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"
56 volatile bool force_quit;
58 #define RTE_LOGTYPE_IPSEC RTE_LOGTYPE_USER1
60 #define MAX_JUMBO_PKT_LEN 9600
62 #define MEMPOOL_CACHE_SIZE 256
64 #define NB_MBUF (32000)
66 #define CDEV_QUEUE_DESC 2048
67 #define CDEV_MAP_ENTRIES 16384
68 #define CDEV_MP_NB_OBJS 1024
69 #define CDEV_MP_CACHE_SZ 64
70 #define MAX_QUEUE_PAIRS 1
72 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
74 /* Configure how many packets ahead to prefetch, when reading packets */
75 #define PREFETCH_OFFSET 3
77 #define MAX_RX_QUEUE_PER_LCORE 16
79 #define MAX_LCORE_PARAMS 1024
82 * Configurable number of RX/TX ring descriptors
84 #define IPSEC_SECGW_RX_DESC_DEFAULT 1024
85 #define IPSEC_SECGW_TX_DESC_DEFAULT 1024
86 static uint16_t nb_rxd = IPSEC_SECGW_RX_DESC_DEFAULT;
87 static uint16_t nb_txd = IPSEC_SECGW_TX_DESC_DEFAULT;
89 #if RTE_BYTE_ORDER != RTE_LITTLE_ENDIAN
90 #define __BYTES_TO_UINT64(a, b, c, d, e, f, g, h) \
91 (((uint64_t)((a) & 0xff) << 56) | \
92 ((uint64_t)((b) & 0xff) << 48) | \
93 ((uint64_t)((c) & 0xff) << 40) | \
94 ((uint64_t)((d) & 0xff) << 32) | \
95 ((uint64_t)((e) & 0xff) << 24) | \
96 ((uint64_t)((f) & 0xff) << 16) | \
97 ((uint64_t)((g) & 0xff) << 8) | \
98 ((uint64_t)(h) & 0xff))
100 #define __BYTES_TO_UINT64(a, b, c, d, e, f, g, h) \
101 (((uint64_t)((h) & 0xff) << 56) | \
102 ((uint64_t)((g) & 0xff) << 48) | \
103 ((uint64_t)((f) & 0xff) << 40) | \
104 ((uint64_t)((e) & 0xff) << 32) | \
105 ((uint64_t)((d) & 0xff) << 24) | \
106 ((uint64_t)((c) & 0xff) << 16) | \
107 ((uint64_t)((b) & 0xff) << 8) | \
108 ((uint64_t)(a) & 0xff))
110 #define ETHADDR(a, b, c, d, e, f) (__BYTES_TO_UINT64(a, b, c, d, e, f, 0, 0))
112 #define ETHADDR_TO_UINT64(addr) __BYTES_TO_UINT64( \
113 (addr)->addr_bytes[0], (addr)->addr_bytes[1], \
114 (addr)->addr_bytes[2], (addr)->addr_bytes[3], \
115 (addr)->addr_bytes[4], (addr)->addr_bytes[5], \
118 #define FRAG_TBL_BUCKET_ENTRIES 4
119 #define MAX_FRAG_TTL_NS (10LL * NS_PER_S)
121 #define MTU_TO_FRAMELEN(x) ((x) + RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN)
123 /* port/source ethernet addr and destination ethernet addr */
124 struct ethaddr_info {
128 struct ethaddr_info ethaddr_tbl[RTE_MAX_ETHPORTS] = {
129 { 0, ETHADDR(0x00, 0x16, 0x3e, 0x7e, 0x94, 0x9a) },
130 { 0, ETHADDR(0x00, 0x16, 0x3e, 0x22, 0xa1, 0xd9) },
131 { 0, ETHADDR(0x00, 0x16, 0x3e, 0x08, 0x69, 0x26) },
132 { 0, ETHADDR(0x00, 0x16, 0x3e, 0x49, 0x9e, 0xdd) }
135 struct flow_info flow_info_tbl[RTE_MAX_ETHPORTS];
137 #define CMD_LINE_OPT_CONFIG "config"
138 #define CMD_LINE_OPT_SINGLE_SA "single-sa"
139 #define CMD_LINE_OPT_CRYPTODEV_MASK "cryptodev_mask"
140 #define CMD_LINE_OPT_TRANSFER_MODE "transfer-mode"
141 #define CMD_LINE_OPT_SCHEDULE_TYPE "event-schedule-type"
142 #define CMD_LINE_OPT_RX_OFFLOAD "rxoffload"
143 #define CMD_LINE_OPT_TX_OFFLOAD "txoffload"
144 #define CMD_LINE_OPT_REASSEMBLE "reassemble"
145 #define CMD_LINE_OPT_MTU "mtu"
146 #define CMD_LINE_OPT_FRAG_TTL "frag-ttl"
148 #define CMD_LINE_ARG_EVENT "event"
149 #define CMD_LINE_ARG_POLL "poll"
150 #define CMD_LINE_ARG_ORDERED "ordered"
151 #define CMD_LINE_ARG_ATOMIC "atomic"
152 #define CMD_LINE_ARG_PARALLEL "parallel"
155 /* long options mapped to a short option */
157 /* first long only option value must be >= 256, so that we won't
158 * conflict with short options
160 CMD_LINE_OPT_MIN_NUM = 256,
161 CMD_LINE_OPT_CONFIG_NUM,
162 CMD_LINE_OPT_SINGLE_SA_NUM,
163 CMD_LINE_OPT_CRYPTODEV_MASK_NUM,
164 CMD_LINE_OPT_TRANSFER_MODE_NUM,
165 CMD_LINE_OPT_SCHEDULE_TYPE_NUM,
166 CMD_LINE_OPT_RX_OFFLOAD_NUM,
167 CMD_LINE_OPT_TX_OFFLOAD_NUM,
168 CMD_LINE_OPT_REASSEMBLE_NUM,
169 CMD_LINE_OPT_MTU_NUM,
170 CMD_LINE_OPT_FRAG_TTL_NUM,
173 static const struct option lgopts[] = {
174 {CMD_LINE_OPT_CONFIG, 1, 0, CMD_LINE_OPT_CONFIG_NUM},
175 {CMD_LINE_OPT_SINGLE_SA, 1, 0, CMD_LINE_OPT_SINGLE_SA_NUM},
176 {CMD_LINE_OPT_CRYPTODEV_MASK, 1, 0, CMD_LINE_OPT_CRYPTODEV_MASK_NUM},
177 {CMD_LINE_OPT_TRANSFER_MODE, 1, 0, CMD_LINE_OPT_TRANSFER_MODE_NUM},
178 {CMD_LINE_OPT_SCHEDULE_TYPE, 1, 0, CMD_LINE_OPT_SCHEDULE_TYPE_NUM},
179 {CMD_LINE_OPT_RX_OFFLOAD, 1, 0, CMD_LINE_OPT_RX_OFFLOAD_NUM},
180 {CMD_LINE_OPT_TX_OFFLOAD, 1, 0, CMD_LINE_OPT_TX_OFFLOAD_NUM},
181 {CMD_LINE_OPT_REASSEMBLE, 1, 0, CMD_LINE_OPT_REASSEMBLE_NUM},
182 {CMD_LINE_OPT_MTU, 1, 0, CMD_LINE_OPT_MTU_NUM},
183 {CMD_LINE_OPT_FRAG_TTL, 1, 0, CMD_LINE_OPT_FRAG_TTL_NUM},
187 uint32_t unprotected_port_mask;
188 uint32_t single_sa_idx;
189 /* mask of enabled ports */
190 static uint32_t enabled_port_mask;
191 static uint64_t enabled_cryptodev_mask = UINT64_MAX;
192 static int32_t promiscuous_on = 1;
193 static int32_t numa_on = 1; /**< NUMA is enabled by default. */
194 static uint32_t nb_lcores;
195 static uint32_t single_sa;
198 * RX/TX HW offload capabilities to enable/use on ethernet ports.
199 * By default all capabilities are enabled.
