xref: /dpdk/examples/ip_fragmentation/main.c (revision 7bd6f76ee678ec6aa81cb53562f852a43e842718)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2014 Intel Corporation
3  */
4 
5 #include <stdio.h>
6 #include <stdlib.h>
7 #include <stdint.h>
8 #include <inttypes.h>
9 #include <sys/types.h>
10 #include <sys/param.h>
11 #include <string.h>
12 #include <sys/queue.h>
13 #include <stdarg.h>
14 #include <errno.h>
15 #include <getopt.h>
16 
17 #include <rte_common.h>
18 #include <rte_byteorder.h>
19 #include <rte_log.h>
20 #include <rte_memory.h>
21 #include <rte_memcpy.h>
22 #include <rte_eal.h>
23 #include <rte_launch.h>
24 #include <rte_atomic.h>
25 #include <rte_cycles.h>
26 #include <rte_prefetch.h>
27 #include <rte_lcore.h>
28 #include <rte_per_lcore.h>
29 #include <rte_branch_prediction.h>
30 #include <rte_interrupts.h>
31 #include <rte_random.h>
32 #include <rte_debug.h>
33 #include <rte_ether.h>
34 #include <rte_ethdev.h>
35 #include <rte_mempool.h>
36 #include <rte_mbuf.h>
37 #include <rte_lpm.h>
38 #include <rte_lpm6.h>
39 #include <rte_ip.h>
40 #include <rte_string_fns.h>
41 
42 #include <rte_ip_frag.h>
43 
44 #define RTE_LOGTYPE_IP_FRAG RTE_LOGTYPE_USER1
45 
46 /* allow max jumbo frame 9.5 KB */
47 #define JUMBO_FRAME_MAX_SIZE	0x2600
48 
49 #define	ROUNDUP_DIV(a, b)	(((a) + (b) - 1) / (b))
50 
51 /*
52  * Default byte size for the IPv6 Maximum Transfer Unit (MTU).
53  * This value includes the size of IPv6 header.
54  */
55 #define	IPV4_MTU_DEFAULT	ETHER_MTU
56 #define	IPV6_MTU_DEFAULT	ETHER_MTU
57 
58 /*
59  * Default payload in bytes for the IPv6 packet.
60  */
61 #define	IPV4_DEFAULT_PAYLOAD	(IPV4_MTU_DEFAULT - sizeof(struct ipv4_hdr))
62 #define	IPV6_DEFAULT_PAYLOAD	(IPV6_MTU_DEFAULT - sizeof(struct ipv6_hdr))
63 
64 /*
65  * Max number of fragments per packet expected - defined by config file.
66  */
67 #define	MAX_PACKET_FRAG RTE_LIBRTE_IP_FRAG_MAX_FRAG
68 
69 #define NB_MBUF   8192
70 
71 #define MAX_PKT_BURST	32
72 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
73 
74 /* Configure how many packets ahead to prefetch, when reading packets */
75 #define PREFETCH_OFFSET	3
76 
77 /*
78  * Configurable number of RX/TX ring descriptors
79  */
80 #define RTE_TEST_RX_DESC_DEFAULT 1024
81 #define RTE_TEST_TX_DESC_DEFAULT 1024
82 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
83 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
84 
85 /* ethernet addresses of ports */
86 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
87 
88 #ifndef IPv4_BYTES
89 #define IPv4_BYTES_FMT "%" PRIu8 ".%" PRIu8 ".%" PRIu8 ".%" PRIu8
90 #define IPv4_BYTES(addr) \
91 		(uint8_t) (((addr) >> 24) & 0xFF),\
92 		(uint8_t) (((addr) >> 16) & 0xFF),\
93 		(uint8_t) (((addr) >> 8) & 0xFF),\
94 		(uint8_t) ((addr) & 0xFF)
95 #endif
96 
97 #ifndef IPv6_BYTES
98 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
99                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
100 #define IPv6_BYTES(addr) \
101 	addr[0],  addr[1], addr[2],  addr[3], \
102 	addr[4],  addr[5], addr[6],  addr[7], \
103 	addr[8],  addr[9], addr[10], addr[11],\
104 	addr[12], addr[13],addr[14], addr[15]
105 #endif
106 
107 #define IPV6_ADDR_LEN 16
108 
109 /* mask of enabled ports */
110 static int enabled_port_mask = 0;
111 
112 static int rx_queue_per_lcore = 1;
113 
114 #define MBUF_TABLE_SIZE  (2 * MAX(MAX_PKT_BURST, MAX_PACKET_FRAG))
115 
116 struct mbuf_table {
117 	uint16_t len;
118 	struct rte_mbuf *m_table[MBUF_TABLE_SIZE];
119 };
120 
121 struct rx_queue {
122 	struct rte_mempool *direct_pool;
123 	struct rte_mempool *indirect_pool;
124 	struct rte_lpm *lpm;
125 	struct rte_lpm6 *lpm6;
126 	uint16_t portid;
127 };
128 
129 #define MAX_RX_QUEUE_PER_LCORE 16
130 #define MAX_TX_QUEUE_PER_PORT 16
