ff576eb94dbf1e0c0b19f3ca7b46203f7d09c708
[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 <stdio.h>
6 #include <stdlib.h>
7 #include <stdint.h>
8 #include <inttypes.h>
9 #include <sys/types.h>
10 #include <netinet/in.h>
11 #include <netinet/ip.h>
12 #include <netinet/ip6.h>
13 #include <string.h>
14 #include <sys/queue.h>
15 #include <stdarg.h>
16 #include <errno.h>
17 #include <getopt.h>
18
19 #include <rte_common.h>
20 #include <rte_byteorder.h>
21 #include <rte_log.h>
22 #include <rte_eal.h>
23 #include <rte_launch.h>
24 #include <rte_atomic.h>
25 #include <rte_cycles.h>
26 #include <rte_prefetch.h>
27 #include <rte_lcore.h>
28 #include <rte_per_lcore.h>
29 #include <rte_branch_prediction.h>
30 #include <rte_interrupts.h>
31 #include <rte_random.h>
32 #include <rte_debug.h>
33 #include <rte_ether.h>
34 #include <rte_ethdev.h>
35 #include <rte_mempool.h>
36 #include <rte_mbuf.h>
37 #include <rte_acl.h>
38 #include <rte_lpm.h>
39 #include <rte_lpm6.h>
40 #include <rte_hash.h>
41 #include <rte_jhash.h>
42 #include <rte_cryptodev.h>
43 #include <rte_security.h>
44
45 #include "ipsec.h"
46 #include "parser.h"
47
48 #define RTE_LOGTYPE_IPSEC RTE_LOGTYPE_USER1
49
50 #define MAX_JUMBO_PKT_LEN  9600
51
52 #define MEMPOOL_CACHE_SIZE 256
53
54 #define NB_MBUF (32000)
55
56 #define CDEV_QUEUE_DESC 2048
57 #define CDEV_MAP_ENTRIES 1024
58 #define CDEV_MP_NB_OBJS 2048
59 #define CDEV_MP_CACHE_SZ 64
60 #define MAX_QUEUE_PAIRS 1
61
62 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
63
64 #define NB_SOCKETS 4
65
66 /* Configure how many packets ahead to prefetch, when reading packets */
67 #define PREFETCH_OFFSET 3
68
69 #define MAX_RX_QUEUE_PER_LCORE 16
70
71 #define MAX_LCORE_PARAMS 1024
72
73 #define UNPROTECTED_PORT(port) (unprotected_port_mask & (1 << portid))
74
75 /*
76  * Configurable number of RX/TX ring descriptors
77  */
78 #define IPSEC_SECGW_RX_DESC_DEFAULT 1024
79 #define IPSEC_SECGW_TX_DESC_DEFAULT 1024
80 static uint16_t nb_rxd = IPSEC_SECGW_RX_DESC_DEFAULT;
81 static uint16_t nb_txd = IPSEC_SECGW_TX_DESC_DEFAULT;
82
83 #if RTE_BYTE_ORDER != RTE_LITTLE_ENDIAN
84 #define __BYTES_TO_UINT64(a, b, c, d, e, f, g, h) \
85         (((uint64_t)((a) & 0xff) << 56) | \
86         ((uint64_t)((b) & 0xff) << 48) | \
87         ((uint64_t)((c) & 0xff) << 40) | \
88         ((uint64_t)((d) & 0xff) << 32) | \
89         ((uint64_t)((e) & 0xff) << 24) | \
90         ((uint64_t)((f) & 0xff) << 16) | \
91         ((uint64_t)((g) & 0xff) << 8)  | \
92         ((uint64_t)(h) & 0xff))
93 #else
94 #define __BYTES_TO_UINT64(a, b, c, d, e, f, g, h) \
95         (((uint64_t)((h) & 0xff) << 56) | \
96         ((uint64_t)((g) & 0xff) << 48) | \
97         ((uint64_t)((f) & 0xff) << 40) | \
98         ((uint64_t)((e) & 0xff) << 32) | \
99         ((uint64_t)((d) & 0xff) << 24) | \
100         ((uint64_t)((c) & 0xff) << 16) | \
101         ((uint64_t)((b) & 0xff) << 8) | \
102         ((uint64_t)(a) & 0xff))
103 #endif
104 #define ETHADDR(a, b, c, d, e, f) (__BYTES_TO_UINT64(a, b, c, d, e, f, 0, 0))
105
106 #define ETHADDR_TO_UINT64(addr) __BYTES_TO_UINT64( \
107                 addr.addr_bytes[0], addr.addr_bytes[1], \
108                 addr.addr_bytes[2], addr.addr_bytes[3], \
109                 addr.addr_bytes[4], addr.addr_bytes[5], \
110                 0, 0)
111
112 /* port/source ethernet addr and destination ethernet addr */
113 struct ethaddr_info {
114         uint64_t src, dst;
115 };
116
117 struct ethaddr_info ethaddr_tbl[RTE_MAX_ETHPORTS] = {
118         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x7e, 0x94, 0x9a) },
119         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x22, 0xa1, 0xd9) },
120         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x08, 0x69, 0x26) },
121         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x49, 0x9e, 0xdd) }
122 };
123
124 #define CMD_LINE_OPT_CONFIG             "config"
125 #define CMD_LINE_OPT_SINGLE_SA          "single-sa"
126 #define CMD_LINE_OPT_CRYPTODEV_MASK     "cryptodev_mask"
127
128 enum {
129         /* long options mapped to a short option */
130
131         /* first long only option value must be >= 256, so that we won't
132          * conflict with short options
133          */
134         CMD_LINE_OPT_MIN_NUM = 256,
135         CMD_LINE_OPT_CONFIG_NUM,
136         CMD_LINE_OPT_SINGLE_SA_NUM,
137         CMD_LINE_OPT_CRYPTODEV_MASK_NUM,
138 };
139
140 static const struct option lgopts[] = {
141         {CMD_LINE_OPT_CONFIG, 1, 0, CMD_LINE_OPT_CONFIG_NUM},
142         {CMD_LINE_OPT_SINGLE_SA, 1, 0, CMD_LINE_OPT_SINGLE_SA_NUM},
143         {CMD_LINE_OPT_CRYPTODEV_MASK, 1, 0, CMD_LINE_OPT_CRYPTODEV_MASK_NUM},
144         {NULL, 0, 0, 0}
145 };
146
147 /* mask of enabled ports */
148 static uint32_t enabled_port_mask;
149 static uint64_t enabled_cryptodev_mask = UINT64_MAX;
150 static uint32_t unprotected_port_mask;
151 static int32_t promiscuous_on = 1;
152 static int32_t numa_on = 1; /**< NUMA is enabled by default. */
153 static uint32_t nb_lcores;
154 static uint32_t single_sa;
155 static uint32_t single_sa_idx;
156 static uint32_t frame_size;
157
158 struct lcore_rx_queue {
159         uint16_t port_id;
160         uint8_t queue_id;
161 } __rte_cache_aligned;
162
163 struct lcore_params {
164         uint16_t port_id;
165         uint8_t queue_id;
166         uint8_t lcore_id;
167 } __rte_cache_aligned;
168
169 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
170
171 static struct lcore_params *lcore_params;
172 static uint16_t nb_lcore_params;
173
174 static struct rte_hash *cdev_map_in;
175 static struct rte_hash *cdev_map_out;
176
177 struct buffer {
178         uint16_t len;
179         struct rte_mbuf *m_table[MAX_PKT_BURST] __rte_aligned(sizeof(void *));
180 };
181
182 struct lcore_conf {
183         uint16_t nb_rx_queue;
184         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
185         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
186         struct buffer tx_mbufs[RTE_MAX_ETHPORTS];
187         struct ipsec_ctx inbound;
188         struct ipsec_ctx outbound;
189         struct rt_ctx *rt4_ctx;
190         struct rt_ctx *rt6_ctx;
191 } __rte_cache_aligned;
192
193 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
194
195 static struct rte_eth_conf port_conf = {
196         .rxmode = {
197                 .mq_mode        = ETH_MQ_RX_RSS,
198                 .max_rx_pkt_len = ETHER_MAX_LEN,
199                 .split_hdr_size = 0,
200                 .offloads = DEV_RX_OFFLOAD_CHECKSUM |
201                             DEV_RX_OFFLOAD_CRC_STRIP,
202         },
203         .rx_adv_conf = {
204                 .rss_conf = {
205                         .rss_key = NULL,
206                         .rss_hf = ETH_RSS_IP | ETH_RSS_UDP |
207                                 ETH_RSS_TCP | ETH_RSS_SCTP,
208                 },
209         },
210         .txmode = {
211                 .mq_mode = ETH_MQ_TX_NONE,
212                 .offloads = (DEV_TX_OFFLOAD_IPV4_CKSUM |
213                              DEV_TX_OFFLOAD_MULTI_SEGS),
214         },
215 };
216
217 static struct socket_ctx socket_ctx[NB_SOCKETS];
218
219 struct traffic_type {
220         const uint8_t *data[MAX_PKT_BURST * 2];
221         struct rte_mbuf *pkts[MAX_PKT_BURST * 2];
222         uint32_t res[MAX_PKT_BURST * 2];
223         uint32_t num;
224 };
225
226 struct ipsec_traffic {
227         struct traffic_type ipsec;
228         struct traffic_type ip4;
229         struct traffic_type ip6;
230 };
231
232 static inline void
233 prepare_one_packet(struct rte_mbuf *pkt, struct ipsec_traffic *t)
234 {
235         uint8_t *nlp;
236         struct ether_hdr *eth;
237
238         eth = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
239         if (eth->ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4)) {
240                 nlp = (uint8_t *)rte_pktmbuf_adj(pkt, ETHER_HDR_LEN);
241                 nlp = RTE_PTR_ADD(nlp, offsetof(struct ip, ip_p));
242                 if (*nlp == IPPROTO_ESP)
243                         t->ipsec.pkts[(t->ipsec.num)++] = pkt;
244                 else {
245                         t->ip4.data[t->ip4.num] = nlp;
246                         t->ip4.pkts[(t->ip4.num)++] = pkt;
247                 }
248         } else if (eth->ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6)) {
249                 nlp = (uint8_t *)rte_pktmbuf_adj(pkt, ETHER_HDR_LEN);
250                 nlp = RTE_PTR_ADD(nlp, offsetof(struct ip6_hdr, ip6_nxt));
251                 if (*nlp == IPPROTO_ESP)
252                         t->ipsec.pkts[(t->ipsec.num)++] = pkt;
253                 else {
254                         t->ip6.data[t->ip6.num] = nlp;
255                         t->ip6.pkts[(t->ip6.num)++] = pkt;
256                 }
257         } else {
258                 /* Unknown/Unsupported type, drop the packet */
259                 RTE_LOG(ERR, IPSEC, "Unsupported packet type\n");
260                 rte_pktmbuf_free(pkt);
261         }
262
