xref: /dpdk/examples/ip_fragmentation/main.c (revision 3e0ceb9f17fff027fc6c8f18de35e11719ffa61e)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
5  *   All rights reserved.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
11  *     * Redistributions of source code must retain the above copyright
12  *       notice, this list of conditions and the following disclaimer.
13  *     * Redistributions in binary form must reproduce the above copyright
14  *       notice, this list of conditions and the following disclaimer in
15  *       the documentation and/or other materials provided with the
16  *       distribution.
17  *     * Neither the name of Intel Corporation nor the names of its
18  *       contributors may be used to endorse or promote products derived
19  *       from this software without specific prior written permission.
20  *
21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include <stdio.h>
35 #include <stdlib.h>
36 #include <stdint.h>
37 #include <inttypes.h>
38 #include <sys/types.h>
39 #include <sys/param.h>
40 #include <string.h>
41 #include <sys/queue.h>
42 #include <stdarg.h>
43 #include <errno.h>
44 #include <getopt.h>
45 
46 #include <rte_common.h>
47 #include <rte_byteorder.h>
48 #include <rte_log.h>
49 #include <rte_memory.h>
50 #include <rte_memcpy.h>
51 #include <rte_eal.h>
52 #include <rte_launch.h>
53 #include <rte_atomic.h>
54 #include <rte_cycles.h>
55 #include <rte_prefetch.h>
56 #include <rte_lcore.h>
57 #include <rte_per_lcore.h>
58 #include <rte_branch_prediction.h>
59 #include <rte_interrupts.h>
60 #include <rte_random.h>
61 #include <rte_debug.h>
62 #include <rte_ether.h>
63 #include <rte_ethdev.h>
64 #include <rte_mempool.h>
65 #include <rte_mbuf.h>
66 #include <rte_lpm.h>
67 #include <rte_lpm6.h>
68 #include <rte_ip.h>
69 #include <rte_string_fns.h>
70 
71 #include <rte_ip_frag.h>
72 
73 #define RTE_LOGTYPE_IP_FRAG RTE_LOGTYPE_USER1
74 
75 /* allow max jumbo frame 9.5 KB */
76 #define JUMBO_FRAME_MAX_SIZE	0x2600
77 
78 #define	ROUNDUP_DIV(a, b)	(((a) + (b) - 1) / (b))
79 
80 /*
81  * Default byte size for the IPv6 Maximum Transfer Unit (MTU).
82  * This value includes the size of IPv6 header.
83  */
84 #define	IPV4_MTU_DEFAULT	ETHER_MTU
85 #define	IPV6_MTU_DEFAULT	ETHER_MTU
86 
87 /*
88  * Default payload in bytes for the IPv6 packet.
89  */
90 #define	IPV4_DEFAULT_PAYLOAD	(IPV4_MTU_DEFAULT - sizeof(struct ipv4_hdr))
91 #define	IPV6_DEFAULT_PAYLOAD	(IPV6_MTU_DEFAULT - sizeof(struct ipv6_hdr))
92 
93 /*
94  * Max number of fragments per packet expected - defined by config file.
95  */
96 #define	MAX_PACKET_FRAG RTE_LIBRTE_IP_FRAG_MAX_FRAG
97 
98 #define NB_MBUF   8192
99 
100 #define MAX_PKT_BURST	32
101 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
102 
103 /* Configure how many packets ahead to prefetch, when reading packets */
104 #define PREFETCH_OFFSET	3
105 
106 /*
107  * Configurable number of RX/TX ring descriptors
108  */
109 #define RTE_TEST_RX_DESC_DEFAULT 128
110 #define RTE_TEST_TX_DESC_DEFAULT 512
111 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
112 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
113 
114 /* ethernet addresses of ports */
115 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
116 
117 #ifndef IPv4_BYTES
118 #define IPv4_BYTES_FMT "%" PRIu8 ".%" PRIu8 ".%" PRIu8 ".%" PRIu8
119 #define IPv4_BYTES(addr) \
120 		(uint8_t) (((addr) >> 24) & 0xFF),\
121 		(uint8_t) (((addr) >> 16) & 0xFF),\
122 		(uint8_t) (((addr) >> 8) & 0xFF),\
123 		(uint8_t) ((addr) & 0xFF)
124 #endif
125 
126 #ifndef IPv6_BYTES
127 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
128                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
129 #define IPv6_BYTES(addr) \
130 	addr[0],  addr[1], addr[2],  addr[3], \
131 	addr[4],  addr[5], addr[6],  addr[7], \
