xref: /dpdk/examples/ip_reassembly/main.c (revision a103a97e7191179ad6a451ce85182df2ecb10c26)
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 <string.h>
40 #include <sys/queue.h>
41 #include <stdarg.h>
42 #include <errno.h>
43 #include <getopt.h>
44 #include <signal.h>
45 #include <sys/param.h>
46 
47 #include <rte_common.h>
48 #include <rte_byteorder.h>
49 #include <rte_log.h>
50 #include <rte_memory.h>
51 #include <rte_memcpy.h>
52 #include <rte_memzone.h>
53 #include <rte_eal.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_mempool.h>
68 #include <rte_mbuf.h>
69 #include <rte_malloc.h>
70 #include <rte_ip.h>
71 #include <rte_tcp.h>
72 #include <rte_udp.h>
73 #include <rte_string_fns.h>
74 #include <rte_lpm.h>
75 #include <rte_lpm6.h>
76 
77 #include <rte_ip_frag.h>
78 
79 #define MAX_PKT_BURST 32
80 
81 
82 #define RTE_LOGTYPE_IP_RSMBL RTE_LOGTYPE_USER1
83 
84 #define MAX_JUMBO_PKT_LEN  9600
85 
86 #define	BUF_SIZE	RTE_MBUF_DEFAULT_DATAROOM
87 #define MBUF_SIZE	\
88 	(BUF_SIZE + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
89 
90 #define NB_MBUF 8192
91 
92 /* allow max jumbo frame 9.5 KB */
93 #define JUMBO_FRAME_MAX_SIZE	0x2600
94 
95 #define	MAX_FLOW_NUM	UINT16_MAX
96 #define	MIN_FLOW_NUM	1
97 #define	DEF_FLOW_NUM	0x1000
98 
99 /* TTL numbers are in ms. */
100 #define	MAX_FLOW_TTL	(3600 * MS_PER_S)
101 #define	MIN_FLOW_TTL	1
102 #define	DEF_FLOW_TTL	MS_PER_S
103 
104 #define MAX_FRAG_NUM RTE_LIBRTE_IP_FRAG_MAX_FRAG
105 
106 /* Should be power of two. */
107 #define	IP_FRAG_TBL_BUCKET_ENTRIES	16
108 
109 static uint32_t max_flow_num = DEF_FLOW_NUM;
110 static uint32_t max_flow_ttl = DEF_FLOW_TTL;
111 
112 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
113 
114 #define NB_SOCKETS 8
115 
116 /* Configure how many packets ahead to prefetch, when reading packets */
117 #define PREFETCH_OFFSET	3
118 
119 /*
120  * Configurable number of RX/TX ring descriptors
121  */
122 #define RTE_TEST_RX_DESC_DEFAULT 128
123 #define RTE_TEST_TX_DESC_DEFAULT 512
124 
125 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
126 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
127 
128 /* ethernet addresses of ports */
129 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
130 
131 #ifndef IPv4_BYTES
132 #define IPv4_BYTES_FMT "%" PRIu8 ".%" PRIu8 ".%" PRIu8 ".%" PRIu8
133 #define IPv4_BYTES(addr) \
134 		(uint8_t) (((addr) >> 24) & 0xFF),\
135 		(uint8_t) (((addr) >> 16) & 0xFF),\
136 		(uint8_t) (((addr) >> 8) & 0xFF),\
137 		(uint8_t) ((addr) & 0xFF)
138 #endif
139 
140 #ifndef IPv6_BYTES
141 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
142                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
143 #define IPv6_BYTES(addr) \
144 	addr[0],  addr[1], addr[2],  addr[3], \
145 	addr[4],  addr[5], addr[6],  addr[7], \
146 	addr[8],  addr[9], addr[10], addr[11],\
147 	addr[12], addr[13],addr[14], addr[15]
148 #endif
149 
150 #define IPV6_ADDR_LEN 16
151 
152 /* mask of enabled ports */
153 static uint32_t enabled_port_mask = 0;
154 
155 static int rx_queue_per_lcore = 1;
156 
157 struct mbuf_table {
158 	uint32_t len;
159 	uint32_t head;
160 	uint32_t tail;
161 	struct rte_mbuf *m_table[0];
162 };
163 
164 struct rx_queue {
165 	struct rte_ip_frag_tbl *frag_tbl;
166 	struct rte_mempool *pool;
167 	struct rte_lpm *lpm;
168 	struct rte_lpm6 *lpm6;
169 	uint16_t portid;
170 };
171 
172 struct tx_lcore_stat {
173 	uint64_t call;
174 	uint64_t drop;
175 	uint64_t queue;
176 	uint64_t send;
177 };
178 
179 #define MAX_RX_QUEUE_PER_LCORE 16
180 #define MAX_TX_QUEUE_PER_PORT 16
181 #define MAX_RX_QUEUE_PER_PORT 128
182 
183 struct lcore_queue_conf {
184 	uint16_t n_rx_queue;
185 	struct rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
186 	uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
187 	struct rte_ip_frag_death_row death_row;
188 	struct mbuf_table *tx_mbufs[RTE_MAX_ETHPORTS];
189 	struct tx_lcore_stat tx_stat;
190 } __rte_cache_aligned;
