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 CDEV_MP_CACHE_MULTIPLIER 1.5 /* from rte_mempool.c */
69 #define MAX_QUEUE_PAIRS 1
71 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
73 /* Configure how many packets ahead to prefetch, when reading packets */
74 #define PREFETCH_OFFSET 3
76 #define MAX_RX_QUEUE_PER_LCORE 16
78 #define MAX_LCORE_PARAMS 1024
81 * Configurable number of RX/TX ring descriptors
83 #define IPSEC_SECGW_RX_DESC_DEFAULT 1024
84 #define IPSEC_SECGW_TX_DESC_DEFAULT 1024
85 static uint16_t nb_rxd = IPSEC_SECGW_RX_DESC_DEFAULT;
86 static uint16_t nb_txd = IPSEC_SECGW_TX_DESC_DEFAULT;
88 #define ETHADDR_TO_UINT64(addr) __BYTES_TO_UINT64( \
89 (addr)->addr_bytes[0], (addr)->addr_bytes[1], \
90 (addr)->addr_bytes[2], (addr)->addr_bytes[3], \
91 (addr)->addr_bytes[4], (addr)->addr_bytes[5], \
94 #define FRAG_TBL_BUCKET_ENTRIES 4
95 #define MAX_FRAG_TTL_NS (10LL * NS_PER_S)
97 #define MTU_TO_FRAMELEN(x) ((x) + RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN)
99 struct ethaddr_info ethaddr_tbl[RTE_MAX_ETHPORTS] = {
100 { 0, ETHADDR(0x00, 0x16, 0x3e, 0x7e, 0x94, 0x9a) },
101 { 0, ETHADDR(0x00, 0x16, 0x3e, 0x22, 0xa1, 0xd9) },
102 { 0, ETHADDR(0x00, 0x16, 0x3e, 0x08, 0x69, 0x26) },
103 { 0, ETHADDR(0x00, 0x16, 0x3e, 0x49, 0x9e, 0xdd) }
106 struct flow_info flow_info_tbl[RTE_MAX_ETHPORTS];
108 #define CMD_LINE_OPT_CONFIG "config"
109 #define CMD_LINE_OPT_SINGLE_SA "single-sa"
110 #define CMD_LINE_OPT_CRYPTODEV_MASK "cryptodev_mask"
111 #define CMD_LINE_OPT_TRANSFER_MODE "transfer-mode"
112 #define CMD_LINE_OPT_SCHEDULE_TYPE "event-schedule-type"
113 #define CMD_LINE_OPT_RX_OFFLOAD "rxoffload"
114 #define CMD_LINE_OPT_TX_OFFLOAD "txoffload"
115 #define CMD_LINE_OPT_REASSEMBLE "reassemble"
116 #define CMD_LINE_OPT_MTU "mtu"
117 #define CMD_LINE_OPT_FRAG_TTL "frag-ttl"
119 #define CMD_LINE_ARG_EVENT "event"
120 #define CMD_LINE_ARG_POLL "poll"
121 #define CMD_LINE_ARG_ORDERED "ordered"
122 #define CMD_LINE_ARG_ATOMIC "atomic"
123 #define CMD_LINE_ARG_PARALLEL "parallel"
126 /* long options mapped to a short option */
128 /* first long only option value must be >= 256, so that we won't
129 * conflict with short options
131 CMD_LINE_OPT_MIN_NUM = 256,
132 CMD_LINE_OPT_CONFIG_NUM,
133 CMD_LINE_OPT_SINGLE_SA_NUM,
134 CMD_LINE_OPT_CRYPTODEV_MASK_NUM,
135 CMD_LINE_OPT_TRANSFER_MODE_NUM,
136 CMD_LINE_OPT_SCHEDULE_TYPE_NUM,
137 CMD_LINE_OPT_RX_OFFLOAD_NUM,
138 CMD_LINE_OPT_TX_OFFLOAD_NUM,
139 CMD_LINE_OPT_REASSEMBLE_NUM,
140 CMD_LINE_OPT_MTU_NUM,
141 CMD_LINE_OPT_FRAG_TTL_NUM,
144 static const struct option lgopts[] = {
145 {CMD_LINE_OPT_CONFIG, 1, 0, CMD_LINE_OPT_CONFIG_NUM},
146 {CMD_LINE_OPT_SINGLE_SA, 1, 0, CMD_LINE_OPT_SINGLE_SA_NUM},
147 {CMD_LINE_OPT_CRYPTODEV_MASK, 1, 0, CMD_LINE_OPT_CRYPTODEV_MASK_NUM},
148 {CMD_LINE_OPT_TRANSFER_MODE, 1, 0, CMD_LINE_OPT_TRANSFER_MODE_NUM},
149 {CMD_LINE_OPT_SCHEDULE_TYPE, 1, 0, CMD_LINE_OPT_SCHEDULE_TYPE_NUM},
150 {CMD_LINE_OPT_RX_OFFLOAD, 1, 0, CMD_LINE_OPT_RX_OFFLOAD_NUM},
151 {CMD_LINE_OPT_TX_OFFLOAD, 1, 0, CMD_LINE_OPT_TX_OFFLOAD_NUM},
152 {CMD_LINE_OPT_REASSEMBLE, 1, 0, CMD_LINE_OPT_REASSEMBLE_NUM},
153 {CMD_LINE_OPT_MTU, 1, 0, CMD_LINE_OPT_MTU_NUM},
154 {CMD_LINE_OPT_FRAG_TTL, 1, 0, CMD_LINE_OPT_FRAG_TTL_NUM},
158 uint32_t unprotected_port_mask;
159 uint32_t single_sa_idx;
160 /* mask of enabled ports */
161 static uint32_t enabled_port_mask;
162 static uint64_t enabled_cryptodev_mask = UINT64_MAX;
163 static int32_t promiscuous_on = 1;
164 static int32_t numa_on = 1; /**< NUMA is enabled by default. */
165 static uint32_t nb_lcores;
166 static uint32_t single_sa;
167 static uint32_t nb_bufs_in_pool;
170 * RX/TX HW offload capabilities to enable/use on ethernet ports.
171 * By default all capabilities are enabled.
173 static uint64_t dev_rx_offload = UINT64_MAX;
174 static uint64_t dev_tx_offload = UINT64_MAX;
177 * global values that determine multi-seg policy
179 static uint32_t frag_tbl_sz;
180 static uint32_t frame_buf_size = RTE_MBUF_DEFAULT_BUF_SIZE;
181 static uint32_t mtu_size = RTE_ETHER_MTU;
182 static uint64_t frag_ttl_ns = MAX_FRAG_TTL_NS;
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 = ETH_MQ_RX_RSS,
238 .max_rx_pkt_len = RTE_ETHER_MAX_LEN,
240 .offloads = DEV_RX_OFFLOAD_CHECKSUM,
245 .rss_hf = ETH_RSS_IP | ETH_RSS_UDP |
246 ETH_RSS_TCP | ETH_RSS_SCTP,
250 .mq_mode = ETH_MQ_TX_NONE,
254 struct socket_ctx socket_ctx[NB_SOCKETS];
257 * Determine is multi-segment support required:
258 * - either frame buffer size is smaller then mtu
259 * - or reassmeble support is requested
262 multi_seg_required(void)
264 return (MTU_TO_FRAMELEN(mtu_size) + RTE_PKTMBUF_HEADROOM >
265 frame_buf_size || frag_tbl_sz != 0);
269 adjust_ipv4_pktlen(struct rte_mbuf *m, const struct rte_ipv4_hdr *iph,
274 plen = rte_be_to_cpu_16(iph->total_length) + l2_len;
275 if (plen < m->pkt_len) {
276 trim = m->pkt_len - plen;
277 rte_pktmbuf_trim(m, trim);
282 adjust_ipv6_pktlen(struct rte_mbuf *m, const struct rte_ipv6_hdr *iph,
287 plen = rte_be_to_cpu_16(iph->payload_len) + sizeof(*iph) + l2_len;
288 if (plen < m->pkt_len) {
289 trim = m->pkt_len - plen;
290 rte_pktmbuf_trim(m, trim);
294 #if (STATS_INTERVAL > 0)
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);
356 #endif /* STATS_INTERVAL */
359 prepare_one_packet(struct rte_mbuf *pkt, struct ipsec_traffic *t)
361 const struct rte_ether_hdr *eth;
362 const struct rte_ipv4_hdr *iph4;
363 const struct rte_ipv6_hdr *iph6;
364 const struct rte_udp_hdr *udp;
365 uint16_t ip4_hdr_len;
368 eth = rte_pktmbuf_mtod(pkt, const struct rte_ether_hdr *);
369 if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
371 iph4 = (const struct rte_ipv4_hdr *)rte_pktmbuf_adj(pkt,
373 adjust_ipv4_pktlen(pkt, iph4, 0);
375 switch (iph4->next_proto_id) {
377 t->ipsec.pkts[(t->ipsec.num)++] = pkt;
380 if (app_sa_prm.udp_encap == 1) {
381 ip4_hdr_len = ((iph4->version_ihl &
382 RTE_IPV4_HDR_IHL_MASK) *
383 RTE_IPV4_IHL_MULTIPLIER);
384 udp = rte_pktmbuf_mtod_offset(pkt,
385 struct rte_udp_hdr *, ip4_hdr_len);
386 nat_port = rte_cpu_to_be_16(IPSEC_NAT_T_PORT);
387 if (udp->src_port == nat_port ||
388 udp->dst_port == nat_port){
389 t->ipsec.pkts[(t->ipsec.num)++] = pkt;
391 MBUF_PTYPE_TUNNEL_ESP_IN_UDP;
397 t->ip4.data[t->ip4.num] = &iph4->next_proto_id;
398 t->ip4.pkts[(t->ip4.num)++] = pkt;
401 pkt->l3_len = sizeof(*iph4);
402 pkt->packet_type |= RTE_PTYPE_L3_IPV4;
403 } else if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6)) {
405 size_t l3len, ext_len;
408 /* get protocol type */
409 iph6 = (const struct rte_ipv6_hdr *)rte_pktmbuf_adj(pkt,
411 adjust_ipv6_pktlen(pkt, iph6, 0);
413 next_proto = iph6->proto;
415 /* determine l3 header size up to ESP extension */
416 l3len = sizeof(struct ip6_hdr);
417 p = rte_pktmbuf_mtod(pkt, uint8_t *);
418 while (next_proto != IPPROTO_ESP && l3len < pkt->data_len &&
419 (next_proto = rte_ipv6_get_next_ext(p + l3len,
420 next_proto, &ext_len)) >= 0)
423 /* drop packet when IPv6 header exceeds first segment length */
424 if (unlikely(l3len > pkt->data_len)) {
429 switch (next_proto) {
431 t->ipsec.pkts[(t->ipsec.num)++] = pkt;
434 if (app_sa_prm.udp_encap == 1) {
435 udp = rte_pktmbuf_mtod_offset(pkt,
436 struct rte_udp_hdr *, l3len);
437 nat_port = rte_cpu_to_be_16(IPSEC_NAT_T_PORT);
438 if (udp->src_port == nat_port ||
439 udp->dst_port == nat_port){
440 t->ipsec.pkts[(t->ipsec.num)++] = pkt;
442 MBUF_PTYPE_TUNNEL_ESP_IN_UDP;
448 t->ip6.data[t->ip6.num] = &iph6->proto;
449 t->ip6.pkts[(t->ip6.num)++] = pkt;
453 pkt->packet_type |= RTE_PTYPE_L3_IPV6;
455 /* Unknown/Unsupported type, drop the packet */
456 RTE_LOG(ERR, IPSEC, "Unsupported packet type 0x%x\n",
457 rte_be_to_cpu_16(eth->ether_type));
462 /* Check if the packet has been processed inline. For inline protocol
463 * processed packets, the metadata in the mbuf can be used to identify
464 * the security processing done on the packet. The metadata will be
465 * used to retrieve the application registered userdata associated
466 * with the security session.
