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>
51 #include "event_helper.h"
53 #include "ipsec_worker.h"
57 volatile bool force_quit;
59 #define MAX_JUMBO_PKT_LEN 9600
61 #define MEMPOOL_CACHE_SIZE 256
63 #define CDEV_QUEUE_DESC 2048
64 #define CDEV_MAP_ENTRIES 16384
65 #define CDEV_MP_CACHE_SZ 64
66 #define MAX_QUEUE_PAIRS 1
68 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
70 /* Configure how many packets ahead to prefetch, when reading packets */
71 #define PREFETCH_OFFSET 3
73 #define MAX_RX_QUEUE_PER_LCORE 16
75 #define MAX_LCORE_PARAMS 1024
78 * Configurable number of RX/TX ring descriptors
80 #define IPSEC_SECGW_RX_DESC_DEFAULT 1024
81 #define IPSEC_SECGW_TX_DESC_DEFAULT 1024
82 static uint16_t nb_rxd = IPSEC_SECGW_RX_DESC_DEFAULT;
83 static uint16_t nb_txd = IPSEC_SECGW_TX_DESC_DEFAULT;
85 #define ETHADDR_TO_UINT64(addr) __BYTES_TO_UINT64( \
86 (addr)->addr_bytes[0], (addr)->addr_bytes[1], \
87 (addr)->addr_bytes[2], (addr)->addr_bytes[3], \
88 (addr)->addr_bytes[4], (addr)->addr_bytes[5], \
91 #define FRAG_TBL_BUCKET_ENTRIES 4
92 #define MAX_FRAG_TTL_NS (10LL * NS_PER_S)
94 #define MTU_TO_FRAMELEN(x) ((x) + RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN)
96 struct ethaddr_info ethaddr_tbl[RTE_MAX_ETHPORTS] = {
97 { 0, ETHADDR(0x00, 0x16, 0x3e, 0x7e, 0x94, 0x9a) },
98 { 0, ETHADDR(0x00, 0x16, 0x3e, 0x22, 0xa1, 0xd9) },
99 { 0, ETHADDR(0x00, 0x16, 0x3e, 0x08, 0x69, 0x26) },
100 { 0, ETHADDR(0x00, 0x16, 0x3e, 0x49, 0x9e, 0xdd) }
103 struct flow_info flow_info_tbl[RTE_MAX_ETHPORTS];
105 #define CMD_LINE_OPT_CONFIG "config"
106 #define CMD_LINE_OPT_SINGLE_SA "single-sa"
107 #define CMD_LINE_OPT_CRYPTODEV_MASK "cryptodev_mask"
108 #define CMD_LINE_OPT_TRANSFER_MODE "transfer-mode"
109 #define CMD_LINE_OPT_SCHEDULE_TYPE "event-schedule-type"
110 #define CMD_LINE_OPT_RX_OFFLOAD "rxoffload"
111 #define CMD_LINE_OPT_TX_OFFLOAD "txoffload"
112 #define CMD_LINE_OPT_REASSEMBLE "reassemble"
113 #define CMD_LINE_OPT_MTU "mtu"
114 #define CMD_LINE_OPT_FRAG_TTL "frag-ttl"
116 #define CMD_LINE_ARG_EVENT "event"
117 #define CMD_LINE_ARG_POLL "poll"
118 #define CMD_LINE_ARG_ORDERED "ordered"
119 #define CMD_LINE_ARG_ATOMIC "atomic"
120 #define CMD_LINE_ARG_PARALLEL "parallel"
123 /* long options mapped to a short option */
125 /* first long only option value must be >= 256, so that we won't
126 * conflict with short options
128 CMD_LINE_OPT_MIN_NUM = 256,
129 CMD_LINE_OPT_CONFIG_NUM,
130 CMD_LINE_OPT_SINGLE_SA_NUM,
131 CMD_LINE_OPT_CRYPTODEV_MASK_NUM,
132 CMD_LINE_OPT_TRANSFER_MODE_NUM,
133 CMD_LINE_OPT_SCHEDULE_TYPE_NUM,
134 CMD_LINE_OPT_RX_OFFLOAD_NUM,
135 CMD_LINE_OPT_TX_OFFLOAD_NUM,
136 CMD_LINE_OPT_REASSEMBLE_NUM,
137 CMD_LINE_OPT_MTU_NUM,
138 CMD_LINE_OPT_FRAG_TTL_NUM,
141 static const struct option lgopts[] = {
142 {CMD_LINE_OPT_CONFIG, 1, 0, CMD_LINE_OPT_CONFIG_NUM},
143 {CMD_LINE_OPT_SINGLE_SA, 1, 0, CMD_LINE_OPT_SINGLE_SA_NUM},
144 {CMD_LINE_OPT_CRYPTODEV_MASK, 1, 0, CMD_LINE_OPT_CRYPTODEV_MASK_NUM},
145 {CMD_LINE_OPT_TRANSFER_MODE, 1, 0, CMD_LINE_OPT_TRANSFER_MODE_NUM},
146 {CMD_LINE_OPT_SCHEDULE_TYPE, 1, 0, CMD_LINE_OPT_SCHEDULE_TYPE_NUM},
147 {CMD_LINE_OPT_RX_OFFLOAD, 1, 0, CMD_LINE_OPT_RX_OFFLOAD_NUM},
148 {CMD_LINE_OPT_TX_OFFLOAD, 1, 0, CMD_LINE_OPT_TX_OFFLOAD_NUM},
149 {CMD_LINE_OPT_REASSEMBLE, 1, 0, CMD_LINE_OPT_REASSEMBLE_NUM},
150 {CMD_LINE_OPT_MTU, 1, 0, CMD_LINE_OPT_MTU_NUM},
151 {CMD_LINE_OPT_FRAG_TTL, 1, 0, CMD_LINE_OPT_FRAG_TTL_NUM},
155 uint32_t unprotected_port_mask;
156 uint32_t single_sa_idx;
157 /* mask of enabled ports */
158 static uint32_t enabled_port_mask;
159 static uint64_t enabled_cryptodev_mask = UINT64_MAX;
160 static int32_t promiscuous_on = 1;
161 static int32_t numa_on = 1; /**< NUMA is enabled by default. */
162 static uint32_t nb_lcores;
163 static uint32_t single_sa;
164 static uint32_t nb_bufs_in_pool;
167 * RX/TX HW offload capabilities to enable/use on ethernet ports.
168 * By default all capabilities are enabled.
170 static uint64_t dev_rx_offload = UINT64_MAX;
171 static uint64_t dev_tx_offload = UINT64_MAX;
174 * global values that determine multi-seg policy
176 static uint32_t frag_tbl_sz;
177 static uint32_t frame_buf_size = RTE_MBUF_DEFAULT_BUF_SIZE;
178 static uint32_t mtu_size = RTE_ETHER_MTU;
179 static uint64_t frag_ttl_ns = MAX_FRAG_TTL_NS;
181 /* application wide librte_ipsec/SA parameters */
182 struct app_sa_prm app_sa_prm = {
184 .cache_sz = SA_CACHE_SZ
186 static const char *cfgfile;
188 struct lcore_rx_queue {
191 } __rte_cache_aligned;
193 struct lcore_params {
197 } __rte_cache_aligned;
199 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
201 static struct lcore_params *lcore_params;
202 static uint16_t nb_lcore_params;
204 static struct rte_hash *cdev_map_in;
205 static struct rte_hash *cdev_map_out;
209 struct rte_mbuf *m_table[MAX_PKT_BURST] __rte_aligned(sizeof(void *));
213 uint16_t nb_rx_queue;
214 struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
215 uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
216 struct buffer tx_mbufs[RTE_MAX_ETHPORTS];
217 struct ipsec_ctx inbound;
218 struct ipsec_ctx outbound;
219 struct rt_ctx *rt4_ctx;
220 struct rt_ctx *rt6_ctx;
222 struct rte_ip_frag_tbl *tbl;
223 struct rte_mempool *pool_dir;
224 struct rte_mempool *pool_indir;
225 struct rte_ip_frag_death_row dr;
227 } __rte_cache_aligned;
229 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
231 static struct rte_eth_conf port_conf = {
233 .mq_mode = ETH_MQ_RX_RSS,
234 .max_rx_pkt_len = RTE_ETHER_MAX_LEN,
236 .offloads = DEV_RX_OFFLOAD_CHECKSUM,
241 .rss_hf = ETH_RSS_IP | ETH_RSS_UDP |
242 ETH_RSS_TCP | ETH_RSS_SCTP,
246 .mq_mode = ETH_MQ_TX_NONE,
250 struct socket_ctx socket_ctx[NB_SOCKETS];
253 * Determine is multi-segment support required:
254 * - either frame buffer size is smaller then mtu
255 * - or reassmeble support is requested
258 multi_seg_required(void)
260 return (MTU_TO_FRAMELEN(mtu_size) + RTE_PKTMBUF_HEADROOM >
261 frame_buf_size || frag_tbl_sz != 0);
265 adjust_ipv4_pktlen(struct rte_mbuf *m, const struct rte_ipv4_hdr *iph,
270 plen = rte_be_to_cpu_16(iph->total_length) + l2_len;
271 if (plen < m->pkt_len) {
272 trim = m->pkt_len - plen;
273 rte_pktmbuf_trim(m, trim);
278 adjust_ipv6_pktlen(struct rte_mbuf *m, const struct rte_ipv6_hdr *iph,
283 plen = rte_be_to_cpu_16(iph->payload_len) + sizeof(*iph) + l2_len;
284 if (plen < m->pkt_len) {
285 trim = m->pkt_len - plen;
286 rte_pktmbuf_trim(m, trim);
291 prepare_one_packet(struct rte_mbuf *pkt, struct ipsec_traffic *t)
293 const struct rte_ether_hdr *eth;
294 const struct rte_ipv4_hdr *iph4;
295 const struct rte_ipv6_hdr *iph6;
297 eth = rte_pktmbuf_mtod(pkt, const struct rte_ether_hdr *);
298 if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
300 iph4 = (const struct rte_ipv4_hdr *)rte_pktmbuf_adj(pkt,
302 adjust_ipv4_pktlen(pkt, iph4, 0);
304 if (iph4->next_proto_id == IPPROTO_ESP)
305 t->ipsec.pkts[(t->ipsec.num)++] = pkt;
307 t->ip4.data[t->ip4.num] = &iph4->next_proto_id;
308 t->ip4.pkts[(t->ip4.num)++] = pkt;
311 pkt->l3_len = sizeof(*iph4);
312 pkt->packet_type |= RTE_PTYPE_L3_IPV4;
313 } else if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6)) {
315 size_t l3len, ext_len;
318 /* get protocol type */
319 iph6 = (const struct rte_ipv6_hdr *)rte_pktmbuf_adj(pkt,
321 adjust_ipv6_pktlen(pkt, iph6, 0);
323 next_proto = iph6->proto;
325 /* determine l3 header size up to ESP extension */
326 l3len = sizeof(struct ip6_hdr);
327 p = rte_pktmbuf_mtod(pkt, uint8_t *);
328 while (next_proto != IPPROTO_ESP && l3len < pkt->data_len &&
329 (next_proto = rte_ipv6_get_next_ext(p + l3len,
330 next_proto, &ext_len)) >= 0)
333 /* drop packet when IPv6 header exceeds first segment length */
334 if (unlikely(l3len > pkt->data_len)) {
335 rte_pktmbuf_free(pkt);
339 if (next_proto == IPPROTO_ESP)
340 t->ipsec.pkts[(t->ipsec.num)++] = pkt;
342 t->ip6.data[t->ip6.num] = &iph6->proto;
343 t->ip6.pkts[(t->ip6.num)++] = pkt;
347 pkt->packet_type |= RTE_PTYPE_L3_IPV6;
349 /* Unknown/Unsupported type, drop the packet */
350 RTE_LOG(ERR, IPSEC, "Unsupported packet type 0x%x\n",
351 rte_be_to_cpu_16(eth->ether_type));
352 rte_pktmbuf_free(pkt);
356 /* Check if the packet has been processed inline. For inline protocol
357 * processed packets, the metadata in the mbuf can be used to identify
358 * the security processing done on the packet. The metadata will be
359 * used to retrieve the application registered userdata associated
360 * with the security session.