263         /* Check if the packet has been processed inline. For inline protocol
264          * processed packets, the metadata in the mbuf can be used to identify
265          * the security processing done on the packet. The metadata will be
266          * used to retrieve the application registered userdata associated
267          * with the security session.
268          */
269
270         if (pkt->ol_flags & PKT_RX_SEC_OFFLOAD) {
271                 struct ipsec_sa *sa;
272                 struct ipsec_mbuf_metadata *priv;
273                 struct rte_security_ctx *ctx = (struct rte_security_ctx *)
274                                                 rte_eth_dev_get_sec_ctx(
275                                                 pkt->port);
276
277                 /* Retrieve the userdata registered. Here, the userdata
278                  * registered is the SA pointer.
279                  */
280
281                 sa = (struct ipsec_sa *)
282                                 rte_security_get_userdata(ctx, pkt->udata64);
283
284                 if (sa == NULL) {
285                         /* userdata could not be retrieved */
286                         return;
287                 }
288
289                 /* Save SA as priv member in mbuf. This will be used in the
290                  * IPsec selector(SP-SA) check.
291                  */
292
293                 priv = get_priv(pkt);
294                 priv->sa = sa;
295         }
296 }
297
298 static inline void
299 prepare_traffic(struct rte_mbuf **pkts, struct ipsec_traffic *t,
300                 uint16_t nb_pkts)
301 {
302         int32_t i;
303
304         t->ipsec.num = 0;
305         t->ip4.num = 0;
306         t->ip6.num = 0;
307
308         for (i = 0; i < (nb_pkts - PREFETCH_OFFSET); i++) {
309                 rte_prefetch0(rte_pktmbuf_mtod(pkts[i + PREFETCH_OFFSET],
310                                         void *));
311                 prepare_one_packet(pkts[i], t);
312         }
313         /* Process left packets */
314         for (; i < nb_pkts; i++)
315                 prepare_one_packet(pkts[i], t);
316 }
317
318 static inline void
319 prepare_tx_pkt(struct rte_mbuf *pkt, uint16_t port)
320 {
321         struct ip *ip;
322         struct ether_hdr *ethhdr;
323
324         ip = rte_pktmbuf_mtod(pkt, struct ip *);
325
326         ethhdr = (struct ether_hdr *)rte_pktmbuf_prepend(pkt, ETHER_HDR_LEN);
327
328         if (ip->ip_v == IPVERSION) {
329                 pkt->ol_flags |= PKT_TX_IP_CKSUM | PKT_TX_IPV4;
330                 pkt->l3_len = sizeof(struct ip);
331                 pkt->l2_len = ETHER_HDR_LEN;
332
333                 ethhdr->ether_type = rte_cpu_to_be_16(ETHER_TYPE_IPv4);
334         } else {
335                 pkt->ol_flags |= PKT_TX_IPV6;
336                 pkt->l3_len = sizeof(struct ip6_hdr);
337                 pkt->l2_len = ETHER_HDR_LEN;
338
339                 ethhdr->ether_type = rte_cpu_to_be_16(ETHER_TYPE_IPv6);
340         }
341
342         memcpy(&ethhdr->s_addr, &ethaddr_tbl[port].src,
343                         sizeof(struct ether_addr));
344         memcpy(&ethhdr->d_addr, &ethaddr_tbl[port].dst,
345                         sizeof(struct ether_addr));
346 }
347
348 static inline void
349 prepare_tx_burst(struct rte_mbuf *pkts[], uint16_t nb_pkts, uint16_t port)
350 {
351         int32_t i;
352         const int32_t prefetch_offset = 2;
353
354         for (i = 0; i < (nb_pkts - prefetch_offset); i++) {
355                 rte_mbuf_prefetch_part2(pkts[i + prefetch_offset]);
356                 prepare_tx_pkt(pkts[i], port);
357         }
358         /* Process left packets */
359         for (; i < nb_pkts; i++)
360                 prepare_tx_pkt(pkts[i], port);
361 }
362
363 /* Send burst of packets on an output interface */
364 static inline int32_t
365 send_burst(struct lcore_conf *qconf, uint16_t n, uint16_t port)
366 {
367         struct rte_mbuf **m_table;
368         int32_t ret;
369         uint16_t queueid;
370
371         queueid = qconf->tx_queue_id[port];
372         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
373
374         prepare_tx_burst(m_table, n, port);
375
376         ret = rte_eth_tx_burst(port, queueid, m_table, n);
377         if (unlikely(ret < n)) {
378                 do {
379                         rte_pktmbuf_free(m_table[ret]);
380                 } while (++ret < n);
381         }
382
383         return 0;
384 }
385
386 /* Enqueue a single packet, and send burst if queue is filled */
387 static inline int32_t
388 send_single_packet(struct rte_mbuf *m, uint16_t port)
389 {
390         uint32_t lcore_id;
391         uint16_t len;
392         struct lcore_conf *qconf;
393
394         lcore_id = rte_lcore_id();
395
396         qconf = &lcore_conf[lcore_id];
397         len = qconf->tx_mbufs[port].len;
398         qconf->tx_mbufs[port].m_table[len] = m;
399         len++;
400
401         /* enough pkts to be sent */
402         if (unlikely(len == MAX_PKT_BURST)) {
403                 send_burst(qconf, MAX_PKT_BURST, port);
404                 len = 0;
405         }
406
407         qconf->tx_mbufs[port].len = len;
408         return 0;
409 }
410
411 static inline void
412 inbound_sp_sa(struct sp_ctx *sp, struct sa_ctx *sa, struct traffic_type *ip,
413                 uint16_t lim)
414 {
415         struct rte_mbuf *m;
416         uint32_t i, j, res, sa_idx;
417
418         if (ip->num == 0 || sp == NULL)
419                 return;
420
421         rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
422                         ip->num, DEFAULT_MAX_CATEGORIES);
423
424         j = 0;
425         for (i = 0; i < ip->num; i++) {
426                 m = ip->pkts[i];
427                 res = ip->res[i];
428                 if (res & BYPASS) {
429                         ip->pkts[j++] = m;
430                         continue;
431                 }
432                 if (res & DISCARD) {
433                         rte_pktmbuf_free(m);
434                         continue;
435                 }
436
437                 /* Only check SPI match for processed IPSec packets */
438                 if (i < lim && ((m->ol_flags & PKT_RX_SEC_OFFLOAD) == 0)) {
439                         rte_pktmbuf_free(m);
440                         continue;
441                 }
442
443                 sa_idx = ip->res[i] & PROTECT_MASK;
444                 if (sa_idx >= IPSEC_SA_MAX_ENTRIES ||
445                                 !inbound_sa_check(sa, m, sa_idx)) {
446                         rte_pktmbuf_free(m);
447                         continue;
448                 }
449                 ip->pkts[j++] = m;
450         }
451         ip->num = j;
452 }
453
454 static inline void
455 process_pkts_inbound(struct ipsec_ctx *ipsec_ctx,
456                 struct ipsec_traffic *traffic)
457 {
458         struct rte_mbuf *m;
459         uint16_t idx, nb_pkts_in, i, n_ip4, n_ip6;
460
461         nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
462                         traffic->ipsec.num, MAX_PKT_BURST);
463
464         n_ip4 = traffic->ip4.num;
465         n_ip6 = traffic->ip6.num;
466
467         /* SP/ACL Inbound check ipsec and ip4 */
468         for (i = 0; i < nb_pkts_in; i++) {
469                 m = traffic->ipsec.pkts[i];
470                 struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
471                 if (ip->ip_v == IPVERSION) {
472                         idx = traffic->ip4.num++;
473                         traffic->ip4.pkts[idx] = m;
474                         traffic->ip4.data[idx] = rte_pktmbuf_mtod_offset(m,
475                                         uint8_t *, offsetof(struct ip, ip_p));
476                 } else if (ip->ip_v == IP6_VERSION) {
477                         idx = traffic->ip6.num++;
478                         traffic->ip6.pkts[idx] = m;
479                         traffic->ip6.data[idx] = rte_pktmbuf_mtod_offset(m,
480                                         uint8_t *,
481                                         offsetof(struct ip6_hdr, ip6_nxt));
482                 } else
483                         rte_pktmbuf_free(m);
484         }
485
486         inbound_sp_sa(ipsec_ctx->sp4_ctx, ipsec_ctx->sa_ctx, &traffic->ip4,
487                         n_ip4);
488
489         inbound_sp_sa(ipsec_ctx->sp6_ctx, ipsec_ctx->sa_ctx, &traffic->ip6,
490                         n_ip6);
491 }
492
493 static inline void
494 outbound_sp(struct sp_ctx *sp, struct traffic_type *ip,
495                 struct traffic_type *ipsec)
496 {
497         struct rte_mbuf *m;
498         uint32_t i, j, sa_idx;
499
500         if (ip->num == 0 || sp == NULL)
501                 return;
502
503         rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
504                         ip->num, DEFAULT_MAX_CATEGORIES);
505
506         j = 0;
507         for (i = 0; i < ip->num; i++) {
508                 m = ip->pkts[i];
509                 sa_idx = ip->res[i] & PROTECT_MASK;
