net/mlx5: fix VLAN push action validation
[dpdk.git] / examples / ipsec-secgw / ipsec-secgw.c
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2016 Intel Corporation
3  */
4
5 #include <stdbool.h>
6 #include <stdio.h>
7 #include <stdlib.h>
8 #include <stdint.h>
9 #include <inttypes.h>
10 #include <sys/types.h>
11 #include <netinet/in.h>
12 #include <netinet/ip.h>
13 #include <netinet/ip6.h>
14 #include <string.h>
15 #include <sys/queue.h>
16 #include <stdarg.h>
17 #include <errno.h>
18 #include <signal.h>
19 #include <getopt.h>
20
21 #include <rte_common.h>
22 #include <rte_bitmap.h>
23 #include <rte_byteorder.h>
24 #include <rte_log.h>
25 #include <rte_eal.h>
26 #include <rte_launch.h>
27 #include <rte_cycles.h>
28 #include <rte_prefetch.h>
29 #include <rte_lcore.h>
30 #include <rte_per_lcore.h>
31 #include <rte_branch_prediction.h>
32 #include <rte_interrupts.h>
33 #include <rte_random.h>
34 #include <rte_debug.h>
35 #include <rte_ether.h>
36 #include <rte_ethdev.h>
37 #include <rte_mempool.h>
38 #include <rte_mbuf.h>
39 #include <rte_acl.h>
40 #include <rte_lpm.h>
41 #include <rte_lpm6.h>
42 #include <rte_hash.h>
43 #include <rte_jhash.h>
44 #include <rte_cryptodev.h>
45 #include <rte_security.h>
46 #include <rte_eventdev.h>
47 #include <rte_ip.h>
48 #include <rte_ip_frag.h>
49 #include <rte_alarm.h>
50 #include <rte_telemetry.h>
51
52 #include "event_helper.h"
53 #include "flow.h"
54 #include "ipsec.h"
55 #include "ipsec_worker.h"
56 #include "parser.h"
57 #include "sad.h"
58
59 volatile bool force_quit;
60
61 #define MAX_JUMBO_PKT_LEN  9600
62
63 #define MEMPOOL_CACHE_SIZE 256
64
65 #define CDEV_QUEUE_DESC 2048
66 #define CDEV_MAP_ENTRIES 16384
67 #define CDEV_MP_CACHE_SZ 64
68 #define CDEV_MP_CACHE_MULTIPLIER 1.5 /* from rte_mempool.c */
69 #define MAX_QUEUE_PAIRS 1
70
71 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
72
73 /* Configure how many packets ahead to prefetch, when reading packets */
74 #define PREFETCH_OFFSET 3
75
76 #define MAX_RX_QUEUE_PER_LCORE 16
77
78 #define MAX_LCORE_PARAMS 1024
79
80 /*
81  * Configurable number of RX/TX ring descriptors
82  */
83 #define IPSEC_SECGW_RX_DESC_DEFAULT 1024
84 #define IPSEC_SECGW_TX_DESC_DEFAULT 1024
85 static uint16_t nb_rxd = IPSEC_SECGW_RX_DESC_DEFAULT;
86 static uint16_t nb_txd = IPSEC_SECGW_TX_DESC_DEFAULT;
87
88 #define ETHADDR_TO_UINT64(addr) __BYTES_TO_UINT64( \
89                 (addr)->addr_bytes[0], (addr)->addr_bytes[1], \
90                 (addr)->addr_bytes[2], (addr)->addr_bytes[3], \
91                 (addr)->addr_bytes[4], (addr)->addr_bytes[5], \
92                 0, 0)
93
94 #define FRAG_TBL_BUCKET_ENTRIES 4
95 #define MAX_FRAG_TTL_NS         (10LL * NS_PER_S)
96
97 #define MTU_TO_FRAMELEN(x)      ((x) + RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN)
98
99 struct ethaddr_info ethaddr_tbl[RTE_MAX_ETHPORTS] = {
100         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x7e, 0x94, 0x9a) },
101         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x22, 0xa1, 0xd9) },
102         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x08, 0x69, 0x26) },
103         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x49, 0x9e, 0xdd) }
104 };
105
106 struct flow_info flow_info_tbl[RTE_MAX_ETHPORTS];
107
108 #define CMD_LINE_OPT_CONFIG             "config"
109 #define CMD_LINE_OPT_SINGLE_SA          "single-sa"
110 #define CMD_LINE_OPT_CRYPTODEV_MASK     "cryptodev_mask"
111 #define CMD_LINE_OPT_TRANSFER_MODE      "transfer-mode"
112 #define CMD_LINE_OPT_SCHEDULE_TYPE      "event-schedule-type"
113 #define CMD_LINE_OPT_RX_OFFLOAD         "rxoffload"
114 #define CMD_LINE_OPT_TX_OFFLOAD         "txoffload"
115 #define CMD_LINE_OPT_REASSEMBLE         "reassemble"
116 #define CMD_LINE_OPT_MTU                "mtu"
117 #define CMD_LINE_OPT_FRAG_TTL           "frag-ttl"
118 #define CMD_LINE_OPT_EVENT_VECTOR       "event-vector"
119 #define CMD_LINE_OPT_VECTOR_SIZE        "vector-size"
120 #define CMD_LINE_OPT_VECTOR_TIMEOUT     "vector-tmo"
121 #define CMD_LINE_OPT_VECTOR_POOL_SZ     "vector-pool-sz"
122 #define CMD_LINE_OPT_PER_PORT_POOL      "per-port-pool"
123
124 #define CMD_LINE_ARG_EVENT      "event"
125 #define CMD_LINE_ARG_POLL       "poll"
126 #define CMD_LINE_ARG_ORDERED    "ordered"
127 #define CMD_LINE_ARG_ATOMIC     "atomic"
128 #define CMD_LINE_ARG_PARALLEL   "parallel"
129
130 enum {
131         /* long options mapped to a short option */
132
133         /* first long only option value must be >= 256, so that we won't
134          * conflict with short options
135          */
136         CMD_LINE_OPT_MIN_NUM = 256,
137         CMD_LINE_OPT_CONFIG_NUM,
138         CMD_LINE_OPT_SINGLE_SA_NUM,
139         CMD_LINE_OPT_CRYPTODEV_MASK_NUM,
140         CMD_LINE_OPT_TRANSFER_MODE_NUM,
141         CMD_LINE_OPT_SCHEDULE_TYPE_NUM,
142         CMD_LINE_OPT_RX_OFFLOAD_NUM,
143         CMD_LINE_OPT_TX_OFFLOAD_NUM,
144         CMD_LINE_OPT_REASSEMBLE_NUM,
145         CMD_LINE_OPT_MTU_NUM,
146         CMD_LINE_OPT_FRAG_TTL_NUM,
147         CMD_LINE_OPT_EVENT_VECTOR_NUM,
148         CMD_LINE_OPT_VECTOR_SIZE_NUM,
149         CMD_LINE_OPT_VECTOR_TIMEOUT_NUM,
150         CMD_LINE_OPT_VECTOR_POOL_SZ_NUM,
151         CMD_LINE_OPT_PER_PORT_POOL_NUM,
152 };
153
154 static const struct option lgopts[] = {
155         {CMD_LINE_OPT_CONFIG, 1, 0, CMD_LINE_OPT_CONFIG_NUM},
156         {CMD_LINE_OPT_SINGLE_SA, 1, 0, CMD_LINE_OPT_SINGLE_SA_NUM},
157         {CMD_LINE_OPT_CRYPTODEV_MASK, 1, 0, CMD_LINE_OPT_CRYPTODEV_MASK_NUM},
158         {CMD_LINE_OPT_TRANSFER_MODE, 1, 0, CMD_LINE_OPT_TRANSFER_MODE_NUM},
159         {CMD_LINE_OPT_SCHEDULE_TYPE, 1, 0, CMD_LINE_OPT_SCHEDULE_TYPE_NUM},
160         {CMD_LINE_OPT_RX_OFFLOAD, 1, 0, CMD_LINE_OPT_RX_OFFLOAD_NUM},
161         {CMD_LINE_OPT_TX_OFFLOAD, 1, 0, CMD_LINE_OPT_TX_OFFLOAD_NUM},
162         {CMD_LINE_OPT_REASSEMBLE, 1, 0, CMD_LINE_OPT_REASSEMBLE_NUM},
163         {CMD_LINE_OPT_MTU, 1, 0, CMD_LINE_OPT_MTU_NUM},
164         {CMD_LINE_OPT_FRAG_TTL, 1, 0, CMD_LINE_OPT_FRAG_TTL_NUM},
165         {CMD_LINE_OPT_EVENT_VECTOR, 0, 0, CMD_LINE_OPT_EVENT_VECTOR_NUM},
166         {CMD_LINE_OPT_VECTOR_SIZE, 1, 0, CMD_LINE_OPT_VECTOR_SIZE_NUM},
167         {CMD_LINE_OPT_VECTOR_TIMEOUT, 1, 0, CMD_LINE_OPT_VECTOR_TIMEOUT_NUM},
168         {CMD_LINE_OPT_VECTOR_POOL_SZ, 1, 0, CMD_LINE_OPT_VECTOR_POOL_SZ_NUM},
169         {CMD_LINE_OPT_PER_PORT_POOL, 0, 0, CMD_LINE_OPT_PER_PORT_POOL_NUM},
170         {NULL, 0, 0, 0}
171 };
172
173 uint32_t unprotected_port_mask;
174 uint32_t single_sa_idx;
175 /* mask of enabled ports */
176 static uint32_t enabled_port_mask;
177 static uint64_t enabled_cryptodev_mask = UINT64_MAX;
178 static int32_t promiscuous_on = 1;
179 static int32_t numa_on = 1; /**< NUMA is enabled by default. */
180 static uint32_t nb_lcores;
181 static uint32_t single_sa;
182 uint32_t nb_bufs_in_pool;
183
184 /*
185  * RX/TX HW offload capabilities to enable/use on ethernet ports.
186  * By default all capabilities are enabled.
187  */
188 static uint64_t dev_rx_offload = UINT64_MAX;
189 static uint64_t dev_tx_offload = UINT64_MAX;
190
191 /*
192  * global values that determine multi-seg policy
193  */
194 static uint32_t frag_tbl_sz;
195 static uint32_t frame_buf_size = RTE_MBUF_DEFAULT_BUF_SIZE;
196 static uint32_t mtu_size = RTE_ETHER_MTU;
197 static uint64_t frag_ttl_ns = MAX_FRAG_TTL_NS;
198 static uint32_t stats_interval;
199
200 /* application wide librte_ipsec/SA parameters */
201 struct app_sa_prm app_sa_prm = {
202                         .enable = 0,
203                         .cache_sz = SA_CACHE_SZ,
204                         .udp_encap = 0
205                 };
206 static const char *cfgfile;
207
208 struct lcore_rx_queue {
209         uint16_t port_id;
210         uint8_t queue_id;
211 } __rte_cache_aligned;
212
213 struct lcore_params {
214         uint16_t port_id;
215         uint8_t queue_id;
216         uint8_t lcore_id;
217 } __rte_cache_aligned;
218
219 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
220
221 static struct lcore_params *lcore_params;
222 static uint16_t nb_lcore_params;
223
224 static struct rte_hash *cdev_map_in;
225 static struct rte_hash *cdev_map_out;
226
227 struct buffer {
228         uint16_t len;
229         struct rte_mbuf *m_table[MAX_PKT_BURST] __rte_aligned(sizeof(void *));
230 };
231
232 struct lcore_conf {
233         uint16_t nb_rx_queue;
234         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
235         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
236         struct buffer tx_mbufs[RTE_MAX_ETHPORTS];
237         struct ipsec_ctx inbound;
238         struct ipsec_ctx outbound;
239         struct rt_ctx *rt4_ctx;
240         struct rt_ctx *rt6_ctx;
241         struct {
242                 struct rte_ip_frag_tbl *tbl;
243                 struct rte_mempool *pool_indir;
244                 struct rte_ip_frag_death_row dr;
245         } frag;
246 } __rte_cache_aligned;
247
248 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
249
250 static struct rte_eth_conf port_conf = {
251         .rxmode = {
252                 .mq_mode        = RTE_ETH_MQ_RX_RSS,
253                 .split_hdr_size = 0,
254                 .offloads = RTE_ETH_RX_OFFLOAD_CHECKSUM,
255         },
256         .rx_adv_conf = {
257                 .rss_conf = {
258                         .rss_key = NULL,
259                         .rss_hf = RTE_ETH_RSS_IP | RTE_ETH_RSS_UDP |
260                                 RTE_ETH_RSS_TCP | RTE_ETH_RSS_SCTP,
261                 },
262         },
263         .txmode = {
264                 .mq_mode = RTE_ETH_MQ_TX_NONE,
265         },
266 };
267
268 struct socket_ctx socket_ctx[NB_SOCKETS];
269
270 bool per_port_pool;
271
272 /*
273  * Determine is multi-segment support required:
274  *  - either frame buffer size is smaller then mtu
275  *  - or reassemble support is requested
276  */
277 static int
278 multi_seg_required(void)
279 {
280         return (MTU_TO_FRAMELEN(mtu_size) + RTE_PKTMBUF_HEADROOM >
281                 frame_buf_size || frag_tbl_sz != 0);
282 }
283
284 static inline void
285 adjust_ipv4_pktlen(struct rte_mbuf *m, const struct rte_ipv4_hdr *iph,
286         uint32_t l2_len)
287 {
288         uint32_t plen, trim;
289
290         plen = rte_be_to_cpu_16(iph->total_length) + l2_len;
291         if (plen < m->pkt_len) {
292                 trim = m->pkt_len - plen;
293                 rte_pktmbuf_trim(m, trim);
294         }
295 }
296
297 static inline void
298 adjust_ipv6_pktlen(struct rte_mbuf *m, const struct rte_ipv6_hdr *iph,
299         uint32_t l2_len)
300 {
301         uint32_t plen, trim;
302
303         plen = rte_be_to_cpu_16(iph->payload_len) + sizeof(*iph) + l2_len;
304         if (plen < m->pkt_len) {
305                 trim = m->pkt_len - plen;
306                 rte_pktmbuf_trim(m, trim);
307         }
308 }
309
310
311 struct ipsec_core_statistics core_statistics[RTE_MAX_LCORE];
312
313 /* Print out statistics on packet distribution */
314 static void
315 print_stats_cb(__rte_unused void *param)
316 {
317         uint64_t total_packets_dropped, total_packets_tx, total_packets_rx;
318         float burst_percent, rx_per_call, tx_per_call;
319         unsigned int coreid;
320
321         total_packets_dropped = 0;
322         total_packets_tx = 0;
323         total_packets_rx = 0;
324
325         const char clr[] = { 27, '[', '2', 'J', '\0' };
326         const char topLeft[] = { 27, '[', '1', ';', '1', 'H', '\0' };
327
328         /* Clear screen and move to top left */
329         printf("%s%s", clr, topLeft);
330
331         printf("\nCore statistics ====================================");
332
333         for (coreid = 0; coreid < RTE_MAX_LCORE; coreid++) {
334                 /* skip disabled cores */
335                 if (rte_lcore_is_enabled(coreid) == 0)
336                         continue;
337                 burst_percent = (float)(core_statistics[coreid].burst_rx * 100)/
338                                         core_statistics[coreid].rx;
339                 rx_per_call =  (float)(core_statistics[coreid].rx)/
340                                        core_statistics[coreid].rx_call;
341                 tx_per_call =  (float)(core_statistics[coreid].tx)/
342                                        core_statistics[coreid].tx_call;
343                 printf("\nStatistics for core %u ------------------------------"
344                            "\nPackets received: %20"PRIu64
345                            "\nPackets sent: %24"PRIu64
346                            "\nPackets dropped: %21"PRIu64
347                            "\nBurst percent: %23.2f"
348                            "\nPackets per Rx call: %17.2f"
349                            "\nPackets per Tx call: %17.2f",
350                            coreid,
351                            core_statistics[coreid].rx,
352                            core_statistics[coreid].tx,
353                            core_statistics[coreid].dropped,
354                            burst_percent,
355                            rx_per_call,
356                            tx_per_call);
357
358                 total_packets_dropped += core_statistics[coreid].dropped;
359                 total_packets_tx += core_statistics[coreid].tx;
360                 total_packets_rx += core_statistics[coreid].rx;
361         }
362         printf("\nAggregate statistics ==============================="
363                    "\nTotal packets received: %14"PRIu64
364                    "\nTotal packets sent: %18"PRIu64
365                    "\nTotal packets dropped: %15"PRIu64,
366                    total_packets_rx,
367                    total_packets_tx,
368                    total_packets_dropped);
369         printf("\n====================================================\n");
370
371         rte_eal_alarm_set(stats_interval * US_PER_S, print_stats_cb, NULL);
372 }
373
374 static inline void
375 prepare_one_packet(struct rte_mbuf *pkt, struct ipsec_traffic *t)
376 {
377         const struct rte_ether_hdr *eth;
378         const struct rte_ipv4_hdr *iph4;
379         const struct rte_ipv6_hdr *iph6;
380         const struct rte_udp_hdr *udp;
381         uint16_t ip4_hdr_len;
382         uint16_t nat_port;
383
384         eth = rte_pktmbuf_mtod(pkt, const struct rte_ether_hdr *);
385         if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
386
387                 iph4 = (const struct rte_ipv4_hdr *)rte_pktmbuf_adj(pkt,
388                         RTE_ETHER_HDR_LEN);
389                 adjust_ipv4_pktlen(pkt, iph4, 0);
390
391                 switch (iph4->next_proto_id) {
392                 case IPPROTO_ESP:
393                         t->ipsec.pkts[(t->ipsec.num)++] = pkt;
394                         break;
395                 case IPPROTO_UDP:
396                         if (app_sa_prm.udp_encap == 1) {
397                                 ip4_hdr_len = ((iph4->version_ihl &
398                                         RTE_IPV4_HDR_IHL_MASK) *
399                                         RTE_IPV4_IHL_MULTIPLIER);
400                                 udp = rte_pktmbuf_mtod_offset(pkt,
401                                         struct rte_udp_hdr *, ip4_hdr_len);
402                                 nat_port = rte_cpu_to_be_16(IPSEC_NAT_T_PORT);
403                                 if (udp->src_port == nat_port ||
404                                         udp->dst_port == nat_port){
405                                         t->ipsec.pkts[(t->ipsec.num)++] = pkt;
406                                         pkt->packet_type |=
407                                                 MBUF_PTYPE_TUNNEL_ESP_IN_UDP;
408                                         break;
409                                 }
410                         }
411                 /* Fall through */
412                 default:
413                         t->ip4.data[t->ip4.num] = &iph4->next_proto_id;
414                         t->ip4.pkts[(t->ip4.num)++] = pkt;
415                 }
416                 pkt->l2_len = 0;
417                 pkt->l3_len = sizeof(*iph4);
418                 pkt->packet_type |= RTE_PTYPE_L3_IPV4;
419                 if  (pkt->packet_type & RTE_PTYPE_L4_TCP)
420                         pkt->l4_len = sizeof(struct rte_tcp_hdr);
421                 else if (pkt->packet_type & RTE_PTYPE_L4_UDP)
422                         pkt->l4_len = sizeof(struct rte_udp_hdr);
423         } else if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6)) {
424                 int next_proto;
425                 size_t l3len, ext_len;
426                 uint8_t *p;
427
428                 /* get protocol type */
429                 iph6 = (const struct rte_ipv6_hdr *)rte_pktmbuf_adj(pkt,
430                         RTE_ETHER_HDR_LEN);
431                 adjust_ipv6_pktlen(pkt, iph6, 0);
432
433                 next_proto = iph6->proto;
434
435                 /* determine l3 header size up to ESP extension */
436                 l3len = sizeof(struct ip6_hdr);
437                 p = rte_pktmbuf_mtod(pkt, uint8_t *);
438                 while (next_proto != IPPROTO_ESP && l3len < pkt->data_len &&
439                         (next_proto = rte_ipv6_get_next_ext(p + l3len,
440                                                 next_proto, &ext_len)) >= 0)
441                         l3len += ext_len;
442
443                 /* drop packet when IPv6 header exceeds first segment length */
444                 if (unlikely(l3len > pkt->data_len)) {
445                         free_pkts(&pkt, 1);
446                         return;
447                 }
448
449                 switch (next_proto) {
450                 case IPPROTO_ESP:
451                         t->ipsec.pkts[(t->ipsec.num)++] = pkt;
452                         break;
453                 case IPPROTO_UDP:
454                         if (app_sa_prm.udp_encap == 1) {
455                                 udp = rte_pktmbuf_mtod_offset(pkt,
456                                         struct rte_udp_hdr *, l3len);
457                                 nat_port = rte_cpu_to_be_16(IPSEC_NAT_T_PORT);
458                                 if (udp->src_port == nat_port ||
459                                         udp->dst_port == nat_port){
460                                         t->ipsec.pkts[(t->ipsec.num)++] = pkt;
461                                         pkt->packet_type |=
462                                                 MBUF_PTYPE_TUNNEL_ESP_IN_UDP;
463                                         break;
464                                 }
465                         }
466                 /* Fall through */
467                 default:
468                         t->ip6.data[t->ip6.num] = &iph6->proto;
469                         t->ip6.pkts[(t->ip6.num)++] = pkt;
470                 }
471                 pkt->l2_len = 0;
472                 pkt->l3_len = l3len;
473                 pkt->packet_type |= RTE_PTYPE_L3_IPV6;
474         } else {
475                 /* Unknown/Unsupported type, drop the packet */
476                 RTE_LOG(ERR, IPSEC, "Unsupported packet type 0x%x\n",
477                         rte_be_to_cpu_16(eth->ether_type));
478                 free_pkts(&pkt, 1);
479                 return;
480         }
481
482         /* Check if the packet has been processed inline. For inline protocol
483          * processed packets, the metadata in the mbuf can be used to identify
484          * the security processing done on the packet. The metadata will be
485          * used to retrieve the application registered userdata associated
486          * with the security session.
