xref: /dpdk/examples/ip_fragmentation/main.c (revision c6ef1454aee172382f3bc98666ed91985ac3ffc5)
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	RTE_ETHER_MTU
56 #define	IPV6_MTU_DEFAULT	RTE_ETHER_MTU
57 
58 /*
59  * The overhead from max frame size to MTU.
60  * We have to consider the max possible overhead.
61  */
62 #define MTU_OVERHEAD	\
63 	(RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN + \
64 		2 * sizeof(struct rte_vlan_hdr))
65 
66 /*
67  * Default payload in bytes for the IPv6 packet.
68  */
69 #define	IPV4_DEFAULT_PAYLOAD	(IPV4_MTU_DEFAULT - sizeof(struct rte_ipv4_hdr))
70 #define	IPV6_DEFAULT_PAYLOAD	(IPV6_MTU_DEFAULT - sizeof(struct rte_ipv6_hdr))
71 
72 /*
73  * Max number of fragments per packet expected - defined by config file.
74  */
75 #define	MAX_PACKET_FRAG RTE_LIBRTE_IP_FRAG_MAX_FRAG
76 
77 #define NB_MBUF   8192
78 
79 #define MAX_PKT_BURST	32
80 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
81 
82 /* Configure how many packets ahead to prefetch, when reading packets */
83 #define PREFETCH_OFFSET	3
84 
85 /*
86  * Configurable number of RX/TX ring descriptors
87  */
88 #define RTE_TEST_RX_DESC_DEFAULT 1024
89 #define RTE_TEST_TX_DESC_DEFAULT 1024
90 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
91 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
92 
93 /* ethernet addresses of ports */
94 static struct rte_ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
95 
96 #ifndef IPv4_BYTES
97 #define IPv4_BYTES_FMT "%" PRIu8 ".%" PRIu8 ".%" PRIu8 ".%" PRIu8
98 #define IPv4_BYTES(addr) \
99 		(uint8_t) (((addr) >> 24) & 0xFF),\
100 		(uint8_t) (((addr) >> 16) & 0xFF),\
101 		(uint8_t) (((addr) >> 8) & 0xFF),\
102 		(uint8_t) ((addr) & 0xFF)
103 #endif
104 
105 #ifndef IPv6_BYTES
106 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
107                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
108 #define IPv6_BYTES(addr) \
109 	addr[0],  addr[1], addr[2],  addr[3], \
110 	addr[4],  addr[5], addr[6],  addr[7], \
111 	addr[8],  addr[9], addr[10], addr[11],\
112 	addr[12], addr[13],addr[14], addr[15]
113 #endif
114 
115 #define IPV6_ADDR_LEN 16
116 
117 /* mask of enabled ports */
118 static int enabled_port_mask = 0;
119 
120 static int rx_queue_per_lcore = 1;
121 
122 #define MBUF_TABLE_SIZE  (2 * MAX(MAX_PKT_BURST, MAX_PACKET_FRAG))
123 
124 struct mbuf_table {
125 	uint16_t len;
126 	struct rte_mbuf *m_table[MBUF_TABLE_SIZE];
127 };
128 
129 struct rx_queue {
130 	struct rte_mempool *direct_pool;
131 	struct rte_mempool *indirect_pool;
132 	struct rte_lpm *lpm;
133 	struct rte_lpm6 *lpm6;
134 	uint16_t portid;
135 };
136 
137 #define MAX_RX_QUEUE_PER_LCORE 16
138 #define MAX_TX_QUEUE_PER_PORT 16
139 struct lcore_queue_conf {
140 	uint16_t n_rx_queue;
141 	uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
142 	struct rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
143 	struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
144 } __rte_cache_aligned;
145 struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
146 
147 static struct rte_eth_conf port_conf = {
148 	.rxmode = {
149 		.max_rx_pkt_len = JUMBO_FRAME_MAX_SIZE,
150 		.split_hdr_size = 0,
151 		.offloads = (DEV_RX_OFFLOAD_CHECKSUM |
152 			     DEV_RX_OFFLOAD_SCATTER |
153 			     DEV_RX_OFFLOAD_JUMBO_FRAME),
154 	},
155 	.txmode = {
156 		.mq_mode = ETH_MQ_TX_NONE,
157 		.offloads = (DEV_TX_OFFLOAD_IPV4_CKSUM |
158 			     DEV_TX_OFFLOAD_MULTI_SEGS),
159 	},
160 };
161 
162 /*
163  * IPv4 forwarding table
164  */
165 struct l3fwd_ipv4_route {
166 	uint32_t ip;
167 	uint8_t  depth;
168 	uint8_t  if_out;
169 };
170 
171 struct l3fwd_ipv4_route l3fwd_ipv4_route_array[] = {
172 		{RTE_IPV4(100,10,0,0), 16, 0},
