xref: /dpdk/examples/ip_reassembly/main.c (revision 1cde1b9a9b4dbf31cb5e5ccdfc5da3cb079f43a2)
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 <string.h>
11 #include <sys/queue.h>
12 #include <stdarg.h>
13 #include <errno.h>
14 #include <getopt.h>
15 #include <signal.h>
16 #include <sys/param.h>
17 
18 #include <rte_common.h>
19 #include <rte_byteorder.h>
20 #include <rte_log.h>
21 #include <rte_memory.h>
22 #include <rte_memcpy.h>
23 #include <rte_eal.h>
24 #include <rte_launch.h>
25 #include <rte_atomic.h>
26 #include <rte_cycles.h>
27 #include <rte_prefetch.h>
28 #include <rte_lcore.h>
29 #include <rte_per_lcore.h>
30 #include <rte_branch_prediction.h>
31 #include <rte_interrupts.h>
32 #include <rte_random.h>
33 #include <rte_debug.h>
34 #include <rte_ether.h>
35 #include <rte_ethdev.h>
36 #include <rte_mempool.h>
37 #include <rte_mbuf.h>
38 #include <rte_malloc.h>
39 #include <rte_ip.h>
40 #include <rte_tcp.h>
41 #include <rte_udp.h>
42 #include <rte_string_fns.h>
43 #include <rte_lpm.h>
44 #include <rte_lpm6.h>
45 
46 #include <rte_ip_frag.h>
47 
48 #define MAX_PKT_BURST 32
49 
50 
51 #define RTE_LOGTYPE_IP_RSMBL RTE_LOGTYPE_USER1
52 
53 #define MAX_JUMBO_PKT_LEN  9600
54 
55 #define	BUF_SIZE	RTE_MBUF_DEFAULT_DATAROOM
56 #define	MBUF_DATA_SIZE	RTE_MBUF_DEFAULT_BUF_SIZE
57 
58 #define NB_MBUF 8192
59 #define MEMPOOL_CACHE_SIZE 256
60 
61 /* allow max jumbo frame 9.5 KB */
62 #define JUMBO_FRAME_MAX_SIZE	0x2600
63 
64 #define	MAX_FLOW_NUM	UINT16_MAX
65 #define	MIN_FLOW_NUM	1
66 #define	DEF_FLOW_NUM	0x1000
67 
68 /* TTL numbers are in ms. */
69 #define	MAX_FLOW_TTL	(3600 * MS_PER_S)
70 #define	MIN_FLOW_TTL	1
71 #define	DEF_FLOW_TTL	MS_PER_S
72 
73 #define MAX_FRAG_NUM RTE_LIBRTE_IP_FRAG_MAX_FRAG
74 
75 /* Should be power of two. */
76 #define	IP_FRAG_TBL_BUCKET_ENTRIES	16
77 
78 static uint32_t max_flow_num = DEF_FLOW_NUM;
79 static uint32_t max_flow_ttl = DEF_FLOW_TTL;
80 
81 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
82 
83 #define NB_SOCKETS 8
84 
85 /* Configure how many packets ahead to prefetch, when reading packets */
86 #define PREFETCH_OFFSET	3
87 
88 /*
89  * Configurable number of RX/TX ring descriptors
90  */
91 #define RTE_TEST_RX_DESC_DEFAULT 1024
92 #define RTE_TEST_TX_DESC_DEFAULT 1024
93 
94 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
95 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
96 
97 /* ethernet addresses of ports */
98 static struct rte_ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
99 
100 #ifndef IPv4_BYTES
101 #define IPv4_BYTES_FMT "%" PRIu8 ".%" PRIu8 ".%" PRIu8 ".%" PRIu8
102 #define IPv4_BYTES(addr) \
103 		(uint8_t) (((addr) >> 24) & 0xFF),\
104 		(uint8_t) (((addr) >> 16) & 0xFF),\
105 		(uint8_t) (((addr) >> 8) & 0xFF),\
106 		(uint8_t) ((addr) & 0xFF)
107 #endif
108 
109 #ifndef IPv6_BYTES
110 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
111                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
112 #define IPv6_BYTES(addr) \
113 	addr[0],  addr[1], addr[2],  addr[3], \
114 	addr[4],  addr[5], addr[6],  addr[7], \
115 	addr[8],  addr[9], addr[10], addr[11],\
116 	addr[12], addr[13],addr[14], addr[15]
117 #endif
118 
119 #define IPV6_ADDR_LEN 16
120 
121 /* mask of enabled ports */
122 static uint32_t enabled_port_mask = 0;
123 
124 static int rx_queue_per_lcore = 1;
125 
126 struct mbuf_table {
127 	uint32_t len;
128 	uint32_t head;
129 	uint32_t tail;
130 	struct rte_mbuf *m_table[0];
131 };
132 
133 struct rx_queue {
134 	struct rte_ip_frag_tbl *frag_tbl;
135 	struct rte_mempool *pool;
136 	struct rte_lpm *lpm;
137 	struct rte_lpm6 *lpm6;
138 	uint16_t portid;
139 };
140 
141 struct tx_lcore_stat {
142 	uint64_t call;
143 	uint64_t drop;
144 	uint64_t queue;
145 	uint64_t send;
146 };
147 
148 #define MAX_RX_QUEUE_PER_LCORE 16
149 #define MAX_TX_QUEUE_PER_PORT 16
150 #define MAX_RX_QUEUE_PER_PORT 128
151 
152 struct lcore_queue_conf {
153 	uint16_t n_rx_queue;
154 	struct rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
155 	uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
156 	struct rte_ip_frag_death_row death_row;
157 	struct mbuf_table *tx_mbufs[RTE_MAX_ETHPORTS];