510                 if (ip->res[i] & DISCARD)
511                         rte_pktmbuf_free(m);
512                 else if (sa_idx < IPSEC_SA_MAX_ENTRIES) {
513                         ipsec->res[ipsec->num] = sa_idx;
514                         ipsec->pkts[ipsec->num++] = m;
515                 } else /* BYPASS */
516                         ip->pkts[j++] = m;
517         }
518         ip->num = j;
519 }
520
521 static inline void
522 process_pkts_outbound(struct ipsec_ctx *ipsec_ctx,
523                 struct ipsec_traffic *traffic)
524 {
525         struct rte_mbuf *m;
526         uint16_t idx, nb_pkts_out, i;
527
528         /* Drop any IPsec traffic from protected ports */
529         for (i = 0; i < traffic->ipsec.num; i++)
530                 rte_pktmbuf_free(traffic->ipsec.pkts[i]);
531
532         traffic->ipsec.num = 0;
533
534         outbound_sp(ipsec_ctx->sp4_ctx, &traffic->ip4, &traffic->ipsec);
535
536         outbound_sp(ipsec_ctx->sp6_ctx, &traffic->ip6, &traffic->ipsec);
537
538         nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
539                         traffic->ipsec.res, traffic->ipsec.num,
540                         MAX_PKT_BURST);
541
542         for (i = 0; i < nb_pkts_out; i++) {
543                 m = traffic->ipsec.pkts[i];
544                 struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
545                 if (ip->ip_v == IPVERSION) {
546                         idx = traffic->ip4.num++;
547                         traffic->ip4.pkts[idx] = m;
548                 } else {
549                         idx = traffic->ip6.num++;
550                         traffic->ip6.pkts[idx] = m;
551                 }
552         }
553 }
554
555 static inline void
556 process_pkts_inbound_nosp(struct ipsec_ctx *ipsec_ctx,
557                 struct ipsec_traffic *traffic)
558 {
559         struct rte_mbuf *m;
560         uint32_t nb_pkts_in, i, idx;
561
562         /* Drop any IPv4 traffic from unprotected ports */
563         for (i = 0; i < traffic->ip4.num; i++)
564                 rte_pktmbuf_free(traffic->ip4.pkts[i]);
565
566         traffic->ip4.num = 0;
567
568         /* Drop any IPv6 traffic from unprotected ports */
569         for (i = 0; i < traffic->ip6.num; i++)
570                 rte_pktmbuf_free(traffic->ip6.pkts[i]);
571
572         traffic->ip6.num = 0;
573
574         nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
575                         traffic->ipsec.num, MAX_PKT_BURST);
576
577         for (i = 0; i < nb_pkts_in; i++) {
578                 m = traffic->ipsec.pkts[i];
579                 struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
580                 if (ip->ip_v == IPVERSION) {
581                         idx = traffic->ip4.num++;
582                         traffic->ip4.pkts[idx] = m;
583                 } else {
584                         idx = traffic->ip6.num++;
585                         traffic->ip6.pkts[idx] = m;
586                 }
587         }
588 }
589
590 static inline void
591 process_pkts_outbound_nosp(struct ipsec_ctx *ipsec_ctx,
592                 struct ipsec_traffic *traffic)
593 {
594         struct rte_mbuf *m;
595         uint32_t nb_pkts_out, i;
596         struct ip *ip;
597
598         /* Drop any IPsec traffic from protected ports */
599         for (i = 0; i < traffic->ipsec.num; i++)
600                 rte_pktmbuf_free(traffic->ipsec.pkts[i]);
601
602         traffic->ipsec.num = 0;
603
604         for (i = 0; i < traffic->ip4.num; i++)
605                 traffic->ip4.res[i] = single_sa_idx;
606
607         for (i = 0; i < traffic->ip6.num; i++)
608                 traffic->ip6.res[i] = single_sa_idx;
609
610         nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ip4.pkts,
611                         traffic->ip4.res, traffic->ip4.num,
612                         MAX_PKT_BURST);
613
614         /* They all sue the same SA (ip4 or ip6 tunnel) */
615         m = traffic->ipsec.pkts[i];
616         ip = rte_pktmbuf_mtod(m, struct ip *);
617         if (ip->ip_v == IPVERSION)
618                 traffic->ip4.num = nb_pkts_out;
619         else
620                 traffic->ip6.num = nb_pkts_out;
621 }
622
623 static inline int32_t
624 get_hop_for_offload_pkt(struct rte_mbuf *pkt, int is_ipv6)
625 {
626         struct ipsec_mbuf_metadata *priv;
627         struct ipsec_sa *sa;
628
629         priv = get_priv(pkt);
630
631         sa = priv->sa;
632         if (unlikely(sa == NULL)) {
633                 RTE_LOG(ERR, IPSEC, "SA not saved in private data\n");
634                 goto fail;
635         }
636
637         if (is_ipv6)
638                 return sa->portid;
639
640         /* else */
641         return (sa->portid | RTE_LPM_LOOKUP_SUCCESS);
642
643 fail:
644         if (is_ipv6)
645                 return -1;
646
647         /* else */
648         return 0;
649 }
650
651 static inline void
652 route4_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
653 {
654         uint32_t hop[MAX_PKT_BURST * 2];
655         uint32_t dst_ip[MAX_PKT_BURST * 2];
656         int32_t pkt_hop = 0;
657         uint16_t i, offset;
658         uint16_t lpm_pkts = 0;
659
660         if (nb_pkts == 0)
661                 return;
662
663         /* Need to do an LPM lookup for non-inline packets. Inline packets will
664          * have port ID in the SA
665          */
666
667         for (i = 0; i < nb_pkts; i++) {
668                 if (!(pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD)) {
669                         /* Security offload not enabled. So an LPM lookup is
670                          * required to get the hop
671                          */
672                         offset = offsetof(struct ip, ip_dst);
673                         dst_ip[lpm_pkts] = *rte_pktmbuf_mtod_offset(pkts[i],
674                                         uint32_t *, offset);
675                         dst_ip[lpm_pkts] = rte_be_to_cpu_32(dst_ip[lpm_pkts]);
676                         lpm_pkts++;
677                 }
678         }
679
680         rte_lpm_lookup_bulk((struct rte_lpm *)rt_ctx, dst_ip, hop, lpm_pkts);
681
682         lpm_pkts = 0;
683
684         for (i = 0; i < nb_pkts; i++) {
685                 if (pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD) {
686                         /* Read hop from the SA */
687                         pkt_hop = get_hop_for_offload_pkt(pkts[i], 0);
688                 } else {
689                         /* Need to use hop returned by lookup */
690                         pkt_hop = hop[lpm_pkts++];
691                 }
692
693                 if ((pkt_hop & RTE_LPM_LOOKUP_SUCCESS) == 0) {
694                         rte_pktmbuf_free(pkts[i]);
695                         continue;
696                 }
697                 send_single_packet(pkts[i], pkt_hop & 0xff);
698         }
699 }
700
701 static inline void
702 route6_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
703 {
704         int32_t hop[MAX_PKT_BURST * 2];
705         uint8_t dst_ip[MAX_PKT_BURST * 2][16];
706         uint8_t *ip6_dst;
707         int32_t pkt_hop = 0;
708         uint16_t i, offset;
709         uint16_t lpm_pkts = 0;
710
711         if (nb_pkts == 0)
712                 return;
713
714         /* Need to do an LPM lookup for non-inline packets. Inline packets will
715          * have port ID in the SA
716          */
717
718         for (i = 0; i < nb_pkts; i++) {
719                 if (!(pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD)) {
720                         /* Security offload not enabled. So an LPM lookup is
721                          * required to get the hop
722                          */
723                         offset = offsetof(struct ip6_hdr, ip6_dst);
724                         ip6_dst = rte_pktmbuf_mtod_offset(pkts[i], uint8_t *,
725                                         offset);
726                         memcpy(&dst_ip[lpm_pkts][0], ip6_dst, 16);
727                         lpm_pkts++;
728                 }
729         }
730
731         rte_lpm6_lookup_bulk_func((struct rte_lpm6 *)rt_ctx, dst_ip, hop,
732                         lpm_pkts);
733
734         lpm_pkts = 0;
735
736         for (i = 0; i < nb_pkts; i++) {
737                 if (pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD) {
738                         /* Read hop from the SA */
739                         pkt_hop = get_hop_for_offload_pkt(pkts[i], 1);
740                 } else {
741                         /* Need to use hop returned by lookup */
742                         pkt_hop = hop[lpm_pkts++];
743                 }
744
745                 if (pkt_hop == -1) {
746                         rte_pktmbuf_free(pkts[i]);
747                         continue;
748                 }
749                 send_single_packet(pkts[i], pkt_hop & 0xff);
750         }
751 }
752
753 static inline void
754 process_pkts(struct lcore_conf *qconf, struct rte_mbuf **pkts,
755                 uint8_t nb_pkts, uint16_t portid)
