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_cycles.h>
28 #include <rte_prefetch.h>
29 #include <rte_lcore.h>
30 #include <rte_per_lcore.h>
31 #include <rte_branch_prediction.h>
32 #include <rte_interrupts.h>
33 #include <rte_random.h>
34 #include <rte_debug.h>
35 #include <rte_ether.h>
36 #include <rte_ethdev.h>
37 #include <rte_mempool.h>
43 #include <rte_jhash.h>
44 #include <rte_cryptodev.h>
45 #include <rte_security.h>
46 #include <rte_eventdev.h>
48 #include <rte_ip_frag.h>
49 #include <rte_alarm.h>
51 #include "event_helper.h"
54 #include "ipsec_worker.h"
58 volatile bool force_quit;
60 #define MAX_JUMBO_PKT_LEN 9600
62 #define MEMPOOL_CACHE_SIZE 256
64 #define CDEV_QUEUE_DESC 2048
65 #define CDEV_MAP_ENTRIES 16384
66 #define CDEV_MP_CACHE_SZ 64
67 #define CDEV_MP_CACHE_MULTIPLIER 1.5 /* from rte_mempool.c */
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;
182 static uint32_t stats_interval;
184 /* application wide librte_ipsec/SA parameters */
185 struct app_sa_prm app_sa_prm = {
187 .cache_sz = SA_CACHE_SZ,
190 static const char *cfgfile;
192 struct lcore_rx_queue {
195 } __rte_cache_aligned;
197 struct lcore_params {
201 } __rte_cache_aligned;
203 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
205 static struct lcore_params *lcore_params;
206 static uint16_t nb_lcore_params;
208 static struct rte_hash *cdev_map_in;
209 static struct rte_hash *cdev_map_out;
213 struct rte_mbuf *m_table[MAX_PKT_BURST] __rte_aligned(sizeof(void *));
217 uint16_t nb_rx_queue;
218 struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
219 uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
220 struct buffer tx_mbufs[RTE_MAX_ETHPORTS];
221 struct ipsec_ctx inbound;
222 struct ipsec_ctx outbound;
223 struct rt_ctx *rt4_ctx;
224 struct rt_ctx *rt6_ctx;
226 struct rte_ip_frag_tbl *tbl;
227 struct rte_mempool *pool_dir;
228 struct rte_mempool *pool_indir;
229 struct rte_ip_frag_death_row dr;
231 } __rte_cache_aligned;
233 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
235 static struct rte_eth_conf port_conf = {
237 .mq_mode = RTE_ETH_MQ_RX_RSS,
239 .offloads = RTE_ETH_RX_OFFLOAD_CHECKSUM,
244 .rss_hf = RTE_ETH_RSS_IP | RTE_ETH_RSS_UDP |
245 RTE_ETH_RSS_TCP | RTE_ETH_RSS_SCTP,
249 .mq_mode = RTE_ETH_MQ_TX_NONE,
253 struct socket_ctx socket_ctx[NB_SOCKETS];
256 * Determine is multi-segment support required:
257 * - either frame buffer size is smaller then mtu
258 * - or reassmeble support is requested
261 multi_seg_required(void)
263 return (MTU_TO_FRAMELEN(mtu_size) + RTE_PKTMBUF_HEADROOM >
264 frame_buf_size || frag_tbl_sz != 0);
268 adjust_ipv4_pktlen(struct rte_mbuf *m, const struct rte_ipv4_hdr *iph,
273 plen = rte_be_to_cpu_16(iph->total_length) + l2_len;
274 if (plen < m->pkt_len) {
275 trim = m->pkt_len - plen;
276 rte_pktmbuf_trim(m, trim);
281 adjust_ipv6_pktlen(struct rte_mbuf *m, const struct rte_ipv6_hdr *iph,
286 plen = rte_be_to_cpu_16(iph->payload_len) + sizeof(*iph) + l2_len;
287 if (plen < m->pkt_len) {
288 trim = m->pkt_len - plen;
289 rte_pktmbuf_trim(m, trim);
294 struct ipsec_core_statistics core_statistics[RTE_MAX_LCORE];
296 /* Print out statistics on packet distribution */
298 print_stats_cb(__rte_unused void *param)
300 uint64_t total_packets_dropped, total_packets_tx, total_packets_rx;
301 float burst_percent, rx_per_call, tx_per_call;
304 total_packets_dropped = 0;
305 total_packets_tx = 0;
306 total_packets_rx = 0;
308 const char clr[] = { 27, '[', '2', 'J', '\0' };
309 const char topLeft[] = { 27, '[', '1', ';', '1', 'H', '\0' };
311 /* Clear screen and move to top left */
312 printf("%s%s", clr, topLeft);
314 printf("\nCore statistics ====================================");
316 for (coreid = 0; coreid < RTE_MAX_LCORE; coreid++) {
317 /* skip disabled cores */
318 if (rte_lcore_is_enabled(coreid) == 0)
320 burst_percent = (float)(core_statistics[coreid].burst_rx * 100)/
321 core_statistics[coreid].rx;
322 rx_per_call = (float)(core_statistics[coreid].rx)/
323 core_statistics[coreid].rx_call;
324 tx_per_call = (float)(core_statistics[coreid].tx)/
325 core_statistics[coreid].tx_call;
326 printf("\nStatistics for core %u ------------------------------"
327 "\nPackets received: %20"PRIu64
328 "\nPackets sent: %24"PRIu64
329 "\nPackets dropped: %21"PRIu64
330 "\nBurst percent: %23.2f"
331 "\nPackets per Rx call: %17.2f"
332 "\nPackets per Tx call: %17.2f",
334 core_statistics[coreid].rx,
335 core_statistics[coreid].tx,
336 core_statistics[coreid].dropped,
341 total_packets_dropped += core_statistics[coreid].dropped;
342 total_packets_tx += core_statistics[coreid].tx;
343 total_packets_rx += core_statistics[coreid].rx;
345 printf("\nAggregate statistics ==============================="
346 "\nTotal packets received: %14"PRIu64
347 "\nTotal packets sent: %18"PRIu64
348 "\nTotal packets dropped: %15"PRIu64,
351 total_packets_dropped);
352 printf("\n====================================================\n");
354 rte_eal_alarm_set(stats_interval * US_PER_S, print_stats_cb, NULL);
358 prepare_one_packet(struct rte_mbuf *pkt, struct ipsec_traffic *t)
360 const struct rte_ether_hdr *eth;
361 const struct rte_ipv4_hdr *iph4;
362 const struct rte_ipv6_hdr *iph6;
363 const struct rte_udp_hdr *udp;
364 uint16_t ip4_hdr_len;
367 eth = rte_pktmbuf_mtod(pkt, const struct rte_ether_hdr *);
368 if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
370 iph4 = (const struct rte_ipv4_hdr *)rte_pktmbuf_adj(pkt,
372 adjust_ipv4_pktlen(pkt, iph4, 0);
374 switch (iph4->next_proto_id) {
376 t->ipsec.pkts[(t->ipsec.num)++] = pkt;
379 if (app_sa_prm.udp_encap == 1) {
380 ip4_hdr_len = ((iph4->version_ihl &
381 RTE_IPV4_HDR_IHL_MASK) *
382 RTE_IPV4_IHL_MULTIPLIER);
383 udp = rte_pktmbuf_mtod_offset(pkt,
384 struct rte_udp_hdr *, ip4_hdr_len);
385 nat_port = rte_cpu_to_be_16(IPSEC_NAT_T_PORT);
386 if (udp->src_port == nat_port ||
387 udp->dst_port == nat_port){
388 t->ipsec.pkts[(t->ipsec.num)++] = pkt;
390 MBUF_PTYPE_TUNNEL_ESP_IN_UDP;
396 t->ip4.data[t->ip4.num] = &iph4->next_proto_id;
397 t->ip4.pkts[(t->ip4.num)++] = pkt;
400 pkt->l3_len = sizeof(*iph4);
401 pkt->packet_type |= RTE_PTYPE_L3_IPV4;
402 if (pkt->packet_type & RTE_PTYPE_L4_TCP)
403 pkt->l4_len = sizeof(struct rte_tcp_hdr);
404 else if (pkt->packet_type & RTE_PTYPE_L4_UDP)
405 pkt->l4_len = sizeof(struct rte_udp_hdr);
406 } else if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6)) {
408 size_t l3len, ext_len;
411 /* get protocol type */
412 iph6 = (const struct rte_ipv6_hdr *)rte_pktmbuf_adj(pkt,
414 adjust_ipv6_pktlen(pkt, iph6, 0);
416 next_proto = iph6->proto;
418 /* determine l3 header size up to ESP extension */
419 l3len = sizeof(struct ip6_hdr);
420 p = rte_pktmbuf_mtod(pkt, uint8_t *);
421 while (next_proto != IPPROTO_ESP && l3len < pkt->data_len &&
422 (next_proto = rte_ipv6_get_next_ext(p + l3len,
423 next_proto, &ext_len)) >= 0)
426 /* drop packet when IPv6 header exceeds first segment length */
427 if (unlikely(l3len > pkt->data_len)) {
432 switch (next_proto) {
434 t->ipsec.pkts[(t->ipsec.num)++] = pkt;
437 if (app_sa_prm.udp_encap == 1) {
438 udp = rte_pktmbuf_mtod_offset(pkt,
439 struct rte_udp_hdr *, l3len);
440 nat_port = rte_cpu_to_be_16(IPSEC_NAT_T_PORT);
441 if (udp->src_port == nat_port ||
442 udp->dst_port == nat_port){
443 t->ipsec.pkts[(t->ipsec.num)++] = pkt;
445 MBUF_PTYPE_TUNNEL_ESP_IN_UDP;
451 t->ip6.data[t->ip6.num] = &iph6->proto;
452 t->ip6.pkts[(t->ip6.num)++] = pkt;
456 pkt->packet_type |= RTE_PTYPE_L3_IPV6;
458 /* Unknown/Unsupported type, drop the packet */
459 RTE_LOG(ERR, IPSEC, "Unsupported packet type 0x%x\n",
460 rte_be_to_cpu_16(eth->ether_type));
465 /* Check if the packet has been processed inline. For inline protocol
466 * processed packets, the metadata in the mbuf can be used to identify
467 * the security processing done on the packet. The metadata will be
468 * used to retrieve the application registered userdata associated
469 * with the security session.
472 if (pkt->ol_flags & RTE_MBUF_F_RX_SEC_OFFLOAD &&
473 rte_security_dynfield_is_registered()) {
475 struct ipsec_mbuf_metadata *priv;
476 struct rte_security_ctx *ctx = (struct rte_security_ctx *)
477 rte_eth_dev_get_sec_ctx(
480 /* Retrieve the userdata registered. Here, the userdata
481 * registered is the SA pointer.
483 sa = (struct ipsec_sa *)rte_security_get_userdata(ctx,
484 *rte_security_dynfield(pkt));
486 /* userdata could not be retrieved */
490 /* Save SA as priv member in mbuf. This will be used in the
491 * IPsec selector(SP-SA) check.
