1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2014 Intel Corporation
3  */
4 
5 #include <stdio.h>
6 #include <stdlib.h>
7 #include <stdint.h>
8 #include <inttypes.h>
9 #include <sys/types.h>
10 #include <sys/param.h>
11 #include <string.h>
12 #include <sys/queue.h>
13 #include <stdarg.h>
14 #include <errno.h>
15 #include <getopt.h>
16 
17 #include <rte_common.h>
18 #include <rte_byteorder.h>
19 #include <rte_log.h>
20 #include <rte_memory.h>
21 #include <rte_memcpy.h>
22 #include <rte_eal.h>
23 #include <rte_launch.h>
24 #include <rte_atomic.h>
25 #include <rte_cycles.h>
26 #include <rte_prefetch.h>
27 #include <rte_lcore.h>
28 #include <rte_per_lcore.h>
29 #include <rte_branch_prediction.h>
30 #include <rte_interrupts.h>
31 #include <rte_random.h>
32 #include <rte_debug.h>
33 #include <rte_ether.h>
34 #include <rte_ethdev.h>
35 #include <rte_mempool.h>
36 #include <rte_mbuf.h>
37 #include <rte_lpm.h>
38 #include <rte_lpm6.h>
39 #include <rte_ip.h>
40 #include <rte_string_fns.h>
41 
42 #include <rte_ip_frag.h>
43 
44 #define RTE_LOGTYPE_IP_FRAG RTE_LOGTYPE_USER1
45 
46 /* allow max jumbo frame 9.5 KB */
47 #define JUMBO_FRAME_MAX_SIZE	0x2600
48 
49 #define	ROUNDUP_DIV(a, b)	(((a) + (b) - 1) / (b))
50 
51 /*
52  * Default byte size for the IPv6 Maximum Transfer Unit (MTU).
53  * This value includes the size of IPv6 header.
54  */
55 #define	IPV4_MTU_DEFAULT	ETHER_MTU
56 #define	IPV6_MTU_DEFAULT	ETHER_MTU
57 
58 /*
59  * Default payload in bytes for the IPv6 packet.
60  */
61 #define	IPV4_DEFAULT_PAYLOAD	(IPV4_MTU_DEFAULT - sizeof(struct ipv4_hdr))
62 #define	IPV6_DEFAULT_PAYLOAD	(IPV6_MTU_DEFAULT - sizeof(struct ipv6_hdr))
63 
64 /*
65  * Max number of fragments per packet expected - defined by config file.
66  */
67 #define	MAX_PACKET_FRAG RTE_LIBRTE_IP_FRAG_MAX_FRAG
68 
69 #define NB_MBUF   8192
70 
71 #define MAX_PKT_BURST	32
72 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
73 
74 /* Configure how many packets ahead to prefetch, when reading packets */
75 #define PREFETCH_OFFSET	3
76 
77 /*
78  * Configurable number of RX/TX ring descriptors
79  */
80 #define RTE_TEST_RX_DESC_DEFAULT 1024
81 #define RTE_TEST_TX_DESC_DEFAULT 1024
82 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
83 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
84 
85 /* ethernet addresses of ports */
86 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
87 
88 #ifndef IPv4_BYTES
89 #define IPv4_BYTES_FMT "%" PRIu8 ".%" PRIu8 ".%" PRIu8 ".%" PRIu8
90 #define IPv4_BYTES(addr) \
91 		(uint8_t) (((addr) >> 24) & 0xFF),\
92 		(uint8_t) (((addr) >> 16) & 0xFF),\
93 		(uint8_t) (((addr) >> 8) & 0xFF),\
94 		(uint8_t) ((addr) & 0xFF)
95 #endif
96 
97 #ifndef IPv6_BYTES
98 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
99                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
100 #define IPv6_BYTES(addr) \
101 	addr[0],  addr[1], addr[2],  addr[3], \
102 	addr[4],  addr[5], addr[6],  addr[7], \
103 	addr[8],  addr[9], addr[10], addr[11],\
104 	addr[12], addr[13],addr[14], addr[15]
105 #endif
106 
107 #define IPV6_ADDR_LEN 16
108 
109 /* mask of enabled ports */
110 static int enabled_port_mask = 0;
111 
112 static int rx_queue_per_lcore = 1;
113 
114 #define MBUF_TABLE_SIZE  (2 * MAX(MAX_PKT_BURST, MAX_PACKET_FRAG))
115 
116 struct mbuf_table {
117 	uint16_t len;
118 	struct rte_mbuf *m_table[MBUF_TABLE_SIZE];
119 };
120 
121 struct rx_queue {
122 	struct rte_mempool *direct_pool;
123 	struct rte_mempool *indirect_pool;
124 	struct rte_lpm *lpm;
125 	struct rte_lpm6 *lpm6;
126 	uint16_t portid;
127 };
128 
129 #define MAX_RX_QUEUE_PER_LCORE 16
130 #define MAX_TX_QUEUE_PER_PORT 16
131 struct lcore_queue_conf {
132 	uint16_t n_rx_queue;
