xref: /dpdk/examples/ip_fragmentation/main.c (revision fea1d908d39989a27890b29b5c0ec94c85c8257b)
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_memzone.h>
52 #include <rte_eal.h>
53 #include <rte_per_lcore.h>
54 #include <rte_launch.h>
55 #include <rte_atomic.h>
56 #include <rte_cycles.h>
57 #include <rte_prefetch.h>
58 #include <rte_lcore.h>
59 #include <rte_per_lcore.h>
60 #include <rte_branch_prediction.h>
61 #include <rte_interrupts.h>
62 #include <rte_pci.h>
63 #include <rte_random.h>
64 #include <rte_debug.h>
65 #include <rte_ether.h>
66 #include <rte_ethdev.h>
67 #include <rte_ring.h>
68 #include <rte_mempool.h>
69 #include <rte_mbuf.h>
70 #include <rte_lpm.h>
71 #include <rte_lpm6.h>
72 #include <rte_ip.h>
73 #include <rte_string_fns.h>
74 
75 #include <rte_ip_frag.h>
76 
77 #define RTE_LOGTYPE_IP_FRAG RTE_LOGTYPE_USER1
78 
79 #define MBUF_SIZE (2048 + RTE_PKTMBUF_HEADROOM)
80 
81 /* allow max jumbo frame 9.5 KB */
82 #define JUMBO_FRAME_MAX_SIZE	0x2600
83 
84 #define	ROUNDUP_DIV(a, b)	(((a) + (b) - 1) / (b))
85 
86 /*
87  * Default byte size for the IPv6 Maximum Transfer Unit (MTU).
88  * This value includes the size of IPv6 header.
89  */
90 #define	IPV4_MTU_DEFAULT	ETHER_MTU
91 #define	IPV6_MTU_DEFAULT	ETHER_MTU
92 
93 /*
94  * Default payload in bytes for the IPv6 packet.
95  */
96 #define	IPV4_DEFAULT_PAYLOAD	(IPV4_MTU_DEFAULT - sizeof(struct ipv4_hdr))
97 #define	IPV6_DEFAULT_PAYLOAD	(IPV6_MTU_DEFAULT - sizeof(struct ipv6_hdr))
98 
99 /*
100  * Max number of fragments per packet expected - defined by config file.
101  */
102 #define	MAX_PACKET_FRAG RTE_LIBRTE_IP_FRAG_MAX_FRAG
103 
104 #define NB_MBUF   8192
105 
106 #define MAX_PKT_BURST	32
107 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
108 
109 /* Configure how many packets ahead to prefetch, when reading packets */
110 #define PREFETCH_OFFSET	3
111 
112 /*
113  * Configurable number of RX/TX ring descriptors
114  */
115 #define RTE_TEST_RX_DESC_DEFAULT 128
116 #define RTE_TEST_TX_DESC_DEFAULT 512
117 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
118 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
119 
120 /* ethernet addresses of ports */
121 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
122 
123 #ifndef IPv4_BYTES
124 #define IPv4_BYTES_FMT "%" PRIu8 ".%" PRIu8 ".%" PRIu8 ".%" PRIu8
125 #define IPv4_BYTES(addr) \
126 		(uint8_t) (((addr) >> 24) & 0xFF),\
127 		(uint8_t) (((addr) >> 16) & 0xFF),\
128 		(uint8_t) (((addr) >> 8) & 0xFF),\
129 		(uint8_t) ((addr) & 0xFF)
130 #endif
131 
132 #ifndef IPv6_BYTES
133 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
134                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
135 #define IPv6_BYTES(addr) \
136 	addr[0],  addr[1], addr[2],  addr[3], \
137 	addr[4],  addr[5], addr[6],  addr[7], \
138 	addr[8],  addr[9], addr[10], addr[11],\
139 	addr[12], addr[13],addr[14], addr[15]
140 #endif
141 
142 #define IPV6_ADDR_LEN 16
143 
144 /* mask of enabled ports */
145 static int enabled_port_mask = 0;
146 
147 static int rx_queue_per_lcore = 1;
148 
149 #define MBUF_TABLE_SIZE  (2 * MAX(MAX_PKT_BURST, MAX_PACKET_FRAG))
150 
151 struct mbuf_table {
152 	uint16_t len;
153 	struct rte_mbuf *m_table[MBUF_TABLE_SIZE];
154 };
155 
156 struct rx_queue {
157 	struct rte_mempool *direct_pool;
158 	struct rte_mempool *indirect_pool;