173 		{RTE_IPV4(100,20,0,0), 16, 1},
174 		{RTE_IPV4(100,30,0,0), 16, 2},
175 		{RTE_IPV4(100,40,0,0), 16, 3},
176 		{RTE_IPV4(100,50,0,0), 16, 4},
177 		{RTE_IPV4(100,60,0,0), 16, 5},
178 		{RTE_IPV4(100,70,0,0), 16, 6},
179 		{RTE_IPV4(100,80,0,0), 16, 7},
180 };
181 
182 /*
183  * IPv6 forwarding table
184  */
185 
186 struct l3fwd_ipv6_route {
187 	uint8_t ip[IPV6_ADDR_LEN];
188 	uint8_t depth;
189 	uint8_t if_out;
190 };
191 
192 static struct l3fwd_ipv6_route l3fwd_ipv6_route_array[] = {
193 	{{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
194 	{{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
195 	{{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
196 	{{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
197 	{{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
198 	{{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
199 	{{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
200 	{{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
201 };
202 
203 #define LPM_MAX_RULES         1024
204 #define LPM6_MAX_RULES         1024
205 #define LPM6_NUMBER_TBL8S (1 << 16)
206 
207 struct rte_lpm6_config lpm6_config = {
208 		.max_rules = LPM6_MAX_RULES,
209 		.number_tbl8s = LPM6_NUMBER_TBL8S,
210 		.flags = 0
211 };
212 
213 static struct rte_mempool *socket_direct_pool[RTE_MAX_NUMA_NODES];
214 static struct rte_mempool *socket_indirect_pool[RTE_MAX_NUMA_NODES];
215 static struct rte_lpm *socket_lpm[RTE_MAX_NUMA_NODES];
216 static struct rte_lpm6 *socket_lpm6[RTE_MAX_NUMA_NODES];
217 
218 /* Send burst of packets on an output interface */
219 static inline int
220 send_burst(struct lcore_queue_conf *qconf, uint16_t n, uint16_t port)
221 {
222 	struct rte_mbuf **m_table;
223 	int ret;
224 	uint16_t queueid;
225 
226 	queueid = qconf->tx_queue_id[port];
227 	m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
228 
229 	ret = rte_eth_tx_burst(port, queueid, m_table, n);
230 	if (unlikely(ret < n)) {
231 		do {
232 			rte_pktmbuf_free(m_table[ret]);
233 		} while (++ret < n);
234 	}
235 
236 	return 0;
237 }
238 
239 static inline void
240 l3fwd_simple_forward(struct rte_mbuf *m, struct lcore_queue_conf *qconf,
241 		uint8_t queueid, uint16_t port_in)
242 {
243 	struct rx_queue *rxq;
244 	uint32_t i, len, next_hop;
245 	uint8_t ipv6;
246 	uint16_t port_out;
247 	int32_t len2;
248 
249 	ipv6 = 0;
250 	rxq = &qconf->rx_queue_list[queueid];
251 
252 	/* by default, send everything back to the source port */
253 	port_out = port_in;
254 
255 	/* Remove the Ethernet header and trailer from the input packet */
256 	rte_pktmbuf_adj(m, (uint16_t)sizeof(struct rte_ether_hdr));
257 
258 	/* Build transmission burst */
259 	len = qconf->tx_mbufs[port_out].len;
260 
261 	/* if this is an IPv4 packet */
262 	if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
263 		struct rte_ipv4_hdr *ip_hdr;
264 		uint32_t ip_dst;
265 		/* Read the lookup key (i.e. ip_dst) from the input packet */
266 		ip_hdr = rte_pktmbuf_mtod(m, struct rte_ipv4_hdr *);
267 		ip_dst = rte_be_to_cpu_32(ip_hdr->dst_addr);
268 
269 		/* Find destination port */
270 		if (rte_lpm_lookup(rxq->lpm, ip_dst, &next_hop) == 0 &&
271 				(enabled_port_mask & 1 << next_hop) != 0) {
272 			port_out = next_hop;
273 
274 			/* Build transmission burst for new port */
275 			len = qconf->tx_mbufs[port_out].len;
276 		}
277 
278 		/* if we don't need to do any fragmentation */
279 		if (likely (IPV4_MTU_DEFAULT >= m->pkt_len)) {
280 			qconf->tx_mbufs[port_out].m_table[len] = m;
281 			len2 = 1;
282 		} else {
283 			len2 = rte_ipv4_fragment_packet(m,
284 				&qconf->tx_mbufs[port_out].m_table[len],
285 				(uint16_t)(MBUF_TABLE_SIZE - len),
