mbuf: remove control mbuf
[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 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                 .ignore_offload_bitfield = 1,
168                 .offloads = (DEV_RX_OFFLOAD_CHECKSUM |
169                              DEV_RX_OFFLOAD_JUMBO_FRAME |
170                              DEV_RX_OFFLOAD_CRC_STRIP),
171         },
172         .rx_adv_conf = {
173                         .rss_conf = {
174                                 .rss_key = NULL,
175                                 .rss_hf = ETH_RSS_IP,
176                 },
177         },
178         .txmode = {
179                 .mq_mode = ETH_MQ_TX_NONE,
180                 .offloads = (DEV_TX_OFFLOAD_IPV4_CKSUM |
181                              DEV_TX_OFFLOAD_MULTI_SEGS),
182         },
183 };
184
185 /*
186  * IPv4 forwarding table
187  */
188 struct l3fwd_ipv4_route {
189         uint32_t ip;
190         uint8_t  depth;
191         uint8_t  if_out;
192 };
193
194 struct l3fwd_ipv4_route l3fwd_ipv4_route_array[] = {
195                 {IPv4(100,10,0,0), 16, 0},
196                 {IPv4(100,20,0,0), 16, 1},
197                 {IPv4(100,30,0,0), 16, 2},
198                 {IPv4(100,40,0,0), 16, 3},
199                 {IPv4(100,50,0,0), 16, 4},
200                 {IPv4(100,60,0,0), 16, 5},
201                 {IPv4(100,70,0,0), 16, 6},
202                 {IPv4(100,80,0,0), 16, 7},
203 };
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 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
226 #define LPM_MAX_RULES         1024
227 #define LPM6_MAX_RULES         1024
228 #define LPM6_NUMBER_TBL8S (1 << 16)
229
230 struct rte_lpm6_config lpm6_config = {
231                 .max_rules = LPM6_MAX_RULES,
232                 .number_tbl8s = LPM6_NUMBER_TBL8S,
233                 .flags = 0
234 };
235
236 static struct rte_lpm *socket_lpm[RTE_MAX_NUMA_NODES];
237 static struct rte_lpm6 *socket_lpm6[RTE_MAX_NUMA_NODES];
238
239 #ifdef RTE_LIBRTE_IP_FRAG_TBL_STAT
240 #define TX_LCORE_STAT_UPDATE(s, f, v)   ((s)->f += (v))
241 #else
242 #define TX_LCORE_STAT_UPDATE(s, f, v)   do {} while (0)
243 #endif /* RTE_LIBRTE_IP_FRAG_TBL_STAT */
244
245 /*
246  * If number of queued packets reached given threahold, then
247  * send burst of packets on an output interface.
248  */
249 static inline uint32_t
250 send_burst(struct lcore_queue_conf *qconf, uint32_t thresh, uint16_t port)
251 {
252         uint32_t fill, len, k, n;
253         struct mbuf_table *txmb;
254
255         txmb = qconf->tx_mbufs[port];
256         len = txmb->len;
257
258         if ((int32_t)(fill = txmb->head - txmb->tail) < 0)
259                 fill += len;
260
261         if (fill >= thresh) {
262                 n = RTE_MIN(len - txmb->tail, fill);
263
264                 k = rte_eth_tx_burst(port, qconf->tx_queue_id[port],
265                         txmb->m_table + txmb->tail, (uint16_t)n);
266
267                 TX_LCORE_STAT_UPDATE(&qconf->tx_stat, call, 1);
268                 TX_LCORE_STAT_UPDATE(&qconf->tx_stat, send, k);
269
270                 fill -= k;
271                 if ((txmb->tail += k) == len)
272                         txmb->tail = 0;
273         }
274
275         return fill;
276 }
277
278 /* Enqueue a single packet, and send burst if queue is filled */
279 static inline int
280 send_single_packet(struct rte_mbuf *m, uint16_t port)
281 {
282         uint32_t fill, lcore_id, len;
283         struct lcore_queue_conf *qconf;
284         struct mbuf_table *txmb;
285
286         lcore_id = rte_lcore_id();
287         qconf = &lcore_queue_conf[lcore_id];
288
289         txmb = qconf->tx_mbufs[port];
290         len = txmb->len;
291
292         fill = send_burst(qconf, MAX_PKT_BURST, port);
293
294         if (fill == len - 1) {
295                 TX_LCORE_STAT_UPDATE(&qconf->tx_stat, drop, 1);
296                 rte_pktmbuf_free(txmb->m_table[txmb->tail]);
297                 if (++txmb->tail == len)
298                         txmb->tail = 0;
299         }
300
301         TX_LCORE_STAT_UPDATE(&qconf->tx_stat, queue, 1);
302         txmb->m_table[txmb->head] = m;
303         if(++txmb->head == len)
304                 txmb->head = 0;
305
306         return 0;
307 }
308
309 static inline void
310 reassemble(struct rte_mbuf *m, uint16_t portid, uint32_t queue,
311         struct lcore_queue_conf *qconf, uint64_t tms)
312 {
313         struct ether_hdr *eth_hdr;
314         struct rte_ip_frag_tbl *tbl;
315         struct rte_ip_frag_death_row *dr;
316         struct rx_queue *rxq;
317         void *d_addr_bytes;
318         uint32_t next_hop;