469 if (pkt->ol_flags & PKT_RX_SEC_OFFLOAD &&
470 rte_security_dynfield_is_registered()) {
472 struct ipsec_mbuf_metadata *priv;
473 struct rte_security_ctx *ctx = (struct rte_security_ctx *)
474 rte_eth_dev_get_sec_ctx(
477 /* Retrieve the userdata registered. Here, the userdata
478 * registered is the SA pointer.
480 sa = (struct ipsec_sa *)rte_security_get_userdata(ctx,
481 *rte_security_dynfield(pkt));
483 /* userdata could not be retrieved */
487 /* Save SA as priv member in mbuf. This will be used in the
488 * IPsec selector(SP-SA) check.
491 priv = get_priv(pkt);
497 prepare_traffic(struct rte_mbuf **pkts, struct ipsec_traffic *t,
506 for (i = 0; i < (nb_pkts - PREFETCH_OFFSET); i++) {
507 rte_prefetch0(rte_pktmbuf_mtod(pkts[i + PREFETCH_OFFSET],
509 prepare_one_packet(pkts[i], t);
511 /* Process left packets */
512 for (; i < nb_pkts; i++)
513 prepare_one_packet(pkts[i], t);
517 prepare_tx_pkt(struct rte_mbuf *pkt, uint16_t port,
518 const struct lcore_conf *qconf)
521 struct rte_ether_hdr *ethhdr;
523 ip = rte_pktmbuf_mtod(pkt, struct ip *);
525 ethhdr = (struct rte_ether_hdr *)
526 rte_pktmbuf_prepend(pkt, RTE_ETHER_HDR_LEN);
528 if (ip->ip_v == IPVERSION) {
529 pkt->ol_flags |= qconf->outbound.ipv4_offloads;
530 pkt->l3_len = sizeof(struct ip);
531 pkt->l2_len = RTE_ETHER_HDR_LEN;
535 /* calculate IPv4 cksum in SW */
536 if ((pkt->ol_flags & PKT_TX_IP_CKSUM) == 0)
537 ip->ip_sum = rte_ipv4_cksum((struct rte_ipv4_hdr *)ip);
539 ethhdr->ether_type = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4);
541 pkt->ol_flags |= qconf->outbound.ipv6_offloads;
542 pkt->l3_len = sizeof(struct ip6_hdr);
543 pkt->l2_len = RTE_ETHER_HDR_LEN;
545 ethhdr->ether_type = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6);
548 memcpy(ðhdr->s_addr, ðaddr_tbl[port].src,
549 sizeof(struct rte_ether_addr));
550 memcpy(ðhdr->d_addr, ðaddr_tbl[port].dst,
551 sizeof(struct rte_ether_addr));
555 prepare_tx_burst(struct rte_mbuf *pkts[], uint16_t nb_pkts, uint16_t port,
556 const struct lcore_conf *qconf)
559 const int32_t prefetch_offset = 2;
561 for (i = 0; i < (nb_pkts - prefetch_offset); i++) {
562 rte_mbuf_prefetch_part2(pkts[i + prefetch_offset]);
563 prepare_tx_pkt(pkts[i], port, qconf);
565 /* Process left packets */
566 for (; i < nb_pkts; i++)
567 prepare_tx_pkt(pkts[i], port, qconf);
570 /* Send burst of packets on an output interface */
571 static inline int32_t
572 send_burst(struct lcore_conf *qconf, uint16_t n, uint16_t port)
574 struct rte_mbuf **m_table;
578 queueid = qconf->tx_queue_id[port];
579 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
581 prepare_tx_burst(m_table, n, port, qconf);
583 ret = rte_eth_tx_burst(port, queueid, m_table, n);
585 core_stats_update_tx(ret);
587 if (unlikely(ret < n)) {
589 free_pkts(&m_table[ret], 1);
597 * Helper function to fragment and queue for TX one packet.
599 static inline uint32_t
600 send_fragment_packet(struct lcore_conf *qconf, struct rte_mbuf *m,
601 uint16_t port, uint8_t proto)
607 tbl = qconf->tx_mbufs + port;
610 /* free space for new fragments */
611 if (len + RTE_LIBRTE_IP_FRAG_MAX_FRAG >= RTE_DIM(tbl->m_table)) {
612 send_burst(qconf, len, port);
616 n = RTE_DIM(tbl->m_table) - len;
618 if (proto == IPPROTO_IP)
619 rc = rte_ipv4_fragment_packet(m, tbl->m_table + len,
620 n, mtu_size, qconf->frag.pool_dir,
621 qconf->frag.pool_indir);
623 rc = rte_ipv6_fragment_packet(m, tbl->m_table + len,
624 n, mtu_size, qconf->frag.pool_dir,
625 qconf->frag.pool_indir);
631 "%s: failed to fragment packet with size %u, "
633 __func__, m->pkt_len, rte_errno);
639 /* Enqueue a single packet, and send burst if queue is filled */
640 static inline int32_t
641 send_single_packet(struct rte_mbuf *m, uint16_t port, uint8_t proto)
645 struct lcore_conf *qconf;
647 lcore_id = rte_lcore_id();
649 qconf = &lcore_conf[lcore_id];
650 len = qconf->tx_mbufs[port].len;
652 if (m->pkt_len <= mtu_size) {
653 qconf->tx_mbufs[port].m_table[len] = m;
656 /* need to fragment the packet */
657 } else if (frag_tbl_sz > 0)
658 len = send_fragment_packet(qconf, m, port, proto);
662 /* enough pkts to be sent */
663 if (unlikely(len == MAX_PKT_BURST)) {
664 send_burst(qconf, MAX_PKT_BURST, port);
668 qconf->tx_mbufs[port].len = len;
673 inbound_sp_sa(struct sp_ctx *sp, struct sa_ctx *sa, struct traffic_type *ip,
677 uint32_t i, j, res, sa_idx;
679 if (ip->num == 0 || sp == NULL)
682 rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
683 ip->num, DEFAULT_MAX_CATEGORIES);
686 for (i = 0; i < ip->num; i++) {
693 if (res == DISCARD) {
698 /* Only check SPI match for processed IPSec packets */
699 if (i < lim && ((m->ol_flags & PKT_RX_SEC_OFFLOAD) == 0)) {
705 if (!inbound_sa_check(sa, m, sa_idx)) {
715 split46_traffic(struct ipsec_traffic *trf, struct rte_mbuf *mb[], uint32_t num)
724 for (i = 0; i < num; i++) {
727 ip = rte_pktmbuf_mtod(m, struct ip *);
729 if (ip->ip_v == IPVERSION) {
730 trf->ip4.pkts[n4] = m;
731 trf->ip4.data[n4] = rte_pktmbuf_mtod_offset(m,
732 uint8_t *, offsetof(struct ip, ip_p));
734 } else if (ip->ip_v == IP6_VERSION) {
735 trf->ip6.pkts[n6] = m;
736 trf->ip6.data[n6] = rte_pktmbuf_mtod_offset(m,
738 offsetof(struct ip6_hdr, ip6_nxt));
750 process_pkts_inbound(struct ipsec_ctx *ipsec_ctx,
751 struct ipsec_traffic *traffic)
753 uint16_t nb_pkts_in, n_ip4, n_ip6;
755 n_ip4 = traffic->ip4.num;
756 n_ip6 = traffic->ip6.num;
758 if (app_sa_prm.enable == 0) {
759 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
760 traffic->ipsec.num, MAX_PKT_BURST);
761 split46_traffic(traffic, traffic->ipsec.pkts, nb_pkts_in);
763 inbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.pkts,
764 traffic->ipsec.saptr, traffic->ipsec.num);
765 ipsec_process(ipsec_ctx, traffic);
768 inbound_sp_sa(ipsec_ctx->sp4_ctx, ipsec_ctx->sa_ctx, &traffic->ip4,
771 inbound_sp_sa(ipsec_ctx->sp6_ctx, ipsec_ctx->sa_ctx, &traffic->ip6,
776 outbound_sp(struct sp_ctx *sp, struct traffic_type *ip,
777 struct traffic_type *ipsec)
780 uint32_t i, j, sa_idx;
782 if (ip->num == 0 || sp == NULL)
785 rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
786 ip->num, DEFAULT_MAX_CATEGORIES);
789 for (i = 0; i < ip->num; i++) {
791 sa_idx = ip->res[i] - 1;
792 if (ip->res[i] == DISCARD)
794 else if (ip->res[i] == BYPASS)
797 ipsec->res[ipsec->num] = sa_idx;
798 ipsec->pkts[ipsec->num++] = m;
805 process_pkts_outbound(struct ipsec_ctx *ipsec_ctx,
806 struct ipsec_traffic *traffic)
809 uint16_t idx, nb_pkts_out, i;
811 /* Drop any IPsec traffic from protected ports */
812 free_pkts(traffic->ipsec.pkts, traffic->ipsec.num);
814 traffic->ipsec.num = 0;
816 outbound_sp(ipsec_ctx->sp4_ctx, &traffic->ip4, &traffic->ipsec);
818 outbound_sp(ipsec_ctx->sp6_ctx, &traffic->ip6, &traffic->ipsec);
820 if (app_sa_prm.enable == 0) {
822 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
823 traffic->ipsec.res, traffic->ipsec.num,
826 for (i = 0; i < nb_pkts_out; i++) {
827 m = traffic->ipsec.pkts[i];
828 struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
829 if (ip->ip_v == IPVERSION) {
830 idx = traffic->ip4.num++;
831 traffic->ip4.pkts[idx] = m;
833 idx = traffic->ip6.num++;
834 traffic->ip6.pkts[idx] = m;
838 outbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.res,
839 traffic->ipsec.saptr, traffic->ipsec.num);
840 ipsec_process(ipsec_ctx, traffic);
845 process_pkts_inbound_nosp(struct ipsec_ctx *ipsec_ctx,
846 struct ipsec_traffic *traffic)
849 uint32_t nb_pkts_in, i, idx;
851 /* Drop any IPv4 traffic from unprotected ports */
852 free_pkts(traffic->ip4.pkts, traffic->ip4.num);
854 traffic->ip4.num = 0;
856 /* Drop any IPv6 traffic from unprotected ports */
857 free_pkts(traffic->ip6.pkts, traffic->ip6.num);
859 traffic->ip6.num = 0;
861 if (app_sa_prm.enable == 0) {
863 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
864 traffic->ipsec.num, MAX_PKT_BURST);
866 for (i = 0; i < nb_pkts_in; i++) {
867 m = traffic->ipsec.pkts[i];
868 struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
869 if (ip->ip_v == IPVERSION) {
870 idx = traffic->ip4.num++;
871 traffic->ip4.pkts[idx] = m;
873 idx = traffic->ip6.num++;
874 traffic->ip6.pkts[idx] = m;
878 inbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.pkts,
879 traffic->ipsec.saptr, traffic->ipsec.num);
880 ipsec_process(ipsec_ctx, traffic);
885 process_pkts_outbound_nosp(struct ipsec_ctx *ipsec_ctx,
886 struct ipsec_traffic *traffic)
889 uint32_t nb_pkts_out, i, n;
892 /* Drop any IPsec traffic from protected ports */
893 free_pkts(traffic->ipsec.pkts, traffic->ipsec.num);