286 				IPV4_MTU_DEFAULT,
287 				rxq->direct_pool, rxq->indirect_pool);
288 
289 			/* Free input packet */
290 			rte_pktmbuf_free(m);
291 
292 			/* If we fail to fragment the packet */
293 			if (unlikely (len2 < 0))
294 				return;
295 		}
296 	} else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
297 		/* if this is an IPv6 packet */
298 		struct rte_ipv6_hdr *ip_hdr;
299 
300 		ipv6 = 1;
301 
302 		/* Read the lookup key (i.e. ip_dst) from the input packet */
303 		ip_hdr = rte_pktmbuf_mtod(m, struct rte_ipv6_hdr *);
304 
305 		/* Find destination port */
306 		if (rte_lpm6_lookup(rxq->lpm6, ip_hdr->dst_addr,
307 						&next_hop) == 0 &&
308 				(enabled_port_mask & 1 << next_hop) != 0) {
309 			port_out = next_hop;
310 
311 			/* Build transmission burst for new port */
312 			len = qconf->tx_mbufs[port_out].len;
313 		}
314 
315 		/* if we don't need to do any fragmentation */
316 		if (likely (IPV6_MTU_DEFAULT >= m->pkt_len)) {
317 			qconf->tx_mbufs[port_out].m_table[len] = m;
318 			len2 = 1;
319 		} else {
320 			len2 = rte_ipv6_fragment_packet(m,
321 				&qconf->tx_mbufs[port_out].m_table[len],
322 				(uint16_t)(MBUF_TABLE_SIZE - len),
323 				IPV6_MTU_DEFAULT,
324 				rxq->direct_pool, rxq->indirect_pool);
325 
326 			/* Free input packet */
327 			rte_pktmbuf_free(m);
328 
329 			/* If we fail to fragment the packet */
330 			if (unlikely (len2 < 0))
331 				return;
332 		}
333 	}
334 	/* else, just forward the packet */
335 	else {
336 		qconf->tx_mbufs[port_out].m_table[len] = m;
337 		len2 = 1;
338 	}
339 
340 	for (i = len; i < len + len2; i ++) {
341 		void *d_addr_bytes;
342 
343 		m = qconf->tx_mbufs[port_out].m_table[i];
344 		struct rte_ether_hdr *eth_hdr = (struct rte_ether_hdr *)
345 			rte_pktmbuf_prepend(m,
346 				(uint16_t)sizeof(struct rte_ether_hdr));
347 		if (eth_hdr == NULL) {
348 			rte_panic("No headroom in mbuf.\n");
349 		}
350 
351 		m->l2_len = sizeof(struct rte_ether_hdr);
352 
353 		/* 02:00:00:00:00:xx */
354 		d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
355 		*((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)port_out << 40);
356 
357 		/* src addr */
358 		rte_ether_addr_copy(&ports_eth_addr[port_out],
359 				&eth_hdr->s_addr);
360 		if (ipv6) {
361 			eth_hdr->ether_type =
362 				rte_be_to_cpu_16(RTE_ETHER_TYPE_IPV6);
363 		} else {
364 			eth_hdr->ether_type =
365 				rte_be_to_cpu_16(RTE_ETHER_TYPE_IPV4);
366 			m->ol_flags |= (PKT_TX_IPV4 | PKT_TX_IP_CKSUM);
367 		}
368 	}
369 
370 	len += len2;
371 
372 	if (likely(len < MAX_PKT_BURST)) {
373 		qconf->tx_mbufs[port_out].len = (uint16_t)len;
374 		return;
375 	}
376 
377 	/* Transmit packets */
378 	send_burst(qconf, (uint16_t)len, port_out);
379 	qconf->tx_mbufs[port_out].len = 0;
380 }
381 
382 /* main processing loop */
383 static int
384 main_loop(__attribute__((unused)) void *dummy)
385 {
386 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
387 	unsigned lcore_id;
388 	uint64_t prev_tsc, diff_tsc, cur_tsc;
389 	int i, j, nb_rx;
390 	uint16_t portid;
391 	struct lcore_queue_conf *qconf;
392 	const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
393 
394 	prev_tsc = 0;
395 
396 	lcore_id = rte_lcore_id();
397 	qconf = &lcore_queue_conf[lcore_id];
398 
399 	if (qconf->n_rx_queue == 0) {
400 		RTE_LOG(INFO, IP_FRAG, "lcore %u has nothing to do\n", lcore_id);
401 		return 0;
402 	}
403 
404 	RTE_LOG(INFO, IP_FRAG, "entering main loop on lcore %u\n", lcore_id);
405 
406 	for (i = 0; i < qconf->n_rx_queue; i++) {
407 
408 		portid = qconf->rx_queue_list[i].portid;
409 		RTE_LOG(INFO, IP_FRAG, " -- lcoreid=%u portid=%d\n", lcore_id,
410 				portid);
411 	}
412 
413 	while (1) {
414 
415 		cur_tsc = rte_rdtsc();
