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;
183 /* application wide librte_ipsec/SA parameters */
184 struct app_sa_prm app_sa_prm = {
186 .cache_sz = SA_CACHE_SZ,
189 static const char *cfgfile;
191 struct lcore_rx_queue {
194 } __rte_cache_aligned;
196 struct lcore_params {
200 } __rte_cache_aligned;
202 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
204 static struct lcore_params *lcore_params;
205 static uint16_t nb_lcore_params;
207 static struct rte_hash *cdev_map_in;
208 static struct rte_hash *cdev_map_out;
212 struct rte_mbuf *m_table[MAX_PKT_BURST] __rte_aligned(sizeof(void *));
216 uint16_t nb_rx_queue;
217 struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
218 uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
219 struct buffer tx_mbufs[RTE_MAX_ETHPORTS];
220 struct ipsec_ctx inbound;
221 struct ipsec_ctx outbound;
222 struct rt_ctx *rt4_ctx;
223 struct rt_ctx *rt6_ctx;
225 struct rte_ip_frag_tbl *tbl;
226 struct rte_mempool *pool_dir;
227 struct rte_mempool *pool_indir;
228 struct rte_ip_frag_death_row dr;
230 } __rte_cache_aligned;
232 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
234 static struct rte_eth_conf port_conf = {
236 .mq_mode = RTE_ETH_MQ_RX_RSS,
238 .offloads = RTE_ETH_RX_OFFLOAD_CHECKSUM,
243 .rss_hf = RTE_ETH_RSS_IP | RTE_ETH_RSS_UDP |
244 RTE_ETH_RSS_TCP | RTE_ETH_RSS_SCTP,
248 .mq_mode = RTE_ETH_MQ_TX_NONE,
252 struct socket_ctx socket_ctx[NB_SOCKETS];
255 * Determine is multi-segment support required:
256 * - either frame buffer size is smaller then mtu
257 * - or reassmeble support is requested
260 multi_seg_required(void)
262 return (MTU_TO_FRAMELEN(mtu_size) + RTE_PKTMBUF_HEADROOM >
263 frame_buf_size || frag_tbl_sz != 0);
267 adjust_ipv4_pktlen(struct rte_mbuf *m, const struct rte_ipv4_hdr *iph,
272 plen = rte_be_to_cpu_16(iph->total_length) + l2_len;
273 if (plen < m->pkt_len) {
274 trim = m->pkt_len - plen;
275 rte_pktmbuf_trim(m, trim);
280 adjust_ipv6_pktlen(struct rte_mbuf *m, const struct rte_ipv6_hdr *iph,
285 plen = rte_be_to_cpu_16(iph->payload_len) + sizeof(*iph) + l2_len;
286 if (plen < m->pkt_len) {
287 trim = m->pkt_len - plen;
288 rte_pktmbuf_trim(m, trim);
292 #if (STATS_INTERVAL > 0)
294 /* Print out statistics on packet distribution */
296 print_stats_cb(__rte_unused void *param)
298 uint64_t total_packets_dropped, total_packets_tx, total_packets_rx;
299 float burst_percent, rx_per_call, tx_per_call;
302 total_packets_dropped = 0;
303 total_packets_tx = 0;
304 total_packets_rx = 0;
306 const char clr[] = { 27, '[', '2', 'J', '\0' };
307 const char topLeft[] = { 27, '[', '1', ';', '1', 'H', '\0' };
309 /* Clear screen and move to top left */
310 printf("%s%s", clr, topLeft);
312 printf("\nCore statistics ====================================");
314 for (coreid = 0; coreid < RTE_MAX_LCORE; coreid++) {
315 /* skip disabled cores */
316 if (rte_lcore_is_enabled(coreid) == 0)
318 burst_percent = (float)(core_statistics[coreid].burst_rx * 100)/
319 core_statistics[coreid].rx;
320 rx_per_call = (float)(core_statistics[coreid].rx)/
321 core_statistics[coreid].rx_call;
322 tx_per_call = (float)(core_statistics[coreid].tx)/
323 core_statistics[coreid].tx_call;
324 printf("\nStatistics for core %u ------------------------------"
325 "\nPackets received: %20"PRIu64
326 "\nPackets sent: %24"PRIu64
327 "\nPackets dropped: %21"PRIu64
328 "\nBurst percent: %23.2f"
329 "\nPackets per Rx call: %17.2f"
330 "\nPackets per Tx call: %17.2f",
332 core_statistics[coreid].rx,
333 core_statistics[coreid].tx,
334 core_statistics[coreid].dropped,
339 total_packets_dropped += core_statistics[coreid].dropped;
340 total_packets_tx += core_statistics[coreid].tx;
341 total_packets_rx += core_statistics[coreid].rx;
343 printf("\nAggregate statistics ==============================="
344 "\nTotal packets received: %14"PRIu64
345 "\nTotal packets sent: %18"PRIu64
346 "\nTotal packets dropped: %15"PRIu64,
349 total_packets_dropped);
350 printf("\n====================================================\n");
352 rte_eal_alarm_set(STATS_INTERVAL * US_PER_S, print_stats_cb, NULL);
354 #endif /* STATS_INTERVAL */
357 prepare_one_packet(struct rte_mbuf *pkt, struct ipsec_traffic *t)
359 const struct rte_ether_hdr *eth;
360 const struct rte_ipv4_hdr *iph4;
361 const struct rte_ipv6_hdr *iph6;
362 const struct rte_udp_hdr *udp;
363 uint16_t ip4_hdr_len;
366 eth = rte_pktmbuf_mtod(pkt, const struct rte_ether_hdr *);
367 if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
369 iph4 = (const struct rte_ipv4_hdr *)rte_pktmbuf_adj(pkt,
371 adjust_ipv4_pktlen(pkt, iph4, 0);
373 switch (iph4->next_proto_id) {
375 t->ipsec.pkts[(t->ipsec.num)++] = pkt;
378 if (app_sa_prm.udp_encap == 1) {
379 ip4_hdr_len = ((iph4->version_ihl &
380 RTE_IPV4_HDR_IHL_MASK) *
381 RTE_IPV4_IHL_MULTIPLIER);
382 udp = rte_pktmbuf_mtod_offset(pkt,
383 struct rte_udp_hdr *, ip4_hdr_len);
384 nat_port = rte_cpu_to_be_16(IPSEC_NAT_T_PORT);
385 if (udp->src_port == nat_port ||
386 udp->dst_port == nat_port){
387 t->ipsec.pkts[(t->ipsec.num)++] = pkt;
389 MBUF_PTYPE_TUNNEL_ESP_IN_UDP;
395 t->ip4.data[t->ip4.num] = &iph4->next_proto_id;
396 t->ip4.pkts[(t->ip4.num)++] = pkt;
399 pkt->l3_len = sizeof(*iph4);
400 pkt->packet_type |= RTE_PTYPE_L3_IPV4;
401 if (pkt->packet_type & RTE_PTYPE_L4_TCP)
402 pkt->l4_len = sizeof(struct rte_tcp_hdr);
403 else if (pkt->packet_type & RTE_PTYPE_L4_UDP)
404 pkt->l4_len = sizeof(struct rte_udp_hdr);
405 } else if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6)) {
407 size_t l3len, ext_len;
410 /* get protocol type */
411 iph6 = (const struct rte_ipv6_hdr *)rte_pktmbuf_adj(pkt,
413 adjust_ipv6_pktlen(pkt, iph6, 0);
415 next_proto = iph6->proto;
417 /* determine l3 header size up to ESP extension */
418 l3len = sizeof(struct ip6_hdr);
419 p = rte_pktmbuf_mtod(pkt, uint8_t *);
420 while (next_proto != IPPROTO_ESP && l3len < pkt->data_len &&
421 (next_proto = rte_ipv6_get_next_ext(p + l3len,
422 next_proto, &ext_len)) >= 0)
425 /* drop packet when IPv6 header exceeds first segment length */
426 if (unlikely(l3len > pkt->data_len)) {
431 switch (next_proto) {
433 t->ipsec.pkts[(t->ipsec.num)++] = pkt;
436 if (app_sa_prm.udp_encap == 1) {
437 udp = rte_pktmbuf_mtod_offset(pkt,
438 struct rte_udp_hdr *, l3len);
439 nat_port = rte_cpu_to_be_16(IPSEC_NAT_T_PORT);
440 if (udp->src_port == nat_port ||
441 udp->dst_port == nat_port){
442 t->ipsec.pkts[(t->ipsec.num)++] = pkt;
444 MBUF_PTYPE_TUNNEL_ESP_IN_UDP;
450 t->ip6.data[t->ip6.num] = &iph6->proto;
451 t->ip6.pkts[(t->ip6.num)++] = pkt;
455 pkt->packet_type |= RTE_PTYPE_L3_IPV6;
457 /* Unknown/Unsupported type, drop the packet */
458 RTE_LOG(ERR, IPSEC, "Unsupported packet type 0x%x\n",
459 rte_be_to_cpu_16(eth->ether_type));
464 /* Check if the packet has been processed inline. For inline protocol
465 * processed packets, the metadata in the mbuf can be used to identify
466 * the security processing done on the packet. The metadata will be
467 * used to retrieve the application registered userdata associated
468 * with the security session.
471 if (pkt->ol_flags & RTE_MBUF_F_RX_SEC_OFFLOAD &&
472 rte_security_dynfield_is_registered()) {
474 struct ipsec_mbuf_metadata *priv;
475 struct rte_security_ctx *ctx = (struct rte_security_ctx *)
476 rte_eth_dev_get_sec_ctx(
479 /* Retrieve the userdata registered. Here, the userdata
480 * registered is the SA pointer.
482 sa = (struct ipsec_sa *)rte_security_get_userdata(ctx,
483 *rte_security_dynfield(pkt));
485 /* userdata could not be retrieved */
489 /* Save SA as priv member in mbuf. This will be used in the
490 * IPsec selector(SP-SA) check.