319         uint16_t dst_port;
320
321         rxq = &qconf->rx_queue_list[queue];
322
323         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
324
325         dst_port = portid;
326
327         /* if packet is IPv4 */
328         if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
329                 struct ipv4_hdr *ip_hdr;
330                 uint32_t ip_dst;
331
332                 ip_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
333
334                  /* if it is a fragmented packet, then try to reassemble. */
335                 if (rte_ipv4_frag_pkt_is_fragmented(ip_hdr)) {
336                         struct rte_mbuf *mo;
337
338                         tbl = rxq->frag_tbl;
339                         dr = &qconf->death_row;
340
341                         /* prepare mbuf: setup l2_len/l3_len. */
342                         m->l2_len = sizeof(*eth_hdr);
343                         m->l3_len = sizeof(*ip_hdr);
344
345                         /* process this fragment. */
346                         mo = rte_ipv4_frag_reassemble_packet(tbl, dr, m, tms, ip_hdr);
347                         if (mo == NULL)
348                                 /* no packet to send out. */
349                                 return;
350
351                         /* we have our packet reassembled. */
352                         if (mo != m) {
353                                 m = mo;
354                                 eth_hdr = rte_pktmbuf_mtod(m,
355                                         struct ether_hdr *);
356                                 ip_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
357                         }
358                 }
359                 ip_dst = rte_be_to_cpu_32(ip_hdr->dst_addr);
360
361                 /* Find destination port */
362                 if (rte_lpm_lookup(rxq->lpm, ip_dst, &next_hop) == 0 &&
363                                 (enabled_port_mask & 1 << next_hop) != 0) {
364                         dst_port = next_hop;
365                 }
366
367                 eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv4);
368         } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
369                 /* if packet is IPv6 */
370                 struct ipv6_extension_fragment *frag_hdr;
371                 struct ipv6_hdr *ip_hdr;
372
373                 ip_hdr = (struct ipv6_hdr *)(eth_hdr + 1);
374
375                 frag_hdr = rte_ipv6_frag_get_ipv6_fragment_header(ip_hdr);
376
377                 if (frag_hdr != NULL) {
378                         struct rte_mbuf *mo;
379
380                         tbl = rxq->frag_tbl;
381                         dr  = &qconf->death_row;
382
383                         /* prepare mbuf: setup l2_len/l3_len. */
384                         m->l2_len = sizeof(*eth_hdr);
385                         m->l3_len = sizeof(*ip_hdr) + sizeof(*frag_hdr);
386
387                         mo = rte_ipv6_frag_reassemble_packet(tbl, dr, m, tms, ip_hdr, frag_hdr);
388                         if (mo == NULL)
389                                 return;
390
391                         if (mo != m) {
392                                 m = mo;
393                                 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
394                                 ip_hdr = (struct ipv6_hdr *)(eth_hdr + 1);
395                         }
396                 }
397
398                 /* Find destination port */
399                 if (rte_lpm6_lookup(rxq->lpm6, ip_hdr->dst_addr,
400                                                 &next_hop) == 0 &&
401                                 (enabled_port_mask & 1 << next_hop) != 0) {
402                         dst_port = next_hop;
403                 }
404
405                 eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv6);
406         }
407         /* if packet wasn't IPv4 or IPv6, it's forwarded to the port it came from */
408
409         /* 02:00:00:00:00:xx */
410         d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
411         *((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)dst_port << 40);
412
413         /* src addr */
414         ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
415
416         send_single_packet(m, dst_port);
417 }
418
419 /* main processing loop */
420 static int
421 main_loop(__attribute__((unused)) void *dummy)
422 {
423         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
424         unsigned lcore_id;
425         uint64_t diff_tsc, cur_tsc, prev_tsc;
426         int i, j, nb_rx;
427         uint16_t portid;
