net: rename Ethernet header fields
[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_atomic.h>
26 #include <rte_cycles.h>
27 #include <rte_prefetch.h>
28 #include <rte_lcore.h>
29 #include <rte_per_lcore.h>
30 #include <rte_branch_prediction.h>
31 #include <rte_interrupts.h>
32 #include <rte_random.h>
33 #include <rte_debug.h>
34 #include <rte_ether.h>
35 #include <rte_ethdev.h>
36 #include <rte_mempool.h>
37 #include <rte_mbuf.h>
38 #include <rte_malloc.h>
39 #include <rte_ip.h>
40 #include <rte_tcp.h>
41 #include <rte_udp.h>
42 #include <rte_string_fns.h>
43 #include <rte_lpm.h>
44 #include <rte_lpm6.h>
45
46 #include <rte_ip_frag.h>
47
48 #define MAX_PKT_BURST 32
49
50
51 #define RTE_LOGTYPE_IP_RSMBL RTE_LOGTYPE_USER1
52
53 #define MAX_JUMBO_PKT_LEN  9600
54
55 #define BUF_SIZE        RTE_MBUF_DEFAULT_DATAROOM
56 #define MBUF_DATA_SIZE  RTE_MBUF_DEFAULT_BUF_SIZE
57
58 #define NB_MBUF 8192
59 #define MEMPOOL_CACHE_SIZE 256
60
61 /* allow max jumbo frame 9.5 KB */
62 #define JUMBO_FRAME_MAX_SIZE    0x2600
63
64 #define MAX_FLOW_NUM    UINT16_MAX
65 #define MIN_FLOW_NUM    1
66 #define DEF_FLOW_NUM    0x1000
67
68 /* TTL numbers are in ms. */
69 #define MAX_FLOW_TTL    (3600 * MS_PER_S)
70 #define MIN_FLOW_TTL    1
71 #define DEF_FLOW_TTL    MS_PER_S
72
73 #define MAX_FRAG_NUM RTE_LIBRTE_IP_FRAG_MAX_FRAG
74
75 /* Should be power of two. */
76 #define IP_FRAG_TBL_BUCKET_ENTRIES      16
77
78 static uint32_t max_flow_num = DEF_FLOW_NUM;
79 static uint32_t max_flow_ttl = DEF_FLOW_TTL;
80
81 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
82
83 #define NB_SOCKETS 8
84
85 /* Configure how many packets ahead to prefetch, when reading packets */
86 #define PREFETCH_OFFSET 3
87
88 /*
89  * Configurable number of RX/TX ring descriptors
90  */
91 #define RTE_TEST_RX_DESC_DEFAULT 1024
92 #define RTE_TEST_TX_DESC_DEFAULT 1024
93
94 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
95 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
96
97 /* ethernet addresses of ports */
98 static struct rte_ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
99
100 #ifndef IPv4_BYTES
101 #define IPv4_BYTES_FMT "%" PRIu8 ".%" PRIu8 ".%" PRIu8 ".%" PRIu8
102 #define IPv4_BYTES(addr) \
103                 (uint8_t) (((addr) >> 24) & 0xFF),\
104                 (uint8_t) (((addr) >> 16) & 0xFF),\
105                 (uint8_t) (((addr) >> 8) & 0xFF),\
106                 (uint8_t) ((addr) & 0xFF)
107 #endif
108
109 #ifndef IPv6_BYTES
110 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
111                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
112 #define IPv6_BYTES(addr) \
113         addr[0],  addr[1], addr[2],  addr[3], \
114         addr[4],  addr[5], addr[6],  addr[7], \
115         addr[8],  addr[9], addr[10], addr[11],\
116         addr[12], addr[13],addr[14], addr[15]
117 #endif
118
119 #define IPV6_ADDR_LEN 16
120
121 /* mask of enabled ports */
122 static uint32_t enabled_port_mask = 0;
123
124 static int rx_queue_per_lcore = 1;
125
126 struct mbuf_table {
127         uint32_t len;
128         uint32_t head;
129         uint32_t tail;
130         struct rte_mbuf *m_table[0];
131 };
132
133 struct rx_queue {
134         struct rte_ip_frag_tbl *frag_tbl;
135         struct rte_mempool *pool;
136         struct rte_lpm *lpm;
137         struct rte_lpm6 *lpm6;
138         uint16_t portid;
139 };
140
141 struct tx_lcore_stat {
142         uint64_t call;
143         uint64_t drop;
144         uint64_t queue;
145         uint64_t send;
146 };
147
148 #define MAX_RX_QUEUE_PER_LCORE 16
149 #define MAX_TX_QUEUE_PER_PORT 16
150 #define MAX_RX_QUEUE_PER_PORT 128
151
152 struct lcore_queue_conf {
153         uint16_t n_rx_queue;
154         struct rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
155         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
156         struct rte_ip_frag_death_row death_row;
157         struct mbuf_table *tx_mbufs[RTE_MAX_ETHPORTS];
158         struct tx_lcore_stat tx_stat;
159 } __rte_cache_aligned;
160 static struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
161
162 static struct rte_eth_conf port_conf = {
163         .rxmode = {
164                 .mq_mode        = ETH_MQ_RX_RSS,
165                 .max_rx_pkt_len = JUMBO_FRAME_MAX_SIZE,
166                 .split_hdr_size = 0,
167                 .offloads = (DEV_RX_OFFLOAD_CHECKSUM |
168                              DEV_RX_OFFLOAD_JUMBO_FRAME),
169         },
170         .rx_adv_conf = {
171                         .rss_conf = {
172                                 .rss_key = NULL,
173                                 .rss_hf = ETH_RSS_IP,
174                 },
175         },
176         .txmode = {
177                 .mq_mode = ETH_MQ_TX_NONE,
178                 .offloads = (DEV_TX_OFFLOAD_IPV4_CKSUM |
179                              DEV_TX_OFFLOAD_MULTI_SEGS),
180         },
181 };
182
183 /*
184  * IPv4 forwarding table
185  */
186 struct l3fwd_ipv4_route {
187         uint32_t ip;
188         uint8_t  depth;
189         uint8_t  if_out;
190 };
191
192 /* Default l3fwd_ipv4_route_array table. 8< */
193 struct l3fwd_ipv4_route l3fwd_ipv4_route_array[] = {
194                 {RTE_IPV4(100,10,0,0), 16, 0},
195                 {RTE_IPV4(100,20,0,0), 16, 1},
196                 {RTE_IPV4(100,30,0,0), 16, 2},
197                 {RTE_IPV4(100,40,0,0), 16, 3},
198                 {RTE_IPV4(100,50,0,0), 16, 4},
199                 {RTE_IPV4(100,60,0,0), 16, 5},
200                 {RTE_IPV4(100,70,0,0), 16, 6},
201                 {RTE_IPV4(100,80,0,0), 16, 7},
202 };
203 /* >8 End of default l3fwd_ipv4_route_array table. */
204
205 /*
206  * IPv6 forwarding table
207  */
208
209 struct l3fwd_ipv6_route {
210         uint8_t ip[IPV6_ADDR_LEN];
211         uint8_t depth;
212         uint8_t if_out;
213 };
214