133 	uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
134 	struct rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
135 	struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
136 } __rte_cache_aligned;
137 struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
138 
139 static struct rte_eth_conf port_conf = {
140 	.rxmode = {
141 		.max_rx_pkt_len = JUMBO_FRAME_MAX_SIZE,
142 		.split_hdr_size = 0,
143 		.offloads = (DEV_RX_OFFLOAD_CHECKSUM |
144 			     DEV_RX_OFFLOAD_JUMBO_FRAME),
145 	},
146 	.txmode = {
147 		.mq_mode = ETH_MQ_TX_NONE,
148 		.offloads = (DEV_TX_OFFLOAD_IPV4_CKSUM |
149 			     DEV_TX_OFFLOAD_MULTI_SEGS),
150 	},
151 };
152 
153 /*
154  * IPv4 forwarding table
155  */
156 struct l3fwd_ipv4_route {
157 	uint32_t ip;
158 	uint8_t  depth;
159 	uint8_t  if_out;
160 };
161 
162 struct l3fwd_ipv4_route l3fwd_ipv4_route_array[] = {
163 		{IPv4(100,10,0,0), 16, 0},
164 		{IPv4(100,20,0,0), 16, 1},
165 		{IPv4(100,30,0,0), 16, 2},
166 		{IPv4(100,40,0,0), 16, 3},
167 		{IPv4(100,50,0,0), 16, 4},
168 		{IPv4(100,60,0,0), 16, 5},
169 		{IPv4(100,70,0,0), 16, 6},
170 		{IPv4(100,80,0,0), 16, 7},
171 };
172 
173 /*
174  * IPv6 forwarding table
175  */
176 
177 struct l3fwd_ipv6_route {
178 	uint8_t ip[IPV6_ADDR_LEN];
179 	uint8_t depth;
180 	uint8_t if_out;
181 };
182 
183 static struct l3fwd_ipv6_route l3fwd_ipv6_route_array[] = {
184 	{{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
185 	{{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
186 	{{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
187 	{{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
188 	{{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
189 	{{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
190 	{{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
191 	{{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
192 };
193 
194 #define LPM_MAX_RULES         1024
195 #define LPM6_MAX_RULES         1024
196 #define LPM6_NUMBER_TBL8S (1 << 16)
197 
198 struct rte_lpm6_config lpm6_config = {
199 		.max_rules = LPM6_MAX_RULES,
200 		.number_tbl8s = LPM6_NUMBER_TBL8S,
201 		.flags = 0
202 };
203 
204 static struct rte_mempool *socket_direct_pool[RTE_MAX_NUMA_NODES];
205 static struct rte_mempool *socket_indirect_pool[RTE_MAX_NUMA_NODES];
206 static struct rte_lpm *socket_lpm[RTE_MAX_NUMA_NODES];
207 static struct rte_lpm6 *socket_lpm6[RTE_MAX_NUMA_NODES];
208 
209 /* Send burst of packets on an output interface */
210 static inline int
211 send_burst(struct lcore_queue_conf *qconf, uint16_t n, uint16_t port)
212 {
213 	struct rte_mbuf **m_table;
214 	int ret;
215 	uint16_t queueid;
216 
217 	queueid = qconf->tx_queue_id[port];
218 	m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
219 
220 	ret = rte_eth_tx_burst(port, queueid, m_table, n);
221 	if (unlikely(ret < n)) {
222 		do {
223 			rte_pktmbuf_free(m_table[ret]);
224 		} while (++ret < n);
225 	}
226 
227 	return 0;
228 }
229 
230 static inline void
231 l3fwd_simple_forward(struct rte_mbuf *m, struct lcore_queue_conf *qconf,
232 		uint8_t queueid, uint16_t port_in)
233 {
234 	struct rx_queue *rxq;
235 	uint32_t i, len, next_hop;
236 	uint16_t port_out, ether_type;
237 	int32_t len2;
238 	const struct ether_hdr *eth;
239 
240 	rxq = &qconf->rx_queue_list[queueid];
241 
242 	/* by default, send everything back to the source port */
243 	port_out = port_in;
244 
245 	/* save ether type of the incoming packet */
246 	eth = rte_pktmbuf_mtod(m, const struct ether_hdr *);
247 	ether_type = eth->ether_type;
248 
249 	/* Remove the Ethernet header and trailer from the input packet */
250 	rte_pktmbuf_adj(m, (uint16_t)sizeof(struct ether_hdr));
251 
252 	/* Build transmission burst */
253 	len = qconf->tx_mbufs[port_out].len;