493 priv = get_priv(pkt);
499 prepare_traffic(struct rte_mbuf **pkts, struct ipsec_traffic *t,
508 for (i = 0; i < (nb_pkts - PREFETCH_OFFSET); i++) {
509 rte_prefetch0(rte_pktmbuf_mtod(pkts[i + PREFETCH_OFFSET],
511 prepare_one_packet(pkts[i], t);
513 /* Process left packets */
514 for (; i < nb_pkts; i++)
515 prepare_one_packet(pkts[i], t);
519 prepare_tx_pkt(struct rte_mbuf *pkt, uint16_t port,
520 const struct lcore_conf *qconf)
523 struct rte_ether_hdr *ethhdr;
525 ip = rte_pktmbuf_mtod(pkt, struct ip *);
527 ethhdr = (struct rte_ether_hdr *)
528 rte_pktmbuf_prepend(pkt, RTE_ETHER_HDR_LEN);
530 if (ip->ip_v == IPVERSION) {
531 pkt->ol_flags |= qconf->outbound.ipv4_offloads;
532 pkt->l3_len = sizeof(struct ip);
533 pkt->l2_len = RTE_ETHER_HDR_LEN;
537 /* calculate IPv4 cksum in SW */
538 if ((pkt->ol_flags & RTE_MBUF_F_TX_IP_CKSUM) == 0)
539 ip->ip_sum = rte_ipv4_cksum((struct rte_ipv4_hdr *)ip);
541 ethhdr->ether_type = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4);
543 pkt->ol_flags |= qconf->outbound.ipv6_offloads;
544 pkt->l3_len = sizeof(struct ip6_hdr);
545 pkt->l2_len = RTE_ETHER_HDR_LEN;
547 ethhdr->ether_type = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6);
550 memcpy(ðhdr->src_addr, ðaddr_tbl[port].src,
551 sizeof(struct rte_ether_addr));
552 memcpy(ðhdr->dst_addr, ðaddr_tbl[port].dst,
553 sizeof(struct rte_ether_addr));
557 prepare_tx_burst(struct rte_mbuf *pkts[], uint16_t nb_pkts, uint16_t port,
558 const struct lcore_conf *qconf)
561 const int32_t prefetch_offset = 2;
563 for (i = 0; i < (nb_pkts - prefetch_offset); i++) {
564 rte_mbuf_prefetch_part2(pkts[i + prefetch_offset]);
565 prepare_tx_pkt(pkts[i], port, qconf);
567 /* Process left packets */
568 for (; i < nb_pkts; i++)
569 prepare_tx_pkt(pkts[i], port, qconf);
572 /* Send burst of packets on an output interface */
573 static inline int32_t
574 send_burst(struct lcore_conf *qconf, uint16_t n, uint16_t port)
576 struct rte_mbuf **m_table;
580 queueid = qconf->tx_queue_id[port];
581 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
583 prepare_tx_burst(m_table, n, port, qconf);
585 ret = rte_eth_tx_burst(port, queueid, m_table, n);
587 core_stats_update_tx(ret);
589 if (unlikely(ret < n)) {
591 free_pkts(&m_table[ret], 1);
599 * Helper function to fragment and queue for TX one packet.
601 static inline uint32_t
602 send_fragment_packet(struct lcore_conf *qconf, struct rte_mbuf *m,
603 uint16_t port, uint8_t proto)
609 tbl = qconf->tx_mbufs + port;
612 /* free space for new fragments */
613 if (len + RTE_LIBRTE_IP_FRAG_MAX_FRAG >= RTE_DIM(tbl->m_table)) {
614 send_burst(qconf, len, port);
618 n = RTE_DIM(tbl->m_table) - len;
620 if (proto == IPPROTO_IP)
621 rc = rte_ipv4_fragment_packet(m, tbl->m_table + len,
622 n, mtu_size, qconf->frag.pool_dir,
623 qconf->frag.pool_indir);
625 rc = rte_ipv6_fragment_packet(m, tbl->m_table + len,
626 n, mtu_size, qconf->frag.pool_dir,
627 qconf->frag.pool_indir);
633 "%s: failed to fragment packet with size %u, "
635 __func__, m->pkt_len, rte_errno);
641 /* Enqueue a single packet, and send burst if queue is filled */
642 static inline int32_t
643 send_single_packet(struct rte_mbuf *m, uint16_t port, uint8_t proto)
647 struct lcore_conf *qconf;
649 lcore_id = rte_lcore_id();
651 qconf = &lcore_conf[lcore_id];
652 len = qconf->tx_mbufs[port].len;
654 if (m->pkt_len <= mtu_size) {
655 qconf->tx_mbufs[port].m_table[len] = m;
658 /* need to fragment the packet */
659 } else if (frag_tbl_sz > 0)
660 len = send_fragment_packet(qconf, m, port, proto);
664 /* enough pkts to be sent */
665 if (unlikely(len == MAX_PKT_BURST)) {
666 send_burst(qconf, MAX_PKT_BURST, port);
670 qconf->tx_mbufs[port].len = len;
675 inbound_sp_sa(struct sp_ctx *sp, struct sa_ctx *sa, struct traffic_type *ip,
679 uint32_t i, j, res, sa_idx;
681 if (ip->num == 0 || sp == NULL)
684 rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
685 ip->num, DEFAULT_MAX_CATEGORIES);
688 for (i = 0; i < ip->num; i++) {
695 if (res == DISCARD) {
700 /* Only check SPI match for processed IPSec packets */
701 if (i < lim && ((m->ol_flags & RTE_MBUF_F_RX_SEC_OFFLOAD) == 0)) {
707 if (!inbound_sa_check(sa, m, sa_idx)) {
717 split46_traffic(struct ipsec_traffic *trf, struct rte_mbuf *mb[], uint32_t num)
726 for (i = 0; i < num; i++) {
729 ip = rte_pktmbuf_mtod(m, struct ip *);
731 if (ip->ip_v == IPVERSION) {
732 trf->ip4.pkts[n4] = m;
733 trf->ip4.data[n4] = rte_pktmbuf_mtod_offset(m,
734 uint8_t *, offsetof(struct ip, ip_p));
736 } else if (ip->ip_v == IP6_VERSION) {
737 trf->ip6.pkts[n6] = m;
738 trf->ip6.data[n6] = rte_pktmbuf_mtod_offset(m,
740 offsetof(struct ip6_hdr, ip6_nxt));
752 process_pkts_inbound(struct ipsec_ctx *ipsec_ctx,
753 struct ipsec_traffic *traffic)
755 uint16_t nb_pkts_in, n_ip4, n_ip6;
757 n_ip4 = traffic->ip4.num;
758 n_ip6 = traffic->ip6.num;
760 if (app_sa_prm.enable == 0) {
761 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
762 traffic->ipsec.num, MAX_PKT_BURST);
763 split46_traffic(traffic, traffic->ipsec.pkts, nb_pkts_in);
765 inbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.pkts,
766 traffic->ipsec.saptr, traffic->ipsec.num);
767 ipsec_process(ipsec_ctx, traffic);
770 inbound_sp_sa(ipsec_ctx->sp4_ctx, ipsec_ctx->sa_ctx, &traffic->ip4,
773 inbound_sp_sa(ipsec_ctx->sp6_ctx, ipsec_ctx->sa_ctx, &traffic->ip6,
778 outbound_sp(struct sp_ctx *sp, struct traffic_type *ip,
779 struct traffic_type *ipsec)
782 uint32_t i, j, sa_idx;
784 if (ip->num == 0 || sp == NULL)
787 rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
788 ip->num, DEFAULT_MAX_CATEGORIES);
791 for (i = 0; i < ip->num; i++) {
793 sa_idx = ip->res[i] - 1;
794 if (ip->res[i] == DISCARD)
796 else if (ip->res[i] == BYPASS)
799 ipsec->res[ipsec->num] = sa_idx;
800 ipsec->pkts[ipsec->num++] = m;
807 process_pkts_outbound(struct ipsec_ctx *ipsec_ctx,
808 struct ipsec_traffic *traffic)
811 uint16_t idx, nb_pkts_out, i;
813 /* Drop any IPsec traffic from protected ports */
814 free_pkts(traffic->ipsec.pkts, traffic->ipsec.num);
816 traffic->ipsec.num = 0;
818 outbound_sp(ipsec_ctx->sp4_ctx, &traffic->ip4, &traffic->ipsec);
820 outbound_sp(ipsec_ctx->sp6_ctx, &traffic->ip6, &traffic->ipsec);
822 if (app_sa_prm.enable == 0) {
824 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
825 traffic->ipsec.res, traffic->ipsec.num,
828 for (i = 0; i < nb_pkts_out; i++) {
829 m = traffic->ipsec.pkts[i];
830 struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
831 if (ip->ip_v == IPVERSION) {
832 idx = traffic->ip4.num++;
833 traffic->ip4.pkts[idx] = m;
835 idx = traffic->ip6.num++;
836 traffic->ip6.pkts[idx] = m;
840 outbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.res,
841 traffic->ipsec.saptr, traffic->ipsec.num);
842 ipsec_process(ipsec_ctx, traffic);
847 process_pkts_inbound_nosp(struct ipsec_ctx *ipsec_ctx,
848 struct ipsec_traffic *traffic)
851 uint32_t nb_pkts_in, i, idx;
853 if (app_sa_prm.enable == 0) {
855 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
856 traffic->ipsec.num, MAX_PKT_BURST);
858 for (i = 0; i < nb_pkts_in; i++) {
859 m = traffic->ipsec.pkts[i];
860 struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
861 if (ip->ip_v == IPVERSION) {
862 idx = traffic->ip4.num++;
863 traffic->ip4.pkts[idx] = m;
865 idx = traffic->ip6.num++;
866 traffic->ip6.pkts[idx] = m;
870 inbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.pkts,
871 traffic->ipsec.saptr, traffic->ipsec.num);
872 ipsec_process(ipsec_ctx, traffic);
877 process_pkts_outbound_nosp(struct ipsec_ctx *ipsec_ctx,
878 struct ipsec_traffic *traffic)
881 uint32_t nb_pkts_out, i, n;
884 /* Drop any IPsec traffic from protected ports */
885 free_pkts(traffic->ipsec.pkts, traffic->ipsec.num);
889 for (i = 0; i < traffic->ip4.num; i++) {
890 traffic->ipsec.pkts[n] = traffic->ip4.pkts[i];
891 traffic->ipsec.res[n++] = single_sa_idx;
894 for (i = 0; i < traffic->ip6.num; i++) {
895 traffic->ipsec.pkts[n] = traffic->ip6.pkts[i];
896 traffic->ipsec.res[n++] = single_sa_idx;
899 traffic->ip4.num = 0;
900 traffic->ip6.num = 0;
901 traffic->ipsec.num = n;
903 if (app_sa_prm.enable == 0) {
905 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
906 traffic->ipsec.res, traffic->ipsec.num,
909 /* They all sue the same SA (ip4 or ip6 tunnel) */