131 struct lcore_queue_conf {
132 	uint16_t n_rx_queue;
133 	uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
134 	struct rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
135 	struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
136 } __rte_cache_aligned;
137 struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
138 
139 static struct rte_eth_conf port_conf = {
140 	.rxmode = {
141 		.max_rx_pkt_len = JUMBO_FRAME_MAX_SIZE,
142 		.split_hdr_size = 0,
143 		.offloads = (DEV_RX_OFFLOAD_CHECKSUM |
144 			     DEV_RX_OFFLOAD_JUMBO_FRAME),
145 	},
146 	.txmode = {
147 		.mq_mode = ETH_MQ_TX_NONE,
148 		.offloads = (DEV_TX_OFFLOAD_IPV4_CKSUM |
149 			     DEV_TX_OFFLOAD_MULTI_SEGS),
150 	},
151 };
152 
153 /*
154  * IPv4 forwarding table
155  */
156 struct l3fwd_ipv4_route {
157 	uint32_t ip;
158 	uint8_t  depth;
159 	uint8_t  if_out;
160 };
161 
162 struct l3fwd_ipv4_route l3fwd_ipv4_route_array[] = {
163 		{IPv4(100,10,0,0), 16, 0},
164 		{IPv4(100,20,0,0), 16, 1},
165 		{IPv4(100,30,0,0), 16, 2},
166 		{IPv4(100,40,0,0), 16, 3},
167 		{IPv4(100,50,0,0), 16, 4},
168 		{IPv4(100,60,0,0), 16, 5},
169 		{IPv4(100,70,0,0), 16, 6},
170 		{IPv4(100,80,0,0), 16, 7},
171 };
172 
173 /*
174  * IPv6 forwarding table
175  */
176 
177 struct l3fwd_ipv6_route {
178 	uint8_t ip[IPV6_ADDR_LEN];
179 	uint8_t depth;
180 	uint8_t if_out;
181 };
182 
183 static struct l3fwd_ipv6_route l3fwd_ipv6_route_array[] = {
184 	{{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
185 	{{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
186 	{{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
187 	{{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
188 	{{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
189 	{{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
190 	{{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
191 	{{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
192 };
193 
194 #define LPM_MAX_RULES         1024
195 #define LPM6_MAX_RULES         1024
196 #define LPM6_NUMBER_TBL8S (1 << 16)
197 
198 struct rte_lpm6_config lpm6_config = {
199 		.max_rules = LPM6_MAX_RULES,
200 		.number_tbl8s = LPM6_NUMBER_TBL8S,
201 		.flags = 0
202 };
203 
204 static struct rte_mempool *socket_direct_pool[RTE_MAX_NUMA_NODES];
205 static struct rte_mempool *socket_indirect_pool[RTE_MAX_NUMA_NODES];
206 static struct rte_lpm *socket_lpm[RTE_MAX_NUMA_NODES];
207 static struct rte_lpm6 *socket_lpm6[RTE_MAX_NUMA_NODES];
208 
209 /* Send burst of packets on an output interface */
210 static inline int
211 send_burst(struct lcore_queue_conf *qconf, uint16_t n, uint16_t port)
212 {
213 	struct rte_mbuf **m_table;
214 	int ret;
215 	uint16_t queueid;
216 
217 	queueid = qconf->tx_queue_id[port];
218 	m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
219 
220 	ret = rte_eth_tx_burst(port, queueid, m_table, n);
221 	if (unlikely(ret < n)) {
222 		do {
223 			rte_pktmbuf_free(m_table[ret]);
224 		} while (++ret < n);
225 	}
226 
227 	return 0;
228 }
229 
230 static inline void
231 l3fwd_simple_forward(struct rte_mbuf *m, struct lcore_queue_conf *qconf,
232 		uint8_t queueid, uint16_t port_in)
233 {
234 	struct rx_queue *rxq;
235 	uint32_t i, len, next_hop;
236 	uint8_t ipv6;
237 	uint16_t port_out;
238 	int32_t len2;
239 
240 	ipv6 = 0;
241 	rxq = &qconf->rx_queue_list[queueid];
242 
243 	/* by default, send everything back to the source port */
244 	port_out = port_in;
245 
246 	/* Remove the Ethernet header and trailer from the input packet */
247 	rte_pktmbuf_adj(m, (uint16_t)sizeof(struct ether_hdr));
248 
249 	/* Build transmission burst */
250 	len = qconf->tx_mbufs[port_out].len;
251 
252 	/* if this is an IPv4 packet */
253 	if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