494 priv = get_priv(pkt);
500 prepare_traffic(struct rte_mbuf **pkts, struct ipsec_traffic *t,
509 for (i = 0; i < (nb_pkts - PREFETCH_OFFSET); i++) {
510 rte_prefetch0(rte_pktmbuf_mtod(pkts[i + PREFETCH_OFFSET],
512 prepare_one_packet(pkts[i], t);
514 /* Process left packets */
515 for (; i < nb_pkts; i++)
516 prepare_one_packet(pkts[i], t);
520 prepare_tx_pkt(struct rte_mbuf *pkt, uint16_t port,
521 const struct lcore_conf *qconf)
524 struct rte_ether_hdr *ethhdr;
526 ip = rte_pktmbuf_mtod(pkt, struct ip *);
528 ethhdr = (struct rte_ether_hdr *)
529 rte_pktmbuf_prepend(pkt, RTE_ETHER_HDR_LEN);
531 if (ip->ip_v == IPVERSION) {
532 pkt->ol_flags |= qconf->outbound.ipv4_offloads;
533 pkt->l3_len = sizeof(struct ip);
534 pkt->l2_len = RTE_ETHER_HDR_LEN;
538 /* calculate IPv4 cksum in SW */
539 if ((pkt->ol_flags & RTE_MBUF_F_TX_IP_CKSUM) == 0)
540 ip->ip_sum = rte_ipv4_cksum((struct rte_ipv4_hdr *)ip);
542 ethhdr->ether_type = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4);
544 pkt->ol_flags |= qconf->outbound.ipv6_offloads;
545 pkt->l3_len = sizeof(struct ip6_hdr);
546 pkt->l2_len = RTE_ETHER_HDR_LEN;
548 ethhdr->ether_type = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6);
551 memcpy(ðhdr->src_addr, ðaddr_tbl[port].src,
552 sizeof(struct rte_ether_addr));
553 memcpy(ðhdr->dst_addr, ðaddr_tbl[port].dst,
554 sizeof(struct rte_ether_addr));
558 prepare_tx_burst(struct rte_mbuf *pkts[], uint16_t nb_pkts, uint16_t port,
559 const struct lcore_conf *qconf)
562 const int32_t prefetch_offset = 2;
564 for (i = 0; i < (nb_pkts - prefetch_offset); i++) {
565 rte_mbuf_prefetch_part2(pkts[i + prefetch_offset]);
566 prepare_tx_pkt(pkts[i], port, qconf);
568 /* Process left packets */
569 for (; i < nb_pkts; i++)
570 prepare_tx_pkt(pkts[i], port, qconf);
573 /* Send burst of packets on an output interface */
574 static inline int32_t
575 send_burst(struct lcore_conf *qconf, uint16_t n, uint16_t port)
577 struct rte_mbuf **m_table;
581 queueid = qconf->tx_queue_id[port];
582 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
584 prepare_tx_burst(m_table, n, port, qconf);
586 ret = rte_eth_tx_burst(port, queueid, m_table, n);
588 core_stats_update_tx(ret);
590 if (unlikely(ret < n)) {
592 free_pkts(&m_table[ret], 1);
600 * Helper function to fragment and queue for TX one packet.
602 static inline uint32_t
603 send_fragment_packet(struct lcore_conf *qconf, struct rte_mbuf *m,
604 uint16_t port, uint8_t proto)
610 tbl = qconf->tx_mbufs + port;
613 /* free space for new fragments */
614 if (len + RTE_LIBRTE_IP_FRAG_MAX_FRAG >= RTE_DIM(tbl->m_table)) {
615 send_burst(qconf, len, port);
619 n = RTE_DIM(tbl->m_table) - len;
621 if (proto == IPPROTO_IP)
622 rc = rte_ipv4_fragment_packet(m, tbl->m_table + len,
623 n, mtu_size, qconf->frag.pool_dir,
624 qconf->frag.pool_indir);
626 rc = rte_ipv6_fragment_packet(m, tbl->m_table + len,
627 n, mtu_size, qconf->frag.pool_dir,
628 qconf->frag.pool_indir);
634 "%s: failed to fragment packet with size %u, "
636 __func__, m->pkt_len, rte_errno);
642 /* Enqueue a single packet, and send burst if queue is filled */
643 static inline int32_t
644 send_single_packet(struct rte_mbuf *m, uint16_t port, uint8_t proto)
648 struct lcore_conf *qconf;
650 lcore_id = rte_lcore_id();
652 qconf = &lcore_conf[lcore_id];
653 len = qconf->tx_mbufs[port].len;
655 if (m->pkt_len <= mtu_size) {
656 qconf->tx_mbufs[port].m_table[len] = m;
659 /* need to fragment the packet */
660 } else if (frag_tbl_sz > 0)
661 len = send_fragment_packet(qconf, m, port, proto);
665 /* enough pkts to be sent */
666 if (unlikely(len == MAX_PKT_BURST)) {
667 send_burst(qconf, MAX_PKT_BURST, port);
671 qconf->tx_mbufs[port].len = len;
676 inbound_sp_sa(struct sp_ctx *sp, struct sa_ctx *sa, struct traffic_type *ip,
680 uint32_t i, j, res, sa_idx;
682 if (ip->num == 0 || sp == NULL)
685 rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
686 ip->num, DEFAULT_MAX_CATEGORIES);
689 for (i = 0; i < ip->num; i++) {
696 if (res == DISCARD) {
701 /* Only check SPI match for processed IPSec packets */
702 if (i < lim && ((m->ol_flags & RTE_MBUF_F_RX_SEC_OFFLOAD) == 0)) {
708 if (!inbound_sa_check(sa, m, sa_idx)) {
718 split46_traffic(struct ipsec_traffic *trf, struct rte_mbuf *mb[], uint32_t num)
727 for (i = 0; i < num; i++) {
730 ip = rte_pktmbuf_mtod(m, struct ip *);
732 if (ip->ip_v == IPVERSION) {
733 trf->ip4.pkts[n4] = m;
734 trf->ip4.data[n4] = rte_pktmbuf_mtod_offset(m,
735 uint8_t *, offsetof(struct ip, ip_p));
737 } else if (ip->ip_v == IP6_VERSION) {
738 trf->ip6.pkts[n6] = m;
739 trf->ip6.data[n6] = rte_pktmbuf_mtod_offset(m,
741 offsetof(struct ip6_hdr, ip6_nxt));
753 process_pkts_inbound(struct ipsec_ctx *ipsec_ctx,
754 struct ipsec_traffic *traffic)
756 uint16_t nb_pkts_in, n_ip4, n_ip6;
758 n_ip4 = traffic->ip4.num;
759 n_ip6 = traffic->ip6.num;
761 if (app_sa_prm.enable == 0) {
762 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
763 traffic->ipsec.num, MAX_PKT_BURST);
764 split46_traffic(traffic, traffic->ipsec.pkts, nb_pkts_in);
766 inbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.pkts,
767 traffic->ipsec.saptr, traffic->ipsec.num);
768 ipsec_process(ipsec_ctx, traffic);
771 inbound_sp_sa(ipsec_ctx->sp4_ctx, ipsec_ctx->sa_ctx, &traffic->ip4,
774 inbound_sp_sa(ipsec_ctx->sp6_ctx, ipsec_ctx->sa_ctx, &traffic->ip6,
779 outbound_sp(struct sp_ctx *sp, struct traffic_type *ip,
780 struct traffic_type *ipsec)
783 uint32_t i, j, sa_idx;
785 if (ip->num == 0 || sp == NULL)
788 rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
789 ip->num, DEFAULT_MAX_CATEGORIES);
792 for (i = 0; i < ip->num; i++) {
794 sa_idx = ip->res[i] - 1;
795 if (ip->res[i] == DISCARD)
797 else if (ip->res[i] == BYPASS)
800 ipsec->res[ipsec->num] = sa_idx;
801 ipsec->pkts[ipsec->num++] = m;
808 process_pkts_outbound(struct ipsec_ctx *ipsec_ctx,
809 struct ipsec_traffic *traffic)
812 uint16_t idx, nb_pkts_out, i;
814 /* Drop any IPsec traffic from protected ports */
815 free_pkts(traffic->ipsec.pkts, traffic->ipsec.num);
817 traffic->ipsec.num = 0;
819 outbound_sp(ipsec_ctx->sp4_ctx, &traffic->ip4, &traffic->ipsec);
821 outbound_sp(ipsec_ctx->sp6_ctx, &traffic->ip6, &traffic->ipsec);
823 if (app_sa_prm.enable == 0) {
825 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
826 traffic->ipsec.res, traffic->ipsec.num,
829 for (i = 0; i < nb_pkts_out; i++) {
830 m = traffic->ipsec.pkts[i];
831 struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
832 if (ip->ip_v == IPVERSION) {
833 idx = traffic->ip4.num++;
834 traffic->ip4.pkts[idx] = m;
836 idx = traffic->ip6.num++;
837 traffic->ip6.pkts[idx] = m;
841 outbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.res,
842 traffic->ipsec.saptr, traffic->ipsec.num);
843 ipsec_process(ipsec_ctx, traffic);
848 process_pkts_inbound_nosp(struct ipsec_ctx *ipsec_ctx,
849 struct ipsec_traffic *traffic)
852 uint32_t nb_pkts_in, i, idx;
854 if (app_sa_prm.enable == 0) {
856 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
857 traffic->ipsec.num, MAX_PKT_BURST);
859 for (i = 0; i < nb_pkts_in; i++) {
860 m = traffic->ipsec.pkts[i];
861 struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
862 if (ip->ip_v == IPVERSION) {
863 idx = traffic->ip4.num++;
864 traffic->ip4.pkts[idx] = m;
866 idx = traffic->ip6.num++;
867 traffic->ip6.pkts[idx] = m;
871 inbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.pkts,
872 traffic->ipsec.saptr, traffic->ipsec.num);
873 ipsec_process(ipsec_ctx, traffic);
878 process_pkts_outbound_nosp(struct ipsec_ctx *ipsec_ctx,
879 struct ipsec_traffic *traffic)
882 uint32_t nb_pkts_out, i, n;
885 /* Drop any IPsec traffic from protected ports */
886 free_pkts(traffic->ipsec.pkts, traffic->ipsec.num);
890 for (i = 0; i < traffic->ip4.num; i++) {
891 traffic->ipsec.pkts[n] = traffic->ip4.pkts[i];
892 traffic->ipsec.res[n++] = single_sa_idx;
895 for (i = 0; i < traffic->ip6.num; i++) {
896 traffic->ipsec.pkts[n] = traffic->ip6.pkts[i];
897 traffic->ipsec.res[n++] = single_sa_idx;
900 traffic->ip4.num = 0;
901 traffic->ip6.num = 0;
902 traffic->ipsec.num = n;
904 if (app_sa_prm.enable == 0) {
906 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
907 traffic->ipsec.res, traffic->ipsec.num,
910 /* They all sue the same SA (ip4 or ip6 tunnel) */
911 m = traffic->ipsec.pkts[0];
912 ip = rte_pktmbuf_mtod(m, struct ip *);
