1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2016 Intel Corporation
10 #include <sys/types.h>
11 #include <netinet/in.h>
12 #include <netinet/ip.h>
13 #include <netinet/ip6.h>
15 #include <sys/queue.h>
21 #include <rte_common.h>
22 #include <rte_bitmap.h>
23 #include <rte_byteorder.h>
26 #include <rte_launch.h>
27 #include <rte_atomic.h>
28 #include <rte_cycles.h>
29 #include <rte_prefetch.h>
30 #include <rte_lcore.h>
31 #include <rte_per_lcore.h>
32 #include <rte_branch_prediction.h>
33 #include <rte_interrupts.h>
34 #include <rte_random.h>
35 #include <rte_debug.h>
36 #include <rte_ether.h>
37 #include <rte_ethdev.h>
38 #include <rte_mempool.h>
44 #include <rte_jhash.h>
45 #include <rte_cryptodev.h>
46 #include <rte_security.h>
47 #include <rte_eventdev.h>
49 #include <rte_ip_frag.h>
50 #include <rte_alarm.h>
52 #include "event_helper.h"
55 #include "ipsec_worker.h"
59 volatile bool force_quit;
61 #define MAX_JUMBO_PKT_LEN 9600
63 #define MEMPOOL_CACHE_SIZE 256
65 #define CDEV_QUEUE_DESC 2048
66 #define CDEV_MAP_ENTRIES 16384
67 #define CDEV_MP_CACHE_SZ 64
68 #define MAX_QUEUE_PAIRS 1
70 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
72 /* Configure how many packets ahead to prefetch, when reading packets */
73 #define PREFETCH_OFFSET 3
75 #define MAX_RX_QUEUE_PER_LCORE 16
77 #define MAX_LCORE_PARAMS 1024
80 * Configurable number of RX/TX ring descriptors
82 #define IPSEC_SECGW_RX_DESC_DEFAULT 1024
83 #define IPSEC_SECGW_TX_DESC_DEFAULT 1024
84 static uint16_t nb_rxd = IPSEC_SECGW_RX_DESC_DEFAULT;
85 static uint16_t nb_txd = IPSEC_SECGW_TX_DESC_DEFAULT;
87 #define ETHADDR_TO_UINT64(addr) __BYTES_TO_UINT64( \
88 (addr)->addr_bytes[0], (addr)->addr_bytes[1], \
89 (addr)->addr_bytes[2], (addr)->addr_bytes[3], \
90 (addr)->addr_bytes[4], (addr)->addr_bytes[5], \
93 #define FRAG_TBL_BUCKET_ENTRIES 4
94 #define MAX_FRAG_TTL_NS (10LL * NS_PER_S)
96 #define MTU_TO_FRAMELEN(x) ((x) + RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN)
98 struct ethaddr_info ethaddr_tbl[RTE_MAX_ETHPORTS] = {
99 { 0, ETHADDR(0x00, 0x16, 0x3e, 0x7e, 0x94, 0x9a) },
100 { 0, ETHADDR(0x00, 0x16, 0x3e, 0x22, 0xa1, 0xd9) },
101 { 0, ETHADDR(0x00, 0x16, 0x3e, 0x08, 0x69, 0x26) },
102 { 0, ETHADDR(0x00, 0x16, 0x3e, 0x49, 0x9e, 0xdd) }
105 struct flow_info flow_info_tbl[RTE_MAX_ETHPORTS];
107 #define CMD_LINE_OPT_CONFIG "config"
108 #define CMD_LINE_OPT_SINGLE_SA "single-sa"
109 #define CMD_LINE_OPT_CRYPTODEV_MASK "cryptodev_mask"
110 #define CMD_LINE_OPT_TRANSFER_MODE "transfer-mode"
111 #define CMD_LINE_OPT_SCHEDULE_TYPE "event-schedule-type"
112 #define CMD_LINE_OPT_RX_OFFLOAD "rxoffload"
113 #define CMD_LINE_OPT_TX_OFFLOAD "txoffload"
114 #define CMD_LINE_OPT_REASSEMBLE "reassemble"
115 #define CMD_LINE_OPT_MTU "mtu"
116 #define CMD_LINE_OPT_FRAG_TTL "frag-ttl"
118 #define CMD_LINE_ARG_EVENT "event"
119 #define CMD_LINE_ARG_POLL "poll"
120 #define CMD_LINE_ARG_ORDERED "ordered"
121 #define CMD_LINE_ARG_ATOMIC "atomic"
122 #define CMD_LINE_ARG_PARALLEL "parallel"
125 /* long options mapped to a short option */
127 /* first long only option value must be >= 256, so that we won't
128 * conflict with short options
130 CMD_LINE_OPT_MIN_NUM = 256,
131 CMD_LINE_OPT_CONFIG_NUM,
132 CMD_LINE_OPT_SINGLE_SA_NUM,
133 CMD_LINE_OPT_CRYPTODEV_MASK_NUM,
134 CMD_LINE_OPT_TRANSFER_MODE_NUM,
135 CMD_LINE_OPT_SCHEDULE_TYPE_NUM,
136 CMD_LINE_OPT_RX_OFFLOAD_NUM,
137 CMD_LINE_OPT_TX_OFFLOAD_NUM,
138 CMD_LINE_OPT_REASSEMBLE_NUM,
139 CMD_LINE_OPT_MTU_NUM,
140 CMD_LINE_OPT_FRAG_TTL_NUM,
143 static const struct option lgopts[] = {
144 {CMD_LINE_OPT_CONFIG, 1, 0, CMD_LINE_OPT_CONFIG_NUM},
145 {CMD_LINE_OPT_SINGLE_SA, 1, 0, CMD_LINE_OPT_SINGLE_SA_NUM},
146 {CMD_LINE_OPT_CRYPTODEV_MASK, 1, 0, CMD_LINE_OPT_CRYPTODEV_MASK_NUM},
147 {CMD_LINE_OPT_TRANSFER_MODE, 1, 0, CMD_LINE_OPT_TRANSFER_MODE_NUM},
148 {CMD_LINE_OPT_SCHEDULE_TYPE, 1, 0, CMD_LINE_OPT_SCHEDULE_TYPE_NUM},
149 {CMD_LINE_OPT_RX_OFFLOAD, 1, 0, CMD_LINE_OPT_RX_OFFLOAD_NUM},
150 {CMD_LINE_OPT_TX_OFFLOAD, 1, 0, CMD_LINE_OPT_TX_OFFLOAD_NUM},
151 {CMD_LINE_OPT_REASSEMBLE, 1, 0, CMD_LINE_OPT_REASSEMBLE_NUM},
152 {CMD_LINE_OPT_MTU, 1, 0, CMD_LINE_OPT_MTU_NUM},
153 {CMD_LINE_OPT_FRAG_TTL, 1, 0, CMD_LINE_OPT_FRAG_TTL_NUM},
157 uint32_t unprotected_port_mask;
158 uint32_t single_sa_idx;
159 /* mask of enabled ports */
160 static uint32_t enabled_port_mask;
161 static uint64_t enabled_cryptodev_mask = UINT64_MAX;
162 static int32_t promiscuous_on = 1;
163 static int32_t numa_on = 1; /**< NUMA is enabled by default. */
164 static uint32_t nb_lcores;
165 static uint32_t single_sa;
166 static uint32_t nb_bufs_in_pool;
169 * RX/TX HW offload capabilities to enable/use on ethernet ports.
170 * By default all capabilities are enabled.
172 static uint64_t dev_rx_offload = UINT64_MAX;
173 static uint64_t dev_tx_offload = UINT64_MAX;
176 * global values that determine multi-seg policy
178 static uint32_t frag_tbl_sz;
179 static uint32_t frame_buf_size = RTE_MBUF_DEFAULT_BUF_SIZE;
180 static uint32_t mtu_size = RTE_ETHER_MTU;
181 static uint64_t frag_ttl_ns = MAX_FRAG_TTL_NS;
183 /* application wide librte_ipsec/SA parameters */
184 struct app_sa_prm app_sa_prm = {
186 .cache_sz = SA_CACHE_SZ
188 static const char *cfgfile;
190 struct lcore_rx_queue {
193 } __rte_cache_aligned;
195 struct lcore_params {
199 } __rte_cache_aligned;
201 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
203 static struct lcore_params *lcore_params;
204 static uint16_t nb_lcore_params;
206 static struct rte_hash *cdev_map_in;
207 static struct rte_hash *cdev_map_out;
211 struct rte_mbuf *m_table[MAX_PKT_BURST] __rte_aligned(sizeof(void *));
215 uint16_t nb_rx_queue;
216 struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
217 uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
218 struct buffer tx_mbufs[RTE_MAX_ETHPORTS];
219 struct ipsec_ctx inbound;
220 struct ipsec_ctx outbound;
221 struct rt_ctx *rt4_ctx;
222 struct rt_ctx *rt6_ctx;
224 struct rte_ip_frag_tbl *tbl;
225 struct rte_mempool *pool_dir;
226 struct rte_mempool *pool_indir;
227 struct rte_ip_frag_death_row dr;
229 } __rte_cache_aligned;
231 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
233 static struct rte_eth_conf port_conf = {
235 .mq_mode = ETH_MQ_RX_RSS,
236 .max_rx_pkt_len = RTE_ETHER_MAX_LEN,
238 .offloads = DEV_RX_OFFLOAD_CHECKSUM,
243 .rss_hf = ETH_RSS_IP | ETH_RSS_UDP |
244 ETH_RSS_TCP | ETH_RSS_SCTP,
248 .mq_mode = ETH_MQ_TX_NONE,
252 struct socket_ctx socket_ctx[NB_SOCKETS];
255 * Determine is multi-segment support required:
256 * - either frame buffer size is smaller then mtu
257 * - or reassmeble support is requested
260 multi_seg_required(void)
262 return (MTU_TO_FRAMELEN(mtu_size) + RTE_PKTMBUF_HEADROOM >
263 frame_buf_size || frag_tbl_sz != 0);
267 adjust_ipv4_pktlen(struct rte_mbuf *m, const struct rte_ipv4_hdr *iph,
272 plen = rte_be_to_cpu_16(iph->total_length) + l2_len;
273 if (plen < m->pkt_len) {
274 trim = m->pkt_len - plen;
275 rte_pktmbuf_trim(m, trim);
280 adjust_ipv6_pktlen(struct rte_mbuf *m, const struct rte_ipv6_hdr *iph,
285 plen = rte_be_to_cpu_16(iph->payload_len) + sizeof(*iph) + l2_len;
286 if (plen < m->pkt_len) {
287 trim = m->pkt_len - plen;
288 rte_pktmbuf_trim(m, trim);
292 #if (STATS_INTERVAL > 0)
294 /* Print out statistics on packet distribution */
296 print_stats_cb(__rte_unused void *param)
298 uint64_t total_packets_dropped, total_packets_tx, total_packets_rx;
299 float burst_percent, rx_per_call, tx_per_call;
302 total_packets_dropped = 0;
303 total_packets_tx = 0;
304 total_packets_rx = 0;
306 const char clr[] = { 27, '[', '2', 'J', '\0' };
307 const char topLeft[] = { 27, '[', '1', ';', '1', 'H', '\0' };
309 /* Clear screen and move to top left */
310 printf("%s%s", clr, topLeft);
312 printf("\nCore statistics ====================================");
314 for (coreid = 0; coreid < RTE_MAX_LCORE; coreid++) {
315 /* skip disabled cores */
316 if (rte_lcore_is_enabled(coreid) == 0)
318 burst_percent = (float)(core_statistics[coreid].burst_rx * 100)/
319 core_statistics[coreid].rx;
320 rx_per_call = (float)(core_statistics[coreid].rx)/
321 core_statistics[coreid].rx_call;
322 tx_per_call = (float)(core_statistics[coreid].tx)/
323 core_statistics[coreid].tx_call;
324 printf("\nStatistics for core %u ------------------------------"
325 "\nPackets received: %20"PRIu64
326 "\nPackets sent: %24"PRIu64
327 "\nPackets dropped: %21"PRIu64
328 "\nBurst percent: %23.2f"
329 "\nPackets per Rx call: %17.2f"
330 "\nPackets per Tx call: %17.2f",
332 core_statistics[coreid].rx,
333 core_statistics[coreid].tx,
334 core_statistics[coreid].dropped,
339 total_packets_dropped += core_statistics[coreid].dropped;
340 total_packets_tx += core_statistics[coreid].tx;
341 total_packets_rx += core_statistics[coreid].rx;
343 printf("\nAggregate statistics ==============================="
344 "\nTotal packets received: %14"PRIu64
345 "\nTotal packets sent: %18"PRIu64
346 "\nTotal packets dropped: %15"PRIu64,
349 total_packets_dropped);