132 	addr[8],  addr[9], addr[10], addr[11],\
133 	addr[12], addr[13],addr[14], addr[15]
134 #endif
135 
136 #define IPV6_ADDR_LEN 16
137 
138 /* mask of enabled ports */
139 static int enabled_port_mask = 0;
140 
141 static int rx_queue_per_lcore = 1;
142 
143 #define MBUF_TABLE_SIZE  (2 * MAX(MAX_PKT_BURST, MAX_PACKET_FRAG))
144 
145 struct mbuf_table {
146 	uint16_t len;
147 	struct rte_mbuf *m_table[MBUF_TABLE_SIZE];
148 };
149 
150 struct rx_queue {
151 	struct rte_mempool *direct_pool;
152 	struct rte_mempool *indirect_pool;
153 	struct rte_lpm *lpm;
154 	struct rte_lpm6 *lpm6;
155 	uint16_t portid;
156 };
157 
158 #define MAX_RX_QUEUE_PER_LCORE 16
159 #define MAX_TX_QUEUE_PER_PORT 16
160 struct lcore_queue_conf {
161 	uint16_t n_rx_queue;
162 	uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
163 	struct rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
164 	struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
165 } __rte_cache_aligned;
166 struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
167 
168 static struct rte_eth_conf port_conf = {
169 	.rxmode = {
170 		.max_rx_pkt_len = JUMBO_FRAME_MAX_SIZE,
171 		.split_hdr_size = 0,
172 		.header_split   = 0, /**< Header Split disabled */
173 		.hw_ip_checksum = 1, /**< IP checksum offload enabled */
174 		.hw_vlan_filter = 0, /**< VLAN filtering disabled */
175 		.jumbo_frame    = 1, /**< Jumbo Frame Support enabled */
176 		.hw_strip_crc   = 1, /**< CRC stripped by hardware */
177 	},
178 	.txmode = {
179 		.mq_mode = ETH_MQ_TX_NONE,
180 	},
181 };
182 
183 /*
184  * IPv4 forwarding table
185  */
186 struct l3fwd_ipv4_route {
187 	uint32_t ip;
188 	uint8_t  depth;
189 	uint8_t  if_out;
190 };
191 
192 struct l3fwd_ipv4_route l3fwd_ipv4_route_array[] = {
193 		{IPv4(100,10,0,0), 16, 0},
194 		{IPv4(100,20,0,0), 16, 1},
195 		{IPv4(100,30,0,0), 16, 2},
196 		{IPv4(100,40,0,0), 16, 3},
197 		{IPv4(100,50,0,0), 16, 4},
198 		{IPv4(100,60,0,0), 16, 5},
199 		{IPv4(100,70,0,0), 16, 6},
200 		{IPv4(100,80,0,0), 16, 7},
201 };
202 
203 /*
204  * IPv6 forwarding table
205  */
206 
207 struct l3fwd_ipv6_route {
208 	uint8_t ip[IPV6_ADDR_LEN];
209 	uint8_t depth;
210 	uint8_t if_out;
211 };
212 
213 static struct l3fwd_ipv6_route l3fwd_ipv6_route_array[] = {
214 	{{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
215 	{{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
216 	{{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
217 	{{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
218 	{{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
219 	{{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
220 	{{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
221 	{{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
222 };
223 
224 #define LPM_MAX_RULES         1024
225 #define LPM6_MAX_RULES         1024
226 #define LPM6_NUMBER_TBL8S (1 << 16)
227 
228 struct rte_lpm6_config lpm6_config = {
229 		.max_rules = LPM6_MAX_RULES,
230 		.number_tbl8s = LPM6_NUMBER_TBL8S,
231 		.flags = 0
232 };
233 
234 static struct rte_mempool *socket_direct_pool[RTE_MAX_NUMA_NODES];
235 static struct rte_mempool *socket_indirect_pool[RTE_MAX_NUMA_NODES];
236 static struct rte_lpm *socket_lpm[RTE_MAX_NUMA_NODES];
237 static struct rte_lpm6 *socket_lpm6[RTE_MAX_NUMA_NODES];
238 
239 /* Send burst of packets on an output interface */
240 static inline int
241 send_burst(struct lcore_queue_conf *qconf, uint16_t n, uint16_t port)
242 {
243 	struct rte_mbuf **m_table;
244 	int ret;
245 	uint16_t queueid;
246 
247 	queueid = qconf->tx_queue_id[port];
248 	m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
249 
250 	ret = rte_eth_tx_burst(port, queueid, m_table, n);
251 	if (unlikely(ret < n)) {
252 		do {
253 			rte_pktmbuf_free(m_table[ret]);
254 		} while (++ret < n);
255 	}
256 
257 	return 0;
258 }