913 if (ip->ip_v == IPVERSION) {
914 traffic->ip4.num = nb_pkts_out;
915 for (i = 0; i < nb_pkts_out; i++)
916 traffic->ip4.pkts[i] = traffic->ipsec.pkts[i];
918 traffic->ip6.num = nb_pkts_out;
919 for (i = 0; i < nb_pkts_out; i++)
920 traffic->ip6.pkts[i] = traffic->ipsec.pkts[i];
923 outbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.res,
924 traffic->ipsec.saptr, traffic->ipsec.num);
925 ipsec_process(ipsec_ctx, traffic);
929 static inline int32_t
930 get_hop_for_offload_pkt(struct rte_mbuf *pkt, int is_ipv6)
932 struct ipsec_mbuf_metadata *priv;
935 priv = get_priv(pkt);
938 if (unlikely(sa == NULL)) {
939 RTE_LOG(ERR, IPSEC, "SA not saved in private data\n");
947 return (sa->portid | RTE_LPM_LOOKUP_SUCCESS);
958 route4_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
960 uint32_t hop[MAX_PKT_BURST * 2];
961 uint32_t dst_ip[MAX_PKT_BURST * 2];
964 uint16_t lpm_pkts = 0;
969 /* Need to do an LPM lookup for non-inline packets. Inline packets will
970 * have port ID in the SA
973 for (i = 0; i < nb_pkts; i++) {
974 if (!(pkts[i]->ol_flags & RTE_MBUF_F_TX_SEC_OFFLOAD)) {
975 /* Security offload not enabled. So an LPM lookup is
976 * required to get the hop
978 offset = offsetof(struct ip, ip_dst);
979 dst_ip[lpm_pkts] = *rte_pktmbuf_mtod_offset(pkts[i],
981 dst_ip[lpm_pkts] = rte_be_to_cpu_32(dst_ip[lpm_pkts]);
986 rte_lpm_lookup_bulk((struct rte_lpm *)rt_ctx, dst_ip, hop, lpm_pkts);
990 for (i = 0; i < nb_pkts; i++) {
991 if (pkts[i]->ol_flags & RTE_MBUF_F_TX_SEC_OFFLOAD) {
992 /* Read hop from the SA */
993 pkt_hop = get_hop_for_offload_pkt(pkts[i], 0);
995 /* Need to use hop returned by lookup */
996 pkt_hop = hop[lpm_pkts++];
999 if ((pkt_hop & RTE_LPM_LOOKUP_SUCCESS) == 0) {
1000 free_pkts(&pkts[i], 1);
1003 send_single_packet(pkts[i], pkt_hop & 0xff, IPPROTO_IP);
1008 route6_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
1010 int32_t hop[MAX_PKT_BURST * 2];
1011 uint8_t dst_ip[MAX_PKT_BURST * 2][16];
1013 int32_t pkt_hop = 0;
1015 uint16_t lpm_pkts = 0;
1020 /* Need to do an LPM lookup for non-inline packets. Inline packets will
1021 * have port ID in the SA
1024 for (i = 0; i < nb_pkts; i++) {
1025 if (!(pkts[i]->ol_flags & RTE_MBUF_F_TX_SEC_OFFLOAD)) {
1026 /* Security offload not enabled. So an LPM lookup is
1027 * required to get the hop
1029 offset = offsetof(struct ip6_hdr, ip6_dst);
1030 ip6_dst = rte_pktmbuf_mtod_offset(pkts[i], uint8_t *,
1032 memcpy(&dst_ip[lpm_pkts][0], ip6_dst, 16);
1037 rte_lpm6_lookup_bulk_func((struct rte_lpm6 *)rt_ctx, dst_ip, hop,
1042 for (i = 0; i < nb_pkts; i++) {
1043 if (pkts[i]->ol_flags & RTE_MBUF_F_TX_SEC_OFFLOAD) {
1044 /* Read hop from the SA */
1045 pkt_hop = get_hop_for_offload_pkt(pkts[i], 1);
1047 /* Need to use hop returned by lookup */
1048 pkt_hop = hop[lpm_pkts++];
1051 if (pkt_hop == -1) {
1052 free_pkts(&pkts[i], 1);
1055 send_single_packet(pkts[i], pkt_hop & 0xff, IPPROTO_IPV6);
1060 process_pkts(struct lcore_conf *qconf, struct rte_mbuf **pkts,
1061 uint8_t nb_pkts, uint16_t portid)
1063 struct ipsec_traffic traffic;
1065 prepare_traffic(pkts, &traffic, nb_pkts);
1067 if (unlikely(single_sa)) {
1068 if (is_unprotected_port(portid))
1069 process_pkts_inbound_nosp(&qconf->inbound, &traffic);
1071 process_pkts_outbound_nosp(&qconf->outbound, &traffic);
1073 if (is_unprotected_port(portid))
1074 process_pkts_inbound(&qconf->inbound, &traffic);
1076 process_pkts_outbound(&qconf->outbound, &traffic);
1079 route4_pkts(qconf->rt4_ctx, traffic.ip4.pkts, traffic.ip4.num);
1080 route6_pkts(qconf->rt6_ctx, traffic.ip6.pkts, traffic.ip6.num);
1084 drain_tx_buffers(struct lcore_conf *qconf)
1089 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
1090 buf = &qconf->tx_mbufs[portid];
1093 send_burst(qconf, buf->len, portid);
1099 drain_crypto_buffers(struct lcore_conf *qconf)
1102 struct ipsec_ctx *ctx;
1104 /* drain inbound buffers*/
1105 ctx = &qconf->inbound;
1106 for (i = 0; i != ctx->nb_qps; i++) {
1107 if (ctx->tbl[i].len != 0)
1108 enqueue_cop_burst(ctx->tbl + i);
1111 /* drain outbound buffers*/
1112 ctx = &qconf->outbound;
1113 for (i = 0; i != ctx->nb_qps; i++) {
1114 if (ctx->tbl[i].len != 0)
1115 enqueue_cop_burst(ctx->tbl + i);
1120 drain_inbound_crypto_queues(const struct lcore_conf *qconf,
1121 struct ipsec_ctx *ctx)
1124 struct ipsec_traffic trf;
1126 if (app_sa_prm.enable == 0) {
1128 /* dequeue packets from crypto-queue */
1129 n = ipsec_inbound_cqp_dequeue(ctx, trf.ipsec.pkts,
1130 RTE_DIM(trf.ipsec.pkts));
1135 /* split traffic by ipv4-ipv6 */
1136 split46_traffic(&trf, trf.ipsec.pkts, n);
1138 ipsec_cqp_process(ctx, &trf);
1140 /* process ipv4 packets */
1141 if (trf.ip4.num != 0) {
1142 inbound_sp_sa(ctx->sp4_ctx, ctx->sa_ctx, &trf.ip4, 0);
1143 route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
1146 /* process ipv6 packets */
1147 if (trf.ip6.num != 0) {
1148 inbound_sp_sa(ctx->sp6_ctx, ctx->sa_ctx, &trf.ip6, 0);
1149 route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
1154 drain_outbound_crypto_queues(const struct lcore_conf *qconf,
1155 struct ipsec_ctx *ctx)
1158 struct ipsec_traffic trf;
1160 if (app_sa_prm.enable == 0) {
1162 /* dequeue packets from crypto-queue */
1163 n = ipsec_outbound_cqp_dequeue(ctx, trf.ipsec.pkts,
1164 RTE_DIM(trf.ipsec.pkts));
1169 /* split traffic by ipv4-ipv6 */
1170 split46_traffic(&trf, trf.ipsec.pkts, n);
1172 ipsec_cqp_process(ctx, &trf);
1174 /* process ipv4 packets */
1175 if (trf.ip4.num != 0)
1176 route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
1178 /* process ipv6 packets */
1179 if (trf.ip6.num != 0)
1180 route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
1183 /* main processing loop */
1185 ipsec_poll_mode_worker(void)
1187 struct rte_mbuf *pkts[MAX_PKT_BURST];
1189 uint64_t prev_tsc, diff_tsc, cur_tsc;
1193 struct lcore_conf *qconf;
1194 int32_t rc, socket_id;
1195 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
1196 / US_PER_S * BURST_TX_DRAIN_US;
1197 struct lcore_rx_queue *rxql;
1200 lcore_id = rte_lcore_id();
1201 qconf = &lcore_conf[lcore_id];
1202 rxql = qconf->rx_queue_list;
1203 socket_id = rte_lcore_to_socket_id(lcore_id);
1205 qconf->rt4_ctx = socket_ctx[socket_id].rt_ip4;
1206 qconf->rt6_ctx = socket_ctx[socket_id].rt_ip6;
1207 qconf->inbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_in;
1208 qconf->inbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_in;
1209 qconf->inbound.sa_ctx = socket_ctx[socket_id].sa_in;
1210 qconf->inbound.cdev_map = cdev_map_in;
1211 qconf->inbound.session_pool = socket_ctx[socket_id].session_pool;
1212 qconf->inbound.session_priv_pool =
1213 socket_ctx[socket_id].session_priv_pool;
1214 qconf->outbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_out;
1215 qconf->outbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_out;
1216 qconf->outbound.sa_ctx = socket_ctx[socket_id].sa_out;
1217 qconf->outbound.cdev_map = cdev_map_out;
1218 qconf->outbound.session_pool = socket_ctx[socket_id].session_pool;
1219 qconf->outbound.session_priv_pool =
1220 socket_ctx[socket_id].session_priv_pool;
1221 qconf->frag.pool_dir = socket_ctx[socket_id].mbuf_pool;
1222 qconf->frag.pool_indir = socket_ctx[socket_id].mbuf_pool_indir;
1224 rc = ipsec_sad_lcore_cache_init(app_sa_prm.cache_sz);
1227 "SAD cache init on lcore %u, failed with code: %d\n",
1232 if (qconf->nb_rx_queue == 0) {
1233 RTE_LOG(DEBUG, IPSEC, "lcore %u has nothing to do\n",
1238 RTE_LOG(INFO, IPSEC, "entering main loop on lcore %u\n", lcore_id);
1240 for (i = 0; i < qconf->nb_rx_queue; i++) {
1241 portid = rxql[i].port_id;
1242 queueid = rxql[i].queue_id;
1243 RTE_LOG(INFO, IPSEC,
1244 " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
1245 lcore_id, portid, queueid);
1248 while (!force_quit) {
1249 cur_tsc = rte_rdtsc();
1251 /* TX queue buffer drain */
1252 diff_tsc = cur_tsc - prev_tsc;
1254 if (unlikely(diff_tsc > drain_tsc)) {
1255 drain_tx_buffers(qconf);
1256 drain_crypto_buffers(qconf);
1260 for (i = 0; i < qconf->nb_rx_queue; ++i) {
1262 /* Read packets from RX queues */
1263 portid = rxql[i].port_id;
1264 queueid = rxql[i].queue_id;
1265 nb_rx = rte_eth_rx_burst(portid, queueid,
1266 pkts, MAX_PKT_BURST);
1269 core_stats_update_rx(nb_rx);
1270 process_pkts(qconf, pkts, nb_rx, portid);
1273 /* dequeue and process completed crypto-ops */
1274 if (is_unprotected_port(portid))
1275 drain_inbound_crypto_queues(qconf,