350 printf("\n====================================================\n");
352 rte_eal_alarm_set(STATS_INTERVAL * US_PER_S, print_stats_cb, NULL);
354 #endif /* STATS_INTERVAL */
357 prepare_one_packet(struct rte_mbuf *pkt, struct ipsec_traffic *t)
359 const struct rte_ether_hdr *eth;
360 const struct rte_ipv4_hdr *iph4;
361 const struct rte_ipv6_hdr *iph6;
363 eth = rte_pktmbuf_mtod(pkt, const struct rte_ether_hdr *);
364 if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
366 iph4 = (const struct rte_ipv4_hdr *)rte_pktmbuf_adj(pkt,
368 adjust_ipv4_pktlen(pkt, iph4, 0);
370 if (iph4->next_proto_id == IPPROTO_ESP)
371 t->ipsec.pkts[(t->ipsec.num)++] = pkt;
373 t->ip4.data[t->ip4.num] = &iph4->next_proto_id;
374 t->ip4.pkts[(t->ip4.num)++] = pkt;
377 pkt->l3_len = sizeof(*iph4);
378 pkt->packet_type |= RTE_PTYPE_L3_IPV4;
379 } else if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6)) {
381 size_t l3len, ext_len;
384 /* get protocol type */
385 iph6 = (const struct rte_ipv6_hdr *)rte_pktmbuf_adj(pkt,
387 adjust_ipv6_pktlen(pkt, iph6, 0);
389 next_proto = iph6->proto;
391 /* determine l3 header size up to ESP extension */
392 l3len = sizeof(struct ip6_hdr);
393 p = rte_pktmbuf_mtod(pkt, uint8_t *);
394 while (next_proto != IPPROTO_ESP && l3len < pkt->data_len &&
395 (next_proto = rte_ipv6_get_next_ext(p + l3len,
396 next_proto, &ext_len)) >= 0)
399 /* drop packet when IPv6 header exceeds first segment length */
400 if (unlikely(l3len > pkt->data_len)) {
405 if (next_proto == IPPROTO_ESP)
406 t->ipsec.pkts[(t->ipsec.num)++] = pkt;
408 t->ip6.data[t->ip6.num] = &iph6->proto;
409 t->ip6.pkts[(t->ip6.num)++] = pkt;
413 pkt->packet_type |= RTE_PTYPE_L3_IPV6;
415 /* Unknown/Unsupported type, drop the packet */
416 RTE_LOG(ERR, IPSEC, "Unsupported packet type 0x%x\n",
417 rte_be_to_cpu_16(eth->ether_type));
422 /* Check if the packet has been processed inline. For inline protocol
423 * processed packets, the metadata in the mbuf can be used to identify
424 * the security processing done on the packet. The metadata will be
425 * used to retrieve the application registered userdata associated
426 * with the security session.
429 if (pkt->ol_flags & PKT_RX_SEC_OFFLOAD) {
431 struct ipsec_mbuf_metadata *priv;
432 struct rte_security_ctx *ctx = (struct rte_security_ctx *)
433 rte_eth_dev_get_sec_ctx(
436 /* Retrieve the userdata registered. Here, the userdata
437 * registered is the SA pointer.
440 sa = (struct ipsec_sa *)
441 rte_security_get_userdata(ctx, pkt->udata64);
444 /* userdata could not be retrieved */
448 /* Save SA as priv member in mbuf. This will be used in the
449 * IPsec selector(SP-SA) check.
452 priv = get_priv(pkt);
458 prepare_traffic(struct rte_mbuf **pkts, struct ipsec_traffic *t,
467 for (i = 0; i < (nb_pkts - PREFETCH_OFFSET); i++) {
468 rte_prefetch0(rte_pktmbuf_mtod(pkts[i + PREFETCH_OFFSET],
470 prepare_one_packet(pkts[i], t);
472 /* Process left packets */
473 for (; i < nb_pkts; i++)
474 prepare_one_packet(pkts[i], t);
478 prepare_tx_pkt(struct rte_mbuf *pkt, uint16_t port,
479 const struct lcore_conf *qconf)
482 struct rte_ether_hdr *ethhdr;
484 ip = rte_pktmbuf_mtod(pkt, struct ip *);
486 ethhdr = (struct rte_ether_hdr *)
487 rte_pktmbuf_prepend(pkt, RTE_ETHER_HDR_LEN);
489 if (ip->ip_v == IPVERSION) {
490 pkt->ol_flags |= qconf->outbound.ipv4_offloads;
491 pkt->l3_len = sizeof(struct ip);
492 pkt->l2_len = RTE_ETHER_HDR_LEN;
496 /* calculate IPv4 cksum in SW */
497 if ((pkt->ol_flags & PKT_TX_IP_CKSUM) == 0)
498 ip->ip_sum = rte_ipv4_cksum((struct rte_ipv4_hdr *)ip);
500 ethhdr->ether_type = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4);
502 pkt->ol_flags |= qconf->outbound.ipv6_offloads;
503 pkt->l3_len = sizeof(struct ip6_hdr);
504 pkt->l2_len = RTE_ETHER_HDR_LEN;
506 ethhdr->ether_type = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6);
509 memcpy(ðhdr->s_addr, ðaddr_tbl[port].src,
510 sizeof(struct rte_ether_addr));
511 memcpy(ðhdr->d_addr, ðaddr_tbl[port].dst,
512 sizeof(struct rte_ether_addr));
516 prepare_tx_burst(struct rte_mbuf *pkts[], uint16_t nb_pkts, uint16_t port,
517 const struct lcore_conf *qconf)
520 const int32_t prefetch_offset = 2;
522 for (i = 0; i < (nb_pkts - prefetch_offset); i++) {
523 rte_mbuf_prefetch_part2(pkts[i + prefetch_offset]);
524 prepare_tx_pkt(pkts[i], port, qconf);
526 /* Process left packets */
527 for (; i < nb_pkts; i++)
528 prepare_tx_pkt(pkts[i], port, qconf);
531 /* Send burst of packets on an output interface */
532 static inline int32_t
533 send_burst(struct lcore_conf *qconf, uint16_t n, uint16_t port)
535 struct rte_mbuf **m_table;
539 queueid = qconf->tx_queue_id[port];
540 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
542 prepare_tx_burst(m_table, n, port, qconf);
544 ret = rte_eth_tx_burst(port, queueid, m_table, n);
546 core_stats_update_tx(ret);
548 if (unlikely(ret < n)) {
550 free_pkts(&m_table[ret], 1);
558 * Helper function to fragment and queue for TX one packet.
560 static inline uint32_t
561 send_fragment_packet(struct lcore_conf *qconf, struct rte_mbuf *m,
562 uint16_t port, uint8_t proto)
568 tbl = qconf->tx_mbufs + port;
571 /* free space for new fragments */
572 if (len + RTE_LIBRTE_IP_FRAG_MAX_FRAG >= RTE_DIM(tbl->m_table)) {
573 send_burst(qconf, len, port);
577 n = RTE_DIM(tbl->m_table) - len;
579 if (proto == IPPROTO_IP)
580 rc = rte_ipv4_fragment_packet(m, tbl->m_table + len,
581 n, mtu_size, qconf->frag.pool_dir,
582 qconf->frag.pool_indir);
584 rc = rte_ipv6_fragment_packet(m, tbl->m_table + len,
585 n, mtu_size, qconf->frag.pool_dir,
586 qconf->frag.pool_indir);
592 "%s: failed to fragment packet with size %u, "
594 __func__, m->pkt_len, rte_errno);
600 /* Enqueue a single packet, and send burst if queue is filled */
601 static inline int32_t
602 send_single_packet(struct rte_mbuf *m, uint16_t port, uint8_t proto)
606 struct lcore_conf *qconf;
608 lcore_id = rte_lcore_id();
610 qconf = &lcore_conf[lcore_id];
611 len = qconf->tx_mbufs[port].len;
613 if (m->pkt_len <= mtu_size) {
614 qconf->tx_mbufs[port].m_table[len] = m;
617 /* need to fragment the packet */
618 } else if (frag_tbl_sz > 0)
619 len = send_fragment_packet(qconf, m, port, proto);
623 /* enough pkts to be sent */
624 if (unlikely(len == MAX_PKT_BURST)) {
625 send_burst(qconf, MAX_PKT_BURST, port);
629 qconf->tx_mbufs[port].len = len;
634 inbound_sp_sa(struct sp_ctx *sp, struct sa_ctx *sa, struct traffic_type *ip,
638 uint32_t i, j, res, sa_idx;
640 if (ip->num == 0 || sp == NULL)
643 rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
644 ip->num, DEFAULT_MAX_CATEGORIES);
647 for (i = 0; i < ip->num; i++) {
654 if (res == DISCARD) {
659 /* Only check SPI match for processed IPSec packets */
660 if (i < lim && ((m->ol_flags & PKT_RX_SEC_OFFLOAD) == 0)) {
666 if (!inbound_sa_check(sa, m, sa_idx)) {
676 split46_traffic(struct ipsec_traffic *trf, struct rte_mbuf *mb[], uint32_t num)
685 for (i = 0; i < num; i++) {
688 ip = rte_pktmbuf_mtod(m, struct ip *);
690 if (ip->ip_v == IPVERSION) {
691 trf->ip4.pkts[n4] = m;
692 trf->ip4.data[n4] = rte_pktmbuf_mtod_offset(m,
693 uint8_t *, offsetof(struct ip, ip_p));
695 } else if (ip->ip_v == IP6_VERSION) {
696 trf->ip6.pkts[n6] = m;
697 trf->ip6.data[n6] = rte_pktmbuf_mtod_offset(m,
699 offsetof(struct ip6_hdr, ip6_nxt));
711 process_pkts_inbound(struct ipsec_ctx *ipsec_ctx,
712 struct ipsec_traffic *traffic)
714 uint16_t nb_pkts_in, n_ip4, n_ip6;
716 n_ip4 = traffic->ip4.num;
717 n_ip6 = traffic->ip6.num;
719 if (app_sa_prm.enable == 0) {
720 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
721 traffic->ipsec.num, MAX_PKT_BURST);
722 split46_traffic(traffic, traffic->ipsec.pkts, nb_pkts_in);
724 inbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.pkts,
725 traffic->ipsec.saptr, traffic->ipsec.num);
726 ipsec_process(ipsec_ctx, traffic);
729 inbound_sp_sa(ipsec_ctx->sp4_ctx, ipsec_ctx->sa_ctx, &traffic->ip4,
732 inbound_sp_sa(ipsec_ctx->sp6_ctx, ipsec_ctx->sa_ctx, &traffic->ip6,
737 outbound_sp(struct sp_ctx *sp, struct traffic_type *ip,
738 struct traffic_type *ipsec)
741 uint32_t i, j, sa_idx;
743 if (ip->num == 0 || sp == NULL)
746 rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
747 ip->num, DEFAULT_MAX_CATEGORIES);
750 for (i = 0; i < ip->num; i++) {
752 sa_idx = ip->res[i] - 1;
753 if (ip->res[i] == DISCARD)
755 else if (ip->res[i] == BYPASS)
758 ipsec->res[ipsec->num] = sa_idx;
759 ipsec->pkts[ipsec->num++] = m;
766 process_pkts_outbound(struct ipsec_ctx *ipsec_ctx,
767 struct ipsec_traffic *traffic)
770 uint16_t idx, nb_pkts_out, i;
772 /* Drop any IPsec traffic from protected ports */
773 free_pkts(traffic->ipsec.pkts, traffic->ipsec.num);
775 traffic->ipsec.num = 0;
777 outbound_sp(ipsec_ctx->sp4_ctx, &traffic->ip4, &traffic->ipsec);
779 outbound_sp(ipsec_ctx->sp6_ctx, &traffic->ip6, &traffic->ipsec);
781 if (app_sa_prm.enable == 0) {
783 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
784 traffic->ipsec.res, traffic->ipsec.num,
787 for (i = 0; i < nb_pkts_out; i++) {
788 m = traffic->ipsec.pkts[i];