215 /* Default l3fwd_ipv6_route_array table. 8< */
216 static struct l3fwd_ipv6_route l3fwd_ipv6_route_array[] = {
217         {{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
218         {{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
219         {{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
220         {{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
221         {{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
222         {{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
223         {{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
224         {{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
225 };
226 /* >8 End of default l3fwd_ipv6_route_array table. */
227
228 #define LPM_MAX_RULES         1024
229 #define LPM6_MAX_RULES         1024
230 #define LPM6_NUMBER_TBL8S (1 << 16)
231
232 struct rte_lpm6_config lpm6_config = {
233                 .max_rules = LPM6_MAX_RULES,
234                 .number_tbl8s = LPM6_NUMBER_TBL8S,
235                 .flags = 0
236 };
237
238 static struct rte_lpm *socket_lpm[RTE_MAX_NUMA_NODES];
239 static struct rte_lpm6 *socket_lpm6[RTE_MAX_NUMA_NODES];
240
241 #ifdef RTE_LIBRTE_IP_FRAG_TBL_STAT
242 #define TX_LCORE_STAT_UPDATE(s, f, v)   ((s)->f += (v))
243 #else
244 #define TX_LCORE_STAT_UPDATE(s, f, v)   do {} while (0)
245 #endif /* RTE_LIBRTE_IP_FRAG_TBL_STAT */
246
247 /*
248  * If number of queued packets reached given threahold, then
249  * send burst of packets on an output interface.
250  */
251 static inline uint32_t
252 send_burst(struct lcore_queue_conf *qconf, uint32_t thresh, uint16_t port)
253 {
254         uint32_t fill, len, k, n;
255         struct mbuf_table *txmb;
256
257         txmb = qconf->tx_mbufs[port];
258         len = txmb->len;
259
260         if ((int32_t)(fill = txmb->head - txmb->tail) < 0)
261                 fill += len;
262
263         if (fill >= thresh) {
264                 n = RTE_MIN(len - txmb->tail, fill);
265
266                 k = rte_eth_tx_burst(port, qconf->tx_queue_id[port],
267                         txmb->m_table + txmb->tail, (uint16_t)n);
268
269                 TX_LCORE_STAT_UPDATE(&qconf->tx_stat, call, 1);
270                 TX_LCORE_STAT_UPDATE(&qconf->tx_stat, send, k);
271
272                 fill -= k;
273                 if ((txmb->tail += k) == len)
274                         txmb->tail = 0;
275         }
276
277         return fill;
278 }
279
280 /* Enqueue a single packet, and send burst if queue is filled */
281 static inline int
282 send_single_packet(struct rte_mbuf *m, uint16_t port)
283 {
284         uint32_t fill, lcore_id, len;
285         struct lcore_queue_conf *qconf;
286         struct mbuf_table *txmb;
287
288         lcore_id = rte_lcore_id();
289         qconf = &lcore_queue_conf[lcore_id];
290
291         txmb = qconf->tx_mbufs[port];
292         len = txmb->len;
293
294         fill = send_burst(qconf, MAX_PKT_BURST, port);
295
296         if (fill == len - 1) {
297                 TX_LCORE_STAT_UPDATE(&qconf->tx_stat, drop, 1);
298                 rte_pktmbuf_free(txmb->m_table[txmb->tail]);
299                 if (++txmb->tail == len)
300                         txmb->tail = 0;
301         }
302
303         TX_LCORE_STAT_UPDATE(&qconf->tx_stat, queue, 1);
304         txmb->m_table[txmb->head] = m;
305         if(++txmb->head == len)
306                 txmb->head = 0;
307
308         return 0;
309 }
310
311 static inline void
312 reassemble(struct rte_mbuf *m, uint16_t portid, uint32_t queue,
313         struct lcore_queue_conf *qconf, uint64_t tms)
314 {
315         struct rte_ether_hdr *eth_hdr;
316         struct rte_ip_frag_tbl *tbl;
317         struct rte_ip_frag_death_row *dr;
318         struct rx_queue *rxq;
319         void *d_addr_bytes;
320         uint32_t next_hop;
321         uint16_t dst_port;
322
323         rxq = &qconf->rx_queue_list[queue];
324
325         eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
326
327         dst_port = portid;
328
329         /* if packet is IPv4 */
330         if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
331                 struct rte_ipv4_hdr *ip_hdr;
332                 uint32_t ip_dst;
333
334                 ip_hdr = (struct rte_ipv4_hdr *)(eth_hdr + 1);
335
336                  /* if it is a fragmented packet, then try to reassemble. */
337                 if (rte_ipv4_frag_pkt_is_fragmented(ip_hdr)) {
338                         struct rte_mbuf *mo;
339
340                         tbl = rxq->frag_tbl;
341                         dr = &qconf->death_row;
342
343                         /* prepare mbuf: setup l2_len/l3_len. */
344                         m->l2_len = sizeof(*eth_hdr);
345                         m->l3_len = sizeof(*ip_hdr);
346
347                         /* process this fragment. */
348                         mo = rte_ipv4_frag_reassemble_packet(tbl, dr, m, tms, ip_hdr);
349                         if (mo == NULL)
350                                 /* no packet to send out. */
351                                 return;
352
353                         /* we have our packet reassembled. */
354                         if (mo != m) {
355                                 m = mo;
356                                 eth_hdr = rte_pktmbuf_mtod(m,
357                                         struct rte_ether_hdr *);
358                                 ip_hdr = (struct rte_ipv4_hdr *)(eth_hdr + 1);
359                         }
360
361                         /* update offloading flags */
362                         m->ol_flags |= (PKT_TX_IPV4 | PKT_TX_IP_CKSUM);
363                 }
364                 ip_dst = rte_be_to_cpu_32(ip_hdr->dst_addr);
365
366                 /* Find destination port */
367                 if (rte_lpm_lookup(rxq->lpm, ip_dst, &next_hop) == 0 &&
