xref: /dpdk/examples/ip_fragmentation/main.c (revision 2808423a9ce42a748aed77a4b487be27d2b6acfa)
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 			     DEV_RX_OFFLOAD_CRC_STRIP),
146 	},
147 	.txmode = {
148 		.mq_mode = ETH_MQ_TX_NONE,
149 		.offloads = (DEV_TX_OFFLOAD_IPV4_CKSUM |
150 			     DEV_TX_OFFLOAD_MULTI_SEGS),
151 	},
152 };
153 
154 /*
155  * IPv4 forwarding table
156  */
157 struct l3fwd_ipv4_route {
158 	uint32_t ip;
159 	uint8_t  depth;
160 	uint8_t  if_out;
161 };
162 
163 struct l3fwd_ipv4_route l3fwd_ipv4_route_array[] = {
164 		{IPv4(100,10,0,0), 16, 0},
165 		{IPv4(100,20,0,0), 16, 1},
166 		{IPv4(100,30,0,0), 16, 2},
167 		{IPv4(100,40,0,0), 16, 3},
168 		{IPv4(100,50,0,0), 16, 4},
169 		{IPv4(100,60,0,0), 16, 5},
170 		{IPv4(100,70,0,0), 16, 6},
171 		{IPv4(100,80,0,0), 16, 7},
172 };
173 
174 /*
175  * IPv6 forwarding table
176  */
177 
178 struct l3fwd_ipv6_route {
179 	uint8_t ip[IPV6_ADDR_LEN];
180 	uint8_t depth;
181 	uint8_t if_out;
182 };
183 
184 static struct l3fwd_ipv6_route l3fwd_ipv6_route_array[] = {
185 	{{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
186 	{{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
187 	{{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
188 	{{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
189 	{{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
190 	{{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
191 	{{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
192 	{{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
193 };
194 
195 #define LPM_MAX_RULES         1024
196 #define LPM6_MAX_RULES         1024
197 #define LPM6_NUMBER_TBL8S (1 << 16)
198 
199 struct rte_lpm6_config lpm6_config = {
200 		.max_rules = LPM6_MAX_RULES,
201 		.number_tbl8s = LPM6_NUMBER_TBL8S,
202 		.flags = 0
203 };
204 
205 static struct rte_mempool *socket_direct_pool[RTE_MAX_NUMA_NODES];
206 static struct rte_mempool *socket_indirect_pool[RTE_MAX_NUMA_NODES];
207 static struct rte_lpm *socket_lpm[RTE_MAX_NUMA_NODES];
208 static struct rte_lpm6 *socket_lpm6[RTE_MAX_NUMA_NODES];
209 
210 /* Send burst of packets on an output interface */
211 static inline int
212 send_burst(struct lcore_queue_conf *qconf, uint16_t n, uint16_t port)
213 {
214 	struct rte_mbuf **m_table;
215 	int ret;
216 	uint16_t queueid;
217 
218 	queueid = qconf->tx_queue_id[port];
219 	m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
220 
221 	ret = rte_eth_tx_burst(port, queueid, m_table, n);
222 	if (unlikely(ret < n)) {
223 		do {
224 			rte_pktmbuf_free(m_table[ret]);
225 		} while (++ret < n);
226 	}
227 
228 	return 0;
229 }
230 
231 static inline void
232 l3fwd_simple_forward(struct rte_mbuf *m, struct lcore_queue_conf *qconf,
233 		uint8_t queueid, uint16_t port_in)
234 {
235 	struct rx_queue *rxq;
236 	uint32_t i, len, next_hop;
237 	uint8_t ipv6;
238 	uint16_t port_out;
239 	int32_t len2;
240 
241 	ipv6 = 0;
242 	rxq = &qconf->rx_queue_list[queueid];
243 
244 	/* by default, send everything back to the source port */
245 	port_out = port_in;
246 
247 	/* Remove the Ethernet header and trailer from the input packet */
248 	rte_pktmbuf_adj(m, (uint16_t)sizeof(struct ether_hdr));
249 
250 	/* Build transmission burst */
251 	len = qconf->tx_mbufs[port_out].len;
252 
253 	/* if this is an IPv4 packet */
254 	if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
255 		struct ipv4_hdr *ip_hdr;
256 		uint32_t ip_dst;
