mbuf: add namespace to offload flags
[dpdk.git] / examples / ip_reassembly / main.c
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_cycles.h>
26 #include <rte_prefetch.h>
27 #include <rte_lcore.h>
28 #include <rte_per_lcore.h>
29 #include <rte_branch_prediction.h>
30 #include <rte_interrupts.h>
31 #include <rte_random.h>
32 #include <rte_debug.h>
33 #include <rte_ether.h>
34 #include <rte_ethdev.h>
35 #include <rte_mempool.h>
36 #include <rte_mbuf.h>
37 #include <rte_malloc.h>
38 #include <rte_ip.h>
39 #include <rte_tcp.h>
40 #include <rte_udp.h>
41 #include <rte_string_fns.h>
42 #include <rte_lpm.h>
43 #include <rte_lpm6.h>
44
45 #include <rte_ip_frag.h>
46
47 #define MAX_PKT_BURST 32
48
49
50 #define RTE_LOGTYPE_IP_RSMBL RTE_LOGTYPE_USER1
51
52 #define MAX_JUMBO_PKT_LEN  9600
53
54 #define BUF_SIZE        RTE_MBUF_DEFAULT_DATAROOM
55 #define MBUF_DATA_SIZE  RTE_MBUF_DEFAULT_BUF_SIZE
56
57 #define NB_MBUF 8192
58 #define MEMPOOL_CACHE_SIZE 256
59
60 /* allow max jumbo frame 9.5 KB */
61 #define JUMBO_FRAME_MAX_SIZE    0x2600
62
63 #define MAX_FLOW_NUM    UINT16_MAX
64 #define MIN_FLOW_NUM    1
65 #define DEF_FLOW_NUM    0x1000
66
67 /* TTL numbers are in ms. */
68 #define MAX_FLOW_TTL    (3600 * MS_PER_S)
69 #define MIN_FLOW_TTL    1
70 #define DEF_FLOW_TTL    MS_PER_S
71
72 #define MAX_FRAG_NUM RTE_LIBRTE_IP_FRAG_MAX_FRAG
73
74 /* Should be power of two. */
75 #define IP_FRAG_TBL_BUCKET_ENTRIES      16
76
77 static uint32_t max_flow_num = DEF_FLOW_NUM;
78 static uint32_t max_flow_ttl = DEF_FLOW_TTL;
79
80 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
81
82 #define NB_SOCKETS 8
83
84 /* Configure how many packets ahead to prefetch, when reading packets */
85 #define PREFETCH_OFFSET 3
86
87 /*
88  * Configurable number of RX/TX ring descriptors
89  */
90 #define RTE_TEST_RX_DESC_DEFAULT 1024
91 #define RTE_TEST_TX_DESC_DEFAULT 1024
92
93 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
94 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
95
96 /* ethernet addresses of ports */
97 static struct rte_ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
98
99 #ifndef IPv4_BYTES
100 #define IPv4_BYTES_FMT "%" PRIu8 ".%" PRIu8 ".%" PRIu8 ".%" PRIu8
101 #define IPv4_BYTES(addr) \
102                 (uint8_t) (((addr) >> 24) & 0xFF),\
103                 (uint8_t) (((addr) >> 16) & 0xFF),\
104                 (uint8_t) (((addr) >> 8) & 0xFF),\
105                 (uint8_t) ((addr) & 0xFF)
106 #endif
107
108 #ifndef IPv6_BYTES
109 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
110                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
111 #define IPv6_BYTES(addr) \
112         addr[0],  addr[1], addr[2],  addr[3], \
113         addr[4],  addr[5], addr[6],  addr[7], \
114         addr[8],  addr[9], addr[10], addr[11],\
115         addr[12], addr[13],addr[14], addr[15]
116 #endif
117
118 #define IPV6_ADDR_LEN 16
119
120 /* mask of enabled ports */
121 static uint32_t enabled_port_mask = 0;
122
123 static int rx_queue_per_lcore = 1;
124
125 struct mbuf_table {
126         uint32_t len;
127         uint32_t head;
128         uint32_t tail;
129         struct rte_mbuf *m_table[0];
130 };
131
132 struct rx_queue {
133         struct rte_ip_frag_tbl *frag_tbl;
134         struct rte_mempool *pool;
135         struct rte_lpm *lpm;
136         struct rte_lpm6 *lpm6;
137         uint16_t portid;
138 };
139
140 struct tx_lcore_stat {
141         uint64_t call;
142         uint64_t drop;
143         uint64_t queue;
144         uint64_t send;
145 };
146
147 #define MAX_RX_QUEUE_PER_LCORE 16
148 #define MAX_TX_QUEUE_PER_PORT 16
149 #define MAX_RX_QUEUE_PER_PORT 128
150
151 struct lcore_queue_conf {
152         uint16_t n_rx_queue;
153         struct rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
154         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
155         struct rte_ip_frag_death_row death_row;
156         struct mbuf_table *tx_mbufs[RTE_MAX_ETHPORTS];
157         struct tx_lcore_stat tx_stat;
158 } __rte_cache_aligned;
159 static struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
160
161 static struct rte_eth_conf port_conf = {
162         .rxmode = {
163                 .mq_mode        = RTE_ETH_MQ_RX_RSS,
164                 .mtu = JUMBO_FRAME_MAX_SIZE - RTE_ETHER_HDR_LEN -
165                         RTE_ETHER_CRC_LEN,
166                 .split_hdr_size = 0,
167                 .offloads = RTE_ETH_RX_OFFLOAD_CHECKSUM,
168         },
169         .rx_adv_conf = {
170                         .rss_conf = {
171                                 .rss_key = NULL,
172                                 .rss_hf = RTE_ETH_RSS_IP,
173                 },
174         },
175         .txmode = {
176                 .mq_mode = RTE_ETH_MQ_TX_NONE,
177                 .offloads = (RTE_ETH_TX_OFFLOAD_IPV4_CKSUM |
178                              RTE_ETH_TX_OFFLOAD_MULTI_SEGS),
179         },
180 };
181
182 /*
183  * IPv4 forwarding table
184  */
185 struct l3fwd_ipv4_route {
186         uint32_t ip;
187         uint8_t  depth;
188         uint8_t  if_out;
189 };
190
191 /* Default l3fwd_ipv4_route_array table. 8< */
192 struct l3fwd_ipv4_route l3fwd_ipv4_route_array[] = {
193                 {RTE_IPV4(100,10,0,0), 16, 0},
194                 {RTE_IPV4(100,20,0,0), 16, 1},
195                 {RTE_IPV4(100,30,0,0), 16, 2},
196                 {RTE_IPV4(100,40,0,0), 16, 3},
197                 {RTE_IPV4(100,50,0,0), 16, 4},
198                 {RTE_IPV4(100,60,0,0), 16, 5},
199                 {RTE_IPV4(100,70,0,0), 16, 6},
200                 {RTE_IPV4(100,80,0,0), 16, 7},
201 };
202 /* >8 End of default l3fwd_ipv4_route_array table. */
203
204 /*
205  * IPv6 forwarding table
206  */
207
208 struct l3fwd_ipv6_route {
209         uint8_t ip[IPV6_ADDR_LEN];
210         uint8_t depth;
211         uint8_t if_out;