416 
417 		/*
418 		 * TX burst queue drain
419 		 */
420 		diff_tsc = cur_tsc - prev_tsc;
421 		if (unlikely(diff_tsc > drain_tsc)) {
422 
423 			/*
424 			 * This could be optimized (use queueid instead of
425 			 * portid), but it is not called so often
426 			 */
427 			for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
428 				if (qconf->tx_mbufs[portid].len == 0)
429 					continue;
430 				send_burst(&lcore_queue_conf[lcore_id],
431 					   qconf->tx_mbufs[portid].len,
432 					   portid);
433 				qconf->tx_mbufs[portid].len = 0;
434 			}
435 
436 			prev_tsc = cur_tsc;
437 		}
438 
439 		/*
440 		 * Read packet from RX queues
441 		 */
442 		for (i = 0; i < qconf->n_rx_queue; i++) {
443 
444 			portid = qconf->rx_queue_list[i].portid;
445 			nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
446 						 MAX_PKT_BURST);
447 
448 			/* Prefetch first packets */
449 			for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
450 				rte_prefetch0(rte_pktmbuf_mtod(
451 						pkts_burst[j], void *));
452 			}
453 
454 			/* Prefetch and forward already prefetched packets */
455 			for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
456 				rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
457 						j + PREFETCH_OFFSET], void *));
458 				l3fwd_simple_forward(pkts_burst[j], qconf, i, portid);
459 			}
460 
461 			/* Forward remaining prefetched packets */
462 			for (; j < nb_rx; j++) {
463 				l3fwd_simple_forward(pkts_burst[j], qconf, i, portid);
464 			}
465 		}
466 	}
467 }
468 
469 /* display usage */
470 static void
471 print_usage(const char *prgname)
472 {
473 	printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
474 	       "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
475 	       "  -q NQ: number of queue (=ports) per lcore (default is 1)\n",
476 	       prgname);
477 }
478 
479 static int
480 parse_portmask(const char *portmask)
481 {
482 	char *end = NULL;
483 	unsigned long pm;
484 
485 	/* parse hexadecimal string */
486 	pm = strtoul(portmask, &end, 16);
487 	if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
488 		return -1;
489 
490 	if (pm == 0)
491 		return -1;
492 
493 	return pm;
494 }
495 
496 static int
497 parse_nqueue(const char *q_arg)
498 {
499 	char *end = NULL;
500 	unsigned long n;
501 
502 	/* parse hexadecimal string */
503 	n = strtoul(q_arg, &end, 10);
504 	if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
505 		return -1;
506 	if (n == 0)
507 		return -1;
508 	if (n >= MAX_RX_QUEUE_PER_LCORE)
509 		return -1;
510 
511 	return n;
512 }
513 
514 /* Parse the argument given in the command line of the application */
515 static int
516 parse_args(int argc, char **argv)
517 {
518 	int opt, ret;
519 	char **argvopt;
520 	int option_index;
521 	char *prgname = argv[0];
522 	static struct option lgopts[] = {
523 		{NULL, 0, 0, 0}
524 	};
525 
526 	argvopt = argv;
527 
528 	while ((opt = getopt_long(argc, argvopt, "p:q:",
529 				  lgopts, &option_index)) != EOF) {
530 
531 		switch (opt) {
532 		/* portmask */
533 		case 'p':
534 			enabled_port_mask = parse_portmask(optarg);
535 			if (enabled_port_mask < 0) {
536 				printf("invalid portmask\n");
537 				print_usage(prgname);
538 				return -1;
539 			}
540 			break;
541 
542 		/* nqueue */
543 		case 'q':
544 			rx_queue_per_lcore = parse_nqueue(optarg);
545 			if (rx_queue_per_lcore < 0) {
546 				printf("invalid queue number\n");
547 				print_usage(prgname);
548 				return -1;
549 			}
550 			break;
551 
552 		/* long options */
553 		case 0:
554 			print_usage(prgname);
555 			return -1;
556 
557 		default:
558 			print_usage(prgname);
559 			return -1;
560 		}
561 	}
562 
563 	if (enabled_port_mask == 0) {