368                                 (enabled_port_mask & 1 << next_hop) != 0) {
369                         dst_port = next_hop;
370                 }
371
372                 eth_hdr->ether_type = rte_be_to_cpu_16(RTE_ETHER_TYPE_IPV4);
373         } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
374                 /* if packet is IPv6 */
375                 struct ipv6_extension_fragment *frag_hdr;
376                 struct rte_ipv6_hdr *ip_hdr;
377
378                 ip_hdr = (struct rte_ipv6_hdr *)(eth_hdr + 1);
379
380                 frag_hdr = rte_ipv6_frag_get_ipv6_fragment_header(ip_hdr);
381
382                 if (frag_hdr != NULL) {
383                         struct rte_mbuf *mo;
384
385                         tbl = rxq->frag_tbl;
386                         dr  = &qconf->death_row;
387
388                         /* prepare mbuf: setup l2_len/l3_len. */
389                         m->l2_len = sizeof(*eth_hdr);
390                         m->l3_len = sizeof(*ip_hdr) + sizeof(*frag_hdr);
391
392                         mo = rte_ipv6_frag_reassemble_packet(tbl, dr, m, tms, ip_hdr, frag_hdr);
393                         if (mo == NULL)
394                                 return;
395
396                         if (mo != m) {
397                                 m = mo;
398                                 eth_hdr = rte_pktmbuf_mtod(m,
399                                                         struct rte_ether_hdr *);
400                                 ip_hdr = (struct rte_ipv6_hdr *)(eth_hdr + 1);
401                         }
402                 }
403
404                 /* Find destination port */
405                 if (rte_lpm6_lookup(rxq->lpm6, ip_hdr->dst_addr,
406                                                 &next_hop) == 0 &&
407                                 (enabled_port_mask & 1 << next_hop) != 0) {
408                         dst_port = next_hop;
409                 }
410
411                 eth_hdr->ether_type = rte_be_to_cpu_16(RTE_ETHER_TYPE_IPV6);
412         }
413         /* if packet wasn't IPv4 or IPv6, it's forwarded to the port it came from */
414
415         /* 02:00:00:00:00:xx */
416         d_addr_bytes = &eth_hdr->dst_addr.addr_bytes[0];
417         *((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)dst_port << 40);
418
419         /* src addr */
420         rte_ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->src_addr);
421
422         send_single_packet(m, dst_port);
423 }
424
425 /* main processing loop */
426 static int
427 main_loop(__rte_unused void *dummy)
428 {
429         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
430         unsigned lcore_id;
431         uint64_t diff_tsc, cur_tsc, prev_tsc;
432         int i, j, nb_rx;
433         uint16_t portid;
434         struct lcore_queue_conf *qconf;
435         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
436
437         prev_tsc = 0;
438
439         lcore_id = rte_lcore_id();
440         qconf = &lcore_queue_conf[lcore_id];
441
442         if (qconf->n_rx_queue == 0) {
443                 RTE_LOG(INFO, IP_RSMBL, "lcore %u has nothing to do\n", lcore_id);
444                 return 0;
445         }
446
447         RTE_LOG(INFO, IP_RSMBL, "entering main loop on lcore %u\n", lcore_id);
448
449         for (i = 0; i < qconf->n_rx_queue; i++) {
450
451                 portid = qconf->rx_queue_list[i].portid;
452                 RTE_LOG(INFO, IP_RSMBL, " -- lcoreid=%u portid=%u\n", lcore_id,
453                         portid);
454         }
455
456         while (1) {
457
458                 cur_tsc = rte_rdtsc();
459
460                 /*
461                  * TX burst queue drain
462                  */
463                 diff_tsc = cur_tsc - prev_tsc;
464                 if (unlikely(diff_tsc > drain_tsc)) {
465
466                         /*
467                          * This could be optimized (use queueid instead of
468                          * portid), but it is not called so often
469                          */
470                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
471                                 if ((enabled_port_mask & (1 << portid)) != 0)
472                                         send_burst(qconf, 1, portid);
473                         }
474
475                         prev_tsc = cur_tsc;
476                 }
477
478                 /*
479                  * Read packet from RX queues
480                  */
481                 for (i = 0; i < qconf->n_rx_queue; ++i) {
482
483                         portid = qconf->rx_queue_list[i].portid;
484
485                         nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
486                                 MAX_PKT_BURST);
487
488                         /* Prefetch first packets */
489                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
490                                 rte_prefetch0(rte_pktmbuf_mtod(
491                                                 pkts_burst[j], void *));
492                         }
493
494                         /* Prefetch and forward already prefetched packets */
495                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
496                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
497                                         j + PREFETCH_OFFSET], void *));
498                                 reassemble(pkts_burst[j], portid,
499                                         i, qconf, cur_tsc);
500                         }
501
502                         /* Forward remaining prefetched packets */
503                         for (; j < nb_rx; j++) {
504                                 reassemble(pkts_burst[j], portid,
505                                         i, qconf, cur_tsc);
506                         }
507
508                         rte_ip_frag_free_death_row(&qconf->death_row,