254 		struct ipv4_hdr *ip_hdr;
255 		uint32_t ip_dst;
256 		/* Read the lookup key (i.e. ip_dst) from the input packet */
257 		ip_hdr = rte_pktmbuf_mtod(m, struct ipv4_hdr *);
258 		ip_dst = rte_be_to_cpu_32(ip_hdr->dst_addr);
259 
260 		/* Find destination port */
261 		if (rte_lpm_lookup(rxq->lpm, ip_dst, &next_hop) == 0 &&
262 				(enabled_port_mask & 1 << next_hop) != 0) {
263 			port_out = next_hop;
264 
265 			/* Build transmission burst for new port */
266 			len = qconf->tx_mbufs[port_out].len;
267 		}
268 
269 		/* if we don't need to do any fragmentation */
270 		if (likely (IPV4_MTU_DEFAULT >= m->pkt_len)) {
271 			qconf->tx_mbufs[port_out].m_table[len] = m;
272 			len2 = 1;
273 		} else {
274 			len2 = rte_ipv4_fragment_packet(m,
275 				&qconf->tx_mbufs[port_out].m_table[len],
276 				(uint16_t)(MBUF_TABLE_SIZE - len),
277 				IPV4_MTU_DEFAULT,
278 				rxq->direct_pool, rxq->indirect_pool);
279 
280 			/* Free input packet */
281 			rte_pktmbuf_free(m);
282 
283 			/* If we fail to fragment the packet */
284 			if (unlikely (len2 < 0))
285 				return;
286 		}
287 	} else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
288 		/* if this is an IPv6 packet */
289 		struct ipv6_hdr *ip_hdr;
290 
291 		ipv6 = 1;
292 
293 		/* Read the lookup key (i.e. ip_dst) from the input packet */
294 		ip_hdr = rte_pktmbuf_mtod(m, struct ipv6_hdr *);
295 
296 		/* Find destination port */
297 		if (rte_lpm6_lookup(rxq->lpm6, ip_hdr->dst_addr,
298 						&next_hop) == 0 &&
299 				(enabled_port_mask & 1 << next_hop) != 0) {
300 			port_out = next_hop;
301 
302 			/* Build transmission burst for new port */
303 			len = qconf->tx_mbufs[port_out].len;
304 		}
305 
306 		/* if we don't need to do any fragmentation */
307 		if (likely (IPV6_MTU_DEFAULT >= m->pkt_len)) {
308 			qconf->tx_mbufs[port_out].m_table[len] = m;
309 			len2 = 1;
310 		} else {
311 			len2 = rte_ipv6_fragment_packet(m,
312 				&qconf->tx_mbufs[port_out].m_table[len],
313 				(uint16_t)(MBUF_TABLE_SIZE - len),
314 				IPV6_MTU_DEFAULT,
315 				rxq->direct_pool, rxq->indirect_pool);
316 
317 			/* Free input packet */
318 			rte_pktmbuf_free(m);
319 
320 			/* If we fail to fragment the packet */
321 			if (unlikely (len2 < 0))
322 				return;
323 		}
324 	}
325 	/* else, just forward the packet */
326 	else {
327 		qconf->tx_mbufs[port_out].m_table[len] = m;
328 		len2 = 1;
329 	}
330 
331 	for (i = len; i < len + len2; i ++) {
332 		void *d_addr_bytes;
333 
334 		m = qconf->tx_mbufs[port_out].m_table[i];
335 		struct ether_hdr *eth_hdr = (struct ether_hdr *)
336 			rte_pktmbuf_prepend(m, (uint16_t)sizeof(struct ether_hdr));
337 		if (eth_hdr == NULL) {
338 			rte_panic("No headroom in mbuf.\n");
339 		}
340 
341 		m->l2_len = sizeof(struct ether_hdr);
342 
343 		/* 02:00:00:00:00:xx */
344 		d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
345 		*((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)port_out << 40);
346 
347 		/* src addr */
348 		ether_addr_copy(&ports_eth_addr[port_out], &eth_hdr->s_addr);
349 		if (ipv6)
350 			eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv6);
351 		else
352 			eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv4);
353 	}
354 
355 	len += len2;
356 
357 	if (likely(len < MAX_PKT_BURST)) {
358 		qconf->tx_mbufs[port_out].len = (uint16_t)len;
359 		return;
360 	}
361 
362 	/* Transmit packets */
363 	send_burst(qconf, (uint16_t)len, port_out);
364 	qconf->tx_mbufs[port_out].len = 0;
365 }
366 
367 /* main processing loop */
368 static int
369 main_loop(__attribute__((unused)) void *dummy)
370 {
371 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
372 	unsigned lcore_id;
373 	uint64_t prev_tsc, diff_tsc, cur_tsc;