1278 drain_outbound_crypto_queues(qconf,
1285 check_flow_params(uint16_t fdir_portid, uint8_t fdir_qid)
1291 for (i = 0; i < nb_lcore_params; ++i) {
1292 portid = lcore_params_array[i].port_id;
1293 if (portid == fdir_portid) {
1294 queueid = lcore_params_array[i].queue_id;
1295 if (queueid == fdir_qid)
1299 if (i == nb_lcore_params - 1)
1307 check_poll_mode_params(struct eh_conf *eh_conf)
1317 if (eh_conf->mode != EH_PKT_TRANSFER_MODE_POLL)
1320 if (lcore_params == NULL) {
1321 printf("Error: No port/queue/core mappings\n");
1325 for (i = 0; i < nb_lcore_params; ++i) {
1326 lcore = lcore_params[i].lcore_id;
1327 if (!rte_lcore_is_enabled(lcore)) {
1328 printf("error: lcore %hhu is not enabled in "
1329 "lcore mask\n", lcore);
1332 socket_id = rte_lcore_to_socket_id(lcore);
1333 if (socket_id != 0 && numa_on == 0) {
1334 printf("warning: lcore %hhu is on socket %d "
1338 portid = lcore_params[i].port_id;
1339 if ((enabled_port_mask & (1 << portid)) == 0) {
1340 printf("port %u is not enabled in port mask\n", portid);
1343 if (!rte_eth_dev_is_valid_port(portid)) {
1344 printf("port %u is not present on the board\n", portid);
1352 get_port_nb_rx_queues(const uint16_t port)
1357 for (i = 0; i < nb_lcore_params; ++i) {
1358 if (lcore_params[i].port_id == port &&
1359 lcore_params[i].queue_id > queue)
1360 queue = lcore_params[i].queue_id;
1362 return (uint8_t)(++queue);
1366 init_lcore_rx_queues(void)
1368 uint16_t i, nb_rx_queue;
1371 for (i = 0; i < nb_lcore_params; ++i) {
1372 lcore = lcore_params[i].lcore_id;
1373 nb_rx_queue = lcore_conf[lcore].nb_rx_queue;
1374 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1375 printf("error: too many queues (%u) for lcore: %u\n",
1376 nb_rx_queue + 1, lcore);
1379 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1380 lcore_params[i].port_id;
1381 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1382 lcore_params[i].queue_id;
1383 lcore_conf[lcore].nb_rx_queue++;
1390 print_usage(const char *prgname)
1392 fprintf(stderr, "%s [EAL options] --"
1398 " [-w REPLAY_WINDOW_SIZE]"
1402 " [-t STATS_INTERVAL]"
1403 " [-s NUMBER_OF_MBUFS_IN_PKT_POOL]"
1405 " --config (port,queue,lcore)[,(port,queue,lcore)]"
1406 " [--single-sa SAIDX]"
1407 " [--cryptodev_mask MASK]"
1408 " [--transfer-mode MODE]"
1409 " [--event-schedule-type TYPE]"
1410 " [--" CMD_LINE_OPT_RX_OFFLOAD " RX_OFFLOAD_MASK]"
1411 " [--" CMD_LINE_OPT_TX_OFFLOAD " TX_OFFLOAD_MASK]"
1412 " [--" CMD_LINE_OPT_REASSEMBLE " REASSEMBLE_TABLE_SIZE]"
1413 " [--" CMD_LINE_OPT_MTU " MTU]"
1415 " -p PORTMASK: Hexadecimal bitmask of ports to configure\n"
1416 " -P : Enable promiscuous mode\n"
1417 " -u PORTMASK: Hexadecimal bitmask of unprotected ports\n"
1418 " -j FRAMESIZE: Data buffer size, minimum (and default)\n"
1419 " value: RTE_MBUF_DEFAULT_BUF_SIZE\n"
1420 " -l enables code-path that uses librte_ipsec\n"
1421 " -w REPLAY_WINDOW_SIZE specifies IPsec SQN replay window\n"
1422 " size for each SA\n"
1424 " -a enables SA SQN atomic behaviour\n"
1425 " -c specifies inbound SAD cache size,\n"
1426 " zero value disables the cache (default value: 128)\n"
1427 " -t specifies statistics screen update interval,\n"
1428 " zero disables statistics screen (default value: 0)\n"
1429 " -s number of mbufs in packet pool, if not specified number\n"
1430 " of mbufs will be calculated based on number of cores,\n"
1431 " ports and crypto queues\n"
1432 " -f CONFIG_FILE: Configuration file\n"
1433 " --config (port,queue,lcore): Rx queue configuration. In poll\n"
1434 " mode determines which queues from\n"
1435 " which ports are mapped to which cores.\n"
1436 " In event mode this option is not used\n"
1437 " as packets are dynamically scheduled\n"
1438 " to cores by HW.\n"
1439 " --single-sa SAIDX: In poll mode use single SA index for\n"
1440 " outbound traffic, bypassing the SP\n"
1441 " In event mode selects driver submode,\n"
1442 " SA index value is ignored\n"
1443 " --cryptodev_mask MASK: Hexadecimal bitmask of the crypto\n"
1444 " devices to configure\n"
1445 " --transfer-mode MODE\n"
1446 " \"poll\" : Packet transfer via polling (default)\n"
1447 " \"event\" : Packet transfer via event device\n"
1448 " --event-schedule-type TYPE queue schedule type, used only when\n"
1449 " transfer mode is set to event\n"
1450 " \"ordered\" : Ordered (default)\n"
1451 " \"atomic\" : Atomic\n"
1452 " \"parallel\" : Parallel\n"
1453 " --" CMD_LINE_OPT_RX_OFFLOAD
1454 ": bitmask of the RX HW offload capabilities to enable/use\n"
1455 " (RTE_ETH_RX_OFFLOAD_*)\n"
1456 " --" CMD_LINE_OPT_TX_OFFLOAD
1457 ": bitmask of the TX HW offload capabilities to enable/use\n"
1458 " (RTE_ETH_TX_OFFLOAD_*)\n"
1459 " --" CMD_LINE_OPT_REASSEMBLE " NUM"
1460 ": max number of entries in reassemble(fragment) table\n"
1461 " (zero (default value) disables reassembly)\n"
1462 " --" CMD_LINE_OPT_MTU " MTU"
1463 ": MTU value on all ports (default value: 1500)\n"
1464 " outgoing packets with bigger size will be fragmented\n"
1465 " incoming packets with bigger size will be discarded\n"
1466 " --" CMD_LINE_OPT_FRAG_TTL " FRAG_TTL_NS"
1467 ": fragments lifetime in nanoseconds, default\n"
1468 " and maximum value is 10.000.000.000 ns (10 s)\n"
1474 parse_mask(const char *str, uint64_t *val)
1480 t = strtoul(str, &end, 0);
1481 if (errno != 0 || end[0] != 0)
1489 parse_portmask(const char *portmask)
1496 /* parse hexadecimal string */
1497 pm = strtoul(portmask, &end, 16);
1498 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1501 if ((pm == 0) && errno)
1508 parse_decimal(const char *str)
1513 num = strtoull(str, &end, 10);
1514 if ((str[0] == '\0') || (end == NULL) || (*end != '\0')
1522 parse_config(const char *q_arg)
1525 const char *p, *p0 = q_arg;
1533 unsigned long int_fld[_NUM_FLD];
1534 char *str_fld[_NUM_FLD];
1538 nb_lcore_params = 0;
1540 while ((p = strchr(p0, '(')) != NULL) {
1542 p0 = strchr(p, ')');
1547 if (size >= sizeof(s))
1550 snprintf(s, sizeof(s), "%.*s", size, p);
1551 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1554 for (i = 0; i < _NUM_FLD; i++) {
1556 int_fld[i] = strtoul(str_fld[i], &end, 0);
1557 if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
1560 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1561 printf("exceeded max number of lcore params: %hu\n",
1565 lcore_params_array[nb_lcore_params].port_id =
1566 (uint8_t)int_fld[FLD_PORT];
1567 lcore_params_array[nb_lcore_params].queue_id =
1568 (uint8_t)int_fld[FLD_QUEUE];
1569 lcore_params_array[nb_lcore_params].lcore_id =
1570 (uint8_t)int_fld[FLD_LCORE];
1573 lcore_params = lcore_params_array;
1578 print_app_sa_prm(const struct app_sa_prm *prm)
1580 printf("librte_ipsec usage: %s\n",
1581 (prm->enable == 0) ? "disabled" : "enabled");
1583 printf("replay window size: %u\n", prm->window_size);
1584 printf("ESN: %s\n", (prm->enable_esn == 0) ? "disabled" : "enabled");
1585 printf("SA flags: %#" PRIx64 "\n", prm->flags);
1586 printf("Frag TTL: %" PRIu64 " ns\n", frag_ttl_ns);
1590 parse_transfer_mode(struct eh_conf *conf, const char *optarg)
1592 if (!strcmp(CMD_LINE_ARG_POLL, optarg))
1593 conf->mode = EH_PKT_TRANSFER_MODE_POLL;
1594 else if (!strcmp(CMD_LINE_ARG_EVENT, optarg))
1595 conf->mode = EH_PKT_TRANSFER_MODE_EVENT;
1597 printf("Unsupported packet transfer mode\n");
1605 parse_schedule_type(struct eh_conf *conf, const char *optarg)
1607 struct eventmode_conf *em_conf = NULL;
1609 /* Get eventmode conf */
1610 em_conf = conf->mode_params;
1612 if (!strcmp(CMD_LINE_ARG_ORDERED, optarg))
1613 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_ORDERED;
1614 else if (!strcmp(CMD_LINE_ARG_ATOMIC, optarg))
1615 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_ATOMIC;
1616 else if (!strcmp(CMD_LINE_ARG_PARALLEL, optarg))
1617 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_PARALLEL;
1619 printf("Unsupported queue schedule type\n");
1627 parse_args(int32_t argc, char **argv, struct eh_conf *eh_conf)
1632 int32_t option_index;
1633 char *prgname = argv[0];
1634 int32_t f_present = 0;
1638 while ((opt = getopt_long(argc, argvopt, "aelp:Pu:f:j:w:c:t:s:",
1639 lgopts, &option_index)) != EOF) {
1643 enabled_port_mask = parse_portmask(optarg);
1644 if (enabled_port_mask == 0) {
1645 printf("invalid portmask\n");
1646 print_usage(prgname);
1651 printf("Promiscuous mode selected\n");
1655 unprotected_port_mask = parse_portmask(optarg);
