xref: /dpdk/examples/ip_fragmentation/main.c (revision c39d1e082a4b426e915074ce30eb6f410ee2654a)
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 	uint16_t port_out, ether_type;
246 	int32_t len2;
247 	uint64_t ol_flags;
248 	const struct rte_ether_hdr *eth;
249 
250 	ol_flags = 0;
251 	rxq = &qconf->rx_queue_list[queueid];
252 
253 	/* by default, send everything back to the source port */
254 	port_out = port_in;
255 
256 	/* save ether type of the incoming packet */
257 	eth = rte_pktmbuf_mtod(m, const struct rte_ether_hdr *);
258 	ether_type = eth->ether_type;
259 
260 	/* Remove the Ethernet header and trailer from the input packet */
261 	rte_pktmbuf_adj(m, (uint16_t)sizeof(struct rte_ether_hdr));
262 
263 	/* Build transmission burst */
264 	len = qconf->tx_mbufs[port_out].len;
265 
266 	/* if this is an IPv4 packet */
267 	if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
268 		struct rte_ipv4_hdr *ip_hdr;
269 		uint32_t ip_dst;
270 		/* Read the lookup key (i.e. ip_dst) from the input packet */
271 		ip_hdr = rte_pktmbuf_mtod(m, struct rte_ipv4_hdr *);
272 		ip_dst = rte_be_to_cpu_32(ip_hdr->dst_addr);
273 
274 		/* Find destination port */
275 		if (rte_lpm_lookup(rxq->lpm, ip_dst, &next_hop) == 0 &&
276 				(enabled_port_mask & 1 << next_hop) != 0) {
277 			port_out = next_hop;
278 
279 			/* Build transmission burst for new port */
280 			len = qconf->tx_mbufs[port_out].len;
281 		}
282 
283 		/* if we don't need to do any fragmentation */
284 		if (likely (IPV4_MTU_DEFAULT >= m->pkt_len)) {
285 			qconf->tx_mbufs[port_out].m_table[len] = m;
286 			len2 = 1;
287 		} else {
288 			len2 = rte_ipv4_fragment_packet(m,
289 				&qconf->tx_mbufs[port_out].m_table[len],
290 				(uint16_t)(MBUF_TABLE_SIZE - len),
291 				IPV4_MTU_DEFAULT,
292 				rxq->direct_pool, rxq->indirect_pool);
293 
294 			/* Free input packet */
295 			rte_pktmbuf_free(m);
296 
297 			/* request HW to regenerate IPv4 cksum */
298 			ol_flags |= (PKT_TX_IPV4 | PKT_TX_IP_CKSUM);
299 
300 			/* If we fail to fragment the packet */
301 			if (unlikely (len2 < 0))
302 				return;
303 		}
304 	} else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
305 		/* if this is an IPv6 packet */
306 		struct rte_ipv6_hdr *ip_hdr;
307 
308 		/* Read the lookup key (i.e. ip_dst) from the input packet */
309 		ip_hdr = rte_pktmbuf_mtod(m, struct rte_ipv6_hdr *);
310 
311 		/* Find destination port */
312 		if (rte_lpm6_lookup(rxq->lpm6, ip_hdr->dst_addr,
313 						&next_hop) == 0 &&
314 				(enabled_port_mask & 1 << next_hop) != 0) {
315 			port_out = next_hop;
316 
317 			/* Build transmission burst for new port */
318 			len = qconf->tx_mbufs[port_out].len;
319 		}
320 
321 		/* if we don't need to do any fragmentation */
322 		if (likely (IPV6_MTU_DEFAULT >= m->pkt_len)) {
323 			qconf->tx_mbufs[port_out].m_table[len] = m;
324 			len2 = 1;
325 		} else {
326 			len2 = rte_ipv6_fragment_packet(m,
327 				&qconf->tx_mbufs[port_out].m_table[len],
328 				(uint16_t)(MBUF_TABLE_SIZE - len),
329 				IPV6_MTU_DEFAULT,
330 				rxq->direct_pool, rxq->indirect_pool);
331 
332 			/* Free input packet */
333 			rte_pktmbuf_free(m);
334 
335 			/* If we fail to fragment the packet */
336 			if (unlikely (len2 < 0))
337 				return;
338 		}
339 	}
340 	/* else, just forward the packet */
341 	else {
342 		qconf->tx_mbufs[port_out].m_table[len] = m;
343 		len2 = 1;
344 	}
345 
346 	for (i = len; i < len + len2; i ++) {
347 		void *d_addr_bytes;
348 
349 		m = qconf->tx_mbufs[port_out].m_table[i];