428         struct lcore_queue_conf *qconf;
429         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
430
431         prev_tsc = 0;
432
433         lcore_id = rte_lcore_id();
434         qconf = &lcore_queue_conf[lcore_id];
435
436         if (qconf->n_rx_queue == 0) {
437                 RTE_LOG(INFO, IP_RSMBL, "lcore %u has nothing to do\n", lcore_id);
438                 return 0;
439         }
440
441         RTE_LOG(INFO, IP_RSMBL, "entering main loop on lcore %u\n", lcore_id);
442
443         for (i = 0; i < qconf->n_rx_queue; i++) {
444
445                 portid = qconf->rx_queue_list[i].portid;
446                 RTE_LOG(INFO, IP_RSMBL, " -- lcoreid=%u portid=%u\n", lcore_id,
447                         portid);
448         }
449
450         while (1) {
451
452                 cur_tsc = rte_rdtsc();
453
454                 /*
455                  * TX burst queue drain
456                  */
457                 diff_tsc = cur_tsc - prev_tsc;
458                 if (unlikely(diff_tsc > drain_tsc)) {
459
460                         /*
461                          * This could be optimized (use queueid instead of
462                          * portid), but it is not called so often
463                          */
464                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
465                                 if ((enabled_port_mask & (1 << portid)) != 0)
466                                         send_burst(qconf, 1, portid);
467                         }
468
469                         prev_tsc = cur_tsc;
470                 }
471
472                 /*
473                  * Read packet from RX queues
474                  */
475                 for (i = 0; i < qconf->n_rx_queue; ++i) {
476
477                         portid = qconf->rx_queue_list[i].portid;
478
479                         nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
480                                 MAX_PKT_BURST);
481
482                         /* Prefetch first packets */
483                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
484                                 rte_prefetch0(rte_pktmbuf_mtod(
485                                                 pkts_burst[j], void *));
486                         }
487
488                         /* Prefetch and forward already prefetched packets */
489                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
490                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
491                                         j + PREFETCH_OFFSET], void *));
492                                 reassemble(pkts_burst[j], portid,
493                                         i, qconf, cur_tsc);
494                         }
495
496                         /* Forward remaining prefetched packets */
497                         for (; j < nb_rx; j++) {
498                                 reassemble(pkts_burst[j], portid,
499                                         i, qconf, cur_tsc);
500                         }
501
502                         rte_ip_frag_free_death_row(&qconf->death_row,
503                                 PREFETCH_OFFSET);
504                 }
505         }
506 }
507
508 /* display usage */
509 static void
510 print_usage(const char *prgname)
511 {
512         printf("%s [EAL options] -- -p PORTMASK [-q NQ]"
513                 "  [--max-pkt-len PKTLEN]"
514                 "  [--maxflows=<flows>]  [--flowttl=<ttl>[(s|ms)]]\n"
515                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
516                 "  -q NQ: number of RX queues per lcore\n"
517                 "  --maxflows=<flows>: optional, maximum number of flows "
518                 "supported\n"
519                 "  --flowttl=<ttl>[(s|ms)]: optional, maximum TTL for each "
520                 "flow\n",
521                 prgname);
522 }
523
524 static uint32_t
525 parse_flow_num(const char *str, uint32_t min, uint32_t max, uint32_t *val)
526 {
527         char *end;
528         uint64_t v;
529
530         /* parse decimal string */
531         errno = 0;
532         v = strtoul(str, &end, 10);
533         if (errno != 0 || *end != '\0')
534                 return -EINVAL;
535
536         if (v < min || v > max)
537                 return -EINVAL;
538
539         *val = (uint32_t)v;
540         return 0;
541 }
542
543 static int
544 parse_flow_ttl(const char *str, uint32_t min, uint32_t max, uint32_t *val)
545 {
546         char *end;
547         uint64_t v;
548
549         static const char frmt_sec[] = "s";
550         static const char frmt_msec[] = "ms";
551
552         /* parse decimal string */