910 m = traffic->ipsec.pkts[0];
911 ip = rte_pktmbuf_mtod(m, struct ip *);
912 if (ip->ip_v == IPVERSION) {
913 traffic->ip4.num = nb_pkts_out;
914 for (i = 0; i < nb_pkts_out; i++)
915 traffic->ip4.pkts[i] = traffic->ipsec.pkts[i];
917 traffic->ip6.num = nb_pkts_out;
918 for (i = 0; i < nb_pkts_out; i++)
919 traffic->ip6.pkts[i] = traffic->ipsec.pkts[i];
922 outbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.res,
923 traffic->ipsec.saptr, traffic->ipsec.num);
924 ipsec_process(ipsec_ctx, traffic);
928 static inline int32_t
929 get_hop_for_offload_pkt(struct rte_mbuf *pkt, int is_ipv6)
931 struct ipsec_mbuf_metadata *priv;
934 priv = get_priv(pkt);
937 if (unlikely(sa == NULL)) {
938 RTE_LOG(ERR, IPSEC, "SA not saved in private data\n");
946 return (sa->portid | RTE_LPM_LOOKUP_SUCCESS);
957 route4_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
959 uint32_t hop[MAX_PKT_BURST * 2];
960 uint32_t dst_ip[MAX_PKT_BURST * 2];
963 uint16_t lpm_pkts = 0;
968 /* Need to do an LPM lookup for non-inline packets. Inline packets will
969 * have port ID in the SA
972 for (i = 0; i < nb_pkts; i++) {
973 if (!(pkts[i]->ol_flags & RTE_MBUF_F_TX_SEC_OFFLOAD)) {
974 /* Security offload not enabled. So an LPM lookup is
975 * required to get the hop
977 offset = offsetof(struct ip, ip_dst);
978 dst_ip[lpm_pkts] = *rte_pktmbuf_mtod_offset(pkts[i],
980 dst_ip[lpm_pkts] = rte_be_to_cpu_32(dst_ip[lpm_pkts]);
985 rte_lpm_lookup_bulk((struct rte_lpm *)rt_ctx, dst_ip, hop, lpm_pkts);
989 for (i = 0; i < nb_pkts; i++) {
990 if (pkts[i]->ol_flags & RTE_MBUF_F_TX_SEC_OFFLOAD) {
991 /* Read hop from the SA */
992 pkt_hop = get_hop_for_offload_pkt(pkts[i], 0);
994 /* Need to use hop returned by lookup */
995 pkt_hop = hop[lpm_pkts++];
998 if ((pkt_hop & RTE_LPM_LOOKUP_SUCCESS) == 0) {
999 free_pkts(&pkts[i], 1);
1002 send_single_packet(pkts[i], pkt_hop & 0xff, IPPROTO_IP);
1007 route6_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
1009 int32_t hop[MAX_PKT_BURST * 2];
1010 uint8_t dst_ip[MAX_PKT_BURST * 2][16];
1012 int32_t pkt_hop = 0;
1014 uint16_t lpm_pkts = 0;
1019 /* Need to do an LPM lookup for non-inline packets. Inline packets will
1020 * have port ID in the SA
1023 for (i = 0; i < nb_pkts; i++) {
1024 if (!(pkts[i]->ol_flags & RTE_MBUF_F_TX_SEC_OFFLOAD)) {
1025 /* Security offload not enabled. So an LPM lookup is
1026 * required to get the hop
1028 offset = offsetof(struct ip6_hdr, ip6_dst);
1029 ip6_dst = rte_pktmbuf_mtod_offset(pkts[i], uint8_t *,
1031 memcpy(&dst_ip[lpm_pkts][0], ip6_dst, 16);
1036 rte_lpm6_lookup_bulk_func((struct rte_lpm6 *)rt_ctx, dst_ip, hop,
1041 for (i = 0; i < nb_pkts; i++) {
1042 if (pkts[i]->ol_flags & RTE_MBUF_F_TX_SEC_OFFLOAD) {
1043 /* Read hop from the SA */
1044 pkt_hop = get_hop_for_offload_pkt(pkts[i], 1);
1046 /* Need to use hop returned by lookup */
1047 pkt_hop = hop[lpm_pkts++];
1050 if (pkt_hop == -1) {
1051 free_pkts(&pkts[i], 1);
1054 send_single_packet(pkts[i], pkt_hop & 0xff, IPPROTO_IPV6);
1059 process_pkts(struct lcore_conf *qconf, struct rte_mbuf **pkts,
1060 uint8_t nb_pkts, uint16_t portid)
1062 struct ipsec_traffic traffic;
1064 prepare_traffic(pkts, &traffic, nb_pkts);
1066 if (unlikely(single_sa)) {
1067 if (is_unprotected_port(portid))
1068 process_pkts_inbound_nosp(&qconf->inbound, &traffic);
1070 process_pkts_outbound_nosp(&qconf->outbound, &traffic);
1072 if (is_unprotected_port(portid))
1073 process_pkts_inbound(&qconf->inbound, &traffic);
1075 process_pkts_outbound(&qconf->outbound, &traffic);
1078 route4_pkts(qconf->rt4_ctx, traffic.ip4.pkts, traffic.ip4.num);
1079 route6_pkts(qconf->rt6_ctx, traffic.ip6.pkts, traffic.ip6.num);
1083 drain_tx_buffers(struct lcore_conf *qconf)
1088 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
1089 buf = &qconf->tx_mbufs[portid];
1092 send_burst(qconf, buf->len, portid);
1098 drain_crypto_buffers(struct lcore_conf *qconf)
1101 struct ipsec_ctx *ctx;
1103 /* drain inbound buffers*/
1104 ctx = &qconf->inbound;
1105 for (i = 0; i != ctx->nb_qps; i++) {
1106 if (ctx->tbl[i].len != 0)
1107 enqueue_cop_burst(ctx->tbl + i);
1110 /* drain outbound buffers*/
1111 ctx = &qconf->outbound;
1112 for (i = 0; i != ctx->nb_qps; i++) {
1113 if (ctx->tbl[i].len != 0)
1114 enqueue_cop_burst(ctx->tbl + i);
1119 drain_inbound_crypto_queues(const struct lcore_conf *qconf,
1120 struct ipsec_ctx *ctx)
1123 struct ipsec_traffic trf;
1125 if (app_sa_prm.enable == 0) {
1127 /* dequeue packets from crypto-queue */
1128 n = ipsec_inbound_cqp_dequeue(ctx, trf.ipsec.pkts,
1129 RTE_DIM(trf.ipsec.pkts));
1134 /* split traffic by ipv4-ipv6 */
1135 split46_traffic(&trf, trf.ipsec.pkts, n);
1137 ipsec_cqp_process(ctx, &trf);
1139 /* process ipv4 packets */
1140 if (trf.ip4.num != 0) {
1141 inbound_sp_sa(ctx->sp4_ctx, ctx->sa_ctx, &trf.ip4, 0);
1142 route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
1145 /* process ipv6 packets */
1146 if (trf.ip6.num != 0) {
1147 inbound_sp_sa(ctx->sp6_ctx, ctx->sa_ctx, &trf.ip6, 0);
1148 route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
1153 drain_outbound_crypto_queues(const struct lcore_conf *qconf,
1154 struct ipsec_ctx *ctx)
1157 struct ipsec_traffic trf;
1159 if (app_sa_prm.enable == 0) {
1161 /* dequeue packets from crypto-queue */
1162 n = ipsec_outbound_cqp_dequeue(ctx, trf.ipsec.pkts,
1163 RTE_DIM(trf.ipsec.pkts));
1168 /* split traffic by ipv4-ipv6 */
1169 split46_traffic(&trf, trf.ipsec.pkts, n);
1171 ipsec_cqp_process(ctx, &trf);
1173 /* process ipv4 packets */
1174 if (trf.ip4.num != 0)
1175 route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
1177 /* process ipv6 packets */
1178 if (trf.ip6.num != 0)
1179 route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
1182 /* main processing loop */
1184 ipsec_poll_mode_worker(void)
1186 struct rte_mbuf *pkts[MAX_PKT_BURST];
1188 uint64_t prev_tsc, diff_tsc, cur_tsc;
1192 struct lcore_conf *qconf;
1193 int32_t rc, socket_id;
1194 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
1195 / US_PER_S * BURST_TX_DRAIN_US;
1196 struct lcore_rx_queue *rxql;
1199 lcore_id = rte_lcore_id();
1200 qconf = &lcore_conf[lcore_id];
1201 rxql = qconf->rx_queue_list;
1202 socket_id = rte_lcore_to_socket_id(lcore_id);
1204 qconf->rt4_ctx = socket_ctx[socket_id].rt_ip4;
1205 qconf->rt6_ctx = socket_ctx[socket_id].rt_ip6;
1206 qconf->inbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_in;
1207 qconf->inbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_in;
1208 qconf->inbound.sa_ctx = socket_ctx[socket_id].sa_in;
1209 qconf->inbound.cdev_map = cdev_map_in;
1210 qconf->inbound.session_pool = socket_ctx[socket_id].session_pool;
1211 qconf->inbound.session_priv_pool =
1212 socket_ctx[socket_id].session_priv_pool;
1213 qconf->outbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_out;
1214 qconf->outbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_out;
1215 qconf->outbound.sa_ctx = socket_ctx[socket_id].sa_out;
1216 qconf->outbound.cdev_map = cdev_map_out;
1217 qconf->outbound.session_pool = socket_ctx[socket_id].session_pool;
1218 qconf->outbound.session_priv_pool =
1219 socket_ctx[socket_id].session_priv_pool;
1220 qconf->frag.pool_dir = socket_ctx[socket_id].mbuf_pool;
1221 qconf->frag.pool_indir = socket_ctx[socket_id].mbuf_pool_indir;
1223 rc = ipsec_sad_lcore_cache_init(app_sa_prm.cache_sz);
1226 "SAD cache init on lcore %u, failed with code: %d\n",
1231 if (qconf->nb_rx_queue == 0) {
1232 RTE_LOG(DEBUG, IPSEC, "lcore %u has nothing to do\n",
1237 RTE_LOG(INFO, IPSEC, "entering main loop on lcore %u\n", lcore_id);
1239 for (i = 0; i < qconf->nb_rx_queue; i++) {
1240 portid = rxql[i].port_id;
1241 queueid = rxql[i].queue_id;
1242 RTE_LOG(INFO, IPSEC,
1243 " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
1244 lcore_id, portid, queueid);
1247 while (!force_quit) {
1248 cur_tsc = rte_rdtsc();
1250 /* TX queue buffer drain */
1251 diff_tsc = cur_tsc - prev_tsc;
1253 if (unlikely(diff_tsc > drain_tsc)) {
1254 drain_tx_buffers(qconf);
1255 drain_crypto_buffers(qconf);
1259 for (i = 0; i < qconf->nb_rx_queue; ++i) {
1261 /* Read packets from RX queues */
1262 portid = rxql[i].port_id;
1263 queueid = rxql[i].queue_id;
1264 nb_rx = rte_eth_rx_burst(portid, queueid,
1265 pkts, MAX_PKT_BURST);
1268 core_stats_update_rx(nb_rx);
1269 process_pkts(qconf, pkts, nb_rx, portid);