374 	int i, j, nb_rx;
375 	uint16_t portid;
376 	struct lcore_queue_conf *qconf;
377 	const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
378 
379 	prev_tsc = 0;
380 
381 	lcore_id = rte_lcore_id();
382 	qconf = &lcore_queue_conf[lcore_id];
383 
384 	if (qconf->n_rx_queue == 0) {
385 		RTE_LOG(INFO, IP_FRAG, "lcore %u has nothing to do\n", lcore_id);
386 		return 0;
387 	}
388 
389 	RTE_LOG(INFO, IP_FRAG, "entering main loop on lcore %u\n", lcore_id);
390 
391 	for (i = 0; i < qconf->n_rx_queue; i++) {
392 
393 		portid = qconf->rx_queue_list[i].portid;
394 		RTE_LOG(INFO, IP_FRAG, " -- lcoreid=%u portid=%d\n", lcore_id,
395 				portid);
396 	}
397 
398 	while (1) {
399 
400 		cur_tsc = rte_rdtsc();
401 
402 		/*
403 		 * TX burst queue drain
404 		 */
405 		diff_tsc = cur_tsc - prev_tsc;
406 		if (unlikely(diff_tsc > drain_tsc)) {
407 
408 			/*
409 			 * This could be optimized (use queueid instead of
410 			 * portid), but it is not called so often
411 			 */
412 			for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
413 				if (qconf->tx_mbufs[portid].len == 0)
414 					continue;
415 				send_burst(&lcore_queue_conf[lcore_id],
416 					   qconf->tx_mbufs[portid].len,
417 					   portid);
418 				qconf->tx_mbufs[portid].len = 0;
419 			}
420 
421 			prev_tsc = cur_tsc;
422 		}
423 
424 		/*
425 		 * Read packet from RX queues
426 		 */
427 		for (i = 0; i < qconf->n_rx_queue; i++) {
428 
429 			portid = qconf->rx_queue_list[i].portid;
430 			nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
431 						 MAX_PKT_BURST);
432 
433 			/* Prefetch first packets */
434 			for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
435 				rte_prefetch0(rte_pktmbuf_mtod(
436 						pkts_burst[j], void *));
437 			}
438 
439 			/* Prefetch and forward already prefetched packets */
440 			for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
441 				rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
442 						j + PREFETCH_OFFSET], void *));
443 				l3fwd_simple_forward(pkts_burst[j], qconf, i, portid);
444 			}
445 
446 			/* Forward remaining prefetched packets */
447 			for (; j < nb_rx; j++) {
448 				l3fwd_simple_forward(pkts_burst[j], qconf, i, portid);
449 			}
450 		}
451 	}
452 }
453 
454 /* display usage */
455 static void
456 print_usage(const char *prgname)
457 {
458 	printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
459 	       "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
460 	       "  -q NQ: number of queue (=ports) per lcore (default is 1)\n",
461 	       prgname);
462 }
463 
464 static int
465 parse_portmask(const char *portmask)
466 {
467 	char *end = NULL;
468 	unsigned long pm;
469 
470 	/* parse hexadecimal string */
471 	pm = strtoul(portmask, &end, 16);
472 	if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
473 		return -1;
474 
475 	if (pm == 0)
476 		return -1;
477 
478 	return pm;
479 }
480 
481 static int
482 parse_nqueue(const char *q_arg)
483 {
484 	char *end = NULL;
485 	unsigned long n;
486 
487 	/* parse hexadecimal string */
488 	n = strtoul(q_arg, &end, 10);
489 	if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
490 		return -1;
491 	if (n == 0)
492 		return -1;
493 	if (n >= MAX_RX_QUEUE_PER_LCORE)
494 		return -1;
495 
496 	return n;
497 }
498 
499 /* Parse the argument given in the command line of the application */
500 static int
501 parse_args(int argc, char **argv)
502 {
503 	int opt, ret;
504 	char **argvopt;
505 	int option_index;
506 	char *prgname = argv[0];
507 	static struct option lgopts[] = {
508 		{NULL, 0, 0, 0}
509 	};
510 
511 	argvopt = argv;
512 
513 	while ((opt = getopt_long(argc, argvopt, "p:q:",
514 				  lgopts, &option_index)) != EOF) {
515 
516 		switch (opt) {
517 		/* portmask */