789 struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
790 if (ip->ip_v == IPVERSION) {
791 idx = traffic->ip4.num++;
792 traffic->ip4.pkts[idx] = m;
794 idx = traffic->ip6.num++;
795 traffic->ip6.pkts[idx] = m;
799 outbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.res,
800 traffic->ipsec.saptr, traffic->ipsec.num);
801 ipsec_process(ipsec_ctx, traffic);
806 process_pkts_inbound_nosp(struct ipsec_ctx *ipsec_ctx,
807 struct ipsec_traffic *traffic)
810 uint32_t nb_pkts_in, i, idx;
812 /* Drop any IPv4 traffic from unprotected ports */
813 free_pkts(traffic->ip4.pkts, traffic->ip4.num);
815 traffic->ip4.num = 0;
817 /* Drop any IPv6 traffic from unprotected ports */
818 free_pkts(traffic->ip6.pkts, traffic->ip6.num);
820 traffic->ip6.num = 0;
822 if (app_sa_prm.enable == 0) {
824 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
825 traffic->ipsec.num, MAX_PKT_BURST);
827 for (i = 0; i < nb_pkts_in; i++) {
828 m = traffic->ipsec.pkts[i];
829 struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
830 if (ip->ip_v == IPVERSION) {
831 idx = traffic->ip4.num++;
832 traffic->ip4.pkts[idx] = m;
834 idx = traffic->ip6.num++;
835 traffic->ip6.pkts[idx] = m;
839 inbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.pkts,
840 traffic->ipsec.saptr, traffic->ipsec.num);
841 ipsec_process(ipsec_ctx, traffic);
846 process_pkts_outbound_nosp(struct ipsec_ctx *ipsec_ctx,
847 struct ipsec_traffic *traffic)
850 uint32_t nb_pkts_out, i, n;
853 /* Drop any IPsec traffic from protected ports */
854 free_pkts(traffic->ipsec.pkts, traffic->ipsec.num);
858 for (i = 0; i < traffic->ip4.num; i++) {
859 traffic->ipsec.pkts[n] = traffic->ip4.pkts[i];
860 traffic->ipsec.res[n++] = single_sa_idx;
863 for (i = 0; i < traffic->ip6.num; i++) {
864 traffic->ipsec.pkts[n] = traffic->ip6.pkts[i];
865 traffic->ipsec.res[n++] = single_sa_idx;
868 traffic->ip4.num = 0;
869 traffic->ip6.num = 0;
870 traffic->ipsec.num = n;
872 if (app_sa_prm.enable == 0) {
874 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
875 traffic->ipsec.res, traffic->ipsec.num,
878 /* They all sue the same SA (ip4 or ip6 tunnel) */
879 m = traffic->ipsec.pkts[0];
880 ip = rte_pktmbuf_mtod(m, struct ip *);
881 if (ip->ip_v == IPVERSION) {
882 traffic->ip4.num = nb_pkts_out;
883 for (i = 0; i < nb_pkts_out; i++)
884 traffic->ip4.pkts[i] = traffic->ipsec.pkts[i];
886 traffic->ip6.num = nb_pkts_out;
887 for (i = 0; i < nb_pkts_out; i++)
888 traffic->ip6.pkts[i] = traffic->ipsec.pkts[i];
891 outbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.res,
892 traffic->ipsec.saptr, traffic->ipsec.num);
893 ipsec_process(ipsec_ctx, traffic);
897 static inline int32_t
898 get_hop_for_offload_pkt(struct rte_mbuf *pkt, int is_ipv6)
900 struct ipsec_mbuf_metadata *priv;
903 priv = get_priv(pkt);
906 if (unlikely(sa == NULL)) {
907 RTE_LOG(ERR, IPSEC, "SA not saved in private data\n");
915 return (sa->portid | RTE_LPM_LOOKUP_SUCCESS);
926 route4_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
928 uint32_t hop[MAX_PKT_BURST * 2];
929 uint32_t dst_ip[MAX_PKT_BURST * 2];
932 uint16_t lpm_pkts = 0;
937 /* Need to do an LPM lookup for non-inline packets. Inline packets will
938 * have port ID in the SA
941 for (i = 0; i < nb_pkts; i++) {
942 if (!(pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD)) {
943 /* Security offload not enabled. So an LPM lookup is
944 * required to get the hop
946 offset = offsetof(struct ip, ip_dst);
947 dst_ip[lpm_pkts] = *rte_pktmbuf_mtod_offset(pkts[i],
949 dst_ip[lpm_pkts] = rte_be_to_cpu_32(dst_ip[lpm_pkts]);
954 rte_lpm_lookup_bulk((struct rte_lpm *)rt_ctx, dst_ip, hop, lpm_pkts);
958 for (i = 0; i < nb_pkts; i++) {
959 if (pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD) {
960 /* Read hop from the SA */
961 pkt_hop = get_hop_for_offload_pkt(pkts[i], 0);
963 /* Need to use hop returned by lookup */
964 pkt_hop = hop[lpm_pkts++];
967 if ((pkt_hop & RTE_LPM_LOOKUP_SUCCESS) == 0) {
968 free_pkts(&pkts[i], 1);
971 send_single_packet(pkts[i], pkt_hop & 0xff, IPPROTO_IP);
976 route6_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
978 int32_t hop[MAX_PKT_BURST * 2];
979 uint8_t dst_ip[MAX_PKT_BURST * 2][16];
983 uint16_t lpm_pkts = 0;
988 /* Need to do an LPM lookup for non-inline packets. Inline packets will
989 * have port ID in the SA
992 for (i = 0; i < nb_pkts; i++) {
993 if (!(pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD)) {
994 /* Security offload not enabled. So an LPM lookup is
995 * required to get the hop
997 offset = offsetof(struct ip6_hdr, ip6_dst);
998 ip6_dst = rte_pktmbuf_mtod_offset(pkts[i], uint8_t *,
1000 memcpy(&dst_ip[lpm_pkts][0], ip6_dst, 16);
1005 rte_lpm6_lookup_bulk_func((struct rte_lpm6 *)rt_ctx, dst_ip, hop,
1010 for (i = 0; i < nb_pkts; i++) {
1011 if (pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD) {
1012 /* Read hop from the SA */
1013 pkt_hop = get_hop_for_offload_pkt(pkts[i], 1);
1015 /* Need to use hop returned by lookup */
1016 pkt_hop = hop[lpm_pkts++];
1019 if (pkt_hop == -1) {
1020 free_pkts(&pkts[i], 1);
1023 send_single_packet(pkts[i], pkt_hop & 0xff, IPPROTO_IPV6);
1028 process_pkts(struct lcore_conf *qconf, struct rte_mbuf **pkts,
1029 uint8_t nb_pkts, uint16_t portid)
1031 struct ipsec_traffic traffic;
1033 prepare_traffic(pkts, &traffic, nb_pkts);
1035 if (unlikely(single_sa)) {
1036 if (is_unprotected_port(portid))
1037 process_pkts_inbound_nosp(&qconf->inbound, &traffic);
1039 process_pkts_outbound_nosp(&qconf->outbound, &traffic);
1041 if (is_unprotected_port(portid))
1042 process_pkts_inbound(&qconf->inbound, &traffic);
1044 process_pkts_outbound(&qconf->outbound, &traffic);
1047 route4_pkts(qconf->rt4_ctx, traffic.ip4.pkts, traffic.ip4.num);
1048 route6_pkts(qconf->rt6_ctx, traffic.ip6.pkts, traffic.ip6.num);
1052 drain_tx_buffers(struct lcore_conf *qconf)
1057 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
1058 buf = &qconf->tx_mbufs[portid];
1061 send_burst(qconf, buf->len, portid);
1067 drain_crypto_buffers(struct lcore_conf *qconf)
1070 struct ipsec_ctx *ctx;
1072 /* drain inbound buffers*/
1073 ctx = &qconf->inbound;
1074 for (i = 0; i != ctx->nb_qps; i++) {
1075 if (ctx->tbl[i].len != 0)
1076 enqueue_cop_burst(ctx->tbl + i);
1079 /* drain outbound buffers*/
1080 ctx = &qconf->outbound;
1081 for (i = 0; i != ctx->nb_qps; i++) {
1082 if (ctx->tbl[i].len != 0)
1083 enqueue_cop_burst(ctx->tbl + i);
1088 drain_inbound_crypto_queues(const struct lcore_conf *qconf,
1089 struct ipsec_ctx *ctx)
1092 struct ipsec_traffic trf;
1094 if (app_sa_prm.enable == 0) {
1096 /* dequeue packets from crypto-queue */
1097 n = ipsec_inbound_cqp_dequeue(ctx, trf.ipsec.pkts,
1098 RTE_DIM(trf.ipsec.pkts));
1103 /* split traffic by ipv4-ipv6 */
1104 split46_traffic(&trf, trf.ipsec.pkts, n);
1106 ipsec_cqp_process(ctx, &trf);
1108 /* process ipv4 packets */
1109 if (trf.ip4.num != 0) {
1110 inbound_sp_sa(ctx->sp4_ctx, ctx->sa_ctx, &trf.ip4, 0);
1111 route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
1114 /* process ipv6 packets */
1115 if (trf.ip6.num != 0) {
1116 inbound_sp_sa(ctx->sp6_ctx, ctx->sa_ctx, &trf.ip6, 0);
1117 route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
1122 drain_outbound_crypto_queues(const struct lcore_conf *qconf,
1123 struct ipsec_ctx *ctx)
1126 struct ipsec_traffic trf;
1128 if (app_sa_prm.enable == 0) {
1130 /* dequeue packets from crypto-queue */
1131 n = ipsec_outbound_cqp_dequeue(ctx, trf.ipsec.pkts,
1132 RTE_DIM(trf.ipsec.pkts));
1137 /* split traffic by ipv4-ipv6 */
1138 split46_traffic(&trf, trf.ipsec.pkts, n);
1140 ipsec_cqp_process(ctx, &trf);
1142 /* process ipv4 packets */
1143 if (trf.ip4.num != 0)
1144 route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
1146 /* process ipv6 packets */
1147 if (trf.ip6.num != 0)
1148 route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
1151 /* main processing loop */
1153 ipsec_poll_mode_worker(void)
1155 struct rte_mbuf *pkts[MAX_PKT_BURST];
1157 uint64_t prev_tsc, diff_tsc, cur_tsc;
1161 struct lcore_conf *qconf;
1162 int32_t rc, socket_id;
1163 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
1164 / US_PER_S * BURST_TX_DRAIN_US;
1165 struct lcore_rx_queue *rxql;
1168 lcore_id = rte_lcore_id();
1169 qconf = &lcore_conf[lcore_id];
1170 rxql = qconf->rx_queue_list;
1171 socket_id = rte_lcore_to_socket_id(lcore_id);
1173 qconf->rt4_ctx = socket_ctx[socket_id].rt_ip4;
1174 qconf->rt6_ctx = socket_ctx[socket_id].rt_ip6;
1175 qconf->inbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_in;
1176 qconf->inbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_in;
1177 qconf->inbound.sa_ctx = socket_ctx[socket_id].sa_in;
1178 qconf->inbound.cdev_map = cdev_map_in;
1179 qconf->inbound.session_pool = socket_ctx[socket_id].session_pool;
1180 qconf->inbound.session_priv_pool =
1181 socket_ctx[socket_id].session_priv_pool;
1182 qconf->outbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_out;
1183 qconf->outbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_out;