257 		/* Read the lookup key (i.e. ip_dst) from the input packet */
258 		ip_hdr = rte_pktmbuf_mtod(m, struct ipv4_hdr *);
259 		ip_dst = rte_be_to_cpu_32(ip_hdr->dst_addr);
260 
261 		/* Find destination port */
262 		if (rte_lpm_lookup(rxq->lpm, ip_dst, &next_hop) == 0 &&
263 				(enabled_port_mask & 1 << next_hop) != 0) {
264 			port_out = next_hop;
265 
266 			/* Build transmission burst for new port */
267 			len = qconf->tx_mbufs[port_out].len;
268 		}
269 
270 		/* if we don't need to do any fragmentation */
271 		if (likely (IPV4_MTU_DEFAULT >= m->pkt_len)) {
272 			qconf->tx_mbufs[port_out].m_table[len] = m;
273 			len2 = 1;
274 		} else {
275 			len2 = rte_ipv4_fragment_packet(m,
276 				&qconf->tx_mbufs[port_out].m_table[len],
277 				(uint16_t)(MBUF_TABLE_SIZE - len),
278 				IPV4_MTU_DEFAULT,
279 				rxq->direct_pool, rxq->indirect_pool);
280 
281 			/* Free input packet */
282 			rte_pktmbuf_free(m);
283 
284 			/* If we fail to fragment the packet */
285 			if (unlikely (len2 < 0))
286 				return;
287 		}
288 	} else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
289 		/* if this is an IPv6 packet */
290 		struct ipv6_hdr *ip_hdr;
291 
292 		ipv6 = 1;
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 		if (ipv6)
351 			eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv6);
352 		else
353 			eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv4);
354 	}
355 
356 	len += len2;
357 
358 	if (likely(len < MAX_PKT_BURST)) {
359 		qconf->tx_mbufs[port_out].len = (uint16_t)len;
360 		return;
361 	}
362 
363 	/* Transmit packets */
364 	send_burst(qconf, (uint16_t)len, port_out);
365 	qconf->tx_mbufs[port_out].len = 0;
366 }
367 
368 /* main processing loop */
369 static int
370 main_loop(__attribute__((unused)) void *dummy)
371 {
372 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
373 	unsigned lcore_id;
374 	uint64_t prev_tsc, diff_tsc, cur_tsc;
375 	int i, j, nb_rx;
376 	uint16_t portid;
377 	struct lcore_queue_conf *qconf;
378 	const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
379 
380 	prev_tsc = 0;
381 
382 	lcore_id = rte_lcore_id();
383 	qconf = &lcore_queue_conf[lcore_id];
384 
385 	if (qconf->n_rx_queue == 0) {
386 		RTE_LOG(INFO, IP_FRAG, "lcore %u has nothing to do\n", lcore_id);
387 		return 0;
388 	}
389 
390 	RTE_LOG(INFO, IP_FRAG, "entering main loop on lcore %u\n", lcore_id);
391 
392 	for (i = 0; i < qconf->n_rx_queue; i++) {
393 
394 		portid = qconf->rx_queue_list[i].portid;
395 		RTE_LOG(INFO, IP_FRAG, " -- lcoreid=%u portid=%d\n", lcore_id,
396 				portid);
397 	}
398 
399 	while (1) {
400 
401 		cur_tsc = rte_rdtsc();
402 
403 		/*
404 		 * TX burst queue drain
405 		 */
406 		diff_tsc = cur_tsc - prev_tsc;
407 		if (unlikely(diff_tsc > drain_tsc)) {
408 
409 			/*
410 			 * This could be optimized (use queueid instead of
411 			 * portid), but it is not called so often
412 			 */
413 			for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
414 				if (qconf->tx_mbufs[portid].len == 0)
415 					continue;
416 				send_burst(&lcore_queue_conf[lcore_id],
417 					   qconf->tx_mbufs[portid].len,
418 					   portid);
419 				qconf->tx_mbufs[portid].len = 0;
420 			}
421 
422 			prev_tsc = cur_tsc;
423 		}
424 
425 		/*
426 		 * Read packet from RX queues
427 		 */
428 		for (i = 0; i < qconf->n_rx_queue; i++) {
429 
430 			portid = qconf->rx_queue_list[i].portid;
431 			nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