897 for (i = 0; i < traffic->ip4.num; i++) {
898 traffic->ipsec.pkts[n] = traffic->ip4.pkts[i];
899 traffic->ipsec.res[n++] = single_sa_idx;
902 for (i = 0; i < traffic->ip6.num; i++) {
903 traffic->ipsec.pkts[n] = traffic->ip6.pkts[i];
904 traffic->ipsec.res[n++] = single_sa_idx;
907 traffic->ip4.num = 0;
908 traffic->ip6.num = 0;
909 traffic->ipsec.num = n;
911 if (app_sa_prm.enable == 0) {
913 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
914 traffic->ipsec.res, traffic->ipsec.num,
917 /* They all sue the same SA (ip4 or ip6 tunnel) */
918 m = traffic->ipsec.pkts[0];
919 ip = rte_pktmbuf_mtod(m, struct ip *);
920 if (ip->ip_v == IPVERSION) {
921 traffic->ip4.num = nb_pkts_out;
922 for (i = 0; i < nb_pkts_out; i++)
923 traffic->ip4.pkts[i] = traffic->ipsec.pkts[i];
925 traffic->ip6.num = nb_pkts_out;
926 for (i = 0; i < nb_pkts_out; i++)
927 traffic->ip6.pkts[i] = traffic->ipsec.pkts[i];
930 outbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.res,
931 traffic->ipsec.saptr, traffic->ipsec.num);
932 ipsec_process(ipsec_ctx, traffic);
936 static inline int32_t
937 get_hop_for_offload_pkt(struct rte_mbuf *pkt, int is_ipv6)
939 struct ipsec_mbuf_metadata *priv;
942 priv = get_priv(pkt);
945 if (unlikely(sa == NULL)) {
946 RTE_LOG(ERR, IPSEC, "SA not saved in private data\n");
954 return (sa->portid | RTE_LPM_LOOKUP_SUCCESS);
965 route4_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
967 uint32_t hop[MAX_PKT_BURST * 2];
968 uint32_t dst_ip[MAX_PKT_BURST * 2];
971 uint16_t lpm_pkts = 0;
976 /* Need to do an LPM lookup for non-inline packets. Inline packets will
977 * have port ID in the SA
980 for (i = 0; i < nb_pkts; i++) {
981 if (!(pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD)) {
982 /* Security offload not enabled. So an LPM lookup is
983 * required to get the hop
985 offset = offsetof(struct ip, ip_dst);
986 dst_ip[lpm_pkts] = *rte_pktmbuf_mtod_offset(pkts[i],
988 dst_ip[lpm_pkts] = rte_be_to_cpu_32(dst_ip[lpm_pkts]);
993 rte_lpm_lookup_bulk((struct rte_lpm *)rt_ctx, dst_ip, hop, lpm_pkts);
997 for (i = 0; i < nb_pkts; i++) {
998 if (pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD) {
999 /* Read hop from the SA */
1000 pkt_hop = get_hop_for_offload_pkt(pkts[i], 0);
1002 /* Need to use hop returned by lookup */
1003 pkt_hop = hop[lpm_pkts++];
1006 if ((pkt_hop & RTE_LPM_LOOKUP_SUCCESS) == 0) {
1007 free_pkts(&pkts[i], 1);
1010 send_single_packet(pkts[i], pkt_hop & 0xff, IPPROTO_IP);
1015 route6_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
1017 int32_t hop[MAX_PKT_BURST * 2];
1018 uint8_t dst_ip[MAX_PKT_BURST * 2][16];
1020 int32_t pkt_hop = 0;
1022 uint16_t lpm_pkts = 0;
1027 /* Need to do an LPM lookup for non-inline packets. Inline packets will
1028 * have port ID in the SA
1031 for (i = 0; i < nb_pkts; i++) {
1032 if (!(pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD)) {
1033 /* Security offload not enabled. So an LPM lookup is
1034 * required to get the hop
1036 offset = offsetof(struct ip6_hdr, ip6_dst);
1037 ip6_dst = rte_pktmbuf_mtod_offset(pkts[i], uint8_t *,
1039 memcpy(&dst_ip[lpm_pkts][0], ip6_dst, 16);
1044 rte_lpm6_lookup_bulk_func((struct rte_lpm6 *)rt_ctx, dst_ip, hop,
1049 for (i = 0; i < nb_pkts; i++) {
1050 if (pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD) {
1051 /* Read hop from the SA */
1052 pkt_hop = get_hop_for_offload_pkt(pkts[i], 1);
1054 /* Need to use hop returned by lookup */
1055 pkt_hop = hop[lpm_pkts++];
1058 if (pkt_hop == -1) {
1059 free_pkts(&pkts[i], 1);
1062 send_single_packet(pkts[i], pkt_hop & 0xff, IPPROTO_IPV6);
1067 process_pkts(struct lcore_conf *qconf, struct rte_mbuf **pkts,
1068 uint8_t nb_pkts, uint16_t portid)
1070 struct ipsec_traffic traffic;
1072 prepare_traffic(pkts, &traffic, nb_pkts);
1074 if (unlikely(single_sa)) {
1075 if (is_unprotected_port(portid))
1076 process_pkts_inbound_nosp(&qconf->inbound, &traffic);
1078 process_pkts_outbound_nosp(&qconf->outbound, &traffic);
1080 if (is_unprotected_port(portid))
1081 process_pkts_inbound(&qconf->inbound, &traffic);
1083 process_pkts_outbound(&qconf->outbound, &traffic);
1086 route4_pkts(qconf->rt4_ctx, traffic.ip4.pkts, traffic.ip4.num);
1087 route6_pkts(qconf->rt6_ctx, traffic.ip6.pkts, traffic.ip6.num);
1091 drain_tx_buffers(struct lcore_conf *qconf)
1096 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
1097 buf = &qconf->tx_mbufs[portid];
1100 send_burst(qconf, buf->len, portid);
1106 drain_crypto_buffers(struct lcore_conf *qconf)
1109 struct ipsec_ctx *ctx;
1111 /* drain inbound buffers*/
1112 ctx = &qconf->inbound;
1113 for (i = 0; i != ctx->nb_qps; i++) {
1114 if (ctx->tbl[i].len != 0)
1115 enqueue_cop_burst(ctx->tbl + i);
1118 /* drain outbound buffers*/
1119 ctx = &qconf->outbound;
1120 for (i = 0; i != ctx->nb_qps; i++) {
1121 if (ctx->tbl[i].len != 0)
1122 enqueue_cop_burst(ctx->tbl + i);
1127 drain_inbound_crypto_queues(const struct lcore_conf *qconf,
1128 struct ipsec_ctx *ctx)
1131 struct ipsec_traffic trf;
1133 if (app_sa_prm.enable == 0) {
1135 /* dequeue packets from crypto-queue */
1136 n = ipsec_inbound_cqp_dequeue(ctx, trf.ipsec.pkts,
1137 RTE_DIM(trf.ipsec.pkts));
1142 /* split traffic by ipv4-ipv6 */
1143 split46_traffic(&trf, trf.ipsec.pkts, n);
1145 ipsec_cqp_process(ctx, &trf);
1147 /* process ipv4 packets */
1148 if (trf.ip4.num != 0) {
1149 inbound_sp_sa(ctx->sp4_ctx, ctx->sa_ctx, &trf.ip4, 0);
1150 route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
1153 /* process ipv6 packets */
1154 if (trf.ip6.num != 0) {
1155 inbound_sp_sa(ctx->sp6_ctx, ctx->sa_ctx, &trf.ip6, 0);
1156 route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
1161 drain_outbound_crypto_queues(const struct lcore_conf *qconf,
1162 struct ipsec_ctx *ctx)
1165 struct ipsec_traffic trf;
1167 if (app_sa_prm.enable == 0) {
1169 /* dequeue packets from crypto-queue */
1170 n = ipsec_outbound_cqp_dequeue(ctx, trf.ipsec.pkts,
1171 RTE_DIM(trf.ipsec.pkts));
1176 /* split traffic by ipv4-ipv6 */
1177 split46_traffic(&trf, trf.ipsec.pkts, n);
1179 ipsec_cqp_process(ctx, &trf);
1181 /* process ipv4 packets */
1182 if (trf.ip4.num != 0)
1183 route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
1185 /* process ipv6 packets */
1186 if (trf.ip6.num != 0)
1187 route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
1190 /* main processing loop */
1192 ipsec_poll_mode_worker(void)
1194 struct rte_mbuf *pkts[MAX_PKT_BURST];
1196 uint64_t prev_tsc, diff_tsc, cur_tsc;
1200 struct lcore_conf *qconf;
1201 int32_t rc, socket_id;
1202 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
1203 / US_PER_S * BURST_TX_DRAIN_US;
1204 struct lcore_rx_queue *rxql;
1207 lcore_id = rte_lcore_id();
1208 qconf = &lcore_conf[lcore_id];
1209 rxql = qconf->rx_queue_list;
1210 socket_id = rte_lcore_to_socket_id(lcore_id);
1212 qconf->rt4_ctx = socket_ctx[socket_id].rt_ip4;
1213 qconf->rt6_ctx = socket_ctx[socket_id].rt_ip6;
1214 qconf->inbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_in;
1215 qconf->inbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_in;
1216 qconf->inbound.sa_ctx = socket_ctx[socket_id].sa_in;
1217 qconf->inbound.cdev_map = cdev_map_in;
1218 qconf->inbound.session_pool = socket_ctx[socket_id].session_pool;
1219 qconf->inbound.session_priv_pool =
1220 socket_ctx[socket_id].session_priv_pool;
1221 qconf->outbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_out;
1222 qconf->outbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_out;
1223 qconf->outbound.sa_ctx = socket_ctx[socket_id].sa_out;
1224 qconf->outbound.cdev_map = cdev_map_out;
1225 qconf->outbound.session_pool = socket_ctx[socket_id].session_pool;
1226 qconf->outbound.session_priv_pool =
1227 socket_ctx[socket_id].session_priv_pool;
1228 qconf->frag.pool_dir = socket_ctx[socket_id].mbuf_pool;
1229 qconf->frag.pool_indir = socket_ctx[socket_id].mbuf_pool_indir;
1231 rc = ipsec_sad_lcore_cache_init(app_sa_prm.cache_sz);
1234 "SAD cache init on lcore %u, failed with code: %d\n",
1239 if (qconf->nb_rx_queue == 0) {
1240 RTE_LOG(DEBUG, IPSEC, "lcore %u has nothing to do\n",
1245 RTE_LOG(INFO, IPSEC, "entering main loop on lcore %u\n", lcore_id);
1247 for (i = 0; i < qconf->nb_rx_queue; i++) {
1248 portid = rxql[i].port_id;
1249 queueid = rxql[i].queue_id;
1250 RTE_LOG(INFO, IPSEC,
1251 " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
1252 lcore_id, portid, queueid);
1255 while (!force_quit) {
1256 cur_tsc = rte_rdtsc();
1258 /* TX queue buffer drain */