1184 qconf->outbound.sa_ctx = socket_ctx[socket_id].sa_out;
1185 qconf->outbound.cdev_map = cdev_map_out;
1186 qconf->outbound.session_pool = socket_ctx[socket_id].session_pool;
1187 qconf->outbound.session_priv_pool =
1188 socket_ctx[socket_id].session_priv_pool;
1189 qconf->frag.pool_dir = socket_ctx[socket_id].mbuf_pool;
1190 qconf->frag.pool_indir = socket_ctx[socket_id].mbuf_pool_indir;
1192 rc = ipsec_sad_lcore_cache_init(app_sa_prm.cache_sz);
1195 "SAD cache init on lcore %u, failed with code: %d\n",
1200 if (qconf->nb_rx_queue == 0) {
1201 RTE_LOG(DEBUG, IPSEC, "lcore %u has nothing to do\n",
1206 RTE_LOG(INFO, IPSEC, "entering main loop on lcore %u\n", lcore_id);
1208 for (i = 0; i < qconf->nb_rx_queue; i++) {
1209 portid = rxql[i].port_id;
1210 queueid = rxql[i].queue_id;
1211 RTE_LOG(INFO, IPSEC,
1212 " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
1213 lcore_id, portid, queueid);
1216 while (!force_quit) {
1217 cur_tsc = rte_rdtsc();
1219 /* TX queue buffer drain */
1220 diff_tsc = cur_tsc - prev_tsc;
1222 if (unlikely(diff_tsc > drain_tsc)) {
1223 drain_tx_buffers(qconf);
1224 drain_crypto_buffers(qconf);
1228 for (i = 0; i < qconf->nb_rx_queue; ++i) {
1230 /* Read packets from RX queues */
1231 portid = rxql[i].port_id;
1232 queueid = rxql[i].queue_id;
1233 nb_rx = rte_eth_rx_burst(portid, queueid,
1234 pkts, MAX_PKT_BURST);
1237 core_stats_update_rx(nb_rx);
1238 process_pkts(qconf, pkts, nb_rx, portid);
1241 /* dequeue and process completed crypto-ops */
1242 if (is_unprotected_port(portid))
1243 drain_inbound_crypto_queues(qconf,
1246 drain_outbound_crypto_queues(qconf,
1253 check_flow_params(uint16_t fdir_portid, uint8_t fdir_qid)
1259 for (i = 0; i < nb_lcore_params; ++i) {
1260 portid = lcore_params_array[i].port_id;
1261 if (portid == fdir_portid) {
1262 queueid = lcore_params_array[i].queue_id;
1263 if (queueid == fdir_qid)
1267 if (i == nb_lcore_params - 1)
1275 check_poll_mode_params(struct eh_conf *eh_conf)
1285 if (eh_conf->mode != EH_PKT_TRANSFER_MODE_POLL)
1288 if (lcore_params == NULL) {
1289 printf("Error: No port/queue/core mappings\n");
1293 for (i = 0; i < nb_lcore_params; ++i) {
1294 lcore = lcore_params[i].lcore_id;
1295 if (!rte_lcore_is_enabled(lcore)) {
1296 printf("error: lcore %hhu is not enabled in "
1297 "lcore mask\n", lcore);
1300 socket_id = rte_lcore_to_socket_id(lcore);
1301 if (socket_id != 0 && numa_on == 0) {
1302 printf("warning: lcore %hhu is on socket %d "
1306 portid = lcore_params[i].port_id;
1307 if ((enabled_port_mask & (1 << portid)) == 0) {
1308 printf("port %u is not enabled in port mask\n", portid);
1311 if (!rte_eth_dev_is_valid_port(portid)) {
1312 printf("port %u is not present on the board\n", portid);
1320 get_port_nb_rx_queues(const uint16_t port)
1325 for (i = 0; i < nb_lcore_params; ++i) {
1326 if (lcore_params[i].port_id == port &&
1327 lcore_params[i].queue_id > queue)
1328 queue = lcore_params[i].queue_id;
1330 return (uint8_t)(++queue);
1334 init_lcore_rx_queues(void)
1336 uint16_t i, nb_rx_queue;
1339 for (i = 0; i < nb_lcore_params; ++i) {
1340 lcore = lcore_params[i].lcore_id;
1341 nb_rx_queue = lcore_conf[lcore].nb_rx_queue;
1342 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1343 printf("error: too many queues (%u) for lcore: %u\n",
1344 nb_rx_queue + 1, lcore);
1347 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1348 lcore_params[i].port_id;
1349 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1350 lcore_params[i].queue_id;
1351 lcore_conf[lcore].nb_rx_queue++;
1358 print_usage(const char *prgname)
1360 fprintf(stderr, "%s [EAL options] --"
1366 " [-w REPLAY_WINDOW_SIZE]"
1370 " [-s NUMBER_OF_MBUFS_IN_PKT_POOL]"
1372 " --config (port,queue,lcore)[,(port,queue,lcore)]"
1373 " [--single-sa SAIDX]"
1374 " [--cryptodev_mask MASK]"
1375 " [--transfer-mode MODE]"
1376 " [--event-schedule-type TYPE]"
1377 " [--" CMD_LINE_OPT_RX_OFFLOAD " RX_OFFLOAD_MASK]"
1378 " [--" CMD_LINE_OPT_TX_OFFLOAD " TX_OFFLOAD_MASK]"
1379 " [--" CMD_LINE_OPT_REASSEMBLE " REASSEMBLE_TABLE_SIZE]"
1380 " [--" CMD_LINE_OPT_MTU " MTU]"
1382 " -p PORTMASK: Hexadecimal bitmask of ports to configure\n"
1383 " -P : Enable promiscuous mode\n"
1384 " -u PORTMASK: Hexadecimal bitmask of unprotected ports\n"
1385 " -j FRAMESIZE: Data buffer size, minimum (and default)\n"
1386 " value: RTE_MBUF_DEFAULT_BUF_SIZE\n"
1387 " -l enables code-path that uses librte_ipsec\n"
1388 " -w REPLAY_WINDOW_SIZE specifies IPsec SQN replay window\n"
1389 " size for each SA\n"
1391 " -a enables SA SQN atomic behaviour\n"
1392 " -c specifies inbound SAD cache size,\n"
1393 " zero value disables the cache (default value: 128)\n"
1394 " -s number of mbufs in packet pool, if not specified number\n"
1395 " of mbufs will be calculated based on number of cores,\n"
1396 " ports and crypto queues\n"
1397 " -f CONFIG_FILE: Configuration file\n"
1398 " --config (port,queue,lcore): Rx queue configuration. In poll\n"
1399 " mode determines which queues from\n"
1400 " which ports are mapped to which cores.\n"
1401 " In event mode this option is not used\n"
1402 " as packets are dynamically scheduled\n"
1403 " to cores by HW.\n"
1404 " --single-sa SAIDX: In poll mode use single SA index for\n"
1405 " outbound traffic, bypassing the SP\n"
1406 " In event mode selects driver submode,\n"
1407 " SA index value is ignored\n"
1408 " --cryptodev_mask MASK: Hexadecimal bitmask of the crypto\n"
1409 " devices to configure\n"
1410 " --transfer-mode MODE\n"
1411 " \"poll\" : Packet transfer via polling (default)\n"
1412 " \"event\" : Packet transfer via event device\n"
1413 " --event-schedule-type TYPE queue schedule type, used only when\n"
1414 " transfer mode is set to event\n"
1415 " \"ordered\" : Ordered (default)\n"
1416 " \"atomic\" : Atomic\n"
1417 " \"parallel\" : Parallel\n"
1418 " --" CMD_LINE_OPT_RX_OFFLOAD
1419 ": bitmask of the RX HW offload capabilities to enable/use\n"
1420 " (DEV_RX_OFFLOAD_*)\n"
1421 " --" CMD_LINE_OPT_TX_OFFLOAD
1422 ": bitmask of the TX HW offload capabilities to enable/use\n"
1423 " (DEV_TX_OFFLOAD_*)\n"
1424 " --" CMD_LINE_OPT_REASSEMBLE " NUM"
1425 ": max number of entries in reassemble(fragment) table\n"
1426 " (zero (default value) disables reassembly)\n"
1427 " --" CMD_LINE_OPT_MTU " MTU"
1428 ": MTU value on all ports (default value: 1500)\n"
1429 " outgoing packets with bigger size will be fragmented\n"
1430 " incoming packets with bigger size will be discarded\n"
1431 " --" CMD_LINE_OPT_FRAG_TTL " FRAG_TTL_NS"
1432 ": fragments lifetime in nanoseconds, default\n"
1433 " and maximum value is 10.000.000.000 ns (10 s)\n"
1439 parse_mask(const char *str, uint64_t *val)
1445 t = strtoul(str, &end, 0);
1446 if (errno != 0 || end[0] != 0)
1454 parse_portmask(const char *portmask)
1459 /* parse hexadecimal string */
1460 pm = strtoul(portmask, &end, 16);
1461 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1464 if ((pm == 0) && errno)
1471 parse_decimal(const char *str)
1476 num = strtoull(str, &end, 10);
1477 if ((str[0] == '\0') || (end == NULL) || (*end != '\0')
1485 parse_config(const char *q_arg)
1488 const char *p, *p0 = q_arg;
1496 unsigned long int_fld[_NUM_FLD];
1497 char *str_fld[_NUM_FLD];
1501 nb_lcore_params = 0;
1503 while ((p = strchr(p0, '(')) != NULL) {
1505 p0 = strchr(p, ')');
1510 if (size >= sizeof(s))
1513 snprintf(s, sizeof(s), "%.*s", size, p);
1514 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1517 for (i = 0; i < _NUM_FLD; i++) {
1519 int_fld[i] = strtoul(str_fld[i], &end, 0);
1520 if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
1523 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1524 printf("exceeded max number of lcore params: %hu\n",
1528 lcore_params_array[nb_lcore_params].port_id =
1529 (uint8_t)int_fld[FLD_PORT];
1530 lcore_params_array[nb_lcore_params].queue_id =
1531 (uint8_t)int_fld[FLD_QUEUE];
1532 lcore_params_array[nb_lcore_params].lcore_id =
1533 (uint8_t)int_fld[FLD_LCORE];
1536 lcore_params = lcore_params_array;
1541 print_app_sa_prm(const struct app_sa_prm *prm)
1543 printf("librte_ipsec usage: %s\n",
1544 (prm->enable == 0) ? "disabled" : "enabled");
1546 printf("replay window size: %u\n", prm->window_size);
1547 printf("ESN: %s\n", (prm->enable_esn == 0) ? "disabled" : "enabled");
1548 printf("SA flags: %#" PRIx64 "\n", prm->flags);
1549 printf("Frag TTL: %" PRIu64 " ns\n", frag_ttl_ns);
1553 parse_transfer_mode(struct eh_conf *conf, const char *optarg)
1555 if (!strcmp(CMD_LINE_ARG_POLL, optarg))
1556 conf->mode = EH_PKT_TRANSFER_MODE_POLL;
1557 else if (!strcmp(CMD_LINE_ARG_EVENT, optarg))
1558 conf->mode = EH_PKT_TRANSFER_MODE_EVENT;
1560 printf("Unsupported packet transfer mode\n");
1568 parse_schedule_type(struct eh_conf *conf, const char *optarg)
1570 struct eventmode_conf *em_conf = NULL;
1572 /* Get eventmode conf */
1573 em_conf = conf->mode_params;
1575 if (!strcmp(CMD_LINE_ARG_ORDERED, optarg))
1576 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_ORDERED;
1577 else if (!strcmp(CMD_LINE_ARG_ATOMIC, optarg))