201 static uint64_t dev_rx_offload = UINT64_MAX;
202 static uint64_t dev_tx_offload = UINT64_MAX;
205 * global values that determine multi-seg policy
207 static uint32_t frag_tbl_sz;
208 static uint32_t frame_buf_size = RTE_MBUF_DEFAULT_BUF_SIZE;
209 static uint32_t mtu_size = RTE_ETHER_MTU;
210 static uint64_t frag_ttl_ns = MAX_FRAG_TTL_NS;
212 /* application wide librte_ipsec/SA parameters */
213 struct app_sa_prm app_sa_prm = {
215 .cache_sz = SA_CACHE_SZ
217 static const char *cfgfile;
219 struct lcore_rx_queue {
222 } __rte_cache_aligned;
224 struct lcore_params {
228 } __rte_cache_aligned;
230 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
232 static struct lcore_params *lcore_params;
233 static uint16_t nb_lcore_params;
235 static struct rte_hash *cdev_map_in;
236 static struct rte_hash *cdev_map_out;
240 struct rte_mbuf *m_table[MAX_PKT_BURST] __rte_aligned(sizeof(void *));
244 uint16_t nb_rx_queue;
245 struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
246 uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
247 struct buffer tx_mbufs[RTE_MAX_ETHPORTS];
248 struct ipsec_ctx inbound;
249 struct ipsec_ctx outbound;
250 struct rt_ctx *rt4_ctx;
251 struct rt_ctx *rt6_ctx;
253 struct rte_ip_frag_tbl *tbl;
254 struct rte_mempool *pool_dir;
255 struct rte_mempool *pool_indir;
256 struct rte_ip_frag_death_row dr;
258 } __rte_cache_aligned;
260 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
262 static struct rte_eth_conf port_conf = {
264 .mq_mode = ETH_MQ_RX_RSS,
265 .max_rx_pkt_len = RTE_ETHER_MAX_LEN,
267 .offloads = DEV_RX_OFFLOAD_CHECKSUM,
272 .rss_hf = ETH_RSS_IP | ETH_RSS_UDP |
273 ETH_RSS_TCP | ETH_RSS_SCTP,
277 .mq_mode = ETH_MQ_TX_NONE,
281 struct socket_ctx socket_ctx[NB_SOCKETS];
284 * Determine is multi-segment support required:
285 * - either frame buffer size is smaller then mtu
286 * - or reassmeble support is requested
289 multi_seg_required(void)
291 return (MTU_TO_FRAMELEN(mtu_size) + RTE_PKTMBUF_HEADROOM >
292 frame_buf_size || frag_tbl_sz != 0);
296 adjust_ipv4_pktlen(struct rte_mbuf *m, const struct rte_ipv4_hdr *iph,
301 plen = rte_be_to_cpu_16(iph->total_length) + l2_len;
302 if (plen < m->pkt_len) {
303 trim = m->pkt_len - plen;
304 rte_pktmbuf_trim(m, trim);
309 adjust_ipv6_pktlen(struct rte_mbuf *m, const struct rte_ipv6_hdr *iph,
314 plen = rte_be_to_cpu_16(iph->payload_len) + sizeof(*iph) + l2_len;
315 if (plen < m->pkt_len) {
316 trim = m->pkt_len - plen;
317 rte_pktmbuf_trim(m, trim);
322 prepare_one_packet(struct rte_mbuf *pkt, struct ipsec_traffic *t)
324 const struct rte_ether_hdr *eth;
325 const struct rte_ipv4_hdr *iph4;
326 const struct rte_ipv6_hdr *iph6;
328 eth = rte_pktmbuf_mtod(pkt, const struct rte_ether_hdr *);
329 if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
331 iph4 = (const struct rte_ipv4_hdr *)rte_pktmbuf_adj(pkt,
333 adjust_ipv4_pktlen(pkt, iph4, 0);
335 if (iph4->next_proto_id == IPPROTO_ESP)
336 t->ipsec.pkts[(t->ipsec.num)++] = pkt;
338 t->ip4.data[t->ip4.num] = &iph4->next_proto_id;
339 t->ip4.pkts[(t->ip4.num)++] = pkt;
342 pkt->l3_len = sizeof(*iph4);
343 pkt->packet_type |= RTE_PTYPE_L3_IPV4;
344 } else if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6)) {
346 size_t l3len, ext_len;
349 /* get protocol type */
350 iph6 = (const struct rte_ipv6_hdr *)rte_pktmbuf_adj(pkt,
352 adjust_ipv6_pktlen(pkt, iph6, 0);
354 next_proto = iph6->proto;
356 /* determine l3 header size up to ESP extension */
357 l3len = sizeof(struct ip6_hdr);
358 p = rte_pktmbuf_mtod(pkt, uint8_t *);
359 while (next_proto != IPPROTO_ESP && l3len < pkt->data_len &&
360 (next_proto = rte_ipv6_get_next_ext(p + l3len,
361 next_proto, &ext_len)) >= 0)
364 /* drop packet when IPv6 header exceeds first segment length */
365 if (unlikely(l3len > pkt->data_len)) {
366 rte_pktmbuf_free(pkt);
370 if (next_proto == IPPROTO_ESP)
371 t->ipsec.pkts[(t->ipsec.num)++] = pkt;
373 t->ip6.data[t->ip6.num] = &iph6->proto;
374 t->ip6.pkts[(t->ip6.num)++] = pkt;
378 pkt->packet_type |= RTE_PTYPE_L3_IPV6;
380 /* Unknown/Unsupported type, drop the packet */
381 RTE_LOG(ERR, IPSEC, "Unsupported packet type 0x%x\n",
382 rte_be_to_cpu_16(eth->ether_type));
383 rte_pktmbuf_free(pkt);
387 /* Check if the packet has been processed inline. For inline protocol
388 * processed packets, the metadata in the mbuf can be used to identify
389 * the security processing done on the packet. The metadata will be
390 * used to retrieve the application registered userdata associated
391 * with the security session.
394 if (pkt->ol_flags & PKT_RX_SEC_OFFLOAD) {
396 struct ipsec_mbuf_metadata *priv;
397 struct rte_security_ctx *ctx = (struct rte_security_ctx *)
398 rte_eth_dev_get_sec_ctx(
401 /* Retrieve the userdata registered. Here, the userdata
402 * registered is the SA pointer.
405 sa = (struct ipsec_sa *)
406 rte_security_get_userdata(ctx, pkt->udata64);
409 /* userdata could not be retrieved */
413 /* Save SA as priv member in mbuf. This will be used in the
414 * IPsec selector(SP-SA) check.
417 priv = get_priv(pkt);
423 prepare_traffic(struct rte_mbuf **pkts, struct ipsec_traffic *t,
432 for (i = 0; i < (nb_pkts - PREFETCH_OFFSET); i++) {
433 rte_prefetch0(rte_pktmbuf_mtod(pkts[i + PREFETCH_OFFSET],
435 prepare_one_packet(pkts[i], t);
437 /* Process left packets */
438 for (; i < nb_pkts; i++)
439 prepare_one_packet(pkts[i], t);
443 prepare_tx_pkt(struct rte_mbuf *pkt, uint16_t port,
444 const struct lcore_conf *qconf)
447 struct rte_ether_hdr *ethhdr;
449 ip = rte_pktmbuf_mtod(pkt, struct ip *);
451 ethhdr = (struct rte_ether_hdr *)
452 rte_pktmbuf_prepend(pkt, RTE_ETHER_HDR_LEN);
454 if (ip->ip_v == IPVERSION) {
455 pkt->ol_flags |= qconf->outbound.ipv4_offloads;
456 pkt->l3_len = sizeof(struct ip);
457 pkt->l2_len = RTE_ETHER_HDR_LEN;
461 /* calculate IPv4 cksum in SW */
462 if ((pkt->ol_flags & PKT_TX_IP_CKSUM) == 0)
463 ip->ip_sum = rte_ipv4_cksum((struct rte_ipv4_hdr *)ip);
465 ethhdr->ether_type = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4);
467 pkt->ol_flags |= qconf->outbound.ipv6_offloads;
468 pkt->l3_len = sizeof(struct ip6_hdr);
469 pkt->l2_len = RTE_ETHER_HDR_LEN;
471 ethhdr->ether_type = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6);
474 memcpy(ðhdr->s_addr, ðaddr_tbl[port].src,
475 sizeof(struct rte_ether_addr));
476 memcpy(ðhdr->d_addr, ðaddr_tbl[port].dst,
477 sizeof(struct rte_ether_addr));
481 prepare_tx_burst(struct rte_mbuf *pkts[], uint16_t nb_pkts, uint16_t port,
482 const struct lcore_conf *qconf)
485 const int32_t prefetch_offset = 2;
487 for (i = 0; i < (nb_pkts - prefetch_offset); i++) {
488 rte_mbuf_prefetch_part2(pkts[i + prefetch_offset]);
489 prepare_tx_pkt(pkts[i], port, qconf);
491 /* Process left packets */
492 for (; i < nb_pkts; i++)
493 prepare_tx_pkt(pkts[i], port, qconf);
496 /* Send burst of packets on an output interface */
497 static inline int32_t
498 send_burst(struct lcore_conf *qconf, uint16_t n, uint16_t port)
500 struct rte_mbuf **m_table;
504 queueid = qconf->tx_queue_id[port];
505 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
507 prepare_tx_burst(m_table, n, port, qconf);
509 ret = rte_eth_tx_burst(port, queueid, m_table, n);
510 if (unlikely(ret < n)) {
512 rte_pktmbuf_free(m_table[ret]);
520 * Helper function to fragment and queue for TX one packet.
522 static inline uint32_t
523 send_fragment_packet(struct lcore_conf *qconf, struct rte_mbuf *m,
524 uint16_t port, uint8_t proto)
530 tbl = qconf->tx_mbufs + port;
533 /* free space for new fragments */
534 if (len + RTE_LIBRTE_IP_FRAG_MAX_FRAG >= RTE_DIM(tbl->m_table)) {
535 send_burst(qconf, len, port);
539 n = RTE_DIM(tbl->m_table) - len;
541 if (proto == IPPROTO_IP)
542 rc = rte_ipv4_fragment_packet(m, tbl->m_table + len,
543 n, mtu_size, qconf->frag.pool_dir,
544 qconf->frag.pool_indir);
546 rc = rte_ipv6_fragment_packet(m, tbl->m_table + len,
547 n, mtu_size, qconf->frag.pool_dir,
548 qconf->frag.pool_indir);
554 "%s: failed to fragment packet with size %u, "
556 __func__, m->pkt_len, rte_errno);
562 /* Enqueue a single packet, and send burst if queue is filled */
563 static inline int32_t
564 send_single_packet(struct rte_mbuf *m, uint16_t port, uint8_t proto)
568 struct lcore_conf *qconf;
570 lcore_id = rte_lcore_id();
572 qconf = &lcore_conf[lcore_id];
573 len = qconf->tx_mbufs[port].len;