159 	struct rte_lpm *lpm;
160 	struct rte_lpm6 *lpm6;
161 	uint8_t portid;
162 };
163 
164 #define MAX_RX_QUEUE_PER_LCORE 16
165 #define MAX_TX_QUEUE_PER_PORT 16
166 struct lcore_queue_conf {
167 	uint16_t n_rx_queue;
168 	uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
169 	struct rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
170 	struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
171 } __rte_cache_aligned;
172 struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
173 
174 static const struct rte_eth_conf port_conf = {
175 	.rxmode = {
176 		.max_rx_pkt_len = JUMBO_FRAME_MAX_SIZE,
177 		.split_hdr_size = 0,
178 		.header_split   = 0, /**< Header Split disabled */
179 		.hw_ip_checksum = 1, /**< IP checksum offload enabled */
180 		.hw_vlan_filter = 0, /**< VLAN filtering disabled */
181 		.jumbo_frame    = 1, /**< Jumbo Frame Support enabled */
182 		.hw_strip_crc   = 0, /**< CRC stripped by hardware */
183 	},
184 	.txmode = {
185 		.mq_mode = ETH_MQ_TX_NONE,
186 	},
187 };
188 
189 /*
190  * IPv4 forwarding table
191  */
192 struct l3fwd_ipv4_route {
193 	uint32_t ip;
194 	uint8_t  depth;
195 	uint8_t  if_out;
196 };
197 
198 struct l3fwd_ipv4_route l3fwd_ipv4_route_array[] = {
199 		{IPv4(100,10,0,0), 16, 0},
200 		{IPv4(100,20,0,0), 16, 1},
201 		{IPv4(100,30,0,0), 16, 2},
202 		{IPv4(100,40,0,0), 16, 3},
203 		{IPv4(100,50,0,0), 16, 4},
204 		{IPv4(100,60,0,0), 16, 5},
205 		{IPv4(100,70,0,0), 16, 6},
206 		{IPv4(100,80,0,0), 16, 7},
207 };
208 
209 /*
210  * IPv6 forwarding table
211  */
212 
213 struct l3fwd_ipv6_route {
214 	uint8_t ip[IPV6_ADDR_LEN];
215 	uint8_t depth;
216 	uint8_t if_out;
217 };
218 
219 static struct l3fwd_ipv6_route l3fwd_ipv6_route_array[] = {
220 	{{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
221 	{{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
222 	{{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
223 	{{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
224 	{{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
225 	{{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
226 	{{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
227 	{{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
228 };
229 
230 #define LPM_MAX_RULES         1024
231 #define LPM6_MAX_RULES         1024
232 #define LPM6_NUMBER_TBL8S (1 << 16)
233 
234 struct rte_lpm6_config lpm6_config = {
235 		.max_rules = LPM6_MAX_RULES,
236 		.number_tbl8s = LPM6_NUMBER_TBL8S,
237 		.flags = 0
238 };
239 
240 static struct rte_mempool *socket_direct_pool[RTE_MAX_NUMA_NODES];
241 static struct rte_mempool *socket_indirect_pool[RTE_MAX_NUMA_NODES];
242 static struct rte_lpm *socket_lpm[RTE_MAX_NUMA_NODES];
243 static struct rte_lpm6 *socket_lpm6[RTE_MAX_NUMA_NODES];
244 
245 /* Send burst of packets on an output interface */
246 static inline int
247 send_burst(struct lcore_queue_conf *qconf, uint16_t n, uint8_t port)
248 {
249 	struct rte_mbuf **m_table;
250 	int ret;
251 	uint16_t queueid;
252 
253 	queueid = qconf->tx_queue_id[port];
254 	m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
255 
256 	ret = rte_eth_tx_burst(port, queueid, m_table, n);
257 	if (unlikely(ret < n)) {
258 		do {
259 			rte_pktmbuf_free(m_table[ret]);
260 		} while (++ret < n);
261 	}
262 
263 	return 0;
264 }
265 
266 static inline void