487          */
488
489         if (pkt->ol_flags & RTE_MBUF_F_RX_SEC_OFFLOAD &&
490                         rte_security_dynfield_is_registered()) {
491                 struct ipsec_sa *sa;
492                 struct ipsec_mbuf_metadata *priv;
493                 struct rte_security_ctx *ctx = (struct rte_security_ctx *)
494                                                 rte_eth_dev_get_sec_ctx(
495                                                 pkt->port);
496
497                 /* Retrieve the userdata registered. Here, the userdata
498                  * registered is the SA pointer.
499                  */
500                 sa = (struct ipsec_sa *)rte_security_get_userdata(ctx,
501                                 *rte_security_dynfield(pkt));
502                 if (sa == NULL) {
503                         /* userdata could not be retrieved */
504                         return;
505                 }
506
507                 /* Save SA as priv member in mbuf. This will be used in the
508                  * IPsec selector(SP-SA) check.
509                  */
510
511                 priv = get_priv(pkt);
512                 priv->sa = sa;
513         }
514 }
515
516 static inline void
517 prepare_traffic(struct rte_mbuf **pkts, struct ipsec_traffic *t,
518                 uint16_t nb_pkts)
519 {
520         int32_t i;
521
522         t->ipsec.num = 0;
523         t->ip4.num = 0;
524         t->ip6.num = 0;
525
526         for (i = 0; i < (nb_pkts - PREFETCH_OFFSET); i++) {
527                 rte_prefetch0(rte_pktmbuf_mtod(pkts[i + PREFETCH_OFFSET],
528                                         void *));
529                 prepare_one_packet(pkts[i], t);
530         }
531         /* Process left packets */
532         for (; i < nb_pkts; i++)
533                 prepare_one_packet(pkts[i], t);
534 }
535
536 static inline void
537 prepare_tx_pkt(struct rte_mbuf *pkt, uint16_t port,
538                 const struct lcore_conf *qconf)
539 {
540         struct ip *ip;
541         struct rte_ether_hdr *ethhdr;
542
543         ip = rte_pktmbuf_mtod(pkt, struct ip *);
544
545         ethhdr = (struct rte_ether_hdr *)
546                 rte_pktmbuf_prepend(pkt, RTE_ETHER_HDR_LEN);
547
548         if (ip->ip_v == IPVERSION) {
549                 pkt->ol_flags |= qconf->outbound.ipv4_offloads;
550                 pkt->l3_len = sizeof(struct ip);
551                 pkt->l2_len = RTE_ETHER_HDR_LEN;
552
553                 ip->ip_sum = 0;
554
555                 /* calculate IPv4 cksum in SW */
556                 if ((pkt->ol_flags & RTE_MBUF_F_TX_IP_CKSUM) == 0)
557                         ip->ip_sum = rte_ipv4_cksum((struct rte_ipv4_hdr *)ip);
558
559                 ethhdr->ether_type = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4);
560         } else {
561                 pkt->ol_flags |= qconf->outbound.ipv6_offloads;
562                 pkt->l3_len = sizeof(struct ip6_hdr);
563                 pkt->l2_len = RTE_ETHER_HDR_LEN;
564
565                 ethhdr->ether_type = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6);
566         }
567
568         memcpy(&ethhdr->src_addr, &ethaddr_tbl[port].src,
569                         sizeof(struct rte_ether_addr));
570         memcpy(&ethhdr->dst_addr, &ethaddr_tbl[port].dst,
571                         sizeof(struct rte_ether_addr));
572 }
573
574 static inline void
575 prepare_tx_burst(struct rte_mbuf *pkts[], uint16_t nb_pkts, uint16_t port,
576                 const struct lcore_conf *qconf)
577 {
578         int32_t i;
579         const int32_t prefetch_offset = 2;
580
581         for (i = 0; i < (nb_pkts - prefetch_offset); i++) {
582                 rte_mbuf_prefetch_part2(pkts[i + prefetch_offset]);
583                 prepare_tx_pkt(pkts[i], port, qconf);
584         }
585         /* Process left packets */
586         for (; i < nb_pkts; i++)
587                 prepare_tx_pkt(pkts[i], port, qconf);
588 }
589
590 /* Send burst of packets on an output interface */
591 static inline int32_t
592 send_burst(struct lcore_conf *qconf, uint16_t n, uint16_t port)
593 {
594         struct rte_mbuf **m_table;
595         int32_t ret;
596         uint16_t queueid;
597
598         queueid = qconf->tx_queue_id[port];
599         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
600
601         prepare_tx_burst(m_table, n, port, qconf);
602
603         ret = rte_eth_tx_burst(port, queueid, m_table, n);
604
605         core_stats_update_tx(ret);
606
607         if (unlikely(ret < n)) {
608                 do {
609                         free_pkts(&m_table[ret], 1);
610                 } while (++ret < n);
611         }
612
613         return 0;
614 }
615
616 /*
617  * Helper function to fragment and queue for TX one packet.
618  */
619 static inline uint32_t
620 send_fragment_packet(struct lcore_conf *qconf, struct rte_mbuf *m,
621         uint16_t port, uint8_t proto)
622 {
623         struct buffer *tbl;
624         uint32_t len, n;
625         int32_t rc;
626
627         tbl =  qconf->tx_mbufs + port;
628         len = tbl->len;
629
630         /* free space for new fragments */
631         if (len + RTE_LIBRTE_IP_FRAG_MAX_FRAG >=  RTE_DIM(tbl->m_table)) {
632                 send_burst(qconf, len, port);
633                 len = 0;
634         }
635
636         n = RTE_DIM(tbl->m_table) - len;
637
638         if (proto == IPPROTO_IP)
639                 rc = rte_ipv4_fragment_packet(m, tbl->m_table + len,
640                         n, mtu_size, m->pool, qconf->frag.pool_indir);
641         else
642                 rc = rte_ipv6_fragment_packet(m, tbl->m_table + len,
643                         n, mtu_size, m->pool, qconf->frag.pool_indir);
644
645         if (rc >= 0)
646                 len += rc;
647         else
648                 RTE_LOG(ERR, IPSEC,
649                         "%s: failed to fragment packet with size %u, "
650                         "error code: %d\n",
651                         __func__, m->pkt_len, rte_errno);
652
653         free_pkts(&m, 1);
654         return len;
655 }
656
657 /* Enqueue a single packet, and send burst if queue is filled */
658 static inline int32_t
659 send_single_packet(struct rte_mbuf *m, uint16_t port, uint8_t proto)
660 {
661         uint32_t lcore_id;
662         uint16_t len;
663         struct lcore_conf *qconf;
664
665         lcore_id = rte_lcore_id();
666
667         qconf = &lcore_conf[lcore_id];
668         len = qconf->tx_mbufs[port].len;
669
670         if (m->pkt_len <= mtu_size) {
671                 qconf->tx_mbufs[port].m_table[len] = m;
672                 len++;
673
674         /* need to fragment the packet */
675         } else if (frag_tbl_sz > 0)
676                 len = send_fragment_packet(qconf, m, port, proto);
677         else
678                 free_pkts(&m, 1);
679
680         /* enough pkts to be sent */
681         if (unlikely(len == MAX_PKT_BURST)) {
682                 send_burst(qconf, MAX_PKT_BURST, port);
683                 len = 0;
684         }
685
686         qconf->tx_mbufs[port].len = len;
687         return 0;
688 }
689
690 static inline void
691 inbound_sp_sa(struct sp_ctx *sp, struct sa_ctx *sa, struct traffic_type *ip,
692                 uint16_t lim, struct ipsec_spd_stats *stats)
693 {
694         struct rte_mbuf *m;
695         uint32_t i, j, res, sa_idx;
696
697         if (ip->num == 0 || sp == NULL)
698                 return;
699
700         rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
701                         ip->num, DEFAULT_MAX_CATEGORIES);
702
703         j = 0;
704         for (i = 0; i < ip->num; i++) {
705                 m = ip->pkts[i];
706                 res = ip->res[i];
707                 if (res == BYPASS) {
708                         ip->pkts[j++] = m;
709                         stats->bypass++;
710                         continue;
711                 }
712                 if (res == DISCARD) {
713                         free_pkts(&m, 1);
714                         stats->discard++;
715                         continue;
716                 }
717
718                 /* Only check SPI match for processed IPSec packets */
719                 if (i < lim && ((m->ol_flags & RTE_MBUF_F_RX_SEC_OFFLOAD) == 0)) {
720                         stats->discard++;
721                         free_pkts(&m, 1);
722                         continue;
723                 }
724
725                 sa_idx = res - 1;
726                 if (!inbound_sa_check(sa, m, sa_idx)) {
727                         stats->discard++;
728                         free_pkts(&m, 1);
729                         continue;
730                 }
731                 ip->pkts[j++] = m;
732                 stats->protect++;
733         }
734         ip->num = j;
735 }
736
737 static void
738 split46_traffic(struct ipsec_traffic *trf, struct rte_mbuf *mb[], uint32_t num)
739 {
740         uint32_t i, n4, n6;
741         struct ip *ip;
742         struct rte_mbuf *m;
743
744         n4 = trf->ip4.num;
745         n6 = trf->ip6.num;
746
747         for (i = 0; i < num; i++) {
748
749                 m = mb[i];
750                 ip = rte_pktmbuf_mtod(m, struct ip *);
751
752                 if (ip->ip_v == IPVERSION) {
753                         trf->ip4.pkts[n4] = m;
754                         trf->ip4.data[n4] = rte_pktmbuf_mtod_offset(m,
755                                         uint8_t *, offsetof(struct ip, ip_p));
756                         n4++;
757                 } else if (ip->ip_v == IP6_VERSION) {
758                         trf->ip6.pkts[n6] = m;
759                         trf->ip6.data[n6] = rte_pktmbuf_mtod_offset(m,
760                                         uint8_t *,
761                                         offsetof(struct ip6_hdr, ip6_nxt));
762                         n6++;
763                 } else
764                         free_pkts(&m, 1);
765         }
766
767         trf->ip4.num = n4;
768         trf->ip6.num = n6;
769 }
770
771
772 static inline void
773 process_pkts_inbound(struct ipsec_ctx *ipsec_ctx,
774                 struct ipsec_traffic *traffic)
775 {
776         unsigned int lcoreid = rte_lcore_id();
777         uint16_t nb_pkts_in, n_ip4, n_ip6;
778
779         n_ip4 = traffic->ip4.num;
780         n_ip6 = traffic->ip6.num;
781
782         if (app_sa_prm.enable == 0) {
783                 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
784                                 traffic->ipsec.num, MAX_PKT_BURST);
785                 split46_traffic(traffic, traffic->ipsec.pkts, nb_pkts_in);
786         } else {
787                 inbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.pkts,
788                         traffic->ipsec.saptr, traffic->ipsec.num);
789                 ipsec_process(ipsec_ctx, traffic);
790         }
791
792         inbound_sp_sa(ipsec_ctx->sp4_ctx,
793                 ipsec_ctx->sa_ctx, &traffic->ip4, n_ip4,
794                 &core_statistics[lcoreid].inbound.spd4);
795
796         inbound_sp_sa(ipsec_ctx->sp6_ctx,
797                 ipsec_ctx->sa_ctx, &traffic->ip6, n_ip6,
798                 &core_statistics[lcoreid].inbound.spd6);
799 }
800
801 static inline void
802 outbound_spd_lookup(struct sp_ctx *sp,
803                 struct traffic_type *ip,
804                 struct traffic_type *ipsec,
805                 struct ipsec_spd_stats *stats)
806 {
807         struct rte_mbuf *m;
808         uint32_t i, j, sa_idx;
809
810         if (ip->num == 0 || sp == NULL)
811                 return;
812
813         rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
814                         ip->num, DEFAULT_MAX_CATEGORIES);
815
816         for (i = 0, j = 0; i < ip->num; i++) {
817                 m = ip->pkts[i];
818                 sa_idx = ip->res[i] - 1;
819
820                 if (unlikely(ip->res[i] == DISCARD)) {
821                         free_pkts(&m, 1);
822
823                         stats->discard++;
824                 } else if (unlikely(ip->res[i] == BYPASS)) {
825                         ip->pkts[j++] = m;
826
827                         stats->bypass++;
828                 } else {
829                         ipsec->res[ipsec->num] = sa_idx;
830                         ipsec->pkts[ipsec->num++] = m;
831
832                         stats->protect++;
833                 }
834         }
835         ip->num = j;
836 }
837
838 static inline void
839 process_pkts_outbound(struct ipsec_ctx *ipsec_ctx,
840                 struct ipsec_traffic *traffic)
841 {
842         struct rte_mbuf *m;
843         uint16_t idx, nb_pkts_out, i;
844         unsigned int lcoreid = rte_lcore_id();
845
846         /* Drop any IPsec traffic from protected ports */
847         free_pkts(traffic->ipsec.pkts, traffic->ipsec.num);
848
849         traffic->ipsec.num = 0;
850
851         outbound_spd_lookup(ipsec_ctx->sp4_ctx,
852                 &traffic->ip4, &traffic->ipsec,
853                 &core_statistics[lcoreid].outbound.spd4);
854
855         outbound_spd_lookup(ipsec_ctx->sp6_ctx,
856                 &traffic->ip6, &traffic->ipsec,
857                 &core_statistics[lcoreid].outbound.spd6);
858
859         if (app_sa_prm.enable == 0) {
860
861                 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
862                                 traffic->ipsec.res, traffic->ipsec.num,
863                                 MAX_PKT_BURST);
864
865                 for (i = 0; i < nb_pkts_out; i++) {
866                         m = traffic->ipsec.pkts[i];
867                         struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
868                         if (ip->ip_v == IPVERSION) {
869                                 idx = traffic->ip4.num++;
870                                 traffic->ip4.pkts[idx] = m;
871                         } else {
872                                 idx = traffic->ip6.num++;
873                                 traffic->ip6.pkts[idx] = m;
874                         }
875                 }
876         } else {
877                 outbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.res,
878                         traffic->ipsec.saptr, traffic->ipsec.num);
879                 ipsec_process(ipsec_ctx, traffic);
880         }
881 }
882
883 static inline void
884 process_pkts_inbound_nosp(struct ipsec_ctx *ipsec_ctx,
885                 struct ipsec_traffic *traffic)
886 {
887         struct rte_mbuf *m;
888         uint32_t nb_pkts_in, i, idx;
889
890         if (app_sa_prm.enable == 0) {
891
892                 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
893                                 traffic->ipsec.num, MAX_PKT_BURST);
894
895                 for (i = 0; i < nb_pkts_in; i++) {
896                         m = traffic->ipsec.pkts[i];
897                         struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
898                         if (ip->ip_v == IPVERSION) {
899                                 idx = traffic->ip4.num++;
900                                 traffic->ip4.pkts[idx] = m;
901                         } else {
902                                 idx = traffic->ip6.num++;
903                                 traffic->ip6.pkts[idx] = m;
904                         }
905                 }
906         } else {
907                 inbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.pkts,
908                         traffic->ipsec.saptr, traffic->ipsec.num);
909                 ipsec_process(ipsec_ctx, traffic);
910         }
911 }
912
913 static inline void
914 process_pkts_outbound_nosp(struct ipsec_ctx *ipsec_ctx,
915                 struct ipsec_traffic *traffic)
916 {
917         struct rte_mbuf *m;
918         uint32_t nb_pkts_out, i, n;
919         struct ip *ip;
920
921         /* Drop any IPsec traffic from protected ports */
922         free_pkts(traffic->ipsec.pkts, traffic->ipsec.num);
923
924         n = 0;
925
926         for (i = 0; i < traffic->ip4.num; i++) {
927                 traffic->ipsec.pkts[n] = traffic->ip4.pkts[i];
928                 traffic->ipsec.res[n++] = single_sa_idx;
929         }
930
931         for (i = 0; i < traffic->ip6.num; i++) {
932                 traffic->ipsec.pkts[n] = traffic->ip6.pkts[i];
933                 traffic->ipsec.res[n++] = single_sa_idx;
934         }
935
936         traffic->ip4.num = 0;
937         traffic->ip6.num = 0;
938         traffic->ipsec.num = n;
939
940         if (app_sa_prm.enable == 0) {
941
942                 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
943                                 traffic->ipsec.res, traffic->ipsec.num,
944                                 MAX_PKT_BURST);
945
946                 /* They all sue the same SA (ip4 or ip6 tunnel) */
947                 m = traffic->ipsec.pkts[0];
948                 ip = rte_pktmbuf_mtod(m, struct ip *);
949                 if (ip->ip_v == IPVERSION) {