564 		printf("portmask not specified\n");
565 		print_usage(prgname);
566 		return -1;
567 	}
568 
569 	if (optind >= 0)
570 		argv[optind-1] = prgname;
571 
572 	ret = optind-1;
573 	optind = 1; /* reset getopt lib */
574 	return ret;
575 }
576 
577 static void
578 print_ethaddr(const char *name, struct rte_ether_addr *eth_addr)
579 {
580 	char buf[RTE_ETHER_ADDR_FMT_SIZE];
581 	rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
582 	printf("%s%s", name, buf);
583 }
584 
585 /* Check the link status of all ports in up to 9s, and print them finally */
586 static void
587 check_all_ports_link_status(uint32_t port_mask)
588 {
589 #define CHECK_INTERVAL 100 /* 100ms */
590 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
591 	uint16_t portid;
592 	uint8_t count, all_ports_up, print_flag = 0;
593 	struct rte_eth_link link;
594 
595 	printf("\nChecking link status");
596 	fflush(stdout);
597 	for (count = 0; count <= MAX_CHECK_TIME; count++) {
598 		all_ports_up = 1;
599 		RTE_ETH_FOREACH_DEV(portid) {
600 			if ((port_mask & (1 << portid)) == 0)
601 				continue;
602 			memset(&link, 0, sizeof(link));
603 			rte_eth_link_get_nowait(portid, &link);
604 			/* print link status if flag set */
605 			if (print_flag == 1) {
606 				if (link.link_status)
607 					printf(
608 					"Port%d Link Up .Speed %u Mbps - %s\n",
609 						portid, link.link_speed,
610 				(link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
611 					("full-duplex") : ("half-duplex\n"));
612 				else
613 					printf("Port %d Link Down\n", portid);
614 				continue;
615 			}
616 			/* clear all_ports_up flag if any link down */
617 			if (link.link_status == ETH_LINK_DOWN) {
618 				all_ports_up = 0;
619 				break;
620 			}
621 		}
622 		/* after finally printing all link status, get out */
623 		if (print_flag == 1)
624 			break;
625 
626 		if (all_ports_up == 0) {
627 			printf(".");
628 			fflush(stdout);
629 			rte_delay_ms(CHECK_INTERVAL);
630 		}
631 
632 		/* set the print_flag if all ports up or timeout */
633 		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
634 			print_flag = 1;
635 			printf("\ndone\n");
636 		}
637 	}
638 }
639 
640 /* Check L3 packet type detection capablity of the NIC port */
641 static int
642 check_ptype(int portid)
643 {
644 	int i, ret;
645 	int ptype_l3_ipv4 = 0, ptype_l3_ipv6 = 0;
646 	uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
647 
648 	ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
649 	if (ret <= 0)
650 		return 0;
651 
652 	uint32_t ptypes[ret];
653 
654 	ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
655 	for (i = 0; i < ret; ++i) {
656 		if (ptypes[i] & RTE_PTYPE_L3_IPV4)
657 			ptype_l3_ipv4 = 1;
658 		if (ptypes[i] & RTE_PTYPE_L3_IPV6)
659 			ptype_l3_ipv6 = 1;
660 	}
661 
662 	if (ptype_l3_ipv4 == 0)
663 		printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
664 
665 	if (ptype_l3_ipv6 == 0)
666 		printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
667 
668 	if (ptype_l3_ipv4 && ptype_l3_ipv6)
669 		return 1;
670 
671 	return 0;
672 
673 }
674 
675 /* Parse packet type of a packet by SW */
676 static inline void
677 parse_ptype(struct rte_mbuf *m)
678 {
679 	struct rte_ether_hdr *eth_hdr;
680 	uint32_t packet_type = RTE_PTYPE_UNKNOWN;
681 	uint16_t ether_type;
682 
683 	eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
684 	ether_type = eth_hdr->ether_type;
685 	if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4))
686 		packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
687 	else if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6))
688 		packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
689 
690 	m->packet_type = packet_type;
691 }
692 