350 		struct rte_ether_hdr *eth_hdr = (struct rte_ether_hdr *)
351 			rte_pktmbuf_prepend(m,
352 				(uint16_t)sizeof(struct rte_ether_hdr));
353 		if (eth_hdr == NULL) {
354 			rte_panic("No headroom in mbuf.\n");
355 		}
356 
357 		m->ol_flags |= ol_flags;
358 		m->l2_len = sizeof(struct rte_ether_hdr);
359 
360 		/* 02:00:00:00:00:xx */
361 		d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
362 		*((uint64_t *)d_addr_bytes) = 0x000000000002 +
363 			((uint64_t)port_out << 40);
364 
365 		/* src addr */
366 		rte_ether_addr_copy(&ports_eth_addr[port_out],
367 				&eth_hdr->s_addr);
368 		eth_hdr->ether_type = ether_type;
369 	}
370 
371 	len += len2;
372 
373 	if (likely(len < MAX_PKT_BURST)) {
374 		qconf->tx_mbufs[port_out].len = (uint16_t)len;
375 		return;
376 	}
377 
378 	/* Transmit packets */
379 	send_burst(qconf, (uint16_t)len, port_out);
380 	qconf->tx_mbufs[port_out].len = 0;
381 }
382 
383 /* main processing loop */
384 static int
385 main_loop(__attribute__((unused)) void *dummy)
386 {
387 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
388 	unsigned lcore_id;
389 	uint64_t prev_tsc, diff_tsc, cur_tsc;
390 	int i, j, nb_rx;
391 	uint16_t portid;
392 	struct lcore_queue_conf *qconf;
393 	const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
394 
395 	prev_tsc = 0;
396 
397 	lcore_id = rte_lcore_id();
398 	qconf = &lcore_queue_conf[lcore_id];
399 
400 	if (qconf->n_rx_queue == 0) {
401 		RTE_LOG(INFO, IP_FRAG, "lcore %u has nothing to do\n", lcore_id);
402 		return 0;
403 	}
404 
405 	RTE_LOG(INFO, IP_FRAG, "entering main loop on lcore %u\n", lcore_id);
406 
407 	for (i = 0; i < qconf->n_rx_queue; i++) {
408 
409 		portid = qconf->rx_queue_list[i].portid;
410 		RTE_LOG(INFO, IP_FRAG, " -- lcoreid=%u portid=%d\n", lcore_id,
411 				portid);
412 	}
413 
414 	while (1) {
415 
416 		cur_tsc = rte_rdtsc();
417 
418 		/*
419 		 * TX burst queue drain
420 		 */
421 		diff_tsc = cur_tsc - prev_tsc;
422 		if (unlikely(diff_tsc > drain_tsc)) {
423 
424 			/*
425 			 * This could be optimized (use queueid instead of
426 			 * portid), but it is not called so often
427 			 */
428 			for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
429 				if (qconf->tx_mbufs[portid].len == 0)
430 					continue;
431 				send_burst(&lcore_queue_conf[lcore_id],
432 					   qconf->tx_mbufs[portid].len,
433 					   portid);
434 				qconf->tx_mbufs[portid].len = 0;
435 			}
436 
437 			prev_tsc = cur_tsc;
438 		}
439 
440 		/*
441 		 * Read packet from RX queues
442 		 */
443 		for (i = 0; i < qconf->n_rx_queue; i++) {
444 
445 			portid = qconf->rx_queue_list[i].portid;
446 			nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
447 						 MAX_PKT_BURST);
448 
449 			/* Prefetch first packets */
450 			for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
451 				rte_prefetch0(rte_pktmbuf_mtod(
452 						pkts_burst[j], void *));
453 			}
454 
455 			/* Prefetch and forward already prefetched packets */
456 			for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
457 				rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
458 						j + PREFETCH_OFFSET], void *));
459 				l3fwd_simple_forward(pkts_burst[j], qconf, i, portid);
460 			}
461 
462 			/* Forward remaining prefetched packets */
463 			for (; j < nb_rx; j++) {
464 				l3fwd_simple_forward(pkts_burst[j], qconf, i, portid);
465 			}
466 		}
467 	}
468 }
469 
470 /* display usage */
471 static void
472 print_usage(const char *prgname)
473 {
474 	printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