267 l3fwd_simple_forward(struct rte_mbuf *m, struct lcore_queue_conf *qconf,
268 		uint8_t queueid, uint8_t port_in)
269 {
270 	struct rx_queue *rxq;
271 	uint32_t i, len;
272 	uint8_t next_hop, port_out, ipv6;
273 	int32_t len2;
274 
275 	ipv6 = 0;
276 	rxq = &qconf->rx_queue_list[queueid];
277 
278 	/* by default, send everything back to the source port */
279 	port_out = port_in;
280 
281 	/* Remove the Ethernet header and trailer from the input packet */
282 	rte_pktmbuf_adj(m, (uint16_t)sizeof(struct ether_hdr));
283 
284 	/* Build transmission burst */
285 	len = qconf->tx_mbufs[port_out].len;
286 
287 	/* if this is an IPv4 packet */
288 	if (m->ol_flags & PKT_RX_IPV4_HDR) {
289 		struct ipv4_hdr *ip_hdr;
290 		uint32_t ip_dst;
291 		/* Read the lookup key (i.e. ip_dst) from the input packet */
292 		ip_hdr = rte_pktmbuf_mtod(m, struct ipv4_hdr *);
293 		ip_dst = rte_be_to_cpu_32(ip_hdr->dst_addr);
294 
295 		/* Find destination port */
296 		if (rte_lpm_lookup(rxq->lpm, ip_dst, &next_hop) == 0 &&
297 				(enabled_port_mask & 1 << next_hop) != 0) {
298 			port_out = next_hop;
299 
300 			/* Build transmission burst for new port */
301 			len = qconf->tx_mbufs[port_out].len;
302 		}
303 
304 		/* if we don't need to do any fragmentation */
305 		if (likely (IPV4_MTU_DEFAULT >= m->pkt_len)) {
306 			qconf->tx_mbufs[port_out].m_table[len] = m;
307 			len2 = 1;
308 		} else {
309 			len2 = rte_ipv4_fragment_packet(m,
310 				&qconf->tx_mbufs[port_out].m_table[len],
311 				(uint16_t)(MBUF_TABLE_SIZE - len),
312 				IPV4_MTU_DEFAULT,
313 				rxq->direct_pool, rxq->indirect_pool);
314 
315 			/* Free input packet */
316 			rte_pktmbuf_free(m);
317 
318 			/* If we fail to fragment the packet */
319 			if (unlikely (len2 < 0))
320 				return;
321 		}
322 	}
323 	/* if this is an IPv6 packet */
324 	else if (m->ol_flags & PKT_RX_IPV6_HDR) {
325 		struct ipv6_hdr *ip_hdr;
326 
327 		ipv6 = 1;
328 
329 		/* Read the lookup key (i.e. ip_dst) from the input packet */
330 		ip_hdr = rte_pktmbuf_mtod(m, struct ipv6_hdr *);
331 
332 		/* Find destination port */
333 		if (rte_lpm6_lookup(rxq->lpm6, ip_hdr->dst_addr, &next_hop) == 0 &&
334 				(enabled_port_mask & 1 << next_hop) != 0) {
335 			port_out = next_hop;
336 
337 			/* Build transmission burst for new port */
338 			len = qconf->tx_mbufs[port_out].len;
339 		}
340 
341 		/* if we don't need to do any fragmentation */
342 		if (likely (IPV6_MTU_DEFAULT >= m->pkt_len)) {
343 			qconf->tx_mbufs[port_out].m_table[len] = m;
344 			len2 = 1;
345 		} else {
346 			len2 = rte_ipv6_fragment_packet(m,
347 				&qconf->tx_mbufs[port_out].m_table[len],
348 				(uint16_t)(MBUF_TABLE_SIZE - len),
349 				IPV6_MTU_DEFAULT,
350 				rxq->direct_pool, rxq->indirect_pool);
351 
352 			/* Free input packet */
353 			rte_pktmbuf_free(m);
354 
355 			/* If we fail to fragment the packet */
356 			if (unlikely (len2 < 0))
357 				return;
358 		}
359 	}
360 	/* else, just forward the packet */
361 	else {
362 		qconf->tx_mbufs[port_out].m_table[len] = m;
363 		len2 = 1;
364 	}
365 
366 	for (i = len; i < len + len2; i ++) {
367 		void *d_addr_bytes;
368 
369 		m = qconf->tx_mbufs[port_out].m_table[i];
370 		struct ether_hdr *eth_hdr = (struct ether_hdr *)
371 			rte_pktmbuf_prepend(m, (uint16_t)sizeof(struct ether_hdr));
372 		if (eth_hdr == NULL) {
373 			rte_panic("No headroom in mbuf.\n");