756 {
757         struct ipsec_traffic traffic;
758
759         prepare_traffic(pkts, &traffic, nb_pkts);
760
761         if (unlikely(single_sa)) {
762                 if (UNPROTECTED_PORT(portid))
763                         process_pkts_inbound_nosp(&qconf->inbound, &traffic);
764                 else
765                         process_pkts_outbound_nosp(&qconf->outbound, &traffic);
766         } else {
767                 if (UNPROTECTED_PORT(portid))
768                         process_pkts_inbound(&qconf->inbound, &traffic);
769                 else
770                         process_pkts_outbound(&qconf->outbound, &traffic);
771         }
772
773         route4_pkts(qconf->rt4_ctx, traffic.ip4.pkts, traffic.ip4.num);
774         route6_pkts(qconf->rt6_ctx, traffic.ip6.pkts, traffic.ip6.num);
775 }
776
777 static inline void
778 drain_buffers(struct lcore_conf *qconf)
779 {
780         struct buffer *buf;
781         uint32_t portid;
782
783         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
784                 buf = &qconf->tx_mbufs[portid];
785                 if (buf->len == 0)
786                         continue;
787                 send_burst(qconf, buf->len, portid);
788                 buf->len = 0;
789         }
790 }
791
792 /* main processing loop */
793 static int32_t
794 main_loop(__attribute__((unused)) void *dummy)
795 {
796         struct rte_mbuf *pkts[MAX_PKT_BURST];
797         uint32_t lcore_id;
798         uint64_t prev_tsc, diff_tsc, cur_tsc;
799         int32_t i, nb_rx;
800         uint16_t portid;
801         uint8_t queueid;
802         struct lcore_conf *qconf;
803         int32_t socket_id;
804         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
805                         / US_PER_S * BURST_TX_DRAIN_US;
806         struct lcore_rx_queue *rxql;
807
808         prev_tsc = 0;
809         lcore_id = rte_lcore_id();
810         qconf = &lcore_conf[lcore_id];
811         rxql = qconf->rx_queue_list;
812         socket_id = rte_lcore_to_socket_id(lcore_id);
813
814         qconf->rt4_ctx = socket_ctx[socket_id].rt_ip4;
815         qconf->rt6_ctx = socket_ctx[socket_id].rt_ip6;
816         qconf->inbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_in;
817         qconf->inbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_in;
818         qconf->inbound.sa_ctx = socket_ctx[socket_id].sa_in;
819         qconf->inbound.cdev_map = cdev_map_in;
820         qconf->inbound.session_pool = socket_ctx[socket_id].session_pool;
821         qconf->outbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_out;
822         qconf->outbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_out;
823         qconf->outbound.sa_ctx = socket_ctx[socket_id].sa_out;
824         qconf->outbound.cdev_map = cdev_map_out;
825         qconf->outbound.session_pool = socket_ctx[socket_id].session_pool;
826
827         if (qconf->nb_rx_queue == 0) {
828                 RTE_LOG(INFO, IPSEC, "lcore %u has nothing to do\n", lcore_id);
829                 return 0;
830         }
831
832         RTE_LOG(INFO, IPSEC, "entering main loop on lcore %u\n", lcore_id);
833
834         for (i = 0; i < qconf->nb_rx_queue; i++) {
835                 portid = rxql[i].port_id;
836                 queueid = rxql[i].queue_id;
837                 RTE_LOG(INFO, IPSEC,
838                         " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
839                         lcore_id, portid, queueid);
840         }
841
842         while (1) {
843                 cur_tsc = rte_rdtsc();
844
845                 /* TX queue buffer drain */
846                 diff_tsc = cur_tsc - prev_tsc;
847
848                 if (unlikely(diff_tsc > drain_tsc)) {
849                         drain_buffers(qconf);
850                         prev_tsc = cur_tsc;
851                 }
852
853                 /* Read packet from RX queues */
854                 for (i = 0; i < qconf->nb_rx_queue; ++i) {
855                         portid = rxql[i].port_id;
856                         queueid = rxql[i].queue_id;
857                         nb_rx = rte_eth_rx_burst(portid, queueid,
858                                         pkts, MAX_PKT_BURST);
859
860                         if (nb_rx > 0)
861                                 process_pkts(qconf, pkts, nb_rx, portid);
862                 }
863         }
864 }
865
866 static int32_t
867 check_params(void)
868 {
869         uint8_t lcore;
870         uint16_t portid;
871         uint16_t i;
872         int32_t socket_id;
873
874         if (lcore_params == NULL) {
875                 printf("Error: No port/queue/core mappings\n");
876                 return -1;
877         }
878
879         for (i = 0; i < nb_lcore_params; ++i) {
880                 lcore = lcore_params[i].lcore_id;
881                 if (!rte_lcore_is_enabled(lcore)) {
882                         printf("error: lcore %hhu is not enabled in "
883                                 "lcore mask\n", lcore);
884                         return -1;
885                 }
886                 socket_id = rte_lcore_to_socket_id(lcore);
887                 if (socket_id != 0 && numa_on == 0) {
888                         printf("warning: lcore %hhu is on socket %d "
889                                 "with numa off\n",
890                                 lcore, socket_id);
891                 }
892                 portid = lcore_params[i].port_id;
893                 if ((enabled_port_mask & (1 << portid)) == 0) {
894                         printf("port %u is not enabled in port mask\n", portid);
895                         return -1;
896                 }
897                 if (!rte_eth_dev_is_valid_port(portid)) {
898                         printf("port %u is not present on the board\n", portid);
899                         return -1;
900                 }
901         }
902         return 0;
903 }
904
905 static uint8_t
906 get_port_nb_rx_queues(const uint16_t port)
907 {
908         int32_t queue = -1;
909         uint16_t i;
910
911         for (i = 0; i < nb_lcore_params; ++i) {
912                 if (lcore_params[i].port_id == port &&
913                                 lcore_params[i].queue_id > queue)
914                         queue = lcore_params[i].queue_id;
915         }
916         return (uint8_t)(++queue);
917 }
918
919 static int32_t
920 init_lcore_rx_queues(void)
921 {
922         uint16_t i, nb_rx_queue;
923         uint8_t lcore;
924
925         for (i = 0; i < nb_lcore_params; ++i) {
926                 lcore = lcore_params[i].lcore_id;
927                 nb_rx_queue = lcore_conf[lcore].nb_rx_queue;
928                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
929                         printf("error: too many queues (%u) for lcore: %u\n",
930                                         nb_rx_queue + 1, lcore);
931                         return -1;
932                 }
933                 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
934                         lcore_params[i].port_id;
935                 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
936                         lcore_params[i].queue_id;
937                 lcore_conf[lcore].nb_rx_queue++;
938         }
939         return 0;
940 }
941
942 /* display usage */
943 static void
944 print_usage(const char *prgname)
945 {
946         fprintf(stderr, "%s [EAL options] --"
947                 " -p PORTMASK"
948                 " [-P]"
949                 " [-u PORTMASK]"
950                 " [-j FRAMESIZE]"
951                 " -f CONFIG_FILE"
952                 " --config (port,queue,lcore)[,(port,queue,lcore)]"
953                 " [--single-sa SAIDX]"
954                 " [--cryptodev_mask MASK]"
955                 "\n\n"
956                 "  -p PORTMASK: Hexadecimal bitmask of ports to configure\n"
957                 "  -P : Enable promiscuous mode\n"
958                 "  -u PORTMASK: Hexadecimal bitmask of unprotected ports\n"
959                 "  -j FRAMESIZE: Enable jumbo frame with 'FRAMESIZE' as maximum\n"
960                 "                packet size\n"
961                 "  -f CONFIG_FILE: Configuration file\n"
962                 "  --config (port,queue,lcore): Rx queue configuration\n"
963                 "  --single-sa SAIDX: Use single SA index for outbound traffic,\n"
964                 "                     bypassing the SP\n"
965                 "  --cryptodev_mask MASK: Hexadecimal bitmask of the crypto\n"
966                 "                         devices to configure\n"
967                 "\n",
968                 prgname);
969 }
970
971 static int32_t
972 parse_portmask(const char *portmask)
973 {
974         char *end = NULL;
975         unsigned long pm;
976
977         /* parse hexadecimal string */
978         pm = strtoul(portmask, &end, 16);
979         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
980                 return -1;
981
982         if ((pm == 0) && errno)