1578 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_ATOMIC;
1579 else if (!strcmp(CMD_LINE_ARG_PARALLEL, optarg))
1580 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_PARALLEL;
1582 printf("Unsupported queue schedule type\n");
1590 parse_args(int32_t argc, char **argv, struct eh_conf *eh_conf)
1595 int32_t option_index;
1596 char *prgname = argv[0];
1597 int32_t f_present = 0;
1601 while ((opt = getopt_long(argc, argvopt, "aelp:Pu:f:j:w:c:s:",
1602 lgopts, &option_index)) != EOF) {
1606 enabled_port_mask = parse_portmask(optarg);
1607 if (enabled_port_mask == 0) {
1608 printf("invalid portmask\n");
1609 print_usage(prgname);
1614 printf("Promiscuous mode selected\n");
1618 unprotected_port_mask = parse_portmask(optarg);
1619 if (unprotected_port_mask == 0) {
1620 printf("invalid unprotected portmask\n");
1621 print_usage(prgname);
1626 if (f_present == 1) {
1627 printf("\"-f\" option present more than "
1629 print_usage(prgname);
1637 ret = parse_decimal(optarg);
1639 printf("Invalid number of buffers in a pool: "
1641 print_usage(prgname);
1645 nb_bufs_in_pool = ret;
1649 ret = parse_decimal(optarg);
1650 if (ret < RTE_MBUF_DEFAULT_BUF_SIZE ||
1652 printf("Invalid frame buffer size value: %s\n",
1654 print_usage(prgname);
1657 frame_buf_size = ret;
1658 printf("Custom frame buffer size %u\n", frame_buf_size);
1661 app_sa_prm.enable = 1;
1664 app_sa_prm.window_size = parse_decimal(optarg);
1667 app_sa_prm.enable_esn = 1;
1670 app_sa_prm.enable = 1;
1671 app_sa_prm.flags |= RTE_IPSEC_SAFLAG_SQN_ATOM;
1674 ret = parse_decimal(optarg);
1676 printf("Invalid SA cache size: %s\n", optarg);
1677 print_usage(prgname);
1680 app_sa_prm.cache_sz = ret;
1682 case CMD_LINE_OPT_CONFIG_NUM:
1683 ret = parse_config(optarg);
1685 printf("Invalid config\n");
1686 print_usage(prgname);
1690 case CMD_LINE_OPT_SINGLE_SA_NUM:
1691 ret = parse_decimal(optarg);
1692 if (ret == -1 || ret > UINT32_MAX) {
1693 printf("Invalid argument[sa_idx]\n");
1694 print_usage(prgname);
1700 single_sa_idx = ret;
1701 eh_conf->ipsec_mode = EH_IPSEC_MODE_TYPE_DRIVER;
1702 printf("Configured with single SA index %u\n",
1705 case CMD_LINE_OPT_CRYPTODEV_MASK_NUM:
1706 ret = parse_portmask(optarg);
1708 printf("Invalid argument[portmask]\n");
1709 print_usage(prgname);
1714 enabled_cryptodev_mask = ret;
1717 case CMD_LINE_OPT_TRANSFER_MODE_NUM:
1718 ret = parse_transfer_mode(eh_conf, optarg);
1720 printf("Invalid packet transfer mode\n");
1721 print_usage(prgname);
1726 case CMD_LINE_OPT_SCHEDULE_TYPE_NUM:
1727 ret = parse_schedule_type(eh_conf, optarg);
1729 printf("Invalid queue schedule type\n");
1730 print_usage(prgname);
1735 case CMD_LINE_OPT_RX_OFFLOAD_NUM:
1736 ret = parse_mask(optarg, &dev_rx_offload);
1738 printf("Invalid argument for \'%s\': %s\n",
1739 CMD_LINE_OPT_RX_OFFLOAD, optarg);
1740 print_usage(prgname);
1744 case CMD_LINE_OPT_TX_OFFLOAD_NUM:
1745 ret = parse_mask(optarg, &dev_tx_offload);
1747 printf("Invalid argument for \'%s\': %s\n",
1748 CMD_LINE_OPT_TX_OFFLOAD, optarg);
1749 print_usage(prgname);
1753 case CMD_LINE_OPT_REASSEMBLE_NUM:
1754 ret = parse_decimal(optarg);
1755 if (ret < 0 || ret > UINT32_MAX) {
1756 printf("Invalid argument for \'%s\': %s\n",
1757 CMD_LINE_OPT_REASSEMBLE, optarg);
1758 print_usage(prgname);
1763 case CMD_LINE_OPT_MTU_NUM:
1764 ret = parse_decimal(optarg);
1765 if (ret < 0 || ret > RTE_IPV4_MAX_PKT_LEN) {
1766 printf("Invalid argument for \'%s\': %s\n",
1767 CMD_LINE_OPT_MTU, optarg);
1768 print_usage(prgname);
1773 case CMD_LINE_OPT_FRAG_TTL_NUM:
1774 ret = parse_decimal(optarg);
1775 if (ret < 0 || ret > MAX_FRAG_TTL_NS) {
1776 printf("Invalid argument for \'%s\': %s\n",
1777 CMD_LINE_OPT_MTU, optarg);
1778 print_usage(prgname);
1784 print_usage(prgname);
1789 if (f_present == 0) {
1790 printf("Mandatory option \"-f\" not present\n");
1794 /* check do we need to enable multi-seg support */
1795 if (multi_seg_required()) {
1796 /* legacy mode doesn't support multi-seg */
1797 app_sa_prm.enable = 1;
1798 printf("frame buf size: %u, mtu: %u, "
1799 "number of reassemble entries: %u\n"
1800 "multi-segment support is required\n",
1801 frame_buf_size, mtu_size, frag_tbl_sz);
1804 print_app_sa_prm(&app_sa_prm);
1807 argv[optind-1] = prgname;
1810 optind = 1; /* reset getopt lib */
1815 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
1817 char buf[RTE_ETHER_ADDR_FMT_SIZE];
1818 rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
1819 printf("%s%s", name, buf);
1823 * Update destination ethaddr for the port.
1826 add_dst_ethaddr(uint16_t port, const struct rte_ether_addr *addr)
1828 if (port >= RTE_DIM(ethaddr_tbl))
1831 ethaddr_tbl[port].dst = ETHADDR_TO_UINT64(addr);
1835 /* Check the link status of all ports in up to 9s, and print them finally */
1837 check_all_ports_link_status(uint32_t port_mask)
1839 #define CHECK_INTERVAL 100 /* 100ms */
1840 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1842 uint8_t count, all_ports_up, print_flag = 0;
1843 struct rte_eth_link link;
1846 printf("\nChecking link status");
1848 for (count = 0; count <= MAX_CHECK_TIME; count++) {
1850 RTE_ETH_FOREACH_DEV(portid) {
1851 if ((port_mask & (1 << portid)) == 0)
1853 memset(&link, 0, sizeof(link));
1854 ret = rte_eth_link_get_nowait(portid, &link);
1857 if (print_flag == 1)
1858 printf("Port %u link get failed: %s\n",
1859 portid, rte_strerror(-ret));
1862 /* print link status if flag set */
1863 if (print_flag == 1) {
1864 if (link.link_status)
1866 "Port%d Link Up - speed %u Mbps -%s\n",
1867 portid, link.link_speed,
1868 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1869 ("full-duplex") : ("half-duplex"));
1871 printf("Port %d Link Down\n", portid);
1874 /* clear all_ports_up flag if any link down */
1875 if (link.link_status == ETH_LINK_DOWN) {
1880 /* after finally printing all link status, get out */
1881 if (print_flag == 1)
1884 if (all_ports_up == 0) {
1887 rte_delay_ms(CHECK_INTERVAL);
1890 /* set the print_flag if all ports up or timeout */
1891 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1899 add_mapping(struct rte_hash *map, const char *str, uint16_t cdev_id,
1900 uint16_t qp, struct lcore_params *params,
1901 struct ipsec_ctx *ipsec_ctx,
1902 const struct rte_cryptodev_capabilities *cipher,
1903 const struct rte_cryptodev_capabilities *auth,
1904 const struct rte_cryptodev_capabilities *aead)
1908 struct cdev_key key = { 0 };
1910 key.lcore_id = params->lcore_id;
1912 key.cipher_algo = cipher->sym.cipher.algo;
1914 key.auth_algo = auth->sym.auth.algo;
1916 key.aead_algo = aead->sym.aead.algo;
1918 ret = rte_hash_lookup(map, &key);
1922 for (i = 0; i < ipsec_ctx->nb_qps; i++)
1923 if (ipsec_ctx->tbl[i].id == cdev_id)
1926 if (i == ipsec_ctx->nb_qps) {
1927 if (ipsec_ctx->nb_qps == MAX_QP_PER_LCORE) {
1928 printf("Maximum number of crypto devices assigned to "
1929 "a core, increase MAX_QP_PER_LCORE value\n");
1932 ipsec_ctx->tbl[i].id = cdev_id;
1933 ipsec_ctx->tbl[i].qp = qp;
1934 ipsec_ctx->nb_qps++;
1935 printf("%s cdev mapping: lcore %u using cdev %u qp %u "
1936 "(cdev_id_qp %lu)\n", str, key.lcore_id,
1940 ret = rte_hash_add_key_data(map, &key, (void *)i);
1942 printf("Faled to insert cdev mapping for (lcore %u, "
1943 "cdev %u, qp %u), errno %d\n",
1944 key.lcore_id, ipsec_ctx->tbl[i].id,
1945 ipsec_ctx->tbl[i].qp, ret);
1953 add_cdev_mapping(struct rte_cryptodev_info *dev_info, uint16_t cdev_id,
1954 uint16_t qp, struct lcore_params *params)
1957 const struct rte_cryptodev_capabilities *i, *j;
1958 struct rte_hash *map;
1959 struct lcore_conf *qconf;
1960 struct ipsec_ctx *ipsec_ctx;
1963 qconf = &lcore_conf[params->lcore_id];
1965 if ((unprotected_port_mask & (1 << params->port_id)) == 0) {
1967 ipsec_ctx = &qconf->outbound;
1971 ipsec_ctx = &qconf->inbound;
1975 /* Required cryptodevs with operation chainning */
1976 if (!(dev_info->feature_flags &
1977 RTE_CRYPTODEV_FF_SYM_OPERATION_CHAINING))
1980 for (i = dev_info->capabilities;
1981 i->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; i++) {
1982 if (i->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1985 if (i->sym.xform_type == RTE_CRYPTO_SYM_XFORM_AEAD) {
1986 ret |= add_mapping(map, str, cdev_id, qp, params,
1987 ipsec_ctx, NULL, NULL, i);
1991 if (i->sym.xform_type != RTE_CRYPTO_SYM_XFORM_CIPHER)
1994 for (j = dev_info->capabilities;
1995 j->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; j++) {
1996 if (j->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1999 if (j->sym.xform_type != RTE_CRYPTO_SYM_XFORM_AUTH)
2002 ret |= add_mapping(map, str, cdev_id, qp, params,
2003 ipsec_ctx, i, j, NULL);
2010 /* Check if the device is enabled by cryptodev_mask */
2012 check_cryptodev_mask(uint8_t cdev_id)
2014 if (enabled_cryptodev_mask & (1 << cdev_id))
2021 cryptodevs_init(uint16_t req_queue_num)
2023 struct rte_cryptodev_config dev_conf;
2024 struct rte_cryptodev_qp_conf qp_conf;
2025 uint16_t idx, max_nb_qps, qp, total_nb_qps, i;
2027 struct rte_hash_parameters params = { 0 };
2029 const uint64_t mseg_flag = multi_seg_required() ?