553         errno = 0;
554         v = strtoul(str, &end, 10);
555         if (errno != 0)
556                 return -EINVAL;
557
558         if (*end != '\0') {
559                 if (strncmp(frmt_sec, end, sizeof(frmt_sec)) == 0)
560                         v *= MS_PER_S;
561                 else if (strncmp(frmt_msec, end, sizeof (frmt_msec)) != 0)
562                         return -EINVAL;
563         }
564
565         if (v < min || v > max)
566                 return -EINVAL;
567
568         *val = (uint32_t)v;
569         return 0;
570 }
571
572 static int
573 parse_portmask(const char *portmask)
574 {
575         char *end = NULL;
576         unsigned long pm;
577
578         /* parse hexadecimal string */
579         pm = strtoul(portmask, &end, 16);
580         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
581                 return -1;
582
583         if (pm == 0)
584                 return -1;
585
586         return pm;
587 }
588
589 static int
590 parse_nqueue(const char *q_arg)
591 {
592         char *end = NULL;
593         unsigned long n;
594
595         printf("%p\n", q_arg);
596
597         /* parse hexadecimal string */
598         n = strtoul(q_arg, &end, 10);
599         if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
600                 return -1;
601         if (n == 0)
602                 return -1;
603         if (n >= MAX_RX_QUEUE_PER_LCORE)
604                 return -1;
605
606         return n;
607 }
608
609 /* Parse the argument given in the command line of the application */
610 static int
611 parse_args(int argc, char **argv)
612 {
613         int opt, ret;
614         char **argvopt;
615         int option_index;
616         char *prgname = argv[0];
617         static struct option lgopts[] = {
618                 {"max-pkt-len", 1, 0, 0},
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 ether_addr *eth_addr)
697 {
698         char buf[ETHER_ADDR_FMT_SIZE];
699         ether_format_addr(buf, 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(uint16_t port_num, 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
713         printf("\nChecking link status");
714         fflush(stdout);
715         for (count = 0; count <= MAX_CHECK_TIME; count++) {
716                 all_ports_up = 1;
717                 for (portid = 0; portid < port_num; portid++) {
718                         if ((port_mask & (1 << portid)) == 0)
719                                 continue;
720                         memset(&link, 0, sizeof(link));
721                         rte_eth_link_get_nowait(portid, &link);
722                         /* print link status if flag set */
723                         if (print_flag == 1) {
724                                 if (link.link_status)
725                                         printf(
726                                         "Port%d Link Up. Speed %u Mbps - %s\n",
727                                                 portid, link.link_speed,
728                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
729                                         ("full-duplex") : ("half-duplex\n"));
730                                 else
731                                         printf("Port %d Link Down\n", portid);
732                                 continue;
733                         }
734                         /* clear all_ports_up flag if any link down */
735                         if (link.link_status == ETH_LINK_DOWN) {
736                                 all_ports_up = 0;
737                                 break;
738                         }
739                 }
740                 /* after finally printing all link status, get out */
741                 if (print_flag == 1)
742                         break;
743
744                 if (all_ports_up == 0) {
745                         printf(".");
746                         fflush(stdout);
747                         rte_delay_ms(CHECK_INTERVAL);
748                 }
749
750                 /* set the print_flag if all ports up or timeout */
751                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
752                         print_flag = 1;
753                         printf("\ndone\n");
754                 }
755         }
756 }
757
758 static int
759 init_routing_table(void)
760 {
761         struct rte_lpm *lpm;
762         struct rte_lpm6 *lpm6;
763         int socket, ret;
764         unsigned i;
765
766         for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) {
767                 if (socket_lpm[socket]) {
768                         lpm = socket_lpm[socket];