259 
260 static inline void
261 l3fwd_simple_forward(struct rte_mbuf *m, struct lcore_queue_conf *qconf,
262 		uint8_t queueid, uint16_t port_in)
263 {
264 	struct rx_queue *rxq;
265 	uint32_t i, len, next_hop;
266 	uint8_t ipv6;
267 	uint16_t port_out;
268 	int32_t len2;
269 
270 	ipv6 = 0;
271 	rxq = &qconf->rx_queue_list[queueid];
272 
273 	/* by default, send everything back to the source port */
274 	port_out = port_in;
275 
276 	/* Remove the Ethernet header and trailer from the input packet */
277 	rte_pktmbuf_adj(m, (uint16_t)sizeof(struct ether_hdr));
278 
279 	/* Build transmission burst */
280 	len = qconf->tx_mbufs[port_out].len;
281 
282 	/* if this is an IPv4 packet */
283 	if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
284 		struct ipv4_hdr *ip_hdr;
285 		uint32_t ip_dst;
286 		/* Read the lookup key (i.e. ip_dst) from the input packet */
287 		ip_hdr = rte_pktmbuf_mtod(m, struct ipv4_hdr *);
288 		ip_dst = rte_be_to_cpu_32(ip_hdr->dst_addr);
289 
290 		/* Find destination port */
291 		if (rte_lpm_lookup(rxq->lpm, ip_dst, &next_hop) == 0 &&
292 				(enabled_port_mask & 1 << next_hop) != 0) {
293 			port_out = next_hop;
294 
295 			/* Build transmission burst for new port */
296 			len = qconf->tx_mbufs[port_out].len;
297 		}
298 
299 		/* if we don't need to do any fragmentation */
300 		if (likely (IPV4_MTU_DEFAULT >= m->pkt_len)) {
301 			qconf->tx_mbufs[port_out].m_table[len] = m;
302 			len2 = 1;
303 		} else {
304 			len2 = rte_ipv4_fragment_packet(m,
305 				&qconf->tx_mbufs[port_out].m_table[len],
306 				(uint16_t)(MBUF_TABLE_SIZE - len),
307 				IPV4_MTU_DEFAULT,
308 				rxq->direct_pool, rxq->indirect_pool);
309 
310 			/* Free input packet */
311 			rte_pktmbuf_free(m);
312 
313 			/* If we fail to fragment the packet */
314 			if (unlikely (len2 < 0))
315 				return;
316 		}
317 	} else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
318 		/* if this is an IPv6 packet */
319 		struct ipv6_hdr *ip_hdr;
320 
321 		ipv6 = 1;
322 
323 		/* Read the lookup key (i.e. ip_dst) from the input packet */
324 		ip_hdr = rte_pktmbuf_mtod(m, struct ipv6_hdr *);
325 
326 		/* Find destination port */
327 		if (rte_lpm6_lookup(rxq->lpm6, ip_hdr->dst_addr,
328 						&next_hop) == 0 &&
329 				(enabled_port_mask & 1 << next_hop) != 0) {
330 			port_out = next_hop;
331 
332 			/* Build transmission burst for new port */
333 			len = qconf->tx_mbufs[port_out].len;
334 		}
335 
336 		/* if we don't need to do any fragmentation */
337 		if (likely (IPV6_MTU_DEFAULT >= m->pkt_len)) {
338 			qconf->tx_mbufs[port_out].m_table[len] = m;
339 			len2 = 1;
340 		} else {
341 			len2 = rte_ipv6_fragment_packet(m,
342 				&qconf->tx_mbufs[port_out].m_table[len],
343 				(uint16_t)(MBUF_TABLE_SIZE - len),
344 				IPV6_MTU_DEFAULT,
345 				rxq->direct_pool, rxq->indirect_pool);
346 
347 			/* Free input packet */
348 			rte_pktmbuf_free(m);
349 
350 			/* If we fail to fragment the packet */
351 			if (unlikely (len2 < 0))
352 				return;
353 		}
354 	}
355 	/* else, just forward the packet */
356 	else {
357 		qconf->tx_mbufs[port_out].m_table[len] = m;
358 		len2 = 1;
359 	}
360 
361 	for (i = len; i < len + len2; i ++) {
362 		void *d_addr_bytes;
363 
364 		m = qconf->tx_mbufs[port_out].m_table[i];
365 		struct ether_hdr *eth_hdr = (struct ether_hdr *)
366 			rte_pktmbuf_prepend(m, (uint16_t)sizeof(struct ether_hdr));
367 		if (eth_hdr == NULL) {
368 			rte_panic("No headroom in mbuf.\n");
369 		}
370 
371 		m->l2_len = sizeof(struct ether_hdr);
372 
373 		/* 02:00:00:00:00:xx */
374 		d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
375 		*((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)port_out << 40);
376 
377 		/* src addr */
378 		ether_addr_copy(&ports_eth_addr[port_out], &eth_hdr->s_addr);
379 		if (ipv6)