212 };
213
214 /* Default l3fwd_ipv6_route_array table. 8< */
215 static struct l3fwd_ipv6_route l3fwd_ipv6_route_array[] = {
216         {{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
217         {{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
218         {{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
219         {{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
220         {{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
221         {{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
222         {{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
223         {{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
224 };
225 /* >8 End of default l3fwd_ipv6_route_array table. */
226
227 #define LPM_MAX_RULES         1024
228 #define LPM6_MAX_RULES         1024
229 #define LPM6_NUMBER_TBL8S (1 << 16)
230
231 struct rte_lpm6_config lpm6_config = {
232                 .max_rules = LPM6_MAX_RULES,
233                 .number_tbl8s = LPM6_NUMBER_TBL8S,
234                 .flags = 0
235 };
236
237 static struct rte_lpm *socket_lpm[RTE_MAX_NUMA_NODES];
238 static struct rte_lpm6 *socket_lpm6[RTE_MAX_NUMA_NODES];
239
240 #ifdef RTE_LIBRTE_IP_FRAG_TBL_STAT
241 #define TX_LCORE_STAT_UPDATE(s, f, v)   ((s)->f += (v))
242 #else
243 #define TX_LCORE_STAT_UPDATE(s, f, v)   do {} while (0)
244 #endif /* RTE_LIBRTE_IP_FRAG_TBL_STAT */
245
246 /*
247  * If number of queued packets reached given threahold, then
248  * send burst of packets on an output interface.
249  */
250 static inline uint32_t
251 send_burst(struct lcore_queue_conf *qconf, uint32_t thresh, uint16_t port)
252 {
253         uint32_t fill, len, k, n;
254         struct mbuf_table *txmb;
255
256         txmb = qconf->tx_mbufs[port];
257         len = txmb->len;
258
259         if ((int32_t)(fill = txmb->head - txmb->tail) < 0)
260                 fill += len;
261
262         if (fill >= thresh) {
263                 n = RTE_MIN(len - txmb->tail, fill);
264
265                 k = rte_eth_tx_burst(port, qconf->tx_queue_id[port],
266                         txmb->m_table + txmb->tail, (uint16_t)n);
267
268                 TX_LCORE_STAT_UPDATE(&qconf->tx_stat, call, 1);
269                 TX_LCORE_STAT_UPDATE(&qconf->tx_stat, send, k);
270
271                 fill -= k;
272                 if ((txmb->tail += k) == len)
273                         txmb->tail = 0;
274         }
275
276         return fill;
277 }
278
279 /* Enqueue a single packet, and send burst if queue is filled */
280 static inline int
281 send_single_packet(struct rte_mbuf *m, uint16_t port)
282 {
283         uint32_t fill, lcore_id, len;
284         struct lcore_queue_conf *qconf;
285         struct mbuf_table *txmb;
286
287         lcore_id = rte_lcore_id();
288         qconf = &lcore_queue_conf[lcore_id];
289
290         txmb = qconf->tx_mbufs[port];
291         len = txmb->len;
292
293         fill = send_burst(qconf, MAX_PKT_BURST, port);
294
295         if (fill == len - 1) {
296                 TX_LCORE_STAT_UPDATE(&qconf->tx_stat, drop, 1);
297                 rte_pktmbuf_free(txmb->m_table[txmb->tail]);
298                 if (++txmb->tail == len)
299                         txmb->tail = 0;
300         }
301
302         TX_LCORE_STAT_UPDATE(&qconf->tx_stat, queue, 1);
303         txmb->m_table[txmb->head] = m;
304         if(++txmb->head == len)
305                 txmb->head = 0;
306
307         return 0;
308 }
309
310 static inline void
311 reassemble(struct rte_mbuf *m, uint16_t portid, uint32_t queue,
312         struct lcore_queue_conf *qconf, uint64_t tms)
313 {
314         struct rte_ether_hdr *eth_hdr;
315         struct rte_ip_frag_tbl *tbl;
316         struct rte_ip_frag_death_row *dr;
317         struct rx_queue *rxq;
318         void *d_addr_bytes;
319         uint32_t next_hop;
320         uint16_t dst_port;
321
322         rxq = &qconf->rx_queue_list[queue];
323
324         eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
325
326         dst_port = portid;
327
328         /* if packet is IPv4 */
329         if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
330                 struct rte_ipv4_hdr *ip_hdr;
331                 uint32_t ip_dst;
332
333                 ip_hdr = (struct rte_ipv4_hdr *)(eth_hdr + 1);
334
335                  /* if it is a fragmented packet, then try to reassemble. */
336                 if (rte_ipv4_frag_pkt_is_fragmented(ip_hdr)) {
337                         struct rte_mbuf *mo;
338
339                         tbl = rxq->frag_tbl;
340                         dr = &qconf->death_row;
341
342                         /* prepare mbuf: setup l2_len/l3_len. */
343                         m->l2_len = sizeof(*eth_hdr);
344                         m->l3_len = sizeof(*ip_hdr);
345
346                         /* process this fragment. */
347                         mo = rte_ipv4_frag_reassemble_packet(tbl, dr, m, tms, ip_hdr);
348                         if (mo == NULL)
349                                 /* no packet to send out. */
350                                 return;
351
352                         /* we have our packet reassembled. */
353                         if (mo != m) {
354                                 m = mo;
355                                 eth_hdr = rte_pktmbuf_mtod(m,
356                                         struct rte_ether_hdr *);
357                                 ip_hdr = (struct rte_ipv4_hdr *)(eth_hdr + 1);
358                         }
359
360                         /* update offloading flags */
361                         m->ol_flags |= (RTE_MBUF_F_TX_IPV4 | RTE_MBUF_F_TX_IP_CKSUM);
362                 }
363                 ip_dst = rte_be_to_cpu_32(ip_hdr->dst_addr);
364
365                 /* Find destination port */
366                 if (rte_lpm_lookup(rxq->lpm, ip_dst, &next_hop) == 0 &&