950                         traffic->ip4.num = nb_pkts_out;
951                         for (i = 0; i < nb_pkts_out; i++)
952                                 traffic->ip4.pkts[i] = traffic->ipsec.pkts[i];
953                 } else {
954                         traffic->ip6.num = nb_pkts_out;
955                         for (i = 0; i < nb_pkts_out; i++)
956                                 traffic->ip6.pkts[i] = traffic->ipsec.pkts[i];
957                 }
958         } else {
959                 outbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.res,
960                         traffic->ipsec.saptr, traffic->ipsec.num);
961                 ipsec_process(ipsec_ctx, traffic);
962         }
963 }
964
965 static inline int32_t
966 get_hop_for_offload_pkt(struct rte_mbuf *pkt, int is_ipv6)
967 {
968         struct ipsec_mbuf_metadata *priv;
969         struct ipsec_sa *sa;
970
971         priv = get_priv(pkt);
972
973         sa = priv->sa;
974         if (unlikely(sa == NULL)) {
975                 RTE_LOG(ERR, IPSEC, "SA not saved in private data\n");
976                 goto fail;
977         }
978
979         if (is_ipv6)
980                 return sa->portid;
981
982         /* else */
983         return (sa->portid | RTE_LPM_LOOKUP_SUCCESS);
984
985 fail:
986         if (is_ipv6)
987                 return -1;
988
989         /* else */
990         return 0;
991 }
992
993 static inline void
994 route4_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
995 {
996         uint32_t hop[MAX_PKT_BURST * 2];
997         uint32_t dst_ip[MAX_PKT_BURST * 2];
998         int32_t pkt_hop = 0;
999         uint16_t i, offset;
1000         uint16_t lpm_pkts = 0;
1001         unsigned int lcoreid = rte_lcore_id();
1002
1003         if (nb_pkts == 0)
1004                 return;
1005
1006         /* Need to do an LPM lookup for non-inline packets. Inline packets will
1007          * have port ID in the SA
1008          */
1009
1010         for (i = 0; i < nb_pkts; i++) {
1011                 if (!(pkts[i]->ol_flags & RTE_MBUF_F_TX_SEC_OFFLOAD)) {
1012                         /* Security offload not enabled. So an LPM lookup is
1013                          * required to get the hop
1014                          */
1015                         offset = offsetof(struct ip, ip_dst);
1016                         dst_ip[lpm_pkts] = *rte_pktmbuf_mtod_offset(pkts[i],
1017                                         uint32_t *, offset);
1018                         dst_ip[lpm_pkts] = rte_be_to_cpu_32(dst_ip[lpm_pkts]);
1019                         lpm_pkts++;
1020                 }
1021         }
1022
1023         rte_lpm_lookup_bulk((struct rte_lpm *)rt_ctx, dst_ip, hop, lpm_pkts);
1024
1025         lpm_pkts = 0;
1026
1027         for (i = 0; i < nb_pkts; i++) {
1028                 if (pkts[i]->ol_flags & RTE_MBUF_F_TX_SEC_OFFLOAD) {
1029                         /* Read hop from the SA */
1030                         pkt_hop = get_hop_for_offload_pkt(pkts[i], 0);
1031                 } else {
1032                         /* Need to use hop returned by lookup */
1033                         pkt_hop = hop[lpm_pkts++];
1034                 }
1035
1036                 if ((pkt_hop & RTE_LPM_LOOKUP_SUCCESS) == 0) {
1037                         core_statistics[lcoreid].lpm4.miss++;
1038                         free_pkts(&pkts[i], 1);
1039                         continue;
1040                 }
1041                 send_single_packet(pkts[i], pkt_hop & 0xff, IPPROTO_IP);
1042         }
1043 }
1044
1045 static inline void
1046 route6_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
1047 {
1048         int32_t hop[MAX_PKT_BURST * 2];
1049         uint8_t dst_ip[MAX_PKT_BURST * 2][16];
1050         uint8_t *ip6_dst;
1051         int32_t pkt_hop = 0;
1052         uint16_t i, offset;
1053         uint16_t lpm_pkts = 0;
1054         unsigned int lcoreid = rte_lcore_id();
1055
1056         if (nb_pkts == 0)
1057                 return;
1058
1059         /* Need to do an LPM lookup for non-inline packets. Inline packets will
1060          * have port ID in the SA
1061          */
1062
1063         for (i = 0; i < nb_pkts; i++) {
1064                 if (!(pkts[i]->ol_flags & RTE_MBUF_F_TX_SEC_OFFLOAD)) {
1065                         /* Security offload not enabled. So an LPM lookup is
1066                          * required to get the hop
1067                          */
1068                         offset = offsetof(struct ip6_hdr, ip6_dst);
1069                         ip6_dst = rte_pktmbuf_mtod_offset(pkts[i], uint8_t *,
1070                                         offset);
1071                         memcpy(&dst_ip[lpm_pkts][0], ip6_dst, 16);
1072                         lpm_pkts++;
1073                 }
1074         }
1075
1076         rte_lpm6_lookup_bulk_func((struct rte_lpm6 *)rt_ctx, dst_ip, hop,
1077                         lpm_pkts);
1078
1079         lpm_pkts = 0;
1080
1081         for (i = 0; i < nb_pkts; i++) {
1082                 if (pkts[i]->ol_flags & RTE_MBUF_F_TX_SEC_OFFLOAD) {
1083                         /* Read hop from the SA */
1084                         pkt_hop = get_hop_for_offload_pkt(pkts[i], 1);
1085                 } else {
1086                         /* Need to use hop returned by lookup */
1087                         pkt_hop = hop[lpm_pkts++];
1088                 }
1089
1090                 if (pkt_hop == -1) {
1091                         core_statistics[lcoreid].lpm6.miss++;
1092                         free_pkts(&pkts[i], 1);
1093                         continue;
1094                 }
1095                 send_single_packet(pkts[i], pkt_hop & 0xff, IPPROTO_IPV6);
1096         }
1097 }
1098
1099 static inline void
1100 process_pkts(struct lcore_conf *qconf, struct rte_mbuf **pkts,
1101                 uint8_t nb_pkts, uint16_t portid)
1102 {
1103         struct ipsec_traffic traffic;
1104
1105         prepare_traffic(pkts, &traffic, nb_pkts);
1106
1107         if (unlikely(single_sa)) {
1108                 if (is_unprotected_port(portid))
1109                         process_pkts_inbound_nosp(&qconf->inbound, &traffic);
1110                 else
1111                         process_pkts_outbound_nosp(&qconf->outbound, &traffic);
1112         } else {
1113                 if (is_unprotected_port(portid))
1114                         process_pkts_inbound(&qconf->inbound, &traffic);
1115                 else
1116                         process_pkts_outbound(&qconf->outbound, &traffic);
1117         }
1118
1119         route4_pkts(qconf->rt4_ctx, traffic.ip4.pkts, traffic.ip4.num);
1120         route6_pkts(qconf->rt6_ctx, traffic.ip6.pkts, traffic.ip6.num);
1121 }
1122
1123 static inline void
1124 drain_tx_buffers(struct lcore_conf *qconf)
1125 {
1126         struct buffer *buf;
1127         uint32_t portid;
1128
1129         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
1130                 buf = &qconf->tx_mbufs[portid];
1131                 if (buf->len == 0)
1132                         continue;
1133                 send_burst(qconf, buf->len, portid);
1134                 buf->len = 0;
1135         }
1136 }
1137
1138 static inline void
1139 drain_crypto_buffers(struct lcore_conf *qconf)
1140 {
1141         uint32_t i;
1142         struct ipsec_ctx *ctx;
1143
1144         /* drain inbound buffers*/
1145         ctx = &qconf->inbound;
1146         for (i = 0; i != ctx->nb_qps; i++) {
1147                 if (ctx->tbl[i].len != 0)
1148                         enqueue_cop_burst(ctx->tbl  + i);
1149         }
1150
1151         /* drain outbound buffers*/
1152         ctx = &qconf->outbound;
1153         for (i = 0; i != ctx->nb_qps; i++) {
1154                 if (ctx->tbl[i].len != 0)
1155                         enqueue_cop_burst(ctx->tbl  + i);
1156         }
1157 }
1158
1159 static void
1160 drain_inbound_crypto_queues(const struct lcore_conf *qconf,
1161                 struct ipsec_ctx *ctx)
1162 {
1163         uint32_t n;
1164         struct ipsec_traffic trf;
1165         unsigned int lcoreid = rte_lcore_id();
1166
1167         if (app_sa_prm.enable == 0) {
1168
1169                 /* dequeue packets from crypto-queue */
1170                 n = ipsec_inbound_cqp_dequeue(ctx, trf.ipsec.pkts,
1171                         RTE_DIM(trf.ipsec.pkts));
1172
1173                 trf.ip4.num = 0;
1174                 trf.ip6.num = 0;
1175
1176                 /* split traffic by ipv4-ipv6 */
1177                 split46_traffic(&trf, trf.ipsec.pkts, n);
1178         } else
1179                 ipsec_cqp_process(ctx, &trf);
1180
1181         /* process ipv4 packets */
1182         if (trf.ip4.num != 0) {
1183                 inbound_sp_sa(ctx->sp4_ctx, ctx->sa_ctx, &trf.ip4, 0,
1184                         &core_statistics[lcoreid].inbound.spd4);
1185                 route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
1186         }
1187
1188         /* process ipv6 packets */
1189         if (trf.ip6.num != 0) {
1190                 inbound_sp_sa(ctx->sp6_ctx, ctx->sa_ctx, &trf.ip6, 0,
1191                         &core_statistics[lcoreid].inbound.spd6);
1192                 route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
1193         }
1194 }
1195
1196 static void
1197 drain_outbound_crypto_queues(const struct lcore_conf *qconf,
1198                 struct ipsec_ctx *ctx)
1199 {
1200         uint32_t n;
1201         struct ipsec_traffic trf;
1202
1203         if (app_sa_prm.enable == 0) {
1204
1205                 /* dequeue packets from crypto-queue */
1206                 n = ipsec_outbound_cqp_dequeue(ctx, trf.ipsec.pkts,
1207                         RTE_DIM(trf.ipsec.pkts));
1208
1209                 trf.ip4.num = 0;
1210                 trf.ip6.num = 0;
1211
1212                 /* split traffic by ipv4-ipv6 */
1213                 split46_traffic(&trf, trf.ipsec.pkts, n);
1214         } else
1215                 ipsec_cqp_process(ctx, &trf);
1216
1217         /* process ipv4 packets */
1218         if (trf.ip4.num != 0)
1219                 route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
1220
1221         /* process ipv6 packets */
1222         if (trf.ip6.num != 0)
1223                 route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
1224 }
1225
1226 /* main processing loop */
1227 void
1228 ipsec_poll_mode_worker(void)
1229 {
1230         struct rte_mbuf *pkts[MAX_PKT_BURST];
1231         uint32_t lcore_id;
1232         uint64_t prev_tsc, diff_tsc, cur_tsc;
1233         int32_t i, nb_rx;
1234         uint16_t portid;
1235         uint8_t queueid;
1236         struct lcore_conf *qconf;
1237         int32_t rc, socket_id;
1238         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
1239                         / US_PER_S * BURST_TX_DRAIN_US;
1240         struct lcore_rx_queue *rxql;
1241
1242         prev_tsc = 0;
1243         lcore_id = rte_lcore_id();
1244         qconf = &lcore_conf[lcore_id];
1245         rxql = qconf->rx_queue_list;
1246         socket_id = rte_lcore_to_socket_id(lcore_id);
1247
1248         qconf->rt4_ctx = socket_ctx[socket_id].rt_ip4;
1249         qconf->rt6_ctx = socket_ctx[socket_id].rt_ip6;
1250         qconf->inbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_in;
1251         qconf->inbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_in;
1252         qconf->inbound.sa_ctx = socket_ctx[socket_id].sa_in;
1253         qconf->inbound.cdev_map = cdev_map_in;
1254         qconf->inbound.session_pool = socket_ctx[socket_id].session_pool;
1255         qconf->inbound.session_priv_pool =
1256                         socket_ctx[socket_id].session_priv_pool;
1257         qconf->outbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_out;
1258         qconf->outbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_out;
1259         qconf->outbound.sa_ctx = socket_ctx[socket_id].sa_out;
1260         qconf->outbound.cdev_map = cdev_map_out;
1261         qconf->outbound.session_pool = socket_ctx[socket_id].session_pool;
1262         qconf->outbound.session_priv_pool =
1263                         socket_ctx[socket_id].session_priv_pool;
1264         qconf->frag.pool_indir = socket_ctx[socket_id].mbuf_pool_indir;
1265
1266         rc = ipsec_sad_lcore_cache_init(app_sa_prm.cache_sz);
1267         if (rc != 0) {
1268                 RTE_LOG(ERR, IPSEC,
1269                         "SAD cache init on lcore %u, failed with code: %d\n",
1270                         lcore_id, rc);
1271                 return;
1272         }
1273
1274         if (qconf->nb_rx_queue == 0) {
1275                 RTE_LOG(DEBUG, IPSEC, "lcore %u has nothing to do\n",
1276                         lcore_id);
1277                 return;
1278         }
1279
1280         RTE_LOG(INFO, IPSEC, "entering main loop on lcore %u\n", lcore_id);
1281
1282         for (i = 0; i < qconf->nb_rx_queue; i++) {
1283                 portid = rxql[i].port_id;
1284                 queueid = rxql[i].queue_id;
1285                 RTE_LOG(INFO, IPSEC,
1286                         " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
1287                         lcore_id, portid, queueid);
1288         }
1289
1290         while (!force_quit) {
1291                 cur_tsc = rte_rdtsc();
1292
1293                 /* TX queue buffer drain */
1294                 diff_tsc = cur_tsc - prev_tsc;
1295
1296                 if (unlikely(diff_tsc > drain_tsc)) {
1297                         drain_tx_buffers(qconf);
1298                         drain_crypto_buffers(qconf);
1299                         prev_tsc = cur_tsc;
1300                 }
1301
1302                 for (i = 0; i < qconf->nb_rx_queue; ++i) {
1303
1304                         /* Read packets from RX queues */
1305                         portid = rxql[i].port_id;
1306                         queueid = rxql[i].queue_id;
1307                         nb_rx = rte_eth_rx_burst(portid, queueid,
1308                                         pkts, MAX_PKT_BURST);
1309
1310                         if (nb_rx > 0) {
1311                                 core_stats_update_rx(nb_rx);
1312                                 process_pkts(qconf, pkts, nb_rx, portid);
1313                         }
1314
1315                         /* dequeue and process completed crypto-ops */
1316                         if (is_unprotected_port(portid))
1317                                 drain_inbound_crypto_queues(qconf,
1318                                         &qconf->inbound);
1319                         else
1320                                 drain_outbound_crypto_queues(qconf,
1321                                         &qconf->outbound);
1322                 }
1323         }
1324 }
1325
1326 int
1327 check_flow_params(uint16_t fdir_portid, uint8_t fdir_qid)
1328 {
1329         uint16_t i;
1330         uint16_t portid;
1331         uint8_t queueid;
1332
1333         for (i = 0; i < nb_lcore_params; ++i) {
1334                 portid = lcore_params_array[i].port_id;
1335                 if (portid == fdir_portid) {
1336                         queueid = lcore_params_array[i].queue_id;
1337                         if (queueid == fdir_qid)
1338                                 break;
1339                 }
1340
1341                 if (i == nb_lcore_params - 1)
1342                         return -1;
1343         }
1344
1345         return 1;
1346 }
1347
1348 static int32_t
1349 check_poll_mode_params(struct eh_conf *eh_conf)
1350 {
1351         uint8_t lcore;
1352         uint16_t portid;
1353         uint16_t i;
1354         int32_t socket_id;
1355
1356         if (!eh_conf)
1357                 return -EINVAL;
1358
1359         if (eh_conf->mode != EH_PKT_TRANSFER_MODE_POLL)
1360                 return 0;
1361
1362         if (lcore_params == NULL) {
1363                 printf("Error: No port/queue/core mappings\n");
1364                 return -1;
1365         }
1366
1367         for (i = 0; i < nb_lcore_params; ++i) {
1368                 lcore = lcore_params[i].lcore_id;
1369                 if (!rte_lcore_is_enabled(lcore)) {
1370                         printf("error: lcore %hhu is not enabled in "
1371                                 "lcore mask\n", lcore);
1372                         return -1;
1373                 }
1374                 socket_id = rte_lcore_to_socket_id(lcore);
1375                 if (socket_id != 0 && numa_on == 0) {
1376                         printf("warning: lcore %hhu is on socket %d "
1377                                 "with numa off\n",
1378                                 lcore, socket_id);
1379                 }
1380                 portid = lcore_params[i].port_id;
1381                 if ((enabled_port_mask & (1 << portid)) == 0) {
1382                         printf("port %u is not enabled in port mask\n", portid);
1383                         return -1;
1384                 }
1385                 if (!rte_eth_dev_is_valid_port(portid)) {