509                                 PREFETCH_OFFSET);
510                 }
511         }
512 }
513
514 /* display usage */
515 static void
516 print_usage(const char *prgname)
517 {
518         printf("%s [EAL options] -- -p PORTMASK [-q NQ]"
519                 "  [--max-pkt-len PKTLEN]"
520                 "  [--maxflows=<flows>]  [--flowttl=<ttl>[(s|ms)]]\n"
521                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
522                 "  -q NQ: number of RX queues per lcore\n"
523                 "  --maxflows=<flows>: optional, maximum number of flows "
524                 "supported\n"
525                 "  --flowttl=<ttl>[(s|ms)]: optional, maximum TTL for each "
526                 "flow\n",
527                 prgname);
528 }
529
530 static uint32_t
531 parse_flow_num(const char *str, uint32_t min, uint32_t max, uint32_t *val)
532 {
533         char *end;
534         uint64_t v;
535
536         /* parse decimal string */
537         errno = 0;
538         v = strtoul(str, &end, 10);
539         if (errno != 0 || *end != '\0')
540                 return -EINVAL;
541
542         if (v < min || v > max)
543                 return -EINVAL;
544
545         *val = (uint32_t)v;
546         return 0;
547 }
548
549 static int
550 parse_flow_ttl(const char *str, uint32_t min, uint32_t max, uint32_t *val)
551 {
552         char *end;
553         uint64_t v;
554
555         static const char frmt_sec[] = "s";
556         static const char frmt_msec[] = "ms";
557
558         /* parse decimal string */
559         errno = 0;
560         v = strtoul(str, &end, 10);
561         if (errno != 0)
562                 return -EINVAL;
563
564         if (*end != '\0') {
565                 if (strncmp(frmt_sec, end, sizeof(frmt_sec)) == 0)
566                         v *= MS_PER_S;
567                 else if (strncmp(frmt_msec, end, sizeof (frmt_msec)) != 0)
568                         return -EINVAL;
569         }
570
571         if (v < min || v > max)
572                 return -EINVAL;
573
574         *val = (uint32_t)v;
575         return 0;
576 }
577
578 static int
579 parse_portmask(const char *portmask)
580 {
581         char *end = NULL;
582         unsigned long pm;
583
584         /* parse hexadecimal string */
585         pm = strtoul(portmask, &end, 16);
586         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
587                 return 0;
588
589         return pm;
590 }
591
592 static int
593 parse_nqueue(const char *q_arg)
594 {
595         char *end = NULL;
596         unsigned long n;
597
598         printf("%p\n", q_arg);
599
600         /* parse hexadecimal string */
601         n = strtoul(q_arg, &end, 10);
602         if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
603                 return -1;
604         if (n == 0)
605                 return -1;
606         if (n >= MAX_RX_QUEUE_PER_LCORE)
607                 return -1;
608
609         return n;
610 }
611
612 /* Parse the argument given in the command line of the application */
613 static int
614 parse_args(int argc, char **argv)
615 {
616         int opt, ret;
617         char **argvopt;
618         int option_index;
619         char *prgname = argv[0];
620         static struct option lgopts[] = {
621                 {"max-pkt-len", 1, 0, 0},
622                 {"maxflows", 1, 0, 0},
623                 {"flowttl", 1, 0, 0},
624                 {NULL, 0, 0, 0}
625         };
626
627         argvopt = argv;
628
629         while ((opt = getopt_long(argc, argvopt, "p:q:",
630                                 lgopts, &option_index)) != EOF) {
631
632                 switch (opt) {
633                 /* portmask */
634                 case 'p':
635                         enabled_port_mask = parse_portmask(optarg);
636                         if (enabled_port_mask == 0) {
637                                 printf("invalid portmask\n");
638                                 print_usage(prgname);
639                                 return -1;
640                         }
641                         break;
642
643                 /* nqueue */
644                 case 'q':
645                         rx_queue_per_lcore = parse_nqueue(optarg);
646                         if (rx_queue_per_lcore < 0) {
647                                 printf("invalid queue number\n");
648                                 print_usage(prgname);
649                                 return -1;
650                         }
651                         break;
652
653                 /* long options */
654                 case 0:
655                         if (!strncmp(lgopts[option_index].name,
656                                         "maxflows", 8)) {
657                                 if ((ret = parse_flow_num(optarg, MIN_FLOW_NUM,
658                                                 MAX_FLOW_NUM,
659                                                 &max_flow_num)) != 0) {
660                                         printf("invalid value: \"%s\" for "
661                                                 "parameter %s\n",
662                                                 optarg,
663                                                 lgopts[option_index].name);
664                                         print_usage(prgname);
665                                         return ret;
666                                 }
667                         }
668
669                         if (!strncmp(lgopts[option_index].name, "flowttl", 7)) {
670                                 if ((ret = parse_flow_ttl(optarg, MIN_FLOW_TTL,
671                                                 MAX_FLOW_TTL,
672                                                 &max_flow_ttl)) != 0) {
673                                         printf("invalid value: \"%s\" for "
674                                                 "parameter %s\n",
