xref: /dpdk/examples/ip_reassembly/main.c (revision fc1f2750a3ec6da919e3c86e59d56f34ec97154b)
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_tailq.h>
54 #include <rte_eal.h>
55 #include <rte_per_lcore.h>
56 #include <rte_launch.h>
57 #include <rte_atomic.h>
58 #include <rte_cycles.h>
59 #include <rte_prefetch.h>
60 #include <rte_lcore.h>
61 #include <rte_per_lcore.h>
62 #include <rte_branch_prediction.h>
63 #include <rte_interrupts.h>
64 #include <rte_pci.h>
65 #include <rte_random.h>
66 #include <rte_debug.h>
67 #include <rte_ether.h>
68 #include <rte_ethdev.h>
69 #include <rte_ring.h>
70 #include <rte_mempool.h>
71 #include <rte_mbuf.h>
72 #include <rte_malloc.h>
73 #include <rte_ip.h>
74 #include <rte_tcp.h>
75 #include <rte_udp.h>
76 #include <rte_string_fns.h>
77 #include <rte_lpm.h>
78 #include <rte_lpm6.h>
79 
80 #include <rte_ip_frag.h>
81 
82 #include "main.h"
83 
84 #define MAX_PKT_BURST 32
85 
86 
87 #define RTE_LOGTYPE_IP_RSMBL RTE_LOGTYPE_USER1
88 
89 #define MAX_JUMBO_PKT_LEN  9600
90 
91 #define	BUF_SIZE	2048
92 #define MBUF_SIZE	\
93 	(BUF_SIZE + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
94 
95 #define NB_MBUF 8192
96 
97 /* allow max jumbo frame 9.5 KB */
98 #define JUMBO_FRAME_MAX_SIZE	0x2600
99 
100 #define	MAX_FLOW_NUM	UINT16_MAX
101 #define	MIN_FLOW_NUM	1
102 #define	DEF_FLOW_NUM	0x1000
103 
104 /* TTL numbers are in ms. */
105 #define	MAX_FLOW_TTL	(3600 * MS_PER_S)
106 #define	MIN_FLOW_TTL	1
107 #define	DEF_FLOW_TTL	MS_PER_S
108 
109 #define MAX_FRAG_NUM RTE_LIBRTE_IP_FRAG_MAX_FRAG
110 
111 /* Should be power of two. */
112 #define	IP_FRAG_TBL_BUCKET_ENTRIES	16
113 
114 static uint32_t max_flow_num = DEF_FLOW_NUM;
115 static uint32_t max_flow_ttl = DEF_FLOW_TTL;
116 
117 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
118 
119 #define NB_SOCKETS 8
120 
121 /* Configure how many packets ahead to prefetch, when reading packets */
122 #define PREFETCH_OFFSET	3
123 
124 /*
125  * Configurable number of RX/TX ring descriptors
126  */
127 #define RTE_TEST_RX_DESC_DEFAULT 128
128 #define RTE_TEST_TX_DESC_DEFAULT 512
129 
130 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
131 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
132 
133 /* ethernet addresses of ports */
134 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
135 
136 #ifndef IPv4_BYTES
137 #define IPv4_BYTES_FMT "%" PRIu8 ".%" PRIu8 ".%" PRIu8 ".%" PRIu8
138 #define IPv4_BYTES(addr) \
139 		(uint8_t) (((addr) >> 24) & 0xFF),\
140 		(uint8_t) (((addr) >> 16) & 0xFF),\
141 		(uint8_t) (((addr) >> 8) & 0xFF),\
142 		(uint8_t) ((addr) & 0xFF)
143 #endif
144 
145 #ifndef IPv6_BYTES
146 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
147                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
148 #define IPv6_BYTES(addr) \
149 	addr[0],  addr[1], addr[2],  addr[3], \
150 	addr[4],  addr[5], addr[6],  addr[7], \
151 	addr[8],  addr[9], addr[10], addr[11],\
152 	addr[12], addr[13],addr[14], addr[15]
153 #endif
154 
155 #define IPV6_ADDR_LEN 16
156 
157 /* mask of enabled ports */
158 static uint32_t enabled_port_mask = 0;
159 
160 static int rx_queue_per_lcore = 1;
161 
162 struct mbuf_table {
163 	uint32_t len;
164 	uint32_t head;
165 	uint32_t tail;
166 	struct rte_mbuf *m_table[0];
167 };
168 
169 struct rx_queue {
170 	struct rte_ip_frag_tbl *frag_tbl;
171 	struct rte_mempool *pool;