254 
255 	/* if this is an IPv4 packet */
256 	if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
257 		struct ipv4_hdr *ip_hdr;
258 		uint32_t ip_dst;
259 		/* Read the lookup key (i.e. ip_dst) from the input packet */
260 		ip_hdr = rte_pktmbuf_mtod(m, struct ipv4_hdr *);
261 		ip_dst = rte_be_to_cpu_32(ip_hdr->dst_addr);
262 
263 		/* Find destination port */
264 		if (rte_lpm_lookup(rxq->lpm, ip_dst, &next_hop) == 0 &&
265 				(enabled_port_mask & 1 << next_hop) != 0) {
266 			port_out = next_hop;
267 
268 			/* Build transmission burst for new port */
269 			len = qconf->tx_mbufs[port_out].len;
270 		}
271 
272 		/* if we don't need to do any fragmentation */
273 		if (likely (IPV4_MTU_DEFAULT >= m->pkt_len)) {
274 			qconf->tx_mbufs[port_out].m_table[len] = m;
275 			len2 = 1;
276 		} else {
277 			len2 = rte_ipv4_fragment_packet(m,
278 				&qconf->tx_mbufs[port_out].m_table[len],
279 				(uint16_t)(MBUF_TABLE_SIZE - len),
280 				IPV4_MTU_DEFAULT,
281 				rxq->direct_pool, rxq->indirect_pool);
282 
283 			/* Free input packet */
284 			rte_pktmbuf_free(m);
285 
286 			/* If we fail to fragment the packet */
287 			if (unlikely (len2 < 0))
288 				return;
289 		}
290 	} else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
291 		/* if this is an IPv6 packet */
292 		struct ipv6_hdr *ip_hdr;
293 
294 		/* Read the lookup key (i.e. ip_dst) from the input packet */
295 		ip_hdr = rte_pktmbuf_mtod(m, struct ipv6_hdr *);
296 
297 		/* Find destination port */
298 		if (rte_lpm6_lookup(rxq->lpm6, ip_hdr->dst_addr,
299 						&next_hop) == 0 &&
300 				(enabled_port_mask & 1 << next_hop) != 0) {
301 			port_out = next_hop;
302 
303 			/* Build transmission burst for new port */
304 			len = qconf->tx_mbufs[port_out].len;
305 		}
306 
307 		/* if we don't need to do any fragmentation */
308 		if (likely (IPV6_MTU_DEFAULT >= m->pkt_len)) {
309 			qconf->tx_mbufs[port_out].m_table[len] = m;
310 			len2 = 1;
311 		} else {
312 			len2 = rte_ipv6_fragment_packet(m,
313 				&qconf->tx_mbufs[port_out].m_table[len],
314 				(uint16_t)(MBUF_TABLE_SIZE - len),
315 				IPV6_MTU_DEFAULT,
316 				rxq->direct_pool, rxq->indirect_pool);
317 
318 			/* Free input packet */
319 			rte_pktmbuf_free(m);
320 
321 			/* If we fail to fragment the packet */
322 			if (unlikely (len2 < 0))
323 				return;
324 		}
325 	}
326 	/* else, just forward the packet */
327 	else {
328 		qconf->tx_mbufs[port_out].m_table[len] = m;
329 		len2 = 1;
330 	}
331 
332 	for (i = len; i < len + len2; i ++) {
333 		void *d_addr_bytes;
334 
335 		m = qconf->tx_mbufs[port_out].m_table[i];
336 		struct ether_hdr *eth_hdr = (struct ether_hdr *)
337 			rte_pktmbuf_prepend(m, (uint16_t)sizeof(struct ether_hdr));
338 		if (eth_hdr == NULL) {
339 			rte_panic("No headroom in mbuf.\n");
340 		}
341 
342 		m->l2_len = sizeof(struct ether_hdr);
343 
344 		/* 02:00:00:00:00:xx */
345 		d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
346 		*((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)port_out << 40);
347 
348 		/* src addr */
349 		ether_addr_copy(&ports_eth_addr[port_out], &eth_hdr->s_addr);
350 		eth_hdr->ether_type = ether_type;
351 	}
352 
353 	len += len2;
354 
355 	if (likely(len < MAX_PKT_BURST)) {
356 		qconf->tx_mbufs[port_out].len = (uint16_t)len;
357 		return;
358 	}
359 
360 	/* Transmit packets */
361 	send_burst(qconf, (uint16_t)len, port_out);
362 	qconf->tx_mbufs[port_out].len = 0;
363 }
364 
365 /* main processing loop */
366 static int
367 main_loop(__attribute__((unused)) void *dummy)
368 {
369 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
370 	unsigned lcore_id;
371 	uint64_t prev_tsc, diff_tsc, cur_tsc;
372 	int i, j, nb_rx;
373 	uint16_t portid;
374 	struct lcore_queue_conf *qconf;