1656 if (unprotected_port_mask == 0) {
1657 printf("invalid unprotected portmask\n");
1658 print_usage(prgname);
1663 if (f_present == 1) {
1664 printf("\"-f\" option present more than "
1666 print_usage(prgname);
1674 ret = parse_decimal(optarg);
1676 printf("Invalid number of buffers in a pool: "
1678 print_usage(prgname);
1682 nb_bufs_in_pool = ret;
1686 ret = parse_decimal(optarg);
1687 if (ret < RTE_MBUF_DEFAULT_BUF_SIZE ||
1689 printf("Invalid frame buffer size value: %s\n",
1691 print_usage(prgname);
1694 frame_buf_size = ret;
1695 printf("Custom frame buffer size %u\n", frame_buf_size);
1698 app_sa_prm.enable = 1;
1701 app_sa_prm.window_size = parse_decimal(optarg);
1704 app_sa_prm.enable_esn = 1;
1707 app_sa_prm.enable = 1;
1708 app_sa_prm.flags |= RTE_IPSEC_SAFLAG_SQN_ATOM;
1711 ret = parse_decimal(optarg);
1713 printf("Invalid SA cache size: %s\n", optarg);
1714 print_usage(prgname);
1717 app_sa_prm.cache_sz = ret;
1720 ret = parse_decimal(optarg);
1722 printf("Invalid interval value: %s\n", optarg);
1723 print_usage(prgname);
1726 stats_interval = ret;
1728 case CMD_LINE_OPT_CONFIG_NUM:
1729 ret = parse_config(optarg);
1731 printf("Invalid config\n");
1732 print_usage(prgname);
1736 case CMD_LINE_OPT_SINGLE_SA_NUM:
1737 ret = parse_decimal(optarg);
1738 if (ret == -1 || ret > UINT32_MAX) {
1739 printf("Invalid argument[sa_idx]\n");
1740 print_usage(prgname);
1746 single_sa_idx = ret;
1747 eh_conf->ipsec_mode = EH_IPSEC_MODE_TYPE_DRIVER;
1748 printf("Configured with single SA index %u\n",
1751 case CMD_LINE_OPT_CRYPTODEV_MASK_NUM:
1752 ret = parse_portmask(optarg);
1754 printf("Invalid argument[portmask]\n");
1755 print_usage(prgname);
1760 enabled_cryptodev_mask = ret;
1763 case CMD_LINE_OPT_TRANSFER_MODE_NUM:
1764 ret = parse_transfer_mode(eh_conf, optarg);
1766 printf("Invalid packet transfer mode\n");
1767 print_usage(prgname);
1772 case CMD_LINE_OPT_SCHEDULE_TYPE_NUM:
1773 ret = parse_schedule_type(eh_conf, optarg);
1775 printf("Invalid queue schedule type\n");
1776 print_usage(prgname);
1781 case CMD_LINE_OPT_RX_OFFLOAD_NUM:
1782 ret = parse_mask(optarg, &dev_rx_offload);
1784 printf("Invalid argument for \'%s\': %s\n",
1785 CMD_LINE_OPT_RX_OFFLOAD, optarg);
1786 print_usage(prgname);
1790 case CMD_LINE_OPT_TX_OFFLOAD_NUM:
1791 ret = parse_mask(optarg, &dev_tx_offload);
1793 printf("Invalid argument for \'%s\': %s\n",
1794 CMD_LINE_OPT_TX_OFFLOAD, optarg);
1795 print_usage(prgname);
1799 case CMD_LINE_OPT_REASSEMBLE_NUM:
1800 ret = parse_decimal(optarg);
1801 if (ret < 0 || ret > UINT32_MAX) {
1802 printf("Invalid argument for \'%s\': %s\n",
1803 CMD_LINE_OPT_REASSEMBLE, optarg);
1804 print_usage(prgname);
1809 case CMD_LINE_OPT_MTU_NUM:
1810 ret = parse_decimal(optarg);
1811 if (ret < 0 || ret > RTE_IPV4_MAX_PKT_LEN) {
1812 printf("Invalid argument for \'%s\': %s\n",
1813 CMD_LINE_OPT_MTU, optarg);
1814 print_usage(prgname);
1819 case CMD_LINE_OPT_FRAG_TTL_NUM:
1820 ret = parse_decimal(optarg);
1821 if (ret < 0 || ret > MAX_FRAG_TTL_NS) {
1822 printf("Invalid argument for \'%s\': %s\n",
1823 CMD_LINE_OPT_MTU, optarg);
1824 print_usage(prgname);
1830 print_usage(prgname);
1835 if (f_present == 0) {
1836 printf("Mandatory option \"-f\" not present\n");
1840 /* check do we need to enable multi-seg support */
1841 if (multi_seg_required()) {
1842 /* legacy mode doesn't support multi-seg */
1843 app_sa_prm.enable = 1;
1844 printf("frame buf size: %u, mtu: %u, "
1845 "number of reassemble entries: %u\n"
1846 "multi-segment support is required\n",
1847 frame_buf_size, mtu_size, frag_tbl_sz);
1850 print_app_sa_prm(&app_sa_prm);
1853 argv[optind-1] = prgname;
1856 optind = 1; /* reset getopt lib */
1861 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
1863 char buf[RTE_ETHER_ADDR_FMT_SIZE];
1864 rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
1865 printf("%s%s", name, buf);
1869 * Update destination ethaddr for the port.
1872 add_dst_ethaddr(uint16_t port, const struct rte_ether_addr *addr)
1874 if (port >= RTE_DIM(ethaddr_tbl))
1877 ethaddr_tbl[port].dst = ETHADDR_TO_UINT64(addr);
1881 /* Check the link status of all ports in up to 9s, and print them finally */
1883 check_all_ports_link_status(uint32_t port_mask)
1885 #define CHECK_INTERVAL 100 /* 100ms */
1886 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1888 uint8_t count, all_ports_up, print_flag = 0;
1889 struct rte_eth_link link;
1891 char link_status_text[RTE_ETH_LINK_MAX_STR_LEN];
1893 printf("\nChecking link status");
1895 for (count = 0; count <= MAX_CHECK_TIME; count++) {
1897 RTE_ETH_FOREACH_DEV(portid) {
1898 if ((port_mask & (1 << portid)) == 0)
1900 memset(&link, 0, sizeof(link));
1901 ret = rte_eth_link_get_nowait(portid, &link);
1904 if (print_flag == 1)
1905 printf("Port %u link get failed: %s\n",
1906 portid, rte_strerror(-ret));
1909 /* print link status if flag set */
1910 if (print_flag == 1) {
1911 rte_eth_link_to_str(link_status_text,
1912 sizeof(link_status_text), &link);
1913 printf("Port %d %s\n", portid,
1917 /* clear all_ports_up flag if any link down */
1918 if (link.link_status == RTE_ETH_LINK_DOWN) {
1923 /* after finally printing all link status, get out */
1924 if (print_flag == 1)
1927 if (all_ports_up == 0) {
1930 rte_delay_ms(CHECK_INTERVAL);
1933 /* set the print_flag if all ports up or timeout */
1934 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1942 add_mapping(struct rte_hash *map, const char *str, uint16_t cdev_id,
1943 uint16_t qp, struct lcore_params *params,
1944 struct ipsec_ctx *ipsec_ctx,
1945 const struct rte_cryptodev_capabilities *cipher,
1946 const struct rte_cryptodev_capabilities *auth,
1947 const struct rte_cryptodev_capabilities *aead)
1951 struct cdev_key key = { 0 };
1953 key.lcore_id = params->lcore_id;
1955 key.cipher_algo = cipher->sym.cipher.algo;
1957 key.auth_algo = auth->sym.auth.algo;
1959 key.aead_algo = aead->sym.aead.algo;
1961 ret = rte_hash_lookup(map, &key);
1965 for (i = 0; i < ipsec_ctx->nb_qps; i++)
1966 if (ipsec_ctx->tbl[i].id == cdev_id)
1969 if (i == ipsec_ctx->nb_qps) {
1970 if (ipsec_ctx->nb_qps == MAX_QP_PER_LCORE) {
1971 printf("Maximum number of crypto devices assigned to "
1972 "a core, increase MAX_QP_PER_LCORE value\n");
1975 ipsec_ctx->tbl[i].id = cdev_id;
1976 ipsec_ctx->tbl[i].qp = qp;
1977 ipsec_ctx->nb_qps++;
1978 printf("%s cdev mapping: lcore %u using cdev %u qp %u "
1979 "(cdev_id_qp %lu)\n", str, key.lcore_id,
1983 ret = rte_hash_add_key_data(map, &key, (void *)i);
1985 printf("Faled to insert cdev mapping for (lcore %u, "
1986 "cdev %u, qp %u), errno %d\n",
1987 key.lcore_id, ipsec_ctx->tbl[i].id,
1988 ipsec_ctx->tbl[i].qp, ret);
1996 add_cdev_mapping(struct rte_cryptodev_info *dev_info, uint16_t cdev_id,
1997 uint16_t qp, struct lcore_params *params)
2000 const struct rte_cryptodev_capabilities *i, *j;
2001 struct rte_hash *map;
2002 struct lcore_conf *qconf;
2003 struct ipsec_ctx *ipsec_ctx;
2006 qconf = &lcore_conf[params->lcore_id];
2008 if ((unprotected_port_mask & (1 << params->port_id)) == 0) {
2010 ipsec_ctx = &qconf->outbound;
2014 ipsec_ctx = &qconf->inbound;
2018 /* Required cryptodevs with operation chainning */
2019 if (!(dev_info->feature_flags &
2020 RTE_CRYPTODEV_FF_SYM_OPERATION_CHAINING))
2023 for (i = dev_info->capabilities;
2024 i->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; i++) {
2025 if (i->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
2028 if (i->sym.xform_type == RTE_CRYPTO_SYM_XFORM_AEAD) {
2029 ret |= add_mapping(map, str, cdev_id, qp, params,
2030 ipsec_ctx, NULL, NULL, i);
2034 if (i->sym.xform_type != RTE_CRYPTO_SYM_XFORM_CIPHER)
2037 for (j = dev_info->capabilities;
2038 j->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; j++) {
2039 if (j->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
2042 if (j->sym.xform_type != RTE_CRYPTO_SYM_XFORM_AUTH)
2045 ret |= add_mapping(map, str, cdev_id, qp, params,
2046 ipsec_ctx, i, j, NULL);
2053 /* Check if the device is enabled by cryptodev_mask */
2055 check_cryptodev_mask(uint8_t cdev_id)
2057 if (enabled_cryptodev_mask & (1 << cdev_id))
2064 cryptodevs_init(uint16_t req_queue_num)
2066 struct rte_cryptodev_config dev_conf;
2067 struct rte_cryptodev_qp_conf qp_conf;
2068 uint16_t idx, max_nb_qps, qp, total_nb_qps, i;
2070 struct rte_hash_parameters params = { 0 };
2072 const uint64_t mseg_flag = multi_seg_required() ?