575 if (m->pkt_len <= mtu_size) {
576 qconf->tx_mbufs[port].m_table[len] = m;
579 /* need to fragment the packet */
580 } else if (frag_tbl_sz > 0)
581 len = send_fragment_packet(qconf, m, port, proto);
585 /* enough pkts to be sent */
586 if (unlikely(len == MAX_PKT_BURST)) {
587 send_burst(qconf, MAX_PKT_BURST, port);
591 qconf->tx_mbufs[port].len = len;
596 inbound_sp_sa(struct sp_ctx *sp, struct sa_ctx *sa, struct traffic_type *ip,
600 uint32_t i, j, res, sa_idx;
602 if (ip->num == 0 || sp == NULL)
605 rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
606 ip->num, DEFAULT_MAX_CATEGORIES);
609 for (i = 0; i < ip->num; i++) {
616 if (res == DISCARD) {
621 /* Only check SPI match for processed IPSec packets */
622 if (i < lim && ((m->ol_flags & PKT_RX_SEC_OFFLOAD) == 0)) {
628 if (!inbound_sa_check(sa, m, sa_idx)) {
638 split46_traffic(struct ipsec_traffic *trf, struct rte_mbuf *mb[], uint32_t num)
647 for (i = 0; i < num; i++) {
650 ip = rte_pktmbuf_mtod(m, struct ip *);
652 if (ip->ip_v == IPVERSION) {
653 trf->ip4.pkts[n4] = m;
654 trf->ip4.data[n4] = rte_pktmbuf_mtod_offset(m,
655 uint8_t *, offsetof(struct ip, ip_p));
657 } else if (ip->ip_v == IP6_VERSION) {
658 trf->ip6.pkts[n6] = m;
659 trf->ip6.data[n6] = rte_pktmbuf_mtod_offset(m,
661 offsetof(struct ip6_hdr, ip6_nxt));
673 process_pkts_inbound(struct ipsec_ctx *ipsec_ctx,
674 struct ipsec_traffic *traffic)
676 uint16_t nb_pkts_in, n_ip4, n_ip6;
678 n_ip4 = traffic->ip4.num;
679 n_ip6 = traffic->ip6.num;
681 if (app_sa_prm.enable == 0) {
682 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
683 traffic->ipsec.num, MAX_PKT_BURST);
684 split46_traffic(traffic, traffic->ipsec.pkts, nb_pkts_in);
686 inbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.pkts,
687 traffic->ipsec.saptr, traffic->ipsec.num);
688 ipsec_process(ipsec_ctx, traffic);
691 inbound_sp_sa(ipsec_ctx->sp4_ctx, ipsec_ctx->sa_ctx, &traffic->ip4,
694 inbound_sp_sa(ipsec_ctx->sp6_ctx, ipsec_ctx->sa_ctx, &traffic->ip6,
699 outbound_sp(struct sp_ctx *sp, struct traffic_type *ip,
700 struct traffic_type *ipsec)
703 uint32_t i, j, sa_idx;
705 if (ip->num == 0 || sp == NULL)
708 rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
709 ip->num, DEFAULT_MAX_CATEGORIES);
712 for (i = 0; i < ip->num; i++) {
714 sa_idx = ip->res[i] - 1;
715 if (ip->res[i] == DISCARD)
717 else if (ip->res[i] == BYPASS)
720 ipsec->res[ipsec->num] = sa_idx;
721 ipsec->pkts[ipsec->num++] = m;
728 process_pkts_outbound(struct ipsec_ctx *ipsec_ctx,
729 struct ipsec_traffic *traffic)
732 uint16_t idx, nb_pkts_out, i;
734 /* Drop any IPsec traffic from protected ports */
735 for (i = 0; i < traffic->ipsec.num; i++)
736 rte_pktmbuf_free(traffic->ipsec.pkts[i]);
738 traffic->ipsec.num = 0;
740 outbound_sp(ipsec_ctx->sp4_ctx, &traffic->ip4, &traffic->ipsec);
742 outbound_sp(ipsec_ctx->sp6_ctx, &traffic->ip6, &traffic->ipsec);
744 if (app_sa_prm.enable == 0) {
746 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
747 traffic->ipsec.res, traffic->ipsec.num,
750 for (i = 0; i < nb_pkts_out; i++) {
751 m = traffic->ipsec.pkts[i];
752 struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
753 if (ip->ip_v == IPVERSION) {
754 idx = traffic->ip4.num++;
755 traffic->ip4.pkts[idx] = m;
757 idx = traffic->ip6.num++;
758 traffic->ip6.pkts[idx] = m;
762 outbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.res,
763 traffic->ipsec.saptr, traffic->ipsec.num);
764 ipsec_process(ipsec_ctx, traffic);
769 process_pkts_inbound_nosp(struct ipsec_ctx *ipsec_ctx,
770 struct ipsec_traffic *traffic)
773 uint32_t nb_pkts_in, i, idx;
775 /* Drop any IPv4 traffic from unprotected ports */
776 for (i = 0; i < traffic->ip4.num; i++)
777 rte_pktmbuf_free(traffic->ip4.pkts[i]);
779 traffic->ip4.num = 0;
781 /* Drop any IPv6 traffic from unprotected ports */
782 for (i = 0; i < traffic->ip6.num; i++)
783 rte_pktmbuf_free(traffic->ip6.pkts[i]);
785 traffic->ip6.num = 0;
787 if (app_sa_prm.enable == 0) {
789 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
790 traffic->ipsec.num, MAX_PKT_BURST);
792 for (i = 0; i < nb_pkts_in; i++) {
793 m = traffic->ipsec.pkts[i];
794 struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
795 if (ip->ip_v == IPVERSION) {
796 idx = traffic->ip4.num++;
797 traffic->ip4.pkts[idx] = m;
799 idx = traffic->ip6.num++;
800 traffic->ip6.pkts[idx] = m;
804 inbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.pkts,
805 traffic->ipsec.saptr, traffic->ipsec.num);
806 ipsec_process(ipsec_ctx, traffic);
811 process_pkts_outbound_nosp(struct ipsec_ctx *ipsec_ctx,
812 struct ipsec_traffic *traffic)
815 uint32_t nb_pkts_out, i, n;
818 /* Drop any IPsec traffic from protected ports */
819 for (i = 0; i < traffic->ipsec.num; i++)
820 rte_pktmbuf_free(traffic->ipsec.pkts[i]);
824 for (i = 0; i < traffic->ip4.num; i++) {
825 traffic->ipsec.pkts[n] = traffic->ip4.pkts[i];
826 traffic->ipsec.res[n++] = single_sa_idx;
829 for (i = 0; i < traffic->ip6.num; i++) {
830 traffic->ipsec.pkts[n] = traffic->ip6.pkts[i];
831 traffic->ipsec.res[n++] = single_sa_idx;
834 traffic->ip4.num = 0;
835 traffic->ip6.num = 0;
836 traffic->ipsec.num = n;
838 if (app_sa_prm.enable == 0) {
840 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
841 traffic->ipsec.res, traffic->ipsec.num,
844 /* They all sue the same SA (ip4 or ip6 tunnel) */
845 m = traffic->ipsec.pkts[0];
846 ip = rte_pktmbuf_mtod(m, struct ip *);
847 if (ip->ip_v == IPVERSION) {
848 traffic->ip4.num = nb_pkts_out;
849 for (i = 0; i < nb_pkts_out; i++)
850 traffic->ip4.pkts[i] = traffic->ipsec.pkts[i];
852 traffic->ip6.num = nb_pkts_out;
853 for (i = 0; i < nb_pkts_out; i++)
854 traffic->ip6.pkts[i] = traffic->ipsec.pkts[i];
857 outbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.res,
858 traffic->ipsec.saptr, traffic->ipsec.num);
859 ipsec_process(ipsec_ctx, traffic);
863 static inline int32_t
864 get_hop_for_offload_pkt(struct rte_mbuf *pkt, int is_ipv6)
866 struct ipsec_mbuf_metadata *priv;
869 priv = get_priv(pkt);
872 if (unlikely(sa == NULL)) {
873 RTE_LOG(ERR, IPSEC, "SA not saved in private data\n");
881 return (sa->portid | RTE_LPM_LOOKUP_SUCCESS);
892 route4_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
894 uint32_t hop[MAX_PKT_BURST * 2];
895 uint32_t dst_ip[MAX_PKT_BURST * 2];
898 uint16_t lpm_pkts = 0;
903 /* Need to do an LPM lookup for non-inline packets. Inline packets will
904 * have port ID in the SA
907 for (i = 0; i < nb_pkts; i++) {
908 if (!(pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD)) {
909 /* Security offload not enabled. So an LPM lookup is
910 * required to get the hop
912 offset = offsetof(struct ip, ip_dst);
913 dst_ip[lpm_pkts] = *rte_pktmbuf_mtod_offset(pkts[i],
915 dst_ip[lpm_pkts] = rte_be_to_cpu_32(dst_ip[lpm_pkts]);
920 rte_lpm_lookup_bulk((struct rte_lpm *)rt_ctx, dst_ip, hop, lpm_pkts);
924 for (i = 0; i < nb_pkts; i++) {
925 if (pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD) {
926 /* Read hop from the SA */
927 pkt_hop = get_hop_for_offload_pkt(pkts[i], 0);
929 /* Need to use hop returned by lookup */
930 pkt_hop = hop[lpm_pkts++];
933 if ((pkt_hop & RTE_LPM_LOOKUP_SUCCESS) == 0) {
934 rte_pktmbuf_free(pkts[i]);
937 send_single_packet(pkts[i], pkt_hop & 0xff, IPPROTO_IP);
942 route6_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
944 int32_t hop[MAX_PKT_BURST * 2];
945 uint8_t dst_ip[MAX_PKT_BURST * 2][16];
949 uint16_t lpm_pkts = 0;
954 /* Need to do an LPM lookup for non-inline packets. Inline packets will
955 * have port ID in the SA
958 for (i = 0; i < nb_pkts; i++) {
959 if (!(pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD)) {
960 /* Security offload not enabled. So an LPM lookup is
961 * required to get the hop
963 offset = offsetof(struct ip6_hdr, ip6_dst);
964 ip6_dst = rte_pktmbuf_mtod_offset(pkts[i], uint8_t *,
966 memcpy(&dst_ip[lpm_pkts][0], ip6_dst, 16);
971 rte_lpm6_lookup_bulk_func((struct rte_lpm6 *)rt_ctx, dst_ip, hop,
976 for (i = 0; i < nb_pkts; i++) {
977 if (pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD) {
978 /* Read hop from the SA */
979 pkt_hop = get_hop_for_offload_pkt(pkts[i], 1);
981 /* Need to use hop returned by lookup */
982 pkt_hop = hop[lpm_pkts++];
986 rte_pktmbuf_free(pkts[i]);
989 send_single_packet(pkts[i], pkt_hop & 0xff, IPPROTO_IPV6);
994 process_pkts(struct lcore_conf *qconf, struct rte_mbuf **pkts,
