xref: /dpdk/examples/ip_reassembly/main.c (revision 25d11a86c56d50947af33d0b79ede622809bd8b9)
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 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 		{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_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 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 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 ipv4_hdr *ip_hdr;
328 		uint32_t ip_dst;
329 
330 		ip_hdr = (struct 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 ether_hdr *);
354 				ip_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
355 			}
356 		}
357 		ip_dst = rte_be_to_cpu_32(ip_hdr->dst_addr);
358 
359 		/* Find destination port */
360 		if (rte_lpm_lookup(rxq->lpm, ip_dst, &next_hop) == 0 &&
361 				(enabled_port_mask & 1 << next_hop) != 0) {
362 			dst_port = next_hop;
363 		}
364 
365 		eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv4);
366 	} else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
367 		/* if packet is IPv6 */
368 		struct ipv6_extension_fragment *frag_hdr;
369 		struct ipv6_hdr *ip_hdr;
370 
371 		ip_hdr = (struct ipv6_hdr *)(eth_hdr + 1);
372 
373 		frag_hdr = rte_ipv6_frag_get_ipv6_fragment_header(ip_hdr);
374 
375 		if (frag_hdr != NULL) {
376 			struct rte_mbuf *mo;
377 
378 			tbl = rxq->frag_tbl;
379 			dr  = &qconf->death_row;
380 
381 			/* prepare mbuf: setup l2_len/l3_len. */
382 			m->l2_len = sizeof(*eth_hdr);
383 			m->l3_len = sizeof(*ip_hdr) + sizeof(*frag_hdr);
384 
385 			mo = rte_ipv6_frag_reassemble_packet(tbl, dr, m, tms, ip_hdr, frag_hdr);
386 			if (mo == NULL)
387 				return;
388 
389 			if (mo != m) {
390 				m = mo;
391 				eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
392 				ip_hdr = (struct ipv6_hdr *)(eth_hdr + 1);
393 			}
394 		}
395 
396 		/* Find destination port */
397 		if (rte_lpm6_lookup(rxq->lpm6, ip_hdr->dst_addr,
398 						&next_hop) == 0 &&
399 				(enabled_port_mask & 1 << next_hop) != 0) {
400 			dst_port = next_hop;
401 		}
402 
403 		eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv6);
404 	}
405 	/* if packet wasn't IPv4 or IPv6, it's forwarded to the port it came from */
406 
407 	/* 02:00:00:00:00:xx */
408 	d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
409 	*((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)dst_port << 40);
410 
411 	/* src addr */
412 	ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
413 
414 	send_single_packet(m, dst_port);
415 }
416 
417 /* main processing loop */
418 static int
419 main_loop(__attribute__((unused)) void *dummy)
420 {
421 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
422 	unsigned lcore_id;
423 	uint64_t diff_tsc, cur_tsc, prev_tsc;
424 	int i, j, nb_rx;
425 	uint16_t portid;
426 	struct lcore_queue_conf *qconf;
427 	const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
428 
429 	prev_tsc = 0;
430 
431 	lcore_id = rte_lcore_id();
432 	qconf = &lcore_queue_conf[lcore_id];
433 
434 	if (qconf->n_rx_queue == 0) {
435 		RTE_LOG(INFO, IP_RSMBL, "lcore %u has nothing to do\n", lcore_id);
436 		return 0;
437 	}
438 
439 	RTE_LOG(INFO, IP_RSMBL, "entering main loop on lcore %u\n", lcore_id);
440 
441 	for (i = 0; i < qconf->n_rx_queue; i++) {
442 
443 		portid = qconf->rx_queue_list[i].portid;
444 		RTE_LOG(INFO, IP_RSMBL, " -- lcoreid=%u portid=%u\n", lcore_id,
445 			portid);
446 	}
447 
448 	while (1) {
449 
450 		cur_tsc = rte_rdtsc();
451 
452 		/*
453 		 * TX burst queue drain
454 		 */
455 		diff_tsc = cur_tsc - prev_tsc;
456 		if (unlikely(diff_tsc > drain_tsc)) {
457 
458 			/*
459 			 * This could be optimized (use queueid instead of