693 /* callback function to detect packet type for a queue of a port */
694 static uint16_t
695 cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
696 		   struct rte_mbuf *pkts[], uint16_t nb_pkts,
697 		   uint16_t max_pkts __rte_unused,
698 		   void *user_param __rte_unused)
699 {
700 	uint16_t i;
701 
702 	for (i = 0; i < nb_pkts; ++i)
703 		parse_ptype(pkts[i]);
704 
705 	return nb_pkts;
706 }
707 
708 static int
709 init_routing_table(void)
710 {
711 	struct rte_lpm *lpm;
712 	struct rte_lpm6 *lpm6;
713 	int socket, ret;
714 	unsigned i;
715 
716 	for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) {
717 		if (socket_lpm[socket]) {
718 			lpm = socket_lpm[socket];
719 			/* populate the LPM table */
720 			for (i = 0; i < RTE_DIM(l3fwd_ipv4_route_array); i++) {
721 				ret = rte_lpm_add(lpm,
722 					l3fwd_ipv4_route_array[i].ip,
723 					l3fwd_ipv4_route_array[i].depth,
724 					l3fwd_ipv4_route_array[i].if_out);
725 
726 				if (ret < 0) {
727 					RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd "
728 						"LPM table\n", i);
729 					return -1;
730 				}
731 
732 				RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv4_BYTES_FMT
733 						"/%d (port %d)\n",
734 					socket,
735 					IPv4_BYTES(l3fwd_ipv4_route_array[i].ip),
736 					l3fwd_ipv4_route_array[i].depth,
737 					l3fwd_ipv4_route_array[i].if_out);
738 			}
739 		}
740 
741 		if (socket_lpm6[socket]) {
742 			lpm6 = socket_lpm6[socket];
743 			/* populate the LPM6 table */
744 			for (i = 0; i < RTE_DIM(l3fwd_ipv6_route_array); i++) {
745 				ret = rte_lpm6_add(lpm6,
746 					l3fwd_ipv6_route_array[i].ip,
747 					l3fwd_ipv6_route_array[i].depth,
748 					l3fwd_ipv6_route_array[i].if_out);
749 
750 				if (ret < 0) {
751 					RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd "
752 						"LPM6 table\n", i);
753 					return -1;
754 				}
755 
756 				RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv6_BYTES_FMT
757 						"/%d (port %d)\n",
758 					socket,
759 					IPv6_BYTES(l3fwd_ipv6_route_array[i].ip),
760 					l3fwd_ipv6_route_array[i].depth,
761 					l3fwd_ipv6_route_array[i].if_out);
762 			}
763 		}
764 	}
765 	return 0;
766 }
767 
768 static int
769 init_mem(void)
770 {
771 	char buf[PATH_MAX];
772 	struct rte_mempool *mp;
773 	struct rte_lpm *lpm;
774 	struct rte_lpm6 *lpm6;
775 	struct rte_lpm_config lpm_config;
776 	int socket;
777 	unsigned lcore_id;
778 
779 	/* traverse through lcores and initialize structures on each socket */
780 
781 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
782 
783 		if (rte_lcore_is_enabled(lcore_id) == 0)
784 			continue;
785 
786 		socket = rte_lcore_to_socket_id(lcore_id);
787 
788 		if (socket == SOCKET_ID_ANY)
789 			socket = 0;
790 
791 		if (socket_direct_pool[socket] == NULL) {
792 			RTE_LOG(INFO, IP_FRAG, "Creating direct mempool on socket %i\n",
793 					socket);
794 			snprintf(buf, sizeof(buf), "pool_direct_%i", socket);
795 
796 			mp = rte_pktmbuf_pool_create(buf, NB_MBUF, 32,
797 				0, RTE_MBUF_DEFAULT_BUF_SIZE, socket);
798 			if (mp == NULL) {
799 				RTE_LOG(ERR, IP_FRAG, "Cannot create direct mempool\n");
800 				return -1;
801 			}
802 			socket_direct_pool[socket] = mp;
803 		}
804 
805 		if (socket_indirect_pool[socket] == NULL) {
806 			RTE_LOG(INFO, IP_FRAG, "Creating indirect mempool on socket %i\n",
807 					socket);
808 			snprintf(buf, sizeof(buf), "pool_indirect_%i", socket);
809 
810 			mp = rte_pktmbuf_pool_create(buf, NB_MBUF, 32, 0, 0,
811 				socket);
812 			if (mp == NULL) {
813 				RTE_LOG(ERR, IP_FRAG, "Cannot create indirect mempool\n");
814 				return -1;
815 			}
816 			socket_indirect_pool[socket] = mp;
817 		}
818 
819 		if (socket_lpm[socket] == NULL) {