995 uint8_t nb_pkts, uint16_t portid)
997 struct ipsec_traffic traffic;
999 prepare_traffic(pkts, &traffic, nb_pkts);
1001 if (unlikely(single_sa)) {
1002 if (is_unprotected_port(portid))
1003 process_pkts_inbound_nosp(&qconf->inbound, &traffic);
1005 process_pkts_outbound_nosp(&qconf->outbound, &traffic);
1007 if (is_unprotected_port(portid))
1008 process_pkts_inbound(&qconf->inbound, &traffic);
1010 process_pkts_outbound(&qconf->outbound, &traffic);
1013 route4_pkts(qconf->rt4_ctx, traffic.ip4.pkts, traffic.ip4.num);
1014 route6_pkts(qconf->rt6_ctx, traffic.ip6.pkts, traffic.ip6.num);
1018 drain_tx_buffers(struct lcore_conf *qconf)
1023 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
1024 buf = &qconf->tx_mbufs[portid];
1027 send_burst(qconf, buf->len, portid);
1033 drain_crypto_buffers(struct lcore_conf *qconf)
1036 struct ipsec_ctx *ctx;
1038 /* drain inbound buffers*/
1039 ctx = &qconf->inbound;
1040 for (i = 0; i != ctx->nb_qps; i++) {
1041 if (ctx->tbl[i].len != 0)
1042 enqueue_cop_burst(ctx->tbl + i);
1045 /* drain outbound buffers*/
1046 ctx = &qconf->outbound;
1047 for (i = 0; i != ctx->nb_qps; i++) {
1048 if (ctx->tbl[i].len != 0)
1049 enqueue_cop_burst(ctx->tbl + i);
1054 drain_inbound_crypto_queues(const struct lcore_conf *qconf,
1055 struct ipsec_ctx *ctx)
1058 struct ipsec_traffic trf;
1060 if (app_sa_prm.enable == 0) {
1062 /* dequeue packets from crypto-queue */
1063 n = ipsec_inbound_cqp_dequeue(ctx, trf.ipsec.pkts,
1064 RTE_DIM(trf.ipsec.pkts));
1069 /* split traffic by ipv4-ipv6 */
1070 split46_traffic(&trf, trf.ipsec.pkts, n);
1072 ipsec_cqp_process(ctx, &trf);
1074 /* process ipv4 packets */
1075 if (trf.ip4.num != 0) {
1076 inbound_sp_sa(ctx->sp4_ctx, ctx->sa_ctx, &trf.ip4, 0);
1077 route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
1080 /* process ipv6 packets */
1081 if (trf.ip6.num != 0) {
1082 inbound_sp_sa(ctx->sp6_ctx, ctx->sa_ctx, &trf.ip6, 0);
1083 route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
1088 drain_outbound_crypto_queues(const struct lcore_conf *qconf,
1089 struct ipsec_ctx *ctx)
1092 struct ipsec_traffic trf;
1094 if (app_sa_prm.enable == 0) {
1096 /* dequeue packets from crypto-queue */
1097 n = ipsec_outbound_cqp_dequeue(ctx, trf.ipsec.pkts,
1098 RTE_DIM(trf.ipsec.pkts));
1103 /* split traffic by ipv4-ipv6 */
1104 split46_traffic(&trf, trf.ipsec.pkts, n);
1106 ipsec_cqp_process(ctx, &trf);
1108 /* process ipv4 packets */
1109 if (trf.ip4.num != 0)
1110 route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
1112 /* process ipv6 packets */
1113 if (trf.ip6.num != 0)
1114 route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
1117 /* main processing loop */
1119 ipsec_poll_mode_worker(void)
1121 struct rte_mbuf *pkts[MAX_PKT_BURST];
1123 uint64_t prev_tsc, diff_tsc, cur_tsc;
1127 struct lcore_conf *qconf;
1128 int32_t rc, socket_id;
1129 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
1130 / US_PER_S * BURST_TX_DRAIN_US;
1131 struct lcore_rx_queue *rxql;
1134 lcore_id = rte_lcore_id();
1135 qconf = &lcore_conf[lcore_id];
1136 rxql = qconf->rx_queue_list;
1137 socket_id = rte_lcore_to_socket_id(lcore_id);
1139 qconf->rt4_ctx = socket_ctx[socket_id].rt_ip4;
1140 qconf->rt6_ctx = socket_ctx[socket_id].rt_ip6;
1141 qconf->inbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_in;
1142 qconf->inbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_in;
1143 qconf->inbound.sa_ctx = socket_ctx[socket_id].sa_in;
1144 qconf->inbound.cdev_map = cdev_map_in;
1145 qconf->inbound.session_pool = socket_ctx[socket_id].session_pool;
1146 qconf->inbound.session_priv_pool =
1147 socket_ctx[socket_id].session_priv_pool;
1148 qconf->outbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_out;
1149 qconf->outbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_out;
1150 qconf->outbound.sa_ctx = socket_ctx[socket_id].sa_out;
1151 qconf->outbound.cdev_map = cdev_map_out;
1152 qconf->outbound.session_pool = socket_ctx[socket_id].session_pool;
1153 qconf->outbound.session_priv_pool =
1154 socket_ctx[socket_id].session_priv_pool;
1155 qconf->frag.pool_dir = socket_ctx[socket_id].mbuf_pool;
1156 qconf->frag.pool_indir = socket_ctx[socket_id].mbuf_pool_indir;
1158 rc = ipsec_sad_lcore_cache_init(app_sa_prm.cache_sz);
1161 "SAD cache init on lcore %u, failed with code: %d\n",
1166 if (qconf->nb_rx_queue == 0) {
1167 RTE_LOG(DEBUG, IPSEC, "lcore %u has nothing to do\n",
1172 RTE_LOG(INFO, IPSEC, "entering main loop on lcore %u\n", lcore_id);
1174 for (i = 0; i < qconf->nb_rx_queue; i++) {
1175 portid = rxql[i].port_id;
1176 queueid = rxql[i].queue_id;
1177 RTE_LOG(INFO, IPSEC,
1178 " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
1179 lcore_id, portid, queueid);
1182 while (!force_quit) {
1183 cur_tsc = rte_rdtsc();
1185 /* TX queue buffer drain */
1186 diff_tsc = cur_tsc - prev_tsc;
1188 if (unlikely(diff_tsc > drain_tsc)) {
1189 drain_tx_buffers(qconf);
1190 drain_crypto_buffers(qconf);
1194 for (i = 0; i < qconf->nb_rx_queue; ++i) {
1196 /* Read packets from RX queues */
1197 portid = rxql[i].port_id;
1198 queueid = rxql[i].queue_id;
1199 nb_rx = rte_eth_rx_burst(portid, queueid,
1200 pkts, MAX_PKT_BURST);
1203 process_pkts(qconf, pkts, nb_rx, portid);
1205 /* dequeue and process completed crypto-ops */
1206 if (is_unprotected_port(portid))
1207 drain_inbound_crypto_queues(qconf,
1210 drain_outbound_crypto_queues(qconf,
1217 check_poll_mode_params(struct eh_conf *eh_conf)
1227 if (eh_conf->mode != EH_PKT_TRANSFER_MODE_POLL)
1230 if (lcore_params == NULL) {
1231 printf("Error: No port/queue/core mappings\n");
1235 for (i = 0; i < nb_lcore_params; ++i) {
1236 lcore = lcore_params[i].lcore_id;
1237 if (!rte_lcore_is_enabled(lcore)) {
1238 printf("error: lcore %hhu is not enabled in "
1239 "lcore mask\n", lcore);
1242 socket_id = rte_lcore_to_socket_id(lcore);
1243 if (socket_id != 0 && numa_on == 0) {
1244 printf("warning: lcore %hhu is on socket %d "
1248 portid = lcore_params[i].port_id;
1249 if ((enabled_port_mask & (1 << portid)) == 0) {
1250 printf("port %u is not enabled in port mask\n", portid);
1253 if (!rte_eth_dev_is_valid_port(portid)) {
1254 printf("port %u is not present on the board\n", portid);
1262 get_port_nb_rx_queues(const uint16_t port)
1267 for (i = 0; i < nb_lcore_params; ++i) {
1268 if (lcore_params[i].port_id == port &&
1269 lcore_params[i].queue_id > queue)
1270 queue = lcore_params[i].queue_id;
1272 return (uint8_t)(++queue);
1276 init_lcore_rx_queues(void)
1278 uint16_t i, nb_rx_queue;
1281 for (i = 0; i < nb_lcore_params; ++i) {
1282 lcore = lcore_params[i].lcore_id;
1283 nb_rx_queue = lcore_conf[lcore].nb_rx_queue;
1284 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1285 printf("error: too many queues (%u) for lcore: %u\n",
1286 nb_rx_queue + 1, lcore);
1289 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1290 lcore_params[i].port_id;
1291 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1292 lcore_params[i].queue_id;
1293 lcore_conf[lcore].nb_rx_queue++;
1300 print_usage(const char *prgname)
1302 fprintf(stderr, "%s [EAL options] --"
1308 " [-w REPLAY_WINDOW_SIZE]"
1313 " --config (port,queue,lcore)[,(port,queue,lcore)]"
1314 " [--single-sa SAIDX]"
1315 " [--cryptodev_mask MASK]"
1316 " [--transfer-mode MODE]"
1317 " [--event-schedule-type TYPE]"
1318 " [--" CMD_LINE_OPT_RX_OFFLOAD " RX_OFFLOAD_MASK]"
1319 " [--" CMD_LINE_OPT_TX_OFFLOAD " TX_OFFLOAD_MASK]"
1320 " [--" CMD_LINE_OPT_REASSEMBLE " REASSEMBLE_TABLE_SIZE]"
1321 " [--" CMD_LINE_OPT_MTU " MTU]"
1323 " -p PORTMASK: Hexadecimal bitmask of ports to configure\n"
1324 " -P : Enable promiscuous mode\n"
1325 " -u PORTMASK: Hexadecimal bitmask of unprotected ports\n"
1326 " -j FRAMESIZE: Data buffer size, minimum (and default)\n"
1327 " value: RTE_MBUF_DEFAULT_BUF_SIZE\n"
1328 " -l enables code-path that uses librte_ipsec\n"
1329 " -w REPLAY_WINDOW_SIZE specifies IPsec SQN replay window\n"
1330 " size for each SA\n"
1332 " -a enables SA SQN atomic behaviour\n"
1333 " -c specifies inbound SAD cache size,\n"
1334 " zero value disables the cache (default value: 128)\n"
1335 " -f CONFIG_FILE: Configuration file\n"
1336 " --config (port,queue,lcore): Rx queue configuration. In poll\n"
1337 " mode determines which queues from\n"
1338 " which ports are mapped to which cores.\n"
1339 " In event mode this option is not used\n"