475 	       "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
476 	       "  -q NQ: number of queue (=ports) per lcore (default is 1)\n",
477 	       prgname);
478 }
479 
480 static int
481 parse_portmask(const char *portmask)
482 {
483 	char *end = NULL;
484 	unsigned long pm;
485 
486 	/* parse hexadecimal string */
487 	pm = strtoul(portmask, &end, 16);
488 	if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
489 		return -1;
490 
491 	if (pm == 0)
492 		return -1;
493 
494 	return pm;
495 }
496 
497 static int
498 parse_nqueue(const char *q_arg)
499 {
500 	char *end = NULL;
501 	unsigned long n;
502 
503 	/* parse hexadecimal string */
504 	n = strtoul(q_arg, &end, 10);
505 	if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
506 		return -1;
507 	if (n == 0)
508 		return -1;
509 	if (n >= MAX_RX_QUEUE_PER_LCORE)
510 		return -1;
511 
512 	return n;
513 }
514 
515 /* Parse the argument given in the command line of the application */
516 static int
517 parse_args(int argc, char **argv)
518 {
519 	int opt, ret;
520 	char **argvopt;
521 	int option_index;
522 	char *prgname = argv[0];
523 	static struct option lgopts[] = {
524 		{NULL, 0, 0, 0}
525 	};
526 
527 	argvopt = argv;
528 
529 	while ((opt = getopt_long(argc, argvopt, "p:q:",
530 				  lgopts, &option_index)) != EOF) {
531 
532 		switch (opt) {
533 		/* portmask */
534 		case 'p':
535 			enabled_port_mask = parse_portmask(optarg);
536 			if (enabled_port_mask < 0) {
537 				printf("invalid portmask\n");
538 				print_usage(prgname);
539 				return -1;
540 			}
541 			break;
542 
543 		/* nqueue */
544 		case 'q':
545 			rx_queue_per_lcore = parse_nqueue(optarg);
546 			if (rx_queue_per_lcore < 0) {
547 				printf("invalid queue number\n");
548 				print_usage(prgname);
549 				return -1;
550 			}
551 			break;
552 
553 		/* long options */
554 		case 0:
555 			print_usage(prgname);
556 			return -1;
557 
558 		default:
559 			print_usage(prgname);
560 			return -1;
561 		}
562 	}
563 
564 	if (enabled_port_mask == 0) {
565 		printf("portmask not specified\n");
566 		print_usage(prgname);
567 		return -1;
568 	}
569 
570 	if (optind >= 0)
571 		argv[optind-1] = prgname;
572 
573 	ret = optind-1;
574 	optind = 1; /* reset getopt lib */
575 	return ret;
576 }
577 
578 static void
579 print_ethaddr(const char *name, struct rte_ether_addr *eth_addr)
580 {
581 	char buf[RTE_ETHER_ADDR_FMT_SIZE];
582 	rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
583 	printf("%s%s", name, buf);
584 }
585 
586 /* Check the link status of all ports in up to 9s, and print them finally */
587 static void
588 check_all_ports_link_status(uint32_t port_mask)
589 {
590 #define CHECK_INTERVAL 100 /* 100ms */
591 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
592 	uint16_t portid;
593 	uint8_t count, all_ports_up, print_flag = 0;
594 	struct rte_eth_link link;
595 	int ret;
596 
597 	printf("\nChecking link status");
598 	fflush(stdout);
599 	for (count = 0; count <= MAX_CHECK_TIME; count++) {
600 		all_ports_up = 1;
601 		RTE_ETH_FOREACH_DEV(portid) {
602 			if ((port_mask & (1 << portid)) == 0)
603 				continue;
604 			memset(&link, 0, sizeof(link));
605 			ret = rte_eth_link_get_nowait(portid, &link);
606 			if (ret < 0) {
607 				all_ports_up = 0;
608 				if (print_flag == 1)
609 					printf("Port %u link get failed: %s\n",
610 						portid, rte_strerror(-ret));
611 				continue;
612 			}
613 			/* print link status if flag set */
614 			if (print_flag == 1) {
615 				if (link.link_status)