172 	struct rte_lpm *lpm;
173 	struct rte_lpm6 *lpm6;
174 	uint8_t portid;
175 };
176 
177 struct tx_lcore_stat {
178 	uint64_t call;
179 	uint64_t drop;
180 	uint64_t queue;
181 	uint64_t send;
182 };
183 
184 #define MAX_RX_QUEUE_PER_LCORE 16
185 #define MAX_TX_QUEUE_PER_PORT 16
186 #define MAX_RX_QUEUE_PER_PORT 128
187 
188 struct lcore_queue_conf {
189 	uint16_t n_rx_queue;
190 	struct rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
191 	uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
192 	struct rte_ip_frag_death_row death_row;
193 	struct mbuf_table *tx_mbufs[RTE_MAX_ETHPORTS];
194 	struct tx_lcore_stat tx_stat;
195 } __rte_cache_aligned;
196 static struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
197 
198 static struct rte_eth_conf port_conf = {
199 	.rxmode = {
200 		.mq_mode        = ETH_MQ_RX_RSS,
201 		.max_rx_pkt_len = JUMBO_FRAME_MAX_SIZE,
202 		.split_hdr_size = 0,
203 		.header_split   = 0, /**< Header Split disabled */
204 		.hw_ip_checksum = 1, /**< IP checksum offload enabled */
205 		.hw_vlan_filter = 0, /**< VLAN filtering disabled */
206 		.jumbo_frame    = 1, /**< Jumbo Frame Support disabled */
207 		.hw_strip_crc   = 0, /**< CRC stripped by hardware */
208 	},
209 	.rx_adv_conf = {
210 			.rss_conf = {
211 				.rss_key = NULL,
212 				.rss_hf = ETH_RSS_IP,
213 		},
214 	},
215 	.txmode = {
216 		.mq_mode = ETH_MQ_TX_NONE,
217 	},
218 };
219 
220 /*
221  * IPv4 forwarding table
222  */
223 struct l3fwd_ipv4_route {
224 	uint32_t ip;
225 	uint8_t  depth;
226 	uint8_t  if_out;
227 };
228 
229 struct l3fwd_ipv4_route l3fwd_ipv4_route_array[] = {
230 		{IPv4(100,10,0,0), 16, 0},
231 		{IPv4(100,20,0,0), 16, 1},
232 		{IPv4(100,30,0,0), 16, 2},
233 		{IPv4(100,40,0,0), 16, 3},
234 		{IPv4(100,50,0,0), 16, 4},
235 		{IPv4(100,60,0,0), 16, 5},
236 		{IPv4(100,70,0,0), 16, 6},
237 		{IPv4(100,80,0,0), 16, 7},
238 };
239 
240 /*
241  * IPv6 forwarding table
242  */
243 
244 struct l3fwd_ipv6_route {
245 	uint8_t ip[IPV6_ADDR_LEN];
246 	uint8_t depth;
247 	uint8_t if_out;
248 };
249 
250 static struct l3fwd_ipv6_route l3fwd_ipv6_route_array[] = {
251 	{{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
252 	{{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
253 	{{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
254 	{{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
255 	{{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
256 	{{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
257 	{{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
258 	{{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
259 };
260 
261 #define LPM_MAX_RULES         1024
262 #define LPM6_MAX_RULES         1024
263 #define LPM6_NUMBER_TBL8S (1 << 16)
264 
265 struct rte_lpm6_config lpm6_config = {
266 		.max_rules = LPM6_MAX_RULES,
267 		.number_tbl8s = LPM6_NUMBER_TBL8S,
268 		.flags = 0
269 };
270 
271 static struct rte_lpm *socket_lpm[RTE_MAX_NUMA_NODES];
272 static struct rte_lpm6 *socket_lpm6[RTE_MAX_NUMA_NODES];
273 
274 #ifdef RTE_LIBRTE_IP_FRAG_TBL_STAT
275 #define TX_LCORE_STAT_UPDATE(s, f, v)   ((s)->f += (v))
276 #else
277 #define TX_LCORE_STAT_UPDATE(s, f, v)   do {} while (0)
278 #endif /* RTE_LIBRTE_IP_FRAG_TBL_STAT */
279 
280 /*
281  * If number of queued packets reached given threahold, then
282  * send burst of packets on an output interface.