191 static struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
192 
193 static struct rte_eth_conf port_conf = {
194 	.rxmode = {
195 		.mq_mode        = ETH_MQ_RX_RSS,
196 		.max_rx_pkt_len = JUMBO_FRAME_MAX_SIZE,
197 		.split_hdr_size = 0,
198 		.header_split   = 0, /**< Header Split disabled */
199 		.hw_ip_checksum = 1, /**< IP checksum offload enabled */
200 		.hw_vlan_filter = 0, /**< VLAN filtering disabled */
201 		.jumbo_frame    = 1, /**< Jumbo Frame Support disabled */
202 		.hw_strip_crc   = 1, /**< CRC stripped by hardware */
203 	},
204 	.rx_adv_conf = {
205 			.rss_conf = {
206 				.rss_key = NULL,
207 				.rss_hf = ETH_RSS_IP,
208 		},
209 	},
210 	.txmode = {
211 		.mq_mode = ETH_MQ_TX_NONE,
212 	},
213 };
214 
215 /*
216  * IPv4 forwarding table
217  */
218 struct l3fwd_ipv4_route {
219 	uint32_t ip;
220 	uint8_t  depth;
221 	uint8_t  if_out;
222 };
223 
224 struct l3fwd_ipv4_route l3fwd_ipv4_route_array[] = {
225 		{IPv4(100,10,0,0), 16, 0},
226 		{IPv4(100,20,0,0), 16, 1},
227 		{IPv4(100,30,0,0), 16, 2},
228 		{IPv4(100,40,0,0), 16, 3},
229 		{IPv4(100,50,0,0), 16, 4},
230 		{IPv4(100,60,0,0), 16, 5},
231 		{IPv4(100,70,0,0), 16, 6},
232 		{IPv4(100,80,0,0), 16, 7},
233 };
234 
235 /*
236  * IPv6 forwarding table
237  */
238 
239 struct l3fwd_ipv6_route {
240 	uint8_t ip[IPV6_ADDR_LEN];
241 	uint8_t depth;
242 	uint8_t if_out;
243 };
244 
245 static struct l3fwd_ipv6_route l3fwd_ipv6_route_array[] = {
246 	{{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
247 	{{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
248 	{{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
249 	{{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
250 	{{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
251 	{{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
252 	{{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
253 	{{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
254 };
255 
256 #define LPM_MAX_RULES         1024
257 #define LPM6_MAX_RULES         1024
258 #define LPM6_NUMBER_TBL8S (1 << 16)
259 
260 struct rte_lpm6_config lpm6_config = {
261 		.max_rules = LPM6_MAX_RULES,
262 		.number_tbl8s = LPM6_NUMBER_TBL8S,
263 		.flags = 0
264 };
265 
266 static struct rte_lpm *socket_lpm[RTE_MAX_NUMA_NODES];
267 static struct rte_lpm6 *socket_lpm6[RTE_MAX_NUMA_NODES];
268 
269 #ifdef RTE_LIBRTE_IP_FRAG_TBL_STAT
270 #define TX_LCORE_STAT_UPDATE(s, f, v)   ((s)->f += (v))
271 #else
272 #define TX_LCORE_STAT_UPDATE(s, f, v)   do {} while (0)
273 #endif /* RTE_LIBRTE_IP_FRAG_TBL_STAT */
274 
275 /*
276  * If number of queued packets reached given threahold, then
277  * send burst of packets on an output interface.
278  */
279 static inline uint32_t
280 send_burst(struct lcore_queue_conf *qconf, uint32_t thresh, uint16_t port)
281 {
282 	uint32_t fill, len, k, n;
283 	struct mbuf_table *txmb;
284 
285 	txmb = qconf->tx_mbufs[port];
286 	len = txmb->len;
287 
288 	if ((int32_t)(fill = txmb->head - txmb->tail) < 0)
289 		fill += len;
290 
291 	if (fill >= thresh) {
292 		n = RTE_MIN(len - txmb->tail, fill);
293 
294 		k = rte_eth_tx_burst(port, qconf->tx_queue_id[port],
295 			txmb->m_table + txmb->tail, (uint16_t)n);
296 
297 		TX_LCORE_STAT_UPDATE(&qconf->tx_stat, call, 1);
298 		TX_LCORE_STAT_UPDATE(&qconf->tx_stat, send, k);
299 
300 		fill -= k;
301 		if ((txmb->tail += k) == len)
302 			txmb->tail = 0;
303 	}
304 
305 	return fill;
306 }
307 
308 /* Enqueue a single packet, and send burst if queue is filled */
309 static inline int
310 send_single_packet(struct rte_mbuf *m, uint16_t port)
311 {
312 	uint32_t fill, lcore_id, len;
313 	struct lcore_queue_conf *qconf;
314 	struct mbuf_table *txmb;
315 
316 	lcore_id = rte_lcore_id();
317 	qconf = &lcore_queue_conf[lcore_id];
318 
319 	txmb = qconf->tx_mbufs[port];
320 	len = txmb->len;
321 
322 	fill = send_burst(qconf, MAX_PKT_BURST, port);
323 
324 	if (fill == len - 1) {