1259 diff_tsc = cur_tsc - prev_tsc;
1261 if (unlikely(diff_tsc > drain_tsc)) {
1262 drain_tx_buffers(qconf);
1263 drain_crypto_buffers(qconf);
1267 for (i = 0; i < qconf->nb_rx_queue; ++i) {
1269 /* Read packets from RX queues */
1270 portid = rxql[i].port_id;
1271 queueid = rxql[i].queue_id;
1272 nb_rx = rte_eth_rx_burst(portid, queueid,
1273 pkts, MAX_PKT_BURST);
1276 core_stats_update_rx(nb_rx);
1277 process_pkts(qconf, pkts, nb_rx, portid);
1280 /* dequeue and process completed crypto-ops */
1281 if (is_unprotected_port(portid))
1282 drain_inbound_crypto_queues(qconf,
1285 drain_outbound_crypto_queues(qconf,
1292 check_flow_params(uint16_t fdir_portid, uint8_t fdir_qid)
1298 for (i = 0; i < nb_lcore_params; ++i) {
1299 portid = lcore_params_array[i].port_id;
1300 if (portid == fdir_portid) {
1301 queueid = lcore_params_array[i].queue_id;
1302 if (queueid == fdir_qid)
1306 if (i == nb_lcore_params - 1)
1314 check_poll_mode_params(struct eh_conf *eh_conf)
1324 if (eh_conf->mode != EH_PKT_TRANSFER_MODE_POLL)
1327 if (lcore_params == NULL) {
1328 printf("Error: No port/queue/core mappings\n");
1332 for (i = 0; i < nb_lcore_params; ++i) {
1333 lcore = lcore_params[i].lcore_id;
1334 if (!rte_lcore_is_enabled(lcore)) {
1335 printf("error: lcore %hhu is not enabled in "
1336 "lcore mask\n", lcore);
1339 socket_id = rte_lcore_to_socket_id(lcore);
1340 if (socket_id != 0 && numa_on == 0) {
1341 printf("warning: lcore %hhu is on socket %d "
1345 portid = lcore_params[i].port_id;
1346 if ((enabled_port_mask & (1 << portid)) == 0) {
1347 printf("port %u is not enabled in port mask\n", portid);
1350 if (!rte_eth_dev_is_valid_port(portid)) {
1351 printf("port %u is not present on the board\n", portid);
1359 get_port_nb_rx_queues(const uint16_t port)
1364 for (i = 0; i < nb_lcore_params; ++i) {
1365 if (lcore_params[i].port_id == port &&
1366 lcore_params[i].queue_id > queue)
1367 queue = lcore_params[i].queue_id;
1369 return (uint8_t)(++queue);
1373 init_lcore_rx_queues(void)
1375 uint16_t i, nb_rx_queue;
1378 for (i = 0; i < nb_lcore_params; ++i) {
1379 lcore = lcore_params[i].lcore_id;
1380 nb_rx_queue = lcore_conf[lcore].nb_rx_queue;
1381 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1382 printf("error: too many queues (%u) for lcore: %u\n",
1383 nb_rx_queue + 1, lcore);
1386 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1387 lcore_params[i].port_id;
1388 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1389 lcore_params[i].queue_id;
1390 lcore_conf[lcore].nb_rx_queue++;
1397 print_usage(const char *prgname)
1399 fprintf(stderr, "%s [EAL options] --"
1405 " [-w REPLAY_WINDOW_SIZE]"
1409 " [-s NUMBER_OF_MBUFS_IN_PKT_POOL]"
1411 " --config (port,queue,lcore)[,(port,queue,lcore)]"
1412 " [--single-sa SAIDX]"
1413 " [--cryptodev_mask MASK]"
1414 " [--transfer-mode MODE]"
1415 " [--event-schedule-type TYPE]"
1416 " [--" CMD_LINE_OPT_RX_OFFLOAD " RX_OFFLOAD_MASK]"
1417 " [--" CMD_LINE_OPT_TX_OFFLOAD " TX_OFFLOAD_MASK]"
1418 " [--" CMD_LINE_OPT_REASSEMBLE " REASSEMBLE_TABLE_SIZE]"
1419 " [--" CMD_LINE_OPT_MTU " MTU]"
1421 " -p PORTMASK: Hexadecimal bitmask of ports to configure\n"
1422 " -P : Enable promiscuous mode\n"
1423 " -u PORTMASK: Hexadecimal bitmask of unprotected ports\n"
1424 " -j FRAMESIZE: Data buffer size, minimum (and default)\n"
1425 " value: RTE_MBUF_DEFAULT_BUF_SIZE\n"
1426 " -l enables code-path that uses librte_ipsec\n"
1427 " -w REPLAY_WINDOW_SIZE specifies IPsec SQN replay window\n"
1428 " size for each SA\n"
1430 " -a enables SA SQN atomic behaviour\n"
1431 " -c specifies inbound SAD cache size,\n"
1432 " zero value disables the cache (default value: 128)\n"
1433 " -s number of mbufs in packet pool, if not specified number\n"
1434 " of mbufs will be calculated based on number of cores,\n"
1435 " ports and crypto queues\n"
1436 " -f CONFIG_FILE: Configuration file\n"
1437 " --config (port,queue,lcore): Rx queue configuration. In poll\n"
1438 " mode determines which queues from\n"
1439 " which ports are mapped to which cores.\n"
1440 " In event mode this option is not used\n"
1441 " as packets are dynamically scheduled\n"
1442 " to cores by HW.\n"
1443 " --single-sa SAIDX: In poll mode use single SA index for\n"
1444 " outbound traffic, bypassing the SP\n"
1445 " In event mode selects driver submode,\n"
1446 " SA index value is ignored\n"
1447 " --cryptodev_mask MASK: Hexadecimal bitmask of the crypto\n"
1448 " devices to configure\n"
1449 " --transfer-mode MODE\n"
1450 " \"poll\" : Packet transfer via polling (default)\n"
1451 " \"event\" : Packet transfer via event device\n"
1452 " --event-schedule-type TYPE queue schedule type, used only when\n"
1453 " transfer mode is set to event\n"
1454 " \"ordered\" : Ordered (default)\n"
1455 " \"atomic\" : Atomic\n"
1456 " \"parallel\" : Parallel\n"
1457 " --" CMD_LINE_OPT_RX_OFFLOAD
1458 ": bitmask of the RX HW offload capabilities to enable/use\n"
1459 " (DEV_RX_OFFLOAD_*)\n"
1460 " --" CMD_LINE_OPT_TX_OFFLOAD
1461 ": bitmask of the TX HW offload capabilities to enable/use\n"
1462 " (DEV_TX_OFFLOAD_*)\n"
1463 " --" CMD_LINE_OPT_REASSEMBLE " NUM"
1464 ": max number of entries in reassemble(fragment) table\n"
1465 " (zero (default value) disables reassembly)\n"
1466 " --" CMD_LINE_OPT_MTU " MTU"
1467 ": MTU value on all ports (default value: 1500)\n"
1468 " outgoing packets with bigger size will be fragmented\n"
1469 " incoming packets with bigger size will be discarded\n"
1470 " --" CMD_LINE_OPT_FRAG_TTL " FRAG_TTL_NS"
1471 ": fragments lifetime in nanoseconds, default\n"
1472 " and maximum value is 10.000.000.000 ns (10 s)\n"
1478 parse_mask(const char *str, uint64_t *val)
1484 t = strtoul(str, &end, 0);
1485 if (errno != 0 || end[0] != 0)
1493 parse_portmask(const char *portmask)
1498 /* parse hexadecimal string */
1499 pm = strtoul(portmask, &end, 16);
1500 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1503 if ((pm == 0) && errno)
1510 parse_decimal(const char *str)
1515 num = strtoull(str, &end, 10);
1516 if ((str[0] == '\0') || (end == NULL) || (*end != '\0')
1524 parse_config(const char *q_arg)
1527 const char *p, *p0 = q_arg;
1535 unsigned long int_fld[_NUM_FLD];
1536 char *str_fld[_NUM_FLD];
1540 nb_lcore_params = 0;
1542 while ((p = strchr(p0, '(')) != NULL) {
1544 p0 = strchr(p, ')');
1549 if (size >= sizeof(s))
1552 snprintf(s, sizeof(s), "%.*s", size, p);
1553 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1556 for (i = 0; i < _NUM_FLD; i++) {
1558 int_fld[i] = strtoul(str_fld[i], &end, 0);
1559 if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
1562 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1563 printf("exceeded max number of lcore params: %hu\n",
1567 lcore_params_array[nb_lcore_params].port_id =
1568 (uint8_t)int_fld[FLD_PORT];
1569 lcore_params_array[nb_lcore_params].queue_id =
1570 (uint8_t)int_fld[FLD_QUEUE];
1571 lcore_params_array[nb_lcore_params].lcore_id =
1572 (uint8_t)int_fld[FLD_LCORE];
1575 lcore_params = lcore_params_array;
1580 print_app_sa_prm(const struct app_sa_prm *prm)
1582 printf("librte_ipsec usage: %s\n",
1583 (prm->enable == 0) ? "disabled" : "enabled");
1585 printf("replay window size: %u\n", prm->window_size);
1586 printf("ESN: %s\n", (prm->enable_esn == 0) ? "disabled" : "enabled");
1587 printf("SA flags: %#" PRIx64 "\n", prm->flags);
1588 printf("Frag TTL: %" PRIu64 " ns\n", frag_ttl_ns);
1592 parse_transfer_mode(struct eh_conf *conf, const char *optarg)
1594 if (!strcmp(CMD_LINE_ARG_POLL, optarg))
1595 conf->mode = EH_PKT_TRANSFER_MODE_POLL;
1596 else if (!strcmp(CMD_LINE_ARG_EVENT, optarg))
1597 conf->mode = EH_PKT_TRANSFER_MODE_EVENT;
1599 printf("Unsupported packet transfer mode\n");
1607 parse_schedule_type(struct eh_conf *conf, const char *optarg)
1609 struct eventmode_conf *em_conf = NULL;
1611 /* Get eventmode conf */
1612 em_conf = conf->mode_params;
1614 if (!strcmp(CMD_LINE_ARG_ORDERED, optarg))
1615 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_ORDERED;
1616 else if (!strcmp(CMD_LINE_ARG_ATOMIC, optarg))
1617 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_ATOMIC;
1618 else if (!strcmp(CMD_LINE_ARG_PARALLEL, optarg))
1619 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_PARALLEL;
1621 printf("Unsupported queue schedule type\n");
1629 parse_args(int32_t argc, char **argv, struct eh_conf *eh_conf)
1634 int32_t option_index;
1635 char *prgname = argv[0];
1636 int32_t f_present = 0;
1640 while ((opt = getopt_long(argc, argvopt, "aelp:Pu:f:j:w:c:s:",