374 		}
375 
376 		m->l2_len = sizeof(struct ether_hdr);
377 
378 		/* 02:00:00:00:00:xx */
379 		d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
380 		*((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)port_out << 40);
381 
382 		/* src addr */
383 		ether_addr_copy(&ports_eth_addr[port_out], &eth_hdr->s_addr);
384 		if (ipv6)
385 			eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv6);
386 		else
387 			eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv4);
388 	}
389 
390 	len += len2;
391 
392 	if (likely(len < MAX_PKT_BURST)) {
393 		qconf->tx_mbufs[port_out].len = (uint16_t)len;
394 		return;
395 	}
396 
397 	/* Transmit packets */
398 	send_burst(qconf, (uint16_t)len, port_out);
399 	qconf->tx_mbufs[port_out].len = 0;
400 }
401 
402 /* main processing loop */
403 static int
404 main_loop(__attribute__((unused)) void *dummy)
405 {
406 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
407 	unsigned lcore_id;
408 	uint64_t prev_tsc, diff_tsc, cur_tsc;
409 	int i, j, nb_rx;
410 	uint8_t portid;
411 	struct lcore_queue_conf *qconf;
412 	const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
413 
414 	prev_tsc = 0;
415 
416 	lcore_id = rte_lcore_id();
417 	qconf = &lcore_queue_conf[lcore_id];
418 
419 	if (qconf->n_rx_queue == 0) {
420 		RTE_LOG(INFO, IP_FRAG, "lcore %u has nothing to do\n", lcore_id);
421 		return 0;
422 	}
423 
424 	RTE_LOG(INFO, IP_FRAG, "entering main loop on lcore %u\n", lcore_id);
425 
426 	for (i = 0; i < qconf->n_rx_queue; i++) {
427 
428 		portid = qconf->rx_queue_list[i].portid;
429 		RTE_LOG(INFO, IP_FRAG, " -- lcoreid=%u portid=%d\n", lcore_id,
430 				(int) portid);
431 	}
432 
433 	while (1) {
434 
435 		cur_tsc = rte_rdtsc();
436 
437 		/*
438 		 * TX burst queue drain
439 		 */
440 		diff_tsc = cur_tsc - prev_tsc;
441 		if (unlikely(diff_tsc > drain_tsc)) {
442 
443 			/*
444 			 * This could be optimized (use queueid instead of
445 			 * portid), but it is not called so often
446 			 */
447 			for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
448 				if (qconf->tx_mbufs[portid].len == 0)
449 					continue;
450 				send_burst(&lcore_queue_conf[lcore_id],
451 					   qconf->tx_mbufs[portid].len,
452 					   portid);
453 				qconf->tx_mbufs[portid].len = 0;
454 			}
455 
456 			prev_tsc = cur_tsc;
457 		}
458 
459 		/*
460 		 * Read packet from RX queues
461 		 */
462 		for (i = 0; i < qconf->n_rx_queue; i++) {
463 
464 			portid = qconf->rx_queue_list[i].portid;
465 			nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
466 						 MAX_PKT_BURST);
467 
468 			/* Prefetch first packets */
469 			for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
470 				rte_prefetch0(rte_pktmbuf_mtod(
471 						pkts_burst[j], void *));
472 			}
473 
474 			/* Prefetch and forward already prefetched packets */
475 			for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
476 				rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
477 						j + PREFETCH_OFFSET], void *));
478 				l3fwd_simple_forward(pkts_burst[j], qconf, i, portid);
479 			}
480 
481 			/* Forward remaining prefetched packets */
482 			for (; j < nb_rx; j++) {
483 				l3fwd_simple_forward(pkts_burst[j], qconf, i, portid);
484 			}
485 		}
486 	}
487 }
488 
489 /* display usage */
490 static void
491 print_usage(const char *prgname)
492 {