375 	const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
376 
377 	prev_tsc = 0;
378 
379 	lcore_id = rte_lcore_id();
380 	qconf = &lcore_queue_conf[lcore_id];
381 
382 	if (qconf->n_rx_queue == 0) {
383 		RTE_LOG(INFO, IP_FRAG, "lcore %u has nothing to do\n", lcore_id);
384 		return 0;
385 	}
386 
387 	RTE_LOG(INFO, IP_FRAG, "entering main loop on lcore %u\n", lcore_id);
388 
389 	for (i = 0; i < qconf->n_rx_queue; i++) {
390 
391 		portid = qconf->rx_queue_list[i].portid;
392 		RTE_LOG(INFO, IP_FRAG, " -- lcoreid=%u portid=%d\n", lcore_id,
393 				portid);
394 	}
395 
396 	while (1) {
397 
398 		cur_tsc = rte_rdtsc();
399 
400 		/*
401 		 * TX burst queue drain
402 		 */
403 		diff_tsc = cur_tsc - prev_tsc;
404 		if (unlikely(diff_tsc > drain_tsc)) {
405 
406 			/*
407 			 * This could be optimized (use queueid instead of
408 			 * portid), but it is not called so often
409 			 */
410 			for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
411 				if (qconf->tx_mbufs[portid].len == 0)
412 					continue;
413 				send_burst(&lcore_queue_conf[lcore_id],
414 					   qconf->tx_mbufs[portid].len,
415 					   portid);
416 				qconf->tx_mbufs[portid].len = 0;
417 			}
418 
419 			prev_tsc = cur_tsc;
420 		}
421 
422 		/*
423 		 * Read packet from RX queues
424 		 */
425 		for (i = 0; i < qconf->n_rx_queue; i++) {
426 
427 			portid = qconf->rx_queue_list[i].portid;
428 			nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
429 						 MAX_PKT_BURST);
430 
431 			/* Prefetch first packets */
432 			for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
433 				rte_prefetch0(rte_pktmbuf_mtod(
434 						pkts_burst[j], void *));
435 			}
436 
437 			/* Prefetch and forward already prefetched packets */
438 			for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
439 				rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
440 						j + PREFETCH_OFFSET], void *));
441 				l3fwd_simple_forward(pkts_burst[j], qconf, i, portid);
442 			}
443 
444 			/* Forward remaining prefetched packets */
445 			for (; j < nb_rx; j++) {
446 				l3fwd_simple_forward(pkts_burst[j], qconf, i, portid);
447 			}
448 		}
449 	}
450 }
451 
452 /* display usage */
453 static void
454 print_usage(const char *prgname)
455 {
456 	printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
457 	       "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
458 	       "  -q NQ: number of queue (=ports) per lcore (default is 1)\n",
459 	       prgname);
460 }
461 
462 static int
463 parse_portmask(const char *portmask)
464 {
465 	char *end = NULL;
466 	unsigned long pm;
467 
468 	/* parse hexadecimal string */
469 	pm = strtoul(portmask, &end, 16);
470 	if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
471 		return -1;
472 
473 	if (pm == 0)
474 		return -1;
475 
476 	return pm;
477 }
478 
479 static int
480 parse_nqueue(const char *q_arg)
481 {
482 	char *end = NULL;
483 	unsigned long n;
484 
485 	/* parse hexadecimal string */
486 	n = strtoul(q_arg, &end, 10);
487 	if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
488 		return -1;
489 	if (n == 0)
490 		return -1;
491 	if (n >= MAX_RX_QUEUE_PER_LCORE)
492 		return -1;
493 
494 	return n;
495 }
496 
497 /* Parse the argument given in the command line of the application */
498 static int
499 parse_args(int argc, char **argv)
500 {
501 	int opt, ret;
502 	char **argvopt;
503 	int option_index;
504 	char *prgname = argv[0];
505 	static struct option lgopts[] = {
506 		{NULL, 0, 0, 0}
507 	};
508 
509 	argvopt = argv;
510 
511 	while ((opt = getopt_long(argc, argvopt, "p:q:",
512 				  lgopts, &option_index)) != EOF) {
513 
514 		switch (opt) {
515 		/* portmask */
516 		case 'p':
517 			enabled_port_mask = parse_portmask(optarg);