325 		TX_LCORE_STAT_UPDATE(&qconf->tx_stat, drop, 1);
326 		rte_pktmbuf_free(txmb->m_table[txmb->tail]);
327 		if (++txmb->tail == len)
328 			txmb->tail = 0;
329 	}
330 
331 	TX_LCORE_STAT_UPDATE(&qconf->tx_stat, queue, 1);
332 	txmb->m_table[txmb->head] = m;
333 	if(++txmb->head == len)
334 		txmb->head = 0;
335 
336 	return 0;
337 }
338 
339 static inline void
340 reassemble(struct rte_mbuf *m, uint16_t portid, uint32_t queue,
341 	struct lcore_queue_conf *qconf, uint64_t tms)
342 {
343 	struct ether_hdr *eth_hdr;
344 	struct rte_ip_frag_tbl *tbl;
345 	struct rte_ip_frag_death_row *dr;
346 	struct rx_queue *rxq;
347 	void *d_addr_bytes;
348 	uint32_t next_hop;
349 	uint16_t dst_port;
350 
351 	rxq = &qconf->rx_queue_list[queue];
352 
353 	eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
354 
355 	dst_port = portid;
356 
357 	/* if packet is IPv4 */
358 	if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
359 		struct ipv4_hdr *ip_hdr;
360 		uint32_t ip_dst;
361 
362 		ip_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
363 
364 		 /* if it is a fragmented packet, then try to reassemble. */
365 		if (rte_ipv4_frag_pkt_is_fragmented(ip_hdr)) {
366 			struct rte_mbuf *mo;
367 
368 			tbl = rxq->frag_tbl;
369 			dr = &qconf->death_row;
370 
371 			/* prepare mbuf: setup l2_len/l3_len. */
372 			m->l2_len = sizeof(*eth_hdr);
373 			m->l3_len = sizeof(*ip_hdr);
374 
375 			/* process this fragment. */
376 			mo = rte_ipv4_frag_reassemble_packet(tbl, dr, m, tms, ip_hdr);
377 			if (mo == NULL)
378 				/* no packet to send out. */
379 				return;
380 
381 			/* we have our packet reassembled. */
382 			if (mo != m) {
383 				m = mo;
384 				eth_hdr = rte_pktmbuf_mtod(m,
385 					struct ether_hdr *);
386 				ip_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
387 			}
388 		}
389 		ip_dst = rte_be_to_cpu_32(ip_hdr->dst_addr);
390 
391 		/* Find destination port */
392 		if (rte_lpm_lookup(rxq->lpm, ip_dst, &next_hop) == 0 &&
393 				(enabled_port_mask & 1 << next_hop) != 0) {
394 			dst_port = next_hop;
395 		}
396 
397 		eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv4);
398 	} else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
399 		/* if packet is IPv6 */
400 		struct ipv6_extension_fragment *frag_hdr;
401 		struct ipv6_hdr *ip_hdr;
402 
403 		ip_hdr = (struct ipv6_hdr *)(eth_hdr + 1);
404 
405 		frag_hdr = rte_ipv6_frag_get_ipv6_fragment_header(ip_hdr);
406 
407 		if (frag_hdr != NULL) {
408 			struct rte_mbuf *mo;
409 
410 			tbl = rxq->frag_tbl;
411 			dr  = &qconf->death_row;
412 
413 			/* prepare mbuf: setup l2_len/l3_len. */
414 			m->l2_len = sizeof(*eth_hdr);
415 			m->l3_len = sizeof(*ip_hdr) + sizeof(*frag_hdr);
416 
417 			mo = rte_ipv6_frag_reassemble_packet(tbl, dr, m, tms, ip_hdr, frag_hdr);
418 			if (mo == NULL)
419 				return;
420 
421 			if (mo != m) {
422 				m = mo;
423 				eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
424 				ip_hdr = (struct ipv6_hdr *)(eth_hdr + 1);
425 			}
426 		}
427 
428 		/* Find destination port */
429 		if (rte_lpm6_lookup(rxq->lpm6, ip_hdr->dst_addr,
430 						&next_hop) == 0 &&
431 				(enabled_port_mask & 1 << next_hop) != 0) {
432 			dst_port = next_hop;
433 		}
434 
435 		eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv6);
436 	}
437 	/* if packet wasn't IPv4 or IPv6, it's forwarded to the port it came from */
438 
439 	/* 02:00:00:00:00:xx */
440 	d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
441 	*((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)dst_port << 40);
442 
443 	/* src addr */
444 	ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
445 
446 	send_single_packet(m, dst_port);
447 }
448 
449 /* main processing loop */
450 static int
451 main_loop(__attribute__((unused)) void *dummy)
452 {
453 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
454 	unsigned lcore_id;
455 	uint64_t diff_tsc, cur_tsc, prev_tsc;
456 	int i, j, nb_rx;
457 	uint16_t portid;
