xref: /dpdk/examples/ip_fragmentation/main.c (revision 117eaa70584b73eebf6f648cf3ee6f2ab03264a0)
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 		.ignore_offload_bitfield = 1,
144 		.offloads = (DEV_RX_OFFLOAD_CHECKSUM |
145 			     DEV_RX_OFFLOAD_JUMBO_FRAME |
146 			     DEV_RX_OFFLOAD_CRC_STRIP),
147 	},
148 	.txmode = {
149 		.mq_mode = ETH_MQ_TX_NONE,
150 		.offloads = (DEV_TX_OFFLOAD_IPV4_CKSUM |
151 			     DEV_TX_OFFLOAD_MULTI_SEGS),
152 	},
153 };
154 
155 /*
156  * IPv4 forwarding table
157  */
158 struct l3fwd_ipv4_route {
159 	uint32_t ip;
160 	uint8_t  depth;
161 	uint8_t  if_out;
162 };
163 
164 struct l3fwd_ipv4_route l3fwd_ipv4_route_array[] = {
165 		{IPv4(100,10,0,0), 16, 0},
166 		{IPv4(100,20,0,0), 16, 1},
167 		{IPv4(100,30,0,0), 16, 2},
168 		{IPv4(100,40,0,0), 16, 3},
169 		{IPv4(100,50,0,0), 16, 4},
170 		{IPv4(100,60,0,0), 16, 5},
171 		{IPv4(100,70,0,0), 16, 6},
172 		{IPv4(100,80,0,0), 16, 7},
173 };
174 
175 /*
176  * IPv6 forwarding table
177  */
178 
179 struct l3fwd_ipv6_route {
180 	uint8_t ip[IPV6_ADDR_LEN];
181 	uint8_t depth;
182 	uint8_t if_out;
183 };
184 
185 static struct l3fwd_ipv6_route l3fwd_ipv6_route_array[] = {
186 	{{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
187 	{{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
188 	{{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
189 	{{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
190 	{{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
191 	{{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
192 	{{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
193 	{{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
194 };
195 
196 #define LPM_MAX_RULES         1024
197 #define LPM6_MAX_RULES         1024
198 #define LPM6_NUMBER_TBL8S (1 << 16)
199 
200 struct rte_lpm6_config lpm6_config = {
201 		.max_rules = LPM6_MAX_RULES,
202 		.number_tbl8s = LPM6_NUMBER_TBL8S,
203 		.flags = 0
204 };
205 
206 static struct rte_mempool *socket_direct_pool[RTE_MAX_NUMA_NODES];
207 static struct rte_mempool *socket_indirect_pool[RTE_MAX_NUMA_NODES];
208 static struct rte_lpm *socket_lpm[RTE_MAX_NUMA_NODES];
209 static struct rte_lpm6 *socket_lpm6[RTE_MAX_NUMA_NODES];
210 
211 /* Send burst of packets on an output interface */
212 static inline int
213 send_burst(struct lcore_queue_conf *qconf, uint16_t n, uint16_t port)
214 {
215 	struct rte_mbuf **m_table;
216 	int ret;
217 	uint16_t queueid;
218 
219 	queueid = qconf->tx_queue_id[port];
220 	m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
221 
222 	ret = rte_eth_tx_burst(port, queueid, m_table, n);
223 	if (unlikely(ret < n)) {
224 		do {
225 			rte_pktmbuf_free(m_table[ret]);
226 		} while (++ret < n);
227 	}
228 
229 	return 0;
230 }
231 
232 static inline void
233 l3fwd_simple_forward(struct rte_mbuf *m, struct lcore_queue_conf *qconf,
234 		uint8_t queueid, uint16_t port_in)
235 {
236 	struct rx_queue *rxq;
237 	uint32_t i, len, next_hop;
238 	uint8_t ipv6;
239 	uint16_t port_out;
240 	int32_t len2;
241 
242 	ipv6 = 0;
243 	rxq = &qconf->rx_queue_list[queueid];
244 
245 	/* by default, send everything back to the source port */
246 	port_out = port_in;
247 
248 	/* Remove the Ethernet header and trailer from the input packet */