1340 " as packets are dynamically scheduled\n"
1341 " to cores by HW.\n"
1342 " --single-sa SAIDX: In poll mode use single SA index for\n"
1343 " outbound traffic, bypassing the SP\n"
1344 " In event mode selects driver submode,\n"
1345 " SA index value is ignored\n"
1346 " --cryptodev_mask MASK: Hexadecimal bitmask of the crypto\n"
1347 " devices to configure\n"
1348 " --transfer-mode MODE\n"
1349 " \"poll\" : Packet transfer via polling (default)\n"
1350 " \"event\" : Packet transfer via event device\n"
1351 " --event-schedule-type TYPE queue schedule type, used only when\n"
1352 " transfer mode is set to event\n"
1353 " \"ordered\" : Ordered (default)\n"
1354 " \"atomic\" : Atomic\n"
1355 " \"parallel\" : Parallel\n"
1356 " --" CMD_LINE_OPT_RX_OFFLOAD
1357 ": bitmask of the RX HW offload capabilities to enable/use\n"
1358 " (DEV_RX_OFFLOAD_*)\n"
1359 " --" CMD_LINE_OPT_TX_OFFLOAD
1360 ": bitmask of the TX HW offload capabilities to enable/use\n"
1361 " (DEV_TX_OFFLOAD_*)\n"
1362 " --" CMD_LINE_OPT_REASSEMBLE " NUM"
1363 ": max number of entries in reassemble(fragment) table\n"
1364 " (zero (default value) disables reassembly)\n"
1365 " --" CMD_LINE_OPT_MTU " MTU"
1366 ": MTU value on all ports (default value: 1500)\n"
1367 " outgoing packets with bigger size will be fragmented\n"
1368 " incoming packets with bigger size will be discarded\n"
1369 " --" CMD_LINE_OPT_FRAG_TTL " FRAG_TTL_NS"
1370 ": fragments lifetime in nanoseconds, default\n"
1371 " and maximum value is 10.000.000.000 ns (10 s)\n"
1377 parse_mask(const char *str, uint64_t *val)
1383 t = strtoul(str, &end, 0);
1384 if (errno != 0 || end[0] != 0)
1392 parse_portmask(const char *portmask)
1397 /* parse hexadecimal string */
1398 pm = strtoul(portmask, &end, 16);
1399 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1402 if ((pm == 0) && errno)
1409 parse_decimal(const char *str)
1414 num = strtoull(str, &end, 10);
1415 if ((str[0] == '\0') || (end == NULL) || (*end != '\0')
1423 parse_config(const char *q_arg)
1426 const char *p, *p0 = q_arg;
1434 unsigned long int_fld[_NUM_FLD];
1435 char *str_fld[_NUM_FLD];
1439 nb_lcore_params = 0;
1441 while ((p = strchr(p0, '(')) != NULL) {
1443 p0 = strchr(p, ')');
1448 if (size >= sizeof(s))
1451 snprintf(s, sizeof(s), "%.*s", size, p);
1452 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1455 for (i = 0; i < _NUM_FLD; i++) {
1457 int_fld[i] = strtoul(str_fld[i], &end, 0);
1458 if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
1461 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1462 printf("exceeded max number of lcore params: %hu\n",
1466 lcore_params_array[nb_lcore_params].port_id =
1467 (uint8_t)int_fld[FLD_PORT];
1468 lcore_params_array[nb_lcore_params].queue_id =
1469 (uint8_t)int_fld[FLD_QUEUE];
1470 lcore_params_array[nb_lcore_params].lcore_id =
1471 (uint8_t)int_fld[FLD_LCORE];
1474 lcore_params = lcore_params_array;
1479 print_app_sa_prm(const struct app_sa_prm *prm)
1481 printf("librte_ipsec usage: %s\n",
1482 (prm->enable == 0) ? "disabled" : "enabled");
1484 printf("replay window size: %u\n", prm->window_size);
1485 printf("ESN: %s\n", (prm->enable_esn == 0) ? "disabled" : "enabled");
1486 printf("SA flags: %#" PRIx64 "\n", prm->flags);
1487 printf("Frag TTL: %" PRIu64 " ns\n", frag_ttl_ns);
1491 parse_transfer_mode(struct eh_conf *conf, const char *optarg)
1493 if (!strcmp(CMD_LINE_ARG_POLL, optarg))
1494 conf->mode = EH_PKT_TRANSFER_MODE_POLL;
1495 else if (!strcmp(CMD_LINE_ARG_EVENT, optarg))
1496 conf->mode = EH_PKT_TRANSFER_MODE_EVENT;
1498 printf("Unsupported packet transfer mode\n");
1506 parse_schedule_type(struct eh_conf *conf, const char *optarg)
1508 struct eventmode_conf *em_conf = NULL;
1510 /* Get eventmode conf */
1511 em_conf = conf->mode_params;
1513 if (!strcmp(CMD_LINE_ARG_ORDERED, optarg))
1514 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_ORDERED;
1515 else if (!strcmp(CMD_LINE_ARG_ATOMIC, optarg))
1516 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_ATOMIC;
1517 else if (!strcmp(CMD_LINE_ARG_PARALLEL, optarg))
1518 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_PARALLEL;
1520 printf("Unsupported queue schedule type\n");
1528 parse_args(int32_t argc, char **argv, struct eh_conf *eh_conf)
1533 int32_t option_index;
1534 char *prgname = argv[0];
1535 int32_t f_present = 0;
1539 while ((opt = getopt_long(argc, argvopt, "aelp:Pu:f:j:w:c:",
1540 lgopts, &option_index)) != EOF) {
1544 enabled_port_mask = parse_portmask(optarg);
1545 if (enabled_port_mask == 0) {
1546 printf("invalid portmask\n");
1547 print_usage(prgname);
1552 printf("Promiscuous mode selected\n");
1556 unprotected_port_mask = parse_portmask(optarg);
1557 if (unprotected_port_mask == 0) {
1558 printf("invalid unprotected portmask\n");
1559 print_usage(prgname);
1564 if (f_present == 1) {
1565 printf("\"-f\" option present more than "
1567 print_usage(prgname);
1574 ret = parse_decimal(optarg);
1575 if (ret < RTE_MBUF_DEFAULT_BUF_SIZE ||
1577 printf("Invalid frame buffer size value: %s\n",
1579 print_usage(prgname);
1582 frame_buf_size = ret;
1583 printf("Custom frame buffer size %u\n", frame_buf_size);
1586 app_sa_prm.enable = 1;
1589 app_sa_prm.window_size = parse_decimal(optarg);
1592 app_sa_prm.enable_esn = 1;
1595 app_sa_prm.enable = 1;
1596 app_sa_prm.flags |= RTE_IPSEC_SAFLAG_SQN_ATOM;
1599 ret = parse_decimal(optarg);
1601 printf("Invalid SA cache size: %s\n", optarg);
1602 print_usage(prgname);
1605 app_sa_prm.cache_sz = ret;
1607 case CMD_LINE_OPT_CONFIG_NUM:
1608 ret = parse_config(optarg);
1610 printf("Invalid config\n");
1611 print_usage(prgname);
1615 case CMD_LINE_OPT_SINGLE_SA_NUM:
1616 ret = parse_decimal(optarg);
1617 if (ret == -1 || ret > UINT32_MAX) {
1618 printf("Invalid argument[sa_idx]\n");
1619 print_usage(prgname);
1625 single_sa_idx = ret;
1626 eh_conf->ipsec_mode = EH_IPSEC_MODE_TYPE_DRIVER;
1627 printf("Configured with single SA index %u\n",
1630 case CMD_LINE_OPT_CRYPTODEV_MASK_NUM:
1631 ret = parse_portmask(optarg);
1633 printf("Invalid argument[portmask]\n");
1634 print_usage(prgname);
1639 enabled_cryptodev_mask = ret;
1642 case CMD_LINE_OPT_TRANSFER_MODE_NUM:
1643 ret = parse_transfer_mode(eh_conf, optarg);
1645 printf("Invalid packet transfer mode\n");
1646 print_usage(prgname);
1651 case CMD_LINE_OPT_SCHEDULE_TYPE_NUM:
1652 ret = parse_schedule_type(eh_conf, optarg);
1654 printf("Invalid queue schedule type\n");
1655 print_usage(prgname);
1660 case CMD_LINE_OPT_RX_OFFLOAD_NUM:
1661 ret = parse_mask(optarg, &dev_rx_offload);
1663 printf("Invalid argument for \'%s\': %s\n",
1664 CMD_LINE_OPT_RX_OFFLOAD, optarg);
1665 print_usage(prgname);
1669 case CMD_LINE_OPT_TX_OFFLOAD_NUM:
1670 ret = parse_mask(optarg, &dev_tx_offload);
1672 printf("Invalid argument for \'%s\': %s\n",
1673 CMD_LINE_OPT_TX_OFFLOAD, optarg);
1674 print_usage(prgname);
1678 case CMD_LINE_OPT_REASSEMBLE_NUM:
1679 ret = parse_decimal(optarg);
1680 if (ret < 0 || ret > UINT32_MAX) {
1681 printf("Invalid argument for \'%s\': %s\n",
1682 CMD_LINE_OPT_REASSEMBLE, optarg);
1683 print_usage(prgname);
1688 case CMD_LINE_OPT_MTU_NUM:
1689 ret = parse_decimal(optarg);
1690 if (ret < 0 || ret > RTE_IPV4_MAX_PKT_LEN) {
1691 printf("Invalid argument for \'%s\': %s\n",
1692 CMD_LINE_OPT_MTU, optarg);
1693 print_usage(prgname);
1698 case CMD_LINE_OPT_FRAG_TTL_NUM:
1699 ret = parse_decimal(optarg);
1700 if (ret < 0 || ret > MAX_FRAG_TTL_NS) {
1701 printf("Invalid argument for \'%s\': %s\n",
1702 CMD_LINE_OPT_MTU, optarg);
1703 print_usage(prgname);
1709 print_usage(prgname);
1714 if (f_present == 0) {
1715 printf("Mandatory option \"-f\" not present\n");
1719 /* check do we need to enable multi-seg support */
1720 if (multi_seg_required()) {
1721 /* legacy mode doesn't support multi-seg */
1722 app_sa_prm.enable = 1;
1723 printf("frame buf size: %u, mtu: %u, "
1724 "number of reassemble entries: %u\n"
1725 "multi-segment support is required\n",
1726 frame_buf_size, mtu_size, frag_tbl_sz);
1729 print_app_sa_prm(&app_sa_prm);
1732 argv[optind-1] = prgname;
1735 optind = 1; /* reset getopt lib */
1740 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
1742 char buf[RTE_ETHER_ADDR_FMT_SIZE];
1743 rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
1744 printf("%s%s", name, buf);
1748 * Update destination ethaddr for the port.