1272 /* dequeue and process completed crypto-ops */
1273 if (is_unprotected_port(portid))
1274 drain_inbound_crypto_queues(qconf,
1277 drain_outbound_crypto_queues(qconf,
1284 check_flow_params(uint16_t fdir_portid, uint8_t fdir_qid)
1290 for (i = 0; i < nb_lcore_params; ++i) {
1291 portid = lcore_params_array[i].port_id;
1292 if (portid == fdir_portid) {
1293 queueid = lcore_params_array[i].queue_id;
1294 if (queueid == fdir_qid)
1298 if (i == nb_lcore_params - 1)
1306 check_poll_mode_params(struct eh_conf *eh_conf)
1316 if (eh_conf->mode != EH_PKT_TRANSFER_MODE_POLL)
1319 if (lcore_params == NULL) {
1320 printf("Error: No port/queue/core mappings\n");
1324 for (i = 0; i < nb_lcore_params; ++i) {
1325 lcore = lcore_params[i].lcore_id;
1326 if (!rte_lcore_is_enabled(lcore)) {
1327 printf("error: lcore %hhu is not enabled in "
1328 "lcore mask\n", lcore);
1331 socket_id = rte_lcore_to_socket_id(lcore);
1332 if (socket_id != 0 && numa_on == 0) {
1333 printf("warning: lcore %hhu is on socket %d "
1337 portid = lcore_params[i].port_id;
1338 if ((enabled_port_mask & (1 << portid)) == 0) {
1339 printf("port %u is not enabled in port mask\n", portid);
1342 if (!rte_eth_dev_is_valid_port(portid)) {
1343 printf("port %u is not present on the board\n", portid);
1351 get_port_nb_rx_queues(const uint16_t port)
1356 for (i = 0; i < nb_lcore_params; ++i) {
1357 if (lcore_params[i].port_id == port &&
1358 lcore_params[i].queue_id > queue)
1359 queue = lcore_params[i].queue_id;
1361 return (uint8_t)(++queue);
1365 init_lcore_rx_queues(void)
1367 uint16_t i, nb_rx_queue;
1370 for (i = 0; i < nb_lcore_params; ++i) {
1371 lcore = lcore_params[i].lcore_id;
1372 nb_rx_queue = lcore_conf[lcore].nb_rx_queue;
1373 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1374 printf("error: too many queues (%u) for lcore: %u\n",
1375 nb_rx_queue + 1, lcore);
1378 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1379 lcore_params[i].port_id;
1380 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1381 lcore_params[i].queue_id;
1382 lcore_conf[lcore].nb_rx_queue++;
1389 print_usage(const char *prgname)
1391 fprintf(stderr, "%s [EAL options] --"
1397 " [-w REPLAY_WINDOW_SIZE]"
1401 " [-s NUMBER_OF_MBUFS_IN_PKT_POOL]"
1403 " --config (port,queue,lcore)[,(port,queue,lcore)]"
1404 " [--single-sa SAIDX]"
1405 " [--cryptodev_mask MASK]"
1406 " [--transfer-mode MODE]"
1407 " [--event-schedule-type TYPE]"
1408 " [--" CMD_LINE_OPT_RX_OFFLOAD " RX_OFFLOAD_MASK]"
1409 " [--" CMD_LINE_OPT_TX_OFFLOAD " TX_OFFLOAD_MASK]"
1410 " [--" CMD_LINE_OPT_REASSEMBLE " REASSEMBLE_TABLE_SIZE]"
1411 " [--" CMD_LINE_OPT_MTU " MTU]"
1413 " -p PORTMASK: Hexadecimal bitmask of ports to configure\n"
1414 " -P : Enable promiscuous mode\n"
1415 " -u PORTMASK: Hexadecimal bitmask of unprotected ports\n"
1416 " -j FRAMESIZE: Data buffer size, minimum (and default)\n"
1417 " value: RTE_MBUF_DEFAULT_BUF_SIZE\n"
1418 " -l enables code-path that uses librte_ipsec\n"
1419 " -w REPLAY_WINDOW_SIZE specifies IPsec SQN replay window\n"
1420 " size for each SA\n"
1422 " -a enables SA SQN atomic behaviour\n"
1423 " -c specifies inbound SAD cache size,\n"
1424 " zero value disables the cache (default value: 128)\n"
1425 " -s number of mbufs in packet pool, if not specified number\n"
1426 " of mbufs will be calculated based on number of cores,\n"
1427 " ports and crypto queues\n"
1428 " -f CONFIG_FILE: Configuration file\n"
1429 " --config (port,queue,lcore): Rx queue configuration. In poll\n"
1430 " mode determines which queues from\n"
1431 " which ports are mapped to which cores.\n"
1432 " In event mode this option is not used\n"
1433 " as packets are dynamically scheduled\n"
1434 " to cores by HW.\n"
1435 " --single-sa SAIDX: In poll mode use single SA index for\n"
1436 " outbound traffic, bypassing the SP\n"
1437 " In event mode selects driver submode,\n"
1438 " SA index value is ignored\n"
1439 " --cryptodev_mask MASK: Hexadecimal bitmask of the crypto\n"
1440 " devices to configure\n"
1441 " --transfer-mode MODE\n"
1442 " \"poll\" : Packet transfer via polling (default)\n"
1443 " \"event\" : Packet transfer via event device\n"
1444 " --event-schedule-type TYPE queue schedule type, used only when\n"
1445 " transfer mode is set to event\n"
1446 " \"ordered\" : Ordered (default)\n"
1447 " \"atomic\" : Atomic\n"
1448 " \"parallel\" : Parallel\n"
1449 " --" CMD_LINE_OPT_RX_OFFLOAD
1450 ": bitmask of the RX HW offload capabilities to enable/use\n"
1451 " (RTE_ETH_RX_OFFLOAD_*)\n"
1452 " --" CMD_LINE_OPT_TX_OFFLOAD
1453 ": bitmask of the TX HW offload capabilities to enable/use\n"
1454 " (RTE_ETH_TX_OFFLOAD_*)\n"
1455 " --" CMD_LINE_OPT_REASSEMBLE " NUM"
1456 ": max number of entries in reassemble(fragment) table\n"
1457 " (zero (default value) disables reassembly)\n"
1458 " --" CMD_LINE_OPT_MTU " MTU"
1459 ": MTU value on all ports (default value: 1500)\n"
1460 " outgoing packets with bigger size will be fragmented\n"
1461 " incoming packets with bigger size will be discarded\n"
1462 " --" CMD_LINE_OPT_FRAG_TTL " FRAG_TTL_NS"
1463 ": fragments lifetime in nanoseconds, default\n"
1464 " and maximum value is 10.000.000.000 ns (10 s)\n"
1470 parse_mask(const char *str, uint64_t *val)
1476 t = strtoul(str, &end, 0);
1477 if (errno != 0 || end[0] != 0)
1485 parse_portmask(const char *portmask)
1492 /* parse hexadecimal string */
1493 pm = strtoul(portmask, &end, 16);
1494 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1497 if ((pm == 0) && errno)
1504 parse_decimal(const char *str)
1509 num = strtoull(str, &end, 10);
1510 if ((str[0] == '\0') || (end == NULL) || (*end != '\0')
1518 parse_config(const char *q_arg)
1521 const char *p, *p0 = q_arg;
1529 unsigned long int_fld[_NUM_FLD];
1530 char *str_fld[_NUM_FLD];
1534 nb_lcore_params = 0;
1536 while ((p = strchr(p0, '(')) != NULL) {
1538 p0 = strchr(p, ')');
1543 if (size >= sizeof(s))
1546 snprintf(s, sizeof(s), "%.*s", size, p);
1547 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1550 for (i = 0; i < _NUM_FLD; i++) {
1552 int_fld[i] = strtoul(str_fld[i], &end, 0);
1553 if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
1556 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1557 printf("exceeded max number of lcore params: %hu\n",
1561 lcore_params_array[nb_lcore_params].port_id =
1562 (uint8_t)int_fld[FLD_PORT];
1563 lcore_params_array[nb_lcore_params].queue_id =
1564 (uint8_t)int_fld[FLD_QUEUE];
1565 lcore_params_array[nb_lcore_params].lcore_id =
1566 (uint8_t)int_fld[FLD_LCORE];
1569 lcore_params = lcore_params_array;
1574 print_app_sa_prm(const struct app_sa_prm *prm)
1576 printf("librte_ipsec usage: %s\n",
1577 (prm->enable == 0) ? "disabled" : "enabled");
1579 printf("replay window size: %u\n", prm->window_size);
1580 printf("ESN: %s\n", (prm->enable_esn == 0) ? "disabled" : "enabled");
1581 printf("SA flags: %#" PRIx64 "\n", prm->flags);
1582 printf("Frag TTL: %" PRIu64 " ns\n", frag_ttl_ns);
1586 parse_transfer_mode(struct eh_conf *conf, const char *optarg)
1588 if (!strcmp(CMD_LINE_ARG_POLL, optarg))
1589 conf->mode = EH_PKT_TRANSFER_MODE_POLL;
1590 else if (!strcmp(CMD_LINE_ARG_EVENT, optarg))
1591 conf->mode = EH_PKT_TRANSFER_MODE_EVENT;
1593 printf("Unsupported packet transfer mode\n");
1601 parse_schedule_type(struct eh_conf *conf, const char *optarg)
1603 struct eventmode_conf *em_conf = NULL;
1605 /* Get eventmode conf */
1606 em_conf = conf->mode_params;
1608 if (!strcmp(CMD_LINE_ARG_ORDERED, optarg))
1609 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_ORDERED;
1610 else if (!strcmp(CMD_LINE_ARG_ATOMIC, optarg))
1611 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_ATOMIC;
1612 else if (!strcmp(CMD_LINE_ARG_PARALLEL, optarg))
1613 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_PARALLEL;
1615 printf("Unsupported queue schedule type\n");
1623 parse_args(int32_t argc, char **argv, struct eh_conf *eh_conf)
1628 int32_t option_index;
1629 char *prgname = argv[0];
1630 int32_t f_present = 0;
1634 while ((opt = getopt_long(argc, argvopt, "aelp:Pu:f:j:w:c:s:",
1635 lgopts, &option_index)) != EOF) {
1639 enabled_port_mask = parse_portmask(optarg);
1640 if (enabled_port_mask == 0) {
1641 printf("invalid portmask\n");
1642 print_usage(prgname);
1647 printf("Promiscuous mode selected\n");