820 			RTE_LOG(INFO, IP_FRAG, "Creating LPM table on socket %i\n", socket);
821 			snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
822 
823 			lpm_config.max_rules = LPM_MAX_RULES;
824 			lpm_config.number_tbl8s = 256;
825 			lpm_config.flags = 0;
826 
827 			lpm = rte_lpm_create(buf, socket, &lpm_config);
828 			if (lpm == NULL) {
829 				RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
830 				return -1;
831 			}
832 			socket_lpm[socket] = lpm;
833 		}
834 
835 		if (socket_lpm6[socket] == NULL) {
836 			RTE_LOG(INFO, IP_FRAG, "Creating LPM6 table on socket %i\n", socket);
837 			snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
838 
839 			lpm6 = rte_lpm6_create(buf, socket, &lpm6_config);
840 			if (lpm6 == NULL) {
841 				RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
842 				return -1;
843 			}
844 			socket_lpm6[socket] = lpm6;
845 		}
846 	}
847 
848 	return 0;
849 }
850 
851 int
852 main(int argc, char **argv)
853 {
854 	struct lcore_queue_conf *qconf;
855 	struct rte_eth_dev_info dev_info;
856 	struct rte_eth_txconf *txconf;
857 	struct rx_queue *rxq;
858 	int socket, ret;
859 	uint16_t nb_ports;
860 	uint16_t queueid = 0;
861 	unsigned lcore_id = 0, rx_lcore_id = 0;
862 	uint32_t n_tx_queue, nb_lcores;
863 	uint16_t portid;
864 
865 	/* init EAL */
866 	ret = rte_eal_init(argc, argv);
867 	if (ret < 0)
868 		rte_exit(EXIT_FAILURE, "rte_eal_init failed");
869 	argc -= ret;
870 	argv += ret;
871 
872 	/* parse application arguments (after the EAL ones) */
873 	ret = parse_args(argc, argv);
874 	if (ret < 0)
875 		rte_exit(EXIT_FAILURE, "Invalid arguments");
876 
877 	nb_ports = rte_eth_dev_count_avail();
878 	if (nb_ports == 0)
879 		rte_exit(EXIT_FAILURE, "No ports found!\n");
880 
881 	nb_lcores = rte_lcore_count();
882 
883 	/* initialize structures (mempools, lpm etc.) */
884 	if (init_mem() < 0)
885 		rte_panic("Cannot initialize memory structures!\n");
886 
887 	/* check if portmask has non-existent ports */
888 	if (enabled_port_mask & ~(RTE_LEN2MASK(nb_ports, unsigned)))
889 		rte_exit(EXIT_FAILURE, "Non-existent ports in portmask!\n");
890 
891 	/* initialize all ports */
892 	RTE_ETH_FOREACH_DEV(portid) {
893 		struct rte_eth_conf local_port_conf = port_conf;
894 		struct rte_eth_rxconf rxq_conf;
895 
896 		/* skip ports that are not enabled */
897 		if ((enabled_port_mask & (1 << portid)) == 0) {
898 			printf("Skipping disabled port %d\n", portid);
899 			continue;
900 		}
901 
902 		qconf = &lcore_queue_conf[rx_lcore_id];
903 
904 		/* limit the frame size to the maximum supported by NIC */
905 		rte_eth_dev_info_get(portid, &dev_info);
906 		local_port_conf.rxmode.max_rx_pkt_len = RTE_MIN(
907 		    dev_info.max_rx_pktlen,
908 		    local_port_conf.rxmode.max_rx_pkt_len);
909 
910 		/* get the lcore_id for this port */
911 		while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
912 		       qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
913 
914 			rx_lcore_id ++;
915 			if (rx_lcore_id >= RTE_MAX_LCORE)
916 				rte_exit(EXIT_FAILURE, "Not enough cores\n");
917 
918 			qconf = &lcore_queue_conf[rx_lcore_id];
919 		}
920 
921 		socket = (int) rte_lcore_to_socket_id(rx_lcore_id);
922 		if (socket == SOCKET_ID_ANY)
923 			socket = 0;
924 
925 		rxq = &qconf->rx_queue_list[qconf->n_rx_queue];
926 		rxq->portid = portid;
927 		rxq->direct_pool = socket_direct_pool[socket];
928 		rxq->indirect_pool = socket_indirect_pool[socket];
929 		rxq->lpm = socket_lpm[socket];
930 		rxq->lpm6 = socket_lpm6[socket];
931 		qconf->n_rx_queue++;
932 
933 		/* init port */
934 		printf("Initializing port %d on lcore %u...", portid,
935 		       rx_lcore_id);
936 		fflush(stdout);
937 