460 			 * portid), but it is not called so often
461 			 */
462 			for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
463 				if ((enabled_port_mask & (1 << portid)) != 0)
464 					send_burst(qconf, 1, portid);
465 			}
466 
467 			prev_tsc = cur_tsc;
468 		}
469 
470 		/*
471 		 * Read packet from RX queues
472 		 */
473 		for (i = 0; i < qconf->n_rx_queue; ++i) {
474 
475 			portid = qconf->rx_queue_list[i].portid;
476 
477 			nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
478 				MAX_PKT_BURST);
479 
480 			/* Prefetch first packets */
481 			for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
482 				rte_prefetch0(rte_pktmbuf_mtod(
483 						pkts_burst[j], void *));
484 			}
485 
486 			/* Prefetch and forward already prefetched packets */
487 			for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
488 				rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
489 					j + PREFETCH_OFFSET], void *));
490 				reassemble(pkts_burst[j], portid,
491 					i, qconf, cur_tsc);
492 			}
493 
494 			/* Forward remaining prefetched packets */
495 			for (; j < nb_rx; j++) {
496 				reassemble(pkts_burst[j], portid,
497 					i, qconf, cur_tsc);
498 			}
499 
500 			rte_ip_frag_free_death_row(&qconf->death_row,
501 				PREFETCH_OFFSET);
502 		}
503 	}
504 }
505 
506 /* display usage */
507 static void
508 print_usage(const char *prgname)
509 {
510 	printf("%s [EAL options] -- -p PORTMASK [-q NQ]"
511 		"  [--max-pkt-len PKTLEN]"
512 		"  [--maxflows=<flows>]  [--flowttl=<ttl>[(s|ms)]]\n"
513 		"  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
514 		"  -q NQ: number of RX queues per lcore\n"
515 		"  --maxflows=<flows>: optional, maximum number of flows "
516 		"supported\n"
517 		"  --flowttl=<ttl>[(s|ms)]: optional, maximum TTL for each "
518 		"flow\n",
519 		prgname);
520 }
521 
522 static uint32_t
523 parse_flow_num(const char *str, uint32_t min, uint32_t max, uint32_t *val)
524 {
525 	char *end;
526 	uint64_t v;
527 
528 	/* parse decimal string */
529 	errno = 0;
530 	v = strtoul(str, &end, 10);
531 	if (errno != 0 || *end != '\0')
532 		return -EINVAL;
533 
534 	if (v < min || v > max)
535 		return -EINVAL;
536 
537 	*val = (uint32_t)v;
538 	return 0;
539 }
540 
541 static int
542 parse_flow_ttl(const char *str, uint32_t min, uint32_t max, uint32_t *val)
543 {
544 	char *end;
545 	uint64_t v;
546 
547 	static const char frmt_sec[] = "s";
548 	static const char frmt_msec[] = "ms";
549 
550 	/* parse decimal string */
551 	errno = 0;
552 	v = strtoul(str, &end, 10);
553 	if (errno != 0)
554 		return -EINVAL;
555 
556 	if (*end != '\0') {
557 		if (strncmp(frmt_sec, end, sizeof(frmt_sec)) == 0)
558 			v *= MS_PER_S;
559 		else if (strncmp(frmt_msec, end, sizeof (frmt_msec)) != 0)
560 			return -EINVAL;
561 	}
562 
563 	if (v < min || v > max)
564 		return -EINVAL;
565 
566 	*val = (uint32_t)v;
567 	return 0;
568 }
569 
570 static int
571 parse_portmask(const char *portmask)
572 {
573 	char *end = NULL;
574 	unsigned long pm;
575 
576 	/* parse hexadecimal string */
577 	pm = strtoul(portmask, &end, 16);
578 	if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
579 		return -1;
580 
581 	if (pm == 0)
582 		return -1;
583 
584 	return pm;
585 }
586 
587 static int
588 parse_nqueue(const char *q_arg)
589 {
590 	char *end = NULL;
591 	unsigned long n;
592 
593 	printf("%p\n", q_arg);
594 
595 	/* parse hexadecimal string */
596 	n = strtoul(q_arg, &end, 10);
597 	if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
598 		return -1;
599 	if (n == 0)
600 		return -1;
601 	if (n >= MAX_RX_QUEUE_PER_LCORE)
602 		return -1;
603 