518 			if (enabled_port_mask < 0) {
519 				printf("invalid portmask\n");
520 				print_usage(prgname);
521 				return -1;
522 			}
523 			break;
524 
525 		/* nqueue */
526 		case 'q':
527 			rx_queue_per_lcore = parse_nqueue(optarg);
528 			if (rx_queue_per_lcore < 0) {
529 				printf("invalid queue number\n");
530 				print_usage(prgname);
531 				return -1;
532 			}
533 			break;
534 
535 		/* long options */
536 		case 0:
537 			print_usage(prgname);
538 			return -1;
539 
540 		default:
541 			print_usage(prgname);
542 			return -1;
543 		}
544 	}
545 
546 	if (enabled_port_mask == 0) {
547 		printf("portmask not specified\n");
548 		print_usage(prgname);
549 		return -1;
550 	}
551 
552 	if (optind >= 0)
553 		argv[optind-1] = prgname;
554 
555 	ret = optind-1;
556 	optind = 1; /* reset getopt lib */
557 	return ret;
558 }
559 
560 static void
561 print_ethaddr(const char *name, struct ether_addr *eth_addr)
562 {
563 	char buf[ETHER_ADDR_FMT_SIZE];
564 	ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
565 	printf("%s%s", name, buf);
566 }
567 
568 /* Check the link status of all ports in up to 9s, and print them finally */
569 static void
570 check_all_ports_link_status(uint32_t port_mask)
571 {
572 #define CHECK_INTERVAL 100 /* 100ms */
573 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
574 	uint16_t portid;
575 	uint8_t count, all_ports_up, print_flag = 0;
576 	struct rte_eth_link link;
577 
578 	printf("\nChecking link status");
579 	fflush(stdout);
580 	for (count = 0; count <= MAX_CHECK_TIME; count++) {
581 		all_ports_up = 1;
582 		RTE_ETH_FOREACH_DEV(portid) {
583 			if ((port_mask & (1 << portid)) == 0)
584 				continue;
585 			memset(&link, 0, sizeof(link));
586 			rte_eth_link_get_nowait(portid, &link);
587 			/* print link status if flag set */
588 			if (print_flag == 1) {
589 				if (link.link_status)
590 					printf(
591 					"Port%d Link Up .Speed %u Mbps - %s\n",
592 						portid, link.link_speed,
593 				(link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
594 					("full-duplex") : ("half-duplex\n"));
595 				else
596 					printf("Port %d Link Down\n", portid);
597 				continue;
598 			}
599 			/* clear all_ports_up flag if any link down */
600 			if (link.link_status == ETH_LINK_DOWN) {
601 				all_ports_up = 0;
602 				break;
603 			}
604 		}
605 		/* after finally printing all link status, get out */
606 		if (print_flag == 1)
607 			break;
608 
609 		if (all_ports_up == 0) {
610 			printf(".");
611 			fflush(stdout);
612 			rte_delay_ms(CHECK_INTERVAL);
613 		}
614 
615 		/* set the print_flag if all ports up or timeout */
616 		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
617 			print_flag = 1;
618 			printf("\ndone\n");
619 		}
620 	}
621 }
622 
623 /* Check L3 packet type detection capablity of the NIC port */
624 static int
625 check_ptype(int portid)
626 {
627 	int i, ret;
628 	int ptype_l3_ipv4 = 0, ptype_l3_ipv6 = 0;
629 	uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
630 
631 	ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
632 	if (ret <= 0)
633 		return 0;
634 
635 	uint32_t ptypes[ret];
636 
637 	ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
638 	for (i = 0; i < ret; ++i) {
639 		if (ptypes[i] & RTE_PTYPE_L3_IPV4)
640 			ptype_l3_ipv4 = 1;
641 		if (ptypes[i] & RTE_PTYPE_L3_IPV6)
642 			ptype_l3_ipv6 = 1;
643 	}
644 
645 	if (ptype_l3_ipv4 == 0)
646 		printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
647 
648 	if (ptype_l3_ipv6 == 0)
649 		printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
650 
651 	if (ptype_l3_ipv4 && ptype_l3_ipv6)
652 		return 1;
653 
654 	return 0;
655 
656 }
657 
658 /* Parse packet type of a packet by SW */
659 static inline void