769                         /* populate the LPM table */
770                         for (i = 0; i < RTE_DIM(l3fwd_ipv4_route_array); i++) {
771                                 ret = rte_lpm_add(lpm,
772                                         l3fwd_ipv4_route_array[i].ip,
773                                         l3fwd_ipv4_route_array[i].depth,
774                                         l3fwd_ipv4_route_array[i].if_out);
775
776                                 if (ret < 0) {
777                                         RTE_LOG(ERR, IP_RSMBL, "Unable to add entry %i to the l3fwd "
778                                                 "LPM table\n", i);
779                                         return -1;
780                                 }
781
782                                 RTE_LOG(INFO, IP_RSMBL, "Socket %i: adding route " IPv4_BYTES_FMT
783                                                 "/%d (port %d)\n",
784                                         socket,
785                                         IPv4_BYTES(l3fwd_ipv4_route_array[i].ip),
786                                         l3fwd_ipv4_route_array[i].depth,
787                                         l3fwd_ipv4_route_array[i].if_out);
788                         }
789                 }
790
791                 if (socket_lpm6[socket]) {
792                         lpm6 = socket_lpm6[socket];
793                         /* populate the LPM6 table */
794                         for (i = 0; i < RTE_DIM(l3fwd_ipv6_route_array); i++) {
795                                 ret = rte_lpm6_add(lpm6,
796                                         l3fwd_ipv6_route_array[i].ip,
797                                         l3fwd_ipv6_route_array[i].depth,
798                                         l3fwd_ipv6_route_array[i].if_out);
799
800                                 if (ret < 0) {
801                                         RTE_LOG(ERR, IP_RSMBL, "Unable to add entry %i to the l3fwd "
802                                                 "LPM6 table\n", i);
803                                         return -1;
804                                 }
805
806                                 RTE_LOG(INFO, IP_RSMBL, "Socket %i: adding route " IPv6_BYTES_FMT
807                                                 "/%d (port %d)\n",
808                                         socket,
809                                         IPv6_BYTES(l3fwd_ipv6_route_array[i].ip),
810                                         l3fwd_ipv6_route_array[i].depth,
811                                         l3fwd_ipv6_route_array[i].if_out);
812                         }
813                 }
814         }
815         return 0;
816 }
817
818 static int
819 setup_port_tbl(struct lcore_queue_conf *qconf, uint32_t lcore, int socket,
820         uint32_t port)
821 {
822         struct mbuf_table *mtb;
823         uint32_t n;
824         size_t sz;
825
826         n = RTE_MAX(max_flow_num, 2UL * MAX_PKT_BURST);
827         sz = sizeof (*mtb) + sizeof (mtb->m_table[0]) *  n;
828
829         if ((mtb = rte_zmalloc_socket(__func__, sz, RTE_CACHE_LINE_SIZE,
830                         socket)) == NULL) {
831                 RTE_LOG(ERR, IP_RSMBL, "%s() for lcore: %u, port: %u "
832                         "failed to allocate %zu bytes\n",
833                         __func__, lcore, port, sz);
834                 return -1;
835         }
836
837         mtb->len = n;
838         qconf->tx_mbufs[port] = mtb;
839
840         return 0;
841 }
842
843 static int
844 setup_queue_tbl(struct rx_queue *rxq, uint32_t lcore, uint32_t queue)
845 {
846         int socket;
847         uint32_t nb_mbuf;
848         uint64_t frag_cycles;
849         char buf[RTE_MEMPOOL_NAMESIZE];
850
851         socket = rte_lcore_to_socket_id(lcore);
852         if (socket == SOCKET_ID_ANY)
853                 socket = 0;
854
855         frag_cycles = (rte_get_tsc_hz() + MS_PER_S - 1) / MS_PER_S *
856                 max_flow_ttl;
857
858         if ((rxq->frag_tbl = rte_ip_frag_table_create(max_flow_num,
859                         IP_FRAG_TBL_BUCKET_ENTRIES, max_flow_num, frag_cycles,
860                         socket)) == NULL) {
861                 RTE_LOG(ERR, IP_RSMBL, "ip_frag_tbl_create(%u) on "
862                         "lcore: %u for queue: %u failed\n",
863                         max_flow_num, lcore, queue);
864                 return -1;
865         }
866
867         /*
868          * At any given moment up to <max_flow_num * (MAX_FRAG_NUM)>
869          * mbufs could be stored int the fragment table.