983                 return -1;
984
985         return pm;
986 }
987
988 static int32_t
989 parse_decimal(const char *str)
990 {
991         char *end = NULL;
992         unsigned long num;
993
994         num = strtoul(str, &end, 10);
995         if ((str[0] == '\0') || (end == NULL) || (*end != '\0'))
996                 return -1;
997
998         return num;
999 }
1000
1001 static int32_t
1002 parse_config(const char *q_arg)
1003 {
1004         char s[256];
1005         const char *p, *p0 = q_arg;
1006         char *end;
1007         enum fieldnames {
1008                 FLD_PORT = 0,
1009                 FLD_QUEUE,
1010                 FLD_LCORE,
1011                 _NUM_FLD
1012         };
1013         unsigned long int_fld[_NUM_FLD];
1014         char *str_fld[_NUM_FLD];
1015         int32_t i;
1016         uint32_t size;
1017
1018         nb_lcore_params = 0;
1019
1020         while ((p = strchr(p0, '(')) != NULL) {
1021                 ++p;
1022                 p0 = strchr(p, ')');
1023                 if (p0 == NULL)
1024                         return -1;
1025
1026                 size = p0 - p;
1027                 if (size >= sizeof(s))
1028                         return -1;
1029
1030                 snprintf(s, sizeof(s), "%.*s", size, p);
1031                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1032                                 _NUM_FLD)
1033                         return -1;
1034                 for (i = 0; i < _NUM_FLD; i++) {
1035                         errno = 0;
1036                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1037                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
1038                                 return -1;
1039                 }
1040                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1041                         printf("exceeded max number of lcore params: %hu\n",
1042                                 nb_lcore_params);
1043                         return -1;
1044                 }
1045                 lcore_params_array[nb_lcore_params].port_id =
1046                         (uint8_t)int_fld[FLD_PORT];
1047                 lcore_params_array[nb_lcore_params].queue_id =
1048                         (uint8_t)int_fld[FLD_QUEUE];
1049                 lcore_params_array[nb_lcore_params].lcore_id =
1050                         (uint8_t)int_fld[FLD_LCORE];
1051                 ++nb_lcore_params;
1052         }
1053         lcore_params = lcore_params_array;
1054         return 0;
1055 }
1056
1057 static int32_t
1058 parse_args(int32_t argc, char **argv)
1059 {
1060         int32_t opt, ret;
1061         char **argvopt;
1062         int32_t option_index;
1063         char *prgname = argv[0];
1064         int32_t f_present = 0;
1065
1066         argvopt = argv;
1067
1068         while ((opt = getopt_long(argc, argvopt, "p:Pu:f:j:",
1069                                 lgopts, &option_index)) != EOF) {
1070
1071                 switch (opt) {
1072                 case 'p':
1073                         enabled_port_mask = parse_portmask(optarg);
1074                         if (enabled_port_mask == 0) {
1075                                 printf("invalid portmask\n");
1076                                 print_usage(prgname);
1077                                 return -1;
1078                         }
1079                         break;
1080                 case 'P':
1081                         printf("Promiscuous mode selected\n");
1082                         promiscuous_on = 1;
1083                         break;
1084                 case 'u':
1085                         unprotected_port_mask = parse_portmask(optarg);
1086                         if (unprotected_port_mask == 0) {
1087                                 printf("invalid unprotected portmask\n");
1088                                 print_usage(prgname);
1089                                 return -1;
1090                         }
1091                         break;
1092                 case 'f':
1093                         if (f_present == 1) {
1094                                 printf("\"-f\" option present more than "
1095                                         "once!\n");
1096                                 print_usage(prgname);
1097                                 return -1;
1098                         }
1099                         if (parse_cfg_file(optarg) < 0) {
1100                                 printf("parsing file \"%s\" failed\n",
1101                                         optarg);
1102                                 print_usage(prgname);
1103                                 return -1;
1104                         }
1105                         f_present = 1;
1106                         break;
1107                 case 'j':
1108                         {
1109                                 int32_t size = parse_decimal(optarg);
1110                                 if (size <= 1518) {
1111                                         printf("Invalid jumbo frame size\n");
1112                                         if (size < 0) {
1113                                                 print_usage(prgname);
1114                                                 return -1;
1115                                         }
1116                                         printf("Using default value 9000\n");
1117                                         frame_size = 9000;
1118                                 } else {
1119                                         frame_size = size;
1120                                 }
1121                         }
1122                         printf("Enabled jumbo frames size %u\n", frame_size);
1123                         break;
1124                 case CMD_LINE_OPT_CONFIG_NUM:
1125                         ret = parse_config(optarg);
1126                         if (ret) {
1127                                 printf("Invalid config\n");
1128                                 print_usage(prgname);
1129                                 return -1;
1130                         }
1131                         break;
1132                 case CMD_LINE_OPT_SINGLE_SA_NUM:
1133                         ret = parse_decimal(optarg);
1134                         if (ret == -1) {
1135                                 printf("Invalid argument[sa_idx]\n");
1136                                 print_usage(prgname);
1137                                 return -1;
1138                         }
1139
1140                         /* else */
1141                         single_sa = 1;
1142                         single_sa_idx = ret;
1143                         printf("Configured with single SA index %u\n",
1144                                         single_sa_idx);
1145                         break;
1146                 case CMD_LINE_OPT_CRYPTODEV_MASK_NUM:
1147                         ret = parse_portmask(optarg);
1148                         if (ret == -1) {
1149                                 printf("Invalid argument[portmask]\n");
1150                                 print_usage(prgname);
1151                                 return -1;
1152                         }
1153
1154                         /* else */
1155                         enabled_cryptodev_mask = ret;
1156                         break;
1157                 default:
1158                         print_usage(prgname);
1159                         return -1;
1160                 }
1161         }
1162
1163         if (f_present == 0) {
1164                 printf("Mandatory option \"-f\" not present\n");
1165                 return -1;
1166         }
1167
1168         if (optind >= 0)
1169                 argv[optind-1] = prgname;
1170
1171         ret = optind-1;
1172         optind = 1; /* reset getopt lib */
1173         return ret;
1174 }
1175
1176 static void
1177 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1178 {
1179         char buf[ETHER_ADDR_FMT_SIZE];
1180         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
1181         printf("%s%s", name, buf);
1182 }
1183
1184 /* Check the link status of all ports in up to 9s, and print them finally */
1185 static void
1186 check_all_ports_link_status(uint32_t port_mask)
1187 {
1188 #define CHECK_INTERVAL 100 /* 100ms */
1189 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1190         uint16_t portid;
1191         uint8_t count, all_ports_up, print_flag = 0;
1192         struct rte_eth_link link;
1193
1194         printf("\nChecking link status");
1195         fflush(stdout);
1196         for (count = 0; count <= MAX_CHECK_TIME; count++) {
1197                 all_ports_up = 1;
1198                 RTE_ETH_FOREACH_DEV(portid) {
1199                         if ((port_mask & (1 << portid)) == 0)
1200                                 continue;
1201                         memset(&link, 0, sizeof(link));
1202                         rte_eth_link_get_nowait(portid, &link);