616 					printf(
617 					"Port%d Link Up .Speed %u Mbps - %s\n",
618 						portid, link.link_speed,
619 				(link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
620 					("full-duplex") : ("half-duplex\n"));
621 				else
622 					printf("Port %d Link Down\n", portid);
623 				continue;
624 			}
625 			/* clear all_ports_up flag if any link down */
626 			if (link.link_status == ETH_LINK_DOWN) {
627 				all_ports_up = 0;
628 				break;
629 			}
630 		}
631 		/* after finally printing all link status, get out */
632 		if (print_flag == 1)
633 			break;
634 
635 		if (all_ports_up == 0) {
636 			printf(".");
637 			fflush(stdout);
638 			rte_delay_ms(CHECK_INTERVAL);
639 		}
640 
641 		/* set the print_flag if all ports up or timeout */
642 		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
643 			print_flag = 1;
644 			printf("\ndone\n");
645 		}
646 	}
647 }
648 
649 /* Check L3 packet type detection capablity of the NIC port */
650 static int
651 check_ptype(int portid)
652 {
653 	int i, ret;
654 	int ptype_l3_ipv4 = 0, ptype_l3_ipv6 = 0;
655 	uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
656 
657 	ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
658 	if (ret <= 0)
659 		return 0;
660 
661 	uint32_t ptypes[ret];
662 
663 	ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
664 	for (i = 0; i < ret; ++i) {
665 		if (ptypes[i] & RTE_PTYPE_L3_IPV4)
666 			ptype_l3_ipv4 = 1;
667 		if (ptypes[i] & RTE_PTYPE_L3_IPV6)
668 			ptype_l3_ipv6 = 1;
669 	}
670 
671 	if (ptype_l3_ipv4 == 0)
672 		printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
673 
674 	if (ptype_l3_ipv6 == 0)
675 		printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
676 
677 	if (ptype_l3_ipv4 && ptype_l3_ipv6)
678 		return 1;
679 
680 	return 0;
681 
682 }
683 
684 /* Parse packet type of a packet by SW */
685 static inline void
686 parse_ptype(struct rte_mbuf *m)
687 {
688 	struct rte_ether_hdr *eth_hdr;
689 	uint32_t packet_type = RTE_PTYPE_UNKNOWN;
690 	uint16_t ether_type;
691 
692 	eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
693 	ether_type = eth_hdr->ether_type;
694 	if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4))
695 		packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
696 	else if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6))
697 		packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
698 
699 	m->packet_type = packet_type;
700 }
701 
702 /* callback function to detect packet type for a queue of a port */
703 static uint16_t
704 cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
705 		   struct rte_mbuf *pkts[], uint16_t nb_pkts,
706 		   uint16_t max_pkts __rte_unused,
707 		   void *user_param __rte_unused)
708 {
709 	uint16_t i;
710 
711 	for (i = 0; i < nb_pkts; ++i)
712 		parse_ptype(pkts[i]);
713 
714 	return nb_pkts;
715 }
716 
717 static int
718 init_routing_table(void)
719 {
720 	struct rte_lpm *lpm;
721 	struct rte_lpm6 *lpm6;
722 	int socket, ret;
723 	unsigned i;
724 
725 	for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) {
726 		if (socket_lpm[socket]) {
727 			lpm = socket_lpm[socket];
728 			/* populate the LPM table */
729 			for (i = 0; i < RTE_DIM(l3fwd_ipv4_route_array); i++) {
730 				ret = rte_lpm_add(lpm,
731 					l3fwd_ipv4_route_array[i].ip,
732 					l3fwd_ipv4_route_array[i].depth,
733 					l3fwd_ipv4_route_array[i].if_out);
734 
735 				if (ret < 0) {
736 					RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd "
737 						"LPM table\n", i);
738 					return -1;