2073 RTE_CRYPTODEV_FF_IN_PLACE_SGL : 0;
2075 params.entries = CDEV_MAP_ENTRIES;
2076 params.key_len = sizeof(struct cdev_key);
2077 params.hash_func = rte_jhash;
2078 params.hash_func_init_val = 0;
2079 params.socket_id = rte_socket_id();
2081 params.name = "cdev_map_in";
2082 cdev_map_in = rte_hash_create(¶ms);
2083 if (cdev_map_in == NULL)
2084 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
2087 params.name = "cdev_map_out";
2088 cdev_map_out = rte_hash_create(¶ms);
2089 if (cdev_map_out == NULL)
2090 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
2093 printf("lcore/cryptodev/qp mappings:\n");
2097 for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
2098 struct rte_cryptodev_info cdev_info;
2100 if (check_cryptodev_mask((uint8_t)cdev_id))
2103 rte_cryptodev_info_get(cdev_id, &cdev_info);
2105 if ((mseg_flag & cdev_info.feature_flags) != mseg_flag)
2106 rte_exit(EXIT_FAILURE,
2107 "Device %hd does not support \'%s\' feature\n",
2109 rte_cryptodev_get_feature_name(mseg_flag));
2111 if (nb_lcore_params > cdev_info.max_nb_queue_pairs)
2112 max_nb_qps = cdev_info.max_nb_queue_pairs;
2114 max_nb_qps = nb_lcore_params;
2118 while (qp < max_nb_qps && i < nb_lcore_params) {
2119 if (add_cdev_mapping(&cdev_info, cdev_id, qp,
2120 &lcore_params[idx]))
2123 idx = idx % nb_lcore_params;
2127 qp = RTE_MIN(max_nb_qps, RTE_MAX(req_queue_num, qp));
2132 dev_conf.socket_id = rte_cryptodev_socket_id(cdev_id);
2133 dev_conf.nb_queue_pairs = qp;
2134 dev_conf.ff_disable = RTE_CRYPTODEV_FF_ASYMMETRIC_CRYPTO;
2136 uint32_t dev_max_sess = cdev_info.sym.max_nb_sessions;
2137 if (dev_max_sess != 0 &&
2138 dev_max_sess < get_nb_crypto_sessions())
2139 rte_exit(EXIT_FAILURE,
2140 "Device does not support at least %u "
2141 "sessions", get_nb_crypto_sessions());
2143 if (rte_cryptodev_configure(cdev_id, &dev_conf))
2144 rte_panic("Failed to initialize cryptodev %u\n",
2147 qp_conf.nb_descriptors = CDEV_QUEUE_DESC;
2148 qp_conf.mp_session =
2149 socket_ctx[dev_conf.socket_id].session_pool;
2150 qp_conf.mp_session_private =
2151 socket_ctx[dev_conf.socket_id].session_priv_pool;
2152 for (qp = 0; qp < dev_conf.nb_queue_pairs; qp++)
2153 if (rte_cryptodev_queue_pair_setup(cdev_id, qp,
2154 &qp_conf, dev_conf.socket_id))
2155 rte_panic("Failed to setup queue %u for "
2156 "cdev_id %u\n", 0, cdev_id);
2158 if (rte_cryptodev_start(cdev_id))
2159 rte_panic("Failed to start cryptodev %u\n",
2165 return total_nb_qps;
2169 port_init(uint16_t portid, uint64_t req_rx_offloads, uint64_t req_tx_offloads)
2171 struct rte_eth_dev_info dev_info;
2172 struct rte_eth_txconf *txconf;
2173 uint16_t nb_tx_queue, nb_rx_queue;
2174 uint16_t tx_queueid, rx_queueid, queue, lcore_id;
2175 int32_t ret, socket_id;
2176 struct lcore_conf *qconf;
2177 struct rte_ether_addr ethaddr;
2178 struct rte_eth_conf local_port_conf = port_conf;
2180 ret = rte_eth_dev_info_get(portid, &dev_info);
2182 rte_exit(EXIT_FAILURE,
2183 "Error during getting device (port %u) info: %s\n",
2184 portid, strerror(-ret));
2186 /* limit allowed HW offloafs, as user requested */
2187 dev_info.rx_offload_capa &= dev_rx_offload;
2188 dev_info.tx_offload_capa &= dev_tx_offload;
2190 printf("Configuring device port %u:\n", portid);
2192 ret = rte_eth_macaddr_get(portid, ðaddr);
2194 rte_exit(EXIT_FAILURE,
2195 "Error getting MAC address (port %u): %s\n",
2196 portid, rte_strerror(-ret));
2198 ethaddr_tbl[portid].src = ETHADDR_TO_UINT64(ðaddr);
2199 print_ethaddr("Address: ", ðaddr);
2202 nb_rx_queue = get_port_nb_rx_queues(portid);
2203 nb_tx_queue = nb_lcores;
2205 if (nb_rx_queue > dev_info.max_rx_queues)
2206 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
2207 "(max rx queue is %u)\n",
2208 nb_rx_queue, dev_info.max_rx_queues);
2210 if (nb_tx_queue > dev_info.max_tx_queues)
2211 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
2212 "(max tx queue is %u)\n",
2213 nb_tx_queue, dev_info.max_tx_queues);
2215 printf("Creating queues: nb_rx_queue=%d nb_tx_queue=%u...\n",
2216 nb_rx_queue, nb_tx_queue);
2218 local_port_conf.rxmode.mtu = mtu_size;
2220 if (multi_seg_required()) {
2221 local_port_conf.rxmode.offloads |= RTE_ETH_RX_OFFLOAD_SCATTER;
2222 local_port_conf.txmode.offloads |= RTE_ETH_TX_OFFLOAD_MULTI_SEGS;
2225 local_port_conf.rxmode.offloads |= req_rx_offloads;
2226 local_port_conf.txmode.offloads |= req_tx_offloads;
2228 /* Check that all required capabilities are supported */
2229 if ((local_port_conf.rxmode.offloads & dev_info.rx_offload_capa) !=
2230 local_port_conf.rxmode.offloads)
2231 rte_exit(EXIT_FAILURE,
2232 "Error: port %u required RX offloads: 0x%" PRIx64
2233 ", avaialbe RX offloads: 0x%" PRIx64 "\n",
2234 portid, local_port_conf.rxmode.offloads,
2235 dev_info.rx_offload_capa);
2237 if ((local_port_conf.txmode.offloads & dev_info.tx_offload_capa) !=
2238 local_port_conf.txmode.offloads)
2239 rte_exit(EXIT_FAILURE,
2240 "Error: port %u required TX offloads: 0x%" PRIx64
2241 ", avaialbe TX offloads: 0x%" PRIx64 "\n",
2242 portid, local_port_conf.txmode.offloads,
2243 dev_info.tx_offload_capa);
2245 if (dev_info.tx_offload_capa & RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE)
2246 local_port_conf.txmode.offloads |=
2247 RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE;
2249 if (dev_info.tx_offload_capa & RTE_ETH_TX_OFFLOAD_IPV4_CKSUM)
2250 local_port_conf.txmode.offloads |= RTE_ETH_TX_OFFLOAD_IPV4_CKSUM;
2252 printf("port %u configurng rx_offloads=0x%" PRIx64
2253 ", tx_offloads=0x%" PRIx64 "\n",
2254 portid, local_port_conf.rxmode.offloads,
2255 local_port_conf.txmode.offloads);
2257 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
2258 dev_info.flow_type_rss_offloads;
2259 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
2260 port_conf.rx_adv_conf.rss_conf.rss_hf) {
2261 printf("Port %u modified RSS hash function based on hardware support,"
2262 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
2264 port_conf.rx_adv_conf.rss_conf.rss_hf,
2265 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
2268 ret = rte_eth_dev_configure(portid, nb_rx_queue, nb_tx_queue,
2271 rte_exit(EXIT_FAILURE, "Cannot configure device: "
2272 "err=%d, port=%d\n", ret, portid);
2274 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd, &nb_txd);
2276 rte_exit(EXIT_FAILURE, "Cannot adjust number of descriptors: "
2277 "err=%d, port=%d\n", ret, portid);
2279 /* init one TX queue per lcore */
2281 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2282 if (rte_lcore_is_enabled(lcore_id) == 0)
2286 socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2291 printf("Setup txq=%u,%d,%d\n", lcore_id, tx_queueid, socket_id);
2293 txconf = &dev_info.default_txconf;
2294 txconf->offloads = local_port_conf.txmode.offloads;
2296 ret = rte_eth_tx_queue_setup(portid, tx_queueid, nb_txd,
2299 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
2300 "err=%d, port=%d\n", ret, portid);
2302 qconf = &lcore_conf[lcore_id];
2303 qconf->tx_queue_id[portid] = tx_queueid;
2305 /* Pre-populate pkt offloads based on capabilities */
2306 qconf->outbound.ipv4_offloads = RTE_MBUF_F_TX_IPV4;
2307 qconf->outbound.ipv6_offloads = RTE_MBUF_F_TX_IPV6;
2308 if (local_port_conf.txmode.offloads & RTE_ETH_TX_OFFLOAD_IPV4_CKSUM)
2309 qconf->outbound.ipv4_offloads |= RTE_MBUF_F_TX_IP_CKSUM;
2313 /* init RX queues */
2314 for (queue = 0; queue < qconf->nb_rx_queue; ++queue) {
2315 struct rte_eth_rxconf rxq_conf;
2317 if (portid != qconf->rx_queue_list[queue].port_id)
2320 rx_queueid = qconf->rx_queue_list[queue].queue_id;
2322 printf("Setup rxq=%d,%d,%d\n", portid, rx_queueid,
2325 rxq_conf = dev_info.default_rxconf;
2326 rxq_conf.offloads = local_port_conf.rxmode.offloads;
2327 ret = rte_eth_rx_queue_setup(portid, rx_queueid,
2328 nb_rxd, socket_id, &rxq_conf,
2329 socket_ctx[socket_id].mbuf_pool);
2331 rte_exit(EXIT_FAILURE,
2332 "rte_eth_rx_queue_setup: err=%d, "
2333 "port=%d\n", ret, portid);
2340 max_session_size(void)
2344 int16_t cdev_id, port_id, n;
2347 n = rte_cryptodev_count();
2348 for (cdev_id = 0; cdev_id != n; cdev_id++) {
2349 sz = rte_cryptodev_sym_get_private_session_size(cdev_id);
2353 * If crypto device is security capable, need to check the
2354 * size of security session as well.