249 	rte_pktmbuf_adj(m, (uint16_t)sizeof(struct ether_hdr));
250 
251 	/* Build transmission burst */
252 	len = qconf->tx_mbufs[port_out].len;
253 
254 	/* if this is an IPv4 packet */
255 	if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
256 		struct ipv4_hdr *ip_hdr;
257 		uint32_t ip_dst;
258 		/* Read the lookup key (i.e. ip_dst) from the input packet */
259 		ip_hdr = rte_pktmbuf_mtod(m, struct ipv4_hdr *);
260 		ip_dst = rte_be_to_cpu_32(ip_hdr->dst_addr);
261 
262 		/* Find destination port */
263 		if (rte_lpm_lookup(rxq->lpm, ip_dst, &next_hop) == 0 &&
264 				(enabled_port_mask & 1 << next_hop) != 0) {
265 			port_out = next_hop;
266 
267 			/* Build transmission burst for new port */
268 			len = qconf->tx_mbufs[port_out].len;
269 		}
270 
271 		/* if we don't need to do any fragmentation */
272 		if (likely (IPV4_MTU_DEFAULT >= m->pkt_len)) {
273 			qconf->tx_mbufs[port_out].m_table[len] = m;
274 			len2 = 1;
275 		} else {
276 			len2 = rte_ipv4_fragment_packet(m,
277 				&qconf->tx_mbufs[port_out].m_table[len],
278 				(uint16_t)(MBUF_TABLE_SIZE - len),
279 				IPV4_MTU_DEFAULT,
280 				rxq->direct_pool, rxq->indirect_pool);
281 
282 			/* Free input packet */
283 			rte_pktmbuf_free(m);
284 
285 			/* If we fail to fragment the packet */
286 			if (unlikely (len2 < 0))
287 				return;
288 		}
289 	} else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
290 		/* if this is an IPv6 packet */
291 		struct ipv6_hdr *ip_hdr;
292 
293 		ipv6 = 1;
294 
295 		/* Read the lookup key (i.e. ip_dst) from the input packet */
296 		ip_hdr = rte_pktmbuf_mtod(m, struct ipv6_hdr *);
297 
298 		/* Find destination port */
299 		if (rte_lpm6_lookup(rxq->lpm6, ip_hdr->dst_addr,
300 						&next_hop) == 0 &&
301 				(enabled_port_mask & 1 << next_hop) != 0) {
302 			port_out = next_hop;
303 
304 			/* Build transmission burst for new port */
305 			len = qconf->tx_mbufs[port_out].len;
306 		}
307 
308 		/* if we don't need to do any fragmentation */
309 		if (likely (IPV6_MTU_DEFAULT >= m->pkt_len)) {
310 			qconf->tx_mbufs[port_out].m_table[len] = m;
311 			len2 = 1;
312 		} else {
313 			len2 = rte_ipv6_fragment_packet(m,
314 				&qconf->tx_mbufs[port_out].m_table[len],
315 				(uint16_t)(MBUF_TABLE_SIZE - len),
316 				IPV6_MTU_DEFAULT,
317 				rxq->direct_pool, rxq->indirect_pool);
318 
319 			/* Free input packet */
320 			rte_pktmbuf_free(m);
321 
322 			/* If we fail to fragment the packet */
323 			if (unlikely (len2 < 0))
324 				return;
325 		}
326 	}
327 	/* else, just forward the packet */
328 	else {
329 		qconf->tx_mbufs[port_out].m_table[len] = m;
330 		len2 = 1;
331 	}
332 
333 	for (i = len; i < len + len2; i ++) {
334 		void *d_addr_bytes;
335 
336 		m = qconf->tx_mbufs[port_out].m_table[i];
337 		struct ether_hdr *eth_hdr = (struct ether_hdr *)
338 			rte_pktmbuf_prepend(m, (uint16_t)sizeof(struct ether_hdr));
339 		if (eth_hdr == NULL) {
340 			rte_panic("No headroom in mbuf.\n");
341 		}
342 
343 		m->l2_len = sizeof(struct ether_hdr);
344 
345 		/* 02:00:00:00:00:xx */
346 		d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
347 		*((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)port_out << 40);
348 
349 		/* src addr */
350 		ether_addr_copy(&ports_eth_addr[port_out], &eth_hdr->s_addr);