1386                         printf("port %u is not present on the board\n", portid);
1387                         return -1;
1388                 }
1389         }
1390         return 0;
1391 }
1392
1393 static uint8_t
1394 get_port_nb_rx_queues(const uint16_t port)
1395 {
1396         int32_t queue = -1;
1397         uint16_t i;
1398
1399         for (i = 0; i < nb_lcore_params; ++i) {
1400                 if (lcore_params[i].port_id == port &&
1401                                 lcore_params[i].queue_id > queue)
1402                         queue = lcore_params[i].queue_id;
1403         }
1404         return (uint8_t)(++queue);
1405 }
1406
1407 static int32_t
1408 init_lcore_rx_queues(void)
1409 {
1410         uint16_t i, nb_rx_queue;
1411         uint8_t lcore;
1412
1413         for (i = 0; i < nb_lcore_params; ++i) {
1414                 lcore = lcore_params[i].lcore_id;
1415                 nb_rx_queue = lcore_conf[lcore].nb_rx_queue;
1416                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1417                         printf("error: too many queues (%u) for lcore: %u\n",
1418                                         nb_rx_queue + 1, lcore);
1419                         return -1;
1420                 }
1421                 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1422                         lcore_params[i].port_id;
1423                 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1424                         lcore_params[i].queue_id;
1425                 lcore_conf[lcore].nb_rx_queue++;
1426         }
1427         return 0;
1428 }
1429
1430 /* display usage */
1431 static void
1432 print_usage(const char *prgname)
1433 {
1434         fprintf(stderr, "%s [EAL options] --"
1435                 " -p PORTMASK"
1436                 " [-P]"
1437                 " [-u PORTMASK]"
1438                 " [-j FRAMESIZE]"
1439                 " [-l]"
1440                 " [-w REPLAY_WINDOW_SIZE]"
1441                 " [-e]"
1442                 " [-a]"
1443                 " [-c]"
1444                 " [-t STATS_INTERVAL]"
1445                 " [-s NUMBER_OF_MBUFS_IN_PKT_POOL]"
1446                 " -f CONFIG_FILE"
1447                 " --config (port,queue,lcore)[,(port,queue,lcore)]"
1448                 " [--single-sa SAIDX]"
1449                 " [--cryptodev_mask MASK]"
1450                 " [--transfer-mode MODE]"
1451                 " [--event-schedule-type TYPE]"
1452                 " [--" CMD_LINE_OPT_RX_OFFLOAD " RX_OFFLOAD_MASK]"
1453                 " [--" CMD_LINE_OPT_TX_OFFLOAD " TX_OFFLOAD_MASK]"
1454                 " [--" CMD_LINE_OPT_REASSEMBLE " REASSEMBLE_TABLE_SIZE]"
1455                 " [--" CMD_LINE_OPT_MTU " MTU]"
1456                 " [--event-vector]"
1457                 " [--vector-size SIZE]"
1458                 " [--vector-tmo TIMEOUT in ns]"
1459                 "\n\n"
1460                 "  -p PORTMASK: Hexadecimal bitmask of ports to configure\n"
1461                 "  -P : Enable promiscuous mode\n"
1462                 "  -u PORTMASK: Hexadecimal bitmask of unprotected ports\n"
1463                 "  -j FRAMESIZE: Data buffer size, minimum (and default)\n"
1464                 "     value: RTE_MBUF_DEFAULT_BUF_SIZE\n"
1465                 "  -l enables code-path that uses librte_ipsec\n"
1466                 "  -w REPLAY_WINDOW_SIZE specifies IPsec SQN replay window\n"
1467                 "     size for each SA\n"
1468                 "  -e enables ESN\n"
1469                 "  -a enables SA SQN atomic behaviour\n"
1470                 "  -c specifies inbound SAD cache size,\n"
1471                 "     zero value disables the cache (default value: 128)\n"
1472                 "  -t specifies statistics screen update interval,\n"
1473                 "     zero disables statistics screen (default value: 0)\n"
1474                 "  -s number of mbufs in packet pool, if not specified number\n"
1475                 "     of mbufs will be calculated based on number of cores,\n"
1476                 "     ports and crypto queues\n"
1477                 "  -f CONFIG_FILE: Configuration file\n"
1478                 "  --config (port,queue,lcore): Rx queue configuration. In poll\n"
1479                 "                               mode determines which queues from\n"
1480                 "                               which ports are mapped to which cores.\n"
1481                 "                               In event mode this option is not used\n"
1482                 "                               as packets are dynamically scheduled\n"
1483                 "                               to cores by HW.\n"
1484                 "  --single-sa SAIDX: In poll mode use single SA index for\n"
1485                 "                     outbound traffic, bypassing the SP\n"
1486                 "                     In event mode selects driver submode,\n"
1487                 "                     SA index value is ignored\n"
1488                 "  --cryptodev_mask MASK: Hexadecimal bitmask of the crypto\n"
1489                 "                         devices to configure\n"
1490                 "  --transfer-mode MODE\n"
1491                 "               \"poll\"  : Packet transfer via polling (default)\n"
1492                 "               \"event\" : Packet transfer via event device\n"
1493                 "  --event-schedule-type TYPE queue schedule type, used only when\n"
1494                 "                             transfer mode is set to event\n"
1495                 "               \"ordered\"  : Ordered (default)\n"
1496                 "               \"atomic\"   : Atomic\n"
1497                 "               \"parallel\" : Parallel\n"
1498                 "  --" CMD_LINE_OPT_RX_OFFLOAD
1499                 ": bitmask of the RX HW offload capabilities to enable/use\n"
1500                 "                         (RTE_ETH_RX_OFFLOAD_*)\n"
1501                 "  --" CMD_LINE_OPT_TX_OFFLOAD
1502                 ": bitmask of the TX HW offload capabilities to enable/use\n"
1503                 "                         (RTE_ETH_TX_OFFLOAD_*)\n"
1504                 "  --" CMD_LINE_OPT_REASSEMBLE " NUM"
1505                 ": max number of entries in reassemble(fragment) table\n"
1506                 "    (zero (default value) disables reassembly)\n"
1507                 "  --" CMD_LINE_OPT_MTU " MTU"
1508                 ": MTU value on all ports (default value: 1500)\n"
1509                 "    outgoing packets with bigger size will be fragmented\n"
1510                 "    incoming packets with bigger size will be discarded\n"
1511                 "  --" CMD_LINE_OPT_FRAG_TTL " FRAG_TTL_NS"
1512                 ": fragments lifetime in nanoseconds, default\n"
1513                 "    and maximum value is 10.000.000.000 ns (10 s)\n"
1514                 "  --event-vector enables event vectorization\n"
1515                 "  --vector-size Max vector size (default value: 16)\n"
1516                 "  --vector-tmo Max vector timeout in nanoseconds"
1517                 "    (default value: 102400)\n"
1518                 "  --" CMD_LINE_OPT_PER_PORT_POOL " Enable per port mbuf pool\n"
1519                 "  --" CMD_LINE_OPT_VECTOR_POOL_SZ " Vector pool size\n"
1520                 "                    (default value is based on mbuf count)\n"
1521                 "\n",
1522                 prgname);
1523 }
1524
1525 static int
1526 parse_mask(const char *str, uint64_t *val)
1527 {
1528         char *end;
1529         unsigned long t;
1530
1531         errno = 0;
1532         t = strtoul(str, &end, 0);
1533         if (errno != 0 || end[0] != 0)
1534                 return -EINVAL;
1535
1536         *val = t;
1537         return 0;
1538 }
1539
1540 static int32_t
1541 parse_portmask(const char *portmask)
1542 {
1543         char *end = NULL;
1544         unsigned long pm;
1545
1546         errno = 0;
1547
1548         /* parse hexadecimal string */
1549         pm = strtoul(portmask, &end, 16);
1550         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1551                 return -1;
1552
1553         if ((pm == 0) && errno)
1554                 return -1;
1555
1556         return pm;
1557 }
1558
1559 static int64_t
1560 parse_decimal(const char *str)
1561 {
1562         char *end = NULL;
1563         uint64_t num;
1564
1565         num = strtoull(str, &end, 10);
1566         if ((str[0] == '\0') || (end == NULL) || (*end != '\0')
1567                 || num > INT64_MAX)
1568                 return -1;
1569
1570         return num;
1571 }
1572
1573 static int32_t
1574 parse_config(const char *q_arg)
1575 {
1576         char s[256];
1577         const char *p, *p0 = q_arg;
1578         char *end;
1579         enum fieldnames {
1580                 FLD_PORT = 0,
1581                 FLD_QUEUE,
1582                 FLD_LCORE,
1583                 _NUM_FLD
1584         };
1585         unsigned long int_fld[_NUM_FLD];
1586         char *str_fld[_NUM_FLD];
1587         int32_t i;
1588         uint32_t size;
1589
1590         nb_lcore_params = 0;
1591
1592         while ((p = strchr(p0, '(')) != NULL) {
1593                 ++p;
1594                 p0 = strchr(p, ')');
1595                 if (p0 == NULL)
1596                         return -1;
1597
1598                 size = p0 - p;
1599                 if (size >= sizeof(s))
1600                         return -1;
1601
1602                 snprintf(s, sizeof(s), "%.*s", size, p);
1603                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1604                                 _NUM_FLD)
1605                         return -1;
1606                 for (i = 0; i < _NUM_FLD; i++) {
1607                         errno = 0;
1608                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1609                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
1610                                 return -1;
1611                 }
1612                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1613                         printf("exceeded max number of lcore params: %hu\n",
1614                                 nb_lcore_params);
1615                         return -1;
1616                 }
1617                 lcore_params_array[nb_lcore_params].port_id =
1618                         (uint8_t)int_fld[FLD_PORT];
1619                 lcore_params_array[nb_lcore_params].queue_id =
1620                         (uint8_t)int_fld[FLD_QUEUE];
1621                 lcore_params_array[nb_lcore_params].lcore_id =
1622                         (uint8_t)int_fld[FLD_LCORE];
1623                 ++nb_lcore_params;
1624         }
1625         lcore_params = lcore_params_array;
1626         return 0;
1627 }
1628
1629 static void
1630 print_app_sa_prm(const struct app_sa_prm *prm)
1631 {
1632         printf("librte_ipsec usage: %s\n",
1633                 (prm->enable == 0) ? "disabled" : "enabled");
1634
1635         printf("replay window size: %u\n", prm->window_size);
1636         printf("ESN: %s\n", (prm->enable_esn == 0) ? "disabled" : "enabled");
1637         printf("SA flags: %#" PRIx64 "\n", prm->flags);
1638         printf("Frag TTL: %" PRIu64 " ns\n", frag_ttl_ns);
1639 }
1640
1641 static int
1642 parse_transfer_mode(struct eh_conf *conf, const char *optarg)
1643 {
1644         if (!strcmp(CMD_LINE_ARG_POLL, optarg))
1645                 conf->mode = EH_PKT_TRANSFER_MODE_POLL;
1646         else if (!strcmp(CMD_LINE_ARG_EVENT, optarg))
1647                 conf->mode = EH_PKT_TRANSFER_MODE_EVENT;
1648         else {
1649                 printf("Unsupported packet transfer mode\n");
1650                 return -EINVAL;
1651         }
1652
1653         return 0;
1654 }
1655
1656 static int
1657 parse_schedule_type(struct eh_conf *conf, const char *optarg)
1658 {
1659         struct eventmode_conf *em_conf = NULL;
1660
1661         /* Get eventmode conf */
1662         em_conf = conf->mode_params;
1663
1664         if (!strcmp(CMD_LINE_ARG_ORDERED, optarg))
1665                 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_ORDERED;
1666         else if (!strcmp(CMD_LINE_ARG_ATOMIC, optarg))
1667                 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_ATOMIC;
1668         else if (!strcmp(CMD_LINE_ARG_PARALLEL, optarg))
1669                 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_PARALLEL;
1670         else {
1671                 printf("Unsupported queue schedule type\n");
1672                 return -EINVAL;
1673         }
1674
1675         return 0;
1676 }
1677
1678 static int32_t
1679 parse_args(int32_t argc, char **argv, struct eh_conf *eh_conf)
1680 {
1681         int opt;
1682         int64_t ret;
1683         char **argvopt;
1684         int32_t option_index;
1685         char *prgname = argv[0];
1686         int32_t f_present = 0;
1687         struct eventmode_conf *em_conf = NULL;
1688
1689         argvopt = argv;
1690
1691         while ((opt = getopt_long(argc, argvopt, "aelp:Pu:f:j:w:c:t:s:",
1692                                 lgopts, &option_index)) != EOF) {
1693
1694                 switch (opt) {
1695                 case 'p':
1696                         enabled_port_mask = parse_portmask(optarg);
1697                         if (enabled_port_mask == 0) {
1698                                 printf("invalid portmask\n");
1699                                 print_usage(prgname);
1700                                 return -1;
1701                         }
1702                         break;
1703                 case 'P':
1704                         printf("Promiscuous mode selected\n");
1705                         promiscuous_on = 1;
1706                         break;
1707                 case 'u':
1708                         unprotected_port_mask = parse_portmask(optarg);
1709                         if (unprotected_port_mask == 0) {
1710                                 printf("invalid unprotected portmask\n");
1711                                 print_usage(prgname);
1712                                 return -1;
1713                         }
1714                         break;
1715                 case 'f':
1716                         if (f_present == 1) {
1717                                 printf("\"-f\" option present more than "
1718                                         "once!\n");
1719                                 print_usage(prgname);
1720                                 return -1;
1721                         }
1722                         cfgfile = optarg;
1723                         f_present = 1;
1724                         break;
1725
1726                 case 's':
1727                         ret = parse_decimal(optarg);
1728                         if (ret < 0) {
1729                                 printf("Invalid number of buffers in a pool: "
1730                                         "%s\n", optarg);
1731                                 print_usage(prgname);
1732                                 return -1;
1733                         }
1734
1735                         nb_bufs_in_pool = ret;
1736                         break;
1737
1738                 case 'j':
1739                         ret = parse_decimal(optarg);
1740                         if (ret < RTE_MBUF_DEFAULT_BUF_SIZE ||
1741                                         ret > UINT16_MAX) {
1742                                 printf("Invalid frame buffer size value: %s\n",
1743                                         optarg);
1744                                 print_usage(prgname);
1745                                 return -1;
1746                         }
1747                         frame_buf_size = ret;
1748                         printf("Custom frame buffer size %u\n", frame_buf_size);
1749                         break;
1750                 case 'l':
1751                         app_sa_prm.enable = 1;
1752                         break;
1753                 case 'w':
1754                         app_sa_prm.window_size = parse_decimal(optarg);
1755                         break;
1756                 case 'e':
1757                         app_sa_prm.enable_esn = 1;
1758                         break;
1759                 case 'a':
1760                         app_sa_prm.enable = 1;
1761                         app_sa_prm.flags |= RTE_IPSEC_SAFLAG_SQN_ATOM;
1762                         break;
1763                 case 'c':
1764                         ret = parse_decimal(optarg);
1765                         if (ret < 0) {
1766                                 printf("Invalid SA cache size: %s\n", optarg);
1767                                 print_usage(prgname);
1768                                 return -1;
1769                         }
1770                         app_sa_prm.cache_sz = ret;
1771                         break;
1772                 case 't':
1773                         ret = parse_decimal(optarg);