2357 /* Get security context of the crypto device */
2358 sec_ctx = rte_cryptodev_get_sec_ctx(cdev_id);
2359 if (sec_ctx == NULL)
2362 /* Get size of security session */
2363 sz = rte_security_session_get_size(sec_ctx);
2368 RTE_ETH_FOREACH_DEV(port_id) {
2369 if ((enabled_port_mask & (1 << port_id)) == 0)
2372 sec_ctx = rte_eth_dev_get_sec_ctx(port_id);
2373 if (sec_ctx == NULL)
2376 sz = rte_security_session_get_size(sec_ctx);
2385 session_pool_init(struct socket_ctx *ctx, int32_t socket_id, size_t sess_sz)
2387 char mp_name[RTE_MEMPOOL_NAMESIZE];
2388 struct rte_mempool *sess_mp;
2391 snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
2392 "sess_mp_%u", socket_id);
2393 nb_sess = (get_nb_crypto_sessions() + CDEV_MP_CACHE_SZ *
2395 nb_sess = RTE_MAX(nb_sess, CDEV_MP_CACHE_SZ *
2396 CDEV_MP_CACHE_MULTIPLIER);
2397 sess_mp = rte_cryptodev_sym_session_pool_create(
2398 mp_name, nb_sess, sess_sz, CDEV_MP_CACHE_SZ, 0,
2400 ctx->session_pool = sess_mp;
2402 if (ctx->session_pool == NULL)
2403 rte_exit(EXIT_FAILURE,
2404 "Cannot init session pool on socket %d\n", socket_id);
2406 printf("Allocated session pool on socket %d\n", socket_id);
2410 session_priv_pool_init(struct socket_ctx *ctx, int32_t socket_id,
2413 char mp_name[RTE_MEMPOOL_NAMESIZE];
2414 struct rte_mempool *sess_mp;
2417 snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
2418 "sess_mp_priv_%u", socket_id);
2419 nb_sess = (get_nb_crypto_sessions() + CDEV_MP_CACHE_SZ *
2421 nb_sess = RTE_MAX(nb_sess, CDEV_MP_CACHE_SZ *
2422 CDEV_MP_CACHE_MULTIPLIER);
2423 sess_mp = rte_mempool_create(mp_name,
2427 0, NULL, NULL, NULL,
2430 ctx->session_priv_pool = sess_mp;
2432 if (ctx->session_priv_pool == NULL)
2433 rte_exit(EXIT_FAILURE,
2434 "Cannot init session priv pool on socket %d\n",
2437 printf("Allocated session priv pool on socket %d\n",
2442 pool_init(struct socket_ctx *ctx, int32_t socket_id, uint32_t nb_mbuf)
2447 snprintf(s, sizeof(s), "mbuf_pool_%d", socket_id);
2448 ctx->mbuf_pool = rte_pktmbuf_pool_create(s, nb_mbuf,
2449 MEMPOOL_CACHE_SIZE, ipsec_metadata_size(),
2450 frame_buf_size, socket_id);
2453 * if multi-segment support is enabled, then create a pool
2454 * for indirect mbufs.
2456 ms = multi_seg_required();
2458 snprintf(s, sizeof(s), "mbuf_pool_indir_%d", socket_id);
2459 ctx->mbuf_pool_indir = rte_pktmbuf_pool_create(s, nb_mbuf,
2460 MEMPOOL_CACHE_SIZE, 0, 0, socket_id);
2463 if (ctx->mbuf_pool == NULL || (ms != 0 && ctx->mbuf_pool_indir == NULL))
2464 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool on socket %d\n",
2467 printf("Allocated mbuf pool on socket %d\n", socket_id);
2471 inline_ipsec_event_esn_overflow(struct rte_security_ctx *ctx, uint64_t md)
2473 struct ipsec_sa *sa;
2475 /* For inline protocol processing, the metadata in the event will
2476 * uniquely identify the security session which raised the event.
2477 * Application would then need the userdata it had registered with the
2478 * security session to process the event.
2481 sa = (struct ipsec_sa *)rte_security_get_userdata(ctx, md);
2484 /* userdata could not be retrieved */
2488 /* Sequence number over flow. SA need to be re-established */
2494 inline_ipsec_event_callback(uint16_t port_id, enum rte_eth_event_type type,
2495 void *param, void *ret_param)
2498 struct rte_eth_event_ipsec_desc *event_desc = NULL;
2499 struct rte_security_ctx *ctx = (struct rte_security_ctx *)
2500 rte_eth_dev_get_sec_ctx(port_id);
2502 RTE_SET_USED(param);
2504 if (type != RTE_ETH_EVENT_IPSEC)
2507 event_desc = ret_param;
2508 if (event_desc == NULL) {
2509 printf("Event descriptor not set\n");
2513 md = event_desc->metadata;
2515 if (event_desc->subtype == RTE_ETH_EVENT_IPSEC_ESN_OVERFLOW)
2516 return inline_ipsec_event_esn_overflow(ctx, md);
2517 else if (event_desc->subtype >= RTE_ETH_EVENT_IPSEC_MAX) {
2518 printf("Invalid IPsec event reported\n");
2526 rx_callback(__rte_unused uint16_t port, __rte_unused uint16_t queue,
2527 struct rte_mbuf *pkt[], uint16_t nb_pkts,
2528 __rte_unused uint16_t max_pkts, void *user_param)
2532 struct lcore_conf *lc;
2533 struct rte_mbuf *mb;
2534 struct rte_ether_hdr *eth;
2540 for (i = 0; i != nb_pkts; i++) {
2543 eth = rte_pktmbuf_mtod(mb, struct rte_ether_hdr *);
2544 if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
2546 struct rte_ipv4_hdr *iph;
2548 iph = (struct rte_ipv4_hdr *)(eth + 1);
2549 if (rte_ipv4_frag_pkt_is_fragmented(iph)) {
2551 mb->l2_len = sizeof(*eth);
2552 mb->l3_len = sizeof(*iph);
2553 tm = (tm != 0) ? tm : rte_rdtsc();
2554 mb = rte_ipv4_frag_reassemble_packet(
2555 lc->frag.tbl, &lc->frag.dr,
2559 /* fix ip cksum after reassemble. */
2560 iph = rte_pktmbuf_mtod_offset(mb,
2561 struct rte_ipv4_hdr *,
2563 iph->hdr_checksum = 0;
2564 iph->hdr_checksum = rte_ipv4_cksum(iph);
2567 } else if (eth->ether_type ==
2568 rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6)) {
2570 struct rte_ipv6_hdr *iph;
2571 struct ipv6_extension_fragment *fh;
2573 iph = (struct rte_ipv6_hdr *)(eth + 1);
2574 fh = rte_ipv6_frag_get_ipv6_fragment_header(iph);
2576 mb->l2_len = sizeof(*eth);
2577 mb->l3_len = (uintptr_t)fh - (uintptr_t)iph +
2579 tm = (tm != 0) ? tm : rte_rdtsc();
2580 mb = rte_ipv6_frag_reassemble_packet(
2581 lc->frag.tbl, &lc->frag.dr,
2584 /* fix l3_len after reassemble. */
2585 mb->l3_len = mb->l3_len - sizeof(*fh);
2593 /* some fragments were encountered, drain death row */
2595 rte_ip_frag_free_death_row(&lc->frag.dr, 0);
2602 reassemble_lcore_init(struct lcore_conf *lc, uint32_t cid)
2606 uint64_t frag_cycles;
2607 const struct lcore_rx_queue *rxq;
2608 const struct rte_eth_rxtx_callback *cb;
2610 /* create fragment table */
2611 sid = rte_lcore_to_socket_id(cid);
2612 frag_cycles = (rte_get_tsc_hz() + NS_PER_S - 1) /
2613 NS_PER_S * frag_ttl_ns;
2615 lc->frag.tbl = rte_ip_frag_table_create(frag_tbl_sz,
2616 FRAG_TBL_BUCKET_ENTRIES, frag_tbl_sz, frag_cycles, sid);
2617 if (lc->frag.tbl == NULL) {
2618 printf("%s(%u): failed to create fragment table of size: %u, "
2620 __func__, cid, frag_tbl_sz, rte_errno);
2624 /* setup reassemble RX callbacks for all queues */
2625 for (i = 0; i != lc->nb_rx_queue; i++) {
2627 rxq = lc->rx_queue_list + i;
2628 cb = rte_eth_add_rx_callback(rxq->port_id, rxq->queue_id,
2631 printf("%s(%u): failed to install RX callback for "
2632 "portid=%u, queueid=%u, error code: %d\n",
2634 rxq->port_id, rxq->queue_id, rte_errno);
2643 reassemble_init(void)
2649 for (i = 0; i != nb_lcore_params; i++) {
2650 lc = lcore_params[i].lcore_id;
2651 rc = reassemble_lcore_init(lcore_conf + lc, lc);
2660 create_default_ipsec_flow(uint16_t port_id, uint64_t rx_offloads)
2662 struct rte_flow_action action[2];
2663 struct rte_flow_item pattern[2];
2664 struct rte_flow_attr attr = {0};
2665 struct rte_flow_error err;
2666 struct rte_flow *flow;
2669 if (!(rx_offloads & RTE_ETH_RX_OFFLOAD_SECURITY))
2672 /* Add the default rte_flow to enable SECURITY for all ESP packets */
2674 pattern[0].type = RTE_FLOW_ITEM_TYPE_ESP;
2675 pattern[0].spec = NULL;
2676 pattern[0].mask = NULL;
2677 pattern[0].last = NULL;
2678 pattern[1].type = RTE_FLOW_ITEM_TYPE_END;
2680 action[0].type = RTE_FLOW_ACTION_TYPE_SECURITY;
2681 action[0].conf = NULL;
2682 action[1].type = RTE_FLOW_ACTION_TYPE_END;
2683 action[1].conf = NULL;
2687 ret = rte_flow_validate(port_id, &attr, pattern, action, &err);
2691 flow = rte_flow_create(port_id, &attr, pattern, action, &err);
2695 flow_info_tbl[port_id].rx_def_flow = flow;
2696 RTE_LOG(INFO, IPSEC,
2697 "Created default flow enabling SECURITY for all ESP traffic on port %d\n",
2702 signal_handler(int signum)
2704 if (signum == SIGINT || signum == SIGTERM) {
2705 printf("\n\nSignal %d received, preparing to exit...\n",
2712 ev_mode_sess_verify(struct ipsec_sa *sa, int nb_sa)
2714 struct rte_ipsec_session *ips;
2720 for (i = 0; i < nb_sa; i++) {
2721 ips = ipsec_get_primary_session(&sa[i]);
2722 if (ips->type != RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL)
2723 rte_exit(EXIT_FAILURE, "Event mode supports only "
2724 "inline protocol sessions\n");
2730 check_event_mode_params(struct eh_conf *eh_conf)
2732 struct eventmode_conf *em_conf = NULL;
2733 struct lcore_params *params;
2736 if (!eh_conf || !eh_conf->mode_params)
2739 /* Get eventmode conf */
2740 em_conf = eh_conf->mode_params;
2742 if (eh_conf->mode == EH_PKT_TRANSFER_MODE_POLL &&
2743 em_conf->ext_params.sched_type != SCHED_TYPE_NOT_SET) {
2744 printf("error: option --event-schedule-type applies only to "
2749 if (eh_conf->mode != EH_PKT_TRANSFER_MODE_EVENT)
2752 /* Set schedule type to ORDERED if it wasn't explicitly set by user */
2753 if (em_conf->ext_params.sched_type == SCHED_TYPE_NOT_SET)
2754 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_ORDERED;
2757 * Event mode currently supports only inline protocol sessions.