432 						 MAX_PKT_BURST);
433 
434 			/* Prefetch first packets */
435 			for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
436 				rte_prefetch0(rte_pktmbuf_mtod(
437 						pkts_burst[j], void *));
438 			}
439 
440 			/* Prefetch and forward already prefetched packets */
441 			for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
442 				rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
443 						j + PREFETCH_OFFSET], void *));
444 				l3fwd_simple_forward(pkts_burst[j], qconf, i, portid);
445 			}
446 
447 			/* Forward remaining prefetched packets */
448 			for (; j < nb_rx; j++) {
449 				l3fwd_simple_forward(pkts_burst[j], qconf, i, portid);
450 			}
451 		}
452 	}
453 }
454 
455 /* display usage */
456 static void
457 print_usage(const char *prgname)
458 {
459 	printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
460 	       "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
461 	       "  -q NQ: number of queue (=ports) per lcore (default is 1)\n",
462 	       prgname);
463 }
464 
465 static int
466 parse_portmask(const char *portmask)
467 {
468 	char *end = NULL;
469 	unsigned long pm;
470 
471 	/* parse hexadecimal string */
472 	pm = strtoul(portmask, &end, 16);
473 	if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
474 		return -1;
475 
476 	if (pm == 0)
477 		return -1;
478 
479 	return pm;
480 }
481 
482 static int
483 parse_nqueue(const char *q_arg)
484 {
485 	char *end = NULL;
486 	unsigned long n;
487 
488 	/* parse hexadecimal string */
489 	n = strtoul(q_arg, &end, 10);
490 	if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
491 		return -1;
492 	if (n == 0)
493 		return -1;
494 	if (n >= MAX_RX_QUEUE_PER_LCORE)
495 		return -1;
496 
497 	return n;
498 }
499 
500 /* Parse the argument given in the command line of the application */
501 static int
502 parse_args(int argc, char **argv)
503 {
504 	int opt, ret;
505 	char **argvopt;
506 	int option_index;
507 	char *prgname = argv[0];
508 	static struct option lgopts[] = {
509 		{NULL, 0, 0, 0}
510 	};
511 
512 	argvopt = argv;
513 
514 	while ((opt = getopt_long(argc, argvopt, "p:q:",
515 				  lgopts, &option_index)) != EOF) {
516 
517 		switch (opt) {
518 		/* portmask */
519 		case 'p':
520 			enabled_port_mask = parse_portmask(optarg);
521 			if (enabled_port_mask < 0) {
522 				printf("invalid portmask\n");
523 				print_usage(prgname);
524 				return -1;
525 			}
526 			break;
527 
528 		/* nqueue */
529 		case 'q':
530 			rx_queue_per_lcore = parse_nqueue(optarg);
531 			if (rx_queue_per_lcore < 0) {
532 				printf("invalid queue number\n");
533 				print_usage(prgname);
534 				return -1;
535 			}
536 			break;
537 
538 		/* long options */
539 		case 0:
540 			print_usage(prgname);
541 			return -1;
542 
543 		default:
544 			print_usage(prgname);
545 			return -1;
546 		}
547 	}
548 
549 	if (enabled_port_mask == 0) {
550 		printf("portmask not specified\n");
551 		print_usage(prgname);
552 		return -1;
553 	}
554 
555 	if (optind >= 0)
556 		argv[optind-1] = prgname;
557 
558 	ret = optind-1;
559 	optind = 1; /* reset getopt lib */
560 	return ret;
561 }
562 
563 static void
564 print_ethaddr(const char *name, struct ether_addr *eth_addr)
565 {
566 	char buf[ETHER_ADDR_FMT_SIZE];
567 	ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
568 	printf("%s%s", name, buf);
569 }
570 
571 /* Check the link status of all ports in up to 9s, and print them finally */
572 static void