158 	struct tx_lcore_stat tx_stat;
159 } __rte_cache_aligned;
160 static struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
161 
162 static struct rte_eth_conf port_conf = {
163 	.rxmode = {
164 		.mq_mode        = ETH_MQ_RX_RSS,
165 		.max_rx_pkt_len = JUMBO_FRAME_MAX_SIZE,
166 		.split_hdr_size = 0,
167 		.offloads = (DEV_RX_OFFLOAD_CHECKSUM |
168 			     DEV_RX_OFFLOAD_JUMBO_FRAME),
169 	},
170 	.rx_adv_conf = {
171 			.rss_conf = {
172 				.rss_key = NULL,
173 				.rss_hf = ETH_RSS_IP,
174 		},
175 	},
176 	.txmode = {
177 		.mq_mode = ETH_MQ_TX_NONE,
178 		.offloads = (DEV_TX_OFFLOAD_IPV4_CKSUM |
179 			     DEV_TX_OFFLOAD_MULTI_SEGS),
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 		{RTE_IPV4(100,10,0,0), 16, 0},
194 		{RTE_IPV4(100,20,0,0), 16, 1},
195 		{RTE_IPV4(100,30,0,0), 16, 2},
196 		{RTE_IPV4(100,40,0,0), 16, 3},
197 		{RTE_IPV4(100,50,0,0), 16, 4},
198 		{RTE_IPV4(100,60,0,0), 16, 5},
199 		{RTE_IPV4(100,70,0,0), 16, 6},
200 		{RTE_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_lpm *socket_lpm[RTE_MAX_NUMA_NODES];
235 static struct rte_lpm6 *socket_lpm6[RTE_MAX_NUMA_NODES];
236 
237 #ifdef RTE_LIBRTE_IP_FRAG_TBL_STAT
238 #define TX_LCORE_STAT_UPDATE(s, f, v)   ((s)->f += (v))
239 #else
240 #define TX_LCORE_STAT_UPDATE(s, f, v)   do {} while (0)
241 #endif /* RTE_LIBRTE_IP_FRAG_TBL_STAT */
242 
243 /*
244  * If number of queued packets reached given threahold, then
245  * send burst of packets on an output interface.
246  */
247 static inline uint32_t
248 send_burst(struct lcore_queue_conf *qconf, uint32_t thresh, uint16_t port)
249 {
250 	uint32_t fill, len, k, n;
251 	struct mbuf_table *txmb;
252 
253 	txmb = qconf->tx_mbufs[port];
254 	len = txmb->len;
255 
256 	if ((int32_t)(fill = txmb->head - txmb->tail) < 0)
257 		fill += len;
258 
259 	if (fill >= thresh) {
260 		n = RTE_MIN(len - txmb->tail, fill);
261 
262 		k = rte_eth_tx_burst(port, qconf->tx_queue_id[port],
263 			txmb->m_table + txmb->tail, (uint16_t)n);
264 
265 		TX_LCORE_STAT_UPDATE(&qconf->tx_stat, call, 1);
266 		TX_LCORE_STAT_UPDATE(&qconf->tx_stat, send, k);
267 
268 		fill -= k;
269 		if ((txmb->tail += k) == len)
270 			txmb->tail = 0;
271 	}
272 
273 	return fill;
274 }
275 
276 /* Enqueue a single packet, and send burst if queue is filled */
277 static inline int
278 send_single_packet(struct rte_mbuf *m, uint16_t port)
279 {
280 	uint32_t fill, lcore_id, len;
281 	struct lcore_queue_conf *qconf;
282 	struct mbuf_table *txmb;
283 
284 	lcore_id = rte_lcore_id();
285 	qconf = &lcore_queue_conf[lcore_id];
286 
287 	txmb = qconf->tx_mbufs[port];
288 	len = txmb->len;
289 
290 	fill = send_burst(qconf, MAX_PKT_BURST, port);
291 
292 	if (fill == len - 1) {
293 		TX_LCORE_STAT_UPDATE(&qconf->tx_stat, drop, 1);
294 		rte_pktmbuf_free(txmb->m_table[txmb->tail]);
295 		if (++txmb->tail == len)
296 			txmb->tail = 0;
297 	}
298 
299 	TX_LCORE_STAT_UPDATE(&qconf->tx_stat, queue, 1);
300 	txmb->m_table[txmb->head] = m;
301 	if(++txmb->head == len)
302 		txmb->head = 0;
303 
304 	return 0;
305 }
306 
307 static inline void
308 reassemble(struct rte_mbuf *m, uint16_t portid, uint32_t queue,
309 	struct lcore_queue_conf *qconf, uint64_t tms)
310 {
311 	struct rte_ether_hdr *eth_hdr;
312 	struct rte_ip_frag_tbl *tbl;
313 	struct rte_ip_frag_death_row *dr;
314 	struct rx_queue *rxq;
315 	void *d_addr_bytes;
316 	uint32_t next_hop;
317 	uint16_t dst_port;
318 
319 	rxq = &qconf->rx_queue_list[queue];
320 
321 	eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
322 
323 	dst_port = portid;
324 
325 	/* if packet is IPv4 */
326 	if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
327 		struct rte_ipv4_hdr *ip_hdr;
328 		uint32_t ip_dst;
329 
330 		ip_hdr = (struct rte_ipv4_hdr *)(eth_hdr + 1);
331 
332 		 /* if it is a fragmented packet, then try to reassemble. */
333 		if (rte_ipv4_frag_pkt_is_fragmented(ip_hdr)) {
334 			struct rte_mbuf *mo;
335 
336 			tbl = rxq->frag_tbl;
337 			dr = &qconf->death_row;
338 
339 			/* prepare mbuf: setup l2_len/l3_len. */
340 			m->l2_len = sizeof(*eth_hdr);
341 			m->l3_len = sizeof(*ip_hdr);
342 
343 			/* process this fragment. */