351 		if (ipv6)
352 			eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv6);
353 		else
354 			eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv4);
355 	}
356 
357 	len += len2;
358 
359 	if (likely(len < MAX_PKT_BURST)) {
360 		qconf->tx_mbufs[port_out].len = (uint16_t)len;
361 		return;
362 	}
363 
364 	/* Transmit packets */
365 	send_burst(qconf, (uint16_t)len, port_out);
366 	qconf->tx_mbufs[port_out].len = 0;
367 }
368 
369 /* main processing loop */
370 static int
371 main_loop(__attribute__((unused)) void *dummy)
372 {
373 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
374 	unsigned lcore_id;
375 	uint64_t prev_tsc, diff_tsc, cur_tsc;
376 	int i, j, nb_rx;
377 	uint16_t portid;
378 	struct lcore_queue_conf *qconf;
379 	const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
380 
381 	prev_tsc = 0;
382 
383 	lcore_id = rte_lcore_id();
384 	qconf = &lcore_queue_conf[lcore_id];
385 
386 	if (qconf->n_rx_queue == 0) {
387 		RTE_LOG(INFO, IP_FRAG, "lcore %u has nothing to do\n", lcore_id);
388 		return 0;
389 	}
390 
391 	RTE_LOG(INFO, IP_FRAG, "entering main loop on lcore %u\n", lcore_id);
392 
393 	for (i = 0; i < qconf->n_rx_queue; i++) {
394 
395 		portid = qconf->rx_queue_list[i].portid;
396 		RTE_LOG(INFO, IP_FRAG, " -- lcoreid=%u portid=%d\n", lcore_id,
397 				portid);
398 	}
399 
400 	while (1) {
401 
402 		cur_tsc = rte_rdtsc();
403 
404 		/*
405 		 * TX burst queue drain
406 		 */
407 		diff_tsc = cur_tsc - prev_tsc;
408 		if (unlikely(diff_tsc > drain_tsc)) {
409 
410 			/*
411 			 * This could be optimized (use queueid instead of
412 			 * portid), but it is not called so often
413 			 */
414 			for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
415 				if (qconf->tx_mbufs[portid].len == 0)
416 					continue;
417 				send_burst(&lcore_queue_conf[lcore_id],
418 					   qconf->tx_mbufs[portid].len,
419 					   portid);
420 				qconf->tx_mbufs[portid].len = 0;
421 			}
422 
423 			prev_tsc = cur_tsc;
424 		}
425 
426 		/*
427 		 * Read packet from RX queues
428 		 */
429 		for (i = 0; i < qconf->n_rx_queue; i++) {
430 
431 			portid = qconf->rx_queue_list[i].portid;
432 			nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
433 						 MAX_PKT_BURST);
434 
435 			/* Prefetch first packets */
436 			for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
437 				rte_prefetch0(rte_pktmbuf_mtod(
438 						pkts_burst[j], void *));
439 			}
440 
441 			/* Prefetch and forward already prefetched packets */
442 			for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
443 				rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
444 						j + PREFETCH_OFFSET], void *));
445 				l3fwd_simple_forward(pkts_burst[j], qconf, i, portid);
446 			}
447 
448 			/* Forward remaining prefetched packets */
449 			for (; j < nb_rx; j++) {
450 				l3fwd_simple_forward(pkts_burst[j], qconf, i, portid);
451 			}
452 		}
453 	}
454 }
455 
456 /* display usage */
457 static void
458 print_usage(const char *prgname)
459 {
460 	printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
461 	       "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
462 	       "  -q NQ: number of queue (=ports) per lcore (default is 1)\n",
463 	       prgname);
464 }
465 
466 static int