573 check_all_ports_link_status(uint32_t port_mask)
574 {
575 #define CHECK_INTERVAL 100 /* 100ms */
576 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
577 	uint16_t portid;
578 	uint8_t count, all_ports_up, print_flag = 0;
579 	struct rte_eth_link link;
580 
581 	printf("\nChecking link status");
582 	fflush(stdout);
583 	for (count = 0; count <= MAX_CHECK_TIME; count++) {
584 		all_ports_up = 1;
585 		RTE_ETH_FOREACH_DEV(portid) {
586 			if ((port_mask & (1 << portid)) == 0)
587 				continue;
588 			memset(&link, 0, sizeof(link));
589 			rte_eth_link_get_nowait(portid, &link);
590 			/* print link status if flag set */
591 			if (print_flag == 1) {
592 				if (link.link_status)
593 					printf(
594 					"Port%d Link Up .Speed %u Mbps - %s\n",
595 						portid, link.link_speed,
596 				(link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
597 					("full-duplex") : ("half-duplex\n"));
598 				else
599 					printf("Port %d Link Down\n", portid);
600 				continue;
601 			}
602 			/* clear all_ports_up flag if any link down */
603 			if (link.link_status == ETH_LINK_DOWN) {
604 				all_ports_up = 0;
605 				break;
606 			}
607 		}
608 		/* after finally printing all link status, get out */
609 		if (print_flag == 1)
610 			break;
611 
612 		if (all_ports_up == 0) {
613 			printf(".");
614 			fflush(stdout);
615 			rte_delay_ms(CHECK_INTERVAL);
616 		}
617 
618 		/* set the print_flag if all ports up or timeout */
619 		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
620 			print_flag = 1;
621 			printf("\ndone\n");
622 		}
623 	}
624 }
625 
626 /* Check L3 packet type detection capablity of the NIC port */
627 static int
628 check_ptype(int portid)
629 {
630 	int i, ret;
631 	int ptype_l3_ipv4 = 0, ptype_l3_ipv6 = 0;
632 	uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
633 
634 	ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
635 	if (ret <= 0)
636 		return 0;
637 
638 	uint32_t ptypes[ret];
639 
640 	ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
641 	for (i = 0; i < ret; ++i) {
642 		if (ptypes[i] & RTE_PTYPE_L3_IPV4)
643 			ptype_l3_ipv4 = 1;
644 		if (ptypes[i] & RTE_PTYPE_L3_IPV6)
645 			ptype_l3_ipv6 = 1;
646 	}
647 
648 	if (ptype_l3_ipv4 == 0)
649 		printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
650 
651 	if (ptype_l3_ipv6 == 0)
652 		printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
653 
654 	if (ptype_l3_ipv4 && ptype_l3_ipv6)
655 		return 1;
656 
657 	return 0;
658 
659 }
660 
661 /* Parse packet type of a packet by SW */
662 static inline void
663 parse_ptype(struct rte_mbuf *m)
664 {
665 	struct ether_hdr *eth_hdr;
666 	uint32_t packet_type = RTE_PTYPE_UNKNOWN;
667 	uint16_t ether_type;
668 
669 	eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
670 	ether_type = eth_hdr->ether_type;
671 	if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4))
672 		packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
673 	else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6))
674 		packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
675 
676 	m->packet_type = packet_type;
677 }
678 
679 /* callback function to detect packet type for a queue of a port */
680 static uint16_t
681 cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
682 		   struct rte_mbuf *pkts[], uint16_t nb_pkts,
683 		   uint16_t max_pkts __rte_unused,
684 		   void *user_param __rte_unused)
685 {
686 	uint16_t i;
687 
688 	for (i = 0; i < nb_pkts; ++i)