1751 add_dst_ethaddr(uint16_t port, const struct rte_ether_addr *addr)
1753 if (port >= RTE_DIM(ethaddr_tbl))
1756 ethaddr_tbl[port].dst = ETHADDR_TO_UINT64(addr);
1760 /* Check the link status of all ports in up to 9s, and print them finally */
1762 check_all_ports_link_status(uint32_t port_mask)
1764 #define CHECK_INTERVAL 100 /* 100ms */
1765 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1767 uint8_t count, all_ports_up, print_flag = 0;
1768 struct rte_eth_link link;
1771 printf("\nChecking link status");
1773 for (count = 0; count <= MAX_CHECK_TIME; count++) {
1775 RTE_ETH_FOREACH_DEV(portid) {
1776 if ((port_mask & (1 << portid)) == 0)
1778 memset(&link, 0, sizeof(link));
1779 ret = rte_eth_link_get_nowait(portid, &link);
1782 if (print_flag == 1)
1783 printf("Port %u link get failed: %s\n",
1784 portid, rte_strerror(-ret));
1787 /* print link status if flag set */
1788 if (print_flag == 1) {
1789 if (link.link_status)
1791 "Port%d Link Up - speed %u Mbps -%s\n",
1792 portid, link.link_speed,
1793 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1794 ("full-duplex") : ("half-duplex\n"));
1796 printf("Port %d Link Down\n", portid);
1799 /* clear all_ports_up flag if any link down */
1800 if (link.link_status == ETH_LINK_DOWN) {
1805 /* after finally printing all link status, get out */
1806 if (print_flag == 1)
1809 if (all_ports_up == 0) {
1812 rte_delay_ms(CHECK_INTERVAL);
1815 /* set the print_flag if all ports up or timeout */
1816 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1824 add_mapping(struct rte_hash *map, const char *str, uint16_t cdev_id,
1825 uint16_t qp, struct lcore_params *params,
1826 struct ipsec_ctx *ipsec_ctx,
1827 const struct rte_cryptodev_capabilities *cipher,
1828 const struct rte_cryptodev_capabilities *auth,
1829 const struct rte_cryptodev_capabilities *aead)
1833 struct cdev_key key = { 0 };
1835 key.lcore_id = params->lcore_id;
1837 key.cipher_algo = cipher->sym.cipher.algo;
1839 key.auth_algo = auth->sym.auth.algo;
1841 key.aead_algo = aead->sym.aead.algo;
1843 ret = rte_hash_lookup(map, &key);
1847 for (i = 0; i < ipsec_ctx->nb_qps; i++)
1848 if (ipsec_ctx->tbl[i].id == cdev_id)
1851 if (i == ipsec_ctx->nb_qps) {
1852 if (ipsec_ctx->nb_qps == MAX_QP_PER_LCORE) {
1853 printf("Maximum number of crypto devices assigned to "
1854 "a core, increase MAX_QP_PER_LCORE value\n");
1857 ipsec_ctx->tbl[i].id = cdev_id;
1858 ipsec_ctx->tbl[i].qp = qp;
1859 ipsec_ctx->nb_qps++;
1860 printf("%s cdev mapping: lcore %u using cdev %u qp %u "
1861 "(cdev_id_qp %lu)\n", str, key.lcore_id,
1865 ret = rte_hash_add_key_data(map, &key, (void *)i);
1867 printf("Faled to insert cdev mapping for (lcore %u, "
1868 "cdev %u, qp %u), errno %d\n",
1869 key.lcore_id, ipsec_ctx->tbl[i].id,
1870 ipsec_ctx->tbl[i].qp, ret);
1878 add_cdev_mapping(struct rte_cryptodev_info *dev_info, uint16_t cdev_id,
1879 uint16_t qp, struct lcore_params *params)
1882 const struct rte_cryptodev_capabilities *i, *j;
1883 struct rte_hash *map;
1884 struct lcore_conf *qconf;
1885 struct ipsec_ctx *ipsec_ctx;
1888 qconf = &lcore_conf[params->lcore_id];
1890 if ((unprotected_port_mask & (1 << params->port_id)) == 0) {
1892 ipsec_ctx = &qconf->outbound;
1896 ipsec_ctx = &qconf->inbound;
1900 /* Required cryptodevs with operation chainning */
1901 if (!(dev_info->feature_flags &
1902 RTE_CRYPTODEV_FF_SYM_OPERATION_CHAINING))
1905 for (i = dev_info->capabilities;
1906 i->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; i++) {
1907 if (i->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1910 if (i->sym.xform_type == RTE_CRYPTO_SYM_XFORM_AEAD) {
1911 ret |= add_mapping(map, str, cdev_id, qp, params,
1912 ipsec_ctx, NULL, NULL, i);
1916 if (i->sym.xform_type != RTE_CRYPTO_SYM_XFORM_CIPHER)
1919 for (j = dev_info->capabilities;
1920 j->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; j++) {
1921 if (j->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1924 if (j->sym.xform_type != RTE_CRYPTO_SYM_XFORM_AUTH)
1927 ret |= add_mapping(map, str, cdev_id, qp, params,
1928 ipsec_ctx, i, j, NULL);
1935 /* Check if the device is enabled by cryptodev_mask */
1937 check_cryptodev_mask(uint8_t cdev_id)
1939 if (enabled_cryptodev_mask & (1 << cdev_id))
1946 cryptodevs_init(void)
1948 struct rte_cryptodev_config dev_conf;
1949 struct rte_cryptodev_qp_conf qp_conf;
1950 uint16_t idx, max_nb_qps, qp, i;
1952 struct rte_hash_parameters params = { 0 };
1954 const uint64_t mseg_flag = multi_seg_required() ?
1955 RTE_CRYPTODEV_FF_IN_PLACE_SGL : 0;
1957 params.entries = CDEV_MAP_ENTRIES;
1958 params.key_len = sizeof(struct cdev_key);
1959 params.hash_func = rte_jhash;
1960 params.hash_func_init_val = 0;
1961 params.socket_id = rte_socket_id();
1963 params.name = "cdev_map_in";
1964 cdev_map_in = rte_hash_create(¶ms);
1965 if (cdev_map_in == NULL)
1966 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1969 params.name = "cdev_map_out";
1970 cdev_map_out = rte_hash_create(¶ms);
1971 if (cdev_map_out == NULL)
1972 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1975 printf("lcore/cryptodev/qp mappings:\n");
1978 for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
1979 struct rte_cryptodev_info cdev_info;
1981 if (check_cryptodev_mask((uint8_t)cdev_id))
1984 rte_cryptodev_info_get(cdev_id, &cdev_info);
1986 if ((mseg_flag & cdev_info.feature_flags) != mseg_flag)
1987 rte_exit(EXIT_FAILURE,
1988 "Device %hd does not support \'%s\' feature\n",
1990 rte_cryptodev_get_feature_name(mseg_flag));
1992 if (nb_lcore_params > cdev_info.max_nb_queue_pairs)
1993 max_nb_qps = cdev_info.max_nb_queue_pairs;
1995 max_nb_qps = nb_lcore_params;
1999 while (qp < max_nb_qps && i < nb_lcore_params) {
2000 if (add_cdev_mapping(&cdev_info, cdev_id, qp,
2001 &lcore_params[idx]))
2004 idx = idx % nb_lcore_params;
2011 dev_conf.socket_id = rte_cryptodev_socket_id(cdev_id);
2012 dev_conf.nb_queue_pairs = qp;
2013 dev_conf.ff_disable = RTE_CRYPTODEV_FF_ASYMMETRIC_CRYPTO;
2015 uint32_t dev_max_sess = cdev_info.sym.max_nb_sessions;
2016 if (dev_max_sess != 0 && dev_max_sess < CDEV_MP_NB_OBJS)
2017 rte_exit(EXIT_FAILURE,
2018 "Device does not support at least %u "
2019 "sessions", CDEV_MP_NB_OBJS);
2021 if (rte_cryptodev_configure(cdev_id, &dev_conf))
2022 rte_panic("Failed to initialize cryptodev %u\n",
2025 qp_conf.nb_descriptors = CDEV_QUEUE_DESC;
2026 qp_conf.mp_session =
2027 socket_ctx[dev_conf.socket_id].session_pool;
2028 qp_conf.mp_session_private =
2029 socket_ctx[dev_conf.socket_id].session_priv_pool;
2030 for (qp = 0; qp < dev_conf.nb_queue_pairs; qp++)
2031 if (rte_cryptodev_queue_pair_setup(cdev_id, qp,
2032 &qp_conf, dev_conf.socket_id))
2033 rte_panic("Failed to setup queue %u for "
2034 "cdev_id %u\n", 0, cdev_id);
2036 if (rte_cryptodev_start(cdev_id))
2037 rte_panic("Failed to start cryptodev %u\n",
2047 port_init(uint16_t portid, uint64_t req_rx_offloads, uint64_t req_tx_offloads)
2049 uint32_t frame_size;
2050 struct rte_eth_dev_info dev_info;
2051 struct rte_eth_txconf *txconf;
2052 uint16_t nb_tx_queue, nb_rx_queue;
2053 uint16_t tx_queueid, rx_queueid, queue, lcore_id;
2054 int32_t ret, socket_id;
2055 struct lcore_conf *qconf;
2056 struct rte_ether_addr ethaddr;
2057 struct rte_eth_conf local_port_conf = port_conf;
2059 ret = rte_eth_dev_info_get(portid, &dev_info);
2061 rte_exit(EXIT_FAILURE,
2062 "Error during getting device (port %u) info: %s\n",
2063 portid, strerror(-ret));
2065 /* limit allowed HW offloafs, as user requested */
2066 dev_info.rx_offload_capa &= dev_rx_offload;
2067 dev_info.tx_offload_capa &= dev_tx_offload;
2069 printf("Configuring device port %u:\n", portid);
2071 ret = rte_eth_macaddr_get(portid, ðaddr);
2073 rte_exit(EXIT_FAILURE,
2074 "Error getting MAC address (port %u): %s\n",
2075 portid, rte_strerror(-ret));
2077 ethaddr_tbl[portid].src = ETHADDR_TO_UINT64(ðaddr);
2078 print_ethaddr("Address: ", ðaddr);
2081 nb_rx_queue = get_port_nb_rx_queues(portid);
2082 nb_tx_queue = nb_lcores;
2084 if (nb_rx_queue > dev_info.max_rx_queues)
2085 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
2086 "(max rx queue is %u)\n",
2087 nb_rx_queue, dev_info.max_rx_queues);
2089 if (nb_tx_queue > dev_info.max_tx_queues)
2090 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
2091 "(max tx queue is %u)\n",
2092 nb_tx_queue, dev_info.max_tx_queues);
2094 printf("Creating queues: nb_rx_queue=%d nb_tx_queue=%u...\n",
2095 nb_rx_queue, nb_tx_queue);
2097 frame_size = MTU_TO_FRAMELEN(mtu_size);
2098 if (frame_size > local_port_conf.rxmode.max_rx_pkt_len)
2099 local_port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2100 local_port_conf.rxmode.max_rx_pkt_len = frame_size;
2102 if (multi_seg_required()) {
2103 local_port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_SCATTER;
2104 local_port_conf.txmode.offloads |= DEV_TX_OFFLOAD_MULTI_SEGS;
2107 local_port_conf.rxmode.offloads |= req_rx_offloads;
2108 local_port_conf.txmode.offloads |= req_tx_offloads;
2110 /* Check that all required capabilities are supported */
2111 if ((local_port_conf.rxmode.offloads & dev_info.rx_offload_capa) !=
2112 local_port_conf.rxmode.offloads)
2113 rte_exit(EXIT_FAILURE,
2114 "Error: port %u required RX offloads: 0x%" PRIx64
2115 ", avaialbe RX offloads: 0x%" PRIx64 "\n",
2116 portid, local_port_conf.rxmode.offloads,
2117 dev_info.rx_offload_capa);
2119 if ((local_port_conf.txmode.offloads & dev_info.tx_offload_capa) !=
2120 local_port_conf.txmode.offloads)
2121 rte_exit(EXIT_FAILURE,
2122 "Error: port %u required TX offloads: 0x%" PRIx64
2123 ", avaialbe TX offloads: 0x%" PRIx64 "\n",
2124 portid, local_port_conf.txmode.offloads,
2125 dev_info.tx_offload_capa);
2127 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
2128 local_port_conf.txmode.offloads |=
2129 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
2131 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_IPV4_CKSUM)
2132 local_port_conf.txmode.offloads |= DEV_TX_OFFLOAD_IPV4_CKSUM;
2134 printf("port %u configurng rx_offloads=0x%" PRIx64
2135 ", tx_offloads=0x%" PRIx64 "\n",
2136 portid, local_port_conf.rxmode.offloads,
2137 local_port_conf.txmode.offloads);
2139 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
2140 dev_info.flow_type_rss_offloads;
2141 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
2142 port_conf.rx_adv_conf.rss_conf.rss_hf) {
2143 printf("Port %u modified RSS hash function based on hardware support,"
2144 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
2146 port_conf.rx_adv_conf.rss_conf.rss_hf,
2147 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
2150 ret = rte_eth_dev_configure(portid, nb_rx_queue, nb_tx_queue,
2153 rte_exit(EXIT_FAILURE, "Cannot configure device: "
2154 "err=%d, port=%d\n", ret, portid);
2156 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd, &nb_txd);
2158 rte_exit(EXIT_FAILURE, "Cannot adjust number of descriptors: "
2159 "err=%d, port=%d\n", ret, portid);
2161 /* init one TX queue per lcore */
2163 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2164 if (rte_lcore_is_enabled(lcore_id) == 0)
2168 socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2173 printf("Setup txq=%u,%d,%d\n", lcore_id, tx_queueid, socket_id);
2175 txconf = &dev_info.default_txconf;
2176 txconf->offloads = local_port_conf.txmode.offloads;
2178 ret = rte_eth_tx_queue_setup(portid, tx_queueid, nb_txd,
2181 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
2182 "err=%d, port=%d\n", ret, portid);
2184 qconf = &lcore_conf[lcore_id];
2185 qconf->tx_queue_id[portid] = tx_queueid;
2187 /* Pre-populate pkt offloads based on capabilities */
2188 qconf->outbound.ipv4_offloads = PKT_TX_IPV4;
2189 qconf->outbound.ipv6_offloads = PKT_TX_IPV6;
2190 if (local_port_conf.txmode.offloads & DEV_TX_OFFLOAD_IPV4_CKSUM)
2191 qconf->outbound.ipv4_offloads |= PKT_TX_IP_CKSUM;
2195 /* init RX queues */
2196 for (queue = 0; queue < qconf->nb_rx_queue; ++queue) {
2197 struct rte_eth_rxconf rxq_conf;
2199 if (portid != qconf->rx_queue_list[queue].port_id)
2202 rx_queueid = qconf->rx_queue_list[queue].queue_id;
2204 printf("Setup rxq=%d,%d,%d\n", portid, rx_queueid,
2207 rxq_conf = dev_info.default_rxconf;
2208 rxq_conf.offloads = local_port_conf.rxmode.offloads;
2209 ret = rte_eth_rx_queue_setup(portid, rx_queueid,
2210 nb_rxd, socket_id, &rxq_conf,
2211 socket_ctx[socket_id].mbuf_pool);
2213 rte_exit(EXIT_FAILURE,
2214 "rte_eth_rx_queue_setup: err=%d, "
2215 "port=%d\n", ret, portid);
2222 max_session_size(void)
2226 int16_t cdev_id, port_id, n;
2229 n = rte_cryptodev_count();
2230 for (cdev_id = 0; cdev_id != n; cdev_id++) {
2231 sz = rte_cryptodev_sym_get_private_session_size(cdev_id);
2235 * If crypto device is security capable, need to check the
2236 * size of security session as well.