467 parse_portmask(const char *portmask)
468 {
469 	char *end = NULL;
470 	unsigned long pm;
471 
472 	/* parse hexadecimal string */
473 	pm = strtoul(portmask, &end, 16);
474 	if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
475 		return -1;
476 
477 	if (pm == 0)
478 		return -1;
479 
480 	return pm;
481 }
482 
483 static int
484 parse_nqueue(const char *q_arg)
485 {
486 	char *end = NULL;
487 	unsigned long n;
488 
489 	/* parse hexadecimal string */
490 	n = strtoul(q_arg, &end, 10);
491 	if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
492 		return -1;
493 	if (n == 0)
494 		return -1;
495 	if (n >= MAX_RX_QUEUE_PER_LCORE)
496 		return -1;
497 
498 	return n;
499 }
500 
501 /* Parse the argument given in the command line of the application */
502 static int
503 parse_args(int argc, char **argv)
504 {
505 	int opt, ret;
506 	char **argvopt;
507 	int option_index;
508 	char *prgname = argv[0];
509 	static struct option lgopts[] = {
510 		{NULL, 0, 0, 0}
511 	};
512 
513 	argvopt = argv;
514 
515 	while ((opt = getopt_long(argc, argvopt, "p:q:",
516 				  lgopts, &option_index)) != EOF) {
517 
518 		switch (opt) {
519 		/* portmask */
520 		case 'p':
521 			enabled_port_mask = parse_portmask(optarg);
522 			if (enabled_port_mask < 0) {
523 				printf("invalid portmask\n");
524 				print_usage(prgname);
525 				return -1;
526 			}
527 			break;
528 
529 		/* nqueue */
530 		case 'q':
531 			rx_queue_per_lcore = parse_nqueue(optarg);
532 			if (rx_queue_per_lcore < 0) {
533 				printf("invalid queue number\n");
534 				print_usage(prgname);
535 				return -1;
536 			}
537 			break;
538 
539 		/* long options */
540 		case 0:
541 			print_usage(prgname);
542 			return -1;
543 
544 		default:
545 			print_usage(prgname);
546 			return -1;
547 		}
548 	}
549 
550 	if (enabled_port_mask == 0) {
551 		printf("portmask not specified\n");
552 		print_usage(prgname);
553 		return -1;
554 	}
555 
556 	if (optind >= 0)
557 		argv[optind-1] = prgname;
558 
559 	ret = optind-1;
560 	optind = 1; /* reset getopt lib */
561 	return ret;
562 }
563 
564 static void
565 print_ethaddr(const char *name, struct ether_addr *eth_addr)
566 {
567 	char buf[ETHER_ADDR_FMT_SIZE];
568 	ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
569 	printf("%s%s", name, buf);
570 }
571 
572 /* Check the link status of all ports in up to 9s, and print them finally */
573 static void
574 check_all_ports_link_status(uint16_t port_num, uint32_t port_mask)
575 {
576 #define CHECK_INTERVAL 100 /* 100ms */
577 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
578 	uint16_t portid;
579 	uint8_t count, all_ports_up, print_flag = 0;
580 	struct rte_eth_link link;
581 
582 	printf("\nChecking link status");
583 	fflush(stdout);
584 	for (count = 0; count <= MAX_CHECK_TIME; count++) {
585 		all_ports_up = 1;
586 		for (portid = 0; portid < port_num; portid++) {
587 			if ((port_mask & (1 << portid)) == 0)
588 				continue;
589 			memset(&link, 0, sizeof(link));
590 			rte_eth_link_get_nowait(portid, &link);
591 			/* print link status if flag set */
592 			if (print_flag == 1) {
593 				if (link.link_status)
594 					printf(
595 					"Port%d Link Up .Speed %u Mbps - %s\n",
596 						portid, link.link_speed,