283  */
284 static inline uint32_t
285 send_burst(struct lcore_queue_conf *qconf, uint32_t thresh, uint8_t port)
286 {
287 	uint32_t fill, len, k, n;
288 	struct mbuf_table *txmb;
289 
290 	txmb = qconf->tx_mbufs[port];
291 	len = txmb->len;
292 
293 	if ((int32_t)(fill = txmb->head - txmb->tail) < 0)
294 		fill += len;
295 
296 	if (fill >= thresh) {
297 		n = RTE_MIN(len - txmb->tail, fill);
298 
299 		k = rte_eth_tx_burst(port, qconf->tx_queue_id[port],
300 			txmb->m_table + txmb->tail, (uint16_t)n);
301 
302 		TX_LCORE_STAT_UPDATE(&qconf->tx_stat, call, 1);
303 		TX_LCORE_STAT_UPDATE(&qconf->tx_stat, send, k);
304 
305 		fill -= k;
306 		if ((txmb->tail += k) == len)
307 			txmb->tail = 0;
308 	}
309 
310 	return (fill);
311 }
312 
313 /* Enqueue a single packet, and send burst if queue is filled */
314 static inline int
315 send_single_packet(struct rte_mbuf *m, uint8_t port)
316 {
317 	uint32_t fill, lcore_id, len;
318 	struct lcore_queue_conf *qconf;
319 	struct mbuf_table *txmb;
320 
321 	lcore_id = rte_lcore_id();
322 	qconf = &lcore_queue_conf[lcore_id];
323 
324 	txmb = qconf->tx_mbufs[port];
325 	len = txmb->len;
326 
327 	fill = send_burst(qconf, MAX_PKT_BURST, port);
328 
329 	if (fill == len - 1) {
330 		TX_LCORE_STAT_UPDATE(&qconf->tx_stat, drop, 1);
331 		rte_pktmbuf_free(txmb->m_table[txmb->tail]);
332 		if (++txmb->tail == len)
333 			txmb->tail = 0;
334 	}
335 
336 	TX_LCORE_STAT_UPDATE(&qconf->tx_stat, queue, 1);
337 	txmb->m_table[txmb->head] = m;
338 	if(++txmb->head == len)
339 		txmb->head = 0;
340 
341 	return (0);
342 }
343 
344 static inline void
345 reassemble(struct rte_mbuf *m, uint8_t portid, uint32_t queue,
346 	struct lcore_queue_conf *qconf, uint64_t tms)
347 {
348 	struct ether_hdr *eth_hdr;
349 	struct rte_ip_frag_tbl *tbl;
350 	struct rte_ip_frag_death_row *dr;
351 	struct rx_queue *rxq;
352 	void *d_addr_bytes;
353 	uint8_t next_hop, dst_port;
354 
355 	rxq = &qconf->rx_queue_list[queue];
356 
357 	eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
358 
359 	dst_port = portid;
360 
361 	/* if packet is IPv4 */
362 	if (m->ol_flags & (PKT_RX_IPV4_HDR)) {
363 		struct ipv4_hdr *ip_hdr;
364 		uint32_t ip_dst;
365 
366 		ip_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
367 
368 		 /* if it is a fragmented packet, then try to reassemble. */
369 		if (rte_ipv4_frag_pkt_is_fragmented(ip_hdr)) {
370 			struct rte_mbuf *mo;
371 
372 			tbl = rxq->frag_tbl;
373 			dr = &qconf->death_row;
374 
375 			/* prepare mbuf: setup l2_len/l3_len. */
376 			m->l2_len = sizeof(*eth_hdr);
377 			m->l3_len = sizeof(*ip_hdr);
378 
379 			/* process this fragment. */
380 			mo = rte_ipv4_frag_reassemble_packet(tbl, dr, m, tms, ip_hdr);
381 			if (mo == NULL)
382 				/* no packet to send out. */
383 				return;
384 
385 			/* we have our packet reassembled. */
386 			if (mo != m) {
387 				m = mo;
388 				eth_hdr = rte_pktmbuf_mtod(m,
389 					struct ether_hdr *);
390 				ip_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
391 			}
392 		}
393 		ip_dst = rte_be_to_cpu_32(ip_hdr->dst_addr);
394 
395 		/* Find destination port */
396 		if (rte_lpm_lookup(rxq->lpm, ip_dst, &next_hop) == 0 &&
397 				(enabled_port_mask & 1 << next_hop) != 0) {
398 			dst_port = next_hop;
399 		}
400 
401 		eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv4);
402 	}
403 	/* if packet is IPv6 */
404 	else if (m->ol_flags & (PKT_RX_IPV6_HDR | PKT_RX_IPV6_HDR_EXT)) {
405 		struct ipv6_extension_fragment *frag_hdr;
406 		struct ipv6_hdr *ip_hdr;
407 
408 		ip_hdr = (struct ipv6_hdr *)(eth_hdr + 1);
409 
410 		frag_hdr = rte_ipv6_frag_get_ipv6_fragment_header(ip_hdr);
411 
412 		if (frag_hdr != NULL) {
413 			struct rte_mbuf *mo;
414 
415 			tbl = rxq->frag_tbl;
416 			dr  = &qconf->death_row;
417 
418 			/* prepare mbuf: setup l2_len/l3_len. */
419 			m->l2_len = sizeof(*eth_hdr);
420 			m->l3_len = sizeof(*ip_hdr) + sizeof(*frag_hdr);
421 
422 			mo = rte_ipv6_frag_reassemble_packet(tbl, dr, m, tms, ip_hdr, frag_hdr);
423 			if (mo == NULL)
424 				return;
425 
426 			if (mo != m) {
427 				m = mo;
428 				eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
429 				ip_hdr = (struct ipv6_hdr *)(eth_hdr + 1);
430 			}
431 		}
432 