689 		parse_ptype(pkts[i]);
690 
691 	return nb_pkts;
692 }
693 
694 static int
695 init_routing_table(void)
696 {
697 	struct rte_lpm *lpm;
698 	struct rte_lpm6 *lpm6;
699 	int socket, ret;
700 	unsigned i;
701 
702 	for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) {
703 		if (socket_lpm[socket]) {
704 			lpm = socket_lpm[socket];
705 			/* populate the LPM table */
706 			for (i = 0; i < RTE_DIM(l3fwd_ipv4_route_array); i++) {
707 				ret = rte_lpm_add(lpm,
708 					l3fwd_ipv4_route_array[i].ip,
709 					l3fwd_ipv4_route_array[i].depth,
710 					l3fwd_ipv4_route_array[i].if_out);
711 
712 				if (ret < 0) {
713 					RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd "
714 						"LPM table\n", i);
715 					return -1;
716 				}
717 
718 				RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv4_BYTES_FMT
719 						"/%d (port %d)\n",
720 					socket,
721 					IPv4_BYTES(l3fwd_ipv4_route_array[i].ip),
722 					l3fwd_ipv4_route_array[i].depth,
723 					l3fwd_ipv4_route_array[i].if_out);
724 			}
725 		}
726 
727 		if (socket_lpm6[socket]) {
728 			lpm6 = socket_lpm6[socket];
729 			/* populate the LPM6 table */
730 			for (i = 0; i < RTE_DIM(l3fwd_ipv6_route_array); i++) {
731 				ret = rte_lpm6_add(lpm6,
732 					l3fwd_ipv6_route_array[i].ip,
733 					l3fwd_ipv6_route_array[i].depth,
734 					l3fwd_ipv6_route_array[i].if_out);
735 
736 				if (ret < 0) {
737 					RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd "
738 						"LPM6 table\n", i);
739 					return -1;
740 				}
741 
742 				RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv6_BYTES_FMT
743 						"/%d (port %d)\n",
744 					socket,
745 					IPv6_BYTES(l3fwd_ipv6_route_array[i].ip),
746 					l3fwd_ipv6_route_array[i].depth,
747 					l3fwd_ipv6_route_array[i].if_out);
748 			}
749 		}
750 	}
751 	return 0;
752 }
753 
754 static int
755 init_mem(void)
756 {
757 	char buf[PATH_MAX];
758 	struct rte_mempool *mp;
759 	struct rte_lpm *lpm;
760 	struct rte_lpm6 *lpm6;
761 	struct rte_lpm_config lpm_config;
762 	int socket;
763 	unsigned lcore_id;
764 
765 	/* traverse through lcores and initialize structures on each socket */
766 
767 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
768 
769 		if (rte_lcore_is_enabled(lcore_id) == 0)
770 			continue;
771 
772 		socket = rte_lcore_to_socket_id(lcore_id);
773 
774 		if (socket == SOCKET_ID_ANY)
775 			socket = 0;
776 
777 		if (socket_direct_pool[socket] == NULL) {
778 			RTE_LOG(INFO, IP_FRAG, "Creating direct mempool on socket %i\n",
779 					socket);
780 			snprintf(buf, sizeof(buf), "pool_direct_%i", socket);
781 
782 			mp = rte_pktmbuf_pool_create(buf, NB_MBUF, 32,
783 				0, RTE_MBUF_DEFAULT_BUF_SIZE, socket);
784 			if (mp == NULL) {
785 				RTE_LOG(ERR, IP_FRAG, "Cannot create direct mempool\n");
786 				return -1;
787 			}
788 			socket_direct_pool[socket] = mp;
789 		}
790 
791 		if (socket_indirect_pool[socket] == NULL) {
792 			RTE_LOG(INFO, IP_FRAG, "Creating indirect mempool on socket %i\n",
793 					socket);
794 			snprintf(buf, sizeof(buf), "pool_indirect_%i", socket);
795 
796 			mp = rte_pktmbuf_pool_create(buf, NB_MBUF, 32, 0, 0,
797 				socket);
798 			if (mp == NULL) {
799 				RTE_LOG(ERR, IP_FRAG, "Cannot create indirect mempool\n");
800 				return -1;
801 			}
802 			socket_indirect_pool[socket] = mp;
803 		}
804 
805 		if (socket_lpm[socket] == NULL) {
806 			RTE_LOG(INFO, IP_FRAG, "Creating LPM table on socket %i\n", socket);