458 	struct lcore_queue_conf *qconf;
459 	const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
460 
461 	prev_tsc = 0;
462 
463 	lcore_id = rte_lcore_id();
464 	qconf = &lcore_queue_conf[lcore_id];
465 
466 	if (qconf->n_rx_queue == 0) {
467 		RTE_LOG(INFO, IP_RSMBL, "lcore %u has nothing to do\n", lcore_id);
468 		return 0;
469 	}
470 
471 	RTE_LOG(INFO, IP_RSMBL, "entering main loop on lcore %u\n", lcore_id);
472 
473 	for (i = 0; i < qconf->n_rx_queue; i++) {
474 
475 		portid = qconf->rx_queue_list[i].portid;
476 		RTE_LOG(INFO, IP_RSMBL, " -- lcoreid=%u portid=%u\n", lcore_id,
477 			portid);
478 	}
479 
480 	while (1) {
481 
482 		cur_tsc = rte_rdtsc();
483 
484 		/*
485 		 * TX burst queue drain
486 		 */
487 		diff_tsc = cur_tsc - prev_tsc;
488 		if (unlikely(diff_tsc > drain_tsc)) {
489 
490 			/*
491 			 * This could be optimized (use queueid instead of
492 			 * portid), but it is not called so often
493 			 */
494 			for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
495 				if ((enabled_port_mask & (1 << portid)) != 0)
496 					send_burst(qconf, 1, portid);
497 			}
498 
499 			prev_tsc = cur_tsc;
500 		}
501 
502 		/*
503 		 * Read packet from RX queues
504 		 */
505 		for (i = 0; i < qconf->n_rx_queue; ++i) {
506 
507 			portid = qconf->rx_queue_list[i].portid;
508 
509 			nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
510 				MAX_PKT_BURST);
511 
512 			/* Prefetch first packets */
513 			for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
514 				rte_prefetch0(rte_pktmbuf_mtod(
515 						pkts_burst[j], void *));
516 			}
517 
518 			/* Prefetch and forward already prefetched packets */
519 			for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
520 				rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
521 					j + PREFETCH_OFFSET], void *));
522 				reassemble(pkts_burst[j], portid,
523 					i, qconf, cur_tsc);
524 			}
525 
526 			/* Forward remaining prefetched packets */
527 			for (; j < nb_rx; j++) {
528 				reassemble(pkts_burst[j], portid,
529 					i, qconf, cur_tsc);
530 			}
531 
532 			rte_ip_frag_free_death_row(&qconf->death_row,
533 				PREFETCH_OFFSET);
534 		}
535 	}
536 }
537 
538 /* display usage */
539 static void
540 print_usage(const char *prgname)
541 {
542 	printf("%s [EAL options] -- -p PORTMASK [-q NQ]"
543 		"  [--max-pkt-len PKTLEN]"
544 		"  [--maxflows=<flows>]  [--flowttl=<ttl>[(s|ms)]]\n"
545 		"  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
546 		"  -q NQ: number of RX queues per lcore\n"
547 		"  --maxflows=<flows>: optional, maximum number of flows "
548 		"supported\n"
549 		"  --flowttl=<ttl>[(s|ms)]: optional, maximum TTL for each "
550 		"flow\n",
551 		prgname);
552 }
553 
554 static uint32_t
555 parse_flow_num(const char *str, uint32_t min, uint32_t max, uint32_t *val)
556 {
557 	char *end;
558 	uint64_t v;
559 
560 	/* parse decimal string */
561 	errno = 0;
562 	v = strtoul(str, &end, 10);
563 	if (errno != 0 || *end != '\0')
564 		return -EINVAL;
565 
566 	if (v < min || v > max)
567 		return -EINVAL;
568 
569 	*val = (uint32_t)v;
570 	return 0;
571 }
572 
573 static int
574 parse_flow_ttl(const char *str, uint32_t min, uint32_t max, uint32_t *val)
575 {
576 	char *end;
577 	uint64_t v;
578 
579 	static const char frmt_sec[] = "s";
580 	static const char frmt_msec[] = "ms";
581 
582 	/* parse decimal string */
583 	errno = 0;
584 	v = strtoul(str, &end, 10);
585 	if (errno != 0)
586 		return -EINVAL;
587 
588 	if (*end != '\0') {
589 		if (strncmp(frmt_sec, end, sizeof(frmt_sec)) == 0)
590 			v *= MS_PER_S;
591 		else if (strncmp(frmt_msec, end, sizeof (frmt_msec)) != 0)
592 			return -EINVAL;
593 	}
594 
595 	if (v < min || v > max)
596 		return -EINVAL;
597 
598 	*val = (uint32_t)v;
599 	return 0;
600 }
601 
602 static int
603 parse_portmask(const char *portmask)
604 {
605 	char *end = NULL;
606 	unsigned long pm;
607 
608 	/* parse hexadecimal string */
609 	pm = strtoul(portmask, &end, 16);