344 			mo = rte_ipv4_frag_reassemble_packet(tbl, dr, m, tms, ip_hdr);
345 			if (mo == NULL)
346 				/* no packet to send out. */
347 				return;
348 
349 			/* we have our packet reassembled. */
350 			if (mo != m) {
351 				m = mo;
352 				eth_hdr = rte_pktmbuf_mtod(m,
353 					struct rte_ether_hdr *);
354 				ip_hdr = (struct rte_ipv4_hdr *)(eth_hdr + 1);
355 			}
356 
357 			/* update offloading flags */
358 			m->ol_flags |= (PKT_TX_IPV4 | PKT_TX_IP_CKSUM);
359 		}
360 		ip_dst = rte_be_to_cpu_32(ip_hdr->dst_addr);
361 
362 		/* Find destination port */
363 		if (rte_lpm_lookup(rxq->lpm, ip_dst, &next_hop) == 0 &&
364 				(enabled_port_mask & 1 << next_hop) != 0) {
365 			dst_port = next_hop;
366 		}
367 
368 		eth_hdr->ether_type = rte_be_to_cpu_16(RTE_ETHER_TYPE_IPV4);
369 	} else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
370 		/* if packet is IPv6 */
371 		struct ipv6_extension_fragment *frag_hdr;
372 		struct rte_ipv6_hdr *ip_hdr;
373 
374 		ip_hdr = (struct rte_ipv6_hdr *)(eth_hdr + 1);
375 
376 		frag_hdr = rte_ipv6_frag_get_ipv6_fragment_header(ip_hdr);
377 
378 		if (frag_hdr != NULL) {
379 			struct rte_mbuf *mo;
380 
381 			tbl = rxq->frag_tbl;
382 			dr  = &qconf->death_row;
383 
384 			/* prepare mbuf: setup l2_len/l3_len. */
385 			m->l2_len = sizeof(*eth_hdr);
386 			m->l3_len = sizeof(*ip_hdr) + sizeof(*frag_hdr);
387 
388 			mo = rte_ipv6_frag_reassemble_packet(tbl, dr, m, tms, ip_hdr, frag_hdr);
389 			if (mo == NULL)
390 				return;
391 
392 			if (mo != m) {
393 				m = mo;
394 				eth_hdr = rte_pktmbuf_mtod(m,
395 							struct rte_ether_hdr *);
396 				ip_hdr = (struct rte_ipv6_hdr *)(eth_hdr + 1);
397 			}
398 		}
399 
400 		/* Find destination port */
401 		if (rte_lpm6_lookup(rxq->lpm6, ip_hdr->dst_addr,
402 						&next_hop) == 0 &&
403 				(enabled_port_mask & 1 << next_hop) != 0) {
404 			dst_port = next_hop;
405 		}
406 
407 		eth_hdr->ether_type = rte_be_to_cpu_16(RTE_ETHER_TYPE_IPV6);
408 	}
409 	/* if packet wasn't IPv4 or IPv6, it's forwarded to the port it came from */
410 
411 	/* 02:00:00:00:00:xx */
412 	d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
413 	*((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)dst_port << 40);
414 
415 	/* src addr */
416 	rte_ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
417 
418 	send_single_packet(m, dst_port);
419 }
420 
421 /* main processing loop */
422 static int
423 main_loop(__attribute__((unused)) void *dummy)
424 {
425 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
426 	unsigned lcore_id;
427 	uint64_t diff_tsc, cur_tsc, prev_tsc;
428 	int i, j, nb_rx;
429 	uint16_t portid;
430 	struct lcore_queue_conf *qconf;
431 	const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
432 
433 	prev_tsc = 0;
434 
435 	lcore_id = rte_lcore_id();
436 	qconf = &lcore_queue_conf[lcore_id];
437 
438 	if (qconf->n_rx_queue == 0) {
439 		RTE_LOG(INFO, IP_RSMBL, "lcore %u has nothing to do\n", lcore_id);
440 		return 0;
441 	}
442 
443 	RTE_LOG(INFO, IP_RSMBL, "entering main loop on lcore %u\n", lcore_id);
444 
445 	for (i = 0; i < qconf->n_rx_queue; i++) {
446 
447 		portid = qconf->rx_queue_list[i].portid;
448 		RTE_LOG(INFO, IP_RSMBL, " -- lcoreid=%u portid=%u\n", lcore_id,
449 			portid);
450 	}
451 
452 	while (1) {
453 
454 		cur_tsc = rte_rdtsc();
455 
456 		/*
457 		 * TX burst queue drain
458 		 */
459 		diff_tsc = cur_tsc - prev_tsc;
460 		if (unlikely(diff_tsc > drain_tsc)) {
461 
462 			/*
463 			 * This could be optimized (use queueid instead of
464 			 * portid), but it is not called so often
465 			 */
466 			for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
467 				if ((enabled_port_mask & (1 << portid)) != 0)
468 					send_burst(qconf, 1, portid);
469 			}
470 
471 			prev_tsc = cur_tsc;
472 		}
473 
474 		/*
475 		 * Read packet from RX queues
476 		 */
477 		for (i = 0; i < qconf->n_rx_queue; ++i) {
478 
479 			portid = qconf->rx_queue_list[i].portid;
480 
481 			nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
482 				MAX_PKT_BURST);
483 
484 			/* Prefetch first packets */
485 			for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
486 				rte_prefetch0(rte_pktmbuf_mtod(