675                                                 optarg,
676                                                 lgopts[option_index].name);
677                                         print_usage(prgname);
678                                         return ret;
679                                 }
680                         }
681
682                         break;
683
684                 default:
685                         print_usage(prgname);
686                         return -1;
687                 }
688         }
689
690         if (optind >= 0)
691                 argv[optind-1] = prgname;
692
693         ret = optind-1;
694         optind = 1; /* reset getopt lib */
695         return ret;
696 }
697
698 static void
699 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
700 {
701         char buf[RTE_ETHER_ADDR_FMT_SIZE];
702         rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
703         printf("%s%s", name, buf);
704 }
705
706 /* Check the link status of all ports in up to 9s, and print them finally */
707 static void
708 check_all_ports_link_status(uint32_t port_mask)
709 {
710 #define CHECK_INTERVAL 100 /* 100ms */
711 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
712         uint16_t portid;
713         uint8_t count, all_ports_up, print_flag = 0;
714         struct rte_eth_link link;
715         int ret;
716         char link_status_text[RTE_ETH_LINK_MAX_STR_LEN];
717
718         printf("\nChecking link status");
719         fflush(stdout);
720         for (count = 0; count <= MAX_CHECK_TIME; count++) {
721                 all_ports_up = 1;
722                 RTE_ETH_FOREACH_DEV(portid) {
723                         if ((port_mask & (1 << portid)) == 0)
724                                 continue;
725                         memset(&link, 0, sizeof(link));
726                         ret = rte_eth_link_get_nowait(portid, &link);
727                         if (ret < 0) {
728                                 all_ports_up = 0;
729                                 if (print_flag == 1)
730                                         printf("Port %u link get failed: %s\n",
731                                                 portid, rte_strerror(-ret));
732                                 continue;
733                         }
734                         /* print link status if flag set */
735                         if (print_flag == 1) {
736                                 rte_eth_link_to_str(link_status_text,
737                                         sizeof(link_status_text), &link);
738                                 printf("Port %d %s\n", portid,
739                                        link_status_text);
740                                 continue;
741                         }
742                         /* clear all_ports_up flag if any link down */
743                         if (link.link_status == ETH_LINK_DOWN) {
744                                 all_ports_up = 0;
745                                 break;
746                         }
747                 }
748                 /* after finally printing all link status, get out */
749                 if (print_flag == 1)
750                         break;
751
752                 if (all_ports_up == 0) {
753                         printf(".");
754                         fflush(stdout);
755                         rte_delay_ms(CHECK_INTERVAL);
756                 }
757
758                 /* set the print_flag if all ports up or timeout */
759                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
760                         print_flag = 1;
761                         printf("\ndone\n");
762                 }
763         }
764 }
765
766 static int
767 init_routing_table(void)
768 {
769         struct rte_lpm *lpm;
770         struct rte_lpm6 *lpm6;
771         int socket, ret;
772         unsigned i;
773
774         for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) {
775                 if (socket_lpm[socket]) {
776                         lpm = socket_lpm[socket];
777                         /* populate the LPM table */
778                         for (i = 0; i < RTE_DIM(l3fwd_ipv4_route_array); i++) {
779                                 ret = rte_lpm_add(lpm,
780                                         l3fwd_ipv4_route_array[i].ip,
781                                         l3fwd_ipv4_route_array[i].depth,
782                                         l3fwd_ipv4_route_array[i].if_out);
783
784                                 if (ret < 0) {
785                                         RTE_LOG(ERR, IP_RSMBL, "Unable to add entry %i to the l3fwd "
786                                                 "LPM table\n", i);
787                                         return -1;
788                                 }
789
790                                 RTE_LOG(INFO, IP_RSMBL, "Socket %i: adding route " IPv4_BYTES_FMT
791                                                 "/%d (port %d)\n",
792                                         socket,
793                                         IPv4_BYTES(l3fwd_ipv4_route_array[i].ip),
794                                         l3fwd_ipv4_route_array[i].depth,
795                                         l3fwd_ipv4_route_array[i].if_out);
796                         }
797                 }
798
799                 if (socket_lpm6[socket]) {
800                         lpm6 = socket_lpm6[socket];
801                         /* populate the LPM6 table */
802                         for (i = 0; i < RTE_DIM(l3fwd_ipv6_route_array); i++) {
803                                 ret = rte_lpm6_add(lpm6,
804                                         l3fwd_ipv6_route_array[i].ip,
805                                         l3fwd_ipv6_route_array[i].depth,