1203                         /* print link status if flag set */
1204                         if (print_flag == 1) {
1205                                 if (link.link_status)
1206                                         printf(
1207                                         "Port%d Link Up - speed %u Mbps -%s\n",
1208                                                 portid, link.link_speed,
1209                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1210                                         ("full-duplex") : ("half-duplex\n"));
1211                                 else
1212                                         printf("Port %d Link Down\n", portid);
1213                                 continue;
1214                         }
1215                         /* clear all_ports_up flag if any link down */
1216                         if (link.link_status == ETH_LINK_DOWN) {
1217                                 all_ports_up = 0;
1218                                 break;
1219                         }
1220                 }
1221                 /* after finally printing all link status, get out */
1222                 if (print_flag == 1)
1223                         break;
1224
1225                 if (all_ports_up == 0) {
1226                         printf(".");
1227                         fflush(stdout);
1228                         rte_delay_ms(CHECK_INTERVAL);
1229                 }
1230
1231                 /* set the print_flag if all ports up or timeout */
1232                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1233                         print_flag = 1;
1234                         printf("done\n");
1235                 }
1236         }
1237 }
1238
1239 static int32_t
1240 add_mapping(struct rte_hash *map, const char *str, uint16_t cdev_id,
1241                 uint16_t qp, struct lcore_params *params,
1242                 struct ipsec_ctx *ipsec_ctx,
1243                 const struct rte_cryptodev_capabilities *cipher,
1244                 const struct rte_cryptodev_capabilities *auth,
1245                 const struct rte_cryptodev_capabilities *aead)
1246 {
1247         int32_t ret = 0;
1248         unsigned long i;
1249         struct cdev_key key = { 0 };
1250
1251         key.lcore_id = params->lcore_id;
1252         if (cipher)
1253                 key.cipher_algo = cipher->sym.cipher.algo;
1254         if (auth)
1255                 key.auth_algo = auth->sym.auth.algo;
1256         if (aead)
1257                 key.aead_algo = aead->sym.aead.algo;
1258
1259         ret = rte_hash_lookup(map, &key);
1260         if (ret != -ENOENT)
1261                 return 0;
1262
1263         for (i = 0; i < ipsec_ctx->nb_qps; i++)
1264                 if (ipsec_ctx->tbl[i].id == cdev_id)
1265                         break;
1266
1267         if (i == ipsec_ctx->nb_qps) {
1268                 if (ipsec_ctx->nb_qps == MAX_QP_PER_LCORE) {
1269                         printf("Maximum number of crypto devices assigned to "
1270                                 "a core, increase MAX_QP_PER_LCORE value\n");
1271                         return 0;
1272                 }
1273                 ipsec_ctx->tbl[i].id = cdev_id;
1274                 ipsec_ctx->tbl[i].qp = qp;
1275                 ipsec_ctx->nb_qps++;
1276                 printf("%s cdev mapping: lcore %u using cdev %u qp %u "
1277                                 "(cdev_id_qp %lu)\n", str, key.lcore_id,
1278                                 cdev_id, qp, i);
1279         }
1280
1281         ret = rte_hash_add_key_data(map, &key, (void *)i);
1282         if (ret < 0) {
1283                 printf("Faled to insert cdev mapping for (lcore %u, "
1284                                 "cdev %u, qp %u), errno %d\n",
1285                                 key.lcore_id, ipsec_ctx->tbl[i].id,
1286                                 ipsec_ctx->tbl[i].qp, ret);
1287                 return 0;
1288         }
1289
1290         return 1;
1291 }
1292
1293 static int32_t
1294 add_cdev_mapping(struct rte_cryptodev_info *dev_info, uint16_t cdev_id,
1295                 uint16_t qp, struct lcore_params *params)
1296 {
1297         int32_t ret = 0;
1298         const struct rte_cryptodev_capabilities *i, *j;
1299         struct rte_hash *map;
1300         struct lcore_conf *qconf;
1301         struct ipsec_ctx *ipsec_ctx;
1302         const char *str;
1303
1304         qconf = &lcore_conf[params->lcore_id];
1305
1306         if ((unprotected_port_mask & (1 << params->port_id)) == 0) {
1307                 map = cdev_map_out;
1308                 ipsec_ctx = &qconf->outbound;
1309                 str = "Outbound";
1310         } else {
1311                 map = cdev_map_in;
1312                 ipsec_ctx = &qconf->inbound;
1313                 str = "Inbound";
1314         }
1315
1316         /* Required cryptodevs with operation chainning */
1317         if (!(dev_info->feature_flags &
1318                                 RTE_CRYPTODEV_FF_SYM_OPERATION_CHAINING))
1319                 return ret;
1320
1321         for (i = dev_info->capabilities;
1322                         i->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; i++) {
1323                 if (i->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1324                         continue;
1325
1326                 if (i->sym.xform_type == RTE_CRYPTO_SYM_XFORM_AEAD) {
1327                         ret |= add_mapping(map, str, cdev_id, qp, params,
1328                                         ipsec_ctx, NULL, NULL, i);
1329                         continue;
1330                 }
1331
1332                 if (i->sym.xform_type != RTE_CRYPTO_SYM_XFORM_CIPHER)
1333                         continue;
1334
1335                 for (j = dev_info->capabilities;
1336                                 j->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; j++) {
1337                         if (j->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1338                                 continue;
1339
1340                         if (j->sym.xform_type != RTE_CRYPTO_SYM_XFORM_AUTH)
1341                                 continue;
1342
1343                         ret |= add_mapping(map, str, cdev_id, qp, params,
1344                                                 ipsec_ctx, i, j, NULL);
1345                 }
1346         }
1347
1348         return ret;
1349 }
1350
1351 /* Check if the device is enabled by cryptodev_mask */
1352 static int
1353 check_cryptodev_mask(uint8_t cdev_id)
1354 {
1355         if (enabled_cryptodev_mask & (1 << cdev_id))
1356                 return 0;
1357
1358         return -1;
1359 }
1360
1361 static int32_t
1362 cryptodevs_init(void)
1363 {
1364         struct rte_cryptodev_config dev_conf;
1365         struct rte_cryptodev_qp_conf qp_conf;
1366         uint16_t idx, max_nb_qps, qp, i;
1367         int16_t cdev_id, port_id;
1368         struct rte_hash_parameters params = { 0 };
1369
1370         params.entries = CDEV_MAP_ENTRIES;
1371         params.key_len = sizeof(struct cdev_key);
1372         params.hash_func = rte_jhash;
1373         params.hash_func_init_val = 0;
1374         params.socket_id = rte_socket_id();
1375
1376         params.name = "cdev_map_in";
1377         cdev_map_in = rte_hash_create(&params);
1378         if (cdev_map_in == NULL)
1379                 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1380                                 rte_errno);
1381
1382         params.name = "cdev_map_out";
1383         cdev_map_out = rte_hash_create(&params);
1384         if (cdev_map_out == NULL)
1385                 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1386                                 rte_errno);
1387
1388         printf("lcore/cryptodev/qp mappings:\n");
1389
1390         uint32_t max_sess_sz = 0, sess_sz;
1391         for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
1392                 sess_sz = rte_cryptodev_sym_get_private_session_size(cdev_id);
1393                 if (sess_sz > max_sess_sz)
1394                         max_sess_sz = sess_sz;
1395         }
1396         RTE_ETH_FOREACH_DEV(port_id) {
1397                 void *sec_ctx;
1398
1399                 if ((enabled_port_mask & (1 << port_id)) == 0)
1400                         continue;
1401
1402                 sec_ctx = rte_eth_dev_get_sec_ctx(port_id);
1403                 if (sec_ctx == NULL)
1404                         continue;
1405
1406                 sess_sz = rte_security_session_get_size(sec_ctx);
1407                 if (sess_sz > max_sess_sz)
1408                         max_sess_sz = sess_sz;
1409         }
1410
1411         idx = 0;
1412         for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
1413                 struct rte_cryptodev_info cdev_info;
1414
1415                 if (check_cryptodev_mask((uint8_t)cdev_id))
1416                         continue;
1417
1418                 rte_cryptodev_info_get(cdev_id, &cdev_info);
1419
1420                 if (nb_lcore_params > cdev_info.max_nb_queue_pairs)