518 		case 'p':
519 			enabled_port_mask = parse_portmask(optarg);
520 			if (enabled_port_mask < 0) {
521 				printf("invalid portmask\n");
522 				print_usage(prgname);
523 				return -1;
524 			}
525 			break;
526 
527 		/* nqueue */
528 		case 'q':
529 			rx_queue_per_lcore = parse_nqueue(optarg);
530 			if (rx_queue_per_lcore < 0) {
531 				printf("invalid queue number\n");
532 				print_usage(prgname);
533 				return -1;
534 			}
535 			break;
536 
537 		/* long options */
538 		case 0:
539 			print_usage(prgname);
540 			return -1;
541 
542 		default:
543 			print_usage(prgname);
544 			return -1;
545 		}
546 	}
547 
548 	if (enabled_port_mask == 0) {
549 		printf("portmask not specified\n");
550 		print_usage(prgname);
551 		return -1;
552 	}
553 
554 	if (optind >= 0)
555 		argv[optind-1] = prgname;
556 
557 	ret = optind-1;
558 	optind = 1; /* reset getopt lib */
559 	return ret;
560 }
561 
562 static void
563 print_ethaddr(const char *name, struct ether_addr *eth_addr)
564 {
565 	char buf[ETHER_ADDR_FMT_SIZE];
566 	ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
567 	printf("%s%s", name, buf);
568 }
569 
570 /* Check the link status of all ports in up to 9s, and print them finally */
571 static void
572 check_all_ports_link_status(uint32_t port_mask)
573 {
574 #define CHECK_INTERVAL 100 /* 100ms */
575 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
576 	uint16_t portid;
577 	uint8_t count, all_ports_up, print_flag = 0;
578 	struct rte_eth_link link;
579 
580 	printf("\nChecking link status");
581 	fflush(stdout);
582 	for (count = 0; count <= MAX_CHECK_TIME; count++) {
583 		all_ports_up = 1;
584 		RTE_ETH_FOREACH_DEV(portid) {
585 			if ((port_mask & (1 << portid)) == 0)
586 				continue;
587 			memset(&link, 0, sizeof(link));
588 			rte_eth_link_get_nowait(portid, &link);
589 			/* print link status if flag set */
590 			if (print_flag == 1) {
591 				if (link.link_status)
592 					printf(
593 					"Port%d Link Up .Speed %u Mbps - %s\n",
594 						portid, link.link_speed,
595 				(link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
596 					("full-duplex") : ("half-duplex\n"));
597 				else
598 					printf("Port %d Link Down\n", portid);
599 				continue;
600 			}
601 			/* clear all_ports_up flag if any link down */
602 			if (link.link_status == ETH_LINK_DOWN) {
603 				all_ports_up = 0;
604 				break;
605 			}
606 		}
607 		/* after finally printing all link status, get out */
608 		if (print_flag == 1)
609 			break;
610 
611 		if (all_ports_up == 0) {
612 			printf(".");
613 			fflush(stdout);
614 			rte_delay_ms(CHECK_INTERVAL);
615 		}
616 
617 		/* set the print_flag if all ports up or timeout */
618 		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
619 			print_flag = 1;
620 			printf("\ndone\n");
621 		}
622 	}
623 }
624 
625 /* Check L3 packet type detection capablity of the NIC port */
626 static int
627 check_ptype(int portid)
628 {
629 	int i, ret;
630 	int ptype_l3_ipv4 = 0, ptype_l3_ipv6 = 0;
631 	uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
632 
633 	ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
634 	if (ret <= 0)
635 		return 0;
636 
637 	uint32_t ptypes[ret];
638 
639 	ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
640 	for (i = 0; i < ret; ++i) {
641 		if (ptypes[i] & RTE_PTYPE_L3_IPV4)
642 			ptype_l3_ipv4 = 1;
643 		if (ptypes[i] & RTE_PTYPE_L3_IPV6)
644 			ptype_l3_ipv6 = 1;
645 	}
646 
647 	if (ptype_l3_ipv4 == 0)
648 		printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
649 
650 	if (ptype_l3_ipv6 == 0)
651 		printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
652 
653 	if (ptype_l3_ipv4 && ptype_l3_ipv6)
654 		return 1;
655 
656 	return 0;
657 
658 }
659 