806                                         l3fwd_ipv6_route_array[i].if_out);
807
808                                 if (ret < 0) {
809                                         RTE_LOG(ERR, IP_RSMBL, "Unable to add entry %i to the l3fwd "
810                                                 "LPM6 table\n", i);
811                                         return -1;
812                                 }
813
814                                 RTE_LOG(INFO, IP_RSMBL, "Socket %i: adding route " IPv6_BYTES_FMT
815                                                 "/%d (port %d)\n",
816                                         socket,
817                                         IPv6_BYTES(l3fwd_ipv6_route_array[i].ip),
818                                         l3fwd_ipv6_route_array[i].depth,
819                                         l3fwd_ipv6_route_array[i].if_out);
820                         }
821                 }
822         }
823         return 0;
824 }
825
826 static int
827 setup_port_tbl(struct lcore_queue_conf *qconf, uint32_t lcore, int socket,
828         uint32_t port)
829 {
830         struct mbuf_table *mtb;
831         uint32_t n;
832         size_t sz;
833
834         n = RTE_MAX(max_flow_num, 2UL * MAX_PKT_BURST);
835         sz = sizeof (*mtb) + sizeof (mtb->m_table[0]) *  n;
836
837         if ((mtb = rte_zmalloc_socket(__func__, sz, RTE_CACHE_LINE_SIZE,
838                         socket)) == NULL) {
839                 RTE_LOG(ERR, IP_RSMBL, "%s() for lcore: %u, port: %u "
840                         "failed to allocate %zu bytes\n",
841                         __func__, lcore, port, sz);
842                 return -1;
843         }
844
845         mtb->len = n;
846         qconf->tx_mbufs[port] = mtb;
847
848         return 0;
849 }
850
851 static int
852 setup_queue_tbl(struct rx_queue *rxq, uint32_t lcore, uint32_t queue)
853 {
854         int socket;
855         uint32_t nb_mbuf;
856         uint64_t frag_cycles;
857         char buf[RTE_MEMPOOL_NAMESIZE];
858
859         socket = rte_lcore_to_socket_id(lcore);
860         if (socket == SOCKET_ID_ANY)
861                 socket = 0;
862
863         /* Each table entry holds information about packet fragmentation. 8< */
864         frag_cycles = (rte_get_tsc_hz() + MS_PER_S - 1) / MS_PER_S *
865                 max_flow_ttl;
866
867         if ((rxq->frag_tbl = rte_ip_frag_table_create(max_flow_num,
868                         IP_FRAG_TBL_BUCKET_ENTRIES, max_flow_num, frag_cycles,
869                         socket)) == NULL) {
870                 RTE_LOG(ERR, IP_RSMBL, "ip_frag_tbl_create(%u) on "
871                         "lcore: %u for queue: %u failed\n",
872                         max_flow_num, lcore, queue);
873                 return -1;
874         }
875         /* >8 End of holding packet fragmentation. */
876
877         /*
878          * At any given moment up to <max_flow_num * (MAX_FRAG_NUM)>
879          * mbufs could be stored int the fragment table.
880          * Plus, each TX queue can hold up to <max_flow_num> packets.
881          */
882
883         /* mbufs stored int the gragment table. 8< */
884         nb_mbuf = RTE_MAX(max_flow_num, 2UL * MAX_PKT_BURST) * MAX_FRAG_NUM;
885         nb_mbuf *= (port_conf.rxmode.max_rx_pkt_len + BUF_SIZE - 1) / BUF_SIZE;
886         nb_mbuf *= 2; /* ipv4 and ipv6 */
887         nb_mbuf += nb_rxd + nb_txd;
888
889         nb_mbuf = RTE_MAX(nb_mbuf, (uint32_t)NB_MBUF);
890
891         snprintf(buf, sizeof(buf), "mbuf_pool_%u_%u", lcore, queue);
892
893         rxq->pool = rte_pktmbuf_pool_create(buf, nb_mbuf, MEMPOOL_CACHE_SIZE, 0,
894                                             MBUF_DATA_SIZE, socket);
895         if (rxq->pool == NULL) {
896                 RTE_LOG(ERR, IP_RSMBL,
897                         "rte_pktmbuf_pool_create(%s) failed", buf);
898                 return -1;
899         }
900         /* >8 End of mbufs stored int the fragmentation table. */
901
902         return 0;
903 }
904
905 static int
906 init_mem(void)
907 {
908         char buf[PATH_MAX];
909         struct rte_lpm *lpm;
910         struct rte_lpm6 *lpm6;
911         struct rte_lpm_config lpm_config;
912         int socket;
913         unsigned lcore_id;
914
915         /* traverse through lcores and initialize structures on each socket */
916
917         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
918
919                 if (rte_lcore_is_enabled(lcore_id) == 0)
920                         continue;
921
922                 socket = rte_lcore_to_socket_id(lcore_id);
923
924                 if (socket == SOCKET_ID_ANY)
925                         socket = 0;
926
927                 if (socket_lpm[socket] == NULL) {
928                         RTE_LOG(INFO, IP_RSMBL, "Creating LPM table on socket %i\n", socket);
929                         snprintf(buf, sizeof(buf), "IP_RSMBL_LPM_%i", socket);
930
931                         lpm_config.max_rules = LPM_MAX_RULES;
932                         lpm_config.number_tbl8s = 256;
933                         lpm_config.flags = 0;
934
935                         lpm = rte_lpm_create(buf, socket, &lpm_config);
936                         if (lpm == NULL) {
937                                 RTE_LOG(ERR, IP_RSMBL, "Cannot create LPM table\n");
938                                 return -1;
939                         }
940                         socket_lpm[socket] = lpm;
941                 }
942
943                 if (socket_lpm6[socket] == NULL) {
944                         RTE_LOG(INFO, IP_RSMBL, "Creating LPM6 table on socket %i\n", socket);
945                         snprintf(buf, sizeof(buf), "IP_RSMBL_LPM_%i", socket);
946
947                         lpm6 = rte_lpm6_create(buf, socket, &lpm6_config);
948                         if (lpm6 == NULL) {
949                                 RTE_LOG(ERR, IP_RSMBL, "Cannot create LPM table\n");