1774                         if (ret < 0) {
1775                                 printf("Invalid interval value: %s\n", optarg);
1776                                 print_usage(prgname);
1777                                 return -1;
1778                         }
1779                         stats_interval = ret;
1780                         break;
1781                 case CMD_LINE_OPT_CONFIG_NUM:
1782                         ret = parse_config(optarg);
1783                         if (ret) {
1784                                 printf("Invalid config\n");
1785                                 print_usage(prgname);
1786                                 return -1;
1787                         }
1788                         break;
1789                 case CMD_LINE_OPT_SINGLE_SA_NUM:
1790                         ret = parse_decimal(optarg);
1791                         if (ret == -1 || ret > UINT32_MAX) {
1792                                 printf("Invalid argument[sa_idx]\n");
1793                                 print_usage(prgname);
1794                                 return -1;
1795                         }
1796
1797                         /* else */
1798                         single_sa = 1;
1799                         single_sa_idx = ret;
1800                         eh_conf->ipsec_mode = EH_IPSEC_MODE_TYPE_DRIVER;
1801                         printf("Configured with single SA index %u\n",
1802                                         single_sa_idx);
1803                         break;
1804                 case CMD_LINE_OPT_CRYPTODEV_MASK_NUM:
1805                         ret = parse_portmask(optarg);
1806                         if (ret == -1) {
1807                                 printf("Invalid argument[portmask]\n");
1808                                 print_usage(prgname);
1809                                 return -1;
1810                         }
1811
1812                         /* else */
1813                         enabled_cryptodev_mask = ret;
1814                         break;
1815
1816                 case CMD_LINE_OPT_TRANSFER_MODE_NUM:
1817                         ret = parse_transfer_mode(eh_conf, optarg);
1818                         if (ret < 0) {
1819                                 printf("Invalid packet transfer mode\n");
1820                                 print_usage(prgname);
1821                                 return -1;
1822                         }
1823                         break;
1824
1825                 case CMD_LINE_OPT_SCHEDULE_TYPE_NUM:
1826                         ret = parse_schedule_type(eh_conf, optarg);
1827                         if (ret < 0) {
1828                                 printf("Invalid queue schedule type\n");
1829                                 print_usage(prgname);
1830                                 return -1;
1831                         }
1832                         break;
1833
1834                 case CMD_LINE_OPT_RX_OFFLOAD_NUM:
1835                         ret = parse_mask(optarg, &dev_rx_offload);
1836                         if (ret != 0) {
1837                                 printf("Invalid argument for \'%s\': %s\n",
1838                                         CMD_LINE_OPT_RX_OFFLOAD, optarg);
1839                                 print_usage(prgname);
1840                                 return -1;
1841                         }
1842                         break;
1843                 case CMD_LINE_OPT_TX_OFFLOAD_NUM:
1844                         ret = parse_mask(optarg, &dev_tx_offload);
1845                         if (ret != 0) {
1846                                 printf("Invalid argument for \'%s\': %s\n",
1847                                         CMD_LINE_OPT_TX_OFFLOAD, optarg);
1848                                 print_usage(prgname);
1849                                 return -1;
1850                         }
1851                         break;
1852                 case CMD_LINE_OPT_REASSEMBLE_NUM:
1853                         ret = parse_decimal(optarg);
1854                         if (ret < 0 || ret > UINT32_MAX) {
1855                                 printf("Invalid argument for \'%s\': %s\n",
1856                                         CMD_LINE_OPT_REASSEMBLE, optarg);
1857                                 print_usage(prgname);
1858                                 return -1;
1859                         }
1860                         frag_tbl_sz = ret;
1861                         break;
1862                 case CMD_LINE_OPT_MTU_NUM:
1863                         ret = parse_decimal(optarg);
1864                         if (ret < 0 || ret > RTE_IPV4_MAX_PKT_LEN) {
1865                                 printf("Invalid argument for \'%s\': %s\n",
1866                                         CMD_LINE_OPT_MTU, optarg);
1867                                 print_usage(prgname);
1868                                 return -1;
1869                         }
1870                         mtu_size = ret;
1871                         break;
1872                 case CMD_LINE_OPT_FRAG_TTL_NUM:
1873                         ret = parse_decimal(optarg);
1874                         if (ret < 0 || ret > MAX_FRAG_TTL_NS) {
1875                                 printf("Invalid argument for \'%s\': %s\n",
1876                                         CMD_LINE_OPT_MTU, optarg);
1877                                 print_usage(prgname);
1878                                 return -1;
1879                         }
1880                         frag_ttl_ns = ret;
1881                         break;
1882                 case CMD_LINE_OPT_EVENT_VECTOR_NUM:
1883                         em_conf = eh_conf->mode_params;
1884                         em_conf->ext_params.event_vector = 1;
1885                         break;
1886                 case CMD_LINE_OPT_VECTOR_SIZE_NUM:
1887                         ret = parse_decimal(optarg);
1888
1889                         if (ret > MAX_PKT_BURST) {
1890                                 printf("Invalid argument for \'%s\': %s\n",
1891                                         CMD_LINE_OPT_VECTOR_SIZE, optarg);
1892                                 print_usage(prgname);
1893                                 return -1;
1894                         }
1895                         em_conf = eh_conf->mode_params;
1896                         em_conf->ext_params.vector_size = ret;
1897                         break;
1898                 case CMD_LINE_OPT_VECTOR_TIMEOUT_NUM:
1899                         ret = parse_decimal(optarg);
1900
1901                         em_conf = eh_conf->mode_params;
1902                         em_conf->vector_tmo_ns = ret;
1903                         break;
1904                 case CMD_LINE_OPT_VECTOR_POOL_SZ_NUM:
1905                         ret = parse_decimal(optarg);
1906
1907                         em_conf = eh_conf->mode_params;
1908                         em_conf->vector_pool_sz = ret;
1909                         break;
1910                 case CMD_LINE_OPT_PER_PORT_POOL_NUM:
1911                         per_port_pool = 1;
1912                         break;
1913                 default:
1914                         print_usage(prgname);
1915                         return -1;
1916                 }
1917         }
1918
1919         if (f_present == 0) {
1920                 printf("Mandatory option \"-f\" not present\n");
1921                 return -1;
1922         }
1923
1924         /* check do we need to enable multi-seg support */
1925         if (multi_seg_required()) {
1926                 /* legacy mode doesn't support multi-seg */
1927                 app_sa_prm.enable = 1;
1928                 printf("frame buf size: %u, mtu: %u, "
1929                         "number of reassemble entries: %u\n"
1930                         "multi-segment support is required\n",
1931                         frame_buf_size, mtu_size, frag_tbl_sz);
1932         }
1933
1934         print_app_sa_prm(&app_sa_prm);
1935
1936         if (optind >= 0)
1937                 argv[optind-1] = prgname;
1938
1939         ret = optind-1;
1940         optind = 1; /* reset getopt lib */
1941         return ret;
1942 }
1943
1944 static void
1945 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
1946 {
1947         char buf[RTE_ETHER_ADDR_FMT_SIZE];
1948         rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
1949         printf("%s%s", name, buf);
1950 }
1951
1952 /*
1953  * Update destination ethaddr for the port.
1954  */
1955 int
1956 add_dst_ethaddr(uint16_t port, const struct rte_ether_addr *addr)
1957 {
1958         if (port >= RTE_DIM(ethaddr_tbl))
1959                 return -EINVAL;
1960
1961         ethaddr_tbl[port].dst = ETHADDR_TO_UINT64(addr);
1962         return 0;
1963 }
1964
1965 /* Check the link status of all ports in up to 9s, and print them finally */
1966 static void
1967 check_all_ports_link_status(uint32_t port_mask)
1968 {
1969 #define CHECK_INTERVAL 100 /* 100ms */
1970 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1971         uint16_t portid;
1972         uint8_t count, all_ports_up, print_flag = 0;
1973         struct rte_eth_link link;
1974         int ret;
1975         char link_status_text[RTE_ETH_LINK_MAX_STR_LEN];
1976
1977         printf("\nChecking link status");
1978         fflush(stdout);
1979         for (count = 0; count <= MAX_CHECK_TIME; count++) {
1980                 all_ports_up = 1;
1981                 RTE_ETH_FOREACH_DEV(portid) {
1982                         if ((port_mask & (1 << portid)) == 0)
1983                                 continue;
1984                         memset(&link, 0, sizeof(link));
1985                         ret = rte_eth_link_get_nowait(portid, &link);
1986                         if (ret < 0) {
1987                                 all_ports_up = 0;
1988                                 if (print_flag == 1)
1989                                         printf("Port %u link get failed: %s\n",
1990                                                 portid, rte_strerror(-ret));
1991                                 continue;
1992                         }
1993                         /* print link status if flag set */
1994                         if (print_flag == 1) {
1995                                 rte_eth_link_to_str(link_status_text,
1996                                         sizeof(link_status_text), &link);
1997                                 printf("Port %d %s\n", portid,
1998                                        link_status_text);
1999                                 continue;
2000                         }
2001                         /* clear all_ports_up flag if any link down */
2002                         if (link.link_status == RTE_ETH_LINK_DOWN) {
2003                                 all_ports_up = 0;
2004                                 break;
2005                         }
2006                 }
2007                 /* after finally printing all link status, get out */
2008                 if (print_flag == 1)
2009                         break;
2010
2011                 if (all_ports_up == 0) {
2012                         printf(".");
2013                         fflush(stdout);
2014                         rte_delay_ms(CHECK_INTERVAL);
2015                 }
2016
2017                 /* set the print_flag if all ports up or timeout */
2018                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
2019                         print_flag = 1;
2020                         printf("done\n");
2021                 }
2022         }
2023 }
2024
2025 static int32_t
2026 add_mapping(struct rte_hash *map, const char *str, uint16_t cdev_id,
2027                 uint16_t qp, struct lcore_params *params,
2028                 struct ipsec_ctx *ipsec_ctx,
2029                 const struct rte_cryptodev_capabilities *cipher,
2030                 const struct rte_cryptodev_capabilities *auth,
2031                 const struct rte_cryptodev_capabilities *aead)
2032 {
2033         int32_t ret = 0;
2034         unsigned long i;
2035         struct cdev_key key = { 0 };
2036
2037         key.lcore_id = params->lcore_id;
2038         if (cipher)
2039                 key.cipher_algo = cipher->sym.cipher.algo;
2040         if (auth)
2041                 key.auth_algo = auth->sym.auth.algo;
2042         if (aead)
2043                 key.aead_algo = aead->sym.aead.algo;
2044
2045         ret = rte_hash_lookup(map, &key);
2046         if (ret != -ENOENT)
2047                 return 0;
2048
2049         for (i = 0; i < ipsec_ctx->nb_qps; i++)
2050                 if (ipsec_ctx->tbl[i].id == cdev_id)
2051                         break;
2052
2053         if (i == ipsec_ctx->nb_qps) {
2054                 if (ipsec_ctx->nb_qps == MAX_QP_PER_LCORE) {
2055                         printf("Maximum number of crypto devices assigned to "
2056                                 "a core, increase MAX_QP_PER_LCORE value\n");
2057                         return 0;
2058                 }
2059                 ipsec_ctx->tbl[i].id = cdev_id;
2060                 ipsec_ctx->tbl[i].qp = qp;
2061                 ipsec_ctx->nb_qps++;
2062                 printf("%s cdev mapping: lcore %u using cdev %u qp %u "
2063                                 "(cdev_id_qp %lu)\n", str, key.lcore_id,
2064                                 cdev_id, qp, i);
2065         }
2066
2067         ret = rte_hash_add_key_data(map, &key, (void *)i);
2068         if (ret < 0) {
2069                 printf("Failed to insert cdev mapping for (lcore %u, "
2070                                 "cdev %u, qp %u), errno %d\n",
2071                                 key.lcore_id, ipsec_ctx->tbl[i].id,
2072                                 ipsec_ctx->tbl[i].qp, ret);
2073                 return 0;
2074         }
2075
2076         return 1;
2077 }
2078
2079 static int32_t
2080 add_cdev_mapping(struct rte_cryptodev_info *dev_info, uint16_t cdev_id,
2081                 uint16_t qp, struct lcore_params *params)
2082 {
2083         int32_t ret = 0;
2084         const struct rte_cryptodev_capabilities *i, *j;
2085         struct rte_hash *map;
2086         struct lcore_conf *qconf;
2087         struct ipsec_ctx *ipsec_ctx;
2088         const char *str;
2089
2090         qconf = &lcore_conf[params->lcore_id];
2091
2092         if ((unprotected_port_mask & (1 << params->port_id)) == 0) {
2093                 map = cdev_map_out;
2094                 ipsec_ctx = &qconf->outbound;
2095                 str = "Outbound";
2096         } else {
2097                 map = cdev_map_in;
2098                 ipsec_ctx = &qconf->inbound;
2099                 str = "Inbound";
2100         }
2101
2102         /* Required cryptodevs with operation chaining */
2103         if (!(dev_info->feature_flags &
2104                                 RTE_CRYPTODEV_FF_SYM_OPERATION_CHAINING))
2105                 return ret;
2106
2107         for (i = dev_info->capabilities;
2108                         i->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; i++) {
2109                 if (i->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
2110                         continue;
2111
2112                 if (i->sym.xform_type == RTE_CRYPTO_SYM_XFORM_AEAD) {
2113                         ret |= add_mapping(map, str, cdev_id, qp, params,
2114                                         ipsec_ctx, NULL, NULL, i);
2115                         continue;
2116                 }
2117
2118                 if (i->sym.xform_type != RTE_CRYPTO_SYM_XFORM_CIPHER)
2119                         continue;
2120
2121                 for (j = dev_info->capabilities;
2122                                 j->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; j++) {
2123                         if (j->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
2124                                 continue;
2125
2126                         if (j->sym.xform_type != RTE_CRYPTO_SYM_XFORM_AUTH)
2127                                 continue;
2128
2129                         ret |= add_mapping(map, str, cdev_id, qp, params,
2130                                                 ipsec_ctx, i, j, NULL);
2131                 }
2132         }
2133
2134         return ret;
2135 }
2136
2137 /* Check if the device is enabled by cryptodev_mask */
2138 static int
2139 check_cryptodev_mask(uint8_t cdev_id)
2140 {
2141         if (enabled_cryptodev_mask & (1 << cdev_id))
2142                 return 0;
2143
2144         return -1;
2145 }
2146
2147 static uint16_t
2148 cryptodevs_init(uint16_t req_queue_num)
2149 {
2150         struct rte_cryptodev_config dev_conf;
2151         struct rte_cryptodev_qp_conf qp_conf;
2152         uint16_t idx, max_nb_qps, qp, total_nb_qps, i;
2153         int16_t cdev_id;
2154         struct rte_hash_parameters params = { 0 };
2155
2156         const uint64_t mseg_flag = multi_seg_required() ?