660 /* Parse packet type of a packet by SW */
661 static inline void
662 parse_ptype(struct rte_mbuf *m)
663 {
664 	struct ether_hdr *eth_hdr;
665 	uint32_t packet_type = RTE_PTYPE_UNKNOWN;
666 	uint16_t ether_type;
667 
668 	eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
669 	ether_type = eth_hdr->ether_type;
670 	if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4))
671 		packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
672 	else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6))
673 		packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
674 
675 	m->packet_type = packet_type;
676 }
677 
678 /* callback function to detect packet type for a queue of a port */
679 static uint16_t
680 cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
681 		   struct rte_mbuf *pkts[], uint16_t nb_pkts,
682 		   uint16_t max_pkts __rte_unused,
683 		   void *user_param __rte_unused)
684 {
685 	uint16_t i;
686 
687 	for (i = 0; i < nb_pkts; ++i)
688 		parse_ptype(pkts[i]);
689 
690 	return nb_pkts;
691 }
692 
693 static int
694 init_routing_table(void)
695 {
696 	struct rte_lpm *lpm;
697 	struct rte_lpm6 *lpm6;
698 	int socket, ret;
699 	unsigned i;
700 
701 	for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) {
702 		if (socket_lpm[socket]) {
703 			lpm = socket_lpm[socket];
704 			/* populate the LPM table */
705 			for (i = 0; i < RTE_DIM(l3fwd_ipv4_route_array); i++) {
706 				ret = rte_lpm_add(lpm,
707 					l3fwd_ipv4_route_array[i].ip,
708 					l3fwd_ipv4_route_array[i].depth,
709 					l3fwd_ipv4_route_array[i].if_out);
710 
711 				if (ret < 0) {
712 					RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd "
713 						"LPM table\n", i);
714 					return -1;
715 				}
716 
717 				RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv4_BYTES_FMT
718 						"/%d (port %d)\n",
719 					socket,
720 					IPv4_BYTES(l3fwd_ipv4_route_array[i].ip),
721 					l3fwd_ipv4_route_array[i].depth,
722 					l3fwd_ipv4_route_array[i].if_out);
723 			}
724 		}
725 
726 		if (socket_lpm6[socket]) {
727 			lpm6 = socket_lpm6[socket];
728 			/* populate the LPM6 table */
729 			for (i = 0; i < RTE_DIM(l3fwd_ipv6_route_array); i++) {
730 				ret = rte_lpm6_add(lpm6,
731 					l3fwd_ipv6_route_array[i].ip,
732 					l3fwd_ipv6_route_array[i].depth,
733 					l3fwd_ipv6_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 						"LPM6 table\n", i);
738 					return -1;
739 				}
740 
741 				RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv6_BYTES_FMT
742 						"/%d (port %d)\n",
743 					socket,
744 					IPv6_BYTES(l3fwd_ipv6_route_array[i].ip),
745 					l3fwd_ipv6_route_array[i].depth,
746 					l3fwd_ipv6_route_array[i].if_out);
747 			}
748 		}
749 	}
750 	return 0;
751 }
752 
753 static int
754 init_mem(void)
755 {
756 	char buf[PATH_MAX];
757 	struct rte_mempool *mp;
758 	struct rte_lpm *lpm;
759 	struct rte_lpm6 *lpm6;
760 	struct rte_lpm_config lpm_config;
761 	int socket;
762 	unsigned lcore_id;
763 
764 	/* traverse through lcores and initialize structures on each socket */
765 
766 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
767 
768 		if (rte_lcore_is_enabled(lcore_id) == 0)
769 			continue;
770 
771 		socket = rte_lcore_to_socket_id(lcore_id);
772 
773 		if (socket == SOCKET_ID_ANY)
774 			socket = 0;
775 
776 		if (socket_direct_pool[socket] == NULL) {
777 			RTE_LOG(INFO, IP_FRAG, "Creating direct mempool on socket %i\n",
778 					socket);
779 			snprintf(buf, sizeof(buf), "pool_direct_%i", socket);
780 
781 			mp = rte_pktmbuf_pool_create(buf, NB_MBUF, 32,
782 				0, RTE_MBUF_DEFAULT_BUF_SIZE, socket);
783 			if (mp == NULL) {
784 				RTE_LOG(ERR, IP_FRAG, "Cannot create direct mempool\n");
785 				return -1;
786 			}
787 			socket_direct_pool[socket] = mp;