950                                 return -1;
951                         }
952                         socket_lpm6[socket] = lpm6;
953                 }
954         }
955
956         return 0;
957 }
958
959 static void
960 queue_dump_stat(void)
961 {
962         uint32_t i, lcore;
963         const struct lcore_queue_conf *qconf;
964
965         for (lcore = 0; lcore < RTE_MAX_LCORE; lcore++) {
966                 if (rte_lcore_is_enabled(lcore) == 0)
967                         continue;
968
969                 qconf = &lcore_queue_conf[lcore];
970                 for (i = 0; i < qconf->n_rx_queue; i++) {
971
972                         fprintf(stdout, " -- lcoreid=%u portid=%u "
973                                 "frag tbl stat:\n",
974                                 lcore,  qconf->rx_queue_list[i].portid);
975                         rte_ip_frag_table_statistics_dump(stdout,
976                                         qconf->rx_queue_list[i].frag_tbl);
977                         fprintf(stdout, "TX bursts:\t%" PRIu64 "\n"
978                                 "TX packets _queued:\t%" PRIu64 "\n"
979                                 "TX packets dropped:\t%" PRIu64 "\n"
980                                 "TX packets send:\t%" PRIu64 "\n",
981                                 qconf->tx_stat.call,
982                                 qconf->tx_stat.queue,
983                                 qconf->tx_stat.drop,
984                                 qconf->tx_stat.send);
985                 }
986         }
987 }
988
989 static void
990 signal_handler(int signum)
991 {
992         queue_dump_stat();
993         if (signum != SIGUSR1)
994                 rte_exit(0, "received signal: %d, exiting\n", signum);
995 }
996
997 int
998 main(int argc, char **argv)
999 {
1000         struct lcore_queue_conf *qconf;
1001         struct rte_eth_dev_info dev_info;
1002         struct rte_eth_txconf *txconf;
1003         struct rx_queue *rxq;
1004         int ret, socket;
1005         unsigned nb_ports;
1006         uint16_t queueid;
1007         unsigned lcore_id = 0, rx_lcore_id = 0;
1008         uint32_t n_tx_queue, nb_lcores;
1009         uint16_t portid;
1010
1011         /* init EAL */
1012         ret = rte_eal_init(argc, argv);
1013         if (ret < 0)
1014                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1015         argc -= ret;
1016         argv += ret;
1017
1018         /* parse application arguments (after the EAL ones) */
1019         ret = parse_args(argc, argv);
1020         if (ret < 0)
1021                 rte_exit(EXIT_FAILURE, "Invalid IP reassembly parameters\n");
1022
1023         nb_ports = rte_eth_dev_count_avail();
1024         if (nb_ports == 0)
1025                 rte_exit(EXIT_FAILURE, "No ports found!\n");
1026
1027         nb_lcores = rte_lcore_count();
1028
1029         /* initialize structures (mempools, lpm etc.) */
1030         if (init_mem() < 0)
1031                 rte_panic("Cannot initialize memory structures!\n");
1032
1033         /* check if portmask has non-existent ports */
1034         if (enabled_port_mask & ~(RTE_LEN2MASK(nb_ports, unsigned)))
1035                 rte_exit(EXIT_FAILURE, "Non-existent ports in portmask!\n");
1036
1037         /* initialize all ports */
1038         RTE_ETH_FOREACH_DEV(portid) {
1039                 struct rte_eth_rxconf rxq_conf;
1040                 struct rte_eth_conf local_port_conf = port_conf;
1041
1042                 /* skip ports that are not enabled */
1043                 if ((enabled_port_mask & (1 << portid)) == 0) {
1044                         printf("\nSkipping disabled port %d\n", portid);
1045                         continue;
1046                 }
1047
1048                 qconf = &lcore_queue_conf[rx_lcore_id];
1049
1050                 /* limit the frame size to the maximum supported by NIC */
1051                 ret = rte_eth_dev_info_get(portid, &dev_info);
1052                 if (ret != 0)
1053                         rte_exit(EXIT_FAILURE,
1054                                 "Error during getting device (port %u) info: %s\n",
1055                                 portid, strerror(-ret));
1056
1057                 local_port_conf.rxmode.max_rx_pkt_len = RTE_MIN(
1058                     dev_info.max_rx_pktlen,
1059                     local_port_conf.rxmode.max_rx_pkt_len);
1060
1061                 /* get the lcore_id for this port */
1062                 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
1063                            qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
1064
1065                         rx_lcore_id++;
1066                         if (rx_lcore_id >= RTE_MAX_LCORE)
1067                                 rte_exit(EXIT_FAILURE, "Not enough cores\n");
1068
1069                         qconf = &lcore_queue_conf[rx_lcore_id];
1070                 }
1071
1072                 socket = rte_lcore_to_socket_id(portid);
1073                 if (socket == SOCKET_ID_ANY)
1074                         socket = 0;
1075
1076                 queueid = qconf->n_rx_queue;
1077                 rxq = &qconf->rx_queue_list[queueid];
1078                 rxq->portid = portid;
1079                 rxq->lpm = socket_lpm[socket];
1080                 rxq->lpm6 = socket_lpm6[socket];
1081
1082                 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
1083                                                        &nb_txd);
1084                 if (ret < 0)
1085                         rte_exit(EXIT_FAILURE,
1086                                  "Cannot adjust number of descriptors: err=%d, port=%d\n",
1087                                  ret, portid);
1088
1089                 if (setup_queue_tbl(rxq, rx_lcore_id, queueid) < 0)
1090                         rte_exit(EXIT_FAILURE, "Failed to set up queue table\n");