367                                 (enabled_port_mask & 1 << next_hop) != 0) {
368                         dst_port = next_hop;
369                 }
370
371                 eth_hdr->ether_type = rte_be_to_cpu_16(RTE_ETHER_TYPE_IPV4);
372         } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
373                 /* if packet is IPv6 */
374                 struct ipv6_extension_fragment *frag_hdr;
375                 struct rte_ipv6_hdr *ip_hdr;
376
377                 ip_hdr = (struct rte_ipv6_hdr *)(eth_hdr + 1);
378
379                 frag_hdr = rte_ipv6_frag_get_ipv6_fragment_header(ip_hdr);
380
381                 if (frag_hdr != NULL) {
382                         struct rte_mbuf *mo;
383
384                         tbl = rxq->frag_tbl;
385                         dr  = &qconf->death_row;
386
387                         /* prepare mbuf: setup l2_len/l3_len. */
388                         m->l2_len = sizeof(*eth_hdr);
389                         m->l3_len = sizeof(*ip_hdr) + sizeof(*frag_hdr);
390
391                         mo = rte_ipv6_frag_reassemble_packet(tbl, dr, m, tms, ip_hdr, frag_hdr);
392                         if (mo == NULL)
393                                 return;
394
395                         if (mo != m) {
396                                 m = mo;
397                                 eth_hdr = rte_pktmbuf_mtod(m,
398                                                         struct rte_ether_hdr *);
399                                 ip_hdr = (struct rte_ipv6_hdr *)(eth_hdr + 1);
400                         }
401                 }
402
403                 /* Find destination port */
404                 if (rte_lpm6_lookup(rxq->lpm6, ip_hdr->dst_addr,
405                                                 &next_hop) == 0 &&
406                                 (enabled_port_mask & 1 << next_hop) != 0) {
407                         dst_port = next_hop;
408                 }
409
410                 eth_hdr->ether_type = rte_be_to_cpu_16(RTE_ETHER_TYPE_IPV6);
411         }
412         /* if packet wasn't IPv4 or IPv6, it's forwarded to the port it came from */
413
414         /* 02:00:00:00:00:xx */
415         d_addr_bytes = &eth_hdr->dst_addr.addr_bytes[0];
416         *((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)dst_port << 40);
417
418         /* src addr */
419         rte_ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->src_addr);
420
421         send_single_packet(m, dst_port);
422 }
423
424 /* main processing loop */
425 static int
426 main_loop(__rte_unused void *dummy)
427 {
428         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
429         unsigned lcore_id;
430         uint64_t diff_tsc, cur_tsc, prev_tsc;
431         int i, j, nb_rx;
432         uint16_t portid;
433         struct lcore_queue_conf *qconf;
434         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
435
436         prev_tsc = 0;
437
438         lcore_id = rte_lcore_id();
439         qconf = &lcore_queue_conf[lcore_id];
440
441         if (qconf->n_rx_queue == 0) {
442                 RTE_LOG(INFO, IP_RSMBL, "lcore %u has nothing to do\n", lcore_id);
443                 return 0;
444         }
445
446         RTE_LOG(INFO, IP_RSMBL, "entering main loop on lcore %u\n", lcore_id);
447
448         for (i = 0; i < qconf->n_rx_queue; i++) {
449
450                 portid = qconf->rx_queue_list[i].portid;
451                 RTE_LOG(INFO, IP_RSMBL, " -- lcoreid=%u portid=%u\n", lcore_id,
452                         portid);
453         }
454
455         while (1) {
456
457                 cur_tsc = rte_rdtsc();
458
459                 /*
460                  * TX burst queue drain
461                  */
462                 diff_tsc = cur_tsc - prev_tsc;
463                 if (unlikely(diff_tsc > drain_tsc)) {
464
465                         /*
466                          * This could be optimized (use queueid instead of
467                          * portid), but it is not called so often
468                          */
469                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
470                                 if ((enabled_port_mask & (1 << portid)) != 0)
471                                         send_burst(qconf, 1, portid);
472                         }
473
474                         prev_tsc = cur_tsc;
475                 }
476
477                 /*
478                  * Read packet from RX queues
479                  */
480                 for (i = 0; i < qconf->n_rx_queue; ++i) {
481
482                         portid = qconf->rx_queue_list[i].portid;
483
484                         nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
485                                 MAX_PKT_BURST);
486
487                         /* Prefetch first packets */
488                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
489                                 rte_prefetch0(rte_pktmbuf_mtod(
490                                                 pkts_burst[j], void *));
491                         }
492
493                         /* Prefetch and forward already prefetched packets */
494                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
495                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
496                                         j + PREFETCH_OFFSET], void *));
497                                 reassemble(pkts_burst[j], portid,
498                                         i, qconf, cur_tsc);
499                         }
500
501                         /* Forward remaining prefetched packets */
502                         for (; j < nb_rx; j++) {
503                                 reassemble(pkts_burst[j], portid,
504                                         i, qconf, cur_tsc);
505                         }
506
507                         rte_ip_frag_free_death_row(&qconf->death_row,