597 				(link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
598 					("full-duplex") : ("half-duplex\n"));
599 				else
600 					printf("Port %d Link Down\n", portid);
601 				continue;
602 			}
603 			/* clear all_ports_up flag if any link down */
604 			if (link.link_status == ETH_LINK_DOWN) {
605 				all_ports_up = 0;
606 				break;
607 			}
608 		}
609 		/* after finally printing all link status, get out */
610 		if (print_flag == 1)
611 			break;
612 
613 		if (all_ports_up == 0) {
614 			printf(".");
615 			fflush(stdout);
616 			rte_delay_ms(CHECK_INTERVAL);
617 		}
618 
619 		/* set the print_flag if all ports up or timeout */
620 		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
621 			print_flag = 1;
622 			printf("\ndone\n");
623 		}
624 	}
625 }
626 
627 /* Check L3 packet type detection capablity of the NIC port */
628 static int
629 check_ptype(int portid)
630 {
631 	int i, ret;
632 	int ptype_l3_ipv4 = 0, ptype_l3_ipv6 = 0;
633 	uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
634 
635 	ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
636 	if (ret <= 0)
637 		return 0;
638 
639 	uint32_t ptypes[ret];
640 
641 	ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
642 	for (i = 0; i < ret; ++i) {
643 		if (ptypes[i] & RTE_PTYPE_L3_IPV4)
644 			ptype_l3_ipv4 = 1;
645 		if (ptypes[i] & RTE_PTYPE_L3_IPV6)
646 			ptype_l3_ipv6 = 1;
647 	}
648 
649 	if (ptype_l3_ipv4 == 0)
650 		printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
651 
652 	if (ptype_l3_ipv6 == 0)
653 		printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
654 
655 	if (ptype_l3_ipv4 && ptype_l3_ipv6)
656 		return 1;
657 
658 	return 0;
659 
660 }
661 
662 /* Parse packet type of a packet by SW */
663 static inline void
664 parse_ptype(struct rte_mbuf *m)
665 {
666 	struct ether_hdr *eth_hdr;
667 	uint32_t packet_type = RTE_PTYPE_UNKNOWN;
668 	uint16_t ether_type;
669 
670 	eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
671 	ether_type = eth_hdr->ether_type;
672 	if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4))
673 		packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
674 	else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6))
675 		packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
676 
677 	m->packet_type = packet_type;
678 }
679 
680 /* callback function to detect packet type for a queue of a port */
681 static uint16_t
682 cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
683 		   struct rte_mbuf *pkts[], uint16_t nb_pkts,
684 		   uint16_t max_pkts __rte_unused,
685 		   void *user_param __rte_unused)
686 {
687 	uint16_t i;
688 
689 	for (i = 0; i < nb_pkts; ++i)
690 		parse_ptype(pkts[i]);
691 
692 	return nb_pkts;
693 }
694 
695 static int
696 init_routing_table(void)
697 {
698 	struct rte_lpm *lpm;
699 	struct rte_lpm6 *lpm6;
700 	int socket, ret;
701 	unsigned i;
702 
703 	for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) {
704 		if (socket_lpm[socket]) {
705 			lpm = socket_lpm[socket];
706 			/* populate the LPM table */
707 			for (i = 0; i < RTE_DIM(l3fwd_ipv4_route_array); i++) {
708 				ret = rte_lpm_add(lpm,
709 					l3fwd_ipv4_route_array[i].ip,
710 					l3fwd_ipv4_route_array[i].depth,
711 					l3fwd_ipv4_route_array[i].if_out);
712 
713 				if (ret < 0) {