870          * Plus, each TX queue can hold up to <max_flow_num> packets.
871          */
872
873         nb_mbuf = RTE_MAX(max_flow_num, 2UL * MAX_PKT_BURST) * MAX_FRAG_NUM;
874         nb_mbuf *= (port_conf.rxmode.max_rx_pkt_len + BUF_SIZE - 1) / BUF_SIZE;
875         nb_mbuf *= 2; /* ipv4 and ipv6 */
876         nb_mbuf += nb_rxd + nb_txd;
877
878         nb_mbuf = RTE_MAX(nb_mbuf, (uint32_t)NB_MBUF);
879
880         snprintf(buf, sizeof(buf), "mbuf_pool_%u_%u", lcore, queue);
881
882         rxq->pool = rte_pktmbuf_pool_create(buf, nb_mbuf, MEMPOOL_CACHE_SIZE, 0,
883                                             MBUF_DATA_SIZE, socket);
884         if (rxq->pool == NULL) {
885                 RTE_LOG(ERR, IP_RSMBL,
886                         "rte_pktmbuf_pool_create(%s) failed", buf);
887                 return -1;
888         }
889
890         return 0;
891 }
892
893 static int
894 init_mem(void)
895 {
896         char buf[PATH_MAX];
897         struct rte_lpm *lpm;
898         struct rte_lpm6 *lpm6;
899         struct rte_lpm_config lpm_config;
900         int socket;
901         unsigned lcore_id;
902
903         /* traverse through lcores and initialize structures on each socket */
904
905         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
906
907                 if (rte_lcore_is_enabled(lcore_id) == 0)
908                         continue;
909
910                 socket = rte_lcore_to_socket_id(lcore_id);
911
912                 if (socket == SOCKET_ID_ANY)
913                         socket = 0;
914
915                 if (socket_lpm[socket] == NULL) {
916                         RTE_LOG(INFO, IP_RSMBL, "Creating LPM table on socket %i\n", socket);
917                         snprintf(buf, sizeof(buf), "IP_RSMBL_LPM_%i", socket);
918
919                         lpm_config.max_rules = LPM_MAX_RULES;
920                         lpm_config.number_tbl8s = 256;
921                         lpm_config.flags = 0;
922
923                         lpm = rte_lpm_create(buf, socket, &lpm_config);
924                         if (lpm == NULL) {
925                                 RTE_LOG(ERR, IP_RSMBL, "Cannot create LPM table\n");
926                                 return -1;
927                         }
928                         socket_lpm[socket] = lpm;
929                 }
930
931                 if (socket_lpm6[socket] == NULL) {
932                         RTE_LOG(INFO, IP_RSMBL, "Creating LPM6 table on socket %i\n", socket);
933                         snprintf(buf, sizeof(buf), "IP_RSMBL_LPM_%i", socket);
934
935                         lpm6 = rte_lpm6_create(buf, socket, &lpm6_config);
936                         if (lpm6 == NULL) {
937                                 RTE_LOG(ERR, IP_RSMBL, "Cannot create LPM table\n");
938                                 return -1;
939                         }
940                         socket_lpm6[socket] = lpm6;
941                 }
942         }
943
944         return 0;
945 }
946
947 static void
948 queue_dump_stat(void)
949 {
950         uint32_t i, lcore;
951         const struct lcore_queue_conf *qconf;
952
953         for (lcore = 0; lcore < RTE_MAX_LCORE; lcore++) {
954                 if (rte_lcore_is_enabled(lcore) == 0)
955                         continue;
956
957                 qconf = &lcore_queue_conf[lcore];
958                 for (i = 0; i < qconf->n_rx_queue; i++) {
959
960                         fprintf(stdout, " -- lcoreid=%u portid=%u "
961                                 "frag tbl stat:\n",
962                                 lcore,  qconf->rx_queue_list[i].portid);
963                         rte_ip_frag_table_statistics_dump(stdout,
964                                         qconf->rx_queue_list[i].frag_tbl);
965                         fprintf(stdout, "TX bursts:\t%" PRIu64 "\n"
966                                 "TX packets _queued:\t%" PRIu64 "\n"
967                                 "TX packets dropped:\t%" PRIu64 "\n"
968                                 "TX packets send:\t%" PRIu64 "\n",
969                                 qconf->tx_stat.call,
970                                 qconf->tx_stat.queue,
971                                 qconf->tx_stat.drop,
972                                 qconf->tx_stat.send);
973                 }
974         }
975 }