487 						pkts_burst[j], void *));
488 			}
489 
490 			/* Prefetch and forward already prefetched packets */
491 			for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
492 				rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
493 					j + PREFETCH_OFFSET], void *));
494 				reassemble(pkts_burst[j], portid,
495 					i, qconf, cur_tsc);
496 			}
497 
498 			/* Forward remaining prefetched packets */
499 			for (; j < nb_rx; j++) {
500 				reassemble(pkts_burst[j], portid,
501 					i, qconf, cur_tsc);
502 			}
503 
504 			rte_ip_frag_free_death_row(&qconf->death_row,
505 				PREFETCH_OFFSET);
506 		}
507 	}
508 }
509 
510 /* display usage */
511 static void
512 print_usage(const char *prgname)
513 {
514 	printf("%s [EAL options] -- -p PORTMASK [-q NQ]"
515 		"  [--max-pkt-len PKTLEN]"
516 		"  [--maxflows=<flows>]  [--flowttl=<ttl>[(s|ms)]]\n"
517 		"  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
518 		"  -q NQ: number of RX queues per lcore\n"
519 		"  --maxflows=<flows>: optional, maximum number of flows "
520 		"supported\n"
521 		"  --flowttl=<ttl>[(s|ms)]: optional, maximum TTL for each "
522 		"flow\n",
523 		prgname);
524 }
525 
526 static uint32_t
527 parse_flow_num(const char *str, uint32_t min, uint32_t max, uint32_t *val)
528 {
529 	char *end;
530 	uint64_t v;
531 
532 	/* parse decimal string */
533 	errno = 0;
534 	v = strtoul(str, &end, 10);
535 	if (errno != 0 || *end != '\0')
536 		return -EINVAL;
537 
538 	if (v < min || v > max)
539 		return -EINVAL;
540 
541 	*val = (uint32_t)v;
542 	return 0;
543 }
544 
545 static int
546 parse_flow_ttl(const char *str, uint32_t min, uint32_t max, uint32_t *val)
547 {
548 	char *end;
549 	uint64_t v;
550 
551 	static const char frmt_sec[] = "s";
552 	static const char frmt_msec[] = "ms";
553 
554 	/* parse decimal string */
555 	errno = 0;
556 	v = strtoul(str, &end, 10);
557 	if (errno != 0)
558 		return -EINVAL;
559 
560 	if (*end != '\0') {
561 		if (strncmp(frmt_sec, end, sizeof(frmt_sec)) == 0)
562 			v *= MS_PER_S;
563 		else if (strncmp(frmt_msec, end, sizeof (frmt_msec)) != 0)
564 			return -EINVAL;
565 	}
566 
567 	if (v < min || v > max)
568 		return -EINVAL;
569 
570 	*val = (uint32_t)v;
571 	return 0;
572 }
573 
574 static int
575 parse_portmask(const char *portmask)
576 {
577 	char *end = NULL;
578 	unsigned long pm;
579 
580 	/* parse hexadecimal string */
581 	pm = strtoul(portmask, &end, 16);
582 	if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
583 		return -1;
584 
585 	if (pm == 0)
586 		return -1;
587 
588 	return pm;
589 }
590 
591 static int
592 parse_nqueue(const char *q_arg)
593 {
594 	char *end = NULL;
595 	unsigned long n;
596 
597 	printf("%p\n", q_arg);
598 
599 	/* parse hexadecimal string */
600 	n = strtoul(q_arg, &end, 10);
601 	if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
602 		return -1;
603 	if (n == 0)
604 		return -1;
605 	if (n >= MAX_RX_QUEUE_PER_LCORE)
606 		return -1;
607 
608 	return n;
609 }
610 
611 /* Parse the argument given in the command line of the application */
612 static int
613 parse_args(int argc, char **argv)
614 {
615 	int opt, ret;
616 	char **argvopt;
617 	int option_index;
618 	char *prgname = argv[0];
619 	static struct option lgopts[] = {
620 		{"max-pkt-len", 1, 0, 0},
621 		{"maxflows", 1, 0, 0},
622 		{"flowttl", 1, 0, 0},
623 		{NULL, 0, 0, 0}
624 	};
625 
626 	argvopt = argv;
627 
628 	while ((opt = getopt_long(argc, argvopt, "p:q:",
629 				lgopts, &option_index)) != EOF) {
630 
631 		switch (opt) {
632 		/* portmask */
633 		case 'p':
634 			enabled_port_mask = parse_portmask(optarg);
635 			if (enabled_port_mask == 0) {
636 				printf("invalid portmask\n");
637 				print_usage(prgname);
638 				return -1;
639 			}
640 			break;
641 
642 		/* nqueue */
643 		case 'q':
644 			rx_queue_per_lcore = parse_nqueue(optarg);
645 			if (rx_queue_per_lcore < 0) {
646 				printf("invalid queue number\n");
647 				print_usage(prgname);
648 				return -1;
649 			}
650 			break;
651 
652 		/* long options */
653 		case 0:
654 			if (!strncmp(lgopts[option_index].name,
655 					"maxflows", 8)) {
656 				if ((ret = parse_flow_num(optarg, MIN_FLOW_NUM,