433 		/* Find destination port */
434 		if (rte_lpm6_lookup(rxq->lpm6, ip_hdr->dst_addr, &next_hop) == 0 &&
435 				(enabled_port_mask & 1 << next_hop) != 0) {
436 			dst_port = next_hop;
437 		}
438 
439 		eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv6);
440 	}
441 	/* if packet wasn't IPv4 or IPv6, it's forwarded to the port it came from */
442 
443 	/* 02:00:00:00:00:xx */
444 	d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
445 	*((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)dst_port << 40);
446 
447 	/* src addr */
448 	ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
449 
450 	send_single_packet(m, dst_port);
451 }
452 
453 /* main processing loop */
454 static int
455 main_loop(__attribute__((unused)) void *dummy)
456 {
457 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
458 	unsigned lcore_id;
459 	uint64_t diff_tsc, cur_tsc, prev_tsc;
460 	int i, j, nb_rx;
461 	uint8_t portid;
462 	struct lcore_queue_conf *qconf;
463 	const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
464 
465 	prev_tsc = 0;
466 
467 	lcore_id = rte_lcore_id();
468 	qconf = &lcore_queue_conf[lcore_id];
469 
470 	if (qconf->n_rx_queue == 0) {
471 		RTE_LOG(INFO, IP_RSMBL, "lcore %u has nothing to do\n", lcore_id);
472 		return 0;
473 	}
474 
475 	RTE_LOG(INFO, IP_RSMBL, "entering main loop on lcore %u\n", lcore_id);
476 
477 	for (i = 0; i < qconf->n_rx_queue; i++) {
478 
479 		portid = qconf->rx_queue_list[i].portid;
480 		RTE_LOG(INFO, IP_RSMBL, " -- lcoreid=%u portid=%hhu\n", lcore_id,
481 			portid);
482 	}
483 
484 	while (1) {
485 
486 		cur_tsc = rte_rdtsc();
487 
488 		/*
489 		 * TX burst queue drain
490 		 */
491 		diff_tsc = cur_tsc - prev_tsc;
492 		if (unlikely(diff_tsc > drain_tsc)) {
493 
494 			/*
495 			 * This could be optimized (use queueid instead of
496 			 * portid), but it is not called so often
497 			 */
498 			for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
499 				if ((enabled_port_mask & (1 << portid)) != 0)
500 					send_burst(qconf, 1, portid);
501 			}
502 
503 			prev_tsc = cur_tsc;
504 		}
505 
506 		/*
507 		 * Read packet from RX queues
508 		 */
509 		for (i = 0; i < qconf->n_rx_queue; ++i) {
510 
511 			portid = qconf->rx_queue_list[i].portid;
512 
513 			nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
514 				MAX_PKT_BURST);
515 
516 			/* Prefetch first packets */
517 			for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
518 				rte_prefetch0(rte_pktmbuf_mtod(
519 						pkts_burst[j], void *));
520 			}
521 
522 			/* Prefetch and forward already prefetched packets */
523 			for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
524 				rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
525 					j + PREFETCH_OFFSET], void *));
526 				reassemble(pkts_burst[j], portid,
527 					i, qconf, cur_tsc);
528 			}
529 
530 			/* Forward remaining prefetched packets */
531 			for (; j < nb_rx; j++) {
532 				reassemble(pkts_burst[j], portid,
533 					i, qconf, cur_tsc);
534 			}
535 
536 			rte_ip_frag_free_death_row(&qconf->death_row,
537 				PREFETCH_OFFSET);
538 		}
539 	}
540 }
541 
542 /* display usage */
543 static void
544 print_usage(const char *prgname)
545 {
546 	printf("%s [EAL options] -- -p PORTMASK [-q NQ]"
547 		"  [--max-pkt-len PKTLEN]"
548 		"  [--maxflows=<flows>]  [--flowttl=<ttl>[(s|ms)]]\n"
549 		"  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
550 		"  -q NQ: number of RX queues per lcore\n"
551 		"  --maxflows=<flows>: optional, maximum number of flows "
552 		"supported\n"
553 		"  --flowttl=<ttl>[(s|ms)]: optional, maximum TTL for each "
554 		"flow\n",
555 		prgname);
556 }
557 
558 static uint32_t
559 parse_flow_num(const char *str, uint32_t min, uint32_t max, uint32_t *val)
560 {
561 	char *end;
562 	uint64_t v;
563 
564 	/* parse decimal string */
565 	errno = 0;
566 	v = strtoul(str, &end, 10);
567 	if (errno != 0 || *end != '\0')
568 		return (-EINVAL);
569 
570 	if (v < min || v > max)
571 		return (-EINVAL);
572 
573 	*val = (uint32_t)v;
574 	return (0);
575 }
576 
577 static int
578 parse_flow_ttl(const char *str, uint32_t min, uint32_t max, uint32_t *val)
579 {
580 	char *end;
581 	uint64_t v;
582 
583 	static const char frmt_sec[] = "s";