976
977 static void
978 signal_handler(int signum)
979 {
980         queue_dump_stat();
981         if (signum != SIGUSR1)
982                 rte_exit(0, "received signal: %d, exiting\n", signum);
983 }
984
985 int
986 main(int argc, char **argv)
987 {
988         struct lcore_queue_conf *qconf;
989         struct rte_eth_dev_info dev_info;
990         struct rte_eth_txconf *txconf;
991         struct rx_queue *rxq;
992         int ret, socket;
993         unsigned nb_ports;
994         uint16_t queueid;
995         unsigned lcore_id = 0, rx_lcore_id = 0;
996         uint32_t n_tx_queue, nb_lcores;
997         uint16_t portid;
998
999         /* init EAL */
1000         ret = rte_eal_init(argc, argv);
1001         if (ret < 0)
1002                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1003         argc -= ret;
1004         argv += ret;
1005
1006         /* parse application arguments (after the EAL ones) */
1007         ret = parse_args(argc, argv);
1008         if (ret < 0)
1009                 rte_exit(EXIT_FAILURE, "Invalid IP reassembly parameters\n");
1010
1011         nb_ports = rte_eth_dev_count();
1012         if (nb_ports == 0)
1013                 rte_exit(EXIT_FAILURE, "No ports found!\n");
1014
1015         nb_lcores = rte_lcore_count();
1016
1017         /* initialize structures (mempools, lpm etc.) */
1018         if (init_mem() < 0)
1019                 rte_panic("Cannot initialize memory structures!\n");
1020
1021         /* check if portmask has non-existent ports */
1022         if (enabled_port_mask & ~(RTE_LEN2MASK(nb_ports, unsigned)))
1023                 rte_exit(EXIT_FAILURE, "Non-existent ports in portmask!\n");
1024
1025         /* initialize all ports */
1026         for (portid = 0; portid < nb_ports; portid++) {
1027                 struct rte_eth_rxconf rxq_conf;
1028                 struct rte_eth_conf local_port_conf = port_conf;
1029
1030                 /* skip ports that are not enabled */
1031                 if ((enabled_port_mask & (1 << portid)) == 0) {
1032                         printf("\nSkipping disabled port %d\n", portid);
1033                         continue;
1034                 }
1035
1036                 qconf = &lcore_queue_conf[rx_lcore_id];
1037
1038                 /* limit the frame size to the maximum supported by NIC */
1039                 rte_eth_dev_info_get(portid, &dev_info);
1040                 local_port_conf.rxmode.max_rx_pkt_len = RTE_MIN(
1041                     dev_info.max_rx_pktlen,
1042                     local_port_conf.rxmode.max_rx_pkt_len);
1043
1044                 /* get the lcore_id for this port */
1045                 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
1046                            qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
1047
1048                         rx_lcore_id++;
1049                         if (rx_lcore_id >= RTE_MAX_LCORE)
1050                                 rte_exit(EXIT_FAILURE, "Not enough cores\n");
1051
1052                         qconf = &lcore_queue_conf[rx_lcore_id];
1053                 }
1054
1055                 socket = rte_lcore_to_socket_id(portid);
1056                 if (socket == SOCKET_ID_ANY)
1057                         socket = 0;
1058
1059                 queueid = qconf->n_rx_queue;
1060                 rxq = &qconf->rx_queue_list[queueid];
1061                 rxq->portid = portid;
1062                 rxq->lpm = socket_lpm[socket];
1063                 rxq->lpm6 = socket_lpm6[socket];
1064
1065                 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
1066                                                        &nb_txd);
1067                 if (ret < 0)
1068                         rte_exit(EXIT_FAILURE,
1069                                  "Cannot adjust number of descriptors: err=%d, port=%d\n",
1070                                  ret, portid);
1071
1072                 if (setup_queue_tbl(rxq, rx_lcore_id, queueid) < 0)
1073                         rte_exit(EXIT_FAILURE, "Failed to set up queue table\n");
1074                 qconf->n_rx_queue++;
1075
1076                 /* init port */
1077                 printf("Initializing port %d ... ", portid );
1078                 fflush(stdout);
1079
1080                 n_tx_queue = nb_lcores;