610 	if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
611 		return -1;
612 
613 	if (pm == 0)
614 		return -1;
615 
616 	return pm;
617 }
618 
619 static int
620 parse_nqueue(const char *q_arg)
621 {
622 	char *end = NULL;
623 	unsigned long n;
624 
625 	printf("%p\n", q_arg);
626 
627 	/* parse hexadecimal string */
628 	n = strtoul(q_arg, &end, 10);
629 	if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
630 		return -1;
631 	if (n == 0)
632 		return -1;
633 	if (n >= MAX_RX_QUEUE_PER_LCORE)
634 		return -1;
635 
636 	return n;
637 }
638 
639 /* Parse the argument given in the command line of the application */
640 static int
641 parse_args(int argc, char **argv)
642 {
643 	int opt, ret;
644 	char **argvopt;
645 	int option_index;
646 	char *prgname = argv[0];
647 	static struct option lgopts[] = {
648 		{"max-pkt-len", 1, 0, 0},
649 		{"maxflows", 1, 0, 0},
650 		{"flowttl", 1, 0, 0},
651 		{NULL, 0, 0, 0}
652 	};
653 
654 	argvopt = argv;
655 
656 	while ((opt = getopt_long(argc, argvopt, "p:q:",
657 				lgopts, &option_index)) != EOF) {
658 
659 		switch (opt) {
660 		/* portmask */
661 		case 'p':
662 			enabled_port_mask = parse_portmask(optarg);
663 			if (enabled_port_mask == 0) {
664 				printf("invalid portmask\n");
665 				print_usage(prgname);
666 				return -1;
667 			}
668 			break;
669 
670 		/* nqueue */
671 		case 'q':
672 			rx_queue_per_lcore = parse_nqueue(optarg);
673 			if (rx_queue_per_lcore < 0) {
674 				printf("invalid queue number\n");
675 				print_usage(prgname);
676 				return -1;
677 			}
678 			break;
679 
680 		/* long options */
681 		case 0:
682 			if (!strncmp(lgopts[option_index].name,
683 					"maxflows", 8)) {
684 				if ((ret = parse_flow_num(optarg, MIN_FLOW_NUM,
685 						MAX_FLOW_NUM,
686 						&max_flow_num)) != 0) {
687 					printf("invalid value: \"%s\" for "
688 						"parameter %s\n",
689 						optarg,
690 						lgopts[option_index].name);
691 					print_usage(prgname);
692 					return ret;
693 				}
694 			}
695 
696 			if (!strncmp(lgopts[option_index].name, "flowttl", 7)) {
697 				if ((ret = parse_flow_ttl(optarg, MIN_FLOW_TTL,
698 						MAX_FLOW_TTL,
699 						&max_flow_ttl)) != 0) {
700 					printf("invalid value: \"%s\" for "
701 						"parameter %s\n",
702 						optarg,
703 						lgopts[option_index].name);
704 					print_usage(prgname);
705 					return ret;
706 				}
707 			}
708 
709 			break;
710 
711 		default:
712 			print_usage(prgname);
713 			return -1;
714 		}
715 	}
716 
717 	if (optind >= 0)
718 		argv[optind-1] = prgname;
719 
720 	ret = optind-1;
721 	optind = 1; /* reset getopt lib */
722 	return ret;
723 }
724 
725 static void
726 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
727 {
728 	char buf[ETHER_ADDR_FMT_SIZE];
729 	ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
730 	printf("%s%s", name, buf);
731 }
732 
733 /* Check the link status of all ports in up to 9s, and print them finally */
734 static void
735 check_all_ports_link_status(uint16_t port_num, uint32_t port_mask)
736 {
737 #define CHECK_INTERVAL 100 /* 100ms */
738 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
739 	uint16_t portid;
740 	uint8_t count, all_ports_up, print_flag = 0;
741 	struct rte_eth_link link;
742 
743 	printf("\nChecking link status");
744 	fflush(stdout);
745 	for (count = 0; count <= MAX_CHECK_TIME; count++) {
746 		all_ports_up = 1;
747 		for (portid = 0; portid < port_num; portid++) {
748 			if ((port_mask & (1 << portid)) == 0)
749 				continue;
750 			memset(&link, 0, sizeof(link));
751 			rte_eth_link_get_nowait(portid, &link);
752 			/* print link status if flag set */
753 			if (print_flag == 1) {
754 				if (link.link_status)
755 					printf(
756 					"Port%d Link Up. Speed %u Mbps - %s\n",
757 						portid, link.link_speed,
758 				(link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
759 					("full-duplex") : ("half-duplex\n"));
760 				else
761 					printf("Port %d Link Down\n", portid);