1641 lgopts, &option_index)) != EOF) {
1645 enabled_port_mask = parse_portmask(optarg);
1646 if (enabled_port_mask == 0) {
1647 printf("invalid portmask\n");
1648 print_usage(prgname);
1653 printf("Promiscuous mode selected\n");
1657 unprotected_port_mask = parse_portmask(optarg);
1658 if (unprotected_port_mask == 0) {
1659 printf("invalid unprotected portmask\n");
1660 print_usage(prgname);
1665 if (f_present == 1) {
1666 printf("\"-f\" option present more than "
1668 print_usage(prgname);
1676 ret = parse_decimal(optarg);
1678 printf("Invalid number of buffers in a pool: "
1680 print_usage(prgname);
1684 nb_bufs_in_pool = ret;
1688 ret = parse_decimal(optarg);
1689 if (ret < RTE_MBUF_DEFAULT_BUF_SIZE ||
1691 printf("Invalid frame buffer size value: %s\n",
1693 print_usage(prgname);
1696 frame_buf_size = ret;
1697 printf("Custom frame buffer size %u\n", frame_buf_size);
1700 app_sa_prm.enable = 1;
1703 app_sa_prm.window_size = parse_decimal(optarg);
1706 app_sa_prm.enable_esn = 1;
1709 app_sa_prm.enable = 1;
1710 app_sa_prm.flags |= RTE_IPSEC_SAFLAG_SQN_ATOM;
1713 ret = parse_decimal(optarg);
1715 printf("Invalid SA cache size: %s\n", optarg);
1716 print_usage(prgname);
1719 app_sa_prm.cache_sz = ret;
1721 case CMD_LINE_OPT_CONFIG_NUM:
1722 ret = parse_config(optarg);
1724 printf("Invalid config\n");
1725 print_usage(prgname);
1729 case CMD_LINE_OPT_SINGLE_SA_NUM:
1730 ret = parse_decimal(optarg);
1731 if (ret == -1 || ret > UINT32_MAX) {
1732 printf("Invalid argument[sa_idx]\n");
1733 print_usage(prgname);
1739 single_sa_idx = ret;
1740 eh_conf->ipsec_mode = EH_IPSEC_MODE_TYPE_DRIVER;
1741 printf("Configured with single SA index %u\n",
1744 case CMD_LINE_OPT_CRYPTODEV_MASK_NUM:
1745 ret = parse_portmask(optarg);
1747 printf("Invalid argument[portmask]\n");
1748 print_usage(prgname);
1753 enabled_cryptodev_mask = ret;
1756 case CMD_LINE_OPT_TRANSFER_MODE_NUM:
1757 ret = parse_transfer_mode(eh_conf, optarg);
1759 printf("Invalid packet transfer mode\n");
1760 print_usage(prgname);
1765 case CMD_LINE_OPT_SCHEDULE_TYPE_NUM:
1766 ret = parse_schedule_type(eh_conf, optarg);
1768 printf("Invalid queue schedule type\n");
1769 print_usage(prgname);
1774 case CMD_LINE_OPT_RX_OFFLOAD_NUM:
1775 ret = parse_mask(optarg, &dev_rx_offload);
1777 printf("Invalid argument for \'%s\': %s\n",
1778 CMD_LINE_OPT_RX_OFFLOAD, optarg);
1779 print_usage(prgname);
1783 case CMD_LINE_OPT_TX_OFFLOAD_NUM:
1784 ret = parse_mask(optarg, &dev_tx_offload);
1786 printf("Invalid argument for \'%s\': %s\n",
1787 CMD_LINE_OPT_TX_OFFLOAD, optarg);
1788 print_usage(prgname);
1792 case CMD_LINE_OPT_REASSEMBLE_NUM:
1793 ret = parse_decimal(optarg);
1794 if (ret < 0 || ret > UINT32_MAX) {
1795 printf("Invalid argument for \'%s\': %s\n",
1796 CMD_LINE_OPT_REASSEMBLE, optarg);
1797 print_usage(prgname);
1802 case CMD_LINE_OPT_MTU_NUM:
1803 ret = parse_decimal(optarg);
1804 if (ret < 0 || ret > RTE_IPV4_MAX_PKT_LEN) {
1805 printf("Invalid argument for \'%s\': %s\n",
1806 CMD_LINE_OPT_MTU, optarg);
1807 print_usage(prgname);
1812 case CMD_LINE_OPT_FRAG_TTL_NUM:
1813 ret = parse_decimal(optarg);
1814 if (ret < 0 || ret > MAX_FRAG_TTL_NS) {
1815 printf("Invalid argument for \'%s\': %s\n",
1816 CMD_LINE_OPT_MTU, optarg);
1817 print_usage(prgname);
1823 print_usage(prgname);
1828 if (f_present == 0) {
1829 printf("Mandatory option \"-f\" not present\n");
1833 /* check do we need to enable multi-seg support */
1834 if (multi_seg_required()) {
1835 /* legacy mode doesn't support multi-seg */
1836 app_sa_prm.enable = 1;
1837 printf("frame buf size: %u, mtu: %u, "
1838 "number of reassemble entries: %u\n"
1839 "multi-segment support is required\n",
1840 frame_buf_size, mtu_size, frag_tbl_sz);
1843 print_app_sa_prm(&app_sa_prm);
1846 argv[optind-1] = prgname;
1849 optind = 1; /* reset getopt lib */
1854 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
1856 char buf[RTE_ETHER_ADDR_FMT_SIZE];
1857 rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
1858 printf("%s%s", name, buf);
1862 * Update destination ethaddr for the port.
1865 add_dst_ethaddr(uint16_t port, const struct rte_ether_addr *addr)
1867 if (port >= RTE_DIM(ethaddr_tbl))
1870 ethaddr_tbl[port].dst = ETHADDR_TO_UINT64(addr);
1874 /* Check the link status of all ports in up to 9s, and print them finally */
1876 check_all_ports_link_status(uint32_t port_mask)
1878 #define CHECK_INTERVAL 100 /* 100ms */
1879 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1881 uint8_t count, all_ports_up, print_flag = 0;
1882 struct rte_eth_link link;
1884 char link_status_text[RTE_ETH_LINK_MAX_STR_LEN];
1886 printf("\nChecking link status");
1888 for (count = 0; count <= MAX_CHECK_TIME; count++) {
1890 RTE_ETH_FOREACH_DEV(portid) {
1891 if ((port_mask & (1 << portid)) == 0)
1893 memset(&link, 0, sizeof(link));
1894 ret = rte_eth_link_get_nowait(portid, &link);
1897 if (print_flag == 1)
1898 printf("Port %u link get failed: %s\n",
1899 portid, rte_strerror(-ret));
1902 /* print link status if flag set */
1903 if (print_flag == 1) {
1904 rte_eth_link_to_str(link_status_text,
1905 sizeof(link_status_text), &link);
1906 printf("Port %d %s\n", portid,
1910 /* clear all_ports_up flag if any link down */
1911 if (link.link_status == ETH_LINK_DOWN) {
1916 /* after finally printing all link status, get out */
1917 if (print_flag == 1)
1920 if (all_ports_up == 0) {
1923 rte_delay_ms(CHECK_INTERVAL);
1926 /* set the print_flag if all ports up or timeout */
1927 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1935 add_mapping(struct rte_hash *map, const char *str, uint16_t cdev_id,
1936 uint16_t qp, struct lcore_params *params,
1937 struct ipsec_ctx *ipsec_ctx,
1938 const struct rte_cryptodev_capabilities *cipher,
1939 const struct rte_cryptodev_capabilities *auth,
1940 const struct rte_cryptodev_capabilities *aead)
1944 struct cdev_key key = { 0 };
1946 key.lcore_id = params->lcore_id;
1948 key.cipher_algo = cipher->sym.cipher.algo;
1950 key.auth_algo = auth->sym.auth.algo;
1952 key.aead_algo = aead->sym.aead.algo;
1954 ret = rte_hash_lookup(map, &key);
1958 for (i = 0; i < ipsec_ctx->nb_qps; i++)
1959 if (ipsec_ctx->tbl[i].id == cdev_id)
1962 if (i == ipsec_ctx->nb_qps) {
1963 if (ipsec_ctx->nb_qps == MAX_QP_PER_LCORE) {
1964 printf("Maximum number of crypto devices assigned to "
1965 "a core, increase MAX_QP_PER_LCORE value\n");
1968 ipsec_ctx->tbl[i].id = cdev_id;
1969 ipsec_ctx->tbl[i].qp = qp;
1970 ipsec_ctx->nb_qps++;
1971 printf("%s cdev mapping: lcore %u using cdev %u qp %u "
1972 "(cdev_id_qp %lu)\n", str, key.lcore_id,
1976 ret = rte_hash_add_key_data(map, &key, (void *)i);
1978 printf("Faled to insert cdev mapping for (lcore %u, "
1979 "cdev %u, qp %u), errno %d\n",
1980 key.lcore_id, ipsec_ctx->tbl[i].id,
1981 ipsec_ctx->tbl[i].qp, ret);
1989 add_cdev_mapping(struct rte_cryptodev_info *dev_info, uint16_t cdev_id,
1990 uint16_t qp, struct lcore_params *params)
1993 const struct rte_cryptodev_capabilities *i, *j;
1994 struct rte_hash *map;
1995 struct lcore_conf *qconf;
1996 struct ipsec_ctx *ipsec_ctx;
1999 qconf = &lcore_conf[params->lcore_id];
2001 if ((unprotected_port_mask & (1 << params->port_id)) == 0) {
2003 ipsec_ctx = &qconf->outbound;
2007 ipsec_ctx = &qconf->inbound;
2011 /* Required cryptodevs with operation chainning */
2012 if (!(dev_info->feature_flags &
2013 RTE_CRYPTODEV_FF_SYM_OPERATION_CHAINING))
2016 for (i = dev_info->capabilities;
2017 i->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; i++) {
2018 if (i->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
2021 if (i->sym.xform_type == RTE_CRYPTO_SYM_XFORM_AEAD) {
2022 ret |= add_mapping(map, str, cdev_id, qp, params,
2023 ipsec_ctx, NULL, NULL, i);
2027 if (i->sym.xform_type != RTE_CRYPTO_SYM_XFORM_CIPHER)
2030 for (j = dev_info->capabilities;
2031 j->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; j++) {
2032 if (j->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
2035 if (j->sym.xform_type != RTE_CRYPTO_SYM_XFORM_AUTH)
2038 ret |= add_mapping(map, str, cdev_id, qp, params,
2039 ipsec_ctx, i, j, NULL);
2046 /* Check if the device is enabled by cryptodev_mask */
2048 check_cryptodev_mask(uint8_t cdev_id)
2050 if (enabled_cryptodev_mask & (1 << cdev_id))
2057 cryptodevs_init(uint16_t req_queue_num)
2059 struct rte_cryptodev_config dev_conf;
2060 struct rte_cryptodev_qp_conf qp_conf;
2061 uint16_t idx, max_nb_qps, qp, total_nb_qps, i;
2063 struct rte_hash_parameters params = { 0 };
2065 const uint64_t mseg_flag = multi_seg_required() ?