2157                                 RTE_CRYPTODEV_FF_IN_PLACE_SGL : 0;
2158
2159         params.entries = CDEV_MAP_ENTRIES;
2160         params.key_len = sizeof(struct cdev_key);
2161         params.hash_func = rte_jhash;
2162         params.hash_func_init_val = 0;
2163         params.socket_id = rte_socket_id();
2164
2165         params.name = "cdev_map_in";
2166         cdev_map_in = rte_hash_create(&params);
2167         if (cdev_map_in == NULL)
2168                 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
2169                                 rte_errno);
2170
2171         params.name = "cdev_map_out";
2172         cdev_map_out = rte_hash_create(&params);
2173         if (cdev_map_out == NULL)
2174                 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
2175                                 rte_errno);
2176
2177         printf("lcore/cryptodev/qp mappings:\n");
2178
2179         idx = 0;
2180         total_nb_qps = 0;
2181         for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
2182                 struct rte_cryptodev_info cdev_info;
2183
2184                 if (check_cryptodev_mask((uint8_t)cdev_id))
2185                         continue;
2186
2187                 rte_cryptodev_info_get(cdev_id, &cdev_info);
2188
2189                 if ((mseg_flag & cdev_info.feature_flags) != mseg_flag)
2190                         rte_exit(EXIT_FAILURE,
2191                                 "Device %hd does not support \'%s\' feature\n",
2192                                 cdev_id,
2193                                 rte_cryptodev_get_feature_name(mseg_flag));
2194
2195                 if (nb_lcore_params > cdev_info.max_nb_queue_pairs)
2196                         max_nb_qps = cdev_info.max_nb_queue_pairs;
2197                 else
2198                         max_nb_qps = nb_lcore_params;
2199
2200                 qp = 0;
2201                 i = 0;
2202                 while (qp < max_nb_qps && i < nb_lcore_params) {
2203                         if (add_cdev_mapping(&cdev_info, cdev_id, qp,
2204                                                 &lcore_params[idx]))
2205                                 qp++;
2206                         idx++;
2207                         idx = idx % nb_lcore_params;
2208                         i++;
2209                 }
2210
2211                 qp = RTE_MIN(max_nb_qps, RTE_MAX(req_queue_num, qp));
2212                 if (qp == 0)
2213                         continue;
2214
2215                 total_nb_qps += qp;
2216                 dev_conf.socket_id = rte_cryptodev_socket_id(cdev_id);
2217                 dev_conf.nb_queue_pairs = qp;
2218                 dev_conf.ff_disable = RTE_CRYPTODEV_FF_ASYMMETRIC_CRYPTO;
2219
2220                 uint32_t dev_max_sess = cdev_info.sym.max_nb_sessions;
2221                 if (dev_max_sess != 0 &&
2222                                 dev_max_sess < get_nb_crypto_sessions())
2223                         rte_exit(EXIT_FAILURE,
2224                                 "Device does not support at least %u "
2225                                 "sessions", get_nb_crypto_sessions());
2226
2227                 if (rte_cryptodev_configure(cdev_id, &dev_conf))
2228                         rte_panic("Failed to initialize cryptodev %u\n",
2229                                         cdev_id);
2230
2231                 qp_conf.nb_descriptors = CDEV_QUEUE_DESC;
2232                 qp_conf.mp_session =
2233                         socket_ctx[dev_conf.socket_id].session_pool;
2234                 qp_conf.mp_session_private =
2235                         socket_ctx[dev_conf.socket_id].session_priv_pool;
2236                 for (qp = 0; qp < dev_conf.nb_queue_pairs; qp++)
2237                         if (rte_cryptodev_queue_pair_setup(cdev_id, qp,
2238                                         &qp_conf, dev_conf.socket_id))
2239                                 rte_panic("Failed to setup queue %u for "
2240                                                 "cdev_id %u\n", 0, cdev_id);
2241
2242                 if (rte_cryptodev_start(cdev_id))
2243                         rte_panic("Failed to start cryptodev %u\n",
2244                                         cdev_id);
2245         }
2246
2247         printf("\n");
2248
2249         return total_nb_qps;
2250 }
2251
2252 static void
2253 port_init(uint16_t portid, uint64_t req_rx_offloads, uint64_t req_tx_offloads)
2254 {
2255         struct rte_eth_dev_info dev_info;
2256         struct rte_eth_txconf *txconf;
2257         uint16_t nb_tx_queue, nb_rx_queue;
2258         uint16_t tx_queueid, rx_queueid, queue, lcore_id;
2259         int32_t ret, socket_id;
2260         struct lcore_conf *qconf;
2261         struct rte_ether_addr ethaddr;
2262         struct rte_eth_conf local_port_conf = port_conf;
2263
2264         ret = rte_eth_dev_info_get(portid, &dev_info);
2265         if (ret != 0)
2266                 rte_exit(EXIT_FAILURE,
2267                         "Error during getting device (port %u) info: %s\n",
2268                         portid, strerror(-ret));
2269
2270         /* limit allowed HW offloads, as user requested */
2271         dev_info.rx_offload_capa &= dev_rx_offload;
2272         dev_info.tx_offload_capa &= dev_tx_offload;
2273
2274         printf("Configuring device port %u:\n", portid);
2275
2276         ret = rte_eth_macaddr_get(portid, &ethaddr);
2277         if (ret != 0)
2278                 rte_exit(EXIT_FAILURE,
2279                         "Error getting MAC address (port %u): %s\n",
2280                         portid, rte_strerror(-ret));
2281
2282         ethaddr_tbl[portid].src = ETHADDR_TO_UINT64(&ethaddr);
2283         print_ethaddr("Address: ", &ethaddr);
2284         printf("\n");
2285
2286         nb_rx_queue = get_port_nb_rx_queues(portid);
2287         nb_tx_queue = nb_lcores;
2288
2289         if (nb_rx_queue > dev_info.max_rx_queues)
2290                 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
2291                                 "(max rx queue is %u)\n",
2292                                 nb_rx_queue, dev_info.max_rx_queues);
2293
2294         if (nb_tx_queue > dev_info.max_tx_queues)
2295                 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
2296                                 "(max tx queue is %u)\n",
2297                                 nb_tx_queue, dev_info.max_tx_queues);
2298
2299         printf("Creating queues: nb_rx_queue=%d nb_tx_queue=%u...\n",
2300                         nb_rx_queue, nb_tx_queue);
2301
2302         local_port_conf.rxmode.mtu = mtu_size;
2303
2304         if (multi_seg_required()) {
2305                 local_port_conf.rxmode.offloads |= RTE_ETH_RX_OFFLOAD_SCATTER;
2306                 local_port_conf.txmode.offloads |= RTE_ETH_TX_OFFLOAD_MULTI_SEGS;
2307         }
2308
2309         local_port_conf.rxmode.offloads |= req_rx_offloads;
2310         local_port_conf.txmode.offloads |= req_tx_offloads;
2311
2312         /* Check that all required capabilities are supported */
2313         if ((local_port_conf.rxmode.offloads & dev_info.rx_offload_capa) !=
2314                         local_port_conf.rxmode.offloads)
2315                 rte_exit(EXIT_FAILURE,
2316                         "Error: port %u required RX offloads: 0x%" PRIx64
2317                         ", available RX offloads: 0x%" PRIx64 "\n",
2318                         portid, local_port_conf.rxmode.offloads,
2319                         dev_info.rx_offload_capa);
2320
2321         if ((local_port_conf.txmode.offloads & dev_info.tx_offload_capa) !=
2322                         local_port_conf.txmode.offloads)
2323                 rte_exit(EXIT_FAILURE,
2324                         "Error: port %u required TX offloads: 0x%" PRIx64
2325                         ", available TX offloads: 0x%" PRIx64 "\n",
2326                         portid, local_port_conf.txmode.offloads,
2327                         dev_info.tx_offload_capa);
2328
2329         if (dev_info.tx_offload_capa & RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE)
2330                 local_port_conf.txmode.offloads |=
2331                         RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE;
2332
2333         if (dev_info.tx_offload_capa & RTE_ETH_TX_OFFLOAD_IPV4_CKSUM)
2334                 local_port_conf.txmode.offloads |= RTE_ETH_TX_OFFLOAD_IPV4_CKSUM;
2335
2336         printf("port %u configuring rx_offloads=0x%" PRIx64
2337                 ", tx_offloads=0x%" PRIx64 "\n",
2338                 portid, local_port_conf.rxmode.offloads,
2339                 local_port_conf.txmode.offloads);
2340
2341         local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
2342                 dev_info.flow_type_rss_offloads;
2343         if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
2344                         port_conf.rx_adv_conf.rss_conf.rss_hf) {
2345                 printf("Port %u modified RSS hash function based on hardware support,"
2346                         "requested:%#"PRIx64" configured:%#"PRIx64"\n",
2347                         portid,
2348                         port_conf.rx_adv_conf.rss_conf.rss_hf,
2349                         local_port_conf.rx_adv_conf.rss_conf.rss_hf);
2350         }
2351
2352         ret = rte_eth_dev_configure(portid, nb_rx_queue, nb_tx_queue,
2353                         &local_port_conf);
2354         if (ret < 0)
2355                 rte_exit(EXIT_FAILURE, "Cannot configure device: "
2356                                 "err=%d, port=%d\n", ret, portid);
2357
2358         ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd, &nb_txd);
2359         if (ret < 0)
2360                 rte_exit(EXIT_FAILURE, "Cannot adjust number of descriptors: "
2361                                 "err=%d, port=%d\n", ret, portid);
2362
2363         /* init one TX queue per lcore */
2364         tx_queueid = 0;
2365         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2366                 if (rte_lcore_is_enabled(lcore_id) == 0)
2367                         continue;
2368
2369                 if (numa_on)
2370                         socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2371                 else
2372                         socket_id = 0;
2373
2374                 /* init TX queue */
2375                 printf("Setup txq=%u,%d,%d\n", lcore_id, tx_queueid, socket_id);
2376
2377                 txconf = &dev_info.default_txconf;
2378                 txconf->offloads = local_port_conf.txmode.offloads;
2379
2380                 ret = rte_eth_tx_queue_setup(portid, tx_queueid, nb_txd,
2381                                 socket_id, txconf);
2382                 if (ret < 0)
2383                         rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
2384                                         "err=%d, port=%d\n", ret, portid);
2385
2386                 qconf = &lcore_conf[lcore_id];
2387                 qconf->tx_queue_id[portid] = tx_queueid;
2388
2389                 /* Pre-populate pkt offloads based on capabilities */
2390                 qconf->outbound.ipv4_offloads = RTE_MBUF_F_TX_IPV4;
2391                 qconf->outbound.ipv6_offloads = RTE_MBUF_F_TX_IPV6;
2392                 if (local_port_conf.txmode.offloads & RTE_ETH_TX_OFFLOAD_IPV4_CKSUM)
2393                         qconf->outbound.ipv4_offloads |= RTE_MBUF_F_TX_IP_CKSUM;
2394
2395                 tx_queueid++;
2396
2397                 /* init RX queues */
2398                 for (queue = 0; queue < qconf->nb_rx_queue; ++queue) {
2399                         struct rte_eth_rxconf rxq_conf;
2400                         struct rte_mempool *pool;
2401
2402                         if (portid != qconf->rx_queue_list[queue].port_id)
2403                                 continue;
2404
2405                         rx_queueid = qconf->rx_queue_list[queue].queue_id;
2406
2407                         printf("Setup rxq=%d,%d,%d\n", portid, rx_queueid,
2408                                         socket_id);
2409
2410                         rxq_conf = dev_info.default_rxconf;
2411                         rxq_conf.offloads = local_port_conf.rxmode.offloads;
2412
2413                         if (per_port_pool)
2414                                 pool = socket_ctx[socket_id].mbuf_pool[portid];
2415                         else
2416                                 pool = socket_ctx[socket_id].mbuf_pool[0];
2417
2418                         ret = rte_eth_rx_queue_setup(portid, rx_queueid,
2419                                         nb_rxd, socket_id, &rxq_conf, pool);
2420                         if (ret < 0)
2421                                 rte_exit(EXIT_FAILURE,
2422                                         "rte_eth_rx_queue_setup: err=%d, "
2423                                         "port=%d\n", ret, portid);
2424                 }
2425         }
2426         printf("\n");
2427 }
2428
2429 static size_t
2430 max_session_size(void)
2431 {
2432         size_t max_sz, sz;
2433         void *sec_ctx;
2434         int16_t cdev_id, port_id, n;
2435
2436         max_sz = 0;
2437         n =  rte_cryptodev_count();
2438         for (cdev_id = 0; cdev_id != n; cdev_id++) {
2439                 sz = rte_cryptodev_sym_get_private_session_size(cdev_id);
2440                 if (sz > max_sz)
2441                         max_sz = sz;
2442                 /*
2443                  * If crypto device is security capable, need to check the
2444                  * size of security session as well.
2445                  */
2446
2447                 /* Get security context of the crypto device */
2448                 sec_ctx = rte_cryptodev_get_sec_ctx(cdev_id);
2449                 if (sec_ctx == NULL)
2450                         continue;
2451
2452                 /* Get size of security session */
2453                 sz = rte_security_session_get_size(sec_ctx);
2454                 if (sz > max_sz)
2455                         max_sz = sz;
2456         }
2457
2458         RTE_ETH_FOREACH_DEV(port_id) {
2459                 if ((enabled_port_mask & (1 << port_id)) == 0)
2460                         continue;
2461
2462                 sec_ctx = rte_eth_dev_get_sec_ctx(port_id);
2463                 if (sec_ctx == NULL)
2464                         continue;
2465
2466                 sz = rte_security_session_get_size(sec_ctx);
2467                 if (sz > max_sz)
2468                         max_sz = sz;
2469         }
2470
2471         return max_sz;
2472 }
2473
2474 static void
2475 session_pool_init(struct socket_ctx *ctx, int32_t socket_id, size_t sess_sz)
2476 {
2477         char mp_name[RTE_MEMPOOL_NAMESIZE];
2478         struct rte_mempool *sess_mp;
2479         uint32_t nb_sess;
2480
2481         snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
2482                         "sess_mp_%u", socket_id);
2483         nb_sess = (get_nb_crypto_sessions() + CDEV_MP_CACHE_SZ *
2484                 rte_lcore_count());
2485         nb_sess = RTE_MAX(nb_sess, CDEV_MP_CACHE_SZ *
2486                         CDEV_MP_CACHE_MULTIPLIER);
2487         sess_mp = rte_cryptodev_sym_session_pool_create(
2488                         mp_name, nb_sess, sess_sz, CDEV_MP_CACHE_SZ, 0,
2489                         socket_id);
2490         ctx->session_pool = sess_mp;
2491
2492         if (ctx->session_pool == NULL)
2493                 rte_exit(EXIT_FAILURE,
2494                         "Cannot init session pool on socket %d\n", socket_id);
2495         else
2496                 printf("Allocated session pool on socket %d\n", socket_id);
2497 }
2498
2499 static void
2500 session_priv_pool_init(struct socket_ctx *ctx, int32_t socket_id,
2501         size_t sess_sz)
2502 {
2503         char mp_name[RTE_MEMPOOL_NAMESIZE];
2504         struct rte_mempool *sess_mp;
2505         uint32_t nb_sess;
2506
2507         snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
2508                         "sess_mp_priv_%u", socket_id);
2509         nb_sess = (get_nb_crypto_sessions() + CDEV_MP_CACHE_SZ *
2510                 rte_lcore_count());
2511         nb_sess = RTE_MAX(nb_sess, CDEV_MP_CACHE_SZ *
2512                         CDEV_MP_CACHE_MULTIPLIER);
2513         sess_mp = rte_mempool_create(mp_name,
2514                         nb_sess,
2515                         sess_sz,
2516                         CDEV_MP_CACHE_SZ,
2517                         0, NULL, NULL, NULL,
2518                         NULL, socket_id,
2519                         0);
2520         ctx->session_priv_pool = sess_mp;
2521
2522         if (ctx->session_priv_pool == NULL)
2523                 rte_exit(EXIT_FAILURE,
2524                         "Cannot init session priv pool on socket %d\n",
2525                         socket_id);
2526         else
2527                 printf("Allocated session priv pool on socket %d\n",
2528                         socket_id);
2529 }
2530
2531 static void
2532 pool_init(struct socket_ctx *ctx, int32_t socket_id, int portid,
2533           uint32_t nb_mbuf)
2534 {
2535         char s[64];
2536         int32_t ms;
2537
2538
2539         /* mbuf_pool is initialised by the pool_init() function*/
2540         if (socket_ctx[socket_id].mbuf_pool[portid])
2541                 return;
2542
2543         snprintf(s, sizeof(s), "mbuf_pool_%d_%d", socket_id, portid);
2544         ctx->mbuf_pool[portid] = rte_pktmbuf_pool_create(s, nb_mbuf,
2545                                                          MEMPOOL_CACHE_SIZE,
2546                                                          ipsec_metadata_size(),
2547                                                          frame_buf_size,
2548                                                          socket_id);
2549
2550         /*
2551          * if multi-segment support is enabled, then create a pool
2552          * for indirect mbufs. This is not per-port but global.
2553          */
2554         ms = multi_seg_required();
2555         if (ms != 0 && !ctx->mbuf_pool_indir) {
2556                 snprintf(s, sizeof(s), "mbuf_pool_indir_%d", socket_id);
2557                 ctx->mbuf_pool_indir = rte_pktmbuf_pool_create(s, nb_mbuf,
2558                         MEMPOOL_CACHE_SIZE, 0, 0, socket_id);
2559         }
2560
2561         if (ctx->mbuf_pool[portid] == NULL ||
2562             (ms != 0 && ctx->mbuf_pool_indir == NULL))
2563                 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool on socket %d\n",
2564                                 socket_id);
2565         else
2566                 printf("Allocated mbuf pool on socket %d\n", socket_id);
2567 }
2568
2569 static inline int
2570 inline_ipsec_event_esn_overflow(struct rte_security_ctx *ctx, uint64_t md)
2571 {
2572         struct ipsec_sa *sa;
2573
2574         /* For inline protocol processing, the metadata in the event will
2575          * uniquely identify the security session which raised the event.
2576          * Application would then need the userdata it had registered with the
2577          * security session to process the event.
