4 * Copyright(c) 2016 Intel Corporation. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * * Neither the name of Intel Corporation nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
38 #include <sys/types.h>
39 #include <netinet/in.h>
40 #include <netinet/ip.h>
41 #include <netinet/ip6.h>
43 #include <sys/queue.h>
48 #include <rte_common.h>
49 #include <rte_byteorder.h>
52 #include <rte_launch.h>
53 #include <rte_atomic.h>
54 #include <rte_cycles.h>
55 #include <rte_prefetch.h>
56 #include <rte_lcore.h>
57 #include <rte_per_lcore.h>
58 #include <rte_branch_prediction.h>
59 #include <rte_interrupts.h>
61 #include <rte_random.h>
62 #include <rte_debug.h>
63 #include <rte_ether.h>
64 #include <rte_ethdev.h>
65 #include <rte_mempool.h>
71 #include <rte_jhash.h>
72 #include <rte_cryptodev.h>
76 #define RTE_LOGTYPE_IPSEC RTE_LOGTYPE_USER1
78 #define MAX_JUMBO_PKT_LEN 9600
80 #define MEMPOOL_CACHE_SIZE 256
82 #define NB_MBUF (32000)
84 #define CDEV_MAP_ENTRIES 1024
85 #define CDEV_MP_NB_OBJS 2048
86 #define CDEV_MP_CACHE_SZ 64
87 #define MAX_QUEUE_PAIRS 1
89 #define OPTION_CONFIG "config"
90 #define OPTION_SINGLE_SA "single-sa"
91 #define OPTION_EP0 "ep0"
92 #define OPTION_EP1 "ep1"
94 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
98 /* Configure how many packets ahead to prefetch, when reading packets */
99 #define PREFETCH_OFFSET 3
101 #define MAX_RX_QUEUE_PER_LCORE 16
103 #define MAX_LCORE_PARAMS 1024
105 #define UNPROTECTED_PORT(port) (unprotected_port_mask & (1 << portid))
108 * Configurable number of RX/TX ring descriptors
110 #define IPSEC_SECGW_RX_DESC_DEFAULT 128
111 #define IPSEC_SECGW_TX_DESC_DEFAULT 512
112 static uint16_t nb_rxd = IPSEC_SECGW_RX_DESC_DEFAULT;
113 static uint16_t nb_txd = IPSEC_SECGW_TX_DESC_DEFAULT;
115 #if RTE_BYTE_ORDER != RTE_LITTLE_ENDIAN
116 #define __BYTES_TO_UINT64(a, b, c, d, e, f, g, h) \
117 (((uint64_t)((a) & 0xff) << 56) | \
118 ((uint64_t)((b) & 0xff) << 48) | \
119 ((uint64_t)((c) & 0xff) << 40) | \
120 ((uint64_t)((d) & 0xff) << 32) | \
121 ((uint64_t)((e) & 0xff) << 24) | \
122 ((uint64_t)((f) & 0xff) << 16) | \
123 ((uint64_t)((g) & 0xff) << 8) | \
124 ((uint64_t)(h) & 0xff))
126 #define __BYTES_TO_UINT64(a, b, c, d, e, f, g, h) \
127 (((uint64_t)((h) & 0xff) << 56) | \
128 ((uint64_t)((g) & 0xff) << 48) | \
129 ((uint64_t)((f) & 0xff) << 40) | \
130 ((uint64_t)((e) & 0xff) << 32) | \
131 ((uint64_t)((d) & 0xff) << 24) | \
132 ((uint64_t)((c) & 0xff) << 16) | \
133 ((uint64_t)((b) & 0xff) << 8) | \
134 ((uint64_t)(a) & 0xff))
136 #define ETHADDR(a, b, c, d, e, f) (__BYTES_TO_UINT64(a, b, c, d, e, f, 0, 0))
138 #define ETHADDR_TO_UINT64(addr) __BYTES_TO_UINT64( \
139 addr.addr_bytes[0], addr.addr_bytes[1], \
140 addr.addr_bytes[2], addr.addr_bytes[3], \
141 addr.addr_bytes[4], addr.addr_bytes[5], \
144 /* port/source ethernet addr and destination ethernet addr */
145 struct ethaddr_info {
149 struct ethaddr_info ethaddr_tbl[RTE_MAX_ETHPORTS] = {
150 { 0, ETHADDR(0x00, 0x16, 0x3e, 0x7e, 0x94, 0x9a) },
151 { 0, ETHADDR(0x00, 0x16, 0x3e, 0x22, 0xa1, 0xd9) },
152 { 0, ETHADDR(0x00, 0x16, 0x3e, 0x08, 0x69, 0x26) },
153 { 0, ETHADDR(0x00, 0x16, 0x3e, 0x49, 0x9e, 0xdd) }
156 /* mask of enabled ports */
157 static uint32_t enabled_port_mask;
158 static uint32_t unprotected_port_mask;
159 static int32_t promiscuous_on = 1;
160 static int32_t numa_on = 1; /**< NUMA is enabled by default. */
161 static int32_t ep = -1; /**< Endpoint configuration (0 or 1) */
162 static uint32_t nb_lcores;
163 static uint32_t single_sa;
164 static uint32_t single_sa_idx;
166 struct lcore_rx_queue {
169 } __rte_cache_aligned;
171 struct lcore_params {
175 } __rte_cache_aligned;
177 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
179 static struct lcore_params *lcore_params;
180 static uint16_t nb_lcore_params;
182 static struct rte_hash *cdev_map_in;
183 static struct rte_hash *cdev_map_out;
187 struct rte_mbuf *m_table[MAX_PKT_BURST] __rte_aligned(sizeof(void *));
191 uint16_t nb_rx_queue;
192 struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
193 uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
194 struct buffer tx_mbufs[RTE_MAX_ETHPORTS];
195 struct ipsec_ctx inbound;
196 struct ipsec_ctx outbound;
197 struct rt_ctx *rt4_ctx;
198 struct rt_ctx *rt6_ctx;
199 } __rte_cache_aligned;
201 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
203 static struct rte_eth_conf port_conf = {
205 .mq_mode = ETH_MQ_RX_RSS,
206 .max_rx_pkt_len = ETHER_MAX_LEN,
208 .header_split = 0, /**< Header Split disabled */
209 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
210 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
211 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */
212 .hw_strip_crc = 0, /**< CRC stripped by hardware */
217 .rss_hf = ETH_RSS_IP | ETH_RSS_UDP |
218 ETH_RSS_TCP | ETH_RSS_SCTP,
222 .mq_mode = ETH_MQ_TX_NONE,
226 static struct socket_ctx socket_ctx[NB_SOCKETS];
228 struct traffic_type {
229 const uint8_t *data[MAX_PKT_BURST * 2];
230 struct rte_mbuf *pkts[MAX_PKT_BURST * 2];
231 uint32_t res[MAX_PKT_BURST * 2];
235 struct ipsec_traffic {
236 struct traffic_type ipsec;
237 struct traffic_type ip4;
238 struct traffic_type ip6;
242 prepare_one_packet(struct rte_mbuf *pkt, struct ipsec_traffic *t)
245 struct ether_hdr *eth;
247 eth = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
248 if (eth->ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4)) {
249 nlp = (uint8_t *)rte_pktmbuf_adj(pkt, ETHER_HDR_LEN);
250 nlp = RTE_PTR_ADD(nlp, offsetof(struct ip, ip_p));
251 if (*nlp == IPPROTO_ESP)
252 t->ipsec.pkts[(t->ipsec.num)++] = pkt;
254 t->ip4.data[t->ip4.num] = nlp;
255 t->ip4.pkts[(t->ip4.num)++] = pkt;
257 } else if (eth->ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6)) {
258 nlp = (uint8_t *)rte_pktmbuf_adj(pkt, ETHER_HDR_LEN);
259 nlp = RTE_PTR_ADD(nlp, offsetof(struct ip6_hdr, ip6_nxt));
260 if (*nlp == IPPROTO_ESP)
261 t->ipsec.pkts[(t->ipsec.num)++] = pkt;
263 t->ip6.data[t->ip6.num] = nlp;
264 t->ip6.pkts[(t->ip6.num)++] = pkt;
267 /* Unknown/Unsupported type, drop the packet */
268 RTE_LOG(ERR, IPSEC, "Unsupported packet type\n");
269 rte_pktmbuf_free(pkt);
274 prepare_traffic(struct rte_mbuf **pkts, struct ipsec_traffic *t,
283 for (i = 0; i < (nb_pkts - PREFETCH_OFFSET); i++) {
284 rte_prefetch0(rte_pktmbuf_mtod(pkts[i + PREFETCH_OFFSET],
286 prepare_one_packet(pkts[i], t);
288 /* Process left packets */
289 for (; i < nb_pkts; i++)
290 prepare_one_packet(pkts[i], t);
294 prepare_tx_pkt(struct rte_mbuf *pkt, uint8_t port)
297 struct ether_hdr *ethhdr;
299 ip = rte_pktmbuf_mtod(pkt, struct ip *);
301 ethhdr = (struct ether_hdr *)rte_pktmbuf_prepend(pkt, ETHER_HDR_LEN);
303 if (ip->ip_v == IPVERSION) {
304 pkt->ol_flags |= PKT_TX_IP_CKSUM | PKT_TX_IPV4;
305 pkt->l3_len = sizeof(struct ip);
306 pkt->l2_len = ETHER_HDR_LEN;
308 ethhdr->ether_type = rte_cpu_to_be_16(ETHER_TYPE_IPv4);
310 pkt->ol_flags |= PKT_TX_IPV6;
311 pkt->l3_len = sizeof(struct ip6_hdr);
312 pkt->l2_len = ETHER_HDR_LEN;
314 ethhdr->ether_type = rte_cpu_to_be_16(ETHER_TYPE_IPv6);
317 memcpy(ðhdr->s_addr, ðaddr_tbl[port].src,
318 sizeof(struct ether_addr));
319 memcpy(ðhdr->d_addr, ðaddr_tbl[port].dst,
320 sizeof(struct ether_addr));
324 prepare_tx_burst(struct rte_mbuf *pkts[], uint16_t nb_pkts, uint8_t port)
327 const int32_t prefetch_offset = 2;
329 for (i = 0; i < (nb_pkts - prefetch_offset); i++) {
330 rte_mbuf_prefetch_part2(pkts[i + prefetch_offset]);
331 prepare_tx_pkt(pkts[i], port);
333 /* Process left packets */
334 for (; i < nb_pkts; i++)
335 prepare_tx_pkt(pkts[i], port);
338 /* Send burst of packets on an output interface */
339 static inline int32_t
340 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
342 struct rte_mbuf **m_table;
346 queueid = qconf->tx_queue_id[port];
347 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
349 prepare_tx_burst(m_table, n, port);
351 ret = rte_eth_tx_burst(port, queueid, m_table, n);
352 if (unlikely(ret < n)) {
354 rte_pktmbuf_free(m_table[ret]);
361 /* Enqueue a single packet, and send burst if queue is filled */
362 static inline int32_t
363 send_single_packet(struct rte_mbuf *m, uint8_t port)
367 struct lcore_conf *qconf;
369 lcore_id = rte_lcore_id();
371 qconf = &lcore_conf[lcore_id];
372 len = qconf->tx_mbufs[port].len;
373 qconf->tx_mbufs[port].m_table[len] = m;
376 /* enough pkts to be sent */
377 if (unlikely(len == MAX_PKT_BURST)) {
378 send_burst(qconf, MAX_PKT_BURST, port);
382 qconf->tx_mbufs[port].len = len;
387 inbound_sp_sa(struct sp_ctx *sp, struct sa_ctx *sa, struct traffic_type *ip)
390 uint32_t i, j, res, sa_idx;
395 rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
396 ip->num, DEFAULT_MAX_CATEGORIES);
399 for (i = 0; i < ip->num; i++) {
410 /* Check return SA SPI matches pkt SPI */
411 sa_idx = ip->res[i] & PROTECT_MASK;
412 if (sa_idx == 0 || !inbound_sa_check(sa, m, sa_idx)) {
422 process_pkts_inbound(struct ipsec_ctx *ipsec_ctx,
423 struct ipsec_traffic *traffic)
426 uint16_t idx, nb_pkts_in, i;
428 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
429 traffic->ipsec.num, MAX_PKT_BURST);
431 /* SP/ACL Inbound check ipsec and ip4 */
432 for (i = 0; i < nb_pkts_in; i++) {
433 m = traffic->ipsec.pkts[i];
434 struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
435 if (ip->ip_v == IPVERSION) {
436 idx = traffic->ip4.num++;
437 traffic->ip4.pkts[idx] = m;
438 traffic->ip4.data[idx] = rte_pktmbuf_mtod_offset(m,
439 uint8_t *, offsetof(struct ip, ip_p));
440 } else if (ip->ip_v == IP6_VERSION) {
441 idx = traffic->ip6.num++;
442 traffic->ip6.pkts[idx] = m;
443 traffic->ip6.data[idx] = rte_pktmbuf_mtod_offset(m,
445 offsetof(struct ip6_hdr, ip6_nxt));
450 inbound_sp_sa(ipsec_ctx->sp4_ctx, ipsec_ctx->sa_ctx, &traffic->ip4);
452 inbound_sp_sa(ipsec_ctx->sp6_ctx, ipsec_ctx->sa_ctx, &traffic->ip6);
456 outbound_sp(struct sp_ctx *sp, struct traffic_type *ip,
457 struct traffic_type *ipsec)
460 uint32_t i, j, sa_idx;
465 rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
466 ip->num, DEFAULT_MAX_CATEGORIES);
469 for (i = 0; i < ip->num; i