584 	static const char frmt_msec[] = "ms";
585 
586 	/* parse decimal string */
587 	errno = 0;
588 	v = strtoul(str, &end, 10);
589 	if (errno != 0)
590 		return (-EINVAL);
591 
592 	if (*end != '\0') {
593 		if (strncmp(frmt_sec, end, sizeof(frmt_sec)) == 0)
594 			v *= MS_PER_S;
595 		else if (strncmp(frmt_msec, end, sizeof (frmt_msec)) != 0)
596 			return (-EINVAL);
597 	}
598 
599 	if (v < min || v > max)
600 		return (-EINVAL);
601 
602 	*val = (uint32_t)v;
603 	return (0);
604 }
605 
606 static int
607 parse_portmask(const char *portmask)
608 {
609 	char *end = NULL;
610 	unsigned long pm;
611 
612 	/* parse hexadecimal string */
613 	pm = strtoul(portmask, &end, 16);
614 	if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
615 		return -1;
616 
617 	if (pm == 0)
618 		return -1;
619 
620 	return pm;
621 }
622 
623 static int
624 parse_nqueue(const char *q_arg)
625 {
626 	char *end = NULL;
627 	unsigned long n;
628 
629 	printf("%p\n", q_arg);
630 
631 	/* parse hexadecimal string */
632 	n = strtoul(q_arg, &end, 10);
633 	if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
634 		return -1;
635 	if (n == 0)
636 		return -1;
637 	if (n >= MAX_RX_QUEUE_PER_LCORE)
638 		return -1;
639 
640 	return n;
641 }
642 
643 /* Parse the argument given in the command line of the application */
644 static int
645 parse_args(int argc, char **argv)
646 {
647 	int opt, ret;
648 	char **argvopt;
649 	int option_index;
650 	char *prgname = argv[0];
651 	static struct option lgopts[] = {
652 		{"max-pkt-len", 1, 0, 0},
653 		{"maxflows", 1, 0, 0},
654 		{"flowttl", 1, 0, 0},
655 		{NULL, 0, 0, 0}
656 	};
657 
658 	argvopt = argv;
659 
660 	while ((opt = getopt_long(argc, argvopt, "p:q:",
661 				lgopts, &option_index)) != EOF) {
662 
663 		switch (opt) {
664 		/* portmask */
665 		case 'p':
666 			enabled_port_mask = parse_portmask(optarg);
667 			if (enabled_port_mask == 0) {
668 				printf("invalid portmask\n");
669 				print_usage(prgname);
670 				return -1;
671 			}
672 			break;
673 
674 		/* nqueue */
675 		case 'q':
676 			rx_queue_per_lcore = parse_nqueue(optarg);
677 			if (rx_queue_per_lcore < 0) {
678 				printf("invalid queue number\n");
679 				print_usage(prgname);
680 				return -1;
681 			}
682 			break;
683 
684 		/* long options */
685 		case 0:
686 			if (!strncmp(lgopts[option_index].name,
687 					"maxflows", 8)) {
688 				if ((ret = parse_flow_num(optarg, MIN_FLOW_NUM,
689 						MAX_FLOW_NUM,
690 						&max_flow_num)) != 0) {
691 					printf("invalid value: \"%s\" for "
692 						"parameter %s\n",
693 						optarg,
694 						lgopts[option_index].name);
695 					print_usage(prgname);
696 					return (ret);
697 				}
698 			}
699 
700 			if (!strncmp(lgopts[option_index].name, "flowttl", 7)) {
701 				if ((ret = parse_flow_ttl(optarg, MIN_FLOW_TTL,
702 						MAX_FLOW_TTL,
703 						&max_flow_ttl)) != 0) {
704 					printf("invalid value: \"%s\" for "
705 						"parameter %s\n",
706 						optarg,
707 						lgopts[option_index].name);
708 					print_usage(prgname);
709 					return (ret);
710 				}
711 			}
712 
713 			break;
714 
715 		default:
716 			print_usage(prgname);
717 			return -1;
718 		}
719 	}
720 
721 	if (optind >= 0)
722 		argv[optind-1] = prgname;
723 
724 	ret = optind-1;
725 	optind = 0; /* reset getopt lib */
726 	return ret;
727 }
728 
729 static void
730 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
731 {
732 	char buf[ETHER_ADDR_FMT_SIZE];
733 	ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
734 	printf("%s%s", name, buf);
735 }
736 
737 /* Check the link status of all ports in up to 9s, and print them finally */
738 static void
739 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
740 {
741 #define CHECK_INTERVAL 100 /* 100ms */
742 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
743 	uint8_t portid, count, all_ports_up, print_flag = 0;
744 	struct rte_eth_link link;
745 
746 	printf("\nChecking link status");
747 	fflush(stdout);
748 	for (count = 0; count <= MAX_CHECK_TIME; count++) {
749 		all_ports_up = 1;
750 		for (portid = 0; portid < port_num; portid++) {
751 			if ((port_mask & (1 << portid)) == 0)
752 				continue;
753 			memset(&link, 0, sizeof(link));
754 			rte_eth_link_get_nowait(portid, &link);