660 parse_ptype(struct rte_mbuf *m)
661 {
662 	struct ether_hdr *eth_hdr;
663 	uint32_t packet_type = RTE_PTYPE_UNKNOWN;
664 	uint16_t ether_type;
665 
666 	eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
667 	ether_type = eth_hdr->ether_type;
668 	if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4))
669 		packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
670 	else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6))
671 		packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
672 
673 	m->packet_type = packet_type;
674 }
675 
676 /* callback function to detect packet type for a queue of a port */
677 static uint16_t
678 cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
679 		   struct rte_mbuf *pkts[], uint16_t nb_pkts,
680 		   uint16_t max_pkts __rte_unused,
681 		   void *user_param __rte_unused)
682 {
683 	uint16_t i;
684 
685 	for (i = 0; i < nb_pkts; ++i)
686 		parse_ptype(pkts[i]);
687 
688 	return nb_pkts;
689 }
690 
691 static int
692 init_routing_table(void)
693 {
694 	struct rte_lpm *lpm;
695 	struct rte_lpm6 *lpm6;
696 	int socket, ret;
697 	unsigned i;
698 
699 	for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) {
700 		if (socket_lpm[socket]) {
701 			lpm = socket_lpm[socket];
702 			/* populate the LPM table */
703 			for (i = 0; i < RTE_DIM(l3fwd_ipv4_route_array); i++) {
704 				ret = rte_lpm_add(lpm,
705 					l3fwd_ipv4_route_array[i].ip,
706 					l3fwd_ipv4_route_array[i].depth,
707 					l3fwd_ipv4_route_array[i].if_out);
708 
709 				if (ret < 0) {
710 					RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd "
711 						"LPM table\n", i);
712 					return -1;
713 				}
714 
715 				RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv4_BYTES_FMT
716 						"/%d (port %d)\n",
717 					socket,
718 					IPv4_BYTES(l3fwd_ipv4_route_array[i].ip),
719 					l3fwd_ipv4_route_array[i].depth,
720 					l3fwd_ipv4_route_array[i].if_out);
721 			}
722 		}
723 
724 		if (socket_lpm6[socket]) {
725 			lpm6 = socket_lpm6[socket];
726 			/* populate the LPM6 table */
727 			for (i = 0; i < RTE_DIM(l3fwd_ipv6_route_array); i++) {
728 				ret = rte_lpm6_add(lpm6,
729 					l3fwd_ipv6_route_array[i].ip,
730 					l3fwd_ipv6_route_array[i].depth,
731 					l3fwd_ipv6_route_array[i].if_out);
732 
733 				if (ret < 0) {
734 					RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd "
735 						"LPM6 table\n", i);
736 					return -1;
737 				}
738 
739 				RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv6_BYTES_FMT
740 						"/%d (port %d)\n",
741 					socket,
742 					IPv6_BYTES(l3fwd_ipv6_route_array[i].ip),
743 					l3fwd_ipv6_route_array[i].depth,
744 					l3fwd_ipv6_route_array[i].if_out);
745 			}
746 		}
747 	}
748 	return 0;
749 }
750 
751 static int
752 init_mem(void)
753 {
754 	char buf[PATH_MAX];
755 	struct rte_mempool *mp;
756 	struct rte_lpm *lpm;
757 	struct rte_lpm6 *lpm6;
758 	struct rte_lpm_config lpm_config;
759 	int socket;
760 	unsigned lcore_id;
761 
762 	/* traverse through lcores and initialize structures on each socket */
763 
764 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
765 
766 		if (rte_lcore_is_enabled(lcore_id) == 0)
767 			continue;
768 
769 		socket = rte_lcore_to_socket_id(lcore_id);
770 
771 		if (socket == SOCKET_ID_ANY)
772 			socket = 0;
773 
774 		if (socket_direct_pool[socket] == NULL) {
775 			RTE_LOG(INFO, IP_FRAG, "Creating direct mempool on socket %i\n",
776 					socket);
777 			snprintf(buf, sizeof(buf), "pool_direct_%i", socket);
778 
779 			mp = rte_pktmbuf_pool_create(buf, NB_MBUF, 32,
780 				0, RTE_MBUF_DEFAULT_BUF_SIZE, socket);
781 			if (mp == NULL) {
782 				RTE_LOG(ERR, IP_FRAG, "Cannot create direct mempool\n");
783 				return -1;
784 			}