657 						MAX_FLOW_NUM,
658 						&max_flow_num)) != 0) {
659 					printf("invalid value: \"%s\" for "
660 						"parameter %s\n",
661 						optarg,
662 						lgopts[option_index].name);
663 					print_usage(prgname);
664 					return ret;
665 				}
666 			}
667 
668 			if (!strncmp(lgopts[option_index].name, "flowttl", 7)) {
669 				if ((ret = parse_flow_ttl(optarg, MIN_FLOW_TTL,
670 						MAX_FLOW_TTL,
671 						&max_flow_ttl)) != 0) {
672 					printf("invalid value: \"%s\" for "
673 						"parameter %s\n",
674 						optarg,
675 						lgopts[option_index].name);
676 					print_usage(prgname);
677 					return ret;
678 				}
679 			}
680 
681 			break;
682 
683 		default:
684 			print_usage(prgname);
685 			return -1;
686 		}
687 	}
688 
689 	if (optind >= 0)
690 		argv[optind-1] = prgname;
691 
692 	ret = optind-1;
693 	optind = 1; /* reset getopt lib */
694 	return ret;
695 }
696 
697 static void
698 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
699 {
700 	char buf[RTE_ETHER_ADDR_FMT_SIZE];
701 	rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
702 	printf("%s%s", name, buf);
703 }
704 
705 /* Check the link status of all ports in up to 9s, and print them finally */
706 static void
707 check_all_ports_link_status(uint32_t port_mask)
708 {
709 #define CHECK_INTERVAL 100 /* 100ms */
710 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
711 	uint16_t portid;
712 	uint8_t count, all_ports_up, print_flag = 0;
713 	struct rte_eth_link link;
714 	int ret;
715 
716 	printf("\nChecking link status");
717 	fflush(stdout);
718 	for (count = 0; count <= MAX_CHECK_TIME; count++) {
719 		all_ports_up = 1;
720 		RTE_ETH_FOREACH_DEV(portid) {
721 			if ((port_mask & (1 << portid)) == 0)
722 				continue;
723 			memset(&link, 0, sizeof(link));
724 			ret = rte_eth_link_get_nowait(portid, &link);
725 			if (ret < 0) {
726 				all_ports_up = 0;
727 				if (print_flag == 1)
728 					printf("Port %u link get failed: %s\n",
729 						portid, rte_strerror(-ret));
730 				continue;
731 			}
732 			/* print link status if flag set */
733 			if (print_flag == 1) {
734 				if (link.link_status)
735 					printf(
736 					"Port%d Link Up. Speed %u Mbps - %s\n",
737 						portid, link.link_speed,
738 				(link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
739 					("full-duplex") : ("half-duplex\n"));
740 				else
741 					printf("Port %d Link Down\n", portid);
742 				continue;
743 			}
744 			/* clear all_ports_up flag if any link down */
745 			if (link.link_status == ETH_LINK_DOWN) {
746 				all_ports_up = 0;
747 				break;
748 			}
749 		}
750 		/* after finally printing all link status, get out */
751 		if (print_flag == 1)
752 			break;
753 
754 		if (all_ports_up == 0) {
755 			printf(".");
756 			fflush(stdout);
757 			rte_delay_ms(CHECK_INTERVAL);
758 		}
759 
760 		/* set the print_flag if all ports up or timeout */
761 		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
762 			print_flag = 1;
763 			printf("\ndone\n");
764 		}
765 	}
766 }
767 
768 static int
769 init_routing_table(void)
770 {
771 	struct rte_lpm *lpm;
772 	struct rte_lpm6 *lpm6;
773 	int socket, ret;
774 	unsigned i;
775 
776 	for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) {
777 		if (socket_lpm[socket]) {
778 			lpm = socket_lpm[socket];
779 			/* populate the LPM table */
780 			for (i = 0; i < RTE_DIM(l3fwd_ipv4_route_array); i++) {
781 				ret = rte_lpm_add(lpm,
782 					l3fwd_ipv4_route_array[i].ip,
783 					l3fwd_ipv4_route_array[i].depth,
784 					l3fwd_ipv4_route_array[i].if_out);
785 
786 				if (ret < 0) {
787 					RTE_LOG(ERR, IP_RSMBL, "Unable to add entry %i to the l3fwd "
788 						"LPM table\n", i);
789 					return -1;
790 				}
791 
792 				RTE_LOG(INFO, IP_RSMBL, "Socket %i: adding route " IPv4_BYTES_FMT
793 						"/%d (port %d)\n",
794 					socket,
795 					IPv4_BYTES(l3fwd_ipv4_route_array[i].ip),
796 					l3fwd_ipv4_route_array[i].depth,
797 					l3fwd_ipv4_route_array[i].if_out);
798 			}
799 		}
800 
801 		if (socket_lpm6[socket]) {
802 			lpm6 = socket_lpm6[socket];
803 			/* populate the LPM6 table */
804 			for (i = 0; i < RTE_DIM(l3fwd_ipv6_route_array); i++) {