739 				}
740 
741 				RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv4_BYTES_FMT
742 						"/%d (port %d)\n",
743 					socket,
744 					IPv4_BYTES(l3fwd_ipv4_route_array[i].ip),
745 					l3fwd_ipv4_route_array[i].depth,
746 					l3fwd_ipv4_route_array[i].if_out);
747 			}
748 		}
749 
750 		if (socket_lpm6[socket]) {
751 			lpm6 = socket_lpm6[socket];
752 			/* populate the LPM6 table */
753 			for (i = 0; i < RTE_DIM(l3fwd_ipv6_route_array); i++) {
754 				ret = rte_lpm6_add(lpm6,
755 					l3fwd_ipv6_route_array[i].ip,
756 					l3fwd_ipv6_route_array[i].depth,
757 					l3fwd_ipv6_route_array[i].if_out);
758 
759 				if (ret < 0) {
760 					RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd "
761 						"LPM6 table\n", i);
762 					return -1;
763 				}
764 
765 				RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv6_BYTES_FMT
766 						"/%d (port %d)\n",
767 					socket,
768 					IPv6_BYTES(l3fwd_ipv6_route_array[i].ip),
769 					l3fwd_ipv6_route_array[i].depth,
770 					l3fwd_ipv6_route_array[i].if_out);
771 			}
772 		}
773 	}
774 	return 0;
775 }
776 
777 static int
778 init_mem(void)
779 {
780 	char buf[PATH_MAX];
781 	struct rte_mempool *mp;
782 	struct rte_lpm *lpm;
783 	struct rte_lpm6 *lpm6;
784 	struct rte_lpm_config lpm_config;
785 	int socket;
786 	unsigned lcore_id;
787 
788 	/* traverse through lcores and initialize structures on each socket */
789 
790 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
791 
792 		if (rte_lcore_is_enabled(lcore_id) == 0)
793 			continue;
794 
795 		socket = rte_lcore_to_socket_id(lcore_id);
796 
797 		if (socket == SOCKET_ID_ANY)
798 			socket = 0;
799 
800 		if (socket_direct_pool[socket] == NULL) {
801 			RTE_LOG(INFO, IP_FRAG, "Creating direct mempool on socket %i\n",
802 					socket);
803 			snprintf(buf, sizeof(buf), "pool_direct_%i", socket);
804 
805 			mp = rte_pktmbuf_pool_create(buf, NB_MBUF, 32,
806 				0, RTE_MBUF_DEFAULT_BUF_SIZE, socket);
807 			if (mp == NULL) {
808 				RTE_LOG(ERR, IP_FRAG, "Cannot create direct mempool\n");
809 				return -1;
810 			}
811 			socket_direct_pool[socket] = mp;
812 		}
813 
814 		if (socket_indirect_pool[socket] == NULL) {
815 			RTE_LOG(INFO, IP_FRAG, "Creating indirect mempool on socket %i\n",
816 					socket);
817 			snprintf(buf, sizeof(buf), "pool_indirect_%i", socket);
818 
819 			mp = rte_pktmbuf_pool_create(buf, NB_MBUF, 32, 0, 0,
820 				socket);
821 			if (mp == NULL) {
822 				RTE_LOG(ERR, IP_FRAG, "Cannot create indirect mempool\n");
823 				return -1;
824 			}
825 			socket_indirect_pool[socket] = mp;
826 		}
827 
828 		if (socket_lpm[socket] == NULL) {
829 			RTE_LOG(INFO, IP_FRAG, "Creating LPM table on socket %i\n", socket);
830 			snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
831 
832 			lpm_config.max_rules = LPM_MAX_RULES;
833 			lpm_config.number_tbl8s = 256;
834 			lpm_config.flags = 0;
835 
836 			lpm = rte_lpm_create(buf, socket, &lpm_config);
837 			if (lpm == NULL) {
838 				RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
839 				return -1;
840 			}
841 			socket_lpm[socket] = lpm;
842 		}
843 
844 		if (socket_lpm6[socket] == NULL) {
845 			RTE_LOG(INFO, IP_FRAG, "Creating LPM6 table on socket %i\n", socket);
846 			snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
847 
848 			lpm6 = rte_lpm6_create(buf, socket, &lpm6_config);
849 			if (lpm6 == NULL) {
850 				RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
851 				return -1;
852 			}
853 			socket_lpm6[socket] = lpm6;
854 		}
855 	}