1651 unprotected_port_mask = parse_portmask(optarg);
1652 if (unprotected_port_mask == 0) {
1653 printf("invalid unprotected portmask\n");
1654 print_usage(prgname);
1659 if (f_present == 1) {
1660 printf("\"-f\" option present more than "
1662 print_usage(prgname);
1670 ret = parse_decimal(optarg);
1672 printf("Invalid number of buffers in a pool: "
1674 print_usage(prgname);
1678 nb_bufs_in_pool = ret;
1682 ret = parse_decimal(optarg);
1683 if (ret < RTE_MBUF_DEFAULT_BUF_SIZE ||
1685 printf("Invalid frame buffer size value: %s\n",
1687 print_usage(prgname);
1690 frame_buf_size = ret;
1691 printf("Custom frame buffer size %u\n", frame_buf_size);
1694 app_sa_prm.enable = 1;
1697 app_sa_prm.window_size = parse_decimal(optarg);
1700 app_sa_prm.enable_esn = 1;
1703 app_sa_prm.enable = 1;
1704 app_sa_prm.flags |= RTE_IPSEC_SAFLAG_SQN_ATOM;
1707 ret = parse_decimal(optarg);
1709 printf("Invalid SA cache size: %s\n", optarg);
1710 print_usage(prgname);
1713 app_sa_prm.cache_sz = ret;
1715 case CMD_LINE_OPT_CONFIG_NUM:
1716 ret = parse_config(optarg);
1718 printf("Invalid config\n");
1719 print_usage(prgname);
1723 case CMD_LINE_OPT_SINGLE_SA_NUM:
1724 ret = parse_decimal(optarg);
1725 if (ret == -1 || ret > UINT32_MAX) {
1726 printf("Invalid argument[sa_idx]\n");
1727 print_usage(prgname);
1733 single_sa_idx = ret;
1734 eh_conf->ipsec_mode = EH_IPSEC_MODE_TYPE_DRIVER;
1735 printf("Configured with single SA index %u\n",
1738 case CMD_LINE_OPT_CRYPTODEV_MASK_NUM:
1739 ret = parse_portmask(optarg);
1741 printf("Invalid argument[portmask]\n");
1742 print_usage(prgname);
1747 enabled_cryptodev_mask = ret;
1750 case CMD_LINE_OPT_TRANSFER_MODE_NUM:
1751 ret = parse_transfer_mode(eh_conf, optarg);
1753 printf("Invalid packet transfer mode\n");
1754 print_usage(prgname);
1759 case CMD_LINE_OPT_SCHEDULE_TYPE_NUM:
1760 ret = parse_schedule_type(eh_conf, optarg);
1762 printf("Invalid queue schedule type\n");
1763 print_usage(prgname);
1768 case CMD_LINE_OPT_RX_OFFLOAD_NUM:
1769 ret = parse_mask(optarg, &dev_rx_offload);
1771 printf("Invalid argument for \'%s\': %s\n",
1772 CMD_LINE_OPT_RX_OFFLOAD, optarg);
1773 print_usage(prgname);
1777 case CMD_LINE_OPT_TX_OFFLOAD_NUM:
1778 ret = parse_mask(optarg, &dev_tx_offload);
1780 printf("Invalid argument for \'%s\': %s\n",
1781 CMD_LINE_OPT_TX_OFFLOAD, optarg);
1782 print_usage(prgname);
1786 case CMD_LINE_OPT_REASSEMBLE_NUM:
1787 ret = parse_decimal(optarg);
1788 if (ret < 0 || ret > UINT32_MAX) {
1789 printf("Invalid argument for \'%s\': %s\n",
1790 CMD_LINE_OPT_REASSEMBLE, optarg);
1791 print_usage(prgname);
1796 case CMD_LINE_OPT_MTU_NUM:
1797 ret = parse_decimal(optarg);
1798 if (ret < 0 || ret > RTE_IPV4_MAX_PKT_LEN) {
1799 printf("Invalid argument for \'%s\': %s\n",
1800 CMD_LINE_OPT_MTU, optarg);
1801 print_usage(prgname);
1806 case CMD_LINE_OPT_FRAG_TTL_NUM:
1807 ret = parse_decimal(optarg);
1808 if (ret < 0 || ret > MAX_FRAG_TTL_NS) {
1809 printf("Invalid argument for \'%s\': %s\n",
1810 CMD_LINE_OPT_MTU, optarg);
1811 print_usage(prgname);
1817 print_usage(prgname);
1822 if (f_present == 0) {
1823 printf("Mandatory option \"-f\" not present\n");
1827 /* check do we need to enable multi-seg support */
1828 if (multi_seg_required()) {
1829 /* legacy mode doesn't support multi-seg */
1830 app_sa_prm.enable = 1;
1831 printf("frame buf size: %u, mtu: %u, "
1832 "number of reassemble entries: %u\n"
1833 "multi-segment support is required\n",
1834 frame_buf_size, mtu_size, frag_tbl_sz);
1837 print_app_sa_prm(&app_sa_prm);
1840 argv[optind-1] = prgname;
1843 optind = 1; /* reset getopt lib */
1848 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
1850 char buf[RTE_ETHER_ADDR_FMT_SIZE];
1851 rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
1852 printf("%s%s", name, buf);
1856 * Update destination ethaddr for the port.
1859 add_dst_ethaddr(uint16_t port, const struct rte_ether_addr *addr)
1861 if (port >= RTE_DIM(ethaddr_tbl))
1864 ethaddr_tbl[port].dst = ETHADDR_TO_UINT64(addr);
1868 /* Check the link status of all ports in up to 9s, and print them finally */
1870 check_all_ports_link_status(uint32_t port_mask)
1872 #define CHECK_INTERVAL 100 /* 100ms */
1873 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1875 uint8_t count, all_ports_up, print_flag = 0;
1876 struct rte_eth_link link;
1878 char link_status_text[RTE_ETH_LINK_MAX_STR_LEN];
1880 printf("\nChecking link status");
1882 for (count = 0; count <= MAX_CHECK_TIME; count++) {
1884 RTE_ETH_FOREACH_DEV(portid) {
1885 if ((port_mask & (1 << portid)) == 0)
1887 memset(&link, 0, sizeof(link));
1888 ret = rte_eth_link_get_nowait(portid, &link);
1891 if (print_flag == 1)
1892 printf("Port %u link get failed: %s\n",
1893 portid, rte_strerror(-ret));
1896 /* print link status if flag set */
1897 if (print_flag == 1) {
1898 rte_eth_link_to_str(link_status_text,
1899 sizeof(link_status_text), &link);
1900 printf("Port %d %s\n", portid,
1904 /* clear all_ports_up flag if any link down */
1905 if (link.link_status == RTE_ETH_LINK_DOWN) {
1910 /* after finally printing all link status, get out */
1911 if (print_flag == 1)
1914 if (all_ports_up == 0) {
1917 rte_delay_ms(CHECK_INTERVAL);
1920 /* set the print_flag if all ports up or timeout */
1921 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1929 add_mapping(struct rte_hash *map, const char *str, uint16_t cdev_id,
1930 uint16_t qp, struct lcore_params *params,
1931 struct ipsec_ctx *ipsec_ctx,
1932 const struct rte_cryptodev_capabilities *cipher,
1933 const struct rte_cryptodev_capabilities *auth,
1934 const struct rte_cryptodev_capabilities *aead)
1938 struct cdev_key key = { 0 };
1940 key.lcore_id = params->lcore_id;
1942 key.cipher_algo = cipher->sym.cipher.algo;
1944 key.auth_algo = auth->sym.auth.algo;
1946 key.aead_algo = aead->sym.aead.algo;
1948 ret = rte_hash_lookup(map, &key);
1952 for (i = 0; i < ipsec_ctx->nb_qps; i++)
1953 if (ipsec_ctx->tbl[i].id == cdev_id)
1956 if (i == ipsec_ctx->nb_qps) {
1957 if (ipsec_ctx->nb_qps == MAX_QP_PER_LCORE) {
1958 printf("Maximum number of crypto devices assigned to "
1959 "a core, increase MAX_QP_PER_LCORE value\n");
1962 ipsec_ctx->tbl[i].id = cdev_id;
1963 ipsec_ctx->tbl[i].qp = qp;
1964 ipsec_ctx->nb_qps++;
1965 printf("%s cdev mapping: lcore %u using cdev %u qp %u "
1966 "(cdev_id_qp %lu)\n", str, key.lcore_id,
1970 ret = rte_hash_add_key_data(map, &key, (void *)i);
1972 printf("Faled to insert cdev mapping for (lcore %u, "
1973 "cdev %u, qp %u), errno %d\n",
1974 key.lcore_id, ipsec_ctx->tbl[i].id,
1975 ipsec_ctx->tbl[i].qp, ret);
1983 add_cdev_mapping(struct rte_cryptodev_info *dev_info, uint16_t cdev_id,
1984 uint16_t qp, struct lcore_params *params)
1987 const struct rte_cryptodev_capabilities *i, *j;
1988 struct rte_hash *map;
1989 struct lcore_conf *qconf;
1990 struct ipsec_ctx *ipsec_ctx;
1993 qconf = &lcore_conf[params->lcore_id];
1995 if ((unprotected_port_mask & (1 << params->port_id)) == 0) {
1997 ipsec_ctx = &qconf->outbound;
2001 ipsec_ctx = &qconf->inbound;
2005 /* Required cryptodevs with operation chainning */
2006 if (!(dev_info->feature_flags &
2007 RTE_CRYPTODEV_FF_SYM_OPERATION_CHAINING))
2010 for (i = dev_info->capabilities;
2011 i->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; i++) {
2012 if (i->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
2015 if (i->sym.xform_type == RTE_CRYPTO_SYM_XFORM_AEAD) {
2016 ret |= add_mapping(map, str, cdev_id, qp, params,
2017 ipsec_ctx, NULL, NULL, i);
2021 if (i->sym.xform_type != RTE_CRYPTO_SYM_XFORM_CIPHER)
2024 for (j = dev_info->capabilities;
2025 j->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; j++) {
2026 if (j->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
2029 if (j->sym.xform_type != RTE_CRYPTO_SYM_XFORM_AUTH)
2032 ret |= add_mapping(map, str, cdev_id, qp, params,
2033 ipsec_ctx, i, j, NULL);
2040 /* Check if the device is enabled by cryptodev_mask */
2042 check_cryptodev_mask(uint8_t cdev_id)
2044 if (enabled_cryptodev_mask & (1 << cdev_id))
2051 cryptodevs_init(uint16_t req_queue_num)
2053 struct rte_cryptodev_config dev_conf;
2054 struct rte_cryptodev_qp_conf qp_conf;
2055 uint16_t idx, max_nb_qps, qp, total_nb_qps, i;
2057 struct rte_hash_parameters params = { 0 };
2059 const uint64_t mseg_flag = multi_seg_required() ?