380 			eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv6);
381 		else
382 			eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv4);
383 	}
384 
385 	len += len2;
386 
387 	if (likely(len < MAX_PKT_BURST)) {
388 		qconf->tx_mbufs[port_out].len = (uint16_t)len;
389 		return;
390 	}
391 
392 	/* Transmit packets */
393 	send_burst(qconf, (uint16_t)len, port_out);
394 	qconf->tx_mbufs[port_out].len = 0;
395 }
396 
397 /* main processing loop */
398 static int
399 main_loop(__attribute__((unused)) void *dummy)
400 {
401 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
402 	unsigned lcore_id;
403 	uint64_t prev_tsc, diff_tsc, cur_tsc;
404 	int i, j, nb_rx;
405 	uint16_t portid;
406 	struct lcore_queue_conf *qconf;
407 	const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
408 
409 	prev_tsc = 0;
410 
411 	lcore_id = rte_lcore_id();
412 	qconf = &lcore_queue_conf[lcore_id];
413 
414 	if (qconf->n_rx_queue == 0) {
415 		RTE_LOG(INFO, IP_FRAG, "lcore %u has nothing to do\n", lcore_id);
416 		return 0;
417 	}
418 
419 	RTE_LOG(INFO, IP_FRAG, "entering main loop on lcore %u\n", lcore_id);
420 
421 	for (i = 0; i < qconf->n_rx_queue; i++) {
422 
423 		portid = qconf->rx_queue_list[i].portid;
424 		RTE_LOG(INFO, IP_FRAG, " -- lcoreid=%u portid=%d\n", lcore_id,
425 				portid);
426 	}
427 
428 	while (1) {
429 
430 		cur_tsc = rte_rdtsc();
431 
432 		/*
433 		 * TX burst queue drain
434 		 */
435 		diff_tsc = cur_tsc - prev_tsc;
436 		if (unlikely(diff_tsc > drain_tsc)) {
437 
438 			/*
439 			 * This could be optimized (use queueid instead of
440 			 * portid), but it is not called so often
441 			 */
442 			for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
443 				if (qconf->tx_mbufs[portid].len == 0)
444 					continue;
445 				send_burst(&lcore_queue_conf[lcore_id],
446 					   qconf->tx_mbufs[portid].len,
447 					   portid);
448 				qconf->tx_mbufs[portid].len = 0;
449 			}
450 
451 			prev_tsc = cur_tsc;
452 		}
453 
454 		/*
455 		 * Read packet from RX queues
456 		 */
457 		for (i = 0; i < qconf->n_rx_queue; i++) {
458 
459 			portid = qconf->rx_queue_list[i].portid;
460 			nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
461 						 MAX_PKT_BURST);
462 
463 			/* Prefetch first packets */
464 			for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
465 				rte_prefetch0(rte_pktmbuf_mtod(
466 						pkts_burst[j], void *));
467 			}
468 
469 			/* Prefetch and forward already prefetched packets */
470 			for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
471 				rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
472 						j + PREFETCH_OFFSET], void *));
473 				l3fwd_simple_forward(pkts_burst[j], qconf, i, portid);
474 			}
475 
476 			/* Forward remaining prefetched packets */
477 			for (; j < nb_rx; j++) {
478 				l3fwd_simple_forward(pkts_burst[j], qconf, i, portid);
479 			}
480 		}
481 	}
482 }
483 
484 /* display usage */
485 static void
486 print_usage(const char *prgname)
487 {
488 	printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
489 	       "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
490 	       "  -q NQ: number of queue (=ports) per lcore (default is 1)\n",
491 	       prgname);
492 }
493 
494 static int
495 parse_portmask(const char *portmask)
496 {
497 	char *end = NULL;
498 	unsigned long pm;
499 
500 	/* parse hexadecimal string */
501 	pm = strtoul(portmask, &end, 16);
502 	if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
503 		return -1;
504 
505 	if (pm == 0)
506 		return -1;
507 
508 	return pm;
509 }
510 
511 static int
512 parse_nqueue(const char *q_arg)
513 {
514 	char *end = NULL;
515 	unsigned long n;
516 
517 	/* parse hexadecimal string */
518 	n = strtoul(q_arg, &end, 10);
519 	if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
520 		return -1;
521 	if (n == 0)