938 		n_tx_queue = nb_lcores;
939 		if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
940 			n_tx_queue = MAX_TX_QUEUE_PER_PORT;
941 		if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
942 			local_port_conf.txmode.offloads |=
943 				DEV_TX_OFFLOAD_MBUF_FAST_FREE;
944 		ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
945 					    &local_port_conf);
946 		if (ret < 0) {
947 			printf("\n");
948 			rte_exit(EXIT_FAILURE, "Cannot configure device: "
949 				"err=%d, port=%d\n",
950 				ret, portid);
951 		}
952 
953 		/* set the mtu to the maximum received packet size */
954 		ret = rte_eth_dev_set_mtu(portid,
955 			local_port_conf.rxmode.max_rx_pkt_len - MTU_OVERHEAD);
956 		if (ret < 0) {
957 			printf("\n");
958 			rte_exit(EXIT_FAILURE, "Set MTU failed: "
959 				"err=%d, port=%d\n",
960 			ret, portid);
961 		}
962 
963 		ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
964 					    &nb_txd);
965 		if (ret < 0) {
966 			printf("\n");
967 			rte_exit(EXIT_FAILURE, "Cannot adjust number of "
968 				"descriptors: err=%d, port=%d\n", ret, portid);
969 		}
970 
971 		/* init one RX queue */
972 		rxq_conf = dev_info.default_rxconf;
973 		rxq_conf.offloads = local_port_conf.rxmode.offloads;
974 		ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
975 					     socket, &rxq_conf,
976 					     socket_direct_pool[socket]);
977 		if (ret < 0) {
978 			printf("\n");
979 			rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: "
980 				"err=%d, port=%d\n",
981 				ret, portid);
982 		}
983 
984 		rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
985 		print_ethaddr(" Address:", &ports_eth_addr[portid]);
986 		printf("\n");
987 
988 		/* init one TX queue per couple (lcore,port) */
989 		queueid = 0;
990 		for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
991 			if (rte_lcore_is_enabled(lcore_id) == 0)
992 				continue;
993 
994 			if (queueid >= rte_eth_devices[portid].data->nb_tx_queues)
995 				break;
996 
997 			socket = (int) rte_lcore_to_socket_id(lcore_id);
998 			printf("txq=%u,%d ", lcore_id, queueid);
999 			fflush(stdout);
1000 
1001 			txconf = &dev_info.default_txconf;
1002 			txconf->offloads = local_port_conf.txmode.offloads;
1003 			ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1004 						     socket, txconf);
1005 			if (ret < 0) {
1006 				printf("\n");
1007 				rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
1008 					"err=%d, port=%d\n", ret, portid);
1009 			}
1010 
1011 			qconf = &lcore_queue_conf[lcore_id];
1012 			qconf->tx_queue_id[portid] = queueid;
1013 			queueid++;
1014 		}
1015 
1016 		printf("\n");
1017 	}
1018 
1019 	printf("\n");
1020 
1021 	/* start ports */
1022 	RTE_ETH_FOREACH_DEV(portid) {
1023 		if ((enabled_port_mask & (1 << portid)) == 0) {
1024 			continue;
1025 		}
1026 		/* Start device */
1027 		ret = rte_eth_dev_start(portid);
1028 		if (ret < 0)
1029 			rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1030 				ret, portid);
1031 
1032 		rte_eth_promiscuous_enable(portid);
1033 
1034 		if (check_ptype(portid) == 0) {
1035 			rte_eth_add_rx_callback(portid, 0, cb_parse_ptype, NULL);
1036 			printf("Add Rx callback function to detect L3 packet type by SW :"
1037 				" port = %d\n", portid);
1038 		}
1039 	}
1040 
1041 	if (init_routing_table() < 0)
1042 		rte_exit(EXIT_FAILURE, "Cannot init routing table\n");
1043 
1044 	check_all_ports_link_status(enabled_port_mask);
1045 
1046 	/* launch per-lcore init on every lcore */
1047 	rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1048 	RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1049 		if (rte_eal_wait_lcore(lcore_id) < 0)
1050 			return -1;
1051 	}
1052 
1053 	return 0;
1054 }
1055