2578          */
2579
2580         sa = (struct ipsec_sa *)rte_security_get_userdata(ctx, md);
2581
2582         if (sa == NULL) {
2583                 /* userdata could not be retrieved */
2584                 return -1;
2585         }
2586
2587         /* Sequence number over flow. SA need to be re-established */
2588         RTE_SET_USED(sa);
2589         return 0;
2590 }
2591
2592 static int
2593 inline_ipsec_event_callback(uint16_t port_id, enum rte_eth_event_type type,
2594                  void *param, void *ret_param)
2595 {
2596         uint64_t md;
2597         struct rte_eth_event_ipsec_desc *event_desc = NULL;
2598         struct rte_security_ctx *ctx = (struct rte_security_ctx *)
2599                                         rte_eth_dev_get_sec_ctx(port_id);
2600
2601         RTE_SET_USED(param);
2602
2603         if (type != RTE_ETH_EVENT_IPSEC)
2604                 return -1;
2605
2606         event_desc = ret_param;
2607         if (event_desc == NULL) {
2608                 printf("Event descriptor not set\n");
2609                 return -1;
2610         }
2611
2612         md = event_desc->metadata;
2613
2614         if (event_desc->subtype == RTE_ETH_EVENT_IPSEC_ESN_OVERFLOW)
2615                 return inline_ipsec_event_esn_overflow(ctx, md);
2616         else if (event_desc->subtype >= RTE_ETH_EVENT_IPSEC_MAX) {
2617                 printf("Invalid IPsec event reported\n");
2618                 return -1;
2619         }
2620
2621         return -1;
2622 }
2623
2624 static int
2625 ethdev_reset_event_callback(uint16_t port_id,
2626                 enum rte_eth_event_type type,
2627                  void *param __rte_unused, void *ret_param __rte_unused)
2628 {
2629         printf("Reset Event on port id %d type %d\n", port_id, type);
2630         printf("Force quit application");
2631         force_quit = true;
2632         return 0;
2633 }
2634
2635 static uint16_t
2636 rx_callback(__rte_unused uint16_t port, __rte_unused uint16_t queue,
2637         struct rte_mbuf *pkt[], uint16_t nb_pkts,
2638         __rte_unused uint16_t max_pkts, void *user_param)
2639 {
2640         uint64_t tm;
2641         uint32_t i, k;
2642         struct lcore_conf *lc;
2643         struct rte_mbuf *mb;
2644         struct rte_ether_hdr *eth;
2645
2646         lc = user_param;
2647         k = 0;
2648         tm = 0;
2649
2650         for (i = 0; i != nb_pkts; i++) {
2651
2652                 mb = pkt[i];
2653                 eth = rte_pktmbuf_mtod(mb, struct rte_ether_hdr *);
2654                 if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
2655
2656                         struct rte_ipv4_hdr *iph;
2657
2658                         iph = (struct rte_ipv4_hdr *)(eth + 1);
2659                         if (rte_ipv4_frag_pkt_is_fragmented(iph)) {
2660
2661                                 mb->l2_len = sizeof(*eth);
2662                                 mb->l3_len = sizeof(*iph);
2663                                 tm = (tm != 0) ? tm : rte_rdtsc();
2664                                 mb = rte_ipv4_frag_reassemble_packet(
2665                                         lc->frag.tbl, &lc->frag.dr,
2666                                         mb, tm, iph);
2667
2668                                 if (mb != NULL) {
2669                                         /* fix ip cksum after reassemble. */
2670                                         iph = rte_pktmbuf_mtod_offset(mb,
2671                                                 struct rte_ipv4_hdr *,
2672                                                 mb->l2_len);
2673                                         iph->hdr_checksum = 0;
2674                                         iph->hdr_checksum = rte_ipv4_cksum(iph);
2675                                 }
2676                         }
2677                 } else if (eth->ether_type ==
2678                                 rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6)) {
2679
2680                         struct rte_ipv6_hdr *iph;
2681                         struct rte_ipv6_fragment_ext *fh;
2682
2683                         iph = (struct rte_ipv6_hdr *)(eth + 1);
2684                         fh = rte_ipv6_frag_get_ipv6_fragment_header(iph);
2685                         if (fh != NULL) {
2686                                 mb->l2_len = sizeof(*eth);
2687                                 mb->l3_len = (uintptr_t)fh - (uintptr_t)iph +
2688                                         sizeof(*fh);
2689                                 tm = (tm != 0) ? tm : rte_rdtsc();
2690                                 mb = rte_ipv6_frag_reassemble_packet(
2691                                         lc->frag.tbl, &lc->frag.dr,
2692                                         mb, tm, iph, fh);
2693                                 if (mb != NULL)
2694                                         /* fix l3_len after reassemble. */
2695                                         mb->l3_len = mb->l3_len - sizeof(*fh);
2696                         }
2697                 }
2698
2699                 pkt[k] = mb;
2700                 k += (mb != NULL);
2701         }
2702
2703         /* some fragments were encountered, drain death row */
2704         if (tm != 0)
2705                 rte_ip_frag_free_death_row(&lc->frag.dr, 0);
2706
2707         return k;
2708 }
2709
2710
2711 static int
2712 reassemble_lcore_init(struct lcore_conf *lc, uint32_t cid)
2713 {
2714         int32_t sid;
2715         uint32_t i;
2716         uint64_t frag_cycles;
2717         const struct lcore_rx_queue *rxq;
2718         const struct rte_eth_rxtx_callback *cb;
2719
2720         /* create fragment table */
2721         sid = rte_lcore_to_socket_id(cid);
2722         frag_cycles = (rte_get_tsc_hz() + NS_PER_S - 1) /
2723                 NS_PER_S * frag_ttl_ns;
2724
2725         lc->frag.tbl = rte_ip_frag_table_create(frag_tbl_sz,
2726                 FRAG_TBL_BUCKET_ENTRIES, frag_tbl_sz, frag_cycles, sid);
2727         if (lc->frag.tbl == NULL) {
2728                 printf("%s(%u): failed to create fragment table of size: %u, "
2729                         "error code: %d\n",
2730                         __func__, cid, frag_tbl_sz, rte_errno);
2731                 return -ENOMEM;
2732         }
2733
2734         /* setup reassemble RX callbacks for all queues */
2735         for (i = 0; i != lc->nb_rx_queue; i++) {
2736
2737                 rxq = lc->rx_queue_list + i;
2738                 cb = rte_eth_add_rx_callback(rxq->port_id, rxq->queue_id,
2739                         rx_callback, lc);
2740                 if (cb == NULL) {
2741                         printf("%s(%u): failed to install RX callback for "
2742                                 "portid=%u, queueid=%u, error code: %d\n",
2743                                 __func__, cid,
2744                                 rxq->port_id, rxq->queue_id, rte_errno);
2745                         return -ENOMEM;
2746                 }
2747         }
2748
2749         return 0;
2750 }
2751
2752 static int
2753 reassemble_init(void)
2754 {
2755         int32_t rc;
2756         uint32_t i, lc;
2757
2758         rc = 0;
2759         for (i = 0; i != nb_lcore_params; i++) {
2760                 lc = lcore_params[i].lcore_id;
2761                 rc = reassemble_lcore_init(lcore_conf + lc, lc);
2762                 if (rc != 0)
2763                         break;
2764         }
2765
2766         return rc;
2767 }
2768
2769 static void
2770 create_default_ipsec_flow(uint16_t port_id, uint64_t rx_offloads)
2771 {
2772         struct rte_flow_action action[2];
2773         struct rte_flow_item pattern[2];
2774         struct rte_flow_attr attr = {0};
2775         struct rte_flow_error err;
2776         struct rte_flow *flow;
2777         int ret;
2778
2779         if (!(rx_offloads & RTE_ETH_RX_OFFLOAD_SECURITY))
2780                 return;
2781
2782         /* Add the default rte_flow to enable SECURITY for all ESP packets */
2783
2784         pattern[0].type = RTE_FLOW_ITEM_TYPE_ESP;
2785         pattern[0].spec = NULL;
2786         pattern[0].mask = NULL;
2787         pattern[0].last = NULL;
2788         pattern[1].type = RTE_FLOW_ITEM_TYPE_END;
2789
2790         action[0].type = RTE_FLOW_ACTION_TYPE_SECURITY;
2791         action[0].conf = NULL;
2792         action[1].type = RTE_FLOW_ACTION_TYPE_END;
2793         action[1].conf = NULL;
2794
2795         attr.ingress = 1;
2796
2797         ret = rte_flow_validate(port_id, &attr, pattern, action, &err);
2798         if (ret)
2799                 return;
2800
2801         flow = rte_flow_create(port_id, &attr, pattern, action, &err);
2802         if (flow == NULL)
2803                 return;
2804
2805         flow_info_tbl[port_id].rx_def_flow = flow;
2806         RTE_LOG(INFO, IPSEC,
2807                 "Created default flow enabling SECURITY for all ESP traffic on port %d\n",
2808                 port_id);
2809 }
2810
2811 static void
2812 signal_handler(int signum)
2813 {
2814         if (signum == SIGINT || signum == SIGTERM) {
2815                 printf("\n\nSignal %d received, preparing to exit...\n",
2816                                 signum);
2817                 force_quit = true;
2818         }
2819 }
2820
2821 static void
2822 ev_mode_sess_verify(struct ipsec_sa *sa, int nb_sa)
2823 {
2824         struct rte_ipsec_session *ips;
2825         int32_t i;
2826
2827         if (!sa || !nb_sa)
2828                 return;
2829
2830         for (i = 0; i < nb_sa; i++) {
2831                 ips = ipsec_get_primary_session(&sa[i]);
2832                 if (ips->type != RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL)
2833                         rte_exit(EXIT_FAILURE, "Event mode supports only "
2834                                  "inline protocol sessions\n");
2835         }
2836
2837 }
2838
2839 static int32_t
2840 check_event_mode_params(struct eh_conf *eh_conf)
2841 {
2842         struct eventmode_conf *em_conf = NULL;
2843         struct lcore_params *params;
2844         uint16_t portid;
2845
2846         if (!eh_conf || !eh_conf->mode_params)
2847                 return -EINVAL;
2848
2849         /* Get eventmode conf */
2850         em_conf = eh_conf->mode_params;
2851
2852         if (eh_conf->mode == EH_PKT_TRANSFER_MODE_POLL &&
2853             em_conf->ext_params.sched_type != SCHED_TYPE_NOT_SET) {
2854                 printf("error: option --event-schedule-type applies only to "
2855                        "event mode\n");
2856                 return -EINVAL;
2857         }
2858
2859         if (eh_conf->mode != EH_PKT_TRANSFER_MODE_EVENT)
2860                 return 0;
2861
2862         /* Set schedule type to ORDERED if it wasn't explicitly set by user */
2863         if (em_conf->ext_params.sched_type == SCHED_TYPE_NOT_SET)
2864                 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_ORDERED;
2865
2866         /*
2867          * Event mode currently supports only inline protocol sessions.
2868          * If there are other types of sessions configured then exit with
2869          * error.
2870          */
2871         ev_mode_sess_verify(sa_in, nb_sa_in);
2872         ev_mode_sess_verify(sa_out, nb_sa_out);
2873
2874
2875         /* Option --config does not apply to event mode */
2876         if (nb_lcore_params > 0) {
2877                 printf("error: option --config applies only to poll mode\n");
2878                 return -EINVAL;
2879         }
2880
2881         /*
2882          * In order to use the same port_init routine for both poll and event
2883          * modes initialize lcore_params with one queue for each eth port
2884          */
2885         lcore_params = lcore_params_array;
2886         RTE_ETH_FOREACH_DEV(portid) {
2887                 if ((enabled_port_mask & (1 << portid)) == 0)
2888                         continue;
2889
2890                 params = &lcore_params[nb_lcore_params++];
2891                 params->port_id = portid;
2892                 params->queue_id = 0;
2893                 params->lcore_id = rte_get_next_lcore(0, 0, 1);
2894         }
2895
2896         return 0;
2897 }
2898
2899 static void
2900 inline_sessions_free(struct sa_ctx *sa_ctx)
2901 {
2902         struct rte_ipsec_session *ips;
2903         struct ipsec_sa *sa;
2904         int32_t ret;
2905         uint32_t i;
2906
2907         if (!sa_ctx)
2908                 return;
2909
2910         for (i = 0; i < sa_ctx->nb_sa; i++) {
2911
2912                 sa = &sa_ctx->sa[i];
2913                 if (!sa->spi)
2914                         continue;
2915
2916                 ips = ipsec_get_primary_session(sa);
2917                 if (ips->type != RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL &&
2918                     ips->type != RTE_SECURITY_ACTION_TYPE_INLINE_CRYPTO)
2919                         continue;
2920
2921                 if (!rte_eth_dev_is_valid_port(sa->portid))
2922                         continue;
2923
2924                 ret = rte_security_session_destroy(
2925                                 rte_eth_dev_get_sec_ctx(sa->portid),
2926                                 ips->security.ses);
2927                 if (ret)
2928                         RTE_LOG(ERR, IPSEC, "Failed to destroy security "
2929                                             "session type %d, spi %d\n",
2930                                             ips->type, sa->spi);
2931         }
2932 }
2933
2934 static uint32_t
2935 calculate_nb_mbufs(uint16_t nb_ports, uint16_t nb_crypto_qp, uint32_t nb_rxq,
2936                 uint32_t nb_txq)
2937 {
2938         return RTE_MAX((nb_rxq * nb_rxd +
2939                         nb_ports * nb_lcores * MAX_PKT_BURST +
2940                         nb_ports * nb_txq * nb_txd +
2941                         nb_lcores * MEMPOOL_CACHE_SIZE +
2942                         nb_crypto_qp * CDEV_QUEUE_DESC +
2943                         nb_lcores * frag_tbl_sz *
2944                         FRAG_TBL_BUCKET_ENTRIES),
2945                        8192U);
2946 }
2947
2948
2949 static int
2950 handle_telemetry_cmd_ipsec_secgw_stats(const char *cmd __rte_unused,
2951                 const char *params, struct rte_tel_data *data)
2952 {
2953         uint64_t total_pkts_dropped = 0, total_pkts_tx = 0, total_pkts_rx = 0;
2954         unsigned int coreid;
2955
2956         rte_tel_data_start_dict(data);
2957
2958         if (params) {
2959                 coreid = (uint32_t)atoi(params);
2960                 if (rte_lcore_is_enabled(coreid) == 0)
2961                         return -EINVAL;
2962
2963                 total_pkts_dropped = core_statistics[coreid].dropped;
2964                 total_pkts_tx = core_statistics[coreid].tx;
2965                 total_pkts_rx = core_statistics[coreid].rx;
2966
2967         } else {
2968                 for (coreid = 0; coreid < RTE_MAX_LCORE; coreid++) {
2969
2970                         /* skip disabled cores */
2971                         if (rte_lcore_is_enabled(coreid) == 0)
2972                                 continue;
2973
2974                         total_pkts_dropped += core_statistics[coreid].dropped;
2975                         total_pkts_tx += core_statistics[coreid].tx;
2976                         total_pkts_rx += core_statistics[coreid].rx;
2977                 }
2978         }
2979
2980         /* add telemetry key/values pairs */
2981         rte_tel_data_add_dict_u64(data, "packets received",
2982                                 total_pkts_rx);
2983
2984         rte_tel_data_add_dict_u64(data, "packets transmitted",
2985                                 total_pkts_tx);
2986
2987         rte_tel_data_add_dict_u64(data, "packets dropped",
2988                                 total_pkts_dropped);
2989
2990
2991         return 0;
2992 }
2993
2994 static void
2995 update_lcore_statistics(struct ipsec_core_statistics *total, uint32_t coreid)
2996 {
2997         struct ipsec_core_statistics *lcore_stats;
2998
2999         /* skip disabled cores */
3000         if (rte_lcore_is_enabled(coreid) == 0)
3001                 return;
3002
3003         lcore_stats = &core_statistics[coreid];
3004
3005         total->rx = lcore_stats->rx;
3006         total->dropped = lcore_stats->dropped;
3007         total->tx = lcore_stats->tx;
3008
3009         /* outbound stats */
3010         total->outbound.spd6.protect += lcore_stats->outbound.spd6.protect;
3011         total->outbound.spd6.bypass += lcore_stats->outbound.spd6.bypass;
3012         total->outbound.spd6.discard += lcore_stats->outbound.spd6.discard;
3013
3014         total->outbound.spd4.protect += lcore_stats->outbound.spd4.protect;
3015         total->outbound.spd4.bypass += lcore_stats->outbound.spd4.bypass;
3016         total->outbound.spd4.discard += lcore_stats->outbound.spd4.discard;
3017
3018         total->outbound.sad.miss += lcore_stats->outbound.sad.miss;
3019
3020         /* inbound stats */
3021         total->inbound.spd6.protect += lcore_stats->inbound.spd6.protect;
3022         total->inbound.spd6.bypass += lcore_stats->inbound.spd6.bypass;
3023         total->inbound.spd6.discard += lcore_stats->inbound.spd6.discard;
3024
3025         total->inbound.spd4.protect += lcore_stats->inbound.spd4.protect;
3026         total->inbound.spd4.bypass += lcore_stats->inbound.spd4.bypass;
3027         total->inbound.spd4.discard += lcore_stats->inbound.spd4.discard;
3028
3029         total->inbound.sad.miss += lcore_stats->inbound.sad.miss;
3030
3031
3032         /* routing stats */
3033         total->lpm4.miss += lcore_stats->lpm4.miss;
3034         total->lpm6.miss += lcore_stats->lpm6.miss;
3035 }
3036
3037 static void
3038 update_statistics(struct ipsec_core_statistics *total, uint32_t coreid)
3039 {
3040         memset(total, 0, sizeof(*total));
3041
3042         if (coreid != UINT32_MAX) {
3043                 update_lcore_statistics(total, coreid);
3044         } else {
3045                 for (coreid = 0; coreid < RTE_MAX_LCORE; coreid++)
3046                         update_lcore_statistics(total, coreid);
3047         }
3048 }
3049
3050 static int
3051 handle_telemetry_cmd_ipsec_secgw_stats_outbound(const char *cmd __rte_unused,
3052                 const char *params, struct rte_tel_data *data)
3053 {
3054         struct ipsec_core_statistics total_stats;
3055
3056         struct rte_tel_data *spd4_data = rte_tel_data_alloc();
3057         struct rte_tel_data *spd6_data = rte_tel_data_alloc();
3058         struct rte_tel_data *sad_data = rte_tel_data_alloc();
3059         unsigned int coreid = UINT32_MAX;
3060         int rc = 0;
3061
3062         /* verify allocated telemetry data structures */
3063         if (!spd4_data || !spd6_data || !sad_data) {
3064                 rc = -ENOMEM;
3065                 goto exit;
3066         }
3067
3068         /* initialize telemetry data structs as dicts */
3069         rte_tel_data_start_dict(data);
3070
3071         rte_tel_data_start_dict(spd4_data);
3072         rte_tel_data_start_dict(spd6_data);
3073         rte_tel_data_start_dict(sad_data);
3074
3075         if (params) {