522 		return -1;
523 	if (n >= MAX_RX_QUEUE_PER_LCORE)
524 		return -1;
525 
526 	return n;
527 }
528 
529 /* Parse the argument given in the command line of the application */
530 static int
531 parse_args(int argc, char **argv)
532 {
533 	int opt, ret;
534 	char **argvopt;
535 	int option_index;
536 	char *prgname = argv[0];
537 	static struct option lgopts[] = {
538 		{NULL, 0, 0, 0}
539 	};
540 
541 	argvopt = argv;
542 
543 	while ((opt = getopt_long(argc, argvopt, "p:q:",
544 				  lgopts, &option_index)) != EOF) {
545 
546 		switch (opt) {
547 		/* portmask */
548 		case 'p':
549 			enabled_port_mask = parse_portmask(optarg);
550 			if (enabled_port_mask < 0) {
551 				printf("invalid portmask\n");
552 				print_usage(prgname);
553 				return -1;
554 			}
555 			break;
556 
557 		/* nqueue */
558 		case 'q':
559 			rx_queue_per_lcore = parse_nqueue(optarg);
560 			if (rx_queue_per_lcore < 0) {
561 				printf("invalid queue number\n");
562 				print_usage(prgname);
563 				return -1;
564 			}
565 			break;
566 
567 		/* long options */
568 		case 0:
569 			print_usage(prgname);
570 			return -1;
571 
572 		default:
573 			print_usage(prgname);
574 			return -1;
575 		}
576 	}
577 
578 	if (enabled_port_mask == 0) {
579 		printf("portmask not specified\n");
580 		print_usage(prgname);
581 		return -1;
582 	}
583 
584 	if (optind >= 0)
585 		argv[optind-1] = prgname;
586 
587 	ret = optind-1;
588 	optind = 1; /* reset getopt lib */
589 	return ret;
590 }
591 
592 static void
593 print_ethaddr(const char *name, struct ether_addr *eth_addr)
594 {
595 	char buf[ETHER_ADDR_FMT_SIZE];
596 	ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
597 	printf("%s%s", name, buf);
598 }
599 
600 /* Check the link status of all ports in up to 9s, and print them finally */
601 static void
602 check_all_ports_link_status(uint16_t port_num, uint32_t port_mask)
603 {
604 #define CHECK_INTERVAL 100 /* 100ms */
605 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
606 	uint16_t portid;
607 	uint8_t count, all_ports_up, print_flag = 0;
608 	struct rte_eth_link link;
609 
610 	printf("\nChecking link status");
611 	fflush(stdout);
612 	for (count = 0; count <= MAX_CHECK_TIME; count++) {
613 		all_ports_up = 1;
614 		for (portid = 0; portid < port_num; portid++) {
615 			if ((port_mask & (1 << portid)) == 0)
616 				continue;
617 			memset(&link, 0, sizeof(link));
618 			rte_eth_link_get_nowait(portid, &link);
619 			/* print link status if flag set */
620 			if (print_flag == 1) {
621 				if (link.link_status)
622 					printf(
623 					"Port%d Link Up .Speed %u Mbps - %s\n",
624 						portid, link.link_speed,
625 				(link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
626 					("full-duplex") : ("half-duplex\n"));
627 				else
628 					printf("Port %d Link Down\n", portid);
629 				continue;
630 			}
631 			/* clear all_ports_up flag if any link down */
632 			if (link.link_status == ETH_LINK_DOWN) {
633 				all_ports_up = 0;
634 				break;
635 			}
636 		}
637 		/* after finally printing all link status, get out */
638 		if (print_flag == 1)
639 			break;
640 
641 		if (all_ports_up == 0) {
642 			printf(".");
643 			fflush(stdout);
644 			rte_delay_ms(CHECK_INTERVAL);
645 		}
646 
647 		/* set the print_flag if all ports up or timeout */
648 		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
649 			print_flag = 1;
650 			printf("\ndone\n");
651 		}
652 	}
653 }
654 
655 /* Check L3 packet type detection capablity of the NIC port */
656 static int
657 check_ptype(int portid)
658 {
659 	int i, ret;
660 	int ptype_l3_ipv4 = 0, ptype_l3_ipv6 = 0;
661 	uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
662 
663 	ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
664 	if (ret <= 0)
665 		return 0;
666 
667 	uint32_t ptypes[ret];
668 