805 				ret = rte_lpm6_add(lpm6,
806 					l3fwd_ipv6_route_array[i].ip,
807 					l3fwd_ipv6_route_array[i].depth,
808 					l3fwd_ipv6_route_array[i].if_out);
809 
810 				if (ret < 0) {
811 					RTE_LOG(ERR, IP_RSMBL, "Unable to add entry %i to the l3fwd "
812 						"LPM6 table\n", i);
813 					return -1;
814 				}
815 
816 				RTE_LOG(INFO, IP_RSMBL, "Socket %i: adding route " IPv6_BYTES_FMT
817 						"/%d (port %d)\n",
818 					socket,
819 					IPv6_BYTES(l3fwd_ipv6_route_array[i].ip),
820 					l3fwd_ipv6_route_array[i].depth,
821 					l3fwd_ipv6_route_array[i].if_out);
822 			}
823 		}
824 	}
825 	return 0;
826 }
827 
828 static int
829 setup_port_tbl(struct lcore_queue_conf *qconf, uint32_t lcore, int socket,
830 	uint32_t port)
831 {
832 	struct mbuf_table *mtb;
833 	uint32_t n;
834 	size_t sz;
835 
836 	n = RTE_MAX(max_flow_num, 2UL * MAX_PKT_BURST);
837 	sz = sizeof (*mtb) + sizeof (mtb->m_table[0]) *  n;
838 
839 	if ((mtb = rte_zmalloc_socket(__func__, sz, RTE_CACHE_LINE_SIZE,
840 			socket)) == NULL) {
841 		RTE_LOG(ERR, IP_RSMBL, "%s() for lcore: %u, port: %u "
842 			"failed to allocate %zu bytes\n",
843 			__func__, lcore, port, sz);
844 		return -1;
845 	}
846 
847 	mtb->len = n;
848 	qconf->tx_mbufs[port] = mtb;
849 
850 	return 0;
851 }
852 
853 static int
854 setup_queue_tbl(struct rx_queue *rxq, uint32_t lcore, uint32_t queue)
855 {
856 	int socket;
857 	uint32_t nb_mbuf;
858 	uint64_t frag_cycles;
859 	char buf[RTE_MEMPOOL_NAMESIZE];
860 
861 	socket = rte_lcore_to_socket_id(lcore);
862 	if (socket == SOCKET_ID_ANY)
863 		socket = 0;
864 
865 	frag_cycles = (rte_get_tsc_hz() + MS_PER_S - 1) / MS_PER_S *
866 		max_flow_ttl;
867 
868 	if ((rxq->frag_tbl = rte_ip_frag_table_create(max_flow_num,
869 			IP_FRAG_TBL_BUCKET_ENTRIES, max_flow_num, frag_cycles,
870 			socket)) == NULL) {
871 		RTE_LOG(ERR, IP_RSMBL, "ip_frag_tbl_create(%u) on "
872 			"lcore: %u for queue: %u failed\n",
873 			max_flow_num, lcore, queue);
874 		return -1;
875 	}
876 
877 	/*
878 	 * At any given moment up to <max_flow_num * (MAX_FRAG_NUM)>
879 	 * mbufs could be stored int the fragment table.
880 	 * Plus, each TX queue can hold up to <max_flow_num> packets.
881 	 */
882 
883 	nb_mbuf = RTE_MAX(max_flow_num, 2UL * MAX_PKT_BURST) * MAX_FRAG_NUM;
884 	nb_mbuf *= (port_conf.rxmode.max_rx_pkt_len + BUF_SIZE - 1) / BUF_SIZE;
885 	nb_mbuf *= 2; /* ipv4 and ipv6 */
886 	nb_mbuf += nb_rxd + nb_txd;
887 
888 	nb_mbuf = RTE_MAX(nb_mbuf, (uint32_t)NB_MBUF);
889 
890 	snprintf(buf, sizeof(buf), "mbuf_pool_%u_%u", lcore, queue);
891 
892 	rxq->pool = rte_pktmbuf_pool_create(buf, nb_mbuf, MEMPOOL_CACHE_SIZE, 0,
893 					    MBUF_DATA_SIZE, socket);
894 	if (rxq->pool == NULL) {
895 		RTE_LOG(ERR, IP_RSMBL,
896 			"rte_pktmbuf_pool_create(%s) failed", buf);
897 		return -1;
898 	}
899 
900 	return 0;
901 }
902 
903 static int
904 init_mem(void)
905 {
906 	char buf[PATH_MAX];
907 	struct rte_lpm *lpm;
908 	struct rte_lpm6 *lpm6;
909 	struct rte_lpm_config lpm_config;
910 	int socket;
911 	unsigned lcore_id;
912 
913 	/* traverse through lcores and initialize structures on each socket */
914 
915 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
916 
917 		if (rte_lcore_is_enabled(lcore_id) == 0)
918 			continue;
919 
920 		socket = rte_lcore_to_socket_id(lcore_id);
921 
922 		if (socket == SOCKET_ID_ANY)
923 			socket = 0;
924 
925 		if (socket_lpm[socket] == NULL) {
926 			RTE_LOG(INFO, IP_RSMBL, "Creating LPM table on socket %i\n", socket);
927 			snprintf(buf, sizeof(buf), "IP_RSMBL_LPM_%i", socket);
928 
929 			lpm_config.max_rules = LPM_MAX_RULES;
930 			lpm_config.number_tbl8s = 256;
931 			lpm_config.flags = 0;
932 
933 			lpm = rte_lpm_create(buf, socket, &lpm_config);
934 			if (lpm == NULL) {
935 				RTE_LOG(ERR, IP_RSMBL, "Cannot create LPM table\n");
936 				return -1;
937 			}
938 			socket_lpm[socket] = lpm;
939 		}
940 
941 		if (socket_lpm6[socket] == NULL) {
942 			RTE_LOG(INFO, IP_RSMBL, "Creating LPM6 table on socket %i\n", socket);
943 			snprintf(buf, sizeof(buf), "IP_RSMBL_LPM_%i", socket);
944 
945 			lpm6 = rte_lpm6_create(buf, socket, &lpm6_config);
946 			if (lpm6 == NULL) {