755 			/* print link status if flag set */
756 			if (print_flag == 1) {
757 				if (link.link_status)
758 					printf("Port %d Link Up - speed %u "
759 						"Mbps - %s\n", (uint8_t)portid,
760 						(unsigned)link.link_speed,
761 				(link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
762 					("full-duplex") : ("half-duplex\n"));
763 				else
764 					printf("Port %d Link Down\n",
765 						(uint8_t)portid);
766 				continue;
767 			}
768 			/* clear all_ports_up flag if any link down */
769 			if (link.link_status == 0) {
770 				all_ports_up = 0;
771 				break;
772 			}
773 		}
774 		/* after finally printing all link status, get out */
775 		if (print_flag == 1)
776 			break;
777 
778 		if (all_ports_up == 0) {
779 			printf(".");
780 			fflush(stdout);
781 			rte_delay_ms(CHECK_INTERVAL);
782 		}
783 
784 		/* set the print_flag if all ports up or timeout */
785 		if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
786 			print_flag = 1;
787 			printf("\ndone\n");
788 		}
789 	}
790 }
791 
792 static int
793 init_routing_table(void)
794 {
795 	struct rte_lpm *lpm;
796 	struct rte_lpm6 *lpm6;
797 	int socket, ret;
798 	unsigned i;
799 
800 	for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) {
801 		if (socket_lpm[socket]) {
802 			lpm = socket_lpm[socket];
803 			/* populate the LPM table */
804 			for (i = 0; i < RTE_DIM(l3fwd_ipv4_route_array); i++) {
805 				ret = rte_lpm_add(lpm,
806 					l3fwd_ipv4_route_array[i].ip,
807 					l3fwd_ipv4_route_array[i].depth,
808 					l3fwd_ipv4_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 						"LPM table\n", i);
813 					return -1;
814 				}
815 
816 				RTE_LOG(INFO, IP_RSMBL, "Socket %i: adding route " IPv4_BYTES_FMT
817 						"/%d (port %d)\n",
818 					socket,
819 					IPv4_BYTES(l3fwd_ipv4_route_array[i].ip),
820 					l3fwd_ipv4_route_array[i].depth,
821 					l3fwd_ipv4_route_array[i].if_out);
822 			}
823 		}
824 
825 		if (socket_lpm6[socket]) {
826 			lpm6 = socket_lpm6[socket];
827 			/* populate the LPM6 table */
828 			for (i = 0; i < RTE_DIM(l3fwd_ipv6_route_array); i++) {
829 				ret = rte_lpm6_add(lpm6,
830 					l3fwd_ipv6_route_array[i].ip,
831 					l3fwd_ipv6_route_array[i].depth,
832 					l3fwd_ipv6_route_array[i].if_out);
833 
834 				if (ret < 0) {
835 					RTE_LOG(ERR, IP_RSMBL, "Unable to add entry %i to the l3fwd "
836 						"LPM6 table\n", i);
837 					return -1;
838 				}
839 
840 				RTE_LOG(INFO, IP_RSMBL, "Socket %i: adding route " IPv6_BYTES_FMT
841 						"/%d (port %d)\n",
842 					socket,
843 					IPv6_BYTES(l3fwd_ipv6_route_array[i].ip),
844 					l3fwd_ipv6_route_array[i].depth,
845 					l3fwd_ipv6_route_array[i].if_out);
846 			}
847 		}
848 	}
849 	return 0;
850 }
851 
852 static int
853 setup_port_tbl(struct lcore_queue_conf *qconf, uint32_t lcore, int socket,
854 	uint32_t port)
855 {
856 	struct mbuf_table *mtb;
857 	uint32_t n;
858 	size_t sz;
859 
860 	n = RTE_MAX(max_flow_num, 2UL * MAX_PKT_BURST);
861 	sz = sizeof (*mtb) + sizeof (mtb->m_table[0]) *  n;
862 
863 	if ((mtb = rte_zmalloc_socket(__func__, sz, CACHE_LINE_SIZE,
864 			socket)) == NULL) {
865 		RTE_LOG(ERR, IP_RSMBL, "%s() for lcore: %u, port: %u "
866 			"failed to allocate %zu bytes\n",
867 			__func__, lcore, port, sz);
868 		return -1;
869 	}
870 
871 	mtb->len = n;
872 	qconf->tx_mbufs[port] = mtb;
873 
874 	return 0;
875 }
876 
877 static int
878 setup_queue_tbl(struct rx_queue *rxq, uint32_t lcore, uint32_t queue)
879 {
880 	int socket;
881 	uint32_t nb_mbuf;
882 	uint64_t frag_cycles;
883 	char buf[RTE_MEMPOOL_NAMESIZE];
884 
885 	socket = rte_lcore_to_socket_id(lcore);
886 	if (socket == SOCKET_ID_ANY)
887 		socket = 0;
888 
889 	frag_cycles = (rte_get_tsc_hz() + MS_PER_S - 1) / MS_PER_S *
890 		max_flow_ttl;
891 
892 	if ((rxq->frag_tbl = rte_ip_frag_table_create(max_flow_num,
893 			IP_FRAG_TBL_BUCKET_ENTRIES, max_flow_num, frag_cycles,
894 			socket)) == NULL) {
895 		RTE_LOG(ERR, IP_RSMBL, "ip_frag_tbl_create(%u) on "
896 			"lcore: %u for queue: %u failed\n",
897 			max_flow_num, lcore, queue);
898 		return -1;
899 	}
900 
901 	/*
902 	 * At any given moment up to <max_flow_num * (MAX_FRAG_NUM)>
903 	 * mbufs could be stored int the fragment table.