1421                         max_nb_qps = cdev_info.max_nb_queue_pairs;
1422                 else
1423                         max_nb_qps = nb_lcore_params;
1424
1425                 qp = 0;
1426                 i = 0;
1427                 while (qp < max_nb_qps && i < nb_lcore_params) {
1428                         if (add_cdev_mapping(&cdev_info, cdev_id, qp,
1429                                                 &lcore_params[idx]))
1430                                 qp++;
1431                         idx++;
1432                         idx = idx % nb_lcore_params;
1433                         i++;
1434                 }
1435
1436                 if (qp == 0)
1437                         continue;
1438
1439                 dev_conf.socket_id = rte_cryptodev_socket_id(cdev_id);
1440                 dev_conf.nb_queue_pairs = qp;
1441
1442                 uint32_t dev_max_sess = cdev_info.sym.max_nb_sessions;
1443                 if (dev_max_sess != 0 && dev_max_sess < (CDEV_MP_NB_OBJS / 2))
1444                         rte_exit(EXIT_FAILURE,
1445                                 "Device does not support at least %u "
1446                                 "sessions", CDEV_MP_NB_OBJS / 2);
1447
1448                 if (!socket_ctx[dev_conf.socket_id].session_pool) {
1449                         char mp_name[RTE_MEMPOOL_NAMESIZE];
1450                         struct rte_mempool *sess_mp;
1451
1452                         snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
1453                                         "sess_mp_%u", dev_conf.socket_id);
1454                         sess_mp = rte_mempool_create(mp_name,
1455                                         CDEV_MP_NB_OBJS,
1456                                         max_sess_sz,
1457                                         CDEV_MP_CACHE_SZ,
1458                                         0, NULL, NULL, NULL,
1459                                         NULL, dev_conf.socket_id,
1460                                         0);
1461                         if (sess_mp == NULL)
1462                                 rte_exit(EXIT_FAILURE,
1463                                         "Cannot create session pool on socket %d\n",
1464                                         dev_conf.socket_id);
1465                         else
1466                                 printf("Allocated session pool on socket %d\n",
1467                                         dev_conf.socket_id);
1468                         socket_ctx[dev_conf.socket_id].session_pool = sess_mp;
1469                 }
1470
1471                 if (rte_cryptodev_configure(cdev_id, &dev_conf))
1472                         rte_panic("Failed to initialize cryptodev %u\n",
1473                                         cdev_id);
1474
1475                 qp_conf.nb_descriptors = CDEV_QUEUE_DESC;
1476                 for (qp = 0; qp < dev_conf.nb_queue_pairs; qp++)
1477                         if (rte_cryptodev_queue_pair_setup(cdev_id, qp,
1478                                         &qp_conf, dev_conf.socket_id,
1479                                         socket_ctx[dev_conf.socket_id].session_pool))
1480                                 rte_panic("Failed to setup queue %u for "
1481                                                 "cdev_id %u\n", 0, cdev_id);
1482
1483                 if (rte_cryptodev_start(cdev_id))
1484                         rte_panic("Failed to start cryptodev %u\n",
1485                                         cdev_id);
1486         }
1487
1488         /* create session pools for eth devices that implement security */
1489         RTE_ETH_FOREACH_DEV(port_id) {
1490                 if ((enabled_port_mask & (1 << port_id)) &&
1491                                 rte_eth_dev_get_sec_ctx(port_id)) {
1492                         int socket_id = rte_eth_dev_socket_id(port_id);
1493
1494                         if (!socket_ctx[socket_id].session_pool) {
1495                                 char mp_name[RTE_MEMPOOL_NAMESIZE];
1496                                 struct rte_mempool *sess_mp;
1497
1498                                 snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
1499                                                 "sess_mp_%u", socket_id);
1500                                 sess_mp = rte_mempool_create(mp_name,
1501                                                 CDEV_MP_NB_OBJS,
1502                                                 max_sess_sz,
1503                                                 CDEV_MP_CACHE_SZ,
1504                                                 0, NULL, NULL, NULL,
1505                                                 NULL, socket_id,
1506                                                 0);
1507                                 if (sess_mp == NULL)
1508                                         rte_exit(EXIT_FAILURE,
1509                                                 "Cannot create session pool "
1510                                                 "on socket %d\n", socket_id);
1511                                 else
1512                                         printf("Allocated session pool "
1513                                                 "on socket %d\n", socket_id);
1514                                 socket_ctx[socket_id].session_pool = sess_mp;
1515                         }
1516                 }
1517         }
1518
1519
1520         printf("\n");
1521
1522         return 0;
1523 }
1524
1525 static void
1526 port_init(uint16_t portid)
1527 {
1528         struct rte_eth_dev_info dev_info;
1529         struct rte_eth_txconf *txconf;
1530         uint16_t nb_tx_queue, nb_rx_queue;
1531         uint16_t tx_queueid, rx_queueid, queue, lcore_id;
1532         int32_t ret, socket_id;
1533         struct lcore_conf *qconf;
1534         struct ether_addr ethaddr;
1535         struct rte_eth_conf local_port_conf = port_conf;
1536
1537         rte_eth_dev_info_get(portid, &dev_info);
1538
1539         printf("Configuring device port %u:\n", portid);
1540
1541         rte_eth_macaddr_get(portid, &ethaddr);
1542         ethaddr_tbl[portid].src = ETHADDR_TO_UINT64(ethaddr);
1543         print_ethaddr("Address: ", &ethaddr);
1544         printf("\n");
1545
1546         nb_rx_queue = get_port_nb_rx_queues(portid);
1547         nb_tx_queue = nb_lcores;
1548
1549         if (nb_rx_queue > dev_info.max_rx_queues)
1550                 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
1551                                 "(max rx queue is %u)\n",
1552                                 nb_rx_queue, dev_info.max_rx_queues);
1553
1554         if (nb_tx_queue > dev_info.max_tx_queues)
1555                 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
1556                                 "(max tx queue is %u)\n",
1557                                 nb_tx_queue, dev_info.max_tx_queues);
1558
1559         printf("Creating queues: nb_rx_queue=%d nb_tx_queue=%u...\n",
1560                         nb_rx_queue, nb_tx_queue);
1561
1562         if (frame_size) {
1563                 local_port_conf.rxmode.max_rx_pkt_len = frame_size;
1564                 local_port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
1565         }
1566
1567         if (dev_info.rx_offload_capa & DEV_RX_OFFLOAD_SECURITY)
1568                 local_port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_SECURITY;
1569         if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_SECURITY)
1570                 local_port_conf.txmode.offloads |= DEV_TX_OFFLOAD_SECURITY;
1571         if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
1572                 local_port_conf.txmode.offloads |=
1573                         DEV_TX_OFFLOAD_MBUF_FAST_FREE;
1574
1575         local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
1576                 dev_info.flow_type_rss_offloads;
1577         if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
1578                         port_conf.rx_adv_conf.rss_conf.rss_hf) {
1579                 printf("Port %u modified RSS hash function based on hardware support,"
1580                         "requested:%#"PRIx64" configured:%#"PRIx64"\n",
1581                         portid,
1582                         port_conf.rx_adv_conf.rss_conf.rss_hf,
1583                         local_port_conf.rx_adv_conf.rss_conf.rss_hf);
1584         }
1585
1586         ret = rte_eth_dev_configure(portid, nb_rx_queue, nb_tx_queue,
1587                         &local_port_conf);
1588         if (ret < 0)
1589                 rte_exit(EXIT_FAILURE, "Cannot configure device: "
1590                                 "err=%d, port=%d\n", ret, portid);
1591
1592         ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd, &nb_txd);
1593         if (ret < 0)
1594                 rte_exit(EXIT_FAILURE, "Cannot adjust number of descriptors: "
1595                                 "err=%d, port=%d\n", ret, portid);
1596
1597         /* init one TX queue per lcore */