762 				continue;
763 			}
764 			/* clear all_ports_up flag if any link down */
765 			if (link.link_status == ETH_LINK_DOWN) {
766 				all_ports_up = 0;
767 				break;
768 			}
769 		}
770 		/* after finally printing all link status, get out */
771 		if (print_flag == 1)
772 			break;
773 
774 		if (all_ports_up == 0) {
775 			printf(".");
776 			fflush(stdout);
777 			rte_delay_ms(CHECK_INTERVAL);
778 		}
779 
780 		/* set the print_flag if all ports up or timeout */
781 		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
782 			print_flag = 1;
783 			printf("\ndone\n");
784 		}
785 	}
786 }
787 
788 static int
789 init_routing_table(void)
790 {
791 	struct rte_lpm *lpm;
792 	struct rte_lpm6 *lpm6;
793 	int socket, ret;
794 	unsigned i;
795 
796 	for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) {
797 		if (socket_lpm[socket]) {
798 			lpm = socket_lpm[socket];
799 			/* populate the LPM table */
800 			for (i = 0; i < RTE_DIM(l3fwd_ipv4_route_array); i++) {
801 				ret = rte_lpm_add(lpm,
802 					l3fwd_ipv4_route_array[i].ip,
803 					l3fwd_ipv4_route_array[i].depth,
804 					l3fwd_ipv4_route_array[i].if_out);
805 
806 				if (ret < 0) {
807 					RTE_LOG(ERR, IP_RSMBL, "Unable to add entry %i to the l3fwd "
808 						"LPM table\n", i);
809 					return -1;
810 				}
811 
812 				RTE_LOG(INFO, IP_RSMBL, "Socket %i: adding route " IPv4_BYTES_FMT
813 						"/%d (port %d)\n",
814 					socket,
815 					IPv4_BYTES(l3fwd_ipv4_route_array[i].ip),
816 					l3fwd_ipv4_route_array[i].depth,
817 					l3fwd_ipv4_route_array[i].if_out);
818 			}
819 		}
820 
821 		if (socket_lpm6[socket]) {
822 			lpm6 = socket_lpm6[socket];
823 			/* populate the LPM6 table */
824 			for (i = 0; i < RTE_DIM(l3fwd_ipv6_route_array); i++) {
825 				ret = rte_lpm6_add(lpm6,
826 					l3fwd_ipv6_route_array[i].ip,
827 					l3fwd_ipv6_route_array[i].depth,
828 					l3fwd_ipv6_route_array[i].if_out);
829 
830 				if (ret < 0) {
831 					RTE_LOG(ERR, IP_RSMBL, "Unable to add entry %i to the l3fwd "
832 						"LPM6 table\n", i);
833 					return -1;
834 				}
835 
836 				RTE_LOG(INFO, IP_RSMBL, "Socket %i: adding route " IPv6_BYTES_FMT
837 						"/%d (port %d)\n",
838 					socket,
839 					IPv6_BYTES(l3fwd_ipv6_route_array[i].ip),
840 					l3fwd_ipv6_route_array[i].depth,
841 					l3fwd_ipv6_route_array[i].if_out);
842 			}
843 		}
844 	}
845 	return 0;
846 }
847 
848 static int
849 setup_port_tbl(struct lcore_queue_conf *qconf, uint32_t lcore, int socket,
850 	uint32_t port)
851 {
852 	struct mbuf_table *mtb;
853 	uint32_t n;
854 	size_t sz;
855 
856 	n = RTE_MAX(max_flow_num, 2UL * MAX_PKT_BURST);
857 	sz = sizeof (*mtb) + sizeof (mtb->m_table[0]) *  n;
858 
859 	if ((mtb = rte_zmalloc_socket(__func__, sz, RTE_CACHE_LINE_SIZE,
860 			socket)) == NULL) {
861 		RTE_LOG(ERR, IP_RSMBL, "%s() for lcore: %u, port: %u "
862 			"failed to allocate %zu bytes\n",
863 			__func__, lcore, port, sz);
864 		return -1;
865 	}
866 
867 	mtb->len = n;
868 	qconf->tx_mbufs[port] = mtb;
869 
870 	return 0;
871 }
872 
873 static int
874 setup_queue_tbl(struct rx_queue *rxq, uint32_t lcore, uint32_t queue)
875 {
876 	int socket;
877 	uint32_t nb_mbuf;
878 	uint64_t frag_cycles;
879 	char buf[RTE_MEMPOOL_NAMESIZE];
880 
881 	socket = rte_lcore_to_socket_id(lcore);
882 	if (socket == SOCKET_ID_ANY)
883 		socket = 0;
884 
885 	frag_cycles = (rte_get_tsc_hz() + MS_PER_S - 1) / MS_PER_S *
886 		max_flow_ttl;
887 
888 	if ((rxq->frag_tbl = rte_ip_frag_table_create(max_flow_num,
889 			IP_FRAG_TBL_BUCKET_ENTRIES, max_flow_num, frag_cycles,
890 			socket)) == NULL) {
891 		RTE_LOG(ERR, IP_RSMBL, "ip_frag_tbl_create(%u) on "
892 			"lcore: %u for queue: %u failed\n",
893 			max_flow_num, lcore, queue);
894 		return -1;
895 	}
896 
897 	/*
898 	 * At any given moment up to <max_flow_num * (MAX_FRAG_NUM)>
899 	 * mbufs could be stored int the fragment table.
900 	 * Plus, each TX queue can hold up to <max_flow_num> packets.