785 			socket_direct_pool[socket] = mp;
786 		}
787 
788 		if (socket_indirect_pool[socket] == NULL) {
789 			RTE_LOG(INFO, IP_FRAG, "Creating indirect mempool on socket %i\n",
790 					socket);
791 			snprintf(buf, sizeof(buf), "pool_indirect_%i", socket);
792 
793 			mp = rte_pktmbuf_pool_create(buf, NB_MBUF, 32, 0, 0,
794 				socket);
795 			if (mp == NULL) {
796 				RTE_LOG(ERR, IP_FRAG, "Cannot create indirect mempool\n");
797 				return -1;
798 			}
799 			socket_indirect_pool[socket] = mp;
800 		}
801 
802 		if (socket_lpm[socket] == NULL) {
803 			RTE_LOG(INFO, IP_FRAG, "Creating LPM table on socket %i\n", socket);
804 			snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
805 
806 			lpm_config.max_rules = LPM_MAX_RULES;
807 			lpm_config.number_tbl8s = 256;
808 			lpm_config.flags = 0;
809 
810 			lpm = rte_lpm_create(buf, socket, &lpm_config);
811 			if (lpm == NULL) {
812 				RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
813 				return -1;
814 			}
815 			socket_lpm[socket] = lpm;
816 		}
817 
818 		if (socket_lpm6[socket] == NULL) {
819 			RTE_LOG(INFO, IP_FRAG, "Creating LPM6 table on socket %i\n", socket);
820 			snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
821 
822 			lpm6 = rte_lpm6_create(buf, socket, &lpm6_config);
823 			if (lpm6 == NULL) {
824 				RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
825 				return -1;
826 			}
827 			socket_lpm6[socket] = lpm6;
828 		}
829 	}
830 
831 	return 0;
832 }
833 
834 int
835 main(int argc, char **argv)
836 {
837 	struct lcore_queue_conf *qconf;
838 	struct rte_eth_dev_info dev_info;
839 	struct rte_eth_txconf *txconf;
840 	struct rx_queue *rxq;
841 	int socket, ret;
842 	uint16_t nb_ports;
843 	uint16_t queueid = 0;
844 	unsigned lcore_id = 0, rx_lcore_id = 0;
845 	uint32_t n_tx_queue, nb_lcores;
846 	uint16_t portid;
847 
848 	/* init EAL */
849 	ret = rte_eal_init(argc, argv);
850 	if (ret < 0)
851 		rte_exit(EXIT_FAILURE, "rte_eal_init failed");
852 	argc -= ret;
853 	argv += ret;
854 
855 	/* parse application arguments (after the EAL ones) */
856 	ret = parse_args(argc, argv);
857 	if (ret < 0)
858 		rte_exit(EXIT_FAILURE, "Invalid arguments");
859 
860 	nb_ports = rte_eth_dev_count_avail();
861 	if (nb_ports == 0)
862 		rte_exit(EXIT_FAILURE, "No ports found!\n");
863 
864 	nb_lcores = rte_lcore_count();
865 
866 	/* initialize structures (mempools, lpm etc.) */
867 	if (init_mem() < 0)
868 		rte_panic("Cannot initialize memory structures!\n");
869 
870 	/* check if portmask has non-existent ports */
871 	if (enabled_port_mask & ~(RTE_LEN2MASK(nb_ports, unsigned)))
872 		rte_exit(EXIT_FAILURE, "Non-existent ports in portmask!\n");
873 
874 	/* initialize all ports */
875 	RTE_ETH_FOREACH_DEV(portid) {
876 		struct rte_eth_conf local_port_conf = port_conf;
877 		struct rte_eth_rxconf rxq_conf;
878 
879 		/* skip ports that are not enabled */
880 		if ((enabled_port_mask & (1 << portid)) == 0) {
881 			printf("Skipping disabled port %d\n", portid);
882 			continue;
883 		}
884 
885 		qconf = &lcore_queue_conf[rx_lcore_id];
886 
887 		/* limit the frame size to the maximum supported by NIC */
888 		rte_eth_dev_info_get(portid, &dev_info);
889 		local_port_conf.rxmode.max_rx_pkt_len = RTE_MIN(
890 		    dev_info.max_rx_pktlen,
891 		    local_port_conf.rxmode.max_rx_pkt_len);
892 
893 		/* get the lcore_id for this port */
894 		while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
895 		       qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
896 
897 			rx_lcore_id ++;
898 			if (rx_lcore_id >= RTE_MAX_LCORE)
899 				rte_exit(EXIT_FAILURE, "Not enough cores\n");
900 
901 			qconf = &lcore_queue_conf[rx_lcore_id];
902 		}
903 
904 		socket = (int) rte_lcore_to_socket_id(rx_lcore_id);