604 	return n;
605 }
606 
607 /* Parse the argument given in the command line of the application */
608 static int
609 parse_args(int argc, char **argv)
610 {
611 	int opt, ret;
612 	char **argvopt;
613 	int option_index;
614 	char *prgname = argv[0];
615 	static struct option lgopts[] = {
616 		{"max-pkt-len", 1, 0, 0},
617 		{"maxflows", 1, 0, 0},
618 		{"flowttl", 1, 0, 0},
619 		{NULL, 0, 0, 0}
620 	};
621 
622 	argvopt = argv;
623 
624 	while ((opt = getopt_long(argc, argvopt, "p:q:",
625 				lgopts, &option_index)) != EOF) {
626 
627 		switch (opt) {
628 		/* portmask */
629 		case 'p':
630 			enabled_port_mask = parse_portmask(optarg);
631 			if (enabled_port_mask == 0) {
632 				printf("invalid portmask\n");
633 				print_usage(prgname);
634 				return -1;
635 			}
636 			break;
637 
638 		/* nqueue */
639 		case 'q':
640 			rx_queue_per_lcore = parse_nqueue(optarg);
641 			if (rx_queue_per_lcore < 0) {
642 				printf("invalid queue number\n");
643 				print_usage(prgname);
644 				return -1;
645 			}
646 			break;
647 
648 		/* long options */
649 		case 0:
650 			if (!strncmp(lgopts[option_index].name,
651 					"maxflows", 8)) {
652 				if ((ret = parse_flow_num(optarg, MIN_FLOW_NUM,
653 						MAX_FLOW_NUM,
654 						&max_flow_num)) != 0) {
655 					printf("invalid value: \"%s\" for "
656 						"parameter %s\n",
657 						optarg,
658 						lgopts[option_index].name);
659 					print_usage(prgname);
660 					return ret;
661 				}
662 			}
663 
664 			if (!strncmp(lgopts[option_index].name, "flowttl", 7)) {
665 				if ((ret = parse_flow_ttl(optarg, MIN_FLOW_TTL,
666 						MAX_FLOW_TTL,
667 						&max_flow_ttl)) != 0) {
668 					printf("invalid value: \"%s\" for "
669 						"parameter %s\n",
670 						optarg,
671 						lgopts[option_index].name);
672 					print_usage(prgname);
673 					return ret;
674 				}
675 			}
676 
677 			break;
678 
679 		default:
680 			print_usage(prgname);
681 			return -1;
682 		}
683 	}
684 
685 	if (optind >= 0)
686 		argv[optind-1] = prgname;
687 
688 	ret = optind-1;
689 	optind = 1; /* reset getopt lib */
690 	return ret;
691 }
692 
693 static void
694 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
695 {
696 	char buf[ETHER_ADDR_FMT_SIZE];
697 	ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
698 	printf("%s%s", name, buf);
699 }
700 
701 /* Check the link status of all ports in up to 9s, and print them finally */
702 static void
703 check_all_ports_link_status(uint32_t port_mask)
704 {
705 #define CHECK_INTERVAL 100 /* 100ms */
706 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
707 	uint16_t portid;
708 	uint8_t count, all_ports_up, print_flag = 0;
709 	struct rte_eth_link link;
710 
711 	printf("\nChecking link status");
712 	fflush(stdout);
713 	for (count = 0; count <= MAX_CHECK_TIME; count++) {
714 		all_ports_up = 1;
715 		RTE_ETH_FOREACH_DEV(portid) {
716 			if ((port_mask & (1 << portid)) == 0)
717 				continue;
718 			memset(&link, 0, sizeof(link));
719 			rte_eth_link_get_nowait(portid, &link);
720 			/* print link status if flag set */
721 			if (print_flag == 1) {
722 				if (link.link_status)
723 					printf(
724 					"Port%d Link Up. Speed %u Mbps - %s\n",
725 						portid, link.link_speed,
726 				(link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
727 					("full-duplex") : ("half-duplex\n"));
728 				else
729 					printf("Port %d Link Down\n", portid);
730 				continue;
731 			}
732 			/* clear all_ports_up flag if any link down */
733 			if (link.link_status == ETH_LINK_DOWN) {
734 				all_ports_up = 0;
735 				break;
736 			}
737 		}
738 		/* after finally printing all link status, get out */