947 				RTE_LOG(ERR, IP_RSMBL, "Cannot create LPM table\n");
948 				return -1;
949 			}
950 			socket_lpm6[socket] = lpm6;
951 		}
952 	}
953 
954 	return 0;
955 }
956 
957 static void
958 queue_dump_stat(void)
959 {
960 	uint32_t i, lcore;
961 	const struct lcore_queue_conf *qconf;
962 
963 	for (lcore = 0; lcore < RTE_MAX_LCORE; lcore++) {
964 		if (rte_lcore_is_enabled(lcore) == 0)
965 			continue;
966 
967 		qconf = &lcore_queue_conf[lcore];
968 		for (i = 0; i < qconf->n_rx_queue; i++) {
969 
970 			fprintf(stdout, " -- lcoreid=%u portid=%u "
971 				"frag tbl stat:\n",
972 				lcore,  qconf->rx_queue_list[i].portid);
973 			rte_ip_frag_table_statistics_dump(stdout,
974 					qconf->rx_queue_list[i].frag_tbl);
975 			fprintf(stdout, "TX bursts:\t%" PRIu64 "\n"
976 				"TX packets _queued:\t%" PRIu64 "\n"
977 				"TX packets dropped:\t%" PRIu64 "\n"
978 				"TX packets send:\t%" PRIu64 "\n",
979 				qconf->tx_stat.call,
980 				qconf->tx_stat.queue,
981 				qconf->tx_stat.drop,
982 				qconf->tx_stat.send);
983 		}
984 	}
985 }
986 
987 static void
988 signal_handler(int signum)
989 {
990 	queue_dump_stat();
991 	if (signum != SIGUSR1)
992 		rte_exit(0, "received signal: %d, exiting\n", signum);
993 }
994 
995 int
996 main(int argc, char **argv)
997 {
998 	struct lcore_queue_conf *qconf;
999 	struct rte_eth_dev_info dev_info;
1000 	struct rte_eth_txconf *txconf;
1001 	struct rx_queue *rxq;
1002 	int ret, socket;
1003 	unsigned nb_ports;
1004 	uint16_t queueid;
1005 	unsigned lcore_id = 0, rx_lcore_id = 0;
1006 	uint32_t n_tx_queue, nb_lcores;
1007 	uint16_t portid;
1008 
1009 	/* init EAL */
1010 	ret = rte_eal_init(argc, argv);
1011 	if (ret < 0)
1012 		rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1013 	argc -= ret;
1014 	argv += ret;
1015 
1016 	/* parse application arguments (after the EAL ones) */
1017 	ret = parse_args(argc, argv);
1018 	if (ret < 0)
1019 		rte_exit(EXIT_FAILURE, "Invalid IP reassembly parameters\n");
1020 
1021 	nb_ports = rte_eth_dev_count_avail();
1022 	if (nb_ports == 0)
1023 		rte_exit(EXIT_FAILURE, "No ports found!\n");
1024 
1025 	nb_lcores = rte_lcore_count();
1026 
1027 	/* initialize structures (mempools, lpm etc.) */
1028 	if (init_mem() < 0)
1029 		rte_panic("Cannot initialize memory structures!\n");
1030 
1031 	/* check if portmask has non-existent ports */
1032 	if (enabled_port_mask & ~(RTE_LEN2MASK(nb_ports, unsigned)))
1033 		rte_exit(EXIT_FAILURE, "Non-existent ports in portmask!\n");
1034 
1035 	/* initialize all ports */
1036 	RTE_ETH_FOREACH_DEV(portid) {
1037 		struct rte_eth_rxconf rxq_conf;
1038 		struct rte_eth_conf local_port_conf = port_conf;
1039 
1040 		/* skip ports that are not enabled */
1041 		if ((enabled_port_mask & (1 << portid)) == 0) {
1042 			printf("\nSkipping disabled port %d\n", portid);
1043 			continue;
1044 		}
1045 
1046 		qconf = &lcore_queue_conf[rx_lcore_id];
1047 
1048 		/* limit the frame size to the maximum supported by NIC */
1049 		ret = rte_eth_dev_info_get(portid, &dev_info);
1050 		if (ret != 0)
1051 			rte_exit(EXIT_FAILURE,
1052 				"Error during getting device (port %u) info: %s\n",
1053 				portid, strerror(-ret));
1054 
1055 		local_port_conf.rxmode.max_rx_pkt_len = RTE_MIN(
1056 		    dev_info.max_rx_pktlen,
1057 		    local_port_conf.rxmode.max_rx_pkt_len);
1058 
1059 		/* get the lcore_id for this port */
1060 		while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
1061 			   qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
1062 
1063 			rx_lcore_id++;
1064 			if (rx_lcore_id >= RTE_MAX_LCORE)
1065 				rte_exit(EXIT_FAILURE, "Not enough cores\n");
1066 
1067 			qconf = &lcore_queue_conf[rx_lcore_id];
1068 		}
1069 
1070 		socket = rte_lcore_to_socket_id(portid);
1071 		if (socket == SOCKET_ID_ANY)
1072 			socket = 0;
1073 
1074 		queueid = qconf->n_rx_queue;
1075 		rxq = &qconf->rx_queue_list[queueid];
1076 		rxq->portid = portid;
1077 		rxq->lpm = socket_lpm[socket];
1078 		rxq->lpm6 = socket_lpm6[socket];
1079 
1080 		ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
1081 						       &nb_txd);
1082 		if (ret < 0)
1083 			rte_exit(EXIT_FAILURE,