905 		if (socket == SOCKET_ID_ANY)
906 			socket = 0;
907 
908 		rxq = &qconf->rx_queue_list[qconf->n_rx_queue];
909 		rxq->portid = portid;
910 		rxq->direct_pool = socket_direct_pool[socket];
911 		rxq->indirect_pool = socket_indirect_pool[socket];
912 		rxq->lpm = socket_lpm[socket];
913 		rxq->lpm6 = socket_lpm6[socket];
914 		qconf->n_rx_queue++;
915 
916 		/* init port */
917 		printf("Initializing port %d on lcore %u...", portid,
918 		       rx_lcore_id);
919 		fflush(stdout);
920 
921 		n_tx_queue = nb_lcores;
922 		if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
923 			n_tx_queue = MAX_TX_QUEUE_PER_PORT;
924 		ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
925 					    &local_port_conf);
926 		if (ret < 0) {
927 			printf("\n");
928 			rte_exit(EXIT_FAILURE, "Cannot configure device: "
929 				"err=%d, port=%d\n",
930 				ret, portid);
931 		}
932 
933 		ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
934 					    &nb_txd);
935 		if (ret < 0) {
936 			printf("\n");
937 			rte_exit(EXIT_FAILURE, "Cannot adjust number of "
938 				"descriptors: err=%d, port=%d\n", ret, portid);
939 		}
940 
941 		/* init one RX queue */
942 		rxq_conf = dev_info.default_rxconf;
943 		rxq_conf.offloads = local_port_conf.rxmode.offloads;
944 		ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
945 					     socket, &rxq_conf,
946 					     socket_direct_pool[socket]);
947 		if (ret < 0) {
948 			printf("\n");
949 			rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: "
950 				"err=%d, port=%d\n",
951 				ret, portid);
952 		}
953 
954 		rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
955 		print_ethaddr(" Address:", &ports_eth_addr[portid]);
956 		printf("\n");
957 
958 		/* init one TX queue per couple (lcore,port) */
959 		rte_eth_dev_info_get(portid, &dev_info);
960 		queueid = 0;
961 		for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
962 			if (rte_lcore_is_enabled(lcore_id) == 0)
963 				continue;
964 
965 			if (queueid >= dev_info.nb_tx_queues)
966 				break;
967 
968 			socket = (int) rte_lcore_to_socket_id(lcore_id);
969 			printf("txq=%u,%d ", lcore_id, queueid);
970 			fflush(stdout);
971 
972 			txconf = &dev_info.default_txconf;
973 			txconf->offloads = local_port_conf.txmode.offloads;
974 			ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
975 						     socket, txconf);
976 			if (ret < 0) {
977 				printf("\n");
978 				rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
979 					"err=%d, port=%d\n", ret, portid);
980 			}
981 
982 			qconf = &lcore_queue_conf[lcore_id];
983 			qconf->tx_queue_id[portid] = queueid;
984 			queueid++;
985 		}
986 
987 		printf("\n");
988 	}
989 
990 	printf("\n");
991 
992 	/* start ports */
993 	RTE_ETH_FOREACH_DEV(portid) {
994 		if ((enabled_port_mask & (1 << portid)) == 0) {
995 			continue;
996 		}
997 		/* Start device */
998 		ret = rte_eth_dev_start(portid);
999 		if (ret < 0)
1000 			rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1001 				ret, portid);
1002 
1003 		rte_eth_promiscuous_enable(portid);
1004 
1005 		if (check_ptype(portid) == 0) {
1006 			rte_eth_add_rx_callback(portid, 0, cb_parse_ptype, NULL);
1007 			printf("Add Rx callback function to detect L3 packet type by SW :"
1008 				" port = %d\n", portid);
1009 		}
1010 	}
1011 
1012 	if (init_routing_table() < 0)
1013 		rte_exit(EXIT_FAILURE, "Cannot init routing table\n");
1014 
1015 	check_all_ports_link_status(enabled_port_mask);
1016 
1017 	/* launch per-lcore init on every lcore */
1018 	rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1019 	RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1020 		if (rte_eal_wait_lcore(lcore_id) < 0)
1021 			return -1;
1022 	}
1023 
1024 	return 0;
1025 }
1026