739 		if (print_flag == 1)
740 			break;
741 
742 		if (all_ports_up == 0) {
743 			printf(".");
744 			fflush(stdout);
745 			rte_delay_ms(CHECK_INTERVAL);
746 		}
747 
748 		/* set the print_flag if all ports up or timeout */
749 		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
750 			print_flag = 1;
751 			printf("\ndone\n");
752 		}
753 	}
754 }
755 
756 static int
757 init_routing_table(void)
758 {
759 	struct rte_lpm *lpm;
760 	struct rte_lpm6 *lpm6;
761 	int socket, ret;
762 	unsigned i;
763 
764 	for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) {
765 		if (socket_lpm[socket]) {
766 			lpm = socket_lpm[socket];
767 			/* populate the LPM table */
768 			for (i = 0; i < RTE_DIM(l3fwd_ipv4_route_array); i++) {
769 				ret = rte_lpm_add(lpm,
770 					l3fwd_ipv4_route_array[i].ip,
771 					l3fwd_ipv4_route_array[i].depth,
772 					l3fwd_ipv4_route_array[i].if_out);
773 
774 				if (ret < 0) {
775 					RTE_LOG(ERR, IP_RSMBL, "Unable to add entry %i to the l3fwd "
776 						"LPM table\n", i);
777 					return -1;
778 				}
779 
780 				RTE_LOG(INFO, IP_RSMBL, "Socket %i: adding route " IPv4_BYTES_FMT
781 						"/%d (port %d)\n",
782 					socket,
783 					IPv4_BYTES(l3fwd_ipv4_route_array[i].ip),
784 					l3fwd_ipv4_route_array[i].depth,
785 					l3fwd_ipv4_route_array[i].if_out);
786 			}
787 		}
788 
789 		if (socket_lpm6[socket]) {
790 			lpm6 = socket_lpm6[socket];
791 			/* populate the LPM6 table */
792 			for (i = 0; i < RTE_DIM(l3fwd_ipv6_route_array); i++) {
793 				ret = rte_lpm6_add(lpm6,
794 					l3fwd_ipv6_route_array[i].ip,
795 					l3fwd_ipv6_route_array[i].depth,
796 					l3fwd_ipv6_route_array[i].if_out);
797 
798 				if (ret < 0) {
799 					RTE_LOG(ERR, IP_RSMBL, "Unable to add entry %i to the l3fwd "
800 						"LPM6 table\n", i);
801 					return -1;
802 				}
803 
804 				RTE_LOG(INFO, IP_RSMBL, "Socket %i: adding route " IPv6_BYTES_FMT
805 						"/%d (port %d)\n",
806 					socket,
807 					IPv6_BYTES(l3fwd_ipv6_route_array[i].ip),
808 					l3fwd_ipv6_route_array[i].depth,
809 					l3fwd_ipv6_route_array[i].if_out);
810 			}
811 		}
812 	}
813 	return 0;
814 }
815 
816 static int
817 setup_port_tbl(struct lcore_queue_conf *qconf, uint32_t lcore, int socket,
818 	uint32_t port)
819 {
820 	struct mbuf_table *mtb;
821 	uint32_t n;
822 	size_t sz;
823 
824 	n = RTE_MAX(max_flow_num, 2UL * MAX_PKT_BURST);
825 	sz = sizeof (*mtb) + sizeof (mtb->m_table[0]) *  n;
826 
827 	if ((mtb = rte_zmalloc_socket(__func__, sz, RTE_CACHE_LINE_SIZE,
828 			socket)) == NULL) {
829 		RTE_LOG(ERR, IP_RSMBL, "%s() for lcore: %u, port: %u "
830 			"failed to allocate %zu bytes\n",
831 			__func__, lcore, port, sz);
832 		return -1;
833 	}
834 
835 	mtb->len = n;
836 	qconf->tx_mbufs[port] = mtb;
837 
838 	return 0;
839 }
840 
841 static int
842 setup_queue_tbl(struct rx_queue *rxq, uint32_t lcore, uint32_t queue)
843 {
844 	int socket;
845 	uint32_t nb_mbuf;
846 	uint64_t frag_cycles;
847 	char buf[RTE_MEMPOOL_NAMESIZE];
848 
849 	socket = rte_lcore_to_socket_id(lcore);
850 	if (socket == SOCKET_ID_ANY)
851 		socket = 0;
852 
853 	frag_cycles = (rte_get_tsc_hz() + MS_PER_S - 1) / MS_PER_S *
854 		max_flow_ttl;
855 
856 	if ((rxq->frag_tbl = rte_ip_frag_table_create(max_flow_num,
857 			IP_FRAG_TBL_BUCKET_ENTRIES, max_flow_num, frag_cycles,
858 			socket)) == NULL) {
859 		RTE_LOG(ERR, IP_RSMBL, "ip_frag_tbl_create(%u) on "
860 			"lcore: %u for queue: %u failed\n",
861 			max_flow_num, lcore, queue);
862 		return -1;
863 	}
864 
865 	/*
866 	 * At any given moment up to <max_flow_num * (MAX_FRAG_NUM)>
867 	 * mbufs could be stored int the fragment table.