807 			snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
808 
809 			lpm_config.max_rules = LPM_MAX_RULES;
810 			lpm_config.number_tbl8s = 256;
811 			lpm_config.flags = 0;
812 
813 			lpm = rte_lpm_create(buf, socket, &lpm_config);
814 			if (lpm == NULL) {
815 				RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
816 				return -1;
817 			}
818 			socket_lpm[socket] = lpm;
819 		}
820 
821 		if (socket_lpm6[socket] == NULL) {
822 			RTE_LOG(INFO, IP_FRAG, "Creating LPM6 table on socket %i\n", socket);
823 			snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
824 
825 			lpm6 = rte_lpm6_create(buf, socket, &lpm6_config);
826 			if (lpm6 == NULL) {
827 				RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
828 				return -1;
829 			}
830 			socket_lpm6[socket] = lpm6;
831 		}
832 	}
833 
834 	return 0;
835 }
836 
837 int
838 main(int argc, char **argv)
839 {
840 	struct lcore_queue_conf *qconf;
841 	struct rte_eth_dev_info dev_info;
842 	struct rte_eth_txconf *txconf;
843 	struct rx_queue *rxq;
844 	int socket, ret;
845 	uint16_t nb_ports;
846 	uint16_t queueid = 0;
847 	unsigned lcore_id = 0, rx_lcore_id = 0;
848 	uint32_t n_tx_queue, nb_lcores;
849 	uint16_t portid;
850 
851 	/* init EAL */
852 	ret = rte_eal_init(argc, argv);
853 	if (ret < 0)
854 		rte_exit(EXIT_FAILURE, "rte_eal_init failed");
855 	argc -= ret;
856 	argv += ret;
857 
858 	/* parse application arguments (after the EAL ones) */
859 	ret = parse_args(argc, argv);
860 	if (ret < 0)
861 		rte_exit(EXIT_FAILURE, "Invalid arguments");
862 
863 	nb_ports = rte_eth_dev_count_avail();
864 	if (nb_ports == 0)
865 		rte_exit(EXIT_FAILURE, "No ports found!\n");
866 
867 	nb_lcores = rte_lcore_count();
868 
869 	/* initialize structures (mempools, lpm etc.) */
870 	if (init_mem() < 0)
871 		rte_panic("Cannot initialize memory structures!\n");
872 
873 	/* check if portmask has non-existent ports */
874 	if (enabled_port_mask & ~(RTE_LEN2MASK(nb_ports, unsigned)))
875 		rte_exit(EXIT_FAILURE, "Non-existent ports in portmask!\n");
876 
877 	/* initialize all ports */
878 	RTE_ETH_FOREACH_DEV(portid) {
879 		struct rte_eth_conf local_port_conf = port_conf;
880 		struct rte_eth_rxconf rxq_conf;
881 
882 		/* skip ports that are not enabled */
883 		if ((enabled_port_mask & (1 << portid)) == 0) {
884 			printf("Skipping disabled port %d\n", portid);
885 			continue;
886 		}
887 
888 		qconf = &lcore_queue_conf[rx_lcore_id];
889 
890 		/* limit the frame size to the maximum supported by NIC */
891 		rte_eth_dev_info_get(portid, &dev_info);
892 		local_port_conf.rxmode.max_rx_pkt_len = RTE_MIN(
893 		    dev_info.max_rx_pktlen,
894 		    local_port_conf.rxmode.max_rx_pkt_len);
895 
896 		/* get the lcore_id for this port */
897 		while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
898 		       qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
899 
900 			rx_lcore_id ++;
901 			if (rx_lcore_id >= RTE_MAX_LCORE)
902 				rte_exit(EXIT_FAILURE, "Not enough cores\n");
903 
904 			qconf = &lcore_queue_conf[rx_lcore_id];
905 		}
906 
907 		socket = (int) rte_lcore_to_socket_id(rx_lcore_id);
908 		if (socket == SOCKET_ID_ANY)
909 			socket = 0;
910 
911 		rxq = &qconf->rx_queue_list[qconf->n_rx_queue];
912 		rxq->portid = portid;
913 		rxq->direct_pool = socket_direct_pool[socket];
914 		rxq->indirect_pool = socket_indirect_pool[socket];
915 		rxq->lpm = socket_lpm[socket];
916 		rxq->lpm6 = socket_lpm6[socket];