2066 RTE_CRYPTODEV_FF_IN_PLACE_SGL : 0;
2068 params.entries = CDEV_MAP_ENTRIES;
2069 params.key_len = sizeof(struct cdev_key);
2070 params.hash_func = rte_jhash;
2071 params.hash_func_init_val = 0;
2072 params.socket_id = rte_socket_id();
2074 params.name = "cdev_map_in";
2075 cdev_map_in = rte_hash_create(¶ms);
2076 if (cdev_map_in == NULL)
2077 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
2080 params.name = "cdev_map_out";
2081 cdev_map_out = rte_hash_create(¶ms);
2082 if (cdev_map_out == NULL)
2083 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
2086 printf("lcore/cryptodev/qp mappings:\n");
2090 for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
2091 struct rte_cryptodev_info cdev_info;
2093 if (check_cryptodev_mask((uint8_t)cdev_id))
2096 rte_cryptodev_info_get(cdev_id, &cdev_info);
2098 if ((mseg_flag & cdev_info.feature_flags) != mseg_flag)
2099 rte_exit(EXIT_FAILURE,
2100 "Device %hd does not support \'%s\' feature\n",
2102 rte_cryptodev_get_feature_name(mseg_flag));
2104 if (nb_lcore_params > cdev_info.max_nb_queue_pairs)
2105 max_nb_qps = cdev_info.max_nb_queue_pairs;
2107 max_nb_qps = nb_lcore_params;
2111 while (qp < max_nb_qps && i < nb_lcore_params) {
2112 if (add_cdev_mapping(&cdev_info, cdev_id, qp,
2113 &lcore_params[idx]))
2116 idx = idx % nb_lcore_params;
2120 qp = RTE_MIN(max_nb_qps, RTE_MAX(req_queue_num, qp));
2125 dev_conf.socket_id = rte_cryptodev_socket_id(cdev_id);
2126 dev_conf.nb_queue_pairs = qp;
2127 dev_conf.ff_disable = RTE_CRYPTODEV_FF_ASYMMETRIC_CRYPTO;
2129 uint32_t dev_max_sess = cdev_info.sym.max_nb_sessions;
2130 if (dev_max_sess != 0 &&
2131 dev_max_sess < get_nb_crypto_sessions())
2132 rte_exit(EXIT_FAILURE,
2133 "Device does not support at least %u "
2134 "sessions", get_nb_crypto_sessions());
2136 if (rte_cryptodev_configure(cdev_id, &dev_conf))
2137 rte_panic("Failed to initialize cryptodev %u\n",
2140 qp_conf.nb_descriptors = CDEV_QUEUE_DESC;
2141 qp_conf.mp_session =
2142 socket_ctx[dev_conf.socket_id].session_pool;
2143 qp_conf.mp_session_private =
2144 socket_ctx[dev_conf.socket_id].session_priv_pool;
2145 for (qp = 0; qp < dev_conf.nb_queue_pairs; qp++)
2146 if (rte_cryptodev_queue_pair_setup(cdev_id, qp,
2147 &qp_conf, dev_conf.socket_id))
2148 rte_panic("Failed to setup queue %u for "
2149 "cdev_id %u\n", 0, cdev_id);
2151 if (rte_cryptodev_start(cdev_id))
2152 rte_panic("Failed to start cryptodev %u\n",
2158 return total_nb_qps;
2162 port_init(uint16_t portid, uint64_t req_rx_offloads, uint64_t req_tx_offloads)
2164 uint32_t frame_size;
2165 struct rte_eth_dev_info dev_info;
2166 struct rte_eth_txconf *txconf;
2167 uint16_t nb_tx_queue, nb_rx_queue;
2168 uint16_t tx_queueid, rx_queueid, queue, lcore_id;
2169 int32_t ret, socket_id;
2170 struct lcore_conf *qconf;
2171 struct rte_ether_addr ethaddr;
2172 struct rte_eth_conf local_port_conf = port_conf;
2174 ret = rte_eth_dev_info_get(portid, &dev_info);
2176 rte_exit(EXIT_FAILURE,
2177 "Error during getting device (port %u) info: %s\n",
2178 portid, strerror(-ret));
2180 /* limit allowed HW offloafs, as user requested */
2181 dev_info.rx_offload_capa &= dev_rx_offload;
2182 dev_info.tx_offload_capa &= dev_tx_offload;
2184 printf("Configuring device port %u:\n", portid);
2186 ret = rte_eth_macaddr_get(portid, ðaddr);
2188 rte_exit(EXIT_FAILURE,
2189 "Error getting MAC address (port %u): %s\n",
2190 portid, rte_strerror(-ret));
2192 ethaddr_tbl[portid].src = ETHADDR_TO_UINT64(ðaddr);
2193 print_ethaddr("Address: ", ðaddr);
2196 nb_rx_queue = get_port_nb_rx_queues(portid);
2197 nb_tx_queue = nb_lcores;
2199 if (nb_rx_queue > dev_info.max_rx_queues)
2200 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
2201 "(max rx queue is %u)\n",
2202 nb_rx_queue, dev_info.max_rx_queues);
2204 if (nb_tx_queue > dev_info.max_tx_queues)
2205 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
2206 "(max tx queue is %u)\n",
2207 nb_tx_queue, dev_info.max_tx_queues);
2209 printf("Creating queues: nb_rx_queue=%d nb_tx_queue=%u...\n",
2210 nb_rx_queue, nb_tx_queue);
2212 frame_size = MTU_TO_FRAMELEN(mtu_size);
2213 if (frame_size > local_port_conf.rxmode.max_rx_pkt_len)
2214 local_port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2215 local_port_conf.rxmode.max_rx_pkt_len = frame_size;
2217 if (multi_seg_required()) {
2218 local_port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_SCATTER;
2219 local_port_conf.txmode.offloads |= DEV_TX_OFFLOAD_MULTI_SEGS;
2222 local_port_conf.rxmode.offloads |= req_rx_offloads;
2223 local_port_conf.txmode.offloads |= req_tx_offloads;
2225 /* Check that all required capabilities are supported */
2226 if ((local_port_conf.rxmode.offloads & dev_info.rx_offload_capa) !=
2227 local_port_conf.rxmode.offloads)
2228 rte_exit(EXIT_FAILURE,
2229 "Error: port %u required RX offloads: 0x%" PRIx64
2230 ", avaialbe RX offloads: 0x%" PRIx64 "\n",
2231 portid, local_port_conf.rxmode.offloads,
2232 dev_info.rx_offload_capa);
2234 if ((local_port_conf.txmode.offloads & dev_info.tx_offload_capa) !=
2235 local_port_conf.txmode.offloads)
2236 rte_exit(EXIT_FAILURE,
2237 "Error: port %u required TX offloads: 0x%" PRIx64
2238 ", avaialbe TX offloads: 0x%" PRIx64 "\n",
2239 portid, local_port_conf.txmode.offloads,
2240 dev_info.tx_offload_capa);
2242 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
2243 local_port_conf.txmode.offloads |=
2244 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
2246 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_IPV4_CKSUM)
2247 local_port_conf.txmode.offloads |= DEV_TX_OFFLOAD_IPV4_CKSUM;
2249 printf("port %u configurng rx_offloads=0x%" PRIx64
2250 ", tx_offloads=0x%" PRIx64 "\n",
2251 portid, local_port_conf.rxmode.offloads,
2252 local_port_conf.txmode.offloads);
2254 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
2255 dev_info.flow_type_rss_offloads;
2256 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
2257 port_conf.rx_adv_conf.rss_conf.rss_hf) {
2258 printf("Port %u modified RSS hash function based on hardware support,"
2259 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
2261 port_conf.rx_adv_conf.rss_conf.rss_hf,
2262 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
2265 ret = rte_eth_dev_configure(portid, nb_rx_queue, nb_tx_queue,
2268 rte_exit(EXIT_FAILURE, "Cannot configure device: "
2269 "err=%d, port=%d\n", ret, portid);
2271 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd, &nb_txd);
2273 rte_exit(EXIT_FAILURE, "Cannot adjust number of descriptors: "
2274 "err=%d, port=%d\n", ret, portid);
2276 /* init one TX queue per lcore */
2278 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2279 if (rte_lcore_is_enabled(lcore_id) == 0)
2283 socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2288 printf("Setup txq=%u,%d,%d\n", lcore_id, tx_queueid, socket_id);
2290 txconf = &dev_info.default_txconf;
2291 txconf->offloads = local_port_conf.txmode.offloads;
2293 ret = rte_eth_tx_queue_setup(portid, tx_queueid, nb_txd,
2296 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
2297 "err=%d, port=%d\n", ret, portid);
2299 qconf = &lcore_conf[lcore_id];
2300 qconf->tx_queue_id[portid] = tx_queueid;
2302 /* Pre-populate pkt offloads based on capabilities */
2303 qconf->outbound.ipv4_offloads = PKT_TX_IPV4;
2304 qconf->outbound.ipv6_offloads = PKT_TX_IPV6;
2305 if (local_port_conf.txmode.offloads & DEV_TX_OFFLOAD_IPV4_CKSUM)
2306 qconf->outbound.ipv4_offloads |= PKT_TX_IP_CKSUM;
2310 /* init RX queues */
2311 for (queue = 0; queue < qconf->nb_rx_queue; ++queue) {
2312 struct rte_eth_rxconf rxq_conf;
2314 if (portid != qconf->rx_queue_list[queue].port_id)
2317 rx_queueid = qconf->rx_queue_list[queue].queue_id;
2319 printf("Setup rxq=%d,%d,%d\n", portid, rx_queueid,
2322 rxq_conf = dev_info.default_rxconf;
2323 rxq_conf.offloads = local_port_conf.rxmode.offloads;
2324 ret = rte_eth_rx_queue_setup(portid, rx_queueid,
2325 nb_rxd, socket_id, &rxq_conf,
2326 socket_ctx[socket_id].mbuf_pool);
2328 rte_exit(EXIT_FAILURE,
2329 "rte_eth_rx_queue_setup: err=%d, "
2330 "port=%d\n", ret, portid);
2337 max_session_size(void)
2341 int16_t cdev_id, port_id, n;
2344 n = rte_cryptodev_count();
2345 for (cdev_id = 0; cdev_id != n; cdev_id++) {
2346 sz = rte_cryptodev_sym_get_private_session_size(cdev_id);
2350 * If crypto device is security capable, need to check the
2351 * size of security session as well.