868 	 * Plus, each TX queue can hold up to <max_flow_num> packets.
869 	 */
870 
871 	nb_mbuf = RTE_MAX(max_flow_num, 2UL * MAX_PKT_BURST) * MAX_FRAG_NUM;
872 	nb_mbuf *= (port_conf.rxmode.max_rx_pkt_len + BUF_SIZE - 1) / BUF_SIZE;
873 	nb_mbuf *= 2; /* ipv4 and ipv6 */
874 	nb_mbuf += nb_rxd + nb_txd;
875 
876 	nb_mbuf = RTE_MAX(nb_mbuf, (uint32_t)NB_MBUF);
877 
878 	snprintf(buf, sizeof(buf), "mbuf_pool_%u_%u", lcore, queue);
879 
880 	rxq->pool = rte_pktmbuf_pool_create(buf, nb_mbuf, MEMPOOL_CACHE_SIZE, 0,
881 					    MBUF_DATA_SIZE, socket);
882 	if (rxq->pool == NULL) {
883 		RTE_LOG(ERR, IP_RSMBL,
884 			"rte_pktmbuf_pool_create(%s) failed", buf);
885 		return -1;
886 	}
887 
888 	return 0;
889 }
890 
891 static int
892 init_mem(void)
893 {
894 	char buf[PATH_MAX];
895 	struct rte_lpm *lpm;
896 	struct rte_lpm6 *lpm6;
897 	struct rte_lpm_config lpm_config;
898 	int socket;
899 	unsigned lcore_id;
900 
901 	/* traverse through lcores and initialize structures on each socket */
902 
903 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
904 
905 		if (rte_lcore_is_enabled(lcore_id) == 0)
906 			continue;
907 
908 		socket = rte_lcore_to_socket_id(lcore_id);
909 
910 		if (socket == SOCKET_ID_ANY)
911 			socket = 0;
912 
913 		if (socket_lpm[socket] == NULL) {
914 			RTE_LOG(INFO, IP_RSMBL, "Creating LPM table on socket %i\n", socket);
915 			snprintf(buf, sizeof(buf), "IP_RSMBL_LPM_%i", socket);
916 
917 			lpm_config.max_rules = LPM_MAX_RULES;
918 			lpm_config.number_tbl8s = 256;
919 			lpm_config.flags = 0;
920 
921 			lpm = rte_lpm_create(buf, socket, &lpm_config);
922 			if (lpm == NULL) {
923 				RTE_LOG(ERR, IP_RSMBL, "Cannot create LPM table\n");
924 				return -1;
925 			}
926 			socket_lpm[socket] = lpm;
927 		}
928 
929 		if (socket_lpm6[socket] == NULL) {
930 			RTE_LOG(INFO, IP_RSMBL, "Creating LPM6 table on socket %i\n", socket);
931 			snprintf(buf, sizeof(buf), "IP_RSMBL_LPM_%i", socket);
932 
933 			lpm6 = rte_lpm6_create(buf, socket, &lpm6_config);
934 			if (lpm6 == NULL) {
935 				RTE_LOG(ERR, IP_RSMBL, "Cannot create LPM table\n");
936 				return -1;
937 			}
938 			socket_lpm6[socket] = lpm6;
939 		}
940 	}
941 
942 	return 0;
943 }
944 
945 static void
946 queue_dump_stat(void)
947 {
948 	uint32_t i, lcore;
949 	const struct lcore_queue_conf *qconf;
950 
951 	for (lcore = 0; lcore < RTE_MAX_LCORE; lcore++) {
952 		if (rte_lcore_is_enabled(lcore) == 0)
953 			continue;
954 
955 		qconf = &lcore_queue_conf[lcore];
956 		for (i = 0; i < qconf->n_rx_queue; i++) {
957 
958 			fprintf(stdout, " -- lcoreid=%u portid=%u "
959 				"frag tbl stat:\n",
960 				lcore,  qconf->rx_queue_list[i].portid);
961 			rte_ip_frag_table_statistics_dump(stdout,
962 					qconf->rx_queue_list[i].frag_tbl);
963 			fprintf(stdout, "TX bursts:\t%" PRIu64 "\n"
964 				"TX packets _queued:\t%" PRIu64 "\n"
965 				"TX packets dropped:\t%" PRIu64 "\n"
966 				"TX packets send:\t%" PRIu64 "\n",
967 				qconf->tx_stat.call,
968 				qconf->tx_stat.queue,
969 				qconf->tx_stat.drop,
970 				qconf->tx_stat.send);
971 		}
972 	}
973 }
974 
975 static void
976 signal_handler(int signum)
977 {
978 	queue_dump_stat();
979 	if (signum != SIGUSR1)
980 		rte_exit(0, "received signal: %d, exiting\n", signum);
981 }
982 
983 int
984 main(int argc, char **argv)
