex") : ("half-duplex\n"));
1832 printf("Port %d Link Down\n", portid);
1835 /* clear all_ports_up flag if any link down */
1836 if (link.link_status == ETH_LINK_DOWN) {
1841 /* after finally printing all link status, get out */
1842 if (print_flag == 1)
1845 if (all_ports_up == 0) {
1848 rte_delay_ms(CHECK_INTERVAL);
1851 /* set the print_flag if all ports up or timeout */
1852 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1860 main(int argc, char **argv)
1862 struct lcore_conf *qconf;
1863 struct rte_eth_dev_info dev_info;
1864 struct rte_eth_txconf *txconf;
1869 uint32_t n_tx_queue, nb_lcores;
1871 uint8_t nb_rx_queue, queue, socketid;
1874 ret = rte_eal_init(argc, argv);
1876 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1880 /* parse application arguments (after the EAL ones) */
1881 ret = parse_args(argc, argv);
1883 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1885 if (check_lcore_params() < 0)
1886 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1888 ret = init_lcore_rx_queues();
1890 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1892 nb_ports = rte_eth_dev_count_avail();
1894 if (check_port_config() < 0)
1895 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1897 /* Add ACL rules and route entries, build trie */
1898 if (app_acl_init() < 0)
1899 rte_exit(EXIT_FAILURE, "app_acl_init failed\n");
1901 nb_lcores = rte_lcore_count();
1903 /* initialize all ports */
1904 RTE_ETH_FOREACH_DEV(portid) {
1905 struct rte_eth_conf local_port_conf = port_conf;
1907 /* skip ports that are not enabled */
1908 if ((enabled_port_mask & (1 << portid)) == 0) {
1909 printf("\nSkipping disabled port %d\n", portid);
1914 printf("Initializing port %d ... ", portid);
1917 nb_rx_queue = get_port_n_rx_queues(portid);
1918 n_tx_queue = nb_lcores;
1919 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
1920 n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1921 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1922 nb_rx_queue, (unsigned)n_tx_queue);
1923 rte_eth_dev_info_get(portid, &dev_info);
1924 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
1925 local_port_conf.txmode.offloads |=
1926 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
1928 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
1929 dev_info.flow_type_rss_offloads;
1930 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
1931 port_conf.rx_adv_conf.rss_conf.rss_hf) {
1932 printf("Port %u modified RSS hash function based on hardware support,"
1933 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
1935 port_conf.rx_adv_conf.rss_conf.rss_hf,
1936 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
1939 ret = rte_eth_dev_configure(portid, nb_rx_queue,
1940 (uint16_t)n_tx_queue, &local_port_conf);
1942 rte_exit(EXIT_FAILURE,
1943 "Cannot configure device: err=%d, port=%d\n",
1946 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
1949 rte_exit(EXIT_FAILURE,
1950 "rte_eth_dev_adjust_nb_rx_tx_desc: err=%d, port=%d\n",
1953 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1954 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1958 ret = init_mem(NB_MBUF);
1960 rte_exit(EXIT_FAILURE, "init_mem failed\n");
1962 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1963 if (rte_lcore_is_enabled(lcore_id) == 0)
1966 /* Initialize TX buffers */
1967 qconf = &lcore_conf[lcore_id];
1968 qconf->tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
1969 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
1970 rte_eth_dev_socket_id(portid));
1971 if (qconf->tx_buffer[portid] == NULL)
1972 rte_exit(EXIT_FAILURE, "Can't allocate tx buffer for port %u\n",
1975 rte_eth_tx_buffer_init(qconf->tx_buffer[portid], MAX_PKT_BURST);
1978 /* init one TX queue per couple (lcore,port) */
1980 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1981 if (rte_lcore_is_enabled(lcore_id) == 0)
1985 socketid = (uint8_t)
1986 rte_lcore_to_socket_id(lcore_id);
1990 printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1993 rte_eth_dev_info_get(portid, &dev_info);
1994 txconf = &dev_info.default_txconf;
1995 txconf->offloads = local_port_conf.txmode.offloads;
1996 ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1999 rte_exit(EXIT_FAILURE,
2000 "rte_eth_tx_queue_setup: err=%d, "
2001 "port=%d\n", ret, portid);
2003 qconf = &lcore_conf[lcore_id];
2004 qconf->tx_queue_id[portid] = queueid;
2007 qconf->tx_port_id[qconf->n_tx_port] = portid;
2013 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2014 if (rte_lcore_is_enabled(lcore_id) == 0)
2016 qconf = &lcore_conf[lcore_id];
2017 printf("\nInitializing rx queues on lcore %u ... ", lcore_id);
2019 /* init RX queues */
2020 for (queue = 0; queue < qconf->n_rx_queue; ++queue) {
2021 struct rte_eth_dev *dev;
2022 struct rte_eth_conf *conf;
2023 struct rte_eth_rxconf rxq_conf;
2025 portid = qconf->rx_queue_list[queue].port_id;
2026 queueid = qconf->rx_queue_list[queue].queue_id;
2027 dev = &rte_eth_devices[portid];
2028 conf = &dev->data->dev_conf;
2031 socketid = (uint8_t)
2032 rte_lcore_to_socket_id(lcore_id);
2036 printf("rxq=%d,%d,%d ", portid, queueid, socketid);
2039 rte_eth_dev_info_get(portid, &dev_info);
2040 rxq_conf = dev_info.default_rxconf;
2041 rxq_conf.offloads = conf->rxmode.offloads;
2042 ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
2043 socketid, &rxq_conf,
2044 pktmbuf_pool[socketid]);
2046 rte_exit(EXIT_FAILURE,
2047 "rte_eth_rx_queue_setup: err=%d,"
2048 "port=%d\n", ret, portid);
2055 RTE_ETH_FOREACH_DEV(portid) {
2056 if ((enabled_port_mask & (1 << portid)) == 0)
2060 ret = rte_eth_dev_start(portid);
2062 rte_exit(EXIT_FAILURE,
2063 "rte_eth_dev_start: err=%d, port=%d\n",
2067 * If enabled, put device in promiscuous mode.
2068 * This allows IO forwarding mode to forward packets
2069 * to itself through 2 cross-connected ports of the
2073 rte_eth_promiscuous_enable(portid);
2076 check_all_ports_link_status(enabled_port_mask);
2078 /* launch per-lcore init on every lcore */
2079 rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
2080 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
2081 if (rte_eal_wait_lcore(lcore_id) < 0)