2239 /* Get security context of the crypto device */
2240 sec_ctx = rte_cryptodev_get_sec_ctx(cdev_id);
2241 if (sec_ctx == NULL)
2244 /* Get size of security session */
2245 sz = rte_security_session_get_size(sec_ctx);
2250 RTE_ETH_FOREACH_DEV(port_id) {
2251 if ((enabled_port_mask & (1 << port_id)) == 0)
2254 sec_ctx = rte_eth_dev_get_sec_ctx(port_id);
2255 if (sec_ctx == NULL)
2258 sz = rte_security_session_get_size(sec_ctx);
2267 session_pool_init(struct socket_ctx *ctx, int32_t socket_id, size_t sess_sz)
2269 char mp_name[RTE_MEMPOOL_NAMESIZE];
2270 struct rte_mempool *sess_mp;
2272 snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
2273 "sess_mp_%u", socket_id);
2274 sess_mp = rte_cryptodev_sym_session_pool_create(
2275 mp_name, CDEV_MP_NB_OBJS,
2276 sess_sz, CDEV_MP_CACHE_SZ, 0,
2278 ctx->session_pool = sess_mp;
2280 if (ctx->session_pool == NULL)
2281 rte_exit(EXIT_FAILURE,
2282 "Cannot init session pool on socket %d\n", socket_id);
2284 printf("Allocated session pool on socket %d\n", socket_id);
2288 session_priv_pool_init(struct socket_ctx *ctx, int32_t socket_id,
2291 char mp_name[RTE_MEMPOOL_NAMESIZE];
2292 struct rte_mempool *sess_mp;
2294 snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
2295 "sess_mp_priv_%u", socket_id);
2296 sess_mp = rte_mempool_create(mp_name,
2300 0, NULL, NULL, NULL,
2303 ctx->session_priv_pool = sess_mp;
2305 if (ctx->session_priv_pool == NULL)
2306 rte_exit(EXIT_FAILURE,
2307 "Cannot init session priv pool on socket %d\n",
2310 printf("Allocated session priv pool on socket %d\n",
2315 pool_init(struct socket_ctx *ctx, int32_t socket_id, uint32_t nb_mbuf)
2320 snprintf(s, sizeof(s), "mbuf_pool_%d", socket_id);
2321 ctx->mbuf_pool = rte_pktmbuf_pool_create(s, nb_mbuf,
2322 MEMPOOL_CACHE_SIZE, ipsec_metadata_size(),
2323 frame_buf_size, socket_id);
2326 * if multi-segment support is enabled, then create a pool
2327 * for indirect mbufs.
2329 ms = multi_seg_required();
2331 snprintf(s, sizeof(s), "mbuf_pool_indir_%d", socket_id);
2332 ctx->mbuf_pool_indir = rte_pktmbuf_pool_create(s, nb_mbuf,
2333 MEMPOOL_CACHE_SIZE, 0, 0, socket_id);
2336 if (ctx->mbuf_pool == NULL || (ms != 0 && ctx->mbuf_pool_indir == NULL))
2337 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool on socket %d\n",
2340 printf("Allocated mbuf pool on socket %d\n", socket_id);
2344 inline_ipsec_event_esn_overflow(struct rte_security_ctx *ctx, uint64_t md)
2346 struct ipsec_sa *sa;
2348 /* For inline protocol processing, the metadata in the event will
2349 * uniquely identify the security session which raised the event.
2350 * Application would then need the userdata it had registered with the
2351 * security session to process the event.
2354 sa = (struct ipsec_sa *)rte_security_get_userdata(ctx, md);
2357 /* userdata could not be retrieved */
2361 /* Sequence number over flow. SA need to be re-established */
2367 inline_ipsec_event_callback(uint16_t port_id, enum rte_eth_event_type type,
2368 void *param, void *ret_param)
2371 struct rte_eth_event_ipsec_desc *event_desc = NULL;
2372 struct rte_security_ctx *ctx = (struct rte_security_ctx *)
2373 rte_eth_dev_get_sec_ctx(port_id);
2375 RTE_SET_USED(param);
2377 if (type != RTE_ETH_EVENT_IPSEC)
2380 event_desc = ret_param;
2381 if (event_desc == NULL) {
2382 printf("Event descriptor not set\n");
2386 md = event_desc->metadata;
2388 if (event_desc->subtype == RTE_ETH_EVENT_IPSEC_ESN_OVERFLOW)
2389 return inline_ipsec_event_esn_overflow(ctx, md);
2390 else if (event_desc->subtype >= RTE_ETH_EVENT_IPSEC_MAX) {
2391 printf("Invalid IPsec event reported\n");
2399 rx_callback(__rte_unused uint16_t port, __rte_unused uint16_t queue,
2400 struct rte_mbuf *pkt[], uint16_t nb_pkts,
2401 __rte_unused uint16_t max_pkts, void *user_param)
2405 struct lcore_conf *lc;
2406 struct rte_mbuf *mb;
2407 struct rte_ether_hdr *eth;
2413 for (i = 0; i != nb_pkts; i++) {
2416 eth = rte_pktmbuf_mtod(mb, struct rte_ether_hdr *);
2417 if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
2419 struct rte_ipv4_hdr *iph;
2421 iph = (struct rte_ipv4_hdr *)(eth + 1);
2422 if (rte_ipv4_frag_pkt_is_fragmented(iph)) {
2424 mb->l2_len = sizeof(*eth);
2425 mb->l3_len = sizeof(*iph);
2426 tm = (tm != 0) ? tm : rte_rdtsc();
2427 mb = rte_ipv4_frag_reassemble_packet(
2428 lc->frag.tbl, &lc->frag.dr,
2432 /* fix ip cksum after reassemble. */
2433 iph = rte_pktmbuf_mtod_offset(mb,
2434 struct rte_ipv4_hdr *,
2436 iph->hdr_checksum = 0;
2437 iph->hdr_checksum = rte_ipv4_cksum(iph);
2440 } else if (eth->ether_type ==
2441 rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6)) {
2443 struct rte_ipv6_hdr *iph;
2444 struct ipv6_extension_fragment *fh;
2446 iph = (struct rte_ipv6_hdr *)(eth + 1);
2447 fh = rte_ipv6_frag_get_ipv6_fragment_header(iph);
2449 mb->l2_len = sizeof(*eth);
2450 mb->l3_len = (uintptr_t)fh - (uintptr_t)iph +
2452 tm = (tm != 0) ? tm : rte_rdtsc();
2453 mb = rte_ipv6_frag_reassemble_packet(
2454 lc->frag.tbl, &lc->frag.dr,
2457 /* fix l3_len after reassemble. */
2458 mb->l3_len = mb->l3_len - sizeof(*fh);
2466 /* some fragments were encountered, drain death row */
2468 rte_ip_frag_free_death_row(&lc->frag.dr, 0);
2475 reassemble_lcore_init(struct lcore_conf *lc, uint32_t cid)
2479 uint64_t frag_cycles;
2480 const struct lcore_rx_queue *rxq;
2481 const struct rte_eth_rxtx_callback *cb;
2483 /* create fragment table */
2484 sid = rte_lcore_to_socket_id(cid);
2485 frag_cycles = (rte_get_tsc_hz() + NS_PER_S - 1) /
2486 NS_PER_S * frag_ttl_ns;
2488 lc->frag.tbl = rte_ip_frag_table_create(frag_tbl_sz,
2489 FRAG_TBL_BUCKET_ENTRIES, frag_tbl_sz, frag_cycles, sid);
2490 if (lc->frag.tbl == NULL) {
2491 printf("%s(%u): failed to create fragment table of size: %u, "
2493 __func__, cid, frag_tbl_sz, rte_errno);
2497 /* setup reassemble RX callbacks for all queues */
2498 for (i = 0; i != lc->nb_rx_queue; i++) {
2500 rxq = lc->rx_queue_list + i;
2501 cb = rte_eth_add_rx_callback(rxq->port_id, rxq->queue_id,
2504 printf("%s(%u): failed to install RX callback for "
2505 "portid=%u, queueid=%u, error code: %d\n",
2507 rxq->port_id, rxq->queue_id, rte_errno);
2516 reassemble_init(void)
2522 for (i = 0; i != nb_lcore_params; i++) {
2523 lc = lcore_params[i].lcore_id;
2524 rc = reassemble_lcore_init(lcore_conf + lc, lc);
2533 create_default_ipsec_flow(uint16_t port_id, uint64_t rx_offloads)
2535 struct rte_flow_action action[2];
2536 struct rte_flow_item pattern[2];
2537 struct rte_flow_attr attr = {0};
2538 struct rte_flow_error err;
2539 struct rte_flow *flow;
2542 if (!(rx_offloads & DEV_RX_OFFLOAD_SECURITY))
2545 /* Add the default rte_flow to enable SECURITY for all ESP packets */
2547 pattern[0].type = RTE_FLOW_ITEM_TYPE_ESP;
2548 pattern[0].spec = NULL;
2549 pattern[0].mask = NULL;
2550 pattern[0].last = NULL;
2551 pattern[1].type = RTE_FLOW_ITEM_TYPE_END;
2553 action[0].type = RTE_FLOW_ACTION_TYPE_SECURITY;
2554 action[0].conf = NULL;
2555 action[1].type = RTE_FLOW_ACTION_TYPE_END;
2556 action[1].conf = NULL;
2560 ret = rte_flow_validate(port_id, &attr, pattern, action, &err);
2564 flow = rte_flow_create(port_id, &attr, pattern, action, &err);
2568 flow_info_tbl[port_id].rx_def_flow = flow;
2569 RTE_LOG(INFO, IPSEC,
2570 "Created default flow enabling SECURITY for all ESP traffic on port %d\n",
2575 signal_handler(int signum)
2577 if (signum == SIGINT || signum == SIGTERM) {
2578 printf("\n\nSignal %d received, preparing to exit...\n",
2585 ev_mode_sess_verify(struct ipsec_sa *sa, int nb_sa)
2587 struct rte_ipsec_session *ips;
2593 for (i = 0; i < nb_sa; i++) {
2594 ips = ipsec_get_primary_session(&sa[i]);
2595 if (ips->type != RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL)
2596 rte_exit(EXIT_FAILURE, "Event mode supports only "
2597 "inline protocol sessions\n");
2603 check_event_mode_params(struct eh_conf *eh_conf)
2605 struct eventmode_conf *em_conf = NULL;
2606 struct lcore_params *params;
2609 if (!eh_conf || !eh_conf->mode_params)
2612 /* Get eventmode conf */
2613 em_conf = eh_conf->mode_params;
2615 if (eh_conf->mode == EH_PKT_TRANSFER_MODE_POLL &&
2616 em_conf->ext_params.sched_type != SCHED_TYPE_NOT_SET) {
2617 printf("error: option --event-schedule-type applies only to "
2622 if (eh_conf->mode != EH_PKT_TRANSFER_MODE_EVENT)
2625 /* Set schedule type to ORDERED if it wasn't explicitly set by user */
2626 if (em_conf->ext_params.sched_type == SCHED_TYPE_NOT_SET)
2627 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_ORDERED;
2630 * Event mode currently supports only inline protocol sessions.