493 	printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
494 	       "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
495 	       "  -q NQ: number of queue (=ports) per lcore (default is 1)\n",
496 	       prgname);
497 }
498 
499 static int
500 parse_portmask(const char *portmask)
501 {
502 	char *end = NULL;
503 	unsigned long pm;
504 
505 	/* parse hexadecimal string */
506 	pm = strtoul(portmask, &end, 16);
507 	if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
508 		return -1;
509 
510 	if (pm == 0)
511 		return -1;
512 
513 	return pm;
514 }
515 
516 static int
517 parse_nqueue(const char *q_arg)
518 {
519 	char *end = NULL;
520 	unsigned long n;
521 
522 	/* parse hexadecimal string */
523 	n = strtoul(q_arg, &end, 10);
524 	if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
525 		return -1;
526 	if (n == 0)
527 		return -1;
528 	if (n >= MAX_RX_QUEUE_PER_LCORE)
529 		return -1;
530 
531 	return n;
532 }
533 
534 /* Parse the argument given in the command line of the application */
535 static int
536 parse_args(int argc, char **argv)
537 {
538 	int opt, ret;
539 	char **argvopt;
540 	int option_index;
541 	char *prgname = argv[0];
542 	static struct option lgopts[] = {
543 		{NULL, 0, 0, 0}
544 	};
545 
546 	argvopt = argv;
547 
548 	while ((opt = getopt_long(argc, argvopt, "p:q:",
549 				  lgopts, &option_index)) != EOF) {
550 
551 		switch (opt) {
552 		/* portmask */
553 		case 'p':
554 			enabled_port_mask = parse_portmask(optarg);
555 			if (enabled_port_mask < 0) {
556 				printf("invalid portmask\n");
557 				print_usage(prgname);
558 				return -1;
559 			}
560 			break;
561 
562 		/* nqueue */
563 		case 'q':
564 			rx_queue_per_lcore = parse_nqueue(optarg);
565 			if (rx_queue_per_lcore < 0) {
566 				printf("invalid queue number\n");
567 				print_usage(prgname);
568 				return -1;
569 			}
570 			break;
571 
572 		/* long options */
573 		case 0:
574 			print_usage(prgname);
575 			return -1;
576 
577 		default:
578 			print_usage(prgname);
579 			return -1;
580 		}
581 	}
582 
583 	if (enabled_port_mask == 0) {
584 		printf("portmask not specified\n");
585 		print_usage(prgname);
586 		return -1;
587 	}
588 
589 	if (optind >= 0)
590 		argv[optind-1] = prgname;
591 
592 	ret = optind-1;
593 	optind = 0; /* reset getopt lib */
594 	return ret;
595 }
596 
597 static void
598 print_ethaddr(const char *name, struct ether_addr *eth_addr)
599 {
600 	char buf[ETHER_ADDR_FMT_SIZE];
601 	ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
602 	printf("%s%s", name, buf);
603 }
604 
605 /* Check the link status of all ports in up to 9s, and print them finally */
606 static void
607 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
608 {
609 #define CHECK_INTERVAL 100 /* 100ms */
610 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
611 	uint8_t portid, count, all_ports_up, print_flag = 0;
612 	struct rte_eth_link link;
613 
614 	printf("\nChecking link status");
615 	fflush(stdout);
616 	for (count = 0; count <= MAX_CHECK_TIME; count++) {
617 		all_ports_up = 1;
618 		for (portid = 0; portid < port_num; portid++) {
619 			if ((port_mask & (1 << portid)) == 0)
620 				continue;
621 			memset(&link, 0, sizeof(link));
622 			rte_eth_link_get_nowait(portid, &link);
623 			/* print link status if flag set */
624 			if (print_flag == 1) {
625 				if (link.link_status)