788 		}
789 
790 		if (socket_indirect_pool[socket] == NULL) {
791 			RTE_LOG(INFO, IP_FRAG, "Creating indirect mempool on socket %i\n",
792 					socket);
793 			snprintf(buf, sizeof(buf), "pool_indirect_%i", socket);
794 
795 			mp = rte_pktmbuf_pool_create(buf, NB_MBUF, 32, 0, 0,
796 				socket);
797 			if (mp == NULL) {
798 				RTE_LOG(ERR, IP_FRAG, "Cannot create indirect mempool\n");
799 				return -1;
800 			}
801 			socket_indirect_pool[socket] = mp;
802 		}
803 
804 		if (socket_lpm[socket] == NULL) {
805 			RTE_LOG(INFO, IP_FRAG, "Creating LPM table on socket %i\n", socket);
806 			snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
807 
808 			lpm_config.max_rules = LPM_MAX_RULES;
809 			lpm_config.number_tbl8s = 256;
810 			lpm_config.flags = 0;
811 
812 			lpm = rte_lpm_create(buf, socket, &lpm_config);
813 			if (lpm == NULL) {
814 				RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
815 				return -1;
816 			}
817 			socket_lpm[socket] = lpm;
818 		}
819 
820 		if (socket_lpm6[socket] == NULL) {
821 			RTE_LOG(INFO, IP_FRAG, "Creating LPM6 table on socket %i\n", socket);
822 			snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
823 
824 			lpm6 = rte_lpm6_create(buf, socket, &lpm6_config);
825 			if (lpm6 == NULL) {
826 				RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
827 				return -1;
828 			}
829 			socket_lpm6[socket] = lpm6;
830 		}
831 	}
832 
833 	return 0;
834 }
835 
836 int
837 main(int argc, char **argv)
838 {
839 	struct lcore_queue_conf *qconf;
840 	struct rte_eth_dev_info dev_info;
841 	struct rte_eth_txconf *txconf;
842 	struct rx_queue *rxq;
843 	int socket, ret;
844 	uint16_t nb_ports;
845 	uint16_t queueid = 0;
846 	unsigned lcore_id = 0, rx_lcore_id = 0;
847 	uint32_t n_tx_queue, nb_lcores;
848 	uint16_t portid;
849 
850 	/* init EAL */
851 	ret = rte_eal_init(argc, argv);
852 	if (ret < 0)
853 		rte_exit(EXIT_FAILURE, "rte_eal_init failed");
854 	argc -= ret;
855 	argv += ret;
856 
857 	/* parse application arguments (after the EAL ones) */
858 	ret = parse_args(argc, argv);
859 	if (ret < 0)
860 		rte_exit(EXIT_FAILURE, "Invalid arguments");
861 
862 	nb_ports = rte_eth_dev_count_avail();
863 	if (nb_ports == 0)
864 		rte_exit(EXIT_FAILURE, "No ports found!\n");
865 
866 	nb_lcores = rte_lcore_count();
867 
868 	/* initialize structures (mempools, lpm etc.) */
869 	if (init_mem() < 0)
870 		rte_panic("Cannot initialize memory structures!\n");
871 
872 	/* check if portmask has non-existent ports */
873 	if (enabled_port_mask & ~(RTE_LEN2MASK(nb_ports, unsigned)))
874 		rte_exit(EXIT_FAILURE, "Non-existent ports in portmask!\n");
875 
876 	/* initialize all ports */
877 	RTE_ETH_FOREACH_DEV(portid) {
878 		struct rte_eth_conf local_port_conf = port_conf;
879 		struct rte_eth_rxconf rxq_conf;
880 
881 		/* skip ports that are not enabled */
882 		if ((enabled_port_mask & (1 << portid)) == 0) {
883 			printf("Skipping disabled port %d\n", portid);
884 			continue;
885 		}
886 
887 		qconf = &lcore_queue_conf[rx_lcore_id];
888 
889 		/* limit the frame size to the maximum supported by NIC */
890 		rte_eth_dev_info_get(portid, &dev_info);
891 		local_port_conf.rxmode.max_rx_pkt_len = RTE_MIN(
892 		    dev_info.max_rx_pktlen,
893 		    local_port_conf.rxmode.max_rx_pkt_len);
894 
895 		/* get the lcore_id for this port */
896 		while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
897 		       qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
898 
899 			rx_lcore_id ++;
900 			if (rx_lcore_id >= RTE_MAX_LCORE)
901 				rte_exit(EXIT_FAILURE, "Not enough cores\n");
902 
903 			qconf = &lcore_queue_conf[rx_lcore_id];
904 		}
905 
906 		socket = (int) rte_lcore_to_socket_id(rx_lcore_id);
907 		if (socket == SOCKET_ID_ANY)