2030 RTE_CRYPTODEV_FF_IN_PLACE_SGL : 0;
2032 params.entries = CDEV_MAP_ENTRIES;
2033 params.key_len = sizeof(struct cdev_key);
2034 params.hash_func = rte_jhash;
2035 params.hash_func_init_val = 0;
2036 params.socket_id = rte_socket_id();
2038 params.name = "cdev_map_in";
2039 cdev_map_in = rte_hash_create(¶ms);
2040 if (cdev_map_in == NULL)
2041 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
2044 params.name = "cdev_map_out";
2045 cdev_map_out = rte_hash_create(¶ms);
2046 if (cdev_map_out == NULL)
2047 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
2050 printf("lcore/cryptodev/qp mappings:\n");
2054 for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
2055 struct rte_cryptodev_info cdev_info;
2057 if (check_cryptodev_mask((uint8_t)cdev_id))
2060 rte_cryptodev_info_get(cdev_id, &cdev_info);
2062 if ((mseg_flag & cdev_info.feature_flags) != mseg_flag)
2063 rte_exit(EXIT_FAILURE,
2064 "Device %hd does not support \'%s\' feature\n",
2066 rte_cryptodev_get_feature_name(mseg_flag));
2068 if (nb_lcore_params > cdev_info.max_nb_queue_pairs)
2069 max_nb_qps = cdev_info.max_nb_queue_pairs;
2071 max_nb_qps = nb_lcore_params;
2075 while (qp < max_nb_qps && i < nb_lcore_params) {
2076 if (add_cdev_mapping(&cdev_info, cdev_id, qp,
2077 &lcore_params[idx]))
2080 idx = idx % nb_lcore_params;
2084 qp = RTE_MIN(max_nb_qps, RTE_MAX(req_queue_num, qp));
2089 dev_conf.socket_id = rte_cryptodev_socket_id(cdev_id);
2090 dev_conf.nb_queue_pairs = qp;
2091 dev_conf.ff_disable = RTE_CRYPTODEV_FF_ASYMMETRIC_CRYPTO;
2093 uint32_t dev_max_sess = cdev_info.sym.max_nb_sessions;
2094 if (dev_max_sess != 0 &&
2095 dev_max_sess < get_nb_crypto_sessions())
2096 rte_exit(EXIT_FAILURE,
2097 "Device does not support at least %u "
2098 "sessions", get_nb_crypto_sessions());
2100 if (rte_cryptodev_configure(cdev_id, &dev_conf))
2101 rte_panic("Failed to initialize cryptodev %u\n",
2104 qp_conf.nb_descriptors = CDEV_QUEUE_DESC;
2105 qp_conf.mp_session =
2106 socket_ctx[dev_conf.socket_id].session_pool;
2107 qp_conf.mp_session_private =
2108 socket_ctx[dev_conf.socket_id].session_priv_pool;
2109 for (qp = 0; qp < dev_conf.nb_queue_pairs; qp++)
2110 if (rte_cryptodev_queue_pair_setup(cdev_id, qp,
2111 &qp_conf, dev_conf.socket_id))
2112 rte_panic("Failed to setup queue %u for "
2113 "cdev_id %u\n", 0, cdev_id);
2115 if (rte_cryptodev_start(cdev_id))
2116 rte_panic("Failed to start cryptodev %u\n",
2122 return total_nb_qps;
2126 port_init(uint16_t portid, uint64_t req_rx_offloads, uint64_t req_tx_offloads)
2128 uint32_t frame_size;
2129 struct rte_eth_dev_info dev_info;
2130 struct rte_eth_txconf *txconf;
2131 uint16_t nb_tx_queue, nb_rx_queue;
2132 uint16_t tx_queueid, rx_queueid, queue, lcore_id;
2133 int32_t ret, socket_id;
2134 struct lcore_conf *qconf;
2135 struct rte_ether_addr ethaddr;
2136 struct rte_eth_conf local_port_conf = port_conf;
2138 ret = rte_eth_dev_info_get(portid, &dev_info);
2140 rte_exit(EXIT_FAILURE,
2141 "Error during getting device (port %u) info: %s\n",
2142 portid, strerror(-ret));
2144 /* limit allowed HW offloafs, as user requested */
2145 dev_info.rx_offload_capa &= dev_rx_offload;
2146 dev_info.tx_offload_capa &= dev_tx_offload;
2148 printf("Configuring device port %u:\n", portid);
2150 ret = rte_eth_macaddr_get(portid, ðaddr);
2152 rte_exit(EXIT_FAILURE,
2153 "Error getting MAC address (port %u): %s\n",
2154 portid, rte_strerror(-ret));
2156 ethaddr_tbl[portid].src = ETHADDR_TO_UINT64(ðaddr);
2157 print_ethaddr("Address: ", ðaddr);
2160 nb_rx_queue = get_port_nb_rx_queues(portid);
2161 nb_tx_queue = nb_lcores;
2163 if (nb_rx_queue > dev_info.max_rx_queues)
2164 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
2165 "(max rx queue is %u)\n",
2166 nb_rx_queue, dev_info.max_rx_queues);
2168 if (nb_tx_queue > dev_info.max_tx_queues)
2169 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
2170 "(max tx queue is %u)\n",
2171 nb_tx_queue, dev_info.max_tx_queues);
2173 printf("Creating queues: nb_rx_queue=%d nb_tx_queue=%u...\n",
2174 nb_rx_queue, nb_tx_queue);
2176 frame_size = MTU_TO_FRAMELEN(mtu_size);
2177 if (frame_size > local_port_conf.rxmode.max_rx_pkt_len)
2178 local_port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2179 local_port_conf.rxmode.max_rx_pkt_len = frame_size;
2181 if (multi_seg_required()) {
2182 local_port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_SCATTER;
2183 local_port_conf.txmode.offloads |= DEV_TX_OFFLOAD_MULTI_SEGS;
2186 local_port_conf.rxmode.offloads |= req_rx_offloads;
2187 local_port_conf.txmode.offloads |= req_tx_offloads;
2189 /* Check that all required capabilities are supported */
2190 if ((local_port_conf.rxmode.offloads & dev_info.rx_offload_capa) !=
2191 local_port_conf.rxmode.offloads)
2192 rte_exit(EXIT_FAILURE,
2193 "Error: port %u required RX offloads: 0x%" PRIx64
2194 ", avaialbe RX offloads: 0x%" PRIx64 "\n",
2195 portid, local_port_conf.rxmode.offloads,
2196 dev_info.rx_offload_capa);
2198 if ((local_port_conf.txmode.offloads & dev_info.tx_offload_capa) !=
2199 local_port_conf.txmode.offloads)
2200 rte_exit(EXIT_FAILURE,
2201 "Error: port %u required TX offloads: 0x%" PRIx64
2202 ", avaialbe TX offloads: 0x%" PRIx64 "\n",
2203 portid, local_port_conf.txmode.offloads,
2204 dev_info.tx_offload_capa);
2206 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
2207 local_port_conf.txmode.offloads |=
2208 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
2210 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_IPV4_CKSUM)
2211 local_port_conf.txmode.offloads |= DEV_TX_OFFLOAD_IPV4_CKSUM;
2213 printf("port %u configurng rx_offloads=0x%" PRIx64
2214 ", tx_offloads=0x%" PRIx64 "\n",
2215 portid, local_port_conf.rxmode.offloads,
2216 local_port_conf.txmode.offloads);
2218 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
2219 dev_info.flow_type_rss_offloads;
2220 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
2221 port_conf.rx_adv_conf.rss_conf.rss_hf) {
2222 printf("Port %u modified RSS hash function based on hardware support,"
2223 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
2225 port_conf.rx_adv_conf.rss_conf.rss_hf,
2226 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
2229 ret = rte_eth_dev_configure(portid, nb_rx_queue, nb_tx_queue,
2232 rte_exit(EXIT_FAILURE, "Cannot configure device: "
2233 "err=%d, port=%d\n", ret, portid);
2235 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd, &nb_txd);
2237 rte_exit(EXIT_FAILURE, "Cannot adjust number of descriptors: "
2238 "err=%d, port=%d\n", ret, portid);
2240 /* init one TX queue per lcore */
2242 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2243 if (rte_lcore_is_enabled(lcore_id) == 0)
2247 socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2252 printf("Setup txq=%u,%d,%d\n", lcore_id, tx_queueid, socket_id);
2254 txconf = &dev_info.default_txconf;
2255 txconf->offloads = local_port_conf.txmode.offloads;
2257 ret = rte_eth_tx_queue_setup(portid, tx_queueid, nb_txd,
2260 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
2261 "err=%d, port=%d\n", ret, portid);
2263 qconf = &lcore_conf[lcore_id];
2264 qconf->tx_queue_id[portid] = tx_queueid;
2266 /* Pre-populate pkt offloads based on capabilities */
2267 qconf->outbound.ipv4_offloads = PKT_TX_IPV4;
2268 qconf->outbound.ipv6_offloads = PKT_TX_IPV6;
2269 if (local_port_conf.txmode.offloads & DEV_TX_OFFLOAD_IPV4_CKSUM)
2270 qconf->outbound.ipv4_offloads |= PKT_TX_IP_CKSUM;
2274 /* init RX queues */
2275 for (queue = 0; queue < qconf->nb_rx_queue; ++queue) {
2276 struct rte_eth_rxconf rxq_conf;
2278 if (portid != qconf->rx_queue_list[queue].port_id)
2281 rx_queueid = qconf->rx_queue_list[queue].queue_id;
2283 printf("Setup rxq=%d,%d,%d\n", portid, rx_queueid,
2286 rxq_conf = dev_info.default_rxconf;
2287 rxq_conf.offloads = local_port_conf.rxmode.offloads;
2288 ret = rte_eth_rx_queue_setup(portid, rx_queueid,
2289 nb_rxd, socket_id, &rxq_conf,
2290 socket_ctx[socket_id].mbuf_pool);
2292 rte_exit(EXIT_FAILURE,
2293 "rte_eth_rx_queue_setup: err=%d, "
2294 "port=%d\n", ret, portid);
2301 max_session_size(void)
2305 int16_t cdev_id, port_id, n;
2308 n = rte_cryptodev_count();
2309 for (cdev_id = 0; cdev_id != n; cdev_id++) {
2310 sz = rte_cryptodev_sym_get_private_session_size(cdev_id);
2314 * If crypto device is security capable, need to check the
2315 * size of security session as well.