669 	ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
670 	for (i = 0; i < ret; ++i) {
671 		if (ptypes[i] & RTE_PTYPE_L3_IPV4)
672 			ptype_l3_ipv4 = 1;
673 		if (ptypes[i] & RTE_PTYPE_L3_IPV6)
674 			ptype_l3_ipv6 = 1;
675 	}
676 
677 	if (ptype_l3_ipv4 == 0)
678 		printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
679 
680 	if (ptype_l3_ipv6 == 0)
681 		printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
682 
683 	if (ptype_l3_ipv4 && ptype_l3_ipv6)
684 		return 1;
685 
686 	return 0;
687 
688 }
689 
690 /* Parse packet type of a packet by SW */
691 static inline void
692 parse_ptype(struct rte_mbuf *m)
693 {
694 	struct ether_hdr *eth_hdr;
695 	uint32_t packet_type = RTE_PTYPE_UNKNOWN;
696 	uint16_t ether_type;
697 
698 	eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
699 	ether_type = eth_hdr->ether_type;
700 	if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4))
701 		packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
702 	else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6))
703 		packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
704 
705 	m->packet_type = packet_type;
706 }
707 
708 /* callback function to detect packet type for a queue of a port */
709 static uint16_t
710 cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
711 		   struct rte_mbuf *pkts[], uint16_t nb_pkts,
712 		   uint16_t max_pkts __rte_unused,
713 		   void *user_param __rte_unused)
714 {
715 	uint16_t i;
716 
717 	for (i = 0; i < nb_pkts; ++i)
718 		parse_ptype(pkts[i]);
719 
720 	return nb_pkts;
721 }
722 
723 static int
724 init_routing_table(void)
725 {
726 	struct rte_lpm *lpm;
727 	struct rte_lpm6 *lpm6;
728 	int socket, ret;
729 	unsigned i;
730 
731 	for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) {
732 		if (socket_lpm[socket]) {
733 			lpm = socket_lpm[socket];
734 			/* populate the LPM table */
735 			for (i = 0; i < RTE_DIM(l3fwd_ipv4_route_array); i++) {
736 				ret = rte_lpm_add(lpm,
737 					l3fwd_ipv4_route_array[i].ip,
738 					l3fwd_ipv4_route_array[i].depth,
739 					l3fwd_ipv4_route_array[i].if_out);
740 
741 				if (ret < 0) {
742 					RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd "
743 						"LPM table\n", i);
744 					return -1;
745 				}
746 
747 				RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv4_BYTES_FMT
748 						"/%d (port %d)\n",
749 					socket,
750 					IPv4_BYTES(l3fwd_ipv4_route_array[i].ip),
751 					l3fwd_ipv4_route_array[i].depth,
752 					l3fwd_ipv4_route_array[i].if_out);
753 			}
754 		}
755 
756 		if (socket_lpm6[socket]) {
757 			lpm6 = socket_lpm6[socket];
758 			/* populate the LPM6 table */
759 			for (i = 0; i < RTE_DIM(l3fwd_ipv6_route_array); i++) {
760 				ret = rte_lpm6_add(lpm6,
761 					l3fwd_ipv6_route_array[i].ip,
762 					l3fwd_ipv6_route_array[i].depth,
763 					l3fwd_ipv6_route_array[i].if_out);
764 
765 				if (ret < 0) {
766 					RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd "
767 						"LPM6 table\n", i);
768 					return -1;
769 				}
770 
771 				RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv6_BYTES_FMT
772 						"/%d (port %d)\n",
773 					socket,
774 					IPv6_BYTES(l3fwd_ipv6_route_array[i].ip),
775 					l3fwd_ipv6_route_array[i].depth,
776 					l3fwd_ipv6_route_array[i].if_out);
777 			}
778 		}
779 	}
780 	return 0;
781 }
782 
783 static int
784 init_mem(void)
785 {
786 	char buf[PATH_MAX];
787 	struct rte_mempool *mp;
788 	struct rte_lpm *lpm;
789 	struct rte_lpm6 *lpm6;
790 	struct rte_lpm_config lpm_config;
791 	int socket;
792 	unsigned lcore_id;
793 
794 	/* traverse through lcores and initialize structures on each socket */
795 
796 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
797 
798 		if (rte_lcore_is_enabled(lcore_id) == 0)
799 			continue;
800 
801 		socket = rte_lcore_to_socket_id(lcore_id);
802 