3076                 coreid = (uint32_t)atoi(params);
3077                 if (rte_lcore_is_enabled(coreid) == 0) {
3078                         rc = -EINVAL;
3079                         goto exit;
3080                 }
3081         }
3082
3083         update_statistics(&total_stats, coreid);
3084
3085         /* add spd 4 telemetry key/values pairs */
3086
3087         rte_tel_data_add_dict_u64(spd4_data, "protect",
3088                 total_stats.outbound.spd4.protect);
3089         rte_tel_data_add_dict_u64(spd4_data, "bypass",
3090                 total_stats.outbound.spd4.bypass);
3091         rte_tel_data_add_dict_u64(spd4_data, "discard",
3092                 total_stats.outbound.spd4.discard);
3093
3094         rte_tel_data_add_dict_container(data, "spd4", spd4_data, 0);
3095
3096         /* add spd 6 telemetry key/values pairs */
3097
3098         rte_tel_data_add_dict_u64(spd6_data, "protect",
3099                 total_stats.outbound.spd6.protect);
3100         rte_tel_data_add_dict_u64(spd6_data, "bypass",
3101                 total_stats.outbound.spd6.bypass);
3102         rte_tel_data_add_dict_u64(spd6_data, "discard",
3103                 total_stats.outbound.spd6.discard);
3104
3105         rte_tel_data_add_dict_container(data, "spd6", spd6_data, 0);
3106
3107         /* add sad telemetry key/values pairs */
3108
3109         rte_tel_data_add_dict_u64(sad_data, "miss",
3110                 total_stats.outbound.sad.miss);
3111
3112         rte_tel_data_add_dict_container(data, "sad", sad_data, 0);
3113
3114 exit:
3115         if (rc) {
3116                 rte_tel_data_free(spd4_data);
3117                 rte_tel_data_free(spd6_data);
3118                 rte_tel_data_free(sad_data);
3119         }
3120         return rc;
3121 }
3122
3123 static int
3124 handle_telemetry_cmd_ipsec_secgw_stats_inbound(const char *cmd __rte_unused,
3125                 const char *params, struct rte_tel_data *data)
3126 {
3127         struct ipsec_core_statistics total_stats;
3128
3129         struct rte_tel_data *spd4_data = rte_tel_data_alloc();
3130         struct rte_tel_data *spd6_data = rte_tel_data_alloc();
3131         struct rte_tel_data *sad_data = rte_tel_data_alloc();
3132         unsigned int coreid = UINT32_MAX;
3133         int rc = 0;
3134
3135         /* verify allocated telemetry data structures */
3136         if (!spd4_data || !spd6_data || !sad_data) {
3137                 rc = -ENOMEM;
3138                 goto exit;
3139         }
3140
3141         /* initialize telemetry data structs as dicts */
3142         rte_tel_data_start_dict(data);
3143         rte_tel_data_start_dict(spd4_data);
3144         rte_tel_data_start_dict(spd6_data);
3145         rte_tel_data_start_dict(sad_data);
3146
3147         /* add children dicts to parent dict */
3148
3149         if (params) {
3150                 coreid = (uint32_t)atoi(params);
3151                 if (rte_lcore_is_enabled(coreid) == 0) {
3152                         rc = -EINVAL;
3153                         goto exit;
3154                 }
3155         }
3156
3157         update_statistics(&total_stats, coreid);
3158
3159         /* add sad telemetry key/values pairs */
3160
3161         rte_tel_data_add_dict_u64(sad_data, "miss",
3162                 total_stats.inbound.sad.miss);
3163
3164         rte_tel_data_add_dict_container(data, "sad", sad_data, 0);
3165
3166         /* add spd 4 telemetry key/values pairs */
3167
3168         rte_tel_data_add_dict_u64(spd4_data, "protect",
3169                 total_stats.inbound.spd4.protect);
3170         rte_tel_data_add_dict_u64(spd4_data, "bypass",
3171                 total_stats.inbound.spd4.bypass);
3172         rte_tel_data_add_dict_u64(spd4_data, "discard",
3173                 total_stats.inbound.spd4.discard);
3174
3175         rte_tel_data_add_dict_container(data, "spd4", spd4_data, 0);
3176
3177         /* add spd 6 telemetry key/values pairs */
3178
3179         rte_tel_data_add_dict_u64(spd6_data, "protect",
3180                 total_stats.inbound.spd6.protect);
3181         rte_tel_data_add_dict_u64(spd6_data, "bypass",
3182                 total_stats.inbound.spd6.bypass);
3183         rte_tel_data_add_dict_u64(spd6_data, "discard",
3184                 total_stats.inbound.spd6.discard);
3185
3186         rte_tel_data_add_dict_container(data, "spd6", spd6_data, 0);
3187
3188 exit:
3189         if (rc) {
3190                 rte_tel_data_free(spd4_data);
3191                 rte_tel_data_free(spd6_data);
3192                 rte_tel_data_free(sad_data);
3193         }
3194         return rc;
3195 }
3196
3197 static int
3198 handle_telemetry_cmd_ipsec_secgw_stats_routing(const char *cmd __rte_unused,
3199                 const char *params, struct rte_tel_data *data)
3200 {
3201         struct ipsec_core_statistics total_stats;
3202
3203         struct rte_tel_data *lpm4_data = rte_tel_data_alloc();
3204         struct rte_tel_data *lpm6_data = rte_tel_data_alloc();
3205         unsigned int coreid = UINT32_MAX;
3206         int rc = 0;
3207
3208         /* verify allocated telemetry data structures */
3209         if (!lpm4_data || !lpm6_data) {
3210                 rc = -ENOMEM;
3211                 goto exit;
3212         }
3213
3214         /* initialize telemetry data structs as dicts */
3215         rte_tel_data_start_dict(data);
3216         rte_tel_data_start_dict(lpm4_data);
3217         rte_tel_data_start_dict(lpm6_data);
3218
3219
3220         if (params) {
3221                 coreid = (uint32_t)atoi(params);
3222                 if (rte_lcore_is_enabled(coreid) == 0) {
3223                         rc = -EINVAL;
3224                         goto exit;
3225                 }
3226         }
3227
3228         update_statistics(&total_stats, coreid);
3229
3230         /* add lpm 4 telemetry key/values pairs */
3231         rte_tel_data_add_dict_u64(lpm4_data, "miss",
3232                 total_stats.lpm4.miss);
3233
3234         rte_tel_data_add_dict_container(data, "IPv4 LPM", lpm4_data, 0);
3235
3236         /* add lpm 6 telemetry key/values pairs */
3237         rte_tel_data_add_dict_u64(lpm6_data, "miss",
3238                 total_stats.lpm6.miss);
3239
3240         rte_tel_data_add_dict_container(data, "IPv6 LPM", lpm6_data, 0);
3241
3242 exit:
3243         if (rc) {
3244                 rte_tel_data_free(lpm4_data);
3245                 rte_tel_data_free(lpm6_data);
3246         }
3247         return rc;
3248 }
3249
3250 static void
3251 ipsec_secgw_telemetry_init(void)
3252 {
3253         rte_telemetry_register_cmd("/examples/ipsec-secgw/stats",
3254                 handle_telemetry_cmd_ipsec_secgw_stats,
3255                 "Returns global stats. "
3256                 "Optional Parameters: int <logical core id>");
3257
3258         rte_telemetry_register_cmd("/examples/ipsec-secgw/stats/outbound",
3259                 handle_telemetry_cmd_ipsec_secgw_stats_outbound,
3260                 "Returns outbound global stats. "
3261                 "Optional Parameters: int <logical core id>");
3262
3263         rte_telemetry_register_cmd("/examples/ipsec-secgw/stats/inbound",
3264                 handle_telemetry_cmd_ipsec_secgw_stats_inbound,
3265                 "Returns inbound global stats. "
3266                 "Optional Parameters: int <logical core id>");
3267
3268         rte_telemetry_register_cmd("/examples/ipsec-secgw/stats/routing",
3269                 handle_telemetry_cmd_ipsec_secgw_stats_routing,
3270                 "Returns routing stats. "
3271                 "Optional Parameters: int <logical core id>");
3272 }
3273
3274
3275 int32_t
3276 main(int32_t argc, char **argv)
3277 {
3278         int32_t ret;
3279         uint32_t lcore_id, nb_txq, nb_rxq = 0;
3280         uint32_t cdev_id;
3281         uint32_t i;
3282         uint8_t socket_id;
3283         uint16_t portid, nb_crypto_qp, nb_ports = 0;
3284         uint64_t req_rx_offloads[RTE_MAX_ETHPORTS];
3285         uint64_t req_tx_offloads[RTE_MAX_ETHPORTS];
3286         struct eh_conf *eh_conf = NULL;
3287         size_t sess_sz;
3288
3289         nb_bufs_in_pool = 0;
3290
3291         /* init EAL */
3292         ret = rte_eal_init(argc, argv);
3293         if (ret < 0)
3294                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
3295         argc -= ret;
3296         argv += ret;
3297
3298         force_quit = false;
3299         signal(SIGINT, signal_handler);
3300         signal(SIGTERM, signal_handler);
3301
3302         /* initialize event helper configuration */
3303         eh_conf = eh_conf_init();
3304         if (eh_conf == NULL)
3305                 rte_exit(EXIT_FAILURE, "Failed to init event helper config");
3306
3307         /* parse application arguments (after the EAL ones) */
3308         ret = parse_args(argc, argv, eh_conf);
3309         if (ret < 0)
3310                 rte_exit(EXIT_FAILURE, "Invalid parameters\n");
3311
3312         ipsec_secgw_telemetry_init();
3313
3314         /* parse configuration file */
3315         if (parse_cfg_file(cfgfile) < 0) {
3316                 printf("parsing file \"%s\" failed\n",
3317                         optarg);
3318                 print_usage(argv[0]);
3319                 return -1;
3320         }
3321
3322         if ((unprotected_port_mask & enabled_port_mask) !=
3323                         unprotected_port_mask)
3324                 rte_exit(EXIT_FAILURE, "Invalid unprotected portmask 0x%x\n",
3325                                 unprotected_port_mask);
3326
3327         if (unprotected_port_mask && !nb_sa_in)
3328                 rte_exit(EXIT_FAILURE, "Cannot use unprotected portmask without configured SA inbound\n");
3329
3330         if (check_poll_mode_params(eh_conf) < 0)
3331                 rte_exit(EXIT_FAILURE, "check_poll_mode_params failed\n");
3332
3333         if (check_event_mode_params(eh_conf) < 0)
3334                 rte_exit(EXIT_FAILURE, "check_event_mode_params failed\n");
3335
3336         ret = init_lcore_rx_queues();
3337         if (ret < 0)
3338                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
3339
3340         nb_lcores = rte_lcore_count();
3341
3342         sess_sz = max_session_size();
3343
3344         /*
3345          * In event mode request minimum number of crypto queues
3346          * to be reserved equal to number of ports.
3347          */
3348         if (eh_conf->mode == EH_PKT_TRANSFER_MODE_EVENT)
3349                 nb_crypto_qp = rte_eth_dev_count_avail();
3350         else
3351                 nb_crypto_qp = 0;
3352
3353         nb_crypto_qp = cryptodevs_init(nb_crypto_qp);
3354
3355         if (nb_bufs_in_pool == 0) {
3356                 RTE_ETH_FOREACH_DEV(portid) {
3357                         if ((enabled_port_mask & (1 << portid)) == 0)
3358                                 continue;
3359                         nb_ports++;
3360                         nb_rxq += get_port_nb_rx_queues(portid);
3361                 }
3362
3363                 nb_txq = nb_lcores;
3364
3365                 nb_bufs_in_pool = calculate_nb_mbufs(nb_ports, nb_crypto_qp,
3366                                                 nb_rxq, nb_txq);
3367         }
3368
3369         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
3370                 if (rte_lcore_is_enabled(lcore_id) == 0)
3371                         continue;
3372
3373                 if (numa_on)
3374                         socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
3375                 else
3376                         socket_id = 0;
3377
3378                 if (per_port_pool) {
3379                         RTE_ETH_FOREACH_DEV(portid) {
3380                                 if ((enabled_port_mask & (1 << portid)) == 0)
3381                                         continue;
3382
3383                                 pool_init(&socket_ctx[socket_id], socket_id,
3384                                           portid, nb_bufs_in_pool);
3385                         }
3386                 } else {
3387                         pool_init(&socket_ctx[socket_id], socket_id, 0,
3388                                   nb_bufs_in_pool);
3389                 }
3390
3391                 if (socket_ctx[socket_id].session_pool)
3392                         continue;
3393
3394                 session_pool_init(&socket_ctx[socket_id], socket_id, sess_sz);
3395                 session_priv_pool_init(&socket_ctx[socket_id], socket_id,
3396                         sess_sz);
3397         }
3398         printf("Number of mbufs in packet pool %d\n", nb_bufs_in_pool);
3399
3400         RTE_ETH_FOREACH_DEV(portid) {
3401                 if ((enabled_port_mask & (1 << portid)) == 0)
3402                         continue;
3403
3404                 sa_check_offloads(portid, &req_rx_offloads[portid],
3405                                 &req_tx_offloads[portid]);
3406                 port_init(portid, req_rx_offloads[portid],
3407                                 req_tx_offloads[portid]);
3408         }
3409
3410         /*
3411          * Set the enabled port mask in helper config for use by helper
3412          * sub-system. This will be used while initializing devices using
3413          * helper sub-system.
3414          */
3415         eh_conf->eth_portmask = enabled_port_mask;
3416
3417         /* Initialize eventmode components */
3418         ret = eh_devs_init(eh_conf);
3419         if (ret < 0)
3420                 rte_exit(EXIT_FAILURE, "eh_devs_init failed, err=%d\n", ret);
3421
3422         /* start ports */
3423         RTE_ETH_FOREACH_DEV(portid) {
3424                 if ((enabled_port_mask & (1 << portid)) == 0)
3425                         continue;
3426
3427                 ret = rte_eth_dev_start(portid);
3428                 if (ret < 0)
3429                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: "
3430                                         "err=%d, port=%d\n", ret, portid);
3431
3432                 /* Create flow after starting the device */
3433                 create_default_ipsec_flow(portid, req_rx_offloads[portid]);
3434
3435                 /*
3436                  * If enabled, put device in promiscuous mode.
3437                  * This allows IO forwarding mode to forward packets
3438                  * to itself through 2 cross-connected  ports of the
3439                  * target machine.
3440                  */
3441                 if (promiscuous_on) {
3442                         ret = rte_eth_promiscuous_enable(portid);
3443                         if (ret != 0)
3444                                 rte_exit(EXIT_FAILURE,
3445                                         "rte_eth_promiscuous_enable: err=%s, port=%d\n",
3446                                         rte_strerror(-ret), portid);
3447                 }
3448
3449                 rte_eth_dev_callback_register(portid, RTE_ETH_EVENT_INTR_RESET,
3450                         ethdev_reset_event_callback, NULL);
3451
3452                 rte_eth_dev_callback_register(portid,
3453                         RTE_ETH_EVENT_IPSEC, inline_ipsec_event_callback, NULL);
3454         }
3455
3456         /* fragment reassemble is enabled */
3457         if (frag_tbl_sz != 0) {
3458                 ret = reassemble_init();
3459                 if (ret != 0)
3460                         rte_exit(EXIT_FAILURE, "failed at reassemble init");
3461         }
3462
3463         /* Replicate each context per socket */
3464         for (i = 0; i < NB_SOCKETS && i < rte_socket_count(); i++) {
3465                 socket_id = rte_socket_id_by_idx(i);
3466                 if ((socket_ctx[socket_id].session_pool != NULL) &&
3467                         (socket_ctx[socket_id].sa_in == NULL) &&
3468                         (socket_ctx[socket_id].sa_out == NULL)) {
3469                         sa_init(&socket_ctx[socket_id], socket_id);
3470                         sp4_init(&socket_ctx[socket_id], socket_id);
3471                         sp6_init(&socket_ctx[socket_id], socket_id);
3472                         rt_init(&socket_ctx[socket_id], socket_id);
3473                 }
3474         }
3475
3476         flow_init();
3477
3478         check_all_ports_link_status(enabled_port_mask);
3479
3480         if (stats_interval > 0)
3481                 rte_eal_alarm_set(stats_interval * US_PER_S,
3482                                 print_stats_cb, NULL);
3483         else
3484                 RTE_LOG(INFO, IPSEC, "Stats display disabled\n");
3485
3486         /* launch per-lcore init on every lcore */
3487         rte_eal_mp_remote_launch(ipsec_launch_one_lcore, eh_conf, CALL_MAIN);
3488         RTE_LCORE_FOREACH_WORKER(lcore_id) {
3489                 if (rte_eal_wait_lcore(lcore_id) < 0)
3490                         return -1;
3491         }
3492
3493         /* Uninitialize eventmode components */
3494         ret = eh_devs_uninit(eh_conf);
3495         if (ret < 0)
3496                 rte_exit(EXIT_FAILURE, "eh_devs_uninit failed, err=%d\n", ret);
3497
3498         /* Free eventmode configuration memory */
3499         eh_conf_uninit(eh_conf);
3500
3501         /* Destroy inline inbound and outbound sessions */
3502         for (i = 0; i < NB_SOCKETS && i < rte_socket_count(); i++) {
3503                 socket_id = rte_socket_id_by_idx(i);
3504                 inline_sessions_free(socket_ctx[socket_id].sa_in);
3505                 inline_sessions_free(socket_ctx[socket_id].sa_out);
3506         }
3507
3508         for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
3509                 printf("Closing cryptodev %d...", cdev_id);
3510                 rte_cryptodev_stop(cdev_id);
3511                 rte_cryptodev_close(cdev_id);
3512                 printf(" Done\n");
3513         }
3514
3515         RTE_ETH_FOREACH_DEV(portid) {
3516                 if ((enabled_port_mask & (1 << portid)) == 0)
3517                         continue;
3518
3519                 printf("Closing port %d...", portid);
3520                 if (flow_info_tbl[portid].rx_def_flow) {
3521                         struct rte_flow_error err;
3522
3523                         ret = rte_flow_destroy(portid,
3524                                 flow_info_tbl[portid].rx_def_flow, &err);
3525                         if (ret)
3526                                 RTE_LOG(ERR, IPSEC, "Failed to destroy flow "
3527                                         " for port %u, err msg: %s\n", portid,
3528                                         err.message);
3529                 }
3530                 ret = rte_eth_dev_stop(portid);
3531                 if (ret != 0)
3532                         RTE_LOG(ERR, IPSEC,
3533                                 "rte_eth_dev_stop: err=%s, port=%u\n",
3534                                 rte_strerror(-ret), portid);
3535
3536                 rte_eth_dev_close(portid);
3537                 printf(" Done\n");
3538         }
3539
3540         /* clean up the EAL */
3541         rte_eal_cleanup();
3542         printf("Bye...\n");
3543
3544         return 0;
3545 }