1084 				 "Cannot adjust number of descriptors: err=%d, port=%d\n",
1085 				 ret, portid);
1086 
1087 		if (setup_queue_tbl(rxq, rx_lcore_id, queueid) < 0)
1088 			rte_exit(EXIT_FAILURE, "Failed to set up queue table\n");
1089 		qconf->n_rx_queue++;
1090 
1091 		/* init port */
1092 		printf("Initializing port %d ... ", portid );
1093 		fflush(stdout);
1094 
1095 		n_tx_queue = nb_lcores;
1096 		if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
1097 			n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1098 		if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
1099 			local_port_conf.txmode.offloads |=
1100 				DEV_TX_OFFLOAD_MBUF_FAST_FREE;
1101 
1102 		local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
1103 			dev_info.flow_type_rss_offloads;
1104 		if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
1105 				port_conf.rx_adv_conf.rss_conf.rss_hf) {
1106 			printf("Port %u modified RSS hash function based on hardware support,"
1107 				"requested:%#"PRIx64" configured:%#"PRIx64"\n",
1108 				portid,
1109 				port_conf.rx_adv_conf.rss_conf.rss_hf,
1110 				local_port_conf.rx_adv_conf.rss_conf.rss_hf);
1111 		}
1112 
1113 		ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
1114 					    &local_port_conf);
1115 		if (ret < 0) {
1116 			printf("\n");
1117 			rte_exit(EXIT_FAILURE, "Cannot configure device: "
1118 				"err=%d, port=%d\n",
1119 				ret, portid);
1120 		}
1121 
1122 		/* init one RX queue */
1123 		rxq_conf = dev_info.default_rxconf;
1124 		rxq_conf.offloads = local_port_conf.rxmode.offloads;
1125 		ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
1126 					     socket, &rxq_conf,
1127 					     rxq->pool);
1128 		if (ret < 0) {
1129 			printf("\n");
1130 			rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: "
1131 				"err=%d, port=%d\n",
1132 				ret, portid);
1133 		}
1134 
1135 		ret = rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1136 		if (ret < 0) {
1137 			printf("\n");
1138 			rte_exit(EXIT_FAILURE,
1139 				"rte_eth_macaddr_get: err=%d, port=%d\n",
1140 				ret, portid);
1141 		}
1142 
1143 		print_ethaddr(" Address:", &ports_eth_addr[portid]);
1144 		printf("\n");
1145 
1146 		/* init one TX queue per couple (lcore,port) */
1147 		queueid = 0;
1148 		for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1149 			if (rte_lcore_is_enabled(lcore_id) == 0)
1150 				continue;
1151 
1152 			socket = (int) rte_lcore_to_socket_id(lcore_id);
1153 
1154 			printf("txq=%u,%d,%d ", lcore_id, queueid, socket);
1155 			fflush(stdout);
1156 
1157 			txconf = &dev_info.default_txconf;
1158 			txconf->offloads = local_port_conf.txmode.offloads;
1159 
1160 			ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1161 					socket, txconf);
1162 			if (ret < 0)
1163 				rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
1164 					"port=%d\n", ret, portid);
1165 
1166 			qconf = &lcore_queue_conf[lcore_id];
1167 			qconf->tx_queue_id[portid] = queueid;
1168 			setup_port_tbl(qconf, lcore_id, socket, portid);
1169 			queueid++;
1170 		}
1171 		printf("\n");
1172 	}
1173 
1174 	printf("\n");
1175 
1176 	/* start ports */
1177 	RTE_ETH_FOREACH_DEV(portid) {
1178 		if ((enabled_port_mask & (1 << portid)) == 0) {
1179 			continue;
1180 		}
1181 		/* Start device */
1182 		ret = rte_eth_dev_start(portid);
1183 		if (ret < 0)
1184 			rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1185 				ret, portid);
1186 
1187 		ret = rte_eth_promiscuous_enable(portid);
1188 		if (ret != 0)
1189 			rte_exit(EXIT_FAILURE,
1190 				"rte_eth_promiscuous_enable: err=%s, port=%d\n",
1191 				rte_strerror(-ret), portid);
1192 	}
1193 
1194 	if (init_routing_table() < 0)
1195 		rte_exit(EXIT_FAILURE, "Cannot init routing table\n");
1196 
1197 	check_all_ports_link_status(enabled_port_mask);
1198 
1199 	signal(SIGUSR1, signal_handler);
1200 	signal(SIGTERM, signal_handler);
1201 	signal(SIGINT, signal_handler);
1202 
1203 	/* launch per-lcore init on every lcore */
1204 	rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1205 	RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1206 		if (rte_eal_wait_lcore(lcore_id) < 0)
1207 			return -1;
1208 	}
1209 
1210 	return 0;
1211 }
1212