2631 * If there are other types of sessions configured then exit with
2634 ev_mode_sess_verify(sa_in, nb_sa_in);
2635 ev_mode_sess_verify(sa_out, nb_sa_out);
2638 /* Option --config does not apply to event mode */
2639 if (nb_lcore_params > 0) {
2640 printf("error: option --config applies only to poll mode\n");
2645 * In order to use the same port_init routine for both poll and event
2646 * modes initialize lcore_params with one queue for each eth port
2648 lcore_params = lcore_params_array;
2649 RTE_ETH_FOREACH_DEV(portid) {
2650 if ((enabled_port_mask & (1 << portid)) == 0)
2653 params = &lcore_params[nb_lcore_params++];
2654 params->port_id = portid;
2655 params->queue_id = 0;
2656 params->lcore_id = rte_get_next_lcore(0, 0, 1);
2663 inline_sessions_free(struct sa_ctx *sa_ctx)
2665 struct rte_ipsec_session *ips;
2666 struct ipsec_sa *sa;
2673 for (i = 0; i < sa_ctx->nb_sa; i++) {
2675 sa = &sa_ctx->sa[i];
2679 ips = ipsec_get_primary_session(sa);
2680 if (ips->type != RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL &&
2681 ips->type != RTE_SECURITY_ACTION_TYPE_INLINE_CRYPTO)
2684 if (!rte_eth_dev_is_valid_port(sa->portid))
2687 ret = rte_security_session_destroy(
2688 rte_eth_dev_get_sec_ctx(sa->portid),
2691 RTE_LOG(ERR, IPSEC, "Failed to destroy security "
2692 "session type %d, spi %d\n",
2693 ips->type, sa->spi);
2698 main(int32_t argc, char **argv)
2706 uint64_t req_rx_offloads[RTE_MAX_ETHPORTS];
2707 uint64_t req_tx_offloads[RTE_MAX_ETHPORTS];
2708 struct eh_conf *eh_conf = NULL;
2712 ret = rte_eal_init(argc, argv);
2714 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
2719 signal(SIGINT, signal_handler);
2720 signal(SIGTERM, signal_handler);
2722 /* initialize event helper configuration */
2723 eh_conf = eh_conf_init();
2724 if (eh_conf == NULL)
2725 rte_exit(EXIT_FAILURE, "Failed to init event helper config");
2727 /* parse application arguments (after the EAL ones) */
2728 ret = parse_args(argc, argv, eh_conf);
2730 rte_exit(EXIT_FAILURE, "Invalid parameters\n");
2732 /* parse configuration file */
2733 if (parse_cfg_file(cfgfile) < 0) {
2734 printf("parsing file \"%s\" failed\n",
2736 print_usage(argv[0]);
2740 if ((unprotected_port_mask & enabled_port_mask) !=
2741 unprotected_port_mask)
2742 rte_exit(EXIT_FAILURE, "Invalid unprotected portmask 0x%x\n",
2743 unprotected_port_mask);
2745 if (check_poll_mode_params(eh_conf) < 0)
2746 rte_exit(EXIT_FAILURE, "check_poll_mode_params failed\n");
2748 if (check_event_mode_params(eh_conf) < 0)
2749 rte_exit(EXIT_FAILURE, "check_event_mode_params failed\n");
2751 ret = init_lcore_rx_queues();
2753 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
2755 nb_lcores = rte_lcore_count();
2757 sess_sz = max_session_size();
2759 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2760 if (rte_lcore_is_enabled(lcore_id) == 0)
2764 socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2768 /* mbuf_pool is initialised by the pool_init() function*/
2769 if (socket_ctx[socket_id].mbuf_pool)
2772 pool_init(&socket_ctx[socket_id], socket_id, NB_MBUF);
2773 session_pool_init(&socket_ctx[socket_id], socket_id, sess_sz);
2774 session_priv_pool_init(&socket_ctx[socket_id], socket_id,
2778 RTE_ETH_FOREACH_DEV(portid) {
2779 if ((enabled_port_mask & (1 << portid)) == 0)
2782 sa_check_offloads(portid, &req_rx_offloads[portid],
2783 &req_tx_offloads[portid]);
2784 port_init(portid, req_rx_offloads[portid],
2785 req_tx_offloads[portid]);
2791 * Set the enabled port mask in helper config for use by helper
2792 * sub-system. This will be used while initializing devices using
2793 * helper sub-system.
2795 eh_conf->eth_portmask = enabled_port_mask;
2797 /* Initialize eventmode components */
2798 ret = eh_devs_init(eh_conf);
2800 rte_exit(EXIT_FAILURE, "eh_devs_init failed, err=%d\n", ret);
2803 RTE_ETH_FOREACH_DEV(portid) {
2804 if ((enabled_port_mask & (1 << portid)) == 0)
2807 /* Create flow before starting the device */
2808 create_default_ipsec_flow(portid, req_rx_offloads[portid]);
2810 ret = rte_eth_dev_start(portid);
2812 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: "
2813 "err=%d, port=%d\n", ret, portid);
2815 * If enabled, put device in promiscuous mode.
2816 * This allows IO forwarding mode to forward packets
2817 * to itself through 2 cross-connected ports of the
2820 if (promiscuous_on) {
2821 ret = rte_eth_promiscuous_enable(portid);
2823 rte_exit(EXIT_FAILURE,
2824 "rte_eth_promiscuous_enable: err=%s, port=%d\n",
2825 rte_strerror(-ret), portid);
2828 rte_eth_dev_callback_register(portid,
2829 RTE_ETH_EVENT_IPSEC, inline_ipsec_event_callback, NULL);
2832 /* fragment reassemble is enabled */
2833 if (frag_tbl_sz != 0) {
2834 ret = reassemble_init();
2836 rte_exit(EXIT_FAILURE, "failed at reassemble init");
2839 /* Replicate each context per socket */
2840 for (i = 0; i < NB_SOCKETS && i < rte_socket_count(); i++) {
2841 socket_id = rte_socket_id_by_idx(i);
2842 if ((socket_ctx[socket_id].mbuf_pool != NULL) &&
2843 (socket_ctx[socket_id].sa_in == NULL) &&
2844 (socket_ctx[socket_id].sa_out == NULL)) {
2845 sa_init(&socket_ctx[socket_id], socket_id);
2846 sp4_init(&socket_ctx[socket_id], socket_id);
2847 sp6_init(&socket_ctx[socket_id], socket_id);
2848 rt_init(&socket_ctx[socket_id], socket_id);
2852 check_all_ports_link_status(enabled_port_mask);
2854 /* launch per-lcore init on every lcore */
2855 rte_eal_mp_remote_launch(ipsec_launch_one_lcore, eh_conf, CALL_MASTER);
2856 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
2857 if (rte_eal_wait_lcore(lcore_id) < 0)
2861 /* Uninitialize eventmode components */
2862 ret = eh_devs_uninit(eh_conf);
2864 rte_exit(EXIT_FAILURE, "eh_devs_uninit failed, err=%d\n", ret);
2866 /* Free eventmode configuration memory */
2867 eh_conf_uninit(eh_conf);
2869 /* Destroy inline inbound and outbound sessions */
2870 for (i = 0; i < NB_SOCKETS && i < rte_socket_count(); i++) {
2871 socket_id = rte_socket_id_by_idx(i);
2872 inline_sessions_free(socket_ctx[socket_id].sa_in);
2873 inline_sessions_free(socket_ctx[socket_id].sa_out);
2876 for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
2877 printf("Closing cryptodev %d...", cdev_id);
2878 rte_cryptodev_stop(cdev_id);
2879 rte_cryptodev_close(cdev_id);
2883 RTE_ETH_FOREACH_DEV(portid) {
2884 if ((enabled_port_mask & (1 << portid)) == 0)
2887 printf("Closing port %d...", portid);
2888 if (flow_info_tbl[portid].rx_def_flow) {
2889 struct rte_flow_error err;
2891 ret = rte_flow_destroy(portid,
2892 flow_info_tbl[portid].rx_def_flow, &err);
2894 RTE_LOG(ERR, IPSEC, "Failed to destroy flow "
2895 " for port %u, err msg: %s\n", portid,
2898 rte_eth_dev_stop(portid);
2899 rte_eth_dev_close(portid);