363 if (pkt->ol_flags & PKT_RX_SEC_OFFLOAD) {
365 struct ipsec_mbuf_metadata *priv;
366 struct rte_security_ctx *ctx = (struct rte_security_ctx *)
367 rte_eth_dev_get_sec_ctx(
370 /* Retrieve the userdata registered. Here, the userdata
371 * registered is the SA pointer.
374 sa = (struct ipsec_sa *)
375 rte_security_get_userdata(ctx, pkt->udata64);
378 /* userdata could not be retrieved */
382 /* Save SA as priv member in mbuf. This will be used in the
383 * IPsec selector(SP-SA) check.
386 priv = get_priv(pkt);
392 prepare_traffic(struct rte_mbuf **pkts, struct ipsec_traffic *t,
401 for (i = 0; i < (nb_pkts - PREFETCH_OFFSET); i++) {
402 rte_prefetch0(rte_pktmbuf_mtod(pkts[i + PREFETCH_OFFSET],
404 prepare_one_packet(pkts[i], t);
406 /* Process left packets */
407 for (; i < nb_pkts; i++)
408 prepare_one_packet(pkts[i], t);
412 prepare_tx_pkt(struct rte_mbuf *pkt, uint16_t port,
413 const struct lcore_conf *qconf)
416 struct rte_ether_hdr *ethhdr;
418 ip = rte_pktmbuf_mtod(pkt, struct ip *);
420 ethhdr = (struct rte_ether_hdr *)
421 rte_pktmbuf_prepend(pkt, RTE_ETHER_HDR_LEN);
423 if (ip->ip_v == IPVERSION) {
424 pkt->ol_flags |= qconf->outbound.ipv4_offloads;
425 pkt->l3_len = sizeof(struct ip);
426 pkt->l2_len = RTE_ETHER_HDR_LEN;
430 /* calculate IPv4 cksum in SW */
431 if ((pkt->ol_flags & PKT_TX_IP_CKSUM) == 0)
432 ip->ip_sum = rte_ipv4_cksum((struct rte_ipv4_hdr *)ip);
434 ethhdr->ether_type = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4);
436 pkt->ol_flags |= qconf->outbound.ipv6_offloads;
437 pkt->l3_len = sizeof(struct ip6_hdr);
438 pkt->l2_len = RTE_ETHER_HDR_LEN;
440 ethhdr->ether_type = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6);
443 memcpy(ðhdr->s_addr, ðaddr_tbl[port].src,
444 sizeof(struct rte_ether_addr));
445 memcpy(ðhdr->d_addr, ðaddr_tbl[port].dst,
446 sizeof(struct rte_ether_addr));
450 prepare_tx_burst(struct rte_mbuf *pkts[], uint16_t nb_pkts, uint16_t port,
451 const struct lcore_conf *qconf)
454 const int32_t prefetch_offset = 2;
456 for (i = 0; i < (nb_pkts - prefetch_offset); i++) {
457 rte_mbuf_prefetch_part2(pkts[i + prefetch_offset]);
458 prepare_tx_pkt(pkts[i], port, qconf);
460 /* Process left packets */
461 for (; i < nb_pkts; i++)
462 prepare_tx_pkt(pkts[i], port, qconf);
465 /* Send burst of packets on an output interface */
466 static inline int32_t
467 send_burst(struct lcore_conf *qconf, uint16_t n, uint16_t port)
469 struct rte_mbuf **m_table;
473 queueid = qconf->tx_queue_id[port];
474 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
476 prepare_tx_burst(m_table, n, port, qconf);
478 ret = rte_eth_tx_burst(port, queueid, m_table, n);
479 if (unlikely(ret < n)) {
481 rte_pktmbuf_free(m_table[ret]);
489 * Helper function to fragment and queue for TX one packet.
491 static inline uint32_t
492 send_fragment_packet(struct lcore_conf *qconf, struct rte_mbuf *m,
493 uint16_t port, uint8_t proto)
499 tbl = qconf->tx_mbufs + port;
502 /* free space for new fragments */
503 if (len + RTE_LIBRTE_IP_FRAG_MAX_FRAG >= RTE_DIM(tbl->m_table)) {
504 send_burst(qconf, len, port);
508 n = RTE_DIM(tbl->m_table) - len;
510 if (proto == IPPROTO_IP)
511 rc = rte_ipv4_fragment_packet(m, tbl->m_table + len,
512 n, mtu_size, qconf->frag.pool_dir,
513 qconf->frag.pool_indir);
515 rc = rte_ipv6_fragment_packet(m, tbl->m_table + len,
516 n, mtu_size, qconf->frag.pool_dir,
517 qconf->frag.pool_indir);
523 "%s: failed to fragment packet with size %u, "
525 __func__, m->pkt_len, rte_errno);
531 /* Enqueue a single packet, and send burst if queue is filled */
532 static inline int32_t
533 send_single_packet(struct rte_mbuf *m, uint16_t port, uint8_t proto)
537 struct lcore_conf *qconf;
539 lcore_id = rte_lcore_id();
541 qconf = &lcore_conf[lcore_id];
542 len = qconf->tx_mbufs[port].len;
544 if (m->pkt_len <= mtu_size) {
545 qconf->tx_mbufs[port].m_table[len] = m;
548 /* need to fragment the packet */
549 } else if (frag_tbl_sz > 0)
550 len = send_fragment_packet(qconf, m, port, proto);
554 /* enough pkts to be sent */
555 if (unlikely(len == MAX_PKT_BURST)) {
556 send_burst(qconf, MAX_PKT_BURST, port);
560 qconf->tx_mbufs[port].len = len;
565 inbound_sp_sa(struct sp_ctx *sp, struct sa_ctx *sa, struct traffic_type *ip,
569 uint32_t i, j, res, sa_idx;
571 if (ip->num == 0 || sp == NULL)
574 rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
575 ip->num, DEFAULT_MAX_CATEGORIES);
578 for (i = 0; i < ip->num; i++) {
585 if (res == DISCARD) {
590 /* Only check SPI match for processed IPSec packets */
591 if (i < lim && ((m->ol_flags & PKT_RX_SEC_OFFLOAD) == 0)) {
597 if (!inbound_sa_check(sa, m, sa_idx)) {
607 split46_traffic(struct ipsec_traffic *trf, struct rte_mbuf *mb[], uint32_t num)
616 for (i = 0; i < num; i++) {
619 ip = rte_pktmbuf_mtod(m, struct ip *);
621 if (ip->ip_v == IPVERSION) {
622 trf->ip4.pkts[n4] = m;
623 trf->ip4.data[n4] = rte_pktmbuf_mtod_offset(m,
624 uint8_t *, offsetof(struct ip, ip_p));
626 } else if (ip->ip_v == IP6_VERSION) {
627 trf->ip6.pkts[n6] = m;
628 trf->ip6.data[n6] = rte_pktmbuf_mtod_offset(m,
630 offsetof(struct ip6_hdr, ip6_nxt));
642 process_pkts_inbound(struct ipsec_ctx *ipsec_ctx,
643 struct ipsec_traffic *traffic)
645 uint16_t nb_pkts_in, n_ip4, n_ip6;
647 n_ip4 = traffic->ip4.num;
648 n_ip6 = traffic->ip6.num;
650 if (app_sa_prm.enable == 0) {
651 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
652 traffic->ipsec.num, MAX_PKT_BURST);
653 split46_traffic(traffic, traffic->ipsec.pkts, nb_pkts_in);
655 inbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.pkts,
656 traffic->ipsec.saptr, traffic->ipsec.num);
657 ipsec_process(ipsec_ctx, traffic);
660 inbound_sp_sa(ipsec_ctx->sp4_ctx, ipsec_ctx->sa_ctx, &traffic->ip4,
663 inbound_sp_sa(ipsec_ctx->sp6_ctx, ipsec_ctx->sa_ctx, &traffic->ip6,
668 outbound_sp(struct sp_ctx *sp, struct traffic_type *ip,
669 struct traffic_type *ipsec)
672 uint32_t i, j, sa_idx;
674 if (ip->num == 0 || sp == NULL)
677 rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
678 ip->num, DEFAULT_MAX_CATEGORIES);
681 for (i = 0; i < ip->num; i++) {
683 sa_idx = ip->res[i] - 1;
684 if (ip->res[i] == DISCARD)
686 else if (ip->res[i] == BYPASS)
689 ipsec->res[ipsec->num] = sa_idx;
690 ipsec->pkts[ipsec->num++] = m;
697 process_pkts_outbound(struct ipsec_ctx *ipsec_ctx,
698 struct ipsec_traffic *traffic)
701 uint16_t idx, nb_pkts_out, i;
703 /* Drop any IPsec traffic from protected ports */
704 for (i = 0; i < traffic->ipsec.num; i++)
705 rte_pktmbuf_free(traffic->ipsec.pkts[i]);
707 traffic->ipsec.num = 0;
709 outbound_sp(ipsec_ctx->sp4_ctx, &traffic->ip4, &traffic->ipsec);
711 outbound_sp(ipsec_ctx->sp6_ctx, &traffic->ip6, &traffic->ipsec);
713 if (app_sa_prm.enable == 0) {
715 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
716 traffic->ipsec.res, traffic->ipsec.num,
719 for (i = 0; i < nb_pkts_out; i++) {
720 m = traffic->ipsec.pkts[i];
721 struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
722 if (ip->ip_v == IPVERSION) {
723 idx = traffic->ip4.num++;
724 traffic->ip4.pkts[idx] = m;
726 idx = traffic->ip6.num++;
727 traffic->ip6.pkts[idx] = m;
731 outbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.res,
732 traffic->ipsec.saptr, traffic->ipsec.num);
733 ipsec_process(ipsec_ctx, traffic);
738 process_pkts_inbound_nosp(struct ipsec_ctx *ipsec_ctx,
739 struct ipsec_traffic *traffic)
742 uint32_t nb_pkts_in, i, idx;
744 /* Drop any IPv4 traffic from unprotected ports */
745 for (i = 0; i < traffic->ip4.num; i++)
746 rte_pktmbuf_free(traffic->ip4.pkts[i]);
748 traffic->ip4.num = 0;
750 /* Drop any IPv6 traffic from unprotected ports */
751 for (i = 0; i < traffic->ip6.num; i++)
752 rte_pktmbuf_free(traffic->ip6.pkts[i]);
754 traffic->ip6.num = 0;
756 if (app_sa_prm.enable == 0) {
758 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
759 traffic->ipsec.num, MAX_PKT_BURST);
761 for (i = 0; i < nb_pkts_in; i++) {
762 m = traffic->ipsec.pkts[i];
763 struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
764 if (ip->ip_v == IPVERSION) {
765 idx = traffic->ip4.num++;
766 traffic->ip4.pkts[idx] = m;
768 idx = traffic->ip6.num++;
769 traffic->ip6.pkts[idx] = m;
773 inbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.pkts,
774 traffic->ipsec.saptr, traffic->ipsec.num);
775 ipsec_process(ipsec_ctx, traffic);
780 process_pkts_outbound_nosp(struct ipsec_ctx *ipsec_ctx,
781 struct ipsec_traffic *traffic)
784 uint32_t nb_pkts_out, i, n;
787 /* Drop any IPsec traffic from protected ports */