508                                 PREFETCH_OFFSET);
509                 }
510         }
511 }
512
513 /* display usage */
514 static void
515 print_usage(const char *prgname)
516 {
517         printf("%s [EAL options] -- -p PORTMASK [-q NQ]"
518                 "  [--maxflows=<flows>]  [--flowttl=<ttl>[(s|ms)]]\n"
519                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
520                 "  -q NQ: number of RX queues per lcore\n"
521                 "  --maxflows=<flows>: optional, maximum number of flows "
522                 "supported\n"
523                 "  --flowttl=<ttl>[(s|ms)]: optional, maximum TTL for each "
524                 "flow\n",
525                 prgname);
526 }
527
528 static uint32_t
529 parse_flow_num(const char *str, uint32_t min, uint32_t max, uint32_t *val)
530 {
531         char *end;
532         uint64_t v;
533
534         /* parse decimal string */
535         errno = 0;
536         v = strtoul(str, &end, 10);
537         if (errno != 0 || *end != '\0')
538                 return -EINVAL;
539
540         if (v < min || v > max)
541                 return -EINVAL;
542
543         *val = (uint32_t)v;
544         return 0;
545 }
546
547 static int
548 parse_flow_ttl(const char *str, uint32_t min, uint32_t max, uint32_t *val)
549 {
550         char *end;
551         uint64_t v;
552
553         static const char frmt_sec[] = "s";
554         static const char frmt_msec[] = "ms";
555
556         /* parse decimal string */
557         errno = 0;
558         v = strtoul(str, &end, 10);
559         if (errno != 0)
560                 return -EINVAL;
561
562         if (*end != '\0') {
563                 if (strncmp(frmt_sec, end, sizeof(frmt_sec)) == 0)
564                         v *= MS_PER_S;
565                 else if (strncmp(frmt_msec, end, sizeof (frmt_msec)) != 0)
566                         return -EINVAL;
567         }
568
569         if (v < min || v > max)
570                 return -EINVAL;
571
572         *val = (uint32_t)v;
573         return 0;
574 }
575
576 static int
577 parse_portmask(const char *portmask)
578 {
579         char *end = NULL;
580         unsigned long pm;
581
582         /* parse hexadecimal string */
583         pm = strtoul(portmask, &end, 16);
584         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
585                 return 0;
586
587         return pm;
588 }
589
590 static int
591 parse_nqueue(const char *q_arg)
592 {
593         char *end = NULL;
594         unsigned long n;
595
596         printf("%p\n", q_arg);
597
598         /* parse hexadecimal string */
599         n = strtoul(q_arg, &end, 10);
600         if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
601                 return -1;
602         if (n == 0)
603                 return -1;
604         if (n >= MAX_RX_QUEUE_PER_LCORE)
605                 return -1;
606
607         return n;
608 }
609
610 /* Parse the argument given in the command line of the application */
611 static int
612 parse_args(int argc, char **argv)
613 {
614         int opt, ret;
615         char **argvopt;
616         int option_index;
617         char *prgname = argv[0];
618         static struct option lgopts[] = {
619                 {"maxflows", 1, 0, 0},
620                 {"flowttl", 1, 0, 0},
621                 {NULL, 0, 0, 0}
622         };
623
624         argvopt = argv;
625
626         while ((opt = getopt_long(argc, argvopt, "p:q:",
627                                 lgopts, &option_index)) != EOF) {
628
629                 switch (opt) {
630                 /* portmask */
631                 case 'p':
632                         enabled_port_mask = parse_portmask(optarg);
633                         if (enabled_port_mask == 0) {
634                                 printf("invalid portmask\n");
635                                 print_usage(prgname);
636                                 return -1;
637                         }
638                         break;
639
640                 /* nqueue */
641                 case 'q':
642                         rx_queue_per_lcore = parse_nqueue(optarg);
643                         if (rx_queue_per_lcore < 0) {
644                                 printf("invalid queue number\n");
645                                 print_usage(prgname);
646                                 return -1;
647                         }
648                         break;
649
650                 /* long options */
651                 case 0:
652                         if (!strncmp(lgopts[option_index].name,
653                                         "maxflows", 8)) {
654                                 if ((ret = parse_flow_num(optarg, MIN_FLOW_NUM,
655                                                 MAX_FLOW_NUM,
656                                                 &max_flow_num)) != 0) {
657                                         printf("invalid value: \"%s\" for "
658                                                 "parameter %s\n",
659                                                 optarg,
660                                                 lgopts[option_index].name);
661                                         print_usage(prgname);
662                                         return ret;
663                                 }
664                         }
665
666                         if (!strncmp(lgopts[option_index].name, "flowttl", 7)) {
667                                 if ((ret = parse_flow_ttl(optarg, MIN_FLOW_TTL,
668                                                 MAX_FLOW_TTL,
669                                                 &max_flow_ttl)) != 0) {
670                                         printf("invalid value: \"%s\" for "
671                                                 "parameter %s\n",
672                                                 optarg,