714 					RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd "
715 						"LPM table\n", i);
716 					return -1;
717 				}
718 
719 				RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv4_BYTES_FMT
720 						"/%d (port %d)\n",
721 					socket,
722 					IPv4_BYTES(l3fwd_ipv4_route_array[i].ip),
723 					l3fwd_ipv4_route_array[i].depth,
724 					l3fwd_ipv4_route_array[i].if_out);
725 			}
726 		}
727 
728 		if (socket_lpm6[socket]) {
729 			lpm6 = socket_lpm6[socket];
730 			/* populate the LPM6 table */
731 			for (i = 0; i < RTE_DIM(l3fwd_ipv6_route_array); i++) {
732 				ret = rte_lpm6_add(lpm6,
733 					l3fwd_ipv6_route_array[i].ip,
734 					l3fwd_ipv6_route_array[i].depth,
735 					l3fwd_ipv6_route_array[i].if_out);
736 
737 				if (ret < 0) {
738 					RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd "
739 						"LPM6 table\n", i);
740 					return -1;
741 				}
742 
743 				RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv6_BYTES_FMT
744 						"/%d (port %d)\n",
745 					socket,
746 					IPv6_BYTES(l3fwd_ipv6_route_array[i].ip),
747 					l3fwd_ipv6_route_array[i].depth,
748 					l3fwd_ipv6_route_array[i].if_out);
749 			}
750 		}
751 	}
752 	return 0;
753 }
754 
755 static int
756 init_mem(void)
757 {
758 	char buf[PATH_MAX];
759 	struct rte_mempool *mp;
760 	struct rte_lpm *lpm;
761 	struct rte_lpm6 *lpm6;
762 	struct rte_lpm_config lpm_config;
763 	int socket;
764 	unsigned lcore_id;
765 
766 	/* traverse through lcores and initialize structures on each socket */
767 
768 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
769 
770 		if (rte_lcore_is_enabled(lcore_id) == 0)
771 			continue;
772 
773 		socket = rte_lcore_to_socket_id(lcore_id);
774 
775 		if (socket == SOCKET_ID_ANY)
776 			socket = 0;
777 
778 		if (socket_direct_pool[socket] == NULL) {
779 			RTE_LOG(INFO, IP_FRAG, "Creating direct mempool on socket %i\n",
780 					socket);
781 			snprintf(buf, sizeof(buf), "pool_direct_%i", socket);
782 
783 			mp = rte_pktmbuf_pool_create(buf, NB_MBUF, 32,
784 				0, RTE_MBUF_DEFAULT_BUF_SIZE, socket);
785 			if (mp == NULL) {
786 				RTE_LOG(ERR, IP_FRAG, "Cannot create direct mempool\n");
787 				return -1;
788 			}
789 			socket_direct_pool[socket] = mp;
790 		}
791 
792 		if (socket_indirect_pool[socket] == NULL) {
793 			RTE_LOG(INFO, IP_FRAG, "Creating indirect mempool on socket %i\n",
794 					socket);
795 			snprintf(buf, sizeof(buf), "pool_indirect_%i", socket);
796 
797 			mp = rte_pktmbuf_pool_create(buf, NB_MBUF, 32, 0, 0,
798 				socket);
799 			if (mp == NULL) {
800 				RTE_LOG(ERR, IP_FRAG, "Cannot create indirect mempool\n");
801 				return -1;
802 			}
803 			socket_indirect_pool[socket] = mp;
804 		}
805 
806 		if (socket_lpm[socket] == NULL) {
807 			RTE_LOG(INFO, IP_FRAG, "Creating LPM table on socket %i\n", socket);
808 			snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
809 
810 			lpm_config.max_rules = LPM_MAX_RULES;
811 			lpm_config.number_tbl8s = 256;
812 			lpm_config.flags = 0;
813 
814 			lpm = rte_lpm_create(buf, socket, &lpm_config);
815 			if (lpm == NULL) {
816 				RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
817 				return -1;
818 			}
819 			socket_lpm[socket] = lpm;
820 		}
821 
822 		if (socket_lpm6[socket] == NULL) {