2758 * If there are other types of sessions configured then exit with
2761 ev_mode_sess_verify(sa_in, nb_sa_in);
2762 ev_mode_sess_verify(sa_out, nb_sa_out);
2765 /* Option --config does not apply to event mode */
2766 if (nb_lcore_params > 0) {
2767 printf("error: option --config applies only to poll mode\n");
2772 * In order to use the same port_init routine for both poll and event
2773 * modes initialize lcore_params with one queue for each eth port
2775 lcore_params = lcore_params_array;
2776 RTE_ETH_FOREACH_DEV(portid) {
2777 if ((enabled_port_mask & (1 << portid)) == 0)
2780 params = &lcore_params[nb_lcore_params++];
2781 params->port_id = portid;
2782 params->queue_id = 0;
2783 params->lcore_id = rte_get_next_lcore(0, 0, 1);
2790 inline_sessions_free(struct sa_ctx *sa_ctx)
2792 struct rte_ipsec_session *ips;
2793 struct ipsec_sa *sa;
2800 for (i = 0; i < sa_ctx->nb_sa; i++) {
2802 sa = &sa_ctx->sa[i];
2806 ips = ipsec_get_primary_session(sa);
2807 if (ips->type != RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL &&
2808 ips->type != RTE_SECURITY_ACTION_TYPE_INLINE_CRYPTO)
2811 if (!rte_eth_dev_is_valid_port(sa->portid))
2814 ret = rte_security_session_destroy(
2815 rte_eth_dev_get_sec_ctx(sa->portid),
2818 RTE_LOG(ERR, IPSEC, "Failed to destroy security "
2819 "session type %d, spi %d\n",
2820 ips->type, sa->spi);
2825 calculate_nb_mbufs(uint16_t nb_ports, uint16_t nb_crypto_qp, uint32_t nb_rxq,
2828 return RTE_MAX((nb_rxq * nb_rxd +
2829 nb_ports * nb_lcores * MAX_PKT_BURST +
2830 nb_ports * nb_txq * nb_txd +
2831 nb_lcores * MEMPOOL_CACHE_SIZE +
2832 nb_crypto_qp * CDEV_QUEUE_DESC +
2833 nb_lcores * frag_tbl_sz *
2834 FRAG_TBL_BUCKET_ENTRIES),
2839 main(int32_t argc, char **argv)
2842 uint32_t lcore_id, nb_txq, nb_rxq = 0;
2846 uint16_t portid, nb_crypto_qp, nb_ports = 0;
2847 uint64_t req_rx_offloads[RTE_MAX_ETHPORTS];
2848 uint64_t req_tx_offloads[RTE_MAX_ETHPORTS];
2849 struct eh_conf *eh_conf = NULL;
2852 nb_bufs_in_pool = 0;
2855 ret = rte_eal_init(argc, argv);
2857 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
2862 signal(SIGINT, signal_handler);
2863 signal(SIGTERM, signal_handler);
2865 /* initialize event helper configuration */
2866 eh_conf = eh_conf_init();
2867 if (eh_conf == NULL)
2868 rte_exit(EXIT_FAILURE, "Failed to init event helper config");
2870 /* parse application arguments (after the EAL ones) */
2871 ret = parse_args(argc, argv, eh_conf);
2873 rte_exit(EXIT_FAILURE, "Invalid parameters\n");
2875 /* parse configuration file */
2876 if (parse_cfg_file(cfgfile) < 0) {
2877 printf("parsing file \"%s\" failed\n",
2879 print_usage(argv[0]);
2883 if ((unprotected_port_mask & enabled_port_mask) !=
2884 unprotected_port_mask)
2885 rte_exit(EXIT_FAILURE, "Invalid unprotected portmask 0x%x\n",
2886 unprotected_port_mask);
2888 if (check_poll_mode_params(eh_conf) < 0)
2889 rte_exit(EXIT_FAILURE, "check_poll_mode_params failed\n");
2891 if (check_event_mode_params(eh_conf) < 0)
2892 rte_exit(EXIT_FAILURE, "check_event_mode_params failed\n");
2894 ret = init_lcore_rx_queues();
2896 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
2898 nb_lcores = rte_lcore_count();
2900 sess_sz = max_session_size();
2903 * In event mode request minimum number of crypto queues
2904 * to be reserved equal to number of ports.
2906 if (eh_conf->mode == EH_PKT_TRANSFER_MODE_EVENT)
2907 nb_crypto_qp = rte_eth_dev_count_avail();
2911 nb_crypto_qp = cryptodevs_init(nb_crypto_qp);
2913 if (nb_bufs_in_pool == 0) {
2914 RTE_ETH_FOREACH_DEV(portid) {
2915 if ((enabled_port_mask & (1 << portid)) == 0)
2918 nb_rxq += get_port_nb_rx_queues(portid);
2923 nb_bufs_in_pool = calculate_nb_mbufs(nb_ports, nb_crypto_qp,
2927 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2928 if (rte_lcore_is_enabled(lcore_id) == 0)
2932 socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2936 /* mbuf_pool is initialised by the pool_init() function*/
2937 if (socket_ctx[socket_id].mbuf_pool)
2940 pool_init(&socket_ctx[socket_id], socket_id, nb_bufs_in_pool);
2941 session_pool_init(&socket_ctx[socket_id], socket_id, sess_sz);
2942 session_priv_pool_init(&socket_ctx[socket_id], socket_id,
2945 printf("Number of mbufs in packet pool %d\n", nb_bufs_in_pool);
2947 RTE_ETH_FOREACH_DEV(portid) {
2948 if ((enabled_port_mask & (1 << portid)) == 0)
2951 sa_check_offloads(portid, &req_rx_offloads[portid],
2952 &req_tx_offloads[portid]);
2953 port_init(portid, req_rx_offloads[portid],
2954 req_tx_offloads[portid]);
2958 * Set the enabled port mask in helper config for use by helper
2959 * sub-system. This will be used while initializing devices using
2960 * helper sub-system.
2962 eh_conf->eth_portmask = enabled_port_mask;
2964 /* Initialize eventmode components */
2965 ret = eh_devs_init(eh_conf);
2967 rte_exit(EXIT_FAILURE, "eh_devs_init failed, err=%d\n", ret);
2970 RTE_ETH_FOREACH_DEV(portid) {
2971 if ((enabled_port_mask & (1 << portid)) == 0)
2974 /* Create flow before starting the device */
2975 create_default_ipsec_flow(portid, req_rx_offloads[portid]);
2977 ret = rte_eth_dev_start(portid);
2979 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: "
2980 "err=%d, port=%d\n", ret, portid);
2982 * If enabled, put device in promiscuous mode.
2983 * This allows IO forwarding mode to forward packets
2984 * to itself through 2 cross-connected ports of the
2987 if (promiscuous_on) {
2988 ret = rte_eth_promiscuous_enable(portid);
2990 rte_exit(EXIT_FAILURE,
2991 "rte_eth_promiscuous_enable: err=%s, port=%d\n",
2992 rte_strerror(-ret), portid);
2995 rte_eth_dev_callback_register(portid,
2996 RTE_ETH_EVENT_IPSEC, inline_ipsec_event_callback, NULL);
2999 /* fragment reassemble is enabled */
3000 if (frag_tbl_sz != 0) {
3001 ret = reassemble_init();
3003 rte_exit(EXIT_FAILURE, "failed at reassemble init");
3006 /* Replicate each context per socket */
3007 for (i = 0; i < NB_SOCKETS && i < rte_socket_count(); i++) {
3008 socket_id = rte_socket_id_by_idx(i);
3009 if ((socket_ctx[socket_id].mbuf_pool != NULL) &&
3010 (socket_ctx[socket_id].sa_in == NULL) &&
3011 (socket_ctx[socket_id].sa_out == NULL)) {
3012 sa_init(&socket_ctx[socket_id], socket_id);
3013 sp4_init(&socket_ctx[socket_id], socket_id);
3014 sp6_init(&socket_ctx[socket_id], socket_id);
3015 rt_init(&socket_ctx[socket_id], socket_id);
3021 check_all_ports_link_status(enabled_port_mask);
3023 if (stats_interval > 0)
3024 rte_eal_alarm_set(stats_interval * US_PER_S,
3025 print_stats_cb, NULL);
3027 RTE_LOG(INFO, IPSEC, "Stats display disabled\n");
3029 /* launch per-lcore init on every lcore */
3030 rte_eal_mp_remote_launch(ipsec_launch_one_lcore, eh_conf, CALL_MAIN);
3031 RTE_LCORE_FOREACH_WORKER(lcore_id) {
3032 if (rte_eal_wait_lcore(lcore_id) < 0)
3036 /* Uninitialize eventmode components */
3037 ret = eh_devs_uninit(eh_conf);
3039 rte_exit(EXIT_FAILURE, "eh_devs_uninit failed, err=%d\n", ret);
3041 /* Free eventmode configuration memory */
3042 eh_conf_uninit(eh_conf);
3044 /* Destroy inline inbound and outbound sessions */
3045 for (i = 0; i < NB_SOCKETS && i < rte_socket_count(); i++) {
3046 socket_id = rte_socket_id_by_idx(i);
3047 inline_sessions_free(socket_ctx[socket_id].sa_in);
3048 inline_sessions_free(socket_ctx[socket_id].sa_out);
3051 for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
3052 printf("Closing cryptodev %d...", cdev_id);
3053 rte_cryptodev_stop(cdev_id);
3054 rte_cryptodev_close(cdev_id);
3058 RTE_ETH_FOREACH_DEV(portid) {
3059 if ((enabled_port_mask & (1 << portid)) == 0)
3062 printf("Closing port %d...", portid);
3063 if (flow_info_tbl[portid].rx_def_flow) {
3064 struct rte_flow_error err;
3066 ret = rte_flow_destroy(portid,
3067 flow_info_tbl[portid].rx_def_flow, &err);
3069 RTE_LOG(ERR, IPSEC, "Failed to destroy flow "
3070 " for port %u, err msg: %s\n", portid,
3073 ret = rte_eth_dev_stop(portid);
3076 "rte_eth_dev_stop: err=%s, port=%u\n",
3077 rte_strerror(-ret), portid);
3079 rte_eth_dev_close(portid);
3083 /* clean up the EAL */