2318 /* Get security context of the crypto device */
2319 sec_ctx = rte_cryptodev_get_sec_ctx(cdev_id);
2320 if (sec_ctx == NULL)
2323 /* Get size of security session */
2324 sz = rte_security_session_get_size(sec_ctx);
2329 RTE_ETH_FOREACH_DEV(port_id) {
2330 if ((enabled_port_mask & (1 << port_id)) == 0)
2333 sec_ctx = rte_eth_dev_get_sec_ctx(port_id);
2334 if (sec_ctx == NULL)
2337 sz = rte_security_session_get_size(sec_ctx);
2346 session_pool_init(struct socket_ctx *ctx, int32_t socket_id, size_t sess_sz)
2348 char mp_name[RTE_MEMPOOL_NAMESIZE];
2349 struct rte_mempool *sess_mp;
2352 snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
2353 "sess_mp_%u", socket_id);
2355 * Doubled due to rte_security_session_create() uses one mempool for
2356 * session and for session private data.
2358 nb_sess = (get_nb_crypto_sessions() + CDEV_MP_CACHE_SZ *
2359 rte_lcore_count()) * 2;
2360 sess_mp = rte_cryptodev_sym_session_pool_create(
2361 mp_name, nb_sess, sess_sz, CDEV_MP_CACHE_SZ, 0,
2363 ctx->session_pool = sess_mp;
2365 if (ctx->session_pool == NULL)
2366 rte_exit(EXIT_FAILURE,
2367 "Cannot init session pool on socket %d\n", socket_id);
2369 printf("Allocated session pool on socket %d\n", socket_id);
2373 session_priv_pool_init(struct socket_ctx *ctx, int32_t socket_id,
2376 char mp_name[RTE_MEMPOOL_NAMESIZE];
2377 struct rte_mempool *sess_mp;
2380 snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
2381 "sess_mp_priv_%u", socket_id);
2383 * Doubled due to rte_security_session_create() uses one mempool for
2384 * session and for session private data.
2386 nb_sess = (get_nb_crypto_sessions() + CDEV_MP_CACHE_SZ *
2387 rte_lcore_count()) * 2;
2388 sess_mp = rte_mempool_create(mp_name,
2392 0, NULL, NULL, NULL,
2395 ctx->session_priv_pool = sess_mp;
2397 if (ctx->session_priv_pool == NULL)
2398 rte_exit(EXIT_FAILURE,
2399 "Cannot init session priv pool on socket %d\n",
2402 printf("Allocated session priv pool on socket %d\n",
2407 pool_init(struct socket_ctx *ctx, int32_t socket_id, uint32_t nb_mbuf)
2412 snprintf(s, sizeof(s), "mbuf_pool_%d", socket_id);
2413 ctx->mbuf_pool = rte_pktmbuf_pool_create(s, nb_mbuf,
2414 MEMPOOL_CACHE_SIZE, ipsec_metadata_size(),
2415 frame_buf_size, socket_id);
2418 * if multi-segment support is enabled, then create a pool
2419 * for indirect mbufs.
2421 ms = multi_seg_required();
2423 snprintf(s, sizeof(s), "mbuf_pool_indir_%d", socket_id);
2424 ctx->mbuf_pool_indir = rte_pktmbuf_pool_create(s, nb_mbuf,
2425 MEMPOOL_CACHE_SIZE, 0, 0, socket_id);
2428 if (ctx->mbuf_pool == NULL || (ms != 0 && ctx->mbuf_pool_indir == NULL))
2429 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool on socket %d\n",
2432 printf("Allocated mbuf pool on socket %d\n", socket_id);
2436 inline_ipsec_event_esn_overflow(struct rte_security_ctx *ctx, uint64_t md)
2438 struct ipsec_sa *sa;
2440 /* For inline protocol processing, the metadata in the event will
2441 * uniquely identify the security session which raised the event.
2442 * Application would then need the userdata it had registered with the
2443 * security session to process the event.
2446 sa = (struct ipsec_sa *)rte_security_get_userdata(ctx, md);
2449 /* userdata could not be retrieved */
2453 /* Sequence number over flow. SA need to be re-established */
2459 inline_ipsec_event_callback(uint16_t port_id, enum rte_eth_event_type type,
2460 void *param, void *ret_param)
2463 struct rte_eth_event_ipsec_desc *event_desc = NULL;
2464 struct rte_security_ctx *ctx = (struct rte_security_ctx *)
2465 rte_eth_dev_get_sec_ctx(port_id);
2467 RTE_SET_USED(param);
2469 if (type != RTE_ETH_EVENT_IPSEC)
2472 event_desc = ret_param;
2473 if (event_desc == NULL) {
2474 printf("Event descriptor not set\n");
2478 md = event_desc->metadata;
2480 if (event_desc->subtype == RTE_ETH_EVENT_IPSEC_ESN_OVERFLOW)
2481 return inline_ipsec_event_esn_overflow(ctx, md);
2482 else if (event_desc->subtype >= RTE_ETH_EVENT_IPSEC_MAX) {
2483 printf("Invalid IPsec event reported\n");
2491 rx_callback(__rte_unused uint16_t port, __rte_unused uint16_t queue,
2492 struct rte_mbuf *pkt[], uint16_t nb_pkts,
2493 __rte_unused uint16_t max_pkts, void *user_param)
2497 struct lcore_conf *lc;
2498 struct rte_mbuf *mb;
2499 struct rte_ether_hdr *eth;
2505 for (i = 0; i != nb_pkts; i++) {
2508 eth = rte_pktmbuf_mtod(mb, struct rte_ether_hdr *);
2509 if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
2511 struct rte_ipv4_hdr *iph;
2513 iph = (struct rte_ipv4_hdr *)(eth + 1);
2514 if (rte_ipv4_frag_pkt_is_fragmented(iph)) {
2516 mb->l2_len = sizeof(*eth);
2517 mb->l3_len = sizeof(*iph);
2518 tm = (tm != 0) ? tm : rte_rdtsc();
2519 mb = rte_ipv4_frag_reassemble_packet(
2520 lc->frag.tbl, &lc->frag.dr,
2524 /* fix ip cksum after reassemble. */
2525 iph = rte_pktmbuf_mtod_offset(mb,
2526 struct rte_ipv4_hdr *,
2528 iph->hdr_checksum = 0;
2529 iph->hdr_checksum = rte_ipv4_cksum(iph);
2532 } else if (eth->ether_type ==
2533 rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6)) {
2535 struct rte_ipv6_hdr *iph;
2536 struct ipv6_extension_fragment *fh;
2538 iph = (struct rte_ipv6_hdr *)(eth + 1);
2539 fh = rte_ipv6_frag_get_ipv6_fragment_header(iph);
2541 mb->l2_len = sizeof(*eth);
2542 mb->l3_len = (uintptr_t)fh - (uintptr_t)iph +
2544 tm = (tm != 0) ? tm : rte_rdtsc();
2545 mb = rte_ipv6_frag_reassemble_packet(
2546 lc->frag.tbl, &lc->frag.dr,
2549 /* fix l3_len after reassemble. */
2550 mb->l3_len = mb->l3_len - sizeof(*fh);
2558 /* some fragments were encountered, drain death row */
2560 rte_ip_frag_free_death_row(&lc->frag.dr, 0);
2567 reassemble_lcore_init(struct lcore_conf *lc, uint32_t cid)
2571 uint64_t frag_cycles;
2572 const struct lcore_rx_queue *rxq;
2573 const struct rte_eth_rxtx_callback *cb;
2575 /* create fragment table */
2576 sid = rte_lcore_to_socket_id(cid);
2577 frag_cycles = (rte_get_tsc_hz() + NS_PER_S - 1) /
2578 NS_PER_S * frag_ttl_ns;
2580 lc->frag.tbl = rte_ip_frag_table_create(frag_tbl_sz,
2581 FRAG_TBL_BUCKET_ENTRIES, frag_tbl_sz, frag_cycles, sid);
2582 if (lc->frag.tbl == NULL) {
2583 printf("%s(%u): failed to create fragment table of size: %u, "
2585 __func__, cid, frag_tbl_sz, rte_errno);
2589 /* setup reassemble RX callbacks for all queues */
2590 for (i = 0; i != lc->nb_rx_queue; i++) {
2592 rxq = lc->rx_queue_list + i;
2593 cb = rte_eth_add_rx_callback(rxq->port_id, rxq->queue_id,
2596 printf("%s(%u): failed to install RX callback for "
2597 "portid=%u, queueid=%u, error code: %d\n",
2599 rxq->port_id, rxq->queue_id, rte_errno);
2608 reassemble_init(void)
2614 for (i = 0; i != nb_lcore_params; i++) {
2615 lc = lcore_params[i].lcore_id;
2616 rc = reassemble_lcore_init(lcore_conf + lc, lc);
2625 create_default_ipsec_flow(uint16_t port_id, uint64_t rx_offloads)
2627 struct rte_flow_action action[2];
2628 struct rte_flow_item pattern[2];
2629 struct rte_flow_attr attr = {0};
2630 struct rte_flow_error err;
2631 struct rte_flow *flow;
2634 if (!(rx_offloads & DEV_RX_OFFLOAD_SECURITY))
2637 /* Add the default rte_flow to enable SECURITY for all ESP packets */
2639 pattern[0].type = RTE_FLOW_ITEM_TYPE_ESP;
2640 pattern[0].spec = NULL;
2641 pattern[0].mask = NULL;
2642 pattern[0].last = NULL;
2643 pattern[1].type = RTE_FLOW_ITEM_TYPE_END;
2645 action[0].type = RTE_FLOW_ACTION_TYPE_SECURITY;
2646 action[0].conf = NULL;
2647 action[1].type = RTE_FLOW_ACTION_TYPE_END;
2648 action[1].conf = NULL;
2652 ret = rte_flow_validate(port_id, &attr, pattern, action, &err);
2656 flow = rte_flow_create(port_id, &attr, pattern, action, &err);
2660 flow_info_tbl[port_id].rx_def_flow = flow;
2661 RTE_LOG(INFO, IPSEC,
2662 "Created default flow enabling SECURITY for all ESP traffic on port %d\n",
2667 signal_handler(int signum)
2669 if (signum == SIGINT || signum == SIGTERM) {
2670 printf("\n\nSignal %d received, preparing to exit...\n",
2677 ev_mode_sess_verify(struct ipsec_sa *sa, int nb_sa)
2679 struct rte_ipsec_session *ips;
2685 for (i = 0; i < nb_sa; i++) {
2686 ips = ipsec_get_primary_session(&sa[i]);
2687 if (ips->type != RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL)
2688 rte_exit(EXIT_FAILURE, "Event mode supports only "
2689 "inline protocol sessions\n");
2695 check_event_mode_params(struct eh_conf *eh_conf)
2697 struct eventmode_conf *em_conf = NULL;
2698 struct lcore_params *params;
2701 if (!eh_conf || !eh_conf->mode_params)
2704 /* Get eventmode conf */
2705 em_conf = eh_conf->mode_params;
2707 if (eh_conf->mode == EH_PKT_TRANSFER_MODE_POLL &&
2708 em_conf->ext_params.sched_type != SCHED_TYPE_NOT_SET) {
2709 printf("error: option --event-schedule-type applies only to "
2714 if (eh_conf->mode != EH_PKT_TRANSFER_MODE_EVENT)
2717 /* Set schedule type to ORDERED if it wasn't explicitly set by user */
2718 if (em_conf->ext_params.sched_type == SCHED_TYPE_NOT_SET)
2719 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_ORDERED;
2722 * Event mode currently supports only inline protocol sessions.