856 
857 	return 0;
858 }
859 
860 int
861 main(int argc, char **argv)
862 {
863 	struct lcore_queue_conf *qconf;
864 	struct rte_eth_dev_info dev_info;
865 	struct rte_eth_txconf *txconf;
866 	struct rx_queue *rxq;
867 	int socket, ret;
868 	uint16_t nb_ports;
869 	uint16_t queueid = 0;
870 	unsigned lcore_id = 0, rx_lcore_id = 0;
871 	uint32_t n_tx_queue, nb_lcores;
872 	uint16_t portid;
873 
874 	/* init EAL */
875 	ret = rte_eal_init(argc, argv);
876 	if (ret < 0)
877 		rte_exit(EXIT_FAILURE, "rte_eal_init failed");
878 	argc -= ret;
879 	argv += ret;
880 
881 	/* parse application arguments (after the EAL ones) */
882 	ret = parse_args(argc, argv);
883 	if (ret < 0)
884 		rte_exit(EXIT_FAILURE, "Invalid arguments");
885 
886 	nb_ports = rte_eth_dev_count_avail();
887 	if (nb_ports == 0)
888 		rte_exit(EXIT_FAILURE, "No ports found!\n");
889 
890 	nb_lcores = rte_lcore_count();
891 
892 	/* initialize structures (mempools, lpm etc.) */
893 	if (init_mem() < 0)
894 		rte_panic("Cannot initialize memory structures!\n");
895 
896 	/* check if portmask has non-existent ports */
897 	if (enabled_port_mask & ~(RTE_LEN2MASK(nb_ports, unsigned)))
898 		rte_exit(EXIT_FAILURE, "Non-existent ports in portmask!\n");
899 
900 	/* initialize all ports */
901 	RTE_ETH_FOREACH_DEV(portid) {
902 		struct rte_eth_conf local_port_conf = port_conf;
903 		struct rte_eth_rxconf rxq_conf;
904 
905 		/* skip ports that are not enabled */
906 		if ((enabled_port_mask & (1 << portid)) == 0) {
907 			printf("Skipping disabled port %d\n", portid);
908 			continue;
909 		}
910 
911 		qconf = &lcore_queue_conf[rx_lcore_id];
912 
913 		/* limit the frame size to the maximum supported by NIC */
914 		ret = rte_eth_dev_info_get(portid, &dev_info);
915 		if (ret != 0)
916 			rte_exit(EXIT_FAILURE,
917 				"Error during getting device (port %u) info: %s\n",
918 				portid, strerror(-ret));
919 
920 		local_port_conf.rxmode.max_rx_pkt_len = RTE_MIN(
921 		    dev_info.max_rx_pktlen,
922 		    local_port_conf.rxmode.max_rx_pkt_len);
923 
924 		/* get the lcore_id for this port */
925 		while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
926 		       qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
927 
928 			rx_lcore_id ++;
929 			if (rx_lcore_id >= RTE_MAX_LCORE)
930 				rte_exit(EXIT_FAILURE, "Not enough cores\n");
931 
932 			qconf = &lcore_queue_conf[rx_lcore_id];
933 		}
934 
935 		socket = (int) rte_lcore_to_socket_id(rx_lcore_id);
936 		if (socket == SOCKET_ID_ANY)
937 			socket = 0;
938 
939 		rxq = &qconf->rx_queue_list[qconf->n_rx_queue];
940 		rxq->portid = portid;
941 		rxq->direct_pool = socket_direct_pool[socket];
942 		rxq->indirect_pool = socket_indirect_pool[socket];
943 		rxq->lpm = socket_lpm[socket];
944 		rxq->lpm6 = socket_lpm6[socket];
945 		qconf->n_rx_queue++;
946 
947 		/* init port */
948 		printf("Initializing port %d on lcore %u...", portid,
949 		       rx_lcore_id);
950 		fflush(stdout);
951 
952 		n_tx_queue = nb_lcores;
953 		if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
954 			n_tx_queue = MAX_TX_QUEUE_PER_PORT;
955 		ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
956 					    &local_port_conf);
957 		if (ret < 0) {
958 			printf("\n");
959 			rte_exit(EXIT_FAILURE, "Cannot configure device: "
960 				"err=%d, port=%d\n",
961 				ret, portid);
962 		}
963 
964 		/* set the mtu to the maximum received packet size */
965 		ret = rte_eth_dev_set_mtu(portid,
966 			local_port_conf.rxmode.max_rx_pkt_len - MTU_OVERHEAD);