673                                                 lgopts[option_index].name);
674                                         print_usage(prgname);
675                                         return ret;
676                                 }
677                         }
678
679                         break;
680
681                 default:
682                         print_usage(prgname);
683                         return -1;
684                 }
685         }
686
687         if (optind >= 0)
688                 argv[optind-1] = prgname;
689
690         ret = optind-1;
691         optind = 1; /* reset getopt lib */
692         return ret;
693 }
694
695 static void
696 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
697 {
698         char buf[RTE_ETHER_ADDR_FMT_SIZE];
699         rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
700         printf("%s%s", name, buf);
701 }
702
703 /* Check the link status of all ports in up to 9s, and print them finally */
704 static void
705 check_all_ports_link_status(uint32_t port_mask)
706 {
707 #define CHECK_INTERVAL 100 /* 100ms */
708 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
709         uint16_t portid;
710         uint8_t count, all_ports_up, print_flag = 0;
711         struct rte_eth_link link;
712         int ret;
713         char link_status_text[RTE_ETH_LINK_MAX_STR_LEN];
714
715         printf("\nChecking link status");
716         fflush(stdout);
717         for (count = 0; count <= MAX_CHECK_TIME; count++) {
718                 all_ports_up = 1;
719                 RTE_ETH_FOREACH_DEV(portid) {
720                         if ((port_mask & (1 << portid)) == 0)
721                                 continue;
722                         memset(&link, 0, sizeof(link));
723                         ret = rte_eth_link_get_nowait(portid, &link);
724                         if (ret < 0) {
725                                 all_ports_up = 0;
726                                 if (print_flag == 1)
727                                         printf("Port %u link get failed: %s\n",
728                                                 portid, rte_strerror(-ret));
729                                 continue;
730                         }
731                         /* print link status if flag set */
732                         if (print_flag == 1) {
733                                 rte_eth_link_to_str(link_status_text,
734                                         sizeof(link_status_text), &link);
735                                 printf("Port %d %s\n", portid,
736                                        link_status_text);
737                                 continue;
738                         }
739                         /* clear all_ports_up flag if any link down */
740                         if (link.link_status == RTE_ETH_LINK_DOWN) {
741                                 all_ports_up = 0;
742                                 break;
743                         }
744                 }
745                 /* after finally printing all link status, get out */
746                 if (print_flag == 1)
747                         break;
748
749                 if (all_ports_up == 0) {
750                         printf(".");
751                         fflush(stdout);
752                         rte_delay_ms(CHECK_INTERVAL);
753                 }
754
755                 /* set the print_flag if all ports up or timeout */
756                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
757                         print_flag = 1;
758                         printf("\ndone\n");
759                 }
760         }
761 }
762
763 static int
764 init_routing_table(void)
765 {
766         struct rte_lpm *lpm;
767         struct rte_lpm6 *lpm6;
768         int socket, ret;
769         unsigned i;
770
771         for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) {
772                 if (socket_lpm[socket]) {
773                         lpm = socket_lpm[socket];
774                         /* populate the LPM table */
775                         for (i = 0; i < RTE_DIM(l3fwd_ipv4_route_array); i++) {
776                                 ret = rte_lpm_add(lpm,
777                                         l3fwd_ipv4_route_array[i].ip,
778                                         l3fwd_ipv4_route_array[i].depth,
779                                         l3fwd_ipv4_route_array[i].if_out);
780
781                                 if (ret < 0) {
782                                         RTE_LOG(ERR, IP_RSMBL, "Unable to add entry %i to the l3fwd "
783                                                 "LPM table\n", i);
784                                         return -1;
785                                 }
786
787                                 RTE_LOG(INFO, IP_RSMBL, "Socket %i: adding route " IPv4_BYTES_FMT
788                                                 "/%d (port %d)\n",
789                                         socket,
790                                         IPv4_BYTES(l3fwd_ipv4_route_array[i].ip),
791                                         l3fwd_ipv4_route_array[i].depth,
792                                         l3fwd_ipv4_route_array[i].if_out);
793                         }
794                 }
795
796                 if (socket_lpm6[socket]) {
797                         lpm6 = socket_lpm6[socket];
798                         /* populate the LPM6 table */
799                         for (i = 0; i < RTE_DIM(l3fwd_ipv6_route_array); i++) {
800                                 ret = rte_lpm6_add(lpm6,
801                                         l3fwd_ipv6_route_array[i].ip,
802                                         l3fwd_ipv6_route_array[i].depth,
803                                         l3fwd_ipv6_route_array[i].if_out);
804