2354 /* Get security context of the crypto device */
2355 sec_ctx = rte_cryptodev_get_sec_ctx(cdev_id);
2356 if (sec_ctx == NULL)
2359 /* Get size of security session */
2360 sz = rte_security_session_get_size(sec_ctx);
2365 RTE_ETH_FOREACH_DEV(port_id) {
2366 if ((enabled_port_mask & (1 << port_id)) == 0)
2369 sec_ctx = rte_eth_dev_get_sec_ctx(port_id);
2370 if (sec_ctx == NULL)
2373 sz = rte_security_session_get_size(sec_ctx);
2382 session_pool_init(struct socket_ctx *ctx, int32_t socket_id, size_t sess_sz)
2384 char mp_name[RTE_MEMPOOL_NAMESIZE];
2385 struct rte_mempool *sess_mp;
2388 snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
2389 "sess_mp_%u", socket_id);
2390 nb_sess = (get_nb_crypto_sessions() + CDEV_MP_CACHE_SZ *
2392 nb_sess = RTE_MAX(nb_sess, CDEV_MP_CACHE_SZ *
2393 CDEV_MP_CACHE_MULTIPLIER);
2394 sess_mp = rte_cryptodev_sym_session_pool_create(
2395 mp_name, nb_sess, sess_sz, CDEV_MP_CACHE_SZ, 0,
2397 ctx->session_pool = sess_mp;
2399 if (ctx->session_pool == NULL)
2400 rte_exit(EXIT_FAILURE,
2401 "Cannot init session pool on socket %d\n", socket_id);
2403 printf("Allocated session pool on socket %d\n", socket_id);
2407 session_priv_pool_init(struct socket_ctx *ctx, int32_t socket_id,
2410 char mp_name[RTE_MEMPOOL_NAMESIZE];
2411 struct rte_mempool *sess_mp;
2414 snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
2415 "sess_mp_priv_%u", socket_id);
2416 nb_sess = (get_nb_crypto_sessions() + CDEV_MP_CACHE_SZ *
2418 nb_sess = RTE_MAX(nb_sess, CDEV_MP_CACHE_SZ *
2419 CDEV_MP_CACHE_MULTIPLIER);
2420 sess_mp = rte_mempool_create(mp_name,
2424 0, NULL, NULL, NULL,
2427 ctx->session_priv_pool = sess_mp;
2429 if (ctx->session_priv_pool == NULL)
2430 rte_exit(EXIT_FAILURE,
2431 "Cannot init session priv pool on socket %d\n",
2434 printf("Allocated session priv pool on socket %d\n",
2439 pool_init(struct socket_ctx *ctx, int32_t socket_id, uint32_t nb_mbuf)
2444 snprintf(s, sizeof(s), "mbuf_pool_%d", socket_id);
2445 ctx->mbuf_pool = rte_pktmbuf_pool_create(s, nb_mbuf,
2446 MEMPOOL_CACHE_SIZE, ipsec_metadata_size(),
2447 frame_buf_size, socket_id);
2450 * if multi-segment support is enabled, then create a pool
2451 * for indirect mbufs.
2453 ms = multi_seg_required();
2455 snprintf(s, sizeof(s), "mbuf_pool_indir_%d", socket_id);
2456 ctx->mbuf_pool_indir = rte_pktmbuf_pool_create(s, nb_mbuf,
2457 MEMPOOL_CACHE_SIZE, 0, 0, socket_id);
2460 if (ctx->mbuf_pool == NULL || (ms != 0 && ctx->mbuf_pool_indir == NULL))
2461 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool on socket %d\n",
2464 printf("Allocated mbuf pool on socket %d\n", socket_id);
2468 inline_ipsec_event_esn_overflow(struct rte_security_ctx *ctx, uint64_t md)
2470 struct ipsec_sa *sa;
2472 /* For inline protocol processing, the metadata in the event will
2473 * uniquely identify the security session which raised the event.
2474 * Application would then need the userdata it had registered with the
2475 * security session to process the event.
2478 sa = (struct ipsec_sa *)rte_security_get_userdata(ctx, md);
2481 /* userdata could not be retrieved */
2485 /* Sequence number over flow. SA need to be re-established */
2491 inline_ipsec_event_callback(uint16_t port_id, enum rte_eth_event_type type,
2492 void *param, void *ret_param)
2495 struct rte_eth_event_ipsec_desc *event_desc = NULL;
2496 struct rte_security_ctx *ctx = (struct rte_security_ctx *)
2497 rte_eth_dev_get_sec_ctx(port_id);
2499 RTE_SET_USED(param);
2501 if (type != RTE_ETH_EVENT_IPSEC)
2504 event_desc = ret_param;
2505 if (event_desc == NULL) {
2506 printf("Event descriptor not set\n");
2510 md = event_desc->metadata;
2512 if (event_desc->subtype == RTE_ETH_EVENT_IPSEC_ESN_OVERFLOW)
2513 return inline_ipsec_event_esn_overflow(ctx, md);
2514 else if (event_desc->subtype >= RTE_ETH_EVENT_IPSEC_MAX) {
2515 printf("Invalid IPsec event reported\n");
2523 rx_callback(__rte_unused uint16_t port, __rte_unused uint16_t queue,
2524 struct rte_mbuf *pkt[], uint16_t nb_pkts,
2525 __rte_unused uint16_t max_pkts, void *user_param)
2529 struct lcore_conf *lc;
2530 struct rte_mbuf *mb;
2531 struct rte_ether_hdr *eth;
2537 for (i = 0; i != nb_pkts; i++) {
2540 eth = rte_pktmbuf_mtod(mb, struct rte_ether_hdr *);
2541 if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
2543 struct rte_ipv4_hdr *iph;
2545 iph = (struct rte_ipv4_hdr *)(eth + 1);
2546 if (rte_ipv4_frag_pkt_is_fragmented(iph)) {
2548 mb->l2_len = sizeof(*eth);
2549 mb->l3_len = sizeof(*iph);
2550 tm = (tm != 0) ? tm : rte_rdtsc();
2551 mb = rte_ipv4_frag_reassemble_packet(
2552 lc->frag.tbl, &lc->frag.dr,
2556 /* fix ip cksum after reassemble. */
2557 iph = rte_pktmbuf_mtod_offset(mb,
2558 struct rte_ipv4_hdr *,
2560 iph->hdr_checksum = 0;
2561 iph->hdr_checksum = rte_ipv4_cksum(iph);
2564 } else if (eth->ether_type ==
2565 rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6)) {
2567 struct rte_ipv6_hdr *iph;
2568 struct ipv6_extension_fragment *fh;
2570 iph = (struct rte_ipv6_hdr *)(eth + 1);
2571 fh = rte_ipv6_frag_get_ipv6_fragment_header(iph);
2573 mb->l2_len = sizeof(*eth);
2574 mb->l3_len = (uintptr_t)fh - (uintptr_t)iph +
2576 tm = (tm != 0) ? tm : rte_rdtsc();
2577 mb = rte_ipv6_frag_reassemble_packet(
2578 lc->frag.tbl, &lc->frag.dr,
2581 /* fix l3_len after reassemble. */
2582 mb->l3_len = mb->l3_len - sizeof(*fh);
2590 /* some fragments were encountered, drain death row */
2592 rte_ip_frag_free_death_row(&lc->frag.dr, 0);
2599 reassemble_lcore_init(struct lcore_conf *lc, uint32_t cid)
2603 uint64_t frag_cycles;
2604 const struct lcore_rx_queue *rxq;
2605 const struct rte_eth_rxtx_callback *cb;
2607 /* create fragment table */
2608 sid = rte_lcore_to_socket_id(cid);
2609 frag_cycles = (rte_get_tsc_hz() + NS_PER_S - 1) /
2610 NS_PER_S * frag_ttl_ns;
2612 lc->frag.tbl = rte_ip_frag_table_create(frag_tbl_sz,
2613 FRAG_TBL_BUCKET_ENTRIES, frag_tbl_sz, frag_cycles, sid);
2614 if (lc->frag.tbl == NULL) {
2615 printf("%s(%u): failed to create fragment table of size: %u, "
2617 __func__, cid, frag_tbl_sz, rte_errno);
2621 /* setup reassemble RX callbacks for all queues */
2622 for (i = 0; i != lc->nb_rx_queue; i++) {
2624 rxq = lc->rx_queue_list + i;
2625 cb = rte_eth_add_rx_callback(rxq->port_id, rxq->queue_id,
2628 printf("%s(%u): failed to install RX callback for "
2629 "portid=%u, queueid=%u, error code: %d\n",
2631 rxq->port_id, rxq->queue_id, rte_errno);
2640 reassemble_init(void)
2646 for (i = 0; i != nb_lcore_params; i++) {
2647 lc = lcore_params[i].lcore_id;
2648 rc = reassemble_lcore_init(lcore_conf + lc, lc);
2657 create_default_ipsec_flow(uint16_t port_id, uint64_t rx_offloads)
2659 struct rte_flow_action action[2];
2660 struct rte_flow_item pattern[2];
2661 struct rte_flow_attr attr = {0};
2662 struct rte_flow_error err;
2663 struct rte_flow *flow;
2666 if (!(rx_offloads & DEV_RX_OFFLOAD_SECURITY))
2669 /* Add the default rte_flow to enable SECURITY for all ESP packets */
2671 pattern[0].type = RTE_FLOW_ITEM_TYPE_ESP;
2672 pattern[0].spec = NULL;
2673 pattern[0].mask = NULL;
2674 pattern[0].last = NULL;
2675 pattern[1].type = RTE_FLOW_ITEM_TYPE_END;
2677 action[0].type = RTE_FLOW_ACTION_TYPE_SECURITY;
2678 action[0].conf = NULL;
2679 action[1].type = RTE_FLOW_ACTION_TYPE_END;
2680 action[1].conf = NULL;
2684 ret = rte_flow_validate(port_id, &attr, pattern, action, &err);
2688 flow = rte_flow_create(port_id, &attr, pattern, action, &err);
2692 flow_info_tbl[port_id].rx_def_flow = flow;
2693 RTE_LOG(INFO, IPSEC,
2694 "Created default flow enabling SECURITY for all ESP traffic on port %d\n",
2699 signal_handler(int signum)
2701 if (signum == SIGINT || signum == SIGTERM) {
2702 printf("\n\nSignal %d received, preparing to exit...\n",
2709 ev_mode_sess_verify(struct ipsec_sa *sa, int nb_sa)
2711 struct rte_ipsec_session *ips;
2717 for (i = 0; i < nb_sa; i++) {
2718 ips = ipsec_get_primary_session(&sa[i]);
2719 if (ips->type != RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL)
2720 rte_exit(EXIT_FAILURE, "Event mode supports only "
2721 "inline protocol sessions\n");
2727 check_event_mode_params(struct eh_conf *eh_conf)
2729 struct eventmode_conf *em_conf = NULL;
2730 struct lcore_params *params;
2733 if (!eh_conf || !eh_conf->mode_params)
2736 /* Get eventmode conf */
2737 em_conf = eh_conf->mode_params;
2739 if (eh_conf->mode == EH_PKT_TRANSFER_MODE_POLL &&
2740 em_conf->ext_params.sched_type != SCHED_TYPE_NOT_SET) {
2741 printf("error: option --event-schedule-type applies only to "
2746 if (eh_conf->mode != EH_PKT_TRANSFER_MODE_EVENT)
2749 /* Set schedule type to ORDERED if it wasn't explicitly set by user */
2750 if (em_conf->ext_params.sched_type == SCHED_TYPE_NOT_SET)
2751 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_ORDERED;
2754 * Event mode currently supports only inline protocol sessions.