967 		if (ret < 0) {
968 			printf("\n");
969 			rte_exit(EXIT_FAILURE, "Set MTU failed: "
970 				"err=%d, port=%d\n",
971 			ret, portid);
972 		}
973 
974 		ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
975 					    &nb_txd);
976 		if (ret < 0) {
977 			printf("\n");
978 			rte_exit(EXIT_FAILURE, "Cannot adjust number of "
979 				"descriptors: err=%d, port=%d\n", ret, portid);
980 		}
981 
982 		/* init one RX queue */
983 		rxq_conf = dev_info.default_rxconf;
984 		rxq_conf.offloads = local_port_conf.rxmode.offloads;
985 		ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
986 					     socket, &rxq_conf,
987 					     socket_direct_pool[socket]);
988 		if (ret < 0) {
989 			printf("\n");
990 			rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: "
991 				"err=%d, port=%d\n",
992 				ret, portid);
993 		}
994 
995 		ret = rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
996 		if (ret < 0) {
997 			printf("\n");
998 			rte_exit(EXIT_FAILURE,
999 				"rte_eth_macaddr_get: err=%d, port=%d\n",
1000 				ret, portid);
1001 		}
1002 
1003 		print_ethaddr(" Address:", &ports_eth_addr[portid]);
1004 		printf("\n");
1005 
1006 		/* init one TX queue per couple (lcore,port) */
1007 		ret = rte_eth_dev_info_get(portid, &dev_info);
1008 		if (ret != 0)
1009 			rte_exit(EXIT_FAILURE,
1010 				"Error during getting device (port %u) info: %s\n",
1011 				portid, strerror(-ret));
1012 
1013 		queueid = 0;
1014 		for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1015 			if (rte_lcore_is_enabled(lcore_id) == 0)
1016 				continue;
1017 
1018 			if (queueid >= dev_info.nb_tx_queues)
1019 				break;
1020 
1021 			socket = (int) rte_lcore_to_socket_id(lcore_id);
1022 			printf("txq=%u,%d ", lcore_id, queueid);
1023 			fflush(stdout);
1024 
1025 			txconf = &dev_info.default_txconf;
1026 			txconf->offloads = local_port_conf.txmode.offloads;
1027 			ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1028 						     socket, txconf);
1029 			if (ret < 0) {
1030 				printf("\n");
1031 				rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
1032 					"err=%d, port=%d\n", ret, portid);
1033 			}
1034 
1035 			qconf = &lcore_queue_conf[lcore_id];
1036 			qconf->tx_queue_id[portid] = queueid;
1037 			queueid++;
1038 		}
1039 
1040 		printf("\n");
1041 	}
1042 
1043 	printf("\n");
1044 
1045 	/* start ports */
1046 	RTE_ETH_FOREACH_DEV(portid) {
1047 		if ((enabled_port_mask & (1 << portid)) == 0) {
1048 			continue;
1049 		}
1050 		/* Start device */
1051 		ret = rte_eth_dev_start(portid);
1052 		if (ret < 0)
1053 			rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1054 				ret, portid);
1055 
1056 		ret = rte_eth_promiscuous_enable(portid);
1057 		if (ret != 0)
1058 			rte_exit(EXIT_FAILURE,
1059 				"rte_eth_promiscuous_enable: err=%s, port=%d\n",
1060 				rte_strerror(-ret), portid);
1061 
1062 		if (check_ptype(portid) == 0) {
1063 			rte_eth_add_rx_callback(portid, 0, cb_parse_ptype, NULL);
1064 			printf("Add Rx callback function to detect L3 packet type by SW :"
1065 				" port = %d\n", portid);
1066 		}
1067 	}
1068 
1069 	if (init_routing_table() < 0)
1070 		rte_exit(EXIT_FAILURE, "Cannot init routing table\n");
1071 
1072 	check_all_ports_link_status(enabled_port_mask);
1073 
1074 	/* launch per-lcore init on every lcore */
1075 	rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1076 	RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1077 		if (rte_eal_wait_lcore(lcore_id) < 0)
1078 			return -1;
1079 	}
1080 
1081 	return 0;
1082 }
1083