805                                 if (ret < 0) {
806                                         RTE_LOG(ERR, IP_RSMBL, "Unable to add entry %i to the l3fwd "
807                                                 "LPM6 table\n", i);
808                                         return -1;
809                                 }
810
811                                 RTE_LOG(INFO, IP_RSMBL, "Socket %i: adding route " IPv6_BYTES_FMT
812                                                 "/%d (port %d)\n",
813                                         socket,
814                                         IPv6_BYTES(l3fwd_ipv6_route_array[i].ip),
815                                         l3fwd_ipv6_route_array[i].depth,
816                                         l3fwd_ipv6_route_array[i].if_out);
817                         }
818                 }
819         }
820         return 0;
821 }
822
823 static int
824 setup_port_tbl(struct lcore_queue_conf *qconf, uint32_t lcore, int socket,
825         uint32_t port)
826 {
827         struct mbuf_table *mtb;
828         uint32_t n;
829         size_t sz;
830
831         n = RTE_MAX(max_flow_num, 2UL * MAX_PKT_BURST);
832         sz = sizeof (*mtb) + sizeof (mtb->m_table[0]) *  n;
833
834         if ((mtb = rte_zmalloc_socket(__func__, sz, RTE_CACHE_LINE_SIZE,
835                         socket)) == NULL) {
836                 RTE_LOG(ERR, IP_RSMBL, "%s() for lcore: %u, port: %u "
837                         "failed to allocate %zu bytes\n",
838                         __func__, lcore, port, sz);
839                 return -1;
840         }
841
842         mtb->len = n;
843         qconf->tx_mbufs[port] = mtb;
844
845         return 0;
846 }
847
848 static int
849 setup_queue_tbl(struct rx_queue *rxq, uint32_t lcore, uint32_t queue)
850 {
851         int socket;
852         uint32_t nb_mbuf;
853         uint64_t frag_cycles;
854         char buf[RTE_MEMPOOL_NAMESIZE];
855
856         socket = rte_lcore_to_socket_id(lcore);
857         if (socket == SOCKET_ID_ANY)
858                 socket = 0;
859
860         /* Each table entry holds information about packet fragmentation. 8< */
861         frag_cycles = (rte_get_tsc_hz() + MS_PER_S - 1) / MS_PER_S *
862                 max_flow_ttl;
863
864         if ((rxq->frag_tbl = rte_ip_frag_table_create(max_flow_num,
865                         IP_FRAG_TBL_BUCKET_ENTRIES, max_flow_num, frag_cycles,
866                         socket)) == NULL) {
867                 RTE_LOG(ERR, IP_RSMBL, "ip_frag_tbl_create(%u) on "
868                         "lcore: %u for queue: %u failed\n",
869                         max_flow_num, lcore, queue);
870                 return -1;
871         }
872         /* >8 End of holding packet fragmentation. */
873
874         /*
875          * At any given moment up to <max_flow_num * (MAX_FRAG_NUM)>
876          * mbufs could be stored int the fragment table.
877          * Plus, each TX queue can hold up to <max_flow_num> packets.
878          */
879
880         /* mbufs stored int the gragment table. 8< */
881         nb_mbuf = RTE_MAX(max_flow_num, 2UL * MAX_PKT_BURST) * MAX_FRAG_NUM;
882         nb_mbuf *= (port_conf.rxmode.mtu + RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN
883                         + BUF_SIZE - 1) / BUF_SIZE;
884         nb_mbuf *= 2; /* ipv4 and ipv6 */
885         nb_mbuf += nb_rxd + nb_txd;
886
887         nb_mbuf = RTE_MAX(nb_mbuf, (uint32_t)NB_MBUF);
888
889         snprintf(buf, sizeof(buf), "mbuf_pool_%u_%u", lcore, queue);
890
891         rxq->pool = rte_pktmbuf_pool_create(buf, nb_mbuf, MEMPOOL_CACHE_SIZE, 0,
892                                             MBUF_DATA_SIZE, socket);
893         if (rxq->pool == NULL) {
894                 RTE_LOG(ERR, IP_RSMBL,
895                         "rte_pktmbuf_pool_create(%s) failed", buf);
896                 return -1;
897         }
898         /* >8 End of mbufs stored int the fragmentation table. */
899
900         return 0;
901 }
902
903 static int
904 init_mem(void)
905 {
906         char buf[PATH_MAX];
907         struct rte_lpm *lpm;
908         struct rte_lpm6 *lpm6;
909         struct rte_lpm_config lpm_config;
910         int socket;
911         unsigned lcore_id;
912
913         /* traverse through lcores and initialize structures on each socket */
914
915         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
916
917                 if (rte_lcore_is_enabled(lcore_id) == 0)
918                         continue;
919
920                 socket = rte_lcore_to_socket_id(lcore_id);
921
922                 if (socket == SOCKET_ID_ANY)
923                         socket = 0;
924
925                 if (socket_lpm[socket] == NULL) {
926                         RTE_LOG(INFO, IP_RSMBL, "Creating LPM table on socket %i\n", socket);
927                         snprintf(buf, sizeof(buf), "IP_RSMBL_LPM_%i", socket);
928
929                         lpm_config.max_rules = LPM_MAX_RULES;
930                         lpm_config.number_tbl8s = 256;
931                         lpm_config.flags = 0;
932
933                         lpm = rte_lpm_create(buf, socket, &lpm_config);
934                         if (lpm == NULL) {
935                                 RTE_LOG(ERR, IP_RSMBL, "Cannot create LPM table\n");
936                                 return -1;
937                         }
938                         socket_lpm[socket] = lpm;
939                 }
940
941                 if (socket_lpm6[socket] == NULL) {
942                         RTE_LOG(INFO, IP_RSMBL, "Creating LPM6 table on socket %i\n", socket);
943                         snprintf(buf, sizeof(buf), "IP_RSMBL_LPM_%i", socket);
944
945                         lpm6 = rte_lpm6_create(buf, socket, &lpm6_config);
946                         if (lpm6 == NULL) {
947                                 RTE_LOG(ERR, IP_RSMBL, "Cannot create LPM table\n");