917 		qconf->n_rx_queue++;
918 
919 		/* init port */
920 		printf("Initializing port %d on lcore %u...", portid,
921 		       rx_lcore_id);
922 		fflush(stdout);
923 
924 		n_tx_queue = nb_lcores;
925 		if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
926 			n_tx_queue = MAX_TX_QUEUE_PER_PORT;
927 		if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
928 			local_port_conf.txmode.offloads |=
929 				DEV_TX_OFFLOAD_MBUF_FAST_FREE;
930 		ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
931 					    &local_port_conf);
932 		if (ret < 0) {
933 			printf("\n");
934 			rte_exit(EXIT_FAILURE, "Cannot configure device: "
935 				"err=%d, port=%d\n",
936 				ret, portid);
937 		}
938 
939 		ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
940 					    &nb_txd);
941 		if (ret < 0) {
942 			printf("\n");
943 			rte_exit(EXIT_FAILURE, "Cannot adjust number of "
944 				"descriptors: err=%d, port=%d\n", ret, portid);
945 		}
946 
947 		/* init one RX queue */
948 		rxq_conf = dev_info.default_rxconf;
949 		rxq_conf.offloads = local_port_conf.rxmode.offloads;
950 		ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
951 					     socket, &rxq_conf,
952 					     socket_direct_pool[socket]);
953 		if (ret < 0) {
954 			printf("\n");
955 			rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: "
956 				"err=%d, port=%d\n",
957 				ret, portid);
958 		}
959 
960 		rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
961 		print_ethaddr(" Address:", &ports_eth_addr[portid]);
962 		printf("\n");
963 
964 		/* init one TX queue per couple (lcore,port) */
965 		queueid = 0;
966 		for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
967 			if (rte_lcore_is_enabled(lcore_id) == 0)
968 				continue;
969 
970 			socket = (int) rte_lcore_to_socket_id(lcore_id);
971 			printf("txq=%u,%d ", lcore_id, queueid);
972 			fflush(stdout);
973 
974 			txconf = &dev_info.default_txconf;
975 			txconf->offloads = local_port_conf.txmode.offloads;
976 			ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
977 						     socket, txconf);
978 			if (ret < 0) {
979 				printf("\n");
980 				rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
981 					"err=%d, port=%d\n", ret, portid);
982 			}
983 
984 			qconf = &lcore_queue_conf[lcore_id];
985 			qconf->tx_queue_id[portid] = queueid;
986 			queueid++;
987 		}
988 
989 		printf("\n");
990 	}
991 
992 	printf("\n");
993 
994 	/* start ports */
995 	RTE_ETH_FOREACH_DEV(portid) {
996 		if ((enabled_port_mask & (1 << portid)) == 0) {
997 			continue;
998 		}
999 		/* Start device */
1000 		ret = rte_eth_dev_start(portid);
1001 		if (ret < 0)
1002 			rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1003 				ret, portid);
1004 
1005 		rte_eth_promiscuous_enable(portid);
1006 
1007 		if (check_ptype(portid) == 0) {
1008 			rte_eth_add_rx_callback(portid, 0, cb_parse_ptype, NULL);
1009 			printf("Add Rx callback function to detect L3 packet type by SW :"
1010 				" port = %d\n", portid);
1011 		}
1012 	}
1013 
1014 	if (init_routing_table() < 0)
1015 		rte_exit(EXIT_FAILURE, "Cannot init routing table\n");
1016 
1017 	check_all_ports_link_status(enabled_port_mask);
1018 
1019 	/* launch per-lcore init on every lcore */
1020 	rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1021 	RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1022 		if (rte_eal_wait_lcore(lcore_id) < 0)
1023 			return -1;
1024 	}
1025 
1026 	return 0;
1027 }
1028