901 	 */
902 
903 	nb_mbuf = RTE_MAX(max_flow_num, 2UL * MAX_PKT_BURST) * MAX_FRAG_NUM;
904 	nb_mbuf *= (port_conf.rxmode.max_rx_pkt_len + BUF_SIZE - 1) / BUF_SIZE;
905 	nb_mbuf *= 2; /* ipv4 and ipv6 */
906 	nb_mbuf += nb_rxd + nb_txd;
907 
908 	nb_mbuf = RTE_MAX(nb_mbuf, (uint32_t)NB_MBUF);
909 
910 	snprintf(buf, sizeof(buf), "mbuf_pool_%u_%u", lcore, queue);
911 
912 	if ((rxq->pool = rte_mempool_create(buf, nb_mbuf, MBUF_SIZE, 0,
913 			sizeof(struct rte_pktmbuf_pool_private),
914 			rte_pktmbuf_pool_init, NULL, rte_pktmbuf_init, NULL,
915 			socket, MEMPOOL_F_SP_PUT | MEMPOOL_F_SC_GET)) == NULL) {
916 		RTE_LOG(ERR, IP_RSMBL, "mempool_create(%s) failed", buf);
917 		return -1;
918 	}
919 
920 	return 0;
921 }
922 
923 static int
924 init_mem(void)
925 {
926 	char buf[PATH_MAX];
927 	struct rte_lpm *lpm;
928 	struct rte_lpm6 *lpm6;
929 	struct rte_lpm_config lpm_config;
930 	int socket;
931 	unsigned lcore_id;
932 
933 	/* traverse through lcores and initialize structures on each socket */
934 
935 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
936 
937 		if (rte_lcore_is_enabled(lcore_id) == 0)
938 			continue;
939 
940 		socket = rte_lcore_to_socket_id(lcore_id);
941 
942 		if (socket == SOCKET_ID_ANY)
943 			socket = 0;
944 
945 		if (socket_lpm[socket] == NULL) {
946 			RTE_LOG(INFO, IP_RSMBL, "Creating LPM table on socket %i\n", socket);
947 			snprintf(buf, sizeof(buf), "IP_RSMBL_LPM_%i", socket);
948 
949 			lpm_config.max_rules = LPM_MAX_RULES;
950 			lpm_config.number_tbl8s = 256;
951 			lpm_config.flags = 0;
952 
953 			lpm = rte_lpm_create(buf, socket, &lpm_config);
954 			if (lpm == NULL) {
955 				RTE_LOG(ERR, IP_RSMBL, "Cannot create LPM table\n");
956 				return -1;
957 			}
958 			socket_lpm[socket] = lpm;
959 		}
960 
961 		if (socket_lpm6[socket] == NULL) {
962 			RTE_LOG(INFO, IP_RSMBL, "Creating LPM6 table on socket %i\n", socket);
963 			snprintf(buf, sizeof(buf), "IP_RSMBL_LPM_%i", socket);
964 
965 			lpm6 = rte_lpm6_create(buf, socket, &lpm6_config);
966 			if (lpm6 == NULL) {
967 				RTE_LOG(ERR, IP_RSMBL, "Cannot create LPM table\n");
968 				return -1;
969 			}
970 			socket_lpm6[socket] = lpm6;
971 		}
972 	}
973 
974 	return 0;
975 }
976 
977 static void
978 queue_dump_stat(void)
979 {
980 	uint32_t i, lcore;
981 	const struct lcore_queue_conf *qconf;
982 
983 	for (lcore = 0; lcore < RTE_MAX_LCORE; lcore++) {
984 		if (rte_lcore_is_enabled(lcore) == 0)
985 			continue;
986 
987 		qconf = &lcore_queue_conf[lcore];
988 		for (i = 0; i < qconf->n_rx_queue; i++) {
989 
990 			fprintf(stdout, " -- lcoreid=%u portid=%u "
991 				"frag tbl stat:\n",
992 				lcore,  qconf->rx_queue_list[i].portid);
993 			rte_ip_frag_table_statistics_dump(stdout,
994 					qconf->rx_queue_list[i].frag_tbl);
995 			fprintf(stdout, "TX bursts:\t%" PRIu64 "\n"
996 				"TX packets _queued:\t%" PRIu64 "\n"
997 				"TX packets dropped:\t%" PRIu64 "\n"
998 				"TX packets send:\t%" PRIu64 "\n",
999 				qconf->tx_stat.call,
1000 				qconf->tx_stat.queue,
1001 				qconf->tx_stat.drop,
1002 				qconf->tx_stat.send);
1003 		}
1004 	}
1005 }
1006 
1007 static void
1008 signal_handler(int signum)
1009 {
1010 	queue_dump_stat();
1011 	if (signum != SIGUSR1)
1012 		rte_exit(0, "received signal: %d, exiting\n", signum);
1013 }
1014 
1015 int
1016 main(int argc, char **argv)
1017 {
1018 	struct lcore_queue_conf *qconf;
1019 	struct rte_eth_dev_info dev_info;
1020 	struct rte_eth_txconf *txconf;
1021 	struct rx_queue *rxq;
1022 	int ret, socket;
1023 	unsigned nb_ports;
1024 	uint16_t queueid;
1025 	unsigned lcore_id = 0, rx_lcore_id = 0;
1026 	uint32_t n_tx_queue, nb_lcores;
1027 	uint16_t portid;
1028 
1029 	/* init EAL */
1030 	ret = rte_eal_init(argc, argv);
1031 	if (ret < 0)
1032 		rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1033 	argc -= ret;
1034 	argv += ret;
1035 
1036 	/* parse application arguments (after the EAL ones) */
1037 	ret = parse_args(argc, argv);
1038 	if (ret < 0)
1039 		rte_exit(EXIT_FAILURE, "Invalid IP reassembly parameters\n");
1040 
1041 	nb_ports = rte_eth_dev_count();
1042 	if (nb_ports == 0)
1043 		rte_exit(EXIT_FAILURE, "No ports found!\n");
1044 
1045 	nb_lcores = rte_lcore_count();
1046 
1047 	/* initialize structures (mempools, lpm etc.) */
1048 	if (init_mem() < 0)
1049 		rte_panic("Cannot initialize memory structures!\n");
1050 
1051 	/* check if portmask has non-existent ports */
1052 	if (enabled_port_mask & ~(RTE_LEN2MASK(nb_ports, unsigned)))
1053 		rte_exit(EXIT_FAILURE, "Non-existent ports in portmask!\n");