948                                 return -1;
949                         }
950                         socket_lpm6[socket] = lpm6;
951                 }
952         }
953
954         return 0;
955 }
956
957 static void
958 queue_dump_stat(void)
959 {
960         uint32_t i, lcore;
961         const struct lcore_queue_conf *qconf;
962
963         for (lcore = 0; lcore < RTE_MAX_LCORE; lcore++) {
964                 if (rte_lcore_is_enabled(lcore) == 0)
965                         continue;
966
967                 qconf = &lcore_queue_conf[lcore];
968                 for (i = 0; i < qconf->n_rx_queue; i++) {
969
970                         fprintf(stdout, " -- lcoreid=%u portid=%u "
971                                 "frag tbl stat:\n",
972                                 lcore,  qconf->rx_queue_list[i].portid);
973                         rte_ip_frag_table_statistics_dump(stdout,
974                                         qconf->rx_queue_list[i].frag_tbl);
975                         fprintf(stdout, "TX bursts:\t%" PRIu64 "\n"
976                                 "TX packets _queued:\t%" PRIu64 "\n"
977                                 "TX packets dropped:\t%" PRIu64 "\n"
978                                 "TX packets send:\t%" PRIu64 "\n",
979                                 qconf->tx_stat.call,
980                                 qconf->tx_stat.queue,
981                                 qconf->tx_stat.drop,
982                                 qconf->tx_stat.send);
983                 }
984         }
985 }
986
987 static void
988 signal_handler(int signum)
989 {
990         queue_dump_stat();
991         if (signum != SIGUSR1)
992                 rte_exit(0, "received signal: %d, exiting\n", signum);
993 }
994
995 int
996 main(int argc, char **argv)
997 {
998         struct lcore_queue_conf *qconf;
999         struct rte_eth_dev_info dev_info;
1000         struct rte_eth_txconf *txconf;
1001         struct rx_queue *rxq;
1002         int ret, socket;
1003         unsigned nb_ports;
1004         uint16_t queueid;
1005         unsigned lcore_id = 0, rx_lcore_id = 0;
1006         uint32_t n_tx_queue, nb_lcores;
1007         uint16_t portid;
1008
1009         /* init EAL */
1010         ret = rte_eal_init(argc, argv);
1011         if (ret < 0)
1012                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1013         argc -= ret;
1014         argv += ret;
1015
1016         /* parse application arguments (after the EAL ones) */
1017         ret = parse_args(argc, argv);
1018         if (ret < 0)
1019                 rte_exit(EXIT_FAILURE, "Invalid IP reassembly parameters\n");
1020
1021         nb_ports = rte_eth_dev_count_avail();
1022         if (nb_ports == 0)
1023                 rte_exit(EXIT_FAILURE, "No ports found!\n");
1024
1025         nb_lcores = rte_lcore_count();
1026
1027         /* initialize structures (mempools, lpm etc.) */
1028         if (init_mem() < 0)
1029                 rte_panic("Cannot initialize memory structures!\n");
1030
1031         /* check if portmask has non-existent ports */
1032         if (enabled_port_mask & ~(RTE_LEN2MASK(nb_ports, unsigned)))
1033                 rte_exit(EXIT_FAILURE, "Non-existent ports in portmask!\n");
1034
1035         /* initialize all ports */
1036         RTE_ETH_FOREACH_DEV(portid) {
1037                 struct rte_eth_rxconf rxq_conf;
1038                 struct rte_eth_conf local_port_conf = port_conf;
1039
1040                 /* skip ports that are not enabled */
1041                 if ((enabled_port_mask & (1 << portid)) == 0) {
1042                         printf("\nSkipping disabled port %d\n", portid);
1043                         continue;
1044                 }
1045
1046                 qconf = &lcore_queue_conf[rx_lcore_id];
1047
1048                 /* limit the frame size to the maximum supported by NIC */
1049                 ret = rte_eth_dev_info_get(portid, &dev_info);
1050                 if (ret != 0)
1051                         rte_exit(EXIT_FAILURE,
1052                                 "Error during getting device (port %u) info: %s\n",
1053                                 portid, strerror(-ret));
1054
1055                 local_port_conf.rxmode.mtu = RTE_MIN(
1056                     dev_info.max_mtu,
1057                     local_port_conf.rxmode.mtu);
1058
1059                 /* get the lcore_id for this port */
1060                 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
1061                            qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
1062
1063                         rx_lcore_id++;
1064                         if (rx_lcore_id >= RTE_MAX_LCORE)
1065                                 rte_exit(EXIT_FAILURE, "Not enough cores\n");
1066
1067                         qconf = &lcore_queue_conf[rx_lcore_id];
1068                 }
1069
1070                 socket = rte_lcore_to_socket_id(portid);
1071                 if (socket == SOCKET_ID_ANY)
1072                         socket = 0;
1073
1074                 queueid = qconf->n_rx_queue;
1075                 rxq = &qconf->rx_queue_list[queueid];
1076                 rxq->portid = portid;
1077                 rxq->lpm = socket_lpm[socket];
1078                 rxq->lpm6 = socket_lpm6[socket];
1079
1080                 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
1081                                                        &nb_txd);
1082                 if (ret < 0)
1083                         rte_exit(EXIT_FAILURE,
1084                                  "Cannot adjust number of descriptors: err=%d, port=%d\n",
1085                                  ret, portid);
1086
1087                 if (setup_queue_tbl(rxq, rx_lcore_id, queueid) < 0)
1088                         rte_exit(EXIT_FAILURE, "Failed to set up queue table\n");