2755 * If there are other types of sessions configured then exit with
2758 ev_mode_sess_verify(sa_in, nb_sa_in);
2759 ev_mode_sess_verify(sa_out, nb_sa_out);
2762 /* Option --config does not apply to event mode */
2763 if (nb_lcore_params > 0) {
2764 printf("error: option --config applies only to poll mode\n");
2769 * In order to use the same port_init routine for both poll and event
2770 * modes initialize lcore_params with one queue for each eth port
2772 lcore_params = lcore_params_array;
2773 RTE_ETH_FOREACH_DEV(portid) {
2774 if ((enabled_port_mask & (1 << portid)) == 0)
2777 params = &lcore_params[nb_lcore_params++];
2778 params->port_id = portid;
2779 params->queue_id = 0;
2780 params->lcore_id = rte_get_next_lcore(0, 0, 1);
2787 inline_sessions_free(struct sa_ctx *sa_ctx)
2789 struct rte_ipsec_session *ips;
2790 struct ipsec_sa *sa;
2797 for (i = 0; i < sa_ctx->nb_sa; i++) {
2799 sa = &sa_ctx->sa[i];
2803 ips = ipsec_get_primary_session(sa);
2804 if (ips->type != RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL &&
2805 ips->type != RTE_SECURITY_ACTION_TYPE_INLINE_CRYPTO)
2808 if (!rte_eth_dev_is_valid_port(sa->portid))
2811 ret = rte_security_session_destroy(
2812 rte_eth_dev_get_sec_ctx(sa->portid),
2815 RTE_LOG(ERR, IPSEC, "Failed to destroy security "
2816 "session type %d, spi %d\n",
2817 ips->type, sa->spi);
2822 calculate_nb_mbufs(uint16_t nb_ports, uint16_t nb_crypto_qp, uint32_t nb_rxq,
2825 return RTE_MAX((nb_rxq * nb_rxd +
2826 nb_ports * nb_lcores * MAX_PKT_BURST +
2827 nb_ports * nb_txq * nb_txd +
2828 nb_lcores * MEMPOOL_CACHE_SIZE +
2829 nb_crypto_qp * CDEV_QUEUE_DESC +
2830 nb_lcores * frag_tbl_sz *
2831 FRAG_TBL_BUCKET_ENTRIES),
2836 main(int32_t argc, char **argv)
2839 uint32_t lcore_id, nb_txq, nb_rxq = 0;
2843 uint16_t portid, nb_crypto_qp, nb_ports = 0;
2844 uint64_t req_rx_offloads[RTE_MAX_ETHPORTS];
2845 uint64_t req_tx_offloads[RTE_MAX_ETHPORTS];
2846 struct eh_conf *eh_conf = NULL;
2849 nb_bufs_in_pool = 0;
2852 ret = rte_eal_init(argc, argv);
2854 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
2859 signal(SIGINT, signal_handler);
2860 signal(SIGTERM, signal_handler);
2862 /* initialize event helper configuration */
2863 eh_conf = eh_conf_init();
2864 if (eh_conf == NULL)
2865 rte_exit(EXIT_FAILURE, "Failed to init event helper config");
2867 /* parse application arguments (after the EAL ones) */
2868 ret = parse_args(argc, argv, eh_conf);
2870 rte_exit(EXIT_FAILURE, "Invalid parameters\n");
2872 /* parse configuration file */
2873 if (parse_cfg_file(cfgfile) < 0) {
2874 printf("parsing file \"%s\" failed\n",
2876 print_usage(argv[0]);
2880 if ((unprotected_port_mask & enabled_port_mask) !=
2881 unprotected_port_mask)
2882 rte_exit(EXIT_FAILURE, "Invalid unprotected portmask 0x%x\n",
2883 unprotected_port_mask);
2885 if (check_poll_mode_params(eh_conf) < 0)
2886 rte_exit(EXIT_FAILURE, "check_poll_mode_params failed\n");
2888 if (check_event_mode_params(eh_conf) < 0)
2889 rte_exit(EXIT_FAILURE, "check_event_mode_params failed\n");
2891 ret = init_lcore_rx_queues();
2893 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
2895 nb_lcores = rte_lcore_count();
2897 sess_sz = max_session_size();
2900 * In event mode request minimum number of crypto queues
2901 * to be reserved equal to number of ports.
2903 if (eh_conf->mode == EH_PKT_TRANSFER_MODE_EVENT)
2904 nb_crypto_qp = rte_eth_dev_count_avail();
2908 nb_crypto_qp = cryptodevs_init(nb_crypto_qp);
2910 if (nb_bufs_in_pool == 0) {
2911 RTE_ETH_FOREACH_DEV(portid) {
2912 if ((enabled_port_mask & (1 << portid)) == 0)
2915 nb_rxq += get_port_nb_rx_queues(portid);
2920 nb_bufs_in_pool = calculate_nb_mbufs(nb_ports, nb_crypto_qp,
2924 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2925 if (rte_lcore_is_enabled(lcore_id) == 0)
2929 socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2933 /* mbuf_pool is initialised by the pool_init() function*/
2934 if (socket_ctx[socket_id].mbuf_pool)
2937 pool_init(&socket_ctx[socket_id], socket_id, nb_bufs_in_pool);
2938 session_pool_init(&socket_ctx[socket_id], socket_id, sess_sz);
2939 session_priv_pool_init(&socket_ctx[socket_id], socket_id,
2942 printf("Number of mbufs in packet pool %d\n", nb_bufs_in_pool);
2944 RTE_ETH_FOREACH_DEV(portid) {
2945 if ((enabled_port_mask & (1 << portid)) == 0)
2948 sa_check_offloads(portid, &req_rx_offloads[portid],
2949 &req_tx_offloads[portid]);
2950 port_init(portid, req_rx_offloads[portid],
2951 req_tx_offloads[portid]);
2955 * Set the enabled port mask in helper config for use by helper
2956 * sub-system. This will be used while initializing devices using
2957 * helper sub-system.
2959 eh_conf->eth_portmask = enabled_port_mask;
2961 /* Initialize eventmode components */
2962 ret = eh_devs_init(eh_conf);
2964 rte_exit(EXIT_FAILURE, "eh_devs_init failed, err=%d\n", ret);
2967 RTE_ETH_FOREACH_DEV(portid) {
2968 if ((enabled_port_mask & (1 << portid)) == 0)
2971 /* Create flow before starting the device */
2972 create_default_ipsec_flow(portid, req_rx_offloads[portid]);
2974 ret = rte_eth_dev_start(portid);
2976 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: "
2977 "err=%d, port=%d\n", ret, portid);
2979 * If enabled, put device in promiscuous mode.
2980 * This allows IO forwarding mode to forward packets
2981 * to itself through 2 cross-connected ports of the
2984 if (promiscuous_on) {
2985 ret = rte_eth_promiscuous_enable(portid);
2987 rte_exit(EXIT_FAILURE,
2988 "rte_eth_promiscuous_enable: err=%s, port=%d\n",
2989 rte_strerror(-ret), portid);
2992 rte_eth_dev_callback_register(portid,
2993 RTE_ETH_EVENT_IPSEC, inline_ipsec_event_callback, NULL);
2996 /* fragment reassemble is enabled */
2997 if (frag_tbl_sz != 0) {
2998 ret = reassemble_init();
3000 rte_exit(EXIT_FAILURE, "failed at reassemble init");
3003 /* Replicate each context per socket */
3004 for (i = 0; i < NB_SOCKETS && i < rte_socket_count(); i++) {
3005 socket_id = rte_socket_id_by_idx(i);
3006 if ((socket_ctx[socket_id].mbuf_pool != NULL) &&
3007 (socket_ctx[socket_id].sa_in == NULL) &&
3008 (socket_ctx[socket_id].sa_out == NULL)) {
3009 sa_init(&socket_ctx[socket_id], socket_id);
3010 sp4_init(&socket_ctx[socket_id], socket_id);
3011 sp6_init(&socket_ctx[socket_id], socket_id);
3012 rt_init(&socket_ctx[socket_id], socket_id);
3018 check_all_ports_link_status(enabled_port_mask);
3020 #if (STATS_INTERVAL > 0)
3021 rte_eal_alarm_set(STATS_INTERVAL * US_PER_S, print_stats_cb, NULL);
3023 RTE_LOG(INFO, IPSEC, "Stats display disabled\n");
3024 #endif /* STATS_INTERVAL */
3026 /* launch per-lcore init on every lcore */
3027 rte_eal_mp_remote_launch(ipsec_launch_one_lcore, eh_conf, CALL_MAIN);
3028 RTE_LCORE_FOREACH_WORKER(lcore_id) {
3029 if (rte_eal_wait_lcore(lcore_id) < 0)
3033 /* Uninitialize eventmode components */
3034 ret = eh_devs_uninit(eh_conf);
3036 rte_exit(EXIT_FAILURE, "eh_devs_uninit failed, err=%d\n", ret);
3038 /* Free eventmode configuration memory */
3039 eh_conf_uninit(eh_conf);
3041 /* Destroy inline inbound and outbound sessions */
3042 for (i = 0; i < NB_SOCKETS && i < rte_socket_count(); i++) {
3043 socket_id = rte_socket_id_by_idx(i);
3044 inline_sessions_free(socket_ctx[socket_id].sa_in);
3045 inline_sessions_free(socket_ctx[socket_id].sa_out);
3048 for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
3049 printf("Closing cryptodev %d...", cdev_id);
3050 rte_cryptodev_stop(cdev_id);
3051 rte_cryptodev_close(cdev_id);
3055 RTE_ETH_FOREACH_DEV(portid) {
3056 if ((enabled_port_mask & (1 << portid)) == 0)
3059 printf("Closing port %d...", portid);
3060 if (flow_info_tbl[portid].rx_def_flow) {
3061 struct rte_flow_error err;
3063 ret = rte_flow_destroy(portid,
3064 flow_info_tbl[portid].rx_def_flow, &err);
3066 RTE_LOG(ERR, IPSEC, "Failed to destroy flow "
3067 " for port %u, err msg: %s\n", portid,
3070 ret = rte_eth_dev_stop(portid);
3073 "rte_eth_dev_stop: err=%s, port=%u\n",
3074 rte_strerror(-ret), portid);
3076 rte_eth_dev_close(portid);
3080 /* clean up the EAL */