626 					printf("Port %d Link Up - speed %u "
627 						"Mbps - %s\n", (uint8_t)portid,
628 						(unsigned)link.link_speed,
629 				(link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
630 					("full-duplex") : ("half-duplex\n"));
631 				else
632 					printf("Port %d Link Down\n",
633 							(uint8_t)portid);
634 				continue;
635 			}
636 			/* clear all_ports_up flag if any link down */
637 			if (link.link_status == 0) {
638 				all_ports_up = 0;
639 				break;
640 			}
641 		}
642 		/* after finally printing all link status, get out */
643 		if (print_flag == 1)
644 			break;
645 
646 		if (all_ports_up == 0) {
647 			printf(".");
648 			fflush(stdout);
649 			rte_delay_ms(CHECK_INTERVAL);
650 		}
651 
652 		/* set the print_flag if all ports up or timeout */
653 		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
654 			print_flag = 1;
655 			printf("\ndone\n");
656 		}
657 	}
658 }
659 
660 static int
661 init_routing_table(void)
662 {
663 	struct rte_lpm *lpm;
664 	struct rte_lpm6 *lpm6;
665 	int socket, ret;
666 	unsigned i;
667 
668 	for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) {
669 		if (socket_lpm[socket]) {
670 			lpm = socket_lpm[socket];
671 			/* populate the LPM table */
672 			for (i = 0; i < RTE_DIM(l3fwd_ipv4_route_array); i++) {
673 				ret = rte_lpm_add(lpm,
674 					l3fwd_ipv4_route_array[i].ip,
675 					l3fwd_ipv4_route_array[i].depth,
676 					l3fwd_ipv4_route_array[i].if_out);
677 
678 				if (ret < 0) {
679 					RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd "
680 						"LPM table\n", i);
681 					return -1;
682 				}
683 
684 				RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv4_BYTES_FMT
685 						"/%d (port %d)\n",
686 					socket,
687 					IPv4_BYTES(l3fwd_ipv4_route_array[i].ip),
688 					l3fwd_ipv4_route_array[i].depth,
689 					l3fwd_ipv4_route_array[i].if_out);
690 			}
691 		}
692 
693 		if (socket_lpm6[socket]) {
694 			lpm6 = socket_lpm6[socket];
695 			/* populate the LPM6 table */
696 			for (i = 0; i < RTE_DIM(l3fwd_ipv6_route_array); i++) {
697 				ret = rte_lpm6_add(lpm6,
698 					l3fwd_ipv6_route_array[i].ip,
699 					l3fwd_ipv6_route_array[i].depth,
700 					l3fwd_ipv6_route_array[i].if_out);
701 
702 				if (ret < 0) {
703 					RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd "
704 						"LPM6 table\n", i);
705 					return -1;
706 				}
707 
708 				RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv6_BYTES_FMT
709 						"/%d (port %d)\n",
710 					socket,
711 					IPv6_BYTES(l3fwd_ipv6_route_array[i].ip),
712 					l3fwd_ipv6_route_array[i].depth,
713 					l3fwd_ipv6_route_array[i].if_out);
714 			}
715 		}
716 	}
717 	return 0;
718 }
719 
720 static int
721 init_mem(void)
722 {
723 	char buf[PATH_MAX];
724 	struct rte_mempool *mp;
725 	struct rte_lpm *lpm;
726 	struct rte_lpm6 *lpm6;
727 	int socket;
728 	unsigned lcore_id;
729 
730 	/* traverse through lcores and initialize structures on each socket */
731 
732 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
733 
734 		if (rte_lcore_is_enabled(lcore_id) == 0)
735 			continue;
736 
737 		socket = rte_lcore_to_socket_id(lcore_id);
738 
739 		if (socket == SOCKET_ID_ANY)
740 			socket = 0;
741 
742 		if (socket_direct_pool[socket] == NULL) {
743 			RTE_LOG(INFO, IP_FRAG, "Creating direct mempool on socket %i\n",
744 					socket);