1089                 qconf->n_rx_queue++;
1090
1091                 /* init port */
1092                 printf("Initializing port %d ... ", portid );
1093                 fflush(stdout);
1094
1095                 n_tx_queue = nb_lcores;
1096                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
1097                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1098                 if (dev_info.tx_offload_capa & RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE)
1099                         local_port_conf.txmode.offloads |=
1100                                 RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE;
1101
1102                 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
1103                         dev_info.flow_type_rss_offloads;
1104                 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
1105                                 port_conf.rx_adv_conf.rss_conf.rss_hf) {
1106                         printf("Port %u modified RSS hash function based on hardware support,"
1107                                 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
1108                                 portid,
1109                                 port_conf.rx_adv_conf.rss_conf.rss_hf,
1110                                 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
1111                 }
1112
1113                 ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
1114                                             &local_port_conf);
1115                 if (ret < 0) {
1116                         printf("\n");
1117                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
1118                                 "err=%d, port=%d\n",
1119                                 ret, portid);
1120                 }
1121
1122                 /* init one RX queue */
1123                 rxq_conf = dev_info.default_rxconf;
1124                 rxq_conf.offloads = local_port_conf.rxmode.offloads;
1125                 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
1126                                              socket, &rxq_conf,
1127                                              rxq->pool);
1128                 if (ret < 0) {
1129                         printf("\n");
1130                         rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: "
1131                                 "err=%d, port=%d\n",
1132                                 ret, portid);
1133                 }
1134
1135                 ret = rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1136                 if (ret < 0) {
1137                         printf("\n");
1138                         rte_exit(EXIT_FAILURE,
1139                                 "rte_eth_macaddr_get: err=%d, port=%d\n",
1140                                 ret, portid);
1141                 }
1142
1143                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1144                 printf("\n");
1145
1146                 /* init one TX queue per couple (lcore,port) */
1147                 queueid = 0;
1148                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1149                         if (rte_lcore_is_enabled(lcore_id) == 0)
1150                                 continue;
1151
1152                         socket = (int) rte_lcore_to_socket_id(lcore_id);
1153
1154                         printf("txq=%u,%d,%d ", lcore_id, queueid, socket);
1155                         fflush(stdout);
1156
1157                         txconf = &dev_info.default_txconf;
1158                         txconf->offloads = local_port_conf.txmode.offloads;
1159
1160                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1161                                         socket, txconf);
1162                         if (ret < 0)
1163                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
1164                                         "port=%d\n", ret, portid);
1165
1166                         qconf = &lcore_queue_conf[lcore_id];
1167                         qconf->tx_queue_id[portid] = queueid;
1168                         setup_port_tbl(qconf, lcore_id, socket, portid);
1169                         queueid++;
1170                 }
1171                 printf("\n");
1172         }
1173
1174         printf("\n");
1175
1176         /* start ports */
1177         RTE_ETH_FOREACH_DEV(portid) {
1178                 if ((enabled_port_mask & (1 << portid)) == 0) {
1179                         continue;
1180                 }
1181                 /* Start device */
1182                 ret = rte_eth_dev_start(portid);
1183                 if (ret < 0)
1184                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1185                                 ret, portid);
1186
1187                 ret = rte_eth_promiscuous_enable(portid);
1188                 if (ret != 0)
1189                         rte_exit(EXIT_FAILURE,
1190                                 "rte_eth_promiscuous_enable: err=%s, port=%d\n",
1191                                 rte_strerror(-ret), portid);
1192         }
1193
1194         if (init_routing_table() < 0)
1195                 rte_exit(EXIT_FAILURE, "Cannot init routing table\n");
1196
1197         check_all_ports_link_status(enabled_port_mask);
1198
1199         signal(SIGUSR1, signal_handler);
1200         signal(SIGTERM, signal_handler);
1201         signal(SIGINT, signal_handler);
1202
1203         /* launch per-lcore init on every lcore */
1204         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MAIN);
1205         RTE_LCORE_FOREACH_WORKER(lcore_id) {
1206                 if (rte_eal_wait_lcore(lcore_id) < 0)
1207                         return -1;
1208         }
1209
1210         /* clean up the EAL */
1211         rte_eal_cleanup();
1212
1213         return 0;
1214 }