823 			RTE_LOG(INFO, IP_FRAG, "Creating LPM6 table on socket %i\n", socket);
824 			snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
825 
826 			lpm6 = rte_lpm6_create(buf, socket, &lpm6_config);
827 			if (lpm6 == NULL) {
828 				RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
829 				return -1;
830 			}
831 			socket_lpm6[socket] = lpm6;
832 		}
833 	}
834 
835 	return 0;
836 }
837 
838 int
839 main(int argc, char **argv)
840 {
841 	struct lcore_queue_conf *qconf;
842 	struct rte_eth_dev_info dev_info;
843 	struct rte_eth_txconf *txconf;
844 	struct rx_queue *rxq;
845 	int socket, ret;
846 	unsigned nb_ports;
847 	uint16_t queueid = 0;
848 	unsigned lcore_id = 0, rx_lcore_id = 0;
849 	uint32_t n_tx_queue, nb_lcores;
850 	uint16_t portid;
851 
852 	/* init EAL */
853 	ret = rte_eal_init(argc, argv);
854 	if (ret < 0)
855 		rte_exit(EXIT_FAILURE, "rte_eal_init failed");
856 	argc -= ret;
857 	argv += ret;
858 
859 	/* parse application arguments (after the EAL ones) */
860 	ret = parse_args(argc, argv);
861 	if (ret < 0)
862 		rte_exit(EXIT_FAILURE, "Invalid arguments");
863 
864 	nb_ports = rte_eth_dev_count();
865 	if (nb_ports == 0)
866 		rte_exit(EXIT_FAILURE, "No ports found!\n");
867 
868 	nb_lcores = rte_lcore_count();
869 
870 	/* initialize structures (mempools, lpm etc.) */
871 	if (init_mem() < 0)
872 		rte_panic("Cannot initialize memory structures!\n");
873 
874 	/* check if portmask has non-existent ports */
875 	if (enabled_port_mask & ~(RTE_LEN2MASK(nb_ports, unsigned)))
876 		rte_exit(EXIT_FAILURE, "Non-existent ports in portmask!\n");
877 
878 	/* initialize all ports */
879 	for (portid = 0; portid < nb_ports; portid++) {
880 		struct rte_eth_conf local_port_conf = port_conf;
881 		struct rte_eth_rxconf rxq_conf;
882 
883 		/* skip ports that are not enabled */
884 		if ((enabled_port_mask & (1 << portid)) == 0) {
885 			printf("Skipping disabled port %d\n", portid);
886 			continue;
887 		}
888 
889 		qconf = &lcore_queue_conf[rx_lcore_id];
890 
891 		/* limit the frame size to the maximum supported by NIC */
892 		rte_eth_dev_info_get(portid, &dev_info);
893 		local_port_conf.rxmode.max_rx_pkt_len = RTE_MIN(
894 		    dev_info.max_rx_pktlen,
895 		    local_port_conf.rxmode.max_rx_pkt_len);
896 
897 		/* get the lcore_id for this port */
898 		while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
899 		       qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
900 
901 			rx_lcore_id ++;
902 			if (rx_lcore_id >= RTE_MAX_LCORE)
903 				rte_exit(EXIT_FAILURE, "Not enough cores\n");
904 
905 			qconf = &lcore_queue_conf[rx_lcore_id];
906 		}
907 
908 		socket = (int) rte_lcore_to_socket_id(rx_lcore_id);
909 		if (socket == SOCKET_ID_ANY)
910 			socket = 0;
911 
912 		rxq = &qconf->rx_queue_list[qconf->n_rx_queue];
913 		rxq->portid = portid;
914 		rxq->direct_pool = socket_direct_pool[socket];
915 		rxq->indirect_pool = socket_indirect_pool[socket];
916 		rxq->lpm = socket_lpm[socket];
917 		rxq->lpm6 = socket_lpm6[socket];
918 		qconf->n_rx_queue++;
919 
920 		/* init port */
921 		printf("Initializing port %d on lcore %u...", portid,
922 		       rx_lcore_id);
923 		fflush(stdout);
924 
925 		n_tx_queue = nb_lcores;
926 		if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