2060 RTE_CRYPTODEV_FF_IN_PLACE_SGL : 0;
2062 params.entries = CDEV_MAP_ENTRIES;
2063 params.key_len = sizeof(struct cdev_key);
2064 params.hash_func = rte_jhash;
2065 params.hash_func_init_val = 0;
2066 params.socket_id = rte_socket_id();
2068 params.name = "cdev_map_in";
2069 cdev_map_in = rte_hash_create(¶ms);
2070 if (cdev_map_in == NULL)
2071 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
2074 params.name = "cdev_map_out";
2075 cdev_map_out = rte_hash_create(¶ms);
2076 if (cdev_map_out == NULL)
2077 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
2080 printf("lcore/cryptodev/qp mappings:\n");
2084 for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
2085 struct rte_cryptodev_info cdev_info;
2087 if (check_cryptodev_mask((uint8_t)cdev_id))
2090 rte_cryptodev_info_get(cdev_id, &cdev_info);
2092 if ((mseg_flag & cdev_info.feature_flags) != mseg_flag)
2093 rte_exit(EXIT_FAILURE,
2094 "Device %hd does not support \'%s\' feature\n",
2096 rte_cryptodev_get_feature_name(mseg_flag));
2098 if (nb_lcore_params > cdev_info.max_nb_queue_pairs)
2099 max_nb_qps = cdev_info.max_nb_queue_pairs;
2101 max_nb_qps = nb_lcore_params;
2105 while (qp < max_nb_qps && i < nb_lcore_params) {
2106 if (add_cdev_mapping(&cdev_info, cdev_id, qp,
2107 &lcore_params[idx]))
2110 idx = idx % nb_lcore_params;
2114 qp = RTE_MIN(max_nb_qps, RTE_MAX(req_queue_num, qp));
2119 dev_conf.socket_id = rte_cryptodev_socket_id(cdev_id);
2120 dev_conf.nb_queue_pairs = qp;
2121 dev_conf.ff_disable = RTE_CRYPTODEV_FF_ASYMMETRIC_CRYPTO;
2123 uint32_t dev_max_sess = cdev_info.sym.max_nb_sessions;
2124 if (dev_max_sess != 0 &&
2125 dev_max_sess < get_nb_crypto_sessions())
2126 rte_exit(EXIT_FAILURE,
2127 "Device does not support at least %u "
2128 "sessions", get_nb_crypto_sessions());
2130 if (rte_cryptodev_configure(cdev_id, &dev_conf))
2131 rte_panic("Failed to initialize cryptodev %u\n",
2134 qp_conf.nb_descriptors = CDEV_QUEUE_DESC;
2135 qp_conf.mp_session =
2136 socket_ctx[dev_conf.socket_id].session_pool;
2137 qp_conf.mp_session_private =
2138 socket_ctx[dev_conf.socket_id].session_priv_pool;
2139 for (qp = 0; qp < dev_conf.nb_queue_pairs; qp++)
2140 if (rte_cryptodev_queue_pair_setup(cdev_id, qp,
2141 &qp_conf, dev_conf.socket_id))
2142 rte_panic("Failed to setup queue %u for "
2143 "cdev_id %u\n", 0, cdev_id);
2145 if (rte_cryptodev_start(cdev_id))
2146 rte_panic("Failed to start cryptodev %u\n",
2152 return total_nb_qps;
2156 port_init(uint16_t portid, uint64_t req_rx_offloads, uint64_t req_tx_offloads)
2158 struct rte_eth_dev_info dev_info;
2159 struct rte_eth_txconf *txconf;
2160 uint16_t nb_tx_queue, nb_rx_queue;
2161 uint16_t tx_queueid, rx_queueid, queue, lcore_id;
2162 int32_t ret, socket_id;
2163 struct lcore_conf *qconf;
2164 struct rte_ether_addr ethaddr;
2165 struct rte_eth_conf local_port_conf = port_conf;
2167 ret = rte_eth_dev_info_get(portid, &dev_info);
2169 rte_exit(EXIT_FAILURE,
2170 "Error during getting device (port %u) info: %s\n",
2171 portid, strerror(-ret));
2173 /* limit allowed HW offloafs, as user requested */
2174 dev_info.rx_offload_capa &= dev_rx_offload;
2175 dev_info.tx_offload_capa &= dev_tx_offload;
2177 printf("Configuring device port %u:\n", portid);
2179 ret = rte_eth_macaddr_get(portid, ðaddr);
2181 rte_exit(EXIT_FAILURE,
2182 "Error getting MAC address (port %u): %s\n",
2183 portid, rte_strerror(-ret));
2185 ethaddr_tbl[portid].src = ETHADDR_TO_UINT64(ðaddr);
2186 print_ethaddr("Address: ", ðaddr);
2189 nb_rx_queue = get_port_nb_rx_queues(portid);
2190 nb_tx_queue = nb_lcores;
2192 if (nb_rx_queue > dev_info.max_rx_queues)
2193 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
2194 "(max rx queue is %u)\n",
2195 nb_rx_queue, dev_info.max_rx_queues);
2197 if (nb_tx_queue > dev_info.max_tx_queues)
2198 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
2199 "(max tx queue is %u)\n",
2200 nb_tx_queue, dev_info.max_tx_queues);
2202 printf("Creating queues: nb_rx_queue=%d nb_tx_queue=%u...\n",
2203 nb_rx_queue, nb_tx_queue);
2205 local_port_conf.rxmode.mtu = mtu_size;
2207 if (multi_seg_required()) {
2208 local_port_conf.rxmode.offloads |= RTE_ETH_RX_OFFLOAD_SCATTER;
2209 local_port_conf.txmode.offloads |= RTE_ETH_TX_OFFLOAD_MULTI_SEGS;
2212 local_port_conf.rxmode.offloads |= req_rx_offloads;
2213 local_port_conf.txmode.offloads |= req_tx_offloads;
2215 /* Check that all required capabilities are supported */
2216 if ((local_port_conf.rxmode.offloads & dev_info.rx_offload_capa) !=
2217 local_port_conf.rxmode.offloads)
2218 rte_exit(EXIT_FAILURE,
2219 "Error: port %u required RX offloads: 0x%" PRIx64
2220 ", avaialbe RX offloads: 0x%" PRIx64 "\n",
2221 portid, local_port_conf.rxmode.offloads,
2222 dev_info.rx_offload_capa);
2224 if ((local_port_conf.txmode.offloads & dev_info.tx_offload_capa) !=
2225 local_port_conf.txmode.offloads)
2226 rte_exit(EXIT_FAILURE,
2227 "Error: port %u required TX offloads: 0x%" PRIx64
2228 ", avaialbe TX offloads: 0x%" PRIx64 "\n",
2229 portid, local_port_conf.txmode.offloads,
2230 dev_info.tx_offload_capa);
2232 if (dev_info.tx_offload_capa & RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE)
2233 local_port_conf.txmode.offloads |=
2234 RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE;
2236 if (dev_info.tx_offload_capa & RTE_ETH_TX_OFFLOAD_IPV4_CKSUM)
2237 local_port_conf.txmode.offloads |= RTE_ETH_TX_OFFLOAD_IPV4_CKSUM;
2239 printf("port %u configurng rx_offloads=0x%" PRIx64
2240 ", tx_offloads=0x%" PRIx64 "\n",
2241 portid, local_port_conf.rxmode.offloads,
2242 local_port_conf.txmode.offloads);
2244 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
2245 dev_info.flow_type_rss_offloads;
2246 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
2247 port_conf.rx_adv_conf.rss_conf.rss_hf) {
2248 printf("Port %u modified RSS hash function based on hardware support,"
2249 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
2251 port_conf.rx_adv_conf.rss_conf.rss_hf,
2252 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
2255 ret = rte_eth_dev_configure(portid, nb_rx_queue, nb_tx_queue,
2258 rte_exit(EXIT_FAILURE, "Cannot configure device: "
2259 "err=%d, port=%d\n", ret, portid);
2261 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd, &nb_txd);
2263 rte_exit(EXIT_FAILURE, "Cannot adjust number of descriptors: "
2264 "err=%d, port=%d\n", ret, portid);
2266 /* init one TX queue per lcore */
2268 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2269 if (rte_lcore_is_enabled(lcore_id) == 0)
2273 socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2278 printf("Setup txq=%u,%d,%d\n", lcore_id, tx_queueid, socket_id);
2280 txconf = &dev_info.default_txconf;
2281 txconf->offloads = local_port_conf.txmode.offloads;
2283 ret = rte_eth_tx_queue_setup(portid, tx_queueid, nb_txd,
2286 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
2287 "err=%d, port=%d\n", ret, portid);
2289 qconf = &lcore_conf[lcore_id];
2290 qconf->tx_queue_id[portid] = tx_queueid;
2292 /* Pre-populate pkt offloads based on capabilities */
2293 qconf->outbound.ipv4_offloads = RTE_MBUF_F_TX_IPV4;
2294 qconf->outbound.ipv6_offloads = RTE_MBUF_F_TX_IPV6;
2295 if (local_port_conf.txmode.offloads & RTE_ETH_TX_OFFLOAD_IPV4_CKSUM)
2296 qconf->outbound.ipv4_offloads |= RTE_MBUF_F_TX_IP_CKSUM;
2300 /* init RX queues */
2301 for (queue = 0; queue < qconf->nb_rx_queue; ++queue) {
2302 struct rte_eth_rxconf rxq_conf;
2304 if (portid != qconf->rx_queue_list[queue].port_id)
2307 rx_queueid = qconf->rx_queue_list[queue].queue_id;
2309 printf("Setup rxq=%d,%d,%d\n", portid, rx_queueid,
2312 rxq_conf = dev_info.default_rxconf;
2313 rxq_conf.offloads = local_port_conf.rxmode.offloads;
2314 ret = rte_eth_rx_queue_setup(portid, rx_queueid,
2315 nb_rxd, socket_id, &rxq_conf,
2316 socket_ctx[socket_id].mbuf_pool);
2318 rte_exit(EXIT_FAILURE,
2319 "rte_eth_rx_queue_setup: err=%d, "
2320 "port=%d\n", ret, portid);
2327 max_session_size(void)
2331 int16_t cdev_id, port_id, n;
2334 n = rte_cryptodev_count();
2335 for (cdev_id = 0; cdev_id != n; cdev_id++) {
2336 sz = rte_cryptodev_sym_get_private_session_size(cdev_id);
2340 * If crypto device is security capable, need to check the
2341 * size of security session as well.