1091                 qconf->n_rx_queue++;
1092
1093                 /* init port */
1094                 printf("Initializing port %d ... ", portid );
1095                 fflush(stdout);
1096
1097                 n_tx_queue = nb_lcores;
1098                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
1099                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1100                 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
1101                         local_port_conf.txmode.offloads |=
1102                                 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
1103
1104                 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
1105                         dev_info.flow_type_rss_offloads;
1106                 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
1107                                 port_conf.rx_adv_conf.rss_conf.rss_hf) {
1108                         printf("Port %u modified RSS hash function based on hardware support,"
1109                                 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
1110                                 portid,
1111                                 port_conf.rx_adv_conf.rss_conf.rss_hf,
1112                                 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
1113                 }
1114
1115                 ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
1116                                             &local_port_conf);
1117                 if (ret < 0) {
1118                         printf("\n");
1119                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
1120                                 "err=%d, port=%d\n",
1121                                 ret, portid);
1122                 }
1123
1124                 /* init one RX queue */
1125                 rxq_conf = dev_info.default_rxconf;
1126                 rxq_conf.offloads = local_port_conf.rxmode.offloads;
1127                 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
1128                                              socket, &rxq_conf,
1129                                              rxq->pool);
1130                 if (ret < 0) {
1131                         printf("\n");
1132                         rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: "
1133                                 "err=%d, port=%d\n",
1134                                 ret, portid);
1135                 }
1136
1137                 ret = rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1138                 if (ret < 0) {
1139                         printf("\n");
1140                         rte_exit(EXIT_FAILURE,
1141                                 "rte_eth_macaddr_get: err=%d, port=%d\n",
1142                                 ret, portid);
1143                 }
1144
1145                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1146                 printf("\n");
1147
1148                 /* init one TX queue per couple (lcore,port) */
1149                 queueid = 0;
1150                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1151                         if (rte_lcore_is_enabled(lcore_id) == 0)
1152                                 continue;
1153
1154                         socket = (int) rte_lcore_to_socket_id(lcore_id);
1155
1156                         printf("txq=%u,%d,%d ", lcore_id, queueid, socket);
1157                         fflush(stdout);
1158
1159                         txconf = &dev_info.default_txconf;
1160                         txconf->offloads = local_port_conf.txmode.offloads;
1161
1162                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1163                                         socket, txconf);
1164                         if (ret < 0)
1165                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
1166                                         "port=%d\n", ret, portid);
1167
1168                         qconf = &lcore_queue_conf[lcore_id];
1169                         qconf->tx_queue_id[portid] = queueid;
1170                         setup_port_tbl(qconf, lcore_id, socket, portid);
1171                         queueid++;
1172                 }
1173                 printf("\n");
1174         }
1175
1176         printf("\n");
1177
1178         /* start ports */
1179         RTE_ETH_FOREACH_DEV(portid) {
1180                 if ((enabled_port_mask & (1 << portid)) == 0) {
1181                         continue;
1182                 }
1183                 /* Start device */
1184                 ret = rte_eth_dev_start(portid);
1185                 if (ret < 0)
1186                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1187                                 ret, portid);
1188
1189                 ret = rte_eth_promiscuous_enable(portid);
1190                 if (ret != 0)
1191                         rte_exit(EXIT_FAILURE,
1192                                 "rte_eth_promiscuous_enable: err=%s, port=%d\n",
1193                                 rte_strerror(-ret), portid);
1194         }
1195
1196         if (init_routing_table() < 0)
1197                 rte_exit(EXIT_FAILURE, "Cannot init routing table\n");
1198
1199         check_all_ports_link_status(enabled_port_mask);
1200
1201         signal(SIGUSR1, signal_handler);
1202         signal(SIGTERM, signal_handler);
1203         signal(SIGINT, signal_handler);
1204
1205         /* launch per-lcore init on every lcore */
1206         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MAIN);
1207         RTE_LCORE_FOREACH_WORKER(lcore_id) {
1208                 if (rte_eal_wait_lcore(lcore_id) < 0)
1209                         return -1;
1210         }
1211
1212         /* clean up the EAL */
1213         rte_eal_cleanup();
1214
1215         return 0;
1216 }