927 			n_tx_queue = MAX_TX_QUEUE_PER_PORT;
928 		if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
929 			local_port_conf.txmode.offloads |=
930 				DEV_TX_OFFLOAD_MBUF_FAST_FREE;
931 		ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
932 					    &local_port_conf);
933 		if (ret < 0) {
934 			printf("\n");
935 			rte_exit(EXIT_FAILURE, "Cannot configure device: "
936 				"err=%d, port=%d\n",
937 				ret, portid);
938 		}
939 
940 		ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
941 					    &nb_txd);
942 		if (ret < 0) {
943 			printf("\n");
944 			rte_exit(EXIT_FAILURE, "Cannot adjust number of "
945 				"descriptors: err=%d, port=%d\n", ret, portid);
946 		}
947 
948 		/* init one RX queue */
949 		rxq_conf = dev_info.default_rxconf;
950 		rxq_conf.offloads = local_port_conf.rxmode.offloads;
951 		ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
952 					     socket, &rxq_conf,
953 					     socket_direct_pool[socket]);
954 		if (ret < 0) {
955 			printf("\n");
956 			rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: "
957 				"err=%d, port=%d\n",
958 				ret, portid);
959 		}
960 
961 		rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
962 		print_ethaddr(" Address:", &ports_eth_addr[portid]);
963 		printf("\n");
964 
965 		/* init one TX queue per couple (lcore,port) */
966 		queueid = 0;
967 		for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
968 			if (rte_lcore_is_enabled(lcore_id) == 0)
969 				continue;
970 
971 			socket = (int) rte_lcore_to_socket_id(lcore_id);
972 			printf("txq=%u,%d ", lcore_id, queueid);
973 			fflush(stdout);
974 
975 			txconf = &dev_info.default_txconf;
976 			txconf->txq_flags = ETH_TXQ_FLAGS_IGNORE;
977 			txconf->offloads = local_port_conf.txmode.offloads;
978 			ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
979 						     socket, txconf);
980 			if (ret < 0) {
981 				printf("\n");
982 				rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
983 					"err=%d, port=%d\n", ret, portid);
984 			}
985 
986 			qconf = &lcore_queue_conf[lcore_id];
987 			qconf->tx_queue_id[portid] = queueid;
988 			queueid++;
989 		}
990 
991 		printf("\n");
992 	}
993 
994 	printf("\n");
995 
996 	/* start ports */
997 	for (portid = 0; portid < nb_ports; portid++) {
998 		if ((enabled_port_mask & (1 << portid)) == 0) {
999 			continue;
1000 		}
1001 		/* Start device */
1002 		ret = rte_eth_dev_start(portid);
1003 		if (ret < 0)
1004 			rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1005 				ret, portid);
1006 
1007 		rte_eth_promiscuous_enable(portid);
1008 
1009 		if (check_ptype(portid) == 0) {
1010 			rte_eth_add_rx_callback(portid, 0, cb_parse_ptype, NULL);
1011 			printf("Add Rx callback function to detect L3 packet type by SW :"
1012 				" port = %d\n", portid);
1013 		}
1014 	}
1015 
1016 	if (init_routing_table() < 0)
1017 		rte_exit(EXIT_FAILURE, "Cannot init routing table\n");
1018 
1019 	check_all_ports_link_status(nb_ports, enabled_port_mask);
1020 
1021 	/* launch per-lcore init on every lcore */
1022 	rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1023 	RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1024 		if (rte_eal_wait_lcore(lcore_id) < 0)
1025 			return -1;
1026 	}
1027 
1028 	return 0;
1029 }
1030