2344 /* Get security context of the crypto device */
2345 sec_ctx = rte_cryptodev_get_sec_ctx(cdev_id);
2346 if (sec_ctx == NULL)
2349 /* Get size of security session */
2350 sz = rte_security_session_get_size(sec_ctx);
2355 RTE_ETH_FOREACH_DEV(port_id) {
2356 if ((enabled_port_mask & (1 << port_id)) == 0)
2359 sec_ctx = rte_eth_dev_get_sec_ctx(port_id);
2360 if (sec_ctx == NULL)
2363 sz = rte_security_session_get_size(sec_ctx);
2372 session_pool_init(struct socket_ctx *ctx, int32_t socket_id, size_t sess_sz)
2374 char mp_name[RTE_MEMPOOL_NAMESIZE];
2375 struct rte_mempool *sess_mp;
2378 snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
2379 "sess_mp_%u", socket_id);
2380 nb_sess = (get_nb_crypto_sessions() + CDEV_MP_CACHE_SZ *
2382 nb_sess = RTE_MAX(nb_sess, CDEV_MP_CACHE_SZ *
2383 CDEV_MP_CACHE_MULTIPLIER);
2384 sess_mp = rte_cryptodev_sym_session_pool_create(
2385 mp_name, nb_sess, sess_sz, CDEV_MP_CACHE_SZ, 0,
2387 ctx->session_pool = sess_mp;
2389 if (ctx->session_pool == NULL)
2390 rte_exit(EXIT_FAILURE,
2391 "Cannot init session pool on socket %d\n", socket_id);
2393 printf("Allocated session pool on socket %d\n", socket_id);
2397 session_priv_pool_init(struct socket_ctx *ctx, int32_t socket_id,
2400 char mp_name[RTE_MEMPOOL_NAMESIZE];
2401 struct rte_mempool *sess_mp;
2404 snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
2405 "sess_mp_priv_%u", socket_id);
2406 nb_sess = (get_nb_crypto_sessions() + CDEV_MP_CACHE_SZ *
2408 nb_sess = RTE_MAX(nb_sess, CDEV_MP_CACHE_SZ *
2409 CDEV_MP_CACHE_MULTIPLIER);
2410 sess_mp = rte_mempool_create(mp_name,
2414 0, NULL, NULL, NULL,
2417 ctx->session_priv_pool = sess_mp;
2419 if (ctx->session_priv_pool == NULL)
2420 rte_exit(EXIT_FAILURE,
2421 "Cannot init session priv pool on socket %d\n",
2424 printf("Allocated session priv pool on socket %d\n",
2429 pool_init(struct socket_ctx *ctx, int32_t socket_id, uint32_t nb_mbuf)
2434 snprintf(s, sizeof(s), "mbuf_pool_%d", socket_id);
2435 ctx->mbuf_pool = rte_pktmbuf_pool_create(s, nb_mbuf,
2436 MEMPOOL_CACHE_SIZE, ipsec_metadata_size(),
2437 frame_buf_size, socket_id);
2440 * if multi-segment support is enabled, then create a pool
2441 * for indirect mbufs.
2443 ms = multi_seg_required();
2445 snprintf(s, sizeof(s), "mbuf_pool_indir_%d", socket_id);
2446 ctx->mbuf_pool_indir = rte_pktmbuf_pool_create(s, nb_mbuf,
2447 MEMPOOL_CACHE_SIZE, 0, 0, socket_id);
2450 if (ctx->mbuf_pool == NULL || (ms != 0 && ctx->mbuf_pool_indir == NULL))
2451 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool on socket %d\n",
2454 printf("Allocated mbuf pool on socket %d\n", socket_id);
2458 inline_ipsec_event_esn_overflow(struct rte_security_ctx *ctx, uint64_t md)
2460 struct ipsec_sa *sa;
2462 /* For inline protocol processing, the metadata in the event will
2463 * uniquely identify the security session which raised the event.
2464 * Application would then need the userdata it had registered with the
2465 * security session to process the event.
2468 sa = (struct ipsec_sa *)rte_security_get_userdata(ctx, md);
2471 /* userdata could not be retrieved */
2475 /* Sequence number over flow. SA need to be re-established */
2481 inline_ipsec_event_callback(uint16_t port_id, enum rte_eth_event_type type,
2482 void *param, void *ret_param)
2485 struct rte_eth_event_ipsec_desc *event_desc = NULL;
2486 struct rte_security_ctx *ctx = (struct rte_security_ctx *)
2487 rte_eth_dev_get_sec_ctx(port_id);
2489 RTE_SET_USED(param);
2491 if (type != RTE_ETH_EVENT_IPSEC)
2494 event_desc = ret_param;
2495 if (event_desc == NULL) {
2496 printf("Event descriptor not set\n");
2500 md = event_desc->metadata;
2502 if (event_desc->subtype == RTE_ETH_EVENT_IPSEC_ESN_OVERFLOW)
2503 return inline_ipsec_event_esn_overflow(ctx, md);
2504 else if (event_desc->subtype >= RTE_ETH_EVENT_IPSEC_MAX) {
2505 printf("Invalid IPsec event reported\n");
2513 rx_callback(__rte_unused uint16_t port, __rte_unused uint16_t queue,
2514 struct rte_mbuf *pkt[], uint16_t nb_pkts,
2515 __rte_unused uint16_t max_pkts, void *user_param)
2519 struct lcore_conf *lc;
2520 struct rte_mbuf *mb;
2521 struct rte_ether_hdr *eth;
2527 for (i = 0; i != nb_pkts; i++) {
2530 eth = rte_pktmbuf_mtod(mb, struct rte_ether_hdr *);
2531 if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
2533 struct rte_ipv4_hdr *iph;
2535 iph = (struct rte_ipv4_hdr *)(eth + 1);
2536 if (rte_ipv4_frag_pkt_is_fragmented(iph)) {
2538 mb->l2_len = sizeof(*eth);
2539 mb->l3_len = sizeof(*iph);
2540 tm = (tm != 0) ? tm : rte_rdtsc();
2541 mb = rte_ipv4_frag_reassemble_packet(
2542 lc->frag.tbl, &lc->frag.dr,
2546 /* fix ip cksum after reassemble. */
2547 iph = rte_pktmbuf_mtod_offset(mb,
2548 struct rte_ipv4_hdr *,
2550 iph->hdr_checksum = 0;
2551 iph->hdr_checksum = rte_ipv4_cksum(iph);
2554 } else if (eth->ether_type ==
2555 rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6)) {
2557 struct rte_ipv6_hdr *iph;
2558 struct ipv6_extension_fragment *fh;
2560 iph = (struct rte_ipv6_hdr *)(eth + 1);
2561 fh = rte_ipv6_frag_get_ipv6_fragment_header(iph);
2563 mb->l2_len = sizeof(*eth);
2564 mb->l3_len = (uintptr_t)fh - (uintptr_t)iph +
2566 tm = (tm != 0) ? tm : rte_rdtsc();
2567 mb = rte_ipv6_frag_reassemble_packet(
2568 lc->frag.tbl, &lc->frag.dr,
2571 /* fix l3_len after reassemble. */
2572 mb->l3_len = mb->l3_len - sizeof(*fh);
2580 /* some fragments were encountered, drain death row */
2582 rte_ip_frag_free_death_row(&lc->frag.dr, 0);
2589 reassemble_lcore_init(struct lcore_conf *lc, uint32_t cid)
2593 uint64_t frag_cycles;
2594 const struct lcore_rx_queue *rxq;
2595 const struct rte_eth_rxtx_callback *cb;
2597 /* create fragment table */
2598 sid = rte_lcore_to_socket_id(cid);
2599 frag_cycles = (rte_get_tsc_hz() + NS_PER_S - 1) /
2600 NS_PER_S * frag_ttl_ns;
2602 lc->frag.tbl = rte_ip_frag_table_create(frag_tbl_sz,
2603 FRAG_TBL_BUCKET_ENTRIES, frag_tbl_sz, frag_cycles, sid);
2604 if (lc->frag.tbl == NULL) {
2605 printf("%s(%u): failed to create fragment table of size: %u, "
2607 __func__, cid, frag_tbl_sz, rte_errno);
2611 /* setup reassemble RX callbacks for all queues */
2612 for (i = 0; i != lc->nb_rx_queue; i++) {
2614 rxq = lc->rx_queue_list + i;
2615 cb = rte_eth_add_rx_callback(rxq->port_id, rxq->queue_id,
2618 printf("%s(%u): failed to install RX callback for "
2619 "portid=%u, queueid=%u, error code: %d\n",
2621 rxq->port_id, rxq->queue_id, rte_errno);
2630 reassemble_init(void)
2636 for (i = 0; i != nb_lcore_params; i++) {
2637 lc = lcore_params[i].lcore_id;
2638 rc = reassemble_lcore_init(lcore_conf + lc, lc);
2647 create_default_ipsec_flow(uint16_t port_id, uint64_t rx_offloads)
2649 struct rte_flow_action action[2];
2650 struct rte_flow_item pattern[2];
2651 struct rte_flow_attr attr = {0};
2652 struct rte_flow_error err;
2653 struct rte_flow *flow;
2656 if (!(rx_offloads & RTE_ETH_RX_OFFLOAD_SECURITY))
2659 /* Add the default rte_flow to enable SECURITY for all ESP packets */
2661 pattern[0].type = RTE_FLOW_ITEM_TYPE_ESP;
2662 pattern[0].spec = NULL;
2663 pattern[0].mask = NULL;
2664 pattern[0].last = NULL;
2665 pattern[1].type = RTE_FLOW_ITEM_TYPE_END;
2667 action[0].type = RTE_FLOW_ACTION_TYPE_SECURITY;
2668 action[0].conf = NULL;
2669 action[1].type = RTE_FLOW_ACTION_TYPE_END;
2670 action[1].conf = NULL;
2674 ret = rte_flow_validate(port_id, &attr, pattern, action, &err);
2678 flow = rte_flow_create(port_id, &attr, pattern, action, &err);
2682 flow_info_tbl[port_id].rx_def_flow = flow;
2683 RTE_LOG(INFO, IPSEC,
2684 "Created default flow enabling SECURITY for all ESP traffic on port %d\n",
2689 signal_handler(int signum)
2691 if (signum == SIGINT || signum == SIGTERM) {
2692 printf("\n\nSignal %d received, preparing to exit...\n",
2699 ev_mode_sess_verify(struct ipsec_sa *sa, int nb_sa)
2701 struct rte_ipsec_session *ips;
2707 for (i = 0; i < nb_sa; i++) {
2708 ips = ipsec_get_primary_session(&sa[i]);
2709 if (ips->type != RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL)
2710 rte_exit(EXIT_FAILURE, "Event mode supports only "
2711 "inline protocol sessions\n");
2717 check_event_mode_params(struct eh_conf *eh_conf)
2719 struct eventmode_conf *em_conf = NULL;
2720 struct lcore_params *params;
2723 if (!eh_conf || !eh_conf->mode_params)
2726 /* Get eventmode conf */
2727 em_conf = eh_conf->mode_params;
2729 if (eh_conf->mode == EH_PKT_TRANSFER_MODE_POLL &&
2730 em_conf->ext_params.sched_type != SCHED_TYPE_NOT_SET) {
2731 printf("error: option --event-schedule-type applies only to "
2736 if (eh_conf->mode != EH_PKT_TRANSFER_MODE_EVENT)
2739 /* Set schedule type to ORDERED if it wasn't explicitly set by user */
2740 if (em_conf->ext_params.sched_type == SCHED_TYPE_NOT_SET)
2741 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_ORDERED;
2744 * Event mode currently supports only inline protocol sessions.