1054 
1055 	/* initialize all ports */
1056 	for (portid = 0; portid < nb_ports; portid++) {
1057 		/* skip ports that are not enabled */
1058 		if ((enabled_port_mask & (1 << portid)) == 0) {
1059 			printf("\nSkipping disabled port %d\n", portid);
1060 			continue;
1061 		}
1062 
1063 		qconf = &lcore_queue_conf[rx_lcore_id];
1064 
1065 		/* limit the frame size to the maximum supported by NIC */
1066 		rte_eth_dev_info_get(portid, &dev_info);
1067 		port_conf.rxmode.max_rx_pkt_len = RTE_MIN(
1068 		    dev_info.max_rx_pktlen, port_conf.rxmode.max_rx_pkt_len);
1069 
1070 		/* get the lcore_id for this port */
1071 		while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
1072 			   qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
1073 
1074 			rx_lcore_id++;
1075 			if (rx_lcore_id >= RTE_MAX_LCORE)
1076 				rte_exit(EXIT_FAILURE, "Not enough cores\n");
1077 
1078 			qconf = &lcore_queue_conf[rx_lcore_id];
1079 		}
1080 
1081 		socket = rte_lcore_to_socket_id(portid);
1082 		if (socket == SOCKET_ID_ANY)
1083 			socket = 0;
1084 
1085 		queueid = qconf->n_rx_queue;
1086 		rxq = &qconf->rx_queue_list[queueid];
1087 		rxq->portid = portid;
1088 		rxq->lpm = socket_lpm[socket];
1089 		rxq->lpm6 = socket_lpm6[socket];
1090 
1091 		ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
1092 						       &nb_txd);
1093 		if (ret < 0)
1094 			rte_exit(EXIT_FAILURE,
1095 				 "Cannot adjust number of descriptors: err=%d, port=%d\n",
1096 				 ret, portid);
1097 
1098 		if (setup_queue_tbl(rxq, rx_lcore_id, queueid) < 0)
1099 			rte_exit(EXIT_FAILURE, "Failed to set up queue table\n");
1100 		qconf->n_rx_queue++;
1101 
1102 		/* init port */
1103 		printf("Initializing port %d ... ", portid );
1104 		fflush(stdout);
1105 
1106 		n_tx_queue = nb_lcores;
1107 		if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
1108 			n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1109 		ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
1110 					    &port_conf);
1111 		if (ret < 0) {
1112 			printf("\n");
1113 			rte_exit(EXIT_FAILURE, "Cannot configure device: "
1114 				"err=%d, port=%d\n",
1115 				ret, portid);
1116 		}
1117 
1118 		/* init one RX queue */
1119 		ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
1120 					     socket, NULL,
1121 					     rxq->pool);
1122 		if (ret < 0) {
1123 			printf("\n");
1124 			rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: "
1125 				"err=%d, port=%d\n",
1126 				ret, portid);
1127 		}
1128 
1129 		rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1130 		print_ethaddr(" Address:", &ports_eth_addr[portid]);
1131 		printf("\n");
1132 
1133 		/* init one TX queue per couple (lcore,port) */
1134 		queueid = 0;
1135 		for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1136 			if (rte_lcore_is_enabled(lcore_id) == 0)
1137 				continue;
1138 
1139 			socket = (int) rte_lcore_to_socket_id(lcore_id);
1140 
1141 			printf("txq=%u,%d,%d ", lcore_id, queueid, socket);
1142 			fflush(stdout);
1143 
1144 			txconf = &dev_info.default_txconf;
1145 			txconf->txq_flags = 0;
1146 
1147 			ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1148 					socket, txconf);
1149 			if (ret < 0)
1150 				rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
1151 					"port=%d\n", ret, portid);
1152 
1153 			qconf = &lcore_queue_conf[lcore_id];
1154 			qconf->tx_queue_id[portid] = queueid;
1155 			setup_port_tbl(qconf, lcore_id, socket, portid);
1156 			queueid++;
1157 		}
1158 		printf("\n");
1159 	}
1160 
1161 	printf("\n");
1162 
1163 	/* start ports */
1164 	for (portid = 0; portid < nb_ports; portid++) {
1165 		if ((enabled_port_mask & (1 << portid)) == 0) {
1166 			continue;
1167 		}
1168 		/* Start device */
1169 		ret = rte_eth_dev_start(portid);
1170 		if (ret < 0)
1171 			rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1172 				ret, portid);
1173 
1174 		rte_eth_promiscuous_enable(portid);
1175 	}
1176 
1177 	if (init_routing_table() < 0)
1178 		rte_exit(EXIT_FAILURE, "Cannot init routing table\n");
1179 
1180 	check_all_ports_link_status(nb_ports, enabled_port_mask);
1181 
1182 	signal(SIGUSR1, signal_handler);
1183 	signal(SIGTERM, signal_handler);
1184 	signal(SIGINT, signal_handler);
1185 
1186 	/* launch per-lcore init on every lcore */
1187 	rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1188 	RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1189 		if (rte_eal_wait_lcore(lcore_id) < 0)
1190 			return -1;
1191 	}
1192 
1193 	return 0;
1194 }
1195