2723 * If there are other types of sessions configured then exit with
2726 ev_mode_sess_verify(sa_in, nb_sa_in);
2727 ev_mode_sess_verify(sa_out, nb_sa_out);
2730 /* Option --config does not apply to event mode */
2731 if (nb_lcore_params > 0) {
2732 printf("error: option --config applies only to poll mode\n");
2737 * In order to use the same port_init routine for both poll and event
2738 * modes initialize lcore_params with one queue for each eth port
2740 lcore_params = lcore_params_array;
2741 RTE_ETH_FOREACH_DEV(portid) {
2742 if ((enabled_port_mask & (1 << portid)) == 0)
2745 params = &lcore_params[nb_lcore_params++];
2746 params->port_id = portid;
2747 params->queue_id = 0;
2748 params->lcore_id = rte_get_next_lcore(0, 0, 1);
2755 inline_sessions_free(struct sa_ctx *sa_ctx)
2757 struct rte_ipsec_session *ips;
2758 struct ipsec_sa *sa;
2765 for (i = 0; i < sa_ctx->nb_sa; i++) {
2767 sa = &sa_ctx->sa[i];
2771 ips = ipsec_get_primary_session(sa);
2772 if (ips->type != RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL &&
2773 ips->type != RTE_SECURITY_ACTION_TYPE_INLINE_CRYPTO)
2776 if (!rte_eth_dev_is_valid_port(sa->portid))
2779 ret = rte_security_session_destroy(
2780 rte_eth_dev_get_sec_ctx(sa->portid),
2783 RTE_LOG(ERR, IPSEC, "Failed to destroy security "
2784 "session type %d, spi %d\n",
2785 ips->type, sa->spi);
2790 calculate_nb_mbufs(uint16_t nb_ports, uint16_t nb_crypto_qp, uint32_t nb_rxq,
2793 return RTE_MAX((nb_rxq * nb_rxd +
2794 nb_ports * nb_lcores * MAX_PKT_BURST +
2795 nb_ports * nb_txq * nb_txd +
2796 nb_lcores * MEMPOOL_CACHE_SIZE +
2797 nb_crypto_qp * CDEV_QUEUE_DESC +
2798 nb_lcores * frag_tbl_sz *
2799 FRAG_TBL_BUCKET_ENTRIES),
2804 main(int32_t argc, char **argv)
2807 uint32_t lcore_id, nb_txq, nb_rxq = 0;
2811 uint16_t portid, nb_crypto_qp, nb_ports = 0;
2812 uint64_t req_rx_offloads[RTE_MAX_ETHPORTS];
2813 uint64_t req_tx_offloads[RTE_MAX_ETHPORTS];
2814 struct eh_conf *eh_conf = NULL;
2817 nb_bufs_in_pool = 0;
2820 ret = rte_eal_init(argc, argv);
2822 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
2827 signal(SIGINT, signal_handler);
2828 signal(SIGTERM, signal_handler);
2830 /* initialize event helper configuration */
2831 eh_conf = eh_conf_init();
2832 if (eh_conf == NULL)
2833 rte_exit(EXIT_FAILURE, "Failed to init event helper config");
2835 /* parse application arguments (after the EAL ones) */
2836 ret = parse_args(argc, argv, eh_conf);
2838 rte_exit(EXIT_FAILURE, "Invalid parameters\n");
2840 /* parse configuration file */
2841 if (parse_cfg_file(cfgfile) < 0) {
2842 printf("parsing file \"%s\" failed\n",
2844 print_usage(argv[0]);
2848 if ((unprotected_port_mask & enabled_port_mask) !=
2849 unprotected_port_mask)
2850 rte_exit(EXIT_FAILURE, "Invalid unprotected portmask 0x%x\n",
2851 unprotected_port_mask);
2853 if (check_poll_mode_params(eh_conf) < 0)
2854 rte_exit(EXIT_FAILURE, "check_poll_mode_params failed\n");
2856 if (check_event_mode_params(eh_conf) < 0)
2857 rte_exit(EXIT_FAILURE, "check_event_mode_params failed\n");
2859 ret = init_lcore_rx_queues();
2861 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
2863 nb_lcores = rte_lcore_count();
2865 sess_sz = max_session_size();
2868 * In event mode request minimum number of crypto queues
2869 * to be reserved equal to number of ports.
2871 if (eh_conf->mode == EH_PKT_TRANSFER_MODE_EVENT)
2872 nb_crypto_qp = rte_eth_dev_count_avail();
2876 nb_crypto_qp = cryptodevs_init(nb_crypto_qp);
2878 if (nb_bufs_in_pool == 0) {
2879 RTE_ETH_FOREACH_DEV(portid) {
2880 if ((enabled_port_mask & (1 << portid)) == 0)
2883 nb_rxq += get_port_nb_rx_queues(portid);
2888 nb_bufs_in_pool = calculate_nb_mbufs(nb_ports, nb_crypto_qp,
2892 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2893 if (rte_lcore_is_enabled(lcore_id) == 0)
2897 socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2901 /* mbuf_pool is initialised by the pool_init() function*/
2902 if (socket_ctx[socket_id].mbuf_pool)
2905 pool_init(&socket_ctx[socket_id], socket_id, nb_bufs_in_pool);
2906 session_pool_init(&socket_ctx[socket_id], socket_id, sess_sz);
2907 session_priv_pool_init(&socket_ctx[socket_id], socket_id,
2910 printf("Number of mbufs in packet pool %d\n", nb_bufs_in_pool);
2912 RTE_ETH_FOREACH_DEV(portid) {
2913 if ((enabled_port_mask & (1 << portid)) == 0)
2916 sa_check_offloads(portid, &req_rx_offloads[portid],
2917 &req_tx_offloads[portid]);
2918 port_init(portid, req_rx_offloads[portid],
2919 req_tx_offloads[portid]);
2923 * Set the enabled port mask in helper config for use by helper
2924 * sub-system. This will be used while initializing devices using
2925 * helper sub-system.
2927 eh_conf->eth_portmask = enabled_port_mask;
2929 /* Initialize eventmode components */
2930 ret = eh_devs_init(eh_conf);
2932 rte_exit(EXIT_FAILURE, "eh_devs_init failed, err=%d\n", ret);
2935 RTE_ETH_FOREACH_DEV(portid) {
2936 if ((enabled_port_mask & (1 << portid)) == 0)
2939 /* Create flow before starting the device */
2940 create_default_ipsec_flow(portid, req_rx_offloads[portid]);
2942 ret = rte_eth_dev_start(portid);
2944 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: "
2945 "err=%d, port=%d\n", ret, portid);
2947 * If enabled, put device in promiscuous mode.
2948 * This allows IO forwarding mode to forward packets
2949 * to itself through 2 cross-connected ports of the
2952 if (promiscuous_on) {
2953 ret = rte_eth_promiscuous_enable(portid);
2955 rte_exit(EXIT_FAILURE,
2956 "rte_eth_promiscuous_enable: err=%s, port=%d\n",
2957 rte_strerror(-ret), portid);
2960 rte_eth_dev_callback_register(portid,
2961 RTE_ETH_EVENT_IPSEC, inline_ipsec_event_callback, NULL);
2964 /* fragment reassemble is enabled */
2965 if (frag_tbl_sz != 0) {
2966 ret = reassemble_init();
2968 rte_exit(EXIT_FAILURE, "failed at reassemble init");
2971 /* Replicate each context per socket */
2972 for (i = 0; i < NB_SOCKETS && i < rte_socket_count(); i++) {
2973 socket_id = rte_socket_id_by_idx(i);
2974 if ((socket_ctx[socket_id].mbuf_pool != NULL) &&
2975 (socket_ctx[socket_id].sa_in == NULL) &&
2976 (socket_ctx[socket_id].sa_out == NULL)) {
2977 sa_init(&socket_ctx[socket_id], socket_id);
2978 sp4_init(&socket_ctx[socket_id], socket_id);
2979 sp6_init(&socket_ctx[socket_id], socket_id);
2980 rt_init(&socket_ctx[socket_id], socket_id);
2986 check_all_ports_link_status(enabled_port_mask);
2988 #if (STATS_INTERVAL > 0)
2989 rte_eal_alarm_set(STATS_INTERVAL * US_PER_S, print_stats_cb, NULL);
2991 RTE_LOG(INFO, IPSEC, "Stats display disabled\n");
2992 #endif /* STATS_INTERVAL */
2994 /* launch per-lcore init on every lcore */
2995 rte_eal_mp_remote_launch(ipsec_launch_one_lcore, eh_conf, CALL_MASTER);
2996 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
2997 if (rte_eal_wait_lcore(lcore_id) < 0)
3001 /* Uninitialize eventmode components */
3002 ret = eh_devs_uninit(eh_conf);
3004 rte_exit(EXIT_FAILURE, "eh_devs_uninit failed, err=%d\n", ret);
3006 /* Free eventmode configuration memory */
3007 eh_conf_uninit(eh_conf);
3009 /* Destroy inline inbound and outbound sessions */
3010 for (i = 0; i < NB_SOCKETS && i < rte_socket_count(); i++) {
3011 socket_id = rte_socket_id_by_idx(i);
3012 inline_sessions_free(socket_ctx[socket_id].sa_in);
3013 inline_sessions_free(socket_ctx[socket_id].sa_out);
3016 for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
3017 printf("Closing cryptodev %d...", cdev_id);
3018 rte_cryptodev_stop(cdev_id);
3019 rte_cryptodev_close(cdev_id);
3023 RTE_ETH_FOREACH_DEV(portid) {
3024 if ((enabled_port_mask & (1 << portid)) == 0)
3027 printf("Closing port %d...", portid);
3028 if (flow_info_tbl[portid].rx_def_flow) {
3029 struct rte_flow_error err;
3031 ret = rte_flow_destroy(portid,
3032 flow_info_tbl[portid].rx_def_flow, &err);
3034 RTE_LOG(ERR, IPSEC, "Failed to destroy flow "
3035 " for port %u, err msg: %s\n", portid,
3038 rte_eth_dev_stop(portid);
3039 rte_eth_dev_close(portid);