803 		if (socket == SOCKET_ID_ANY)
804 			socket = 0;
805 
806 		if (socket_direct_pool[socket] == NULL) {
807 			RTE_LOG(INFO, IP_FRAG, "Creating direct mempool on socket %i\n",
808 					socket);
809 			snprintf(buf, sizeof(buf), "pool_direct_%i", socket);
810 
811 			mp = rte_pktmbuf_pool_create(buf, NB_MBUF, 32,
812 				0, RTE_MBUF_DEFAULT_BUF_SIZE, socket);
813 			if (mp == NULL) {
814 				RTE_LOG(ERR, IP_FRAG, "Cannot create direct mempool\n");
815 				return -1;
816 			}
817 			socket_direct_pool[socket] = mp;
818 		}
819 
820 		if (socket_indirect_pool[socket] == NULL) {
821 			RTE_LOG(INFO, IP_FRAG, "Creating indirect mempool on socket %i\n",
822 					socket);
823 			snprintf(buf, sizeof(buf), "pool_indirect_%i", socket);
824 
825 			mp = rte_pktmbuf_pool_create(buf, NB_MBUF, 32, 0, 0,
826 				socket);
827 			if (mp == NULL) {
828 				RTE_LOG(ERR, IP_FRAG, "Cannot create indirect mempool\n");
829 				return -1;
830 			}
831 			socket_indirect_pool[socket] = mp;
832 		}
833 
834 		if (socket_lpm[socket] == NULL) {
835 			RTE_LOG(INFO, IP_FRAG, "Creating LPM table on socket %i\n", socket);
836 			snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
837 
838 			lpm_config.max_rules = LPM_MAX_RULES;
839 			lpm_config.number_tbl8s = 256;
840 			lpm_config.flags = 0;
841 
842 			lpm = rte_lpm_create(buf, socket, &lpm_config);
843 			if (lpm == NULL) {
844 				RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
845 				return -1;
846 			}
847 			socket_lpm[socket] = lpm;
848 		}
849 
850 		if (socket_lpm6[socket] == NULL) {
851 			RTE_LOG(INFO, IP_FRAG, "Creating LPM6 table on socket %i\n", socket);
852 			snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
853 
854 			lpm6 = rte_lpm6_create(buf, socket, &lpm6_config);
855 			if (lpm6 == NULL) {
856 				RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
857 				return -1;
858 			}
859 			socket_lpm6[socket] = lpm6;
860 		}
861 	}
862 
863 	return 0;
864 }
865 
866 int
867 main(int argc, char **argv)
868 {
869 	struct lcore_queue_conf *qconf;
870 	struct rte_eth_dev_info dev_info;
871 	struct rte_eth_txconf *txconf;
872 	struct rx_queue *rxq;
873 	int socket, ret;
874 	unsigned nb_ports;
875 	uint16_t queueid = 0;
876 	unsigned lcore_id = 0, rx_lcore_id = 0;
877 	uint32_t n_tx_queue, nb_lcores;
878 	uint16_t portid;
879 
880 	/* init EAL */
881 	ret = rte_eal_init(argc, argv);
882 	if (ret < 0)
883 		rte_exit(EXIT_FAILURE, "rte_eal_init failed");
884 	argc -= ret;
885 	argv += ret;
886 
887 	/* parse application arguments (after the EAL ones) */
888 	ret = parse_args(argc, argv);
889 	if (ret < 0)
890 		rte_exit(EXIT_FAILURE, "Invalid arguments");
891 
892 	nb_ports = rte_eth_dev_count();
893 	if (nb_ports == 0)
894 		rte_exit(EXIT_FAILURE, "No ports found!\n");
895 
896 	nb_lcores = rte_lcore_count();
897 
898 	/* initialize structures (mempools, lpm etc.) */
899 	if (init_mem() < 0)
900 		rte_panic("Cannot initialize memory structures!\n");
901 
902 	/* check if portmask has non-existent ports */
903 	if (enabled_port_mask & ~(RTE_LEN2MASK(nb_ports, unsigned)))
904 		rte_exit(EXIT_FAILURE, "Non-existent ports in portmask!\n");
905 
906 	/* initialize all ports */
907 	for (portid = 0; portid < nb_ports; portid++) {
908 		/* skip ports that are not enabled */
909 		if ((enabled_port_mask & (1 << portid)) == 0) {
910 			printf("Skipping disabled port %d\n", portid);
911 			continue;
912 		}
913 
914 		qconf = &lcore_queue_conf[rx_lcore_id];
915 
916 		/* limit the frame size to the maximum supported by NIC */
917 		rte_eth_dev_info_get(portid, &dev_info);
918 		port_conf.rxmode.max_rx_pkt_len = RTE_MIN(
919 		    dev_info.max_rx_pktlen, port_conf.rxmode.max_rx_pkt_len);
920 
921 		/* get the lcore_id for this port */
922 		while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