904 	 * Plus, each TX queue can hold up to <max_flow_num> packets.
905 	 */
906 
907 	nb_mbuf = RTE_MAX(max_flow_num, 2UL * MAX_PKT_BURST) * MAX_FRAG_NUM;
908 	nb_mbuf *= (port_conf.rxmode.max_rx_pkt_len + BUF_SIZE - 1) / BUF_SIZE;
909 	nb_mbuf *= 2; /* ipv4 and ipv6 */
910 	nb_mbuf += RTE_TEST_RX_DESC_DEFAULT + RTE_TEST_TX_DESC_DEFAULT;
911 
912 	nb_mbuf = RTE_MAX(nb_mbuf, (uint32_t)NB_MBUF);
913 
914 	snprintf(buf, sizeof(buf), "mbuf_pool_%u_%u", lcore, queue);
915 
916 	if ((rxq->pool = rte_mempool_create(buf, nb_mbuf, MBUF_SIZE, 0,
917 			sizeof(struct rte_pktmbuf_pool_private),
918 			rte_pktmbuf_pool_init, NULL, rte_pktmbuf_init, NULL,
919 			socket, MEMPOOL_F_SP_PUT | MEMPOOL_F_SC_GET)) == NULL) {
920 		RTE_LOG(ERR, IP_RSMBL, "mempool_create(%s) failed", buf);
921 		return -1;
922 	}
923 
924 	return 0;
925 }
926 
927 static int
928 init_mem(void)
929 {
930 	char buf[PATH_MAX];
931 	struct rte_lpm *lpm;
932 	struct rte_lpm6 *lpm6;
933 	int socket;
934 	unsigned lcore_id;
935 
936 	/* traverse through lcores and initialize structures on each socket */
937 
938 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
939 
940 		if (rte_lcore_is_enabled(lcore_id) == 0)
941 			continue;
942 
943 		socket = rte_lcore_to_socket_id(lcore_id);
944 
945 		if (socket == SOCKET_ID_ANY)
946 			socket = 0;
947 
948 		if (socket_lpm[socket] == NULL) {
949 			RTE_LOG(INFO, IP_RSMBL, "Creating LPM table on socket %i\n", socket);
950 			snprintf(buf, sizeof(buf), "IP_RSMBL_LPM_%i", socket);
951 
952 			lpm = rte_lpm_create(buf, socket, LPM_MAX_RULES, 0);
953 			if (lpm == NULL) {
954 				RTE_LOG(ERR, IP_RSMBL, "Cannot create LPM table\n");
955 				return -1;
956 			}
957 			socket_lpm[socket] = lpm;
958 		}
959 
960 		if (socket_lpm6[socket] == NULL) {
961 			RTE_LOG(INFO, IP_RSMBL, "Creating LPM6 table on socket %i\n", socket);
962 			snprintf(buf, sizeof(buf), "IP_RSMBL_LPM_%i", socket);
963 
964 			lpm6 = rte_lpm6_create("IP_RSMBL_LPM6", socket, &lpm6_config);
965 			if (lpm6 == NULL) {
966 				RTE_LOG(ERR, IP_RSMBL, "Cannot create LPM table\n");
967 				return -1;
968 			}
969 			socket_lpm6[socket] = lpm6;
970 		}
971 	}
972 
973 	return 0;
974 }
975 
976 static void
977 queue_dump_stat(void)
978 {
979 	uint32_t i, lcore;
980 	const struct lcore_queue_conf *qconf;
981 
982 	for (lcore = 0; lcore < RTE_MAX_LCORE; lcore++) {
983 		if (rte_lcore_is_enabled(lcore) == 0)
984 			continue;
985 
986 		qconf = &lcore_queue_conf[lcore];
987 		for (i = 0; i < qconf->n_rx_queue; i++) {
988 
989 			fprintf(stdout, " -- lcoreid=%u portid=%hhu "
990 				"frag tbl stat:\n",
991 				lcore,  qconf->rx_queue_list[i].portid);
992 			rte_ip_frag_table_statistics_dump(stdout,
993 					qconf->rx_queue_list[i].frag_tbl);
994 			fprintf(stdout, "TX bursts:\t%" PRIu64 "\n"
995 				"TX packets _queued:\t%" PRIu64 "\n"
996 				"TX packets dropped:\t%" PRIu64 "\n"
997 				"TX packets send:\t%" PRIu64 "\n",
998 				qconf->tx_stat.call,
999 				qconf->tx_stat.queue,
1000 				qconf->tx_stat.drop,
1001 				qconf->tx_stat.send);
1002 		}
1003 	}
1004 }
1005 
1006 static void
1007 signal_handler(int signum)
1008 {
1009 	queue_dump_stat();
1010 	if (signum != SIGUSR1)
1011 		rte_exit(0, "received signal: %d, exiting\n", signum);
1012 }
1013 
1014 int
1015 MAIN(int argc, char **argv)
1016 {
1017 	struct lcore_queue_conf *qconf;
1018 	struct rte_eth_dev_info dev_info;
1019 	struct rte_eth_txconf *txconf;
1020 	struct rx_queue *rxq;
1021 	int ret, socket;
1022 	unsigned nb_ports;
1023 	uint16_t queueid;
1024 	unsigned lcore_id = 0, rx_lcore_id = 0;
1025 	uint32_t n_tx_queue, nb_lcores;
1026 	uint8_t portid;
1027 
1028 	/* init EAL */
1029 	ret = rte_eal_init(argc, argv);
1030 	if (ret < 0)