1081                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
1082                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1083                 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
1084                         local_port_conf.txmode.offloads |=
1085                                 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
1086                 ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
1087                                             &local_port_conf);
1088                 if (ret < 0) {
1089                         printf("\n");
1090                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
1091                                 "err=%d, port=%d\n",
1092                                 ret, portid);
1093                 }
1094
1095                 /* init one RX queue */
1096                 rxq_conf = dev_info.default_rxconf;
1097                 rxq_conf.offloads = local_port_conf.rxmode.offloads;
1098                 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
1099                                              socket, &rxq_conf,
1100                                              rxq->pool);
1101                 if (ret < 0) {
1102                         printf("\n");
1103                         rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: "
1104                                 "err=%d, port=%d\n",
1105                                 ret, portid);
1106                 }
1107
1108                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1109                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1110                 printf("\n");
1111
1112                 /* init one TX queue per couple (lcore,port) */
1113                 queueid = 0;
1114                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1115                         if (rte_lcore_is_enabled(lcore_id) == 0)
1116                                 continue;
1117
1118                         socket = (int) rte_lcore_to_socket_id(lcore_id);
1119
1120                         printf("txq=%u,%d,%d ", lcore_id, queueid, socket);
1121                         fflush(stdout);
1122
1123                         txconf = &dev_info.default_txconf;
1124                         txconf->txq_flags = ETH_TXQ_FLAGS_IGNORE;
1125                         txconf->offloads = local_port_conf.txmode.offloads;
1126
1127                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1128                                         socket, txconf);
1129                         if (ret < 0)
1130                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
1131                                         "port=%d\n", ret, portid);
1132
1133                         qconf = &lcore_queue_conf[lcore_id];
1134                         qconf->tx_queue_id[portid] = queueid;
1135                         setup_port_tbl(qconf, lcore_id, socket, portid);
1136                         queueid++;
1137                 }
1138                 printf("\n");
1139         }
1140
1141         printf("\n");
1142
1143         /* start ports */
1144         for (portid = 0; portid < nb_ports; portid++) {
1145                 if ((enabled_port_mask & (1 << portid)) == 0) {
1146                         continue;
1147                 }
1148                 /* Start device */
1149                 ret = rte_eth_dev_start(portid);
1150                 if (ret < 0)
1151                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1152                                 ret, portid);
1153
1154                 rte_eth_promiscuous_enable(portid);
1155         }
1156
1157         if (init_routing_table() < 0)
1158                 rte_exit(EXIT_FAILURE, "Cannot init routing table\n");
1159
1160         check_all_ports_link_status(nb_ports, enabled_port_mask);
1161
1162         signal(SIGUSR1, signal_handler);
1163         signal(SIGTERM, signal_handler);
1164         signal(SIGINT, signal_handler);
1165
1166         /* launch per-lcore init on every lcore */
1167         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1168         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1169                 if (rte_eal_wait_lcore(lcore_id) < 0)
1170                         return -1;
1171         }
1172
1173         return 0;
1174 }