923 		       qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
924 
925 			rx_lcore_id ++;
926 			if (rx_lcore_id >= RTE_MAX_LCORE)
927 				rte_exit(EXIT_FAILURE, "Not enough cores\n");
928 
929 			qconf = &lcore_queue_conf[rx_lcore_id];
930 		}
931 
932 		socket = (int) rte_lcore_to_socket_id(rx_lcore_id);
933 		if (socket == SOCKET_ID_ANY)
934 			socket = 0;
935 
936 		rxq = &qconf->rx_queue_list[qconf->n_rx_queue];
937 		rxq->portid = portid;
938 		rxq->direct_pool = socket_direct_pool[socket];
939 		rxq->indirect_pool = socket_indirect_pool[socket];
940 		rxq->lpm = socket_lpm[socket];
941 		rxq->lpm6 = socket_lpm6[socket];
942 		qconf->n_rx_queue++;
943 
944 		/* init port */
945 		printf("Initializing port %d on lcore %u...", portid,
946 		       rx_lcore_id);
947 		fflush(stdout);
948 
949 		n_tx_queue = nb_lcores;
950 		if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
951 			n_tx_queue = MAX_TX_QUEUE_PER_PORT;
952 		ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
953 					    &port_conf);
954 		if (ret < 0) {
955 			printf("\n");
956 			rte_exit(EXIT_FAILURE, "Cannot configure device: "
957 				"err=%d, port=%d\n",
958 				ret, portid);
959 		}
960 
961 		ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
962 					    &nb_txd);
963 		if (ret < 0) {
964 			printf("\n");
965 			rte_exit(EXIT_FAILURE, "Cannot adjust number of "
966 				"descriptors: err=%d, port=%d\n", ret, portid);
967 		}
968 
969 		/* init one RX queue */
970 		ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
971 					     socket, NULL,
972 					     socket_direct_pool[socket]);
973 		if (ret < 0) {
974 			printf("\n");
975 			rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: "
976 				"err=%d, port=%d\n",
977 				ret, portid);
978 		}
979 
980 		rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
981 		print_ethaddr(" Address:", &ports_eth_addr[portid]);
982 		printf("\n");
983 
984 		/* init one TX queue per couple (lcore,port) */
985 		queueid = 0;
986 		for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
987 			if (rte_lcore_is_enabled(lcore_id) == 0)
988 				continue;
989 
990 			socket = (int) rte_lcore_to_socket_id(lcore_id);
991 			printf("txq=%u,%d ", lcore_id, queueid);
992 			fflush(stdout);
993 
994 			txconf = &dev_info.default_txconf;
995 			txconf->txq_flags = 0;
996 			ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
997 						     socket, txconf);
998 			if (ret < 0) {
999 				printf("\n");
1000 				rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
1001 					"err=%d, port=%d\n", ret, portid);
1002 			}
1003 
1004 			qconf = &lcore_queue_conf[lcore_id];
1005 			qconf->tx_queue_id[portid] = queueid;
1006 			queueid++;
1007 		}
1008 
1009 		printf("\n");
1010 	}
1011 
1012 	printf("\n");
1013 
1014 	/* start ports */
1015 	for (portid = 0; portid < nb_ports; portid++) {
1016 		if ((enabled_port_mask & (1 << portid)) == 0) {
1017 			continue;
1018 		}
1019 		/* Start device */
1020 		ret = rte_eth_dev_start(portid);
1021 		if (ret < 0)
1022 			rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1023 				ret, portid);
1024 
1025 		rte_eth_promiscuous_enable(portid);
1026 
1027 		if (check_ptype(portid) == 0) {
1028 			rte_eth_add_rx_callback(portid, 0, cb_parse_ptype, NULL);
1029 			printf("Add Rx callback function to detect L3 packet type by SW :"
1030 				" port = %d\n", portid);
1031 		}
1032 	}
1033 
1034 	if (init_routing_table() < 0)
1035 		rte_exit(EXIT_FAILURE, "Cannot init routing table\n");
1036 
1037 	check_all_ports_link_status(nb_ports, enabled_port_mask);
1038 
1039 	/* launch per-lcore init on every lcore */
1040 	rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1041 	RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1042 		if (rte_eal_wait_lcore(lcore_id) < 0)
1043 			return -1;
1044 	}
1045 
1046 	return 0;
1047 }
1048