1031 		rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1032 	argc -= ret;
1033 	argv += ret;
1034 
1035 	/* parse application arguments (after the EAL ones) */
1036 	ret = parse_args(argc, argv);
1037 	if (ret < 0)
1038 		rte_exit(EXIT_FAILURE, "Invalid IP reassembly parameters\n");
1039 
1040 	nb_ports = rte_eth_dev_count();
1041 	if (nb_ports > RTE_MAX_ETHPORTS)
1042 		nb_ports = RTE_MAX_ETHPORTS;
1043 	else if (nb_ports == 0)
1044 		rte_exit(EXIT_FAILURE, "No ports found!\n");
1045 
1046 	nb_lcores = rte_lcore_count();
1047 
1048 	/* initialize structures (mempools, lpm etc.) */
1049 	if (init_mem() < 0)
1050 		rte_panic("Cannot initialize memory structures!\n");
1051 
1052 	/* check if portmask has non-existent ports */
1053 	if (enabled_port_mask & ~(RTE_LEN2MASK(nb_ports, unsigned)))
1054 		rte_exit(EXIT_FAILURE, "Non-existent ports in portmask!\n");
1055 
1056 	/* initialize all ports */
1057 	for (portid = 0; portid < nb_ports; portid++) {
1058 		/* skip ports that are not enabled */
1059 		if ((enabled_port_mask & (1 << portid)) == 0) {
1060 			printf("\nSkipping disabled port %d\n", portid);
1061 			continue;
1062 		}
1063 
1064 		qconf = &lcore_queue_conf[rx_lcore_id];
1065 
1066 		/* get the lcore_id for this port */
1067 		while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
1068 			   qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
1069 
1070 			rx_lcore_id++;
1071 			if (rx_lcore_id >= RTE_MAX_LCORE)
1072 				rte_exit(EXIT_FAILURE, "Not enough cores\n");
1073 
1074 			qconf = &lcore_queue_conf[rx_lcore_id];
1075 		}
1076 
1077 		socket = rte_lcore_to_socket_id(portid);
1078 		if (socket == SOCKET_ID_ANY)
1079 			socket = 0;
1080 
1081 		queueid = qconf->n_rx_queue;
1082 		rxq = &qconf->rx_queue_list[queueid];
1083 		rxq->portid = portid;
1084 		rxq->lpm = socket_lpm[socket];
1085 		rxq->lpm6 = socket_lpm6[socket];
1086 		if (setup_queue_tbl(rxq, rx_lcore_id, queueid) < 0)
1087 			rte_exit(EXIT_FAILURE, "Failed to set up queue table\n");
1088 		qconf->n_rx_queue++;
1089 
1090 		/* init port */
1091 		printf("Initializing port %d ... ", portid );
1092 		fflush(stdout);
1093 
1094 		n_tx_queue = nb_lcores;
1095 		if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
1096 			n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1097 		ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
1098 					    &port_conf);
1099 		if (ret < 0) {
1100 			printf("\n");
1101 			rte_exit(EXIT_FAILURE, "Cannot configure device: "
1102 				"err=%d, port=%d\n",
1103 				ret, portid);
1104 		}
1105 
1106 		/* init one RX queue */
1107 		ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
1108 					     socket, NULL,
1109 					     rxq->pool);
1110 		if (ret < 0) {
1111 			printf("\n");
1112 			rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: "
1113 				"err=%d, port=%d\n",
1114 				ret, portid);
1115 		}
1116 
1117 		rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1118 		print_ethaddr(" Address:", &ports_eth_addr[portid]);
1119 		printf("\n");
1120 
1121 		/* init one TX queue per couple (lcore,port) */
1122 		queueid = 0;
1123 		for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1124 			if (rte_lcore_is_enabled(lcore_id) == 0)
1125 				continue;
1126 
1127 			socket = (int) rte_lcore_to_socket_id(lcore_id);
1128 
1129 			printf("txq=%u,%d,%d ", lcore_id, queueid, socket);
1130 			fflush(stdout);
1131 
1132 			rte_eth_dev_info_get(portid, &dev_info);
1133 			txconf = &dev_info.default_txconf;
1134 			txconf->txq_flags = 0;
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 	for (portid = 0; portid < nb_ports; 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((uint8_t)nb_ports, 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