985 {
986 	struct lcore_queue_conf *qconf;
987 	struct rte_eth_dev_info dev_info;
988 	struct rte_eth_txconf *txconf;
989 	struct rx_queue *rxq;
990 	int ret, socket;
991 	unsigned nb_ports;
992 	uint16_t queueid;
993 	unsigned lcore_id = 0, rx_lcore_id = 0;
994 	uint32_t n_tx_queue, nb_lcores;
995 	uint16_t portid;
996 
997 	/* init EAL */
998 	ret = rte_eal_init(argc, argv);
999 	if (ret < 0)
1000 		rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1001 	argc -= ret;
1002 	argv += ret;
1003 
1004 	/* parse application arguments (after the EAL ones) */
1005 	ret = parse_args(argc, argv);
1006 	if (ret < 0)
1007 		rte_exit(EXIT_FAILURE, "Invalid IP reassembly parameters\n");
1008 
1009 	nb_ports = rte_eth_dev_count_avail();
1010 	if (nb_ports == 0)
1011 		rte_exit(EXIT_FAILURE, "No ports found!\n");
1012 
1013 	nb_lcores = rte_lcore_count();
1014 
1015 	/* initialize structures (mempools, lpm etc.) */
1016 	if (init_mem() < 0)
1017 		rte_panic("Cannot initialize memory structures!\n");
1018 
1019 	/* check if portmask has non-existent ports */
1020 	if (enabled_port_mask & ~(RTE_LEN2MASK(nb_ports, unsigned)))
1021 		rte_exit(EXIT_FAILURE, "Non-existent ports in portmask!\n");
1022 
1023 	/* initialize all ports */
1024 	RTE_ETH_FOREACH_DEV(portid) {
1025 		struct rte_eth_rxconf rxq_conf;
1026 		struct rte_eth_conf local_port_conf = port_conf;
1027 
1028 		/* skip ports that are not enabled */
1029 		if ((enabled_port_mask & (1 << portid)) == 0) {
1030 			printf("\nSkipping disabled port %d\n", portid);
1031 			continue;
1032 		}
1033 
1034 		qconf = &lcore_queue_conf[rx_lcore_id];
1035 
1036 		/* limit the frame size to the maximum supported by NIC */
1037 		rte_eth_dev_info_get(portid, &dev_info);
1038 		local_port_conf.rxmode.max_rx_pkt_len = RTE_MIN(
1039 		    dev_info.max_rx_pktlen,
1040 		    local_port_conf.rxmode.max_rx_pkt_len);
1041 
1042 		/* get the lcore_id for this port */
1043 		while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
1044 			   qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
1045 
1046 			rx_lcore_id++;
1047 			if (rx_lcore_id >= RTE_MAX_LCORE)
1048 				rte_exit(EXIT_FAILURE, "Not enough cores\n");
1049 
1050 			qconf = &lcore_queue_conf[rx_lcore_id];
1051 		}
1052 
1053 		socket = rte_lcore_to_socket_id(portid);
1054 		if (socket == SOCKET_ID_ANY)
1055 			socket = 0;
1056 
1057 		queueid = qconf->n_rx_queue;
1058 		rxq = &qconf->rx_queue_list[queueid];
1059 		rxq->portid = portid;
1060 		rxq->lpm = socket_lpm[socket];
1061 		rxq->lpm6 = socket_lpm6[socket];
1062 
1063 		ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
1064 						       &nb_txd);
1065 		if (ret < 0)
1066 			rte_exit(EXIT_FAILURE,
1067 				 "Cannot adjust number of descriptors: err=%d, port=%d\n",
1068 				 ret, portid);
1069 
1070 		if (setup_queue_tbl(rxq, rx_lcore_id, queueid) < 0)
1071 			rte_exit(EXIT_FAILURE, "Failed to set up queue table\n");
1072 		qconf->n_rx_queue++;
1073 
1074 		/* init port */
1075 		printf("Initializing port %d ... ", portid );
1076 		fflush(stdout);
1077 
1078 		n_tx_queue = nb_lcores;
1079 		if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
1080 			n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1081 		if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