745 			snprintf(buf, sizeof(buf), "pool_direct_%i", socket);
746 
747 			mp = rte_pktmbuf_pool_create(buf, NB_MBUF, 32,
748 				0, MBUF_DATA_SIZE, socket);
749 			if (mp == NULL) {
750 				RTE_LOG(ERR, IP_FRAG, "Cannot create direct mempool\n");
751 				return -1;
752 			}
753 			socket_direct_pool[socket] = mp;
754 		}
755 
756 		if (socket_indirect_pool[socket] == NULL) {
757 			RTE_LOG(INFO, IP_FRAG, "Creating indirect mempool on socket %i\n",
758 					socket);
759 			snprintf(buf, sizeof(buf), "pool_indirect_%i", socket);
760 
761 			mp = rte_pktmbuf_pool_create(buf, NB_MBUF, 32, 0, 0,
762 				socket);
763 			if (mp == NULL) {
764 				RTE_LOG(ERR, IP_FRAG, "Cannot create indirect mempool\n");
765 				return -1;
766 			}
767 			socket_indirect_pool[socket] = mp;
768 		}
769 
770 		if (socket_lpm[socket] == NULL) {
771 			RTE_LOG(INFO, IP_FRAG, "Creating LPM table on socket %i\n", socket);
772 			snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
773 
774 			lpm = rte_lpm_create(buf, socket, LPM_MAX_RULES, 0);
775 			if (lpm == NULL) {
776 				RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
777 				return -1;
778 			}
779 			socket_lpm[socket] = lpm;
780 		}
781 
782 		if (socket_lpm6[socket] == NULL) {
783 			RTE_LOG(INFO, IP_FRAG, "Creating LPM6 table on socket %i\n", socket);
784 			snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
785 
786 			lpm6 = rte_lpm6_create("IP_FRAG_LPM6", socket, &lpm6_config);
787 			if (lpm6 == NULL) {
788 				RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
789 				return -1;
790 			}
791 			socket_lpm6[socket] = lpm6;
792 		}
793 	}
794 
795 	return 0;
796 }
797 
798 int
799 main(int argc, char **argv)
800 {
801 	struct lcore_queue_conf *qconf;
802 	struct rte_eth_dev_info dev_info;
803 	struct rte_eth_txconf *txconf;
804 	struct rx_queue *rxq;
805 	int socket, ret;
806 	unsigned nb_ports;
807 	uint16_t queueid = 0;
808 	unsigned lcore_id = 0, rx_lcore_id = 0;
809 	uint32_t n_tx_queue, nb_lcores;
810 	uint8_t portid;
811 
812 	/* init EAL */
813 	ret = rte_eal_init(argc, argv);
814 	if (ret < 0)
815 		rte_exit(EXIT_FAILURE, "rte_eal_init failed");
816 	argc -= ret;
817 	argv += ret;
818 
819 	/* parse application arguments (after the EAL ones) */
820 	ret = parse_args(argc, argv);
821 	if (ret < 0)
822 		rte_exit(EXIT_FAILURE, "Invalid arguments");
823 
824 	nb_ports = rte_eth_dev_count();
825 	if (nb_ports > RTE_MAX_ETHPORTS)
826 		nb_ports = RTE_MAX_ETHPORTS;
827 	else if (nb_ports == 0)
828 		rte_exit(EXIT_FAILURE, "No ports found!\n");
829 
830 	nb_lcores = rte_lcore_count();
831 
832 	/* initialize structures (mempools, lpm etc.) */
833 	if (init_mem() < 0)
834 		rte_panic("Cannot initialize memory structures!\n");
835 
836 	/* check if portmask has non-existent ports */
837 	if (enabled_port_mask & ~(RTE_LEN2MASK(nb_ports, unsigned)))
838 		rte_exit(EXIT_FAILURE, "Non-existent ports in portmask!\n");
839 
840 	/* initialize all ports */
841 	for (portid = 0; portid < nb_ports; portid++) {
842 		/* skip ports that are not enabled */
843 		if ((enabled_port_mask & (1 << portid)) == 0) {
844 			printf("Skipping disabled port %d\n", portid);
845 			continue;
846 		}
847 
848 		qconf = &lcore_queue_conf[rx_lcore_id];
849 
850 		/* get the lcore_id for this port */
851 		while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