1598         tx_queueid = 0;
1599         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1600                 if (rte_lcore_is_enabled(lcore_id) == 0)
1601                         continue;
1602
1603                 if (numa_on)
1604                         socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1605                 else
1606                         socket_id = 0;
1607
1608                 /* init TX queue */
1609                 printf("Setup txq=%u,%d,%d\n", lcore_id, tx_queueid, socket_id);
1610
1611                 txconf = &dev_info.default_txconf;
1612                 txconf->offloads = local_port_conf.txmode.offloads;
1613
1614                 ret = rte_eth_tx_queue_setup(portid, tx_queueid, nb_txd,
1615                                 socket_id, txconf);
1616                 if (ret < 0)
1617                         rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
1618                                         "err=%d, port=%d\n", ret, portid);
1619
1620                 qconf = &lcore_conf[lcore_id];
1621                 qconf->tx_queue_id[portid] = tx_queueid;
1622                 tx_queueid++;
1623
1624                 /* init RX queues */
1625                 for (queue = 0; queue < qconf->nb_rx_queue; ++queue) {
1626                         struct rte_eth_rxconf rxq_conf;
1627
1628                         if (portid != qconf->rx_queue_list[queue].port_id)
1629                                 continue;
1630
1631                         rx_queueid = qconf->rx_queue_list[queue].queue_id;
1632
1633                         printf("Setup rxq=%d,%d,%d\n", portid, rx_queueid,
1634                                         socket_id);
1635
1636                         rxq_conf = dev_info.default_rxconf;
1637                         rxq_conf.offloads = local_port_conf.rxmode.offloads;
1638                         ret = rte_eth_rx_queue_setup(portid, rx_queueid,
1639                                         nb_rxd, socket_id, &rxq_conf,
1640                                         socket_ctx[socket_id].mbuf_pool);
1641                         if (ret < 0)
1642                                 rte_exit(EXIT_FAILURE,
1643                                         "rte_eth_rx_queue_setup: err=%d, "
1644                                         "port=%d\n", ret, portid);
1645                 }
1646         }
1647         printf("\n");
1648 }
1649
1650 static void
1651 pool_init(struct socket_ctx *ctx, int32_t socket_id, uint32_t nb_mbuf)
1652 {
1653         char s[64];
1654         uint32_t buff_size = frame_size ? (frame_size + RTE_PKTMBUF_HEADROOM) :
1655                         RTE_MBUF_DEFAULT_BUF_SIZE;
1656
1657
1658         snprintf(s, sizeof(s), "mbuf_pool_%d", socket_id);
1659         ctx->mbuf_pool = rte_pktmbuf_pool_create(s, nb_mbuf,
1660                         MEMPOOL_CACHE_SIZE, ipsec_metadata_size(),
1661                         buff_size,
1662                         socket_id);
1663         if (ctx->mbuf_pool == NULL)
1664                 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool on socket %d\n",
1665                                 socket_id);
1666         else
1667                 printf("Allocated mbuf pool on socket %d\n", socket_id);
1668 }
1669
1670 static inline int
1671 inline_ipsec_event_esn_overflow(struct rte_security_ctx *ctx, uint64_t md)
1672 {
1673         struct ipsec_sa *sa;
1674
1675         /* For inline protocol processing, the metadata in the event will
1676          * uniquely identify the security session which raised the event.
1677          * Application would then need the userdata it had registered with the
1678          * security session to process the event.
1679          */
1680
1681         sa = (struct ipsec_sa *)rte_security_get_userdata(ctx, md);
1682
1683         if (sa == NULL) {
1684                 /* userdata could not be retrieved */
1685                 return -1;
1686         }
1687
1688         /* Sequence number over flow. SA need to be re-established */
1689         RTE_SET_USED(sa);
1690         return 0;
1691 }
1692
1693 static int
1694 inline_ipsec_event_callback(uint16_t port_id, enum rte_eth_event_type type,
1695                  void *param, void *ret_param)
1696 {
1697         uint64_t md;
1698         struct rte_eth_event_ipsec_desc *event_desc = NULL;
1699         struct rte_security_ctx *ctx = (struct rte_security_ctx *)
1700                                         rte_eth_dev_get_sec_ctx(port_id);
1701
1702         RTE_SET_USED(param);
1703
1704         if (type != RTE_ETH_EVENT_IPSEC)
1705                 return -1;
1706
1707         event_desc = ret_param;
1708         if (event_desc == NULL) {
1709                 printf("Event descriptor not set\n");
1710                 return -1;
1711         }
1712
1713         md = event_desc->metadata;
1714
1715         if (event_desc->subtype == RTE_ETH_EVENT_IPSEC_ESN_OVERFLOW)
1716                 return inline_ipsec_event_esn_overflow(ctx, md);
1717         else if (event_desc->subtype >= RTE_ETH_EVENT_IPSEC_MAX) {
1718                 printf("Invalid IPsec event reported\n");
1719                 return -1;
1720         }
1721
1722         return -1;
1723 }
1724
1725 int32_t
1726 main(int32_t argc, char **argv)
1727 {
1728         int32_t ret;
1729         uint32_t lcore_id;
1730         uint8_t socket_id;
1731         uint16_t portid;
1732
1733         /* init EAL */
1734         ret = rte_eal_init(argc, argv);
1735         if (ret < 0)
1736                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1737         argc -= ret;
1738         argv += ret;
1739
1740         /* parse application arguments (after the EAL ones) */
1741         ret = parse_args(argc, argv);
1742         if (ret < 0)
1743                 rte_exit(EXIT_FAILURE, "Invalid parameters\n");
1744
1745         if ((unprotected_port_mask & enabled_port_mask) !=
1746                         unprotected_port_mask)
1747                 rte_exit(EXIT_FAILURE, "Invalid unprotected portmask 0x%x\n",
1748                                 unprotected_port_mask);
1749
1750         if (check_params() < 0)
1751                 rte_exit(EXIT_FAILURE, "check_params failed\n");
1752
1753         ret = init_lcore_rx_queues();
1754         if (ret < 0)
1755                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1756
1757         nb_lcores = rte_lcore_count();
1758
1759         /* Replicate each context per socket */
1760         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1761                 if (rte_lcore_is_enabled(lcore_id) == 0)
1762                         continue;
1763
1764                 if (numa_on)
1765                         socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1766                 else
1767                         socket_id = 0;
1768
1769                 if (socket_ctx[socket_id].mbuf_pool)
1770                         continue;
1771
1772                 sa_init(&socket_ctx[socket_id], socket_id);
1773
1774                 sp4_init(&socket_ctx[socket_id], socket_id);
1775
1776                 sp6_init(&socket_ctx[socket_id], socket_id);
1777
1778                 rt_init(&socket_ctx[socket_id], socket_id);
1779
1780                 pool_init(&socket_ctx[socket_id], socket_id, NB_MBUF);
1781         }
1782
1783         RTE_ETH_FOREACH_DEV(portid) {
1784                 if ((enabled_port_mask & (1 << portid)) == 0)
1785                         continue;
1786
1787                 port_init(portid);
1788         }
1789
1790         cryptodevs_init();
1791
1792         /* start ports */
1793         RTE_ETH_FOREACH_DEV(portid) {
1794                 if ((enabled_port_mask & (1 << portid)) == 0)
1795                         continue;
1796
1797                 /* Start device */
1798                 ret = rte_eth_dev_start(portid);
1799                 if (ret < 0)
1800                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: "
1801                                         "err=%d, port=%d\n", ret, portid);
1802                 /*
1803                  * If enabled, put device in promiscuous mode.
1804                  * This allows IO forwarding mode to forward packets
1805                  * to itself through 2 cross-connected  ports of the
1806                  * target machine.
1807                  */
1808                 if (promiscuous_on)
1809                         rte_eth_promiscuous_enable(portid);
1810
1811                 rte_eth_dev_callback_register(portid,
1812                         RTE_ETH_EVENT_IPSEC, inline_ipsec_event_callback, NULL);
1813         }
1814
1815         check_all_ports_link_status(enabled_port_mask);
1816
1817         /* launch per-lcore init on every lcore */
1818         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1819         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1820                 if (rte_eal_wait_lcore(lcore_id) < 0)
1821                         return -1;
1822         }
1823
1824         return 0;
1825 }