908 			socket = 0;
909 
910 		rxq = &qconf->rx_queue_list[qconf->n_rx_queue];
911 		rxq->portid = portid;
912 		rxq->direct_pool = socket_direct_pool[socket];
913 		rxq->indirect_pool = socket_indirect_pool[socket];
914 		rxq->lpm = socket_lpm[socket];
915 		rxq->lpm6 = socket_lpm6[socket];
916 		qconf->n_rx_queue++;
917 
918 		/* init port */
919 		printf("Initializing port %d on lcore %u...", portid,
920 		       rx_lcore_id);
921 		fflush(stdout);
922 
923 		n_tx_queue = nb_lcores;
924 		if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
925 			n_tx_queue = MAX_TX_QUEUE_PER_PORT;
926 		if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
927 			local_port_conf.txmode.offloads |=
928 				DEV_TX_OFFLOAD_MBUF_FAST_FREE;
929 		ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
930 					    &local_port_conf);
931 		if (ret < 0) {
932 			printf("\n");
933 			rte_exit(EXIT_FAILURE, "Cannot configure device: "
934 				"err=%d, port=%d\n",
935 				ret, portid);
936 		}
937 
938 		ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
939 					    &nb_txd);
940 		if (ret < 0) {
941 			printf("\n");
942 			rte_exit(EXIT_FAILURE, "Cannot adjust number of "
943 				"descriptors: err=%d, port=%d\n", ret, portid);
944 		}
945 
946 		/* init one RX queue */
947 		rxq_conf = dev_info.default_rxconf;
948 		rxq_conf.offloads = local_port_conf.rxmode.offloads;
949 		ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
950 					     socket, &rxq_conf,
951 					     socket_direct_pool[socket]);
952 		if (ret < 0) {
953 			printf("\n");
954 			rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: "
955 				"err=%d, port=%d\n",
956 				ret, portid);
957 		}
958 
959 		rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
960 		print_ethaddr(" Address:", &ports_eth_addr[portid]);
961 		printf("\n");
962 
963 		/* init one TX queue per couple (lcore,port) */
964 		queueid = 0;
965 		for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
966 			if (rte_lcore_is_enabled(lcore_id) == 0)
967 				continue;
968 
969 			socket = (int) rte_lcore_to_socket_id(lcore_id);
970 			printf("txq=%u,%d ", lcore_id, queueid);
971 			fflush(stdout);
972 
973 			txconf = &dev_info.default_txconf;
974 			txconf->offloads = local_port_conf.txmode.offloads;
975 			ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
976 						     socket, txconf);
977 			if (ret < 0) {
978 				printf("\n");
979 				rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
980 					"err=%d, port=%d\n", ret, portid);
981 			}
982 
983 			qconf = &lcore_queue_conf[lcore_id];
984 			qconf->tx_queue_id[portid] = queueid;
985 			queueid++;
986 		}
987 
988 		printf("\n");
989 	}
990 
991 	printf("\n");
992 
993 	/* start ports */
994 	RTE_ETH_FOREACH_DEV(portid) {
995 		if ((enabled_port_mask & (1 << portid)) == 0) {
996 			continue;
997 		}
998 		/* Start device */
999 		ret = rte_eth_dev_start(portid);
1000 		if (ret < 0)
1001 			rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1002 				ret, portid);
1003 
1004 		rte_eth_promiscuous_enable(portid);
1005 
1006 		if (check_ptype(portid) == 0) {
1007 			rte_eth_add_rx_callback(portid, 0, cb_parse_ptype, NULL);
1008 			printf("Add Rx callback function to detect L3 packet type by SW :"
1009 				" port = %d\n", portid);
1010 		}
1011 	}
1012 
1013 	if (init_routing_table() < 0)
1014 		rte_exit(EXIT_FAILURE, "Cannot init routing table\n");
1015 
1016 	check_all_ports_link_status(enabled_port_mask);
1017 
1018 	/* launch per-lcore init on every lcore */
1019 	rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1020 	RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1021 		if (rte_eal_wait_lcore(lcore_id) < 0)
1022 			return -1;
1023 	}
1024 
1025 	return 0;
1026 }
1027