2745 * If there are other types of sessions configured then exit with
2748 ev_mode_sess_verify(sa_in, nb_sa_in);
2749 ev_mode_sess_verify(sa_out, nb_sa_out);
2752 /* Option --config does not apply to event mode */
2753 if (nb_lcore_params > 0) {
2754 printf("error: option --config applies only to poll mode\n");
2759 * In order to use the same port_init routine for both poll and event
2760 * modes initialize lcore_params with one queue for each eth port
2762 lcore_params = lcore_params_array;
2763 RTE_ETH_FOREACH_DEV(portid) {
2764 if ((enabled_port_mask & (1 << portid)) == 0)
2767 params = &lcore_params[nb_lcore_params++];
2768 params->port_id = portid;
2769 params->queue_id = 0;
2770 params->lcore_id = rte_get_next_lcore(0, 0, 1);
2777 inline_sessions_free(struct sa_ctx *sa_ctx)
2779 struct rte_ipsec_session *ips;
2780 struct ipsec_sa *sa;
2787 for (i = 0; i < sa_ctx->nb_sa; i++) {
2789 sa = &sa_ctx->sa[i];
2793 ips = ipsec_get_primary_session(sa);
2794 if (ips->type != RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL &&
2795 ips->type != RTE_SECURITY_ACTION_TYPE_INLINE_CRYPTO)
2798 if (!rte_eth_dev_is_valid_port(sa->portid))
2801 ret = rte_security_session_destroy(
2802 rte_eth_dev_get_sec_ctx(sa->portid),
2805 RTE_LOG(ERR, IPSEC, "Failed to destroy security "
2806 "session type %d, spi %d\n",
2807 ips->type, sa->spi);
2812 calculate_nb_mbufs(uint16_t nb_ports, uint16_t nb_crypto_qp, uint32_t nb_rxq,
2815 return RTE_MAX((nb_rxq * nb_rxd +
2816 nb_ports * nb_lcores * MAX_PKT_BURST +
2817 nb_ports * nb_txq * nb_txd +
2818 nb_lcores * MEMPOOL_CACHE_SIZE +
2819 nb_crypto_qp * CDEV_QUEUE_DESC +
2820 nb_lcores * frag_tbl_sz *
2821 FRAG_TBL_BUCKET_ENTRIES),
2826 main(int32_t argc, char **argv)
2829 uint32_t lcore_id, nb_txq, nb_rxq = 0;
2833 uint16_t portid, nb_crypto_qp, nb_ports = 0;
2834 uint64_t req_rx_offloads[RTE_MAX_ETHPORTS];
2835 uint64_t req_tx_offloads[RTE_MAX_ETHPORTS];
2836 struct eh_conf *eh_conf = NULL;
2839 nb_bufs_in_pool = 0;
2842 ret = rte_eal_init(argc, argv);
2844 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
2849 signal(SIGINT, signal_handler);
2850 signal(SIGTERM, signal_handler);
2852 /* initialize event helper configuration */
2853 eh_conf = eh_conf_init();
2854 if (eh_conf == NULL)
2855 rte_exit(EXIT_FAILURE, "Failed to init event helper config");
2857 /* parse application arguments (after the EAL ones) */
2858 ret = parse_args(argc, argv, eh_conf);
2860 rte_exit(EXIT_FAILURE, "Invalid parameters\n");
2862 /* parse configuration file */
2863 if (parse_cfg_file(cfgfile) < 0) {
2864 printf("parsing file \"%s\" failed\n",
2866 print_usage(argv[0]);
2870 if ((unprotected_port_mask & enabled_port_mask) !=
2871 unprotected_port_mask)
2872 rte_exit(EXIT_FAILURE, "Invalid unprotected portmask 0x%x\n",
2873 unprotected_port_mask);
2875 if (check_poll_mode_params(eh_conf) < 0)
2876 rte_exit(EXIT_FAILURE, "check_poll_mode_params failed\n");
2878 if (check_event_mode_params(eh_conf) < 0)
2879 rte_exit(EXIT_FAILURE, "check_event_mode_params failed\n");
2881 ret = init_lcore_rx_queues();
2883 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
2885 nb_lcores = rte_lcore_count();
2887 sess_sz = max_session_size();
2890 * In event mode request minimum number of crypto queues
2891 * to be reserved equal to number of ports.
2893 if (eh_conf->mode == EH_PKT_TRANSFER_MODE_EVENT)
2894 nb_crypto_qp = rte_eth_dev_count_avail();
2898 nb_crypto_qp = cryptodevs_init(nb_crypto_qp);
2900 if (nb_bufs_in_pool == 0) {
2901 RTE_ETH_FOREACH_DEV(portid) {
2902 if ((enabled_port_mask & (1 << portid)) == 0)
2905 nb_rxq += get_port_nb_rx_queues(portid);
2910 nb_bufs_in_pool = calculate_nb_mbufs(nb_ports, nb_crypto_qp,
2914 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2915 if (rte_lcore_is_enabled(lcore_id) == 0)
2919 socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2923 /* mbuf_pool is initialised by the pool_init() function*/
2924 if (socket_ctx[socket_id].mbuf_pool)
2927 pool_init(&socket_ctx[socket_id], socket_id, nb_bufs_in_pool);
2928 session_pool_init(&socket_ctx[socket_id], socket_id, sess_sz);
2929 session_priv_pool_init(&socket_ctx[socket_id], socket_id,
2932 printf("Number of mbufs in packet pool %d\n", nb_bufs_in_pool);
2934 RTE_ETH_FOREACH_DEV(portid) {
2935 if ((enabled_port_mask & (1 << portid)) == 0)
2938 sa_check_offloads(portid, &req_rx_offloads[portid],
2939 &req_tx_offloads[portid]);
2940 port_init(portid, req_rx_offloads[portid],
2941 req_tx_offloads[portid]);
2945 * Set the enabled port mask in helper config for use by helper
2946 * sub-system. This will be used while initializing devices using
2947 * helper sub-system.
2949 eh_conf->eth_portmask = enabled_port_mask;
2951 /* Initialize eventmode components */
2952 ret = eh_devs_init(eh_conf);
2954 rte_exit(EXIT_FAILURE, "eh_devs_init failed, err=%d\n", ret);
2957 RTE_ETH_FOREACH_DEV(portid) {
2958 if ((enabled_port_mask & (1 << portid)) == 0)
2961 /* Create flow before starting the device */
2962 create_default_ipsec_flow(portid, req_rx_offloads[portid]);
2964 ret = rte_eth_dev_start(portid);
2966 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: "
2967 "err=%d, port=%d\n", ret, portid);
2969 * If enabled, put device in promiscuous mode.
2970 * This allows IO forwarding mode to forward packets
2971 * to itself through 2 cross-connected ports of the
2974 if (promiscuous_on) {
2975 ret = rte_eth_promiscuous_enable(portid);
2977 rte_exit(EXIT_FAILURE,
2978 "rte_eth_promiscuous_enable: err=%s, port=%d\n",
2979 rte_strerror(-ret), portid);
2982 rte_eth_dev_callback_register(portid,
2983 RTE_ETH_EVENT_IPSEC, inline_ipsec_event_callback, NULL);
2986 /* fragment reassemble is enabled */
2987 if (frag_tbl_sz != 0) {
2988 ret = reassemble_init();
2990 rte_exit(EXIT_FAILURE, "failed at reassemble init");
2993 /* Replicate each context per socket */
2994 for (i = 0; i < NB_SOCKETS && i < rte_socket_count(); i++) {
2995 socket_id = rte_socket_id_by_idx(i);
2996 if ((socket_ctx[socket_id].mbuf_pool != NULL) &&
2997 (socket_ctx[socket_id].sa_in == NULL) &&
2998 (socket_ctx[socket_id].sa_out == NULL)) {
2999 sa_init(&socket_ctx[socket_id], socket_id);
3000 sp4_init(&socket_ctx[socket_id], socket_id);
3001 sp6_init(&socket_ctx[socket_id], socket_id);
3002 rt_init(&socket_ctx[socket_id], socket_id);
3008 check_all_ports_link_status(enabled_port_mask);
3010 #if (STATS_INTERVAL > 0)
3011 rte_eal_alarm_set(STATS_INTERVAL * US_PER_S, print_stats_cb, NULL);
3013 RTE_LOG(INFO, IPSEC, "Stats display disabled\n");
3014 #endif /* STATS_INTERVAL */
3016 /* launch per-lcore init on every lcore */
3017 rte_eal_mp_remote_launch(ipsec_launch_one_lcore, eh_conf, CALL_MAIN);
3018 RTE_LCORE_FOREACH_WORKER(lcore_id) {
3019 if (rte_eal_wait_lcore(lcore_id) < 0)
3023 /* Uninitialize eventmode components */
3024 ret = eh_devs_uninit(eh_conf);
3026 rte_exit(EXIT_FAILURE, "eh_devs_uninit failed, err=%d\n", ret);
3028 /* Free eventmode configuration memory */
3029 eh_conf_uninit(eh_conf);
3031 /* Destroy inline inbound and outbound sessions */
3032 for (i = 0; i < NB_SOCKETS && i < rte_socket_count(); i++) {
3033 socket_id = rte_socket_id_by_idx(i);
3034 inline_sessions_free(socket_ctx[socket_id].sa_in);
3035 inline_sessions_free(socket_ctx[socket_id].sa_out);
3038 for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
3039 printf("Closing cryptodev %d...", cdev_id);
3040 rte_cryptodev_stop(cdev_id);
3041 rte_cryptodev_close(cdev_id);
3045 RTE_ETH_FOREACH_DEV(portid) {
3046 if ((enabled_port_mask & (1 << portid)) == 0)
3049 printf("Closing port %d...", portid);
3050 if (flow_info_tbl[portid].rx_def_flow) {
3051 struct rte_flow_error err;
3053 ret = rte_flow_destroy(portid,
3054 flow_info_tbl[portid].rx_def_flow, &err);
3056 RTE_LOG(ERR, IPSEC, "Failed to destroy flow "
3057 " for port %u, err msg: %s\n", portid,
3060 ret = rte_eth_dev_stop(portid);
3063 "rte_eth_dev_stop: err=%s, port=%u\n",
3064 rte_strerror(-ret), portid);
3066 rte_eth_dev_close(portid);
3070 /* clean up the EAL */