1082 			local_port_conf.txmode.offloads |=
1083 				DEV_TX_OFFLOAD_MBUF_FAST_FREE;
1084 
1085 		local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
1086 			dev_info.flow_type_rss_offloads;
1087 		if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
1088 				port_conf.rx_adv_conf.rss_conf.rss_hf) {
1089 			printf("Port %u modified RSS hash function based on hardware support,"
1090 				"requested:%#"PRIx64" configured:%#"PRIx64"\n",
1091 				portid,
1092 				port_conf.rx_adv_conf.rss_conf.rss_hf,
1093 				local_port_conf.rx_adv_conf.rss_conf.rss_hf);
1094 		}
1095 
1096 		ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
1097 					    &local_port_conf);
1098 		if (ret < 0) {
1099 			printf("\n");
1100 			rte_exit(EXIT_FAILURE, "Cannot configure device: "
1101 				"err=%d, port=%d\n",
1102 				ret, portid);
1103 		}
1104 
1105 		/* init one RX queue */
1106 		rxq_conf = dev_info.default_rxconf;
1107 		rxq_conf.offloads = local_port_conf.rxmode.offloads;
1108 		ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
1109 					     socket, &rxq_conf,
1110 					     rxq->pool);
1111 		if (ret < 0) {
1112 			printf("\n");
1113 			rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: "
1114 				"err=%d, port=%d\n",
1115 				ret, portid);
1116 		}
1117 
1118 		rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1119 		print_ethaddr(" Address:", &ports_eth_addr[portid]);
1120 		printf("\n");
1121 
1122 		/* init one TX queue per couple (lcore,port) */
1123 		queueid = 0;
1124 		for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1125 			if (rte_lcore_is_enabled(lcore_id) == 0)
1126 				continue;
1127 
1128 			socket = (int) rte_lcore_to_socket_id(lcore_id);
1129 
1130 			printf("txq=%u,%d,%d ", lcore_id, queueid, socket);
1131 			fflush(stdout);
1132 
1133 			txconf = &dev_info.default_txconf;
1134 			txconf->offloads = local_port_conf.txmode.offloads;
1135 
1136 			ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1137 					socket, txconf);
1138 			if (ret < 0)
1139 				rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
1140 					"port=%d\n", ret, portid);
1141 
1142 			qconf = &lcore_queue_conf[lcore_id];
1143 			qconf->tx_queue_id[portid] = queueid;
1144 			setup_port_tbl(qconf, lcore_id, socket, portid);
1145 			queueid++;
1146 		}
1147 		printf("\n");
1148 	}
1149 
1150 	printf("\n");
1151 
1152 	/* start ports */
1153 	RTE_ETH_FOREACH_DEV(portid) {
1154 		if ((enabled_port_mask & (1 << portid)) == 0) {
1155 			continue;
1156 		}
1157 		/* Start device */
1158 		ret = rte_eth_dev_start(portid);
1159 		if (ret < 0)
1160 			rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1161 				ret, portid);
1162 
1163 		rte_eth_promiscuous_enable(portid);
1164 	}
1165 
1166 	if (init_routing_table() < 0)
1167 		rte_exit(EXIT_FAILURE, "Cannot init routing table\n");
1168 
1169 	check_all_ports_link_status(enabled_port_mask);
1170 
1171 	signal(SIGUSR1, signal_handler);
1172 	signal(SIGTERM, signal_handler);
1173 	signal(SIGINT, signal_handler);
1174 
1175 	/* launch per-lcore init on every lcore */
1176 	rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1177 	RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1178 		if (rte_eal_wait_lcore(lcore_id) < 0)
1179 			return -1;
1180 	}
1181 
1182 	return 0;
1183 }
1184