852 		       qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
853 
854 			rx_lcore_id ++;
855 			if (rx_lcore_id >= RTE_MAX_LCORE)
856 				rte_exit(EXIT_FAILURE, "Not enough cores\n");
857 
858 			qconf = &lcore_queue_conf[rx_lcore_id];
859 		}
860 
861 		socket = (int) rte_lcore_to_socket_id(rx_lcore_id);
862 		if (socket == SOCKET_ID_ANY)
863 			socket = 0;
864 
865 		rxq = &qconf->rx_queue_list[qconf->n_rx_queue];
866 		rxq->portid = portid;
867 		rxq->direct_pool = socket_direct_pool[socket];
868 		rxq->indirect_pool = socket_indirect_pool[socket];
869 		rxq->lpm = socket_lpm[socket];
870 		rxq->lpm6 = socket_lpm6[socket];
871 		qconf->n_rx_queue++;
872 
873 		/* init port */
874 		printf("Initializing port %d on lcore %u...", portid,
875 		       rx_lcore_id);
876 		fflush(stdout);
877 
878 		n_tx_queue = nb_lcores;
879 		if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
880 			n_tx_queue = MAX_TX_QUEUE_PER_PORT;
881 		ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
882 					    &port_conf);
883 		if (ret < 0) {
884 			printf("\n");
885 			rte_exit(EXIT_FAILURE, "Cannot configure device: "
886 				"err=%d, port=%d\n",
887 				ret, portid);
888 		}
889 
890 		/* init one RX queue */
891 		ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
892 					     socket, NULL,
893 					     socket_direct_pool[socket]);
894 		if (ret < 0) {
895 			printf("\n");
896 			rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: "
897 				"err=%d, port=%d\n",
898 				ret, portid);
899 		}
900 
901 		rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
902 		print_ethaddr(" Address:", &ports_eth_addr[portid]);
903 		printf("\n");
904 
905 		/* init one TX queue per couple (lcore,port) */
906 		queueid = 0;
907 		for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
908 			if (rte_lcore_is_enabled(lcore_id) == 0)
909 				continue;
910 
911 			socket = (int) rte_lcore_to_socket_id(lcore_id);
912 			printf("txq=%u,%d ", lcore_id, queueid);
913 			fflush(stdout);
914 
915 			rte_eth_dev_info_get(portid, &dev_info);
916 			txconf = &dev_info.default_txconf;
917 			txconf->txq_flags = 0;
918 			ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
919 						     socket, txconf);
920 			if (ret < 0) {
921 				printf("\n");
922 				rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
923 					"err=%d, port=%d\n", ret, portid);
924 			}
925 
926 			qconf = &lcore_queue_conf[lcore_id];
927 			qconf->tx_queue_id[portid] = queueid;
928 			queueid++;
929 		}
930 
931 		printf("\n");
932 	}
933 
934 	printf("\n");
935 
936 	/* start ports */
937 	for (portid = 0; portid < nb_ports; portid++) {
938 		if ((enabled_port_mask & (1 << portid)) == 0) {
939 			continue;
940 		}
941 		/* Start device */
942 		ret = rte_eth_dev_start(portid);
943 		if (ret < 0)
944 			rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
945 				ret, portid);
946 
947 		rte_eth_promiscuous_enable(portid);
948 	}
949 
950 	if (init_routing_table() < 0)
951 		rte_exit(EXIT_FAILURE, "Cannot init routing table\n");
952 
953 	check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
954 
955 	/* launch per-lcore init on every lcore */
956 	rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
957 	RTE_LCORE_FOREACH_SLAVE(lcore_id) {
958 		if (rte_eal_wait_lcore(lcore_id) < 0)
959 			return -1;
960 	}
961 
962 	return 0;
963 }
964