788 for (i = 0; i < traffic->ipsec.num; i++)
789 rte_pktmbuf_free(traffic->ipsec.pkts[i]);
793 for (i = 0; i < traffic->ip4.num; i++) {
794 traffic->ipsec.pkts[n] = traffic->ip4.pkts[i];
795 traffic->ipsec.res[n++] = single_sa_idx;
798 for (i = 0; i < traffic->ip6.num; i++) {
799 traffic->ipsec.pkts[n] = traffic->ip6.pkts[i];
800 traffic->ipsec.res[n++] = single_sa_idx;
803 traffic->ip4.num = 0;
804 traffic->ip6.num = 0;
805 traffic->ipsec.num = n;
807 if (app_sa_prm.enable == 0) {
809 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
810 traffic->ipsec.res, traffic->ipsec.num,
813 /* They all sue the same SA (ip4 or ip6 tunnel) */
814 m = traffic->ipsec.pkts[0];
815 ip = rte_pktmbuf_mtod(m, struct ip *);
816 if (ip->ip_v == IPVERSION) {
817 traffic->ip4.num = nb_pkts_out;
818 for (i = 0; i < nb_pkts_out; i++)
819 traffic->ip4.pkts[i] = traffic->ipsec.pkts[i];
821 traffic->ip6.num = nb_pkts_out;
822 for (i = 0; i < nb_pkts_out; i++)
823 traffic->ip6.pkts[i] = traffic->ipsec.pkts[i];
826 outbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.res,
827 traffic->ipsec.saptr, traffic->ipsec.num);
828 ipsec_process(ipsec_ctx, traffic);
832 static inline int32_t
833 get_hop_for_offload_pkt(struct rte_mbuf *pkt, int is_ipv6)
835 struct ipsec_mbuf_metadata *priv;
838 priv = get_priv(pkt);
841 if (unlikely(sa == NULL)) {
842 RTE_LOG(ERR, IPSEC, "SA not saved in private data\n");
850 return (sa->portid | RTE_LPM_LOOKUP_SUCCESS);
861 route4_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
863 uint32_t hop[MAX_PKT_BURST * 2];
864 uint32_t dst_ip[MAX_PKT_BURST * 2];
867 uint16_t lpm_pkts = 0;
872 /* Need to do an LPM lookup for non-inline packets. Inline packets will
873 * have port ID in the SA
876 for (i = 0; i < nb_pkts; i++) {
877 if (!(pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD)) {
878 /* Security offload not enabled. So an LPM lookup is
879 * required to get the hop
881 offset = offsetof(struct ip, ip_dst);
882 dst_ip[lpm_pkts] = *rte_pktmbuf_mtod_offset(pkts[i],
884 dst_ip[lpm_pkts] = rte_be_to_cpu_32(dst_ip[lpm_pkts]);
889 rte_lpm_lookup_bulk((struct rte_lpm *)rt_ctx, dst_ip, hop, lpm_pkts);
893 for (i = 0; i < nb_pkts; i++) {
894 if (pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD) {
895 /* Read hop from the SA */
896 pkt_hop = get_hop_for_offload_pkt(pkts[i], 0);
898 /* Need to use hop returned by lookup */
899 pkt_hop = hop[lpm_pkts++];
902 if ((pkt_hop & RTE_LPM_LOOKUP_SUCCESS) == 0) {
903 rte_pktmbuf_free(pkts[i]);
906 send_single_packet(pkts[i], pkt_hop & 0xff, IPPROTO_IP);
911 route6_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
913 int32_t hop[MAX_PKT_BURST * 2];
914 uint8_t dst_ip[MAX_PKT_BURST * 2][16];
918 uint16_t lpm_pkts = 0;
923 /* Need to do an LPM lookup for non-inline packets. Inline packets will
924 * have port ID in the SA
927 for (i = 0; i < nb_pkts; i++) {
928 if (!(pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD)) {
929 /* Security offload not enabled. So an LPM lookup is
930 * required to get the hop
932 offset = offsetof(struct ip6_hdr, ip6_dst);
933 ip6_dst = rte_pktmbuf_mtod_offset(pkts[i], uint8_t *,
935 memcpy(&dst_ip[lpm_pkts][0], ip6_dst, 16);
940 rte_lpm6_lookup_bulk_func((struct rte_lpm6 *)rt_ctx, dst_ip, hop,
945 for (i = 0; i < nb_pkts; i++) {
946 if (pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD) {
947 /* Read hop from the SA */
948 pkt_hop = get_hop_for_offload_pkt(pkts[i], 1);
950 /* Need to use hop returned by lookup */
951 pkt_hop = hop[lpm_pkts++];
955 rte_pktmbuf_free(pkts[i]);
958 send_single_packet(pkts[i], pkt_hop & 0xff, IPPROTO_IPV6);
963 process_pkts(struct lcore_conf *qconf, struct rte_mbuf **pkts,
964 uint8_t nb_pkts, uint16_t portid)
966 struct ipsec_traffic traffic;
968 prepare_traffic(pkts, &traffic, nb_pkts);
970 if (unlikely(single_sa)) {
971 if (is_unprotected_port(portid))
972 process_pkts_inbound_nosp(&qconf->inbound, &traffic);
974 process_pkts_outbound_nosp(&qconf->outbound, &traffic);
976 if (is_unprotected_port(portid))
977 process_pkts_inbound(&qconf->inbound, &traffic);
979 process_pkts_outbound(&qconf->outbound, &traffic);
982 route4_pkts(qconf->rt4_ctx, traffic.ip4.pkts, traffic.ip4.num);
983 route6_pkts(qconf->rt6_ctx, traffic.ip6.pkts, traffic.ip6.num);
987 drain_tx_buffers(struct lcore_conf *qconf)
992 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
993 buf = &qconf->tx_mbufs[portid];
996 send_burst(qconf, buf->len, portid);
1002 drain_crypto_buffers(struct lcore_conf *qconf)
1005 struct ipsec_ctx *ctx;
1007 /* drain inbound buffers*/
1008 ctx = &qconf->inbound;
1009 for (i = 0; i != ctx->nb_qps; i++) {
1010 if (ctx->tbl[i].len != 0)
1011 enqueue_cop_burst(ctx->tbl + i);
1014 /* drain outbound buffers*/
1015 ctx = &qconf->outbound;
1016 for (i = 0; i != ctx->nb_qps; i++) {
1017 if (ctx->tbl[i].len != 0)
1018 enqueue_cop_burst(ctx->tbl + i);
1023 drain_inbound_crypto_queues(const struct lcore_conf *qconf,
1024 struct ipsec_ctx *ctx)
1027 struct ipsec_traffic trf;
1029 if (app_sa_prm.enable == 0) {
1031 /* dequeue packets from crypto-queue */
1032 n = ipsec_inbound_cqp_dequeue(ctx, trf.ipsec.pkts,
1033 RTE_DIM(trf.ipsec.pkts));
1038 /* split traffic by ipv4-ipv6 */
1039 split46_traffic(&trf, trf.ipsec.pkts, n);
1041 ipsec_cqp_process(ctx, &trf);
1043 /* process ipv4 packets */
1044 if (trf.ip4.num != 0) {
1045 inbound_sp_sa(ctx->sp4_ctx, ctx->sa_ctx, &trf.ip4, 0);
1046 route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
1049 /* process ipv6 packets */
1050 if (trf.ip6.num != 0) {
1051 inbound_sp_sa(ctx->sp6_ctx, ctx->sa_ctx, &trf.ip6, 0);
1052 route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
1057 drain_outbound_crypto_queues(const struct lcore_conf *qconf,
1058 struct ipsec_ctx *ctx)
1061 struct ipsec_traffic trf;
1063 if (app_sa_prm.enable == 0) {
1065 /* dequeue packets from crypto-queue */
1066 n = ipsec_outbound_cqp_dequeue(ctx, trf.ipsec.pkts,
1067 RTE_DIM(trf.ipsec.pkts));
1072 /* split traffic by ipv4-ipv6 */
1073 split46_traffic(&trf, trf.ipsec.pkts, n);
1075 ipsec_cqp_process(ctx, &trf);
1077 /* process ipv4 packets */
1078 if (trf.ip4.num != 0)
1079 route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
1081 /* process ipv6 packets */
1082 if (trf.ip6.num != 0)
1083 route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
1086 /* main processing loop */
1088 ipsec_poll_mode_worker(void)
1090 struct rte_mbuf *pkts[MAX_PKT_BURST];
1092 uint64_t prev_tsc, diff_tsc, cur_tsc;
1096 struct lcore_conf *qconf;
1097 int32_t rc, socket_id;
1098 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
1099 / US_PER_S * BURST_TX_DRAIN_US;
1100 struct lcore_rx_queue *rxql;
1103 lcore_id = rte_lcore_id();
1104 qconf = &lcore_conf[lcore_id];
1105 rxql = qconf->rx_queue_list;
1106 socket_id = rte_lcore_to_socket_id(lcore_id);
1108 qconf->rt4_ctx = socket_ctx[socket_id].rt_ip4;
1109 qconf->rt6_ctx = socket_ctx[socket_id].rt_ip6;
1110 qconf->inbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_in;
1111 qconf->inbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_in;
1112 qconf->inbound.sa_ctx = socket_ctx[socket_id].sa_in;
1113 qconf->inbound.cdev_map = cdev_map_in;
1114 qconf->inbound.session_pool = socket_ctx[socket_id].session_pool;
1115 qconf->inbound.session_priv_pool =
1116 socket_ctx[socket_id].session_priv_pool;
1117 qconf->outbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_out;
1118 qconf->outbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_out;
1119 qconf->outbound.sa_ctx = socket_ctx[socket_id].sa_out;
1120 qconf->outbound.cdev_map = cdev_map_out;
1121 qconf->outbound.session_pool = socket_ctx[socket_id].session_pool;
1122 qconf->outbound.session_priv_pool =
1123 socket_ctx[socket_id].session_priv_pool;
1124 qconf->frag.pool_dir = socket_ctx[socket_id].mbuf_pool;
1125 qconf->frag.pool_indir = socket_ctx[socket_id].mbuf_pool_indir;
1127 rc = ipsec_sad_lcore_cache_init(app_sa_prm.cache_sz);
1130 "SAD cache init on lcore %u, failed with code: %d\n",
1135 if (qconf->nb_rx_queue == 0) {
1136 RTE_LOG(DEBUG, IPSEC, "lcore %u has nothing to do\n",
1141 RTE_LOG(INFO, IPSEC, "entering main loop on lcore %u\n", lcore_id);
1143 for (i = 0; i < qconf->nb_rx_queue; i++) {
1144 portid = rxql[i].port_id;
1145 queueid = rxql[i].queue_id;
1146 RTE_LOG(INFO, IPSEC,
1147 " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
1148 lcore_id, portid, queueid);
1151 while (!force_quit) {
1152 cur_tsc = rte_rdtsc();
1154 /* TX queue buffer drain */
1155 diff_tsc = cur_tsc - prev_tsc;
1157 if (unlikely(diff_tsc > drain_tsc)) {
1158 drain_tx_buffers(qconf);
1159 drain_crypto_buffers(qconf);
1163 for (i = 0; i < qconf->nb_rx_queue; ++i) {
1165 /* Read packets from RX queues */
1166 portid = rxql[i].port_id;
1167 queueid = rxql[i].queue_id;
1168 nb_rx = rte_eth_rx_burst(portid, queueid,
1169 pkts, MAX_PKT_BURST);
1172 process_pkts(qconf, pkts, nb_rx, portid);
1174 /* dequeue and process completed crypto-ops */
1175 if (is_unprotected_port(portid))
1176 drain_inbound_crypto_queues(qconf,
1179 drain_outbound_crypto_queues(qconf,
1186 check_flow_params(uint16_t fdir_portid, uint8_t fdir_qid)
1192 for (i = 0; i < nb_lcore_params; ++i) {
1193 portid = lcore_params_array[i].port_id;
1194 if (portid == fdir_portid) {
1195 queueid = lcore_params_array[i].queue_id;
1196 if (queueid == fdir_qid)
1200 if (i == nb_lcore_params - 1)
1208 check_poll_mode_params(struct eh_conf *eh_conf)
1218 if (eh_conf->mode != EH_PKT_TRANSFER_MODE_POLL)
1221 if (lcore_params == NULL) {
1222 printf("Error: No port/queue/core mappings\n");
1226 for (i = 0; i < nb_lcore_params; ++i) {
1227 lcore = lcore_params[i].lcore_id;
1228 if (!rte_lcore_is_enabled(lcore)) {
1229 printf("error: lcore %hhu is not enabled in "
1230 "lcore mask\n", lcore);
1233 socket_id = rte_lcore_to_socket_id(lcore);
1234 if (socket_id != 0 && numa_on == 0) {
1235 printf("warning: lcore %hhu is on socket %d "
1239 portid = lcore_params[i].port_id;
1240 if ((enabled_port_mask & (1 << portid)) == 0) {
1241 printf("port %u is not enabled in port mask\n", portid);
1244 if (!rte_eth_dev_is_valid_port(portid)) {
1245 printf("port %u is not present on the board\n", portid);
1253 get_port_nb_rx_queues(const uint16_t port)
1258 for (i = 0; i < nb_lcore_params; ++i) {
1259 if (lcore_params[i].port_id == port &&
1260 lcore_params[i].queue_id > queue)
1261 queue = lcore_params[i].queue_id;
1263 return (uint8_t)(++queue);
1267 init_lcore_rx_queues(void)
1269 uint16_t i, nb_rx_queue;
1272 for (i = 0; i < nb_lcore_params; ++i) {
1273 lcore = lcore_params[i].lcore_id;
1274 nb_rx_queue = lcore_conf[lcore].nb_rx_queue;
1275 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1276 printf("error: too many queues (%u) for lcore: %u\n",
1277 nb_rx_queue + 1, lcore);
1280 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1281 lcore_params[i].port_id;
1282 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1283 lcore_params[i].queue_id;
1284 lcore_conf[lcore].nb_rx_queue++;
1291 print_usage(const char *prgname)
1293 fprintf(stderr, "%s [EAL options] --"
1299 " [-w REPLAY_WINDOW_SIZE]"
1303 " [-s NUMBER_OF_MBUFS_IN_PKT_POOL]"
1305 " --config (port,queue,lcore)[,(port,queue,lcore)]"
1306 " [--single-sa SAIDX]"
1307 " [--cryptodev_mask MASK]"
1308 " [--transfer-mode MODE]"
1309 " [--event-schedule-type TYPE]"
1310 " [--" CMD_LINE_OPT_RX_OFFLOAD " RX_OFFLOAD_MASK]"
1311 " [--" CMD_LINE_OPT_TX_OFFLOAD " TX_OFFLOAD_MASK]"
1312 " [--" CMD_LINE_OPT_REASSEMBLE " REASSEMBLE_TABLE_SIZE]"
1313 " [--" CMD_LINE_OPT_MTU " MTU]"
1315 " -p PORTMASK: Hexadecimal bitmask of ports to configure\n"
1316 " -P : Enable promiscuous mode\n"
1317 " -u PORTMASK: Hexadecimal bitmask of unprotected ports\n"
1318 " -j FRAMESIZE: Data buffer size, minimum (and default)\n"
1319 " value: RTE_MBUF_DEFAULT_BUF_SIZE\n"
1320 " -l enables code-path that uses librte_ipsec\n"
1321 " -w REPLAY_WINDOW_SIZE specifies IPsec SQN replay window\n"
1322 " size for each SA\n"
1324 " -a enables SA SQN atomic behaviour\n"
1325 " -c specifies inbound SAD cache size,\n"
1326 " zero value disables the cache (default value: 128)\n"
1327 " -s number of mbufs in packet pool, if not specified number\n"
1328 " of mbufs will be calculated based on number of cores,\n"
1329 " ports and crypto queues\n"
1330 " -f CONFIG_FILE: Configuration file\n"
1331 " --config (port,queue,lcore): Rx queue configuration. In poll\n"
1332 " mode determines which queues from\n"
1333 " which ports are mapped to which cores.\n"
1334 " In event mode this option is not used\n"
1335 " as packets are dynamically scheduled\n"
1336 " to cores by HW.\n"
1337 " --single-sa SAIDX: In poll mode use single SA index for\n"
1338 " outbound traffic, bypassing the SP\n"
1339 " In event mode selects driver submode,\n"
1340 " SA index value is ignored\n"
1341 " --cryptodev_mask MASK: Hexadecimal bitmask of the crypto\n"
1342 " devices to configure\n"
1343 " --transfer-mode MODE\n"
1344 " \"poll\" : Packet transfer via polling (default)\n"
1345 " \"event\" : Packet transfer via event device\n"
1346 " --event-schedule-type TYPE queue schedule type, used only when\n"
1347 " transfer mode is set to event\n"
1348 " \"ordered\" : Ordered (default)\n"
1349 " \"atomic\" : Atomic\n"
1350 " \"parallel\" : Parallel\n"
1351 " --" CMD_LINE_OPT_RX_OFFLOAD
1352 ": bitmask of the RX HW offload capabilities to enable/use\n"
1353 " (DEV_RX_OFFLOAD_*)\n"
1354 " --" CMD_LINE_OPT_TX_OFFLOAD
1355 ": bitmask of the TX HW offload capabilities to enable/use\n"
1356 " (DEV_TX_OFFLOAD_*)\n"
1357 " --" CMD_LINE_OPT_REASSEMBLE " NUM"
1358 ": max number of entries in reassemble(fragment) table\n"
1359 " (zero (default value) disables reassembly)\n"
1360 " --" CMD_LINE_OPT_MTU " MTU"
1361 ": MTU value on all ports (default value: 1500)\n"
1362 " outgoing packets with bigger size will be fragmented\n"
1363 " incoming packets with bigger size will be discarded\n"
1364 " --" CMD_LINE_OPT_FRAG_TTL " FRAG_TTL_NS"
1365 ": fragments lifetime in nanoseconds, default\n"
1366 " and maximum value is 10.000.000.000 ns (10 s)\n"
1372 parse_mask(const char *str, uint64_t *val)
1378 t = strtoul(str, &end, 0);
1379 if (errno != 0 || end[0] != 0)
1387 parse_portmask(const char *portmask)
1392 /* parse hexadecimal string */
1393 pm = strtoul(portmask, &end, 16);
1394 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1397 if ((pm == 0) && errno)
1404 parse_decimal(const char *str)
1409 num = strtoull(str, &end, 10);
1410 if ((str[0] == '\0') || (end == NULL) || (*end != '\0')
1418 parse_config(const char *q_arg)
1421 const char *p, *p0 = q_arg;
1429 unsigned long int_fld[_NUM_FLD];
1430 char *str_fld[_NUM_FLD];
1434 nb_lcore_params = 0;
1436 while ((p = strchr(p0, '(')) != NULL) {
1438 p0 = strchr(p, ')');
1443 if (size >= sizeof(s))
1446 snprintf(s, sizeof(s), "%.*s", size, p);
1447 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1450 for (i = 0; i < _NUM_FLD; i++) {
1452 int_fld[i] = strtoul(str_fld[i], &end, 0);
1453 if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
1456 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1457 printf("exceeded max number of lcore params: %hu\n",
1461 lcore_params_array[nb_lcore_params].port_id =
1462 (uint8_t)int_fld[FLD_PORT];
1463 lcore_params_array[nb_lcore_params].queue_id =
1464 (uint8_t)int_fld[FLD_QUEUE];
1465 lcore_params_array[nb_lcore_params].lcore_id =
1466 (uint8_t)int_fld[FLD_LCORE];
1469 lcore_params = lcore_params_array;
1474 print_app_sa_prm(const struct app_sa_prm *prm)
1476 printf("librte_ipsec usage: %s\n",
1477 (prm->enable == 0) ? "disabled" : "enabled");
1479 printf("replay window size: %u\n", prm->window_size);
1480 printf("ESN: %s\n", (prm->enable_esn == 0) ? "disabled" : "enabled");
1481 printf("SA flags: %#" PRIx64 "\n", prm->flags);
1482 printf("Frag TTL: %" PRIu64 " ns\n", frag_ttl_ns);
1486 parse_transfer_mode(struct eh_conf *conf, const char *optarg)
1488 if (!strcmp(CMD_LINE_ARG_POLL, optarg))
1489 conf->mode = EH_PKT_TRANSFER_MODE_POLL;
1490 else if (!strcmp(CMD_LINE_ARG_EVENT, optarg))
1491 conf->mode = EH_PKT_TRANSFER_MODE_EVENT;
1493 printf("Unsupported packet transfer mode\n");
1501 parse_schedule_type(struct eh_conf *conf, const char *optarg)
1503 struct eventmode_conf *em_conf = NULL;
1505 /* Get eventmode conf */
1506 em_conf = conf->mode_params;
1508 if (!strcmp(CMD_LINE_ARG_ORDERED, optarg))
1509 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_ORDERED;
1510 else if (!strcmp(CMD_LINE_ARG_ATOMIC, optarg))
1511 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_ATOMIC;
1512 else if (!strcmp(CMD_LINE_ARG_PARALLEL, optarg))
1513 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_PARALLEL;
1515 printf("Unsupported queue schedule type\n");
1523 parse_args(int32_t argc, char **argv, struct eh_conf *eh_conf)
1528 int32_t option_index;
1529 char *prgname = argv[0];
1530 int32_t f_present = 0;
1534 while ((opt = getopt_long(argc, argvopt, "aelp:Pu:f:j:w:c:s:",
1535 lgopts, &option_index)) != EOF) {
1539 enabled_port_mask = parse_portmask(optarg);
1540 if (enabled_port_mask == 0) {
1541 printf("invalid portmask\n");
1542 print_usage(prgname);
1547 printf("Promiscuous mode selected\n");
1551 unprotected_port_mask = parse_portmask(optarg);
1552 if (unprotected_port_mask == 0) {
1553 printf("invalid unprotected portmask\n");
1554 print_usage(prgname);
1559 if (f_present == 1) {
1560 printf("\"-f\" option present more than "
1562 print_usage(prgname);
1570 ret = parse_decimal(optarg);
1572 printf("Invalid number of buffers in a pool: "
1574 print_usage(prgname);
1578 nb_bufs_in_pool = ret;
1582 ret = parse_decimal(optarg);
1583 if (ret < RTE_MBUF_DEFAULT_BUF_SIZE ||
1585 printf("Invalid frame buffer size value: %s\n",
1587 print_usage(prgname);
1590 frame_buf_size = ret;
1591 printf("Custom frame buffer size %u\n", frame_buf_size);
1594 app_sa_prm.enable = 1;
1597 app_sa_prm.window_size = parse_decimal(optarg);
1600 app_sa_prm.enable_esn = 1;
1603 app_sa_prm.enable = 1;
1604 app_sa_prm.flags |= RTE_IPSEC_SAFLAG_SQN_ATOM;
1607 ret = parse_decimal(optarg);
1609 printf("Invalid SA cache size: %s\n", optarg);
1610 print_usage(prgname);
1613 app_sa_prm.cache_sz = ret;
1615 case CMD_LINE_OPT_CONFIG_NUM:
1616 ret = parse_config(optarg);
1618 printf("Invalid config\n");
1619 print_usage(prgname);
1623 case CMD_LINE_OPT_SINGLE_SA_NUM:
1624 ret = parse_decimal(optarg);
1625 if (ret == -1 || ret > UINT32_MAX) {
1626 printf("Invalid argument[sa_idx]\n");
1627 print_usage(prgname);
1633 single_sa_idx = ret;
1634 eh_conf->ipsec_mode = EH_IPSEC_MODE_TYPE_DRIVER;
1635 printf("Configured with single SA index %u\n",
1638 case CMD_LINE_OPT_CRYPTODEV_MASK_NUM:
1639 ret = parse_portmask(optarg);
1641 printf("Invalid argument[portmask]\n");
1642 print_usage(prgname);
1647 enabled_cryptodev_mask = ret;
1650 case CMD_LINE_OPT_TRANSFER_MODE_NUM:
1651 ret = parse_transfer_mode(eh_conf, optarg);
1653 printf("Invalid packet transfer mode\n");
1654 print_usage(prgname);
1659 case CMD_LINE_OPT_SCHEDULE_TYPE_NUM:
1660 ret = parse_schedule_type(eh_conf, optarg);
1662 printf("Invalid queue schedule type\n");
1663 print_usage(prgname);
1668 case CMD_LINE_OPT_RX_OFFLOAD_NUM:
1669 ret = parse_mask(optarg, &dev_rx_offload);
1671 printf("Invalid argument for \'%s\': %s\n",
1672 CMD_LINE_OPT_RX_OFFLOAD, optarg);
1673 print_usage(prgname);
1677 case CMD_LINE_OPT_TX_OFFLOAD_NUM:
1678 ret = parse_mask(optarg, &dev_tx_offload);
1680 printf("Invalid argument for \'%s\': %s\n",
1681 CMD_LINE_OPT_TX_OFFLOAD, optarg);
1682 print_usage(prgname);
1686 case CMD_LINE_OPT_REASSEMBLE_NUM:
1687 ret = parse_decimal(optarg);
1688 if (ret < 0 || ret > UINT32_MAX) {
1689 printf("Invalid argument for \'%s\': %s\n",
1690 CMD_LINE_OPT_REASSEMBLE, optarg);
1691 print_usage(prgname);
1696 case CMD_LINE_OPT_MTU_NUM:
1697 ret = parse_decimal(optarg);
1698 if (ret < 0 || ret > RTE_IPV4_MAX_PKT_LEN) {
1699 printf("Invalid argument for \'%s\': %s\n",
1700 CMD_LINE_OPT_MTU, optarg);
1701 print_usage(prgname);
1706 case CMD_LINE_OPT_FRAG_TTL_NUM:
1707 ret = parse_decimal(optarg);
1708 if (ret < 0 || ret > MAX_FRAG_TTL_NS) {
1709 printf("Invalid argument for \'%s\': %s\n",
1710 CMD_LINE_OPT_MTU, optarg);
1711 print_usage(prgname);
1717 print_usage(prgname);
1722 if (f_present == 0) {
1723 printf("Mandatory option \"-f\" not present\n");
1727 /* check do we need to enable multi-seg support */
1728 if (multi_seg_required()) {
1729 /* legacy mode doesn't support multi-seg */
1730 app_sa_prm.enable = 1;
1731 printf("frame buf size: %u, mtu: %u, "
1732 "number of reassemble entries: %u\n"
1733 "multi-segment support is required\n",
1734 frame_buf_size, mtu_size, frag_tbl_sz);
1737 print_app_sa_prm(&app_sa_prm);
1740 argv[optind-1] = prgname;
1743 optind = 1; /* reset getopt lib */
1748 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
1750 char buf[RTE_ETHER_ADDR_FMT_SIZE];
1751 rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
1752 printf("%s%s", name, buf);
1756 * Update destination ethaddr for the port.
1759 add_dst_ethaddr(uint16_t port, const struct rte_ether_addr *addr)
1761 if (port >= RTE_DIM(ethaddr_tbl))
1764 ethaddr_tbl[port].dst = ETHADDR_TO_UINT64(addr);
1768 /* Check the link status of all ports in up to 9s, and print them finally */
1770 check_all_ports_link_status(uint32_t port_mask)
1772 #define CHECK_INTERVAL 100 /* 100ms */
1773 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1775 uint8_t count, all_ports_up, print_flag = 0;
1776 struct rte_eth_link link;
1779 printf("\nChecking link status");
1781 for (count = 0; count <= MAX_CHECK_TIME; count++) {
1783 RTE_ETH_FOREACH_DEV(portid) {
1784 if ((port_mask & (1 << portid)) == 0)
1786 memset(&link, 0, sizeof(link));
1787 ret = rte_eth_link_get_nowait(portid, &link);
1790 if (print_flag == 1)
1791 printf("Port %u link get failed: %s\n",
1792 portid, rte_strerror(-ret));
1795 /* print link status if flag set */
1796 if (print_flag == 1) {
1797 if (link.link_status)
1799 "Port%d Link Up - speed %u Mbps -%s\n",
1800 portid, link.link_speed,
1801 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1802 ("full-duplex") : ("half-duplex\n"));
1804 printf("Port %d Link Down\n", portid);
1807 /* clear all_ports_up flag if any link down */
1808 if (link.link_status == ETH_LINK_DOWN) {
1813 /* after finally printing all link status, get out */
1814 if (print_flag == 1)
1817 if (all_ports_up == 0) {
1820 rte_delay_ms(CHECK_INTERVAL);
1823 /* set the print_flag if all ports up or timeout */
1824 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1832 add_mapping(struct rte_hash *map, const char *str, uint16_t cdev_id,
1833 uint16_t qp, struct lcore_params *params,
1834 struct ipsec_ctx *ipsec_ctx,
1835 const struct rte_cryptodev_capabilities *cipher,
1836 const struct rte_cryptodev_capabilities *auth,
1837 const struct rte_cryptodev_capabilities *aead)
1841 struct cdev_key key = { 0 };
1843 key.lcore_id = params->lcore_id;
1845 key.cipher_algo = cipher->sym.cipher.algo;
1847 key.auth_algo = auth->sym.auth.algo;
1849 key.aead_algo = aead->sym.aead.algo;
1851 ret = rte_hash_lookup(map, &key);
1855 for (i = 0; i < ipsec_ctx->nb_qps; i++)
1856 if (ipsec_ctx->tbl[i].id == cdev_id)
1859 if (i == ipsec_ctx->nb_qps) {
1860 if (ipsec_ctx->nb_qps == MAX_QP_PER_LCORE) {
1861 printf("Maximum number of crypto devices assigned to "
1862 "a core, increase MAX_QP_PER_LCORE value\n");
1865 ipsec_ctx->tbl[i].id = cdev_id;
1866 ipsec_ctx->tbl[i].qp = qp;
1867 ipsec_ctx->nb_qps++;
1868 printf("%s cdev mapping: lcore %u using cdev %u qp %u "
1869 "(cdev_id_qp %lu)\n", str, key.lcore_id,
1873 ret = rte_hash_add_key_data(map, &key, (void *)i);
1875 printf("Faled to insert cdev mapping for (lcore %u, "
1876 "cdev %u, qp %u), errno %d\n",
1877 key.lcore_id, ipsec_ctx->tbl[i].id,
1878 ipsec_ctx->tbl[i].qp, ret);
1886 add_cdev_mapping(struct rte_cryptodev_info *dev_info, uint16_t cdev_id,
1887 uint16_t qp, struct lcore_params *params)
1890 const struct rte_cryptodev_capabilities *i, *j;
1891 struct rte_hash *map;
1892 struct lcore_conf *qconf;
1893 struct ipsec_ctx *ipsec_ctx;
1896 qconf = &lcore_conf[params->lcore_id];
1898 if ((unprotected_port_mask & (1 << params->port_id)) == 0) {
1900 ipsec_ctx = &qconf->outbound;
1904 ipsec_ctx = &qconf->inbound;
1908 /* Required cryptodevs with operation chainning */
1909 if (!(dev_info->feature_flags &
1910 RTE_CRYPTODEV_FF_SYM_OPERATION_CHAINING))
1913 for (i = dev_info->capabilities;
1914 i->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; i++) {
1915 if (i->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1918 if (i->sym.xform_type == RTE_CRYPTO_SYM_XFORM_AEAD) {
1919 ret |= add_mapping(map, str, cdev_id, qp, params,
1920 ipsec_ctx, NULL, NULL, i);
1924 if (i->sym.xform_type != RTE_CRYPTO_SYM_XFORM_CIPHER)
1927 for (j = dev_info->capabilities;
1928 j->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; j++) {
1929 if (j->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1932 if (j->sym.xform_type != RTE_CRYPTO_SYM_XFORM_AUTH)
1935 ret |= add_mapping(map, str, cdev_id, qp, params,
1936 ipsec_ctx, i, j, NULL);
1943 /* Check if the device is enabled by cryptodev_mask */
1945 check_cryptodev_mask(uint8_t cdev_id)
1947 if (enabled_cryptodev_mask & (1 << cdev_id))
1954 cryptodevs_init(uint16_t req_queue_num)
1956 struct rte_cryptodev_config dev_conf;
1957 struct rte_cryptodev_qp_conf qp_conf;
1958 uint16_t idx, max_nb_qps, qp, total_nb_qps, i;
1960 struct rte_hash_parameters params = { 0 };
1962 const uint64_t mseg_flag = multi_seg_required() ?
1963 RTE_CRYPTODEV_FF_IN_PLACE_SGL : 0;
1965 params.entries = CDEV_MAP_ENTRIES;
1966 params.key_len = sizeof(struct cdev_key);
1967 params.hash_func = rte_jhash;
1968 params.hash_func_init_val = 0;
1969 params.socket_id = rte_socket_id();
1971 params.name = "cdev_map_in";
1972 cdev_map_in = rte_hash_create(¶ms);
1973 if (cdev_map_in == NULL)
1974 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1977 params.name = "cdev_map_out";
1978 cdev_map_out = rte_hash_create(¶ms);
1979 if (cdev_map_out == NULL)
1980 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1983 printf("lcore/cryptodev/qp mappings:\n");
1987 for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
1988 struct rte_cryptodev_info cdev_info;
1990 if (check_cryptodev_mask((uint8_t)cdev_id))
1993 rte_cryptodev_info_get(cdev_id, &cdev_info);
1995 if ((mseg_flag & cdev_info.feature_flags) != mseg_flag)
1996 rte_exit(EXIT_FAILURE,
1997 "Device %hd does not support \'%s\' feature\n",
1999 rte_cryptodev_get_feature_name(mseg_flag));
2001 if (nb_lcore_params > cdev_info.max_nb_queue_pairs)
2002 max_nb_qps = cdev_info.max_nb_queue_pairs;
2004 max_nb_qps = nb_lcore_params;
2008 while (qp < max_nb_qps && i < nb_lcore_params) {
2009 if (add_cdev_mapping(&cdev_info, cdev_id, qp,
2010 &lcore_params[idx]))
2013 idx = idx % nb_lcore_params;
2017 qp = RTE_MIN(max_nb_qps, RTE_MAX(req_queue_num, qp));
2022 dev_conf.socket_id = rte_cryptodev_socket_id(cdev_id);
2023 dev_conf.nb_queue_pairs = qp;
2024 dev_conf.ff_disable = RTE_CRYPTODEV_FF_ASYMMETRIC_CRYPTO;
2026 uint32_t dev_max_sess = cdev_info.sym.max_nb_sessions;
2027 if (dev_max_sess != 0 &&
2028 dev_max_sess < get_nb_crypto_sessions())
2029 rte_exit(EXIT_FAILURE,
2030 "Device does not support at least %u "
2031 "sessions", get_nb_crypto_sessions());
2033 if (rte_cryptodev_configure(cdev_id, &dev_conf))
2034 rte_panic("Failed to initialize cryptodev %u\n",
2037 qp_conf.nb_descriptors = CDEV_QUEUE_DESC;
2038 qp_conf.mp_session =
2039 socket_ctx[dev_conf.socket_id].session_pool;
2040 qp_conf.mp_session_private =
2041 socket_ctx[dev_conf.socket_id].session_priv_pool;
2042 for (qp = 0; qp < dev_conf.nb_queue_pairs; qp++)
2043 if (rte_cryptodev_queue_pair_setup(cdev_id, qp,
2044 &qp_conf, dev_conf.socket_id))
2045 rte_panic("Failed to setup queue %u for "
2046 "cdev_id %u\n", 0, cdev_id);
2048 if (rte_cryptodev_start(cdev_id))
2049 rte_panic("Failed to start cryptodev %u\n",
2055 return total_nb_qps;
2059 port_init(uint16_t portid, uint64_t req_rx_offloads, uint64_t req_tx_offloads)
2061 uint32_t frame_size;
2062 struct rte_eth_dev_info dev_info;
2063 struct rte_eth_txconf *txconf;
2064 uint16_t nb_tx_queue, nb_rx_queue;
2065 uint16_t tx_queueid, rx_queueid, queue, lcore_id;
2066 int32_t ret, socket_id;
2067 struct lcore_conf *qconf;
2068 struct rte_ether_addr ethaddr;
2069 struct rte_eth_conf local_port_conf = port_conf;
2071 ret = rte_eth_dev_info_get(portid, &dev_info);
2073 rte_exit(EXIT_FAILURE,
2074 "Error during getting device (port %u) info: %s\n",
2075 portid, strerror(-ret));
2077 /* limit allowed HW offloafs, as user requested */
2078 dev_info.rx_offload_capa &= dev_rx_offload;
2079 dev_info.tx_offload_capa &= dev_tx_offload;
2081 printf("Configuring device port %u:\n", portid);
2083 ret = rte_eth_macaddr_get(portid, ðaddr);
2085 rte_exit(EXIT_FAILURE,
2086 "Error getting MAC address (port %u): %s\n",
2087 portid, rte_strerror(-ret));
2089 ethaddr_tbl[portid].src = ETHADDR_TO_UINT64(ðaddr);
2090 print_ethaddr("Address: ", ðaddr);
2093 nb_rx_queue = get_port_nb_rx_queues(portid);
2094 nb_tx_queue = nb_lcores;
2096 if (nb_rx_queue > dev_info.max_rx_queues)
2097 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
2098 "(max rx queue is %u)\n",
2099 nb_rx_queue, dev_info.max_rx_queues);
2101 if (nb_tx_queue > dev_info.max_tx_queues)
2102 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
2103 "(max tx queue is %u)\n",
2104 nb_tx_queue, dev_info.max_tx_queues);
2106 printf("Creating queues: nb_rx_queue=%d nb_tx_queue=%u...\n",
2107 nb_rx_queue, nb_tx_queue);
2109 frame_size = MTU_TO_FRAMELEN(mtu_size);
2110 if (frame_size > local_port_conf.rxmode.max_rx_pkt_len)
2111 local_port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2112 local_port_conf.rxmode.max_rx_pkt_len = frame_size;
2114 if (multi_seg_required()) {
2115 local_port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_SCATTER;
2116 local_port_conf.txmode.offloads |= DEV_TX_OFFLOAD_MULTI_SEGS;
2119 local_port_conf.rxmode.offloads |= req_rx_offloads;
2120 local_port_conf.txmode.offloads |= req_tx_offloads;
2122 /* Check that all required capabilities are supported */
2123 if ((local_port_conf.rxmode.offloads & dev_info.rx_offload_capa) !=
2124 local_port_conf.rxmode.offloads)
2125 rte_exit(EXIT_FAILURE,
2126 "Error: port %u required RX offloads: 0x%" PRIx64
2127 ", avaialbe RX offloads: 0x%" PRIx64 "\n",
2128 portid, local_port_conf.rxmode.offloads,
2129 dev_info.rx_offload_capa);
2131 if ((local_port_conf.txmode.offloads & dev_info.tx_offload_capa) !=
2132 local_port_conf.txmode.offloads)
2133 rte_exit(EXIT_FAILURE,
2134 "Error: port %u required TX offloads: 0x%" PRIx64
2135 ", avaialbe TX offloads: 0x%" PRIx64 "\n",
2136 portid, local_port_conf.txmode.offloads,
2137 dev_info.tx_offload_capa);
2139 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
2140 local_port_conf.txmode.offloads |=
2141 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
2143 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_IPV4_CKSUM)
2144 local_port_conf.txmode.offloads |= DEV_TX_OFFLOAD_IPV4_CKSUM;
2146 printf("port %u configurng rx_offloads=0x%" PRIx64
2147 ", tx_offloads=0x%" PRIx64 "\n",
2148 portid, local_port_conf.rxmode.offloads,
2149 local_port_conf.txmode.offloads);
2151 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
2152 dev_info.flow_type_rss_offloads;
2153 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
2154 port_conf.rx_adv_conf.rss_conf.rss_hf) {
2155 printf("Port %u modified RSS hash function based on hardware support,"
2156 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
2158 port_conf.rx_adv_conf.rss_conf.rss_hf,
2159 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
2162 ret = rte_eth_dev_configure(portid, nb_rx_queue, nb_tx_queue,
2165 rte_exit(EXIT_FAILURE, "Cannot configure device: "
2166 "err=%d, port=%d\n", ret, portid);
2168 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd, &nb_txd);
2170 rte_exit(EXIT_FAILURE, "Cannot adjust number of descriptors: "
2171 "err=%d, port=%d\n", ret, portid);
2173 /* init one TX queue per lcore */
2175 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2176 if (rte_lcore_is_enabled(lcore_id) == 0)
2180 socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2185 printf("Setup txq=%u,%d,%d\n", lcore_id, tx_queueid, socket_id);
2187 txconf = &dev_info.default_txconf;
2188 txconf->offloads = local_port_conf.txmode.offloads;
2190 ret = rte_eth_tx_queue_setup(portid, tx_queueid, nb_txd,
2193 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
2194 "err=%d, port=%d\n", ret, portid);
2196 qconf = &lcore_conf[lcore_id];
2197 qconf->tx_queue_id[portid] = tx_queueid;
2199 /* Pre-populate pkt offloads based on capabilities */
2200 qconf->outbound.ipv4_offloads = PKT_TX_IPV4;
2201 qconf->outbound.ipv6_offloads = PKT_TX_IPV6;
2202 if (local_port_conf.txmode.offloads & DEV_TX_OFFLOAD_IPV4_CKSUM)
2203 qconf->outbound.ipv4_offloads |= PKT_TX_IP_CKSUM;
2207 /* init RX queues */
2208 for (queue = 0; queue < qconf->nb_rx_queue; ++queue) {
2209 struct rte_eth_rxconf rxq_conf;
2211 if (portid != qconf->rx_queue_list[queue].port_id)
2214 rx_queueid = qconf->rx_queue_list[queue].queue_id;
2216 printf("Setup rxq=%d,%d,%d\n", portid, rx_queueid,
2219 rxq_conf = dev_info.default_rxconf;
2220 rxq_conf.offloads = local_port_conf.rxmode.offloads;
2221 ret = rte_eth_rx_queue_setup(portid, rx_queueid,
2222 nb_rxd, socket_id, &rxq_conf,
2223 socket_ctx[socket_id].mbuf_pool);
2225 rte_exit(EXIT_FAILURE,
2226 "rte_eth_rx_queue_setup: err=%d, "
2227 "port=%d\n", ret, portid);
2234 max_session_size(void)
2238 int16_t cdev_id, port_id, n;
2241 n = rte_cryptodev_count();
2242 for (cdev_id = 0; cdev_id != n; cdev_id++) {
2243 sz = rte_cryptodev_sym_get_private_session_size(cdev_id);
2247 * If crypto device is security capable, need to check the
2248 * size of security session as well.
2251 /* Get security context of the crypto device */
2252 sec_ctx = rte_cryptodev_get_sec_ctx(cdev_id);
2253 if (sec_ctx == NULL)
2256 /* Get size of security session */
2257 sz = rte_security_session_get_size(sec_ctx);
2262 RTE_ETH_FOREACH_DEV(port_id) {
2263 if ((enabled_port_mask & (1 << port_id)) == 0)
2266 sec_ctx = rte_eth_dev_get_sec_ctx(port_id);
2267 if (sec_ctx == NULL)
2270 sz = rte_security_session_get_size(sec_ctx);
2279 session_pool_init(struct socket_ctx *ctx, int32_t socket_id, size_t sess_sz)
2281 char mp_name[RTE_MEMPOOL_NAMESIZE];
2282 struct rte_mempool *sess_mp;
2285 snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
2286 "sess_mp_%u", socket_id);
2288 * Doubled due to rte_security_session_create() uses one mempool for
2289 * session and for session private data.
2291 nb_sess = (get_nb_crypto_sessions() + CDEV_MP_CACHE_SZ *
2292 rte_lcore_count()) * 2;
2293 sess_mp = rte_cryptodev_sym_session_pool_create(
2294 mp_name, nb_sess, sess_sz, CDEV_MP_CACHE_SZ, 0,
2296 ctx->session_pool = sess_mp;
2298 if (ctx->session_pool == NULL)
2299 rte_exit(EXIT_FAILURE,
2300 "Cannot init session pool on socket %d\n", socket_id);
2302 printf("Allocated session pool on socket %d\n", socket_id);
2306 session_priv_pool_init(struct socket_ctx *ctx, int32_t socket_id,
2309 char mp_name[RTE_MEMPOOL_NAMESIZE];
2310 struct rte_mempool *sess_mp;
2313 snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
2314 "sess_mp_priv_%u", socket_id);
2316 * Doubled due to rte_security_session_create() uses one mempool for
2317 * session and for session private data.
2319 nb_sess = (get_nb_crypto_sessions() + CDEV_MP_CACHE_SZ *
2320 rte_lcore_count()) * 2;
2321 sess_mp = rte_mempool_create(mp_name,
2325 0, NULL, NULL, NULL,
2328 ctx->session_priv_pool = sess_mp;
2330 if (ctx->session_priv_pool == NULL)
2331 rte_exit(EXIT_FAILURE,
2332 "Cannot init session priv pool on socket %d\n",
2335 printf("Allocated session priv pool on socket %d\n",
2340 pool_init(struct socket_ctx *ctx, int32_t socket_id, uint32_t nb_mbuf)
2345 snprintf(s, sizeof(s), "mbuf_pool_%d", socket_id);
2346 ctx->mbuf_pool = rte_pktmbuf_pool_create(s, nb_mbuf,
2347 MEMPOOL_CACHE_SIZE, ipsec_metadata_size(),
2348 frame_buf_size, socket_id);
2351 * if multi-segment support is enabled, then create a pool
2352 * for indirect mbufs.
2354 ms = multi_seg_required();
2356 snprintf(s, sizeof(s), "mbuf_pool_indir_%d", socket_id);
2357 ctx->mbuf_pool_indir = rte_pktmbuf_pool_create(s, nb_mbuf,
2358 MEMPOOL_CACHE_SIZE, 0, 0, socket_id);
2361 if (ctx->mbuf_pool == NULL || (ms != 0 && ctx->mbuf_pool_indir == NULL))
2362 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool on socket %d\n",
2365 printf("Allocated mbuf pool on socket %d\n", socket_id);
2369 inline_ipsec_event_esn_overflow(struct rte_security_ctx *ctx, uint64_t md)
2371 struct ipsec_sa *sa;
2373 /* For inline protocol processing, the metadata in the event will
2374 * uniquely identify the security session which raised the event.
2375 * Application would then need the userdata it had registered with the
2376 * security session to process the event.
2379 sa = (struct ipsec_sa *)rte_security_get_userdata(ctx, md);
2382 /* userdata could not be retrieved */
2386 /* Sequence number over flow. SA need to be re-established */
2392 inline_ipsec_event_callback(uint16_t port_id, enum rte_eth_event_type type,
2393 void *param, void *ret_param)
2396 struct rte_eth_event_ipsec_desc *event_desc = NULL;
2397 struct rte_security_ctx *ctx = (struct rte_security_ctx *)
2398 rte_eth_dev_get_sec_ctx(port_id);
2400 RTE_SET_USED(param);
2402 if (type != RTE_ETH_EVENT_IPSEC)
2405 event_desc = ret_param;
2406 if (event_desc == NULL) {
2407 printf("Event descriptor not set\n");
2411 md = event_desc->metadata;
2413 if (event_desc->subtype == RTE_ETH_EVENT_IPSEC_ESN_OVERFLOW)
2414 return inline_ipsec_event_esn_overflow(ctx, md);
2415 else if (event_desc->subtype >= RTE_ETH_EVENT_IPSEC_MAX) {
2416 printf("Invalid IPsec event reported\n");
2424 rx_callback(__rte_unused uint16_t port, __rte_unused uint16_t queue,
2425 struct rte_mbuf *pkt[], uint16_t nb_pkts,
2426 __rte_unused uint16_t max_pkts, void *user_param)
2430 struct lcore_conf *lc;
2431 struct rte_mbuf *mb;
2432 struct rte_ether_hdr *eth;
2438 for (i = 0; i != nb_pkts; i++) {
2441 eth = rte_pktmbuf_mtod(mb, struct rte_ether_hdr *);
2442 if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
2444 struct rte_ipv4_hdr *iph;
2446 iph = (struct rte_ipv4_hdr *)(eth + 1);
2447 if (rte_ipv4_frag_pkt_is_fragmented(iph)) {
2449 mb->l2_len = sizeof(*eth);
2450 mb->l3_len = sizeof(*iph);
2451 tm = (tm != 0) ? tm : rte_rdtsc();
2452 mb = rte_ipv4_frag_reassemble_packet(
2453 lc->frag.tbl, &lc->frag.dr,
2457 /* fix ip cksum after reassemble. */
2458 iph = rte_pktmbuf_mtod_offset(mb,
2459 struct rte_ipv4_hdr *,
2461 iph->hdr_checksum = 0;
2462 iph->hdr_checksum = rte_ipv4_cksum(iph);
2465 } else if (eth->ether_type ==
2466 rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6)) {
2468 struct rte_ipv6_hdr *iph;
2469 struct ipv6_extension_fragment *fh;
2471 iph = (struct rte_ipv6_hdr *)(eth + 1);
2472 fh = rte_ipv6_frag_get_ipv6_fragment_header(iph);
2474 mb->l2_len = sizeof(*eth);
2475 mb->l3_len = (uintptr_t)fh - (uintptr_t)iph +
2477 tm = (tm != 0) ? tm : rte_rdtsc();
2478 mb = rte_ipv6_frag_reassemble_packet(
2479 lc->frag.tbl, &lc->frag.dr,
2482 /* fix l3_len after reassemble. */
2483 mb->l3_len = mb->l3_len - sizeof(*fh);
2491 /* some fragments were encountered, drain death row */
2493 rte_ip_frag_free_death_row(&lc->frag.dr, 0);
2500 reassemble_lcore_init(struct lcore_conf *lc, uint32_t cid)
2504 uint64_t frag_cycles;
2505 const struct lcore_rx_queue *rxq;
2506 const struct rte_eth_rxtx_callback *cb;
2508 /* create fragment table */
2509 sid = rte_lcore_to_socket_id(cid);
2510 frag_cycles = (rte_get_tsc_hz() + NS_PER_S - 1) /
2511 NS_PER_S * frag_ttl_ns;
2513 lc->frag.tbl = rte_ip_frag_table_create(frag_tbl_sz,
2514 FRAG_TBL_BUCKET_ENTRIES, frag_tbl_sz, frag_cycles, sid);
2515 if (lc->frag.tbl == NULL) {
2516 printf("%s(%u): failed to create fragment table of size: %u, "
2518 __func__, cid, frag_tbl_sz, rte_errno);
2522 /* setup reassemble RX callbacks for all queues */
2523 for (i = 0; i != lc->nb_rx_queue; i++) {
2525 rxq = lc->rx_queue_list + i;
2526 cb = rte_eth_add_rx_callback(rxq->port_id, rxq->queue_id,
2529 printf("%s(%u): failed to install RX callback for "
2530 "portid=%u, queueid=%u, error code: %d\n",
2532 rxq->port_id, rxq->queue_id, rte_errno);
2541 reassemble_init(void)
2547 for (i = 0; i != nb_lcore_params; i++) {
2548 lc = lcore_params[i].lcore_id;
2549 rc = reassemble_lcore_init(lcore_conf + lc, lc);
2558 create_default_ipsec_flow(uint16_t port_id, uint64_t rx_offloads)
2560 struct rte_flow_action action[2];
2561 struct rte_flow_item pattern[2];
2562 struct rte_flow_attr attr = {0};
2563 struct rte_flow_error err;
2564 struct rte_flow *flow;
2567 if (!(rx_offloads & DEV_RX_OFFLOAD_SECURITY))
2570 /* Add the default rte_flow to enable SECURITY for all ESP packets */
2572 pattern[0].type = RTE_FLOW_ITEM_TYPE_ESP;
2573 pattern[0].spec = NULL;
2574 pattern[0].mask = NULL;
2575 pattern[0].last = NULL;
2576 pattern[1].type = RTE_FLOW_ITEM_TYPE_END;
2578 action[0].type = RTE_FLOW_ACTION_TYPE_SECURITY;
2579 action[0].conf = NULL;
2580 action[1].type = RTE_FLOW_ACTION_TYPE_END;
2581 action[1].conf = NULL;
2585 ret = rte_flow_validate(port_id, &attr, pattern, action, &err);
2589 flow = rte_flow_create(port_id, &attr, pattern, action, &err);
2593 flow_info_tbl[port_id].rx_def_flow = flow;
2594 RTE_LOG(INFO, IPSEC,
2595 "Created default flow enabling SECURITY for all ESP traffic on port %d\n",
2600 signal_handler(int signum)
2602 if (signum == SIGINT || signum == SIGTERM) {
2603 printf("\n\nSignal %d received, preparing to exit...\n",
2610 ev_mode_sess_verify(struct ipsec_sa *sa, int nb_sa)
2612 struct rte_ipsec_session *ips;
2618 for (i = 0; i < nb_sa; i++) {
2619 ips = ipsec_get_primary_session(&sa[i]);
2620 if (ips->type != RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL)
2621 rte_exit(EXIT_FAILURE, "Event mode supports only "
2622 "inline protocol sessions\n");
2628 check_event_mode_params(struct eh_conf *eh_conf)
2630 struct eventmode_conf *em_conf = NULL;
2631 struct lcore_params *params;
2634 if (!eh_conf || !eh_conf->mode_params)
2637 /* Get eventmode conf */
2638 em_conf = eh_conf->mode_params;
2640 if (eh_conf->mode == EH_PKT_TRANSFER_MODE_POLL &&
2641 em_conf->ext_params.sched_type != SCHED_TYPE_NOT_SET) {
2642 printf("error: option --event-schedule-type applies only to "
2647 if (eh_conf->mode != EH_PKT_TRANSFER_MODE_EVENT)
2650 /* Set schedule type to ORDERED if it wasn't explicitly set by user */
2651 if (em_conf->ext_params.sched_type == SCHED_TYPE_NOT_SET)
2652 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_ORDERED;
2655 * Event mode currently supports only inline protocol sessions.
2656 * If there are other types of sessions configured then exit with
2659 ev_mode_sess_verify(sa_in, nb_sa_in);
2660 ev_mode_sess_verify(sa_out, nb_sa_out);
2663 /* Option --config does not apply to event mode */
2664 if (nb_lcore_params > 0) {
2665 printf("error: option --config applies only to poll mode\n");
2670 * In order to use the same port_init routine for both poll and event
2671 * modes initialize lcore_params with one queue for each eth port
2673 lcore_params = lcore_params_array;
2674 RTE_ETH_FOREACH_DEV(portid) {
2675 if ((enabled_port_mask & (1 << portid)) == 0)
2678 params = &lcore_params[nb_lcore_params++];
2679 params->port_id = portid;
2680 params->queue_id = 0;
2681 params->lcore_id = rte_get_next_lcore(0, 0, 1);
2688 inline_sessions_free(struct sa_ctx *sa_ctx)
2690 struct rte_ipsec_session *ips;
2691 struct ipsec_sa *sa;
2698 for (i = 0; i < sa_ctx->nb_sa; i++) {
2700 sa = &sa_ctx->sa[i];
2704 ips = ipsec_get_primary_session(sa);
2705 if (ips->type != RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL &&
2706 ips->type != RTE_SECURITY_ACTION_TYPE_INLINE_CRYPTO)
2709 if (!rte_eth_dev_is_valid_port(sa->portid))
2712 ret = rte_security_session_destroy(
2713 rte_eth_dev_get_sec_ctx(sa->portid),
2716 RTE_LOG(ERR, IPSEC, "Failed to destroy security "
2717 "session type %d, spi %d\n",
2718 ips->type, sa->spi);
2723 calculate_nb_mbufs(uint16_t nb_ports, uint16_t nb_crypto_qp, uint32_t nb_rxq,
2726 return RTE_MAX((nb_rxq * nb_rxd +
2727 nb_ports * nb_lcores * MAX_PKT_BURST +
2728 nb_ports * nb_txq * nb_txd +
2729 nb_lcores * MEMPOOL_CACHE_SIZE +
2730 nb_crypto_qp * CDEV_QUEUE_DESC +
2731 nb_lcores * frag_tbl_sz *
2732 FRAG_TBL_BUCKET_ENTRIES),
2737 main(int32_t argc, char **argv)
2740 uint32_t lcore_id, nb_txq, nb_rxq = 0;
2744 uint16_t portid, nb_crypto_qp, nb_ports = 0;
2745 uint64_t req_rx_offloads[RTE_MAX_ETHPORTS];
2746 uint64_t req_tx_offloads[RTE_MAX_ETHPORTS];
2747 struct eh_conf *eh_conf = NULL;
2750 nb_bufs_in_pool = 0;
2753 ret = rte_eal_init(argc, argv);
2755 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
2760 signal(SIGINT, signal_handler);
2761 signal(SIGTERM, signal_handler);
2763 /* initialize event helper configuration */
2764 eh_conf = eh_conf_init();
2765 if (eh_conf == NULL)
2766 rte_exit(EXIT_FAILURE, "Failed to init event helper config");
2768 /* parse application arguments (after the EAL ones) */
2769 ret = parse_args(argc, argv, eh_conf);
2771 rte_exit(EXIT_FAILURE, "Invalid parameters\n");
2773 /* parse configuration file */
2774 if (parse_cfg_file(cfgfile) < 0) {
2775 printf("parsing file \"%s\" failed\n",
2777 print_usage(argv[0]);
2781 if ((unprotected_port_mask & enabled_port_mask) !=
2782 unprotected_port_mask)
2783 rte_exit(EXIT_FAILURE, "Invalid unprotected portmask 0x%x\n",
2784 unprotected_port_mask);
2786 if (check_poll_mode_params(eh_conf) < 0)
2787 rte_exit(EXIT_FAILURE, "check_poll_mode_params failed\n");
2789 if (check_event_mode_params(eh_conf) < 0)
2790 rte_exit(EXIT_FAILURE, "check_event_mode_params failed\n");
2792 ret = init_lcore_rx_queues();
2794 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
2796 nb_lcores = rte_lcore_count();
2798 sess_sz = max_session_size();
2801 * In event mode request minimum number of crypto queues
2802 * to be reserved equal to number of ports.
2804 if (eh_conf->mode == EH_PKT_TRANSFER_MODE_EVENT)
2805 nb_crypto_qp = rte_eth_dev_count_avail();
2809 nb_crypto_qp = cryptodevs_init(nb_crypto_qp);
2811 if (nb_bufs_in_pool == 0) {
2812 RTE_ETH_FOREACH_DEV(portid) {
2813 if ((enabled_port_mask & (1 << portid)) == 0)
2816 nb_rxq += get_port_nb_rx_queues(portid);
2821 nb_bufs_in_pool = calculate_nb_mbufs(nb_ports, nb_crypto_qp,
2825 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2826 if (rte_lcore_is_enabled(lcore_id) == 0)
2830 socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2834 /* mbuf_pool is initialised by the pool_init() function*/
2835 if (socket_ctx[socket_id].mbuf_pool)
2838 pool_init(&socket_ctx[socket_id], socket_id, nb_bufs_in_pool);
2839 session_pool_init(&socket_ctx[socket_id], socket_id, sess_sz);
2840 session_priv_pool_init(&socket_ctx[socket_id], socket_id,
2843 printf("Number of mbufs in packet pool %d\n", nb_bufs_in_pool);
2845 RTE_ETH_FOREACH_DEV(portid) {
2846 if ((enabled_port_mask & (1 << portid)) == 0)
2849 sa_check_offloads(portid, &req_rx_offloads[portid],
2850 &req_tx_offloads[portid]);
2851 port_init(portid, req_rx_offloads[portid],
2852 req_tx_offloads[portid]);
2856 * Set the enabled port mask in helper config for use by helper
2857 * sub-system. This will be used while initializing devices using
2858 * helper sub-system.
2860 eh_conf->eth_portmask = enabled_port_mask;
2862 /* Initialize eventmode components */
2863 ret = eh_devs_init(eh_conf);
2865 rte_exit(EXIT_FAILURE, "eh_devs_init failed, err=%d\n", ret);
2868 RTE_ETH_FOREACH_DEV(portid) {
2869 if ((enabled_port_mask & (1 << portid)) == 0)
2872 /* Create flow before starting the device */
2873 create_default_ipsec_flow(portid, req_rx_offloads[portid]);
2875 ret = rte_eth_dev_start(portid);
2877 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: "
2878 "err=%d, port=%d\n", ret, portid);
2880 * If enabled, put device in promiscuous mode.
2881 * This allows IO forwarding mode to forward packets
2882 * to itself through 2 cross-connected ports of the
2885 if (promiscuous_on) {
2886 ret = rte_eth_promiscuous_enable(portid);
2888 rte_exit(EXIT_FAILURE,
2889 "rte_eth_promiscuous_enable: err=%s, port=%d\n",
2890 rte_strerror(-ret), portid);
2893 rte_eth_dev_callback_register(portid,
2894 RTE_ETH_EVENT_IPSEC, inline_ipsec_event_callback, NULL);
2897 /* fragment reassemble is enabled */
2898 if (frag_tbl_sz != 0) {
2899 ret = reassemble_init();
2901 rte_exit(EXIT_FAILURE, "failed at reassemble init");
2904 /* Replicate each context per socket */
2905 for (i = 0; i < NB_SOCKETS && i < rte_socket_count(); i++) {
2906 socket_id = rte_socket_id_by_idx(i);
2907 if ((socket_ctx[socket_id].mbuf_pool != NULL) &&
2908 (socket_ctx[socket_id].sa_in == NULL) &&
2909 (socket_ctx[socket_id].sa_out == NULL)) {
2910 sa_init(&socket_ctx[socket_id], socket_id);
2911 sp4_init(&socket_ctx[socket_id], socket_id);
2912 sp6_init(&socket_ctx[socket_id], socket_id);
2913 rt_init(&socket_ctx[socket_id], socket_id);
2917 check_all_ports_link_status(enabled_port_mask);
2919 /* launch per-lcore init on every lcore */
2920 rte_eal_mp_remote_launch(ipsec_launch_one_lcore, eh_conf, CALL_MASTER);
2921 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
2922 if (rte_eal_wait_lcore(lcore_id) < 0)
2926 /* Uninitialize eventmode components */
2927 ret = eh_devs_uninit(eh_conf);
2929 rte_exit(EXIT_FAILURE, "eh_devs_uninit failed, err=%d\n", ret);
2931 /* Free eventmode configuration memory */
2932 eh_conf_uninit(eh_conf);
2934 /* Destroy inline inbound and outbound sessions */
2935 for (i = 0; i < NB_SOCKETS && i < rte_socket_count(); i++) {
2936 socket_id = rte_socket_id_by_idx(i);
2937 inline_sessions_free(socket_ctx[socket_id].sa_in);
2938 inline_sessions_free(socket_ctx[socket_id].sa_out);
2941 for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
2942 printf("Closing cryptodev %d...", cdev_id);
2943 rte_cryptodev_stop(cdev_id);
2944 rte_cryptodev_close(cdev_id);
2948 RTE_ETH_FOREACH_DEV(portid) {
2949 if ((enabled_port_mask & (1 << portid)) == 0)
2952 printf("Closing port %d...", portid);
2953 if (flow_info_tbl[portid].rx_def_flow) {
2954 struct rte_flow_error err;
2956 ret = rte_flow_destroy(portid,
2957 flow_info_tbl[portid].rx_def_flow, &err);
2959 RTE_LOG(ERR, IPSEC, "Failed to destroy flow "
2960 " for port %u, err msg: %s\n", portid,
2963 rte_eth_dev_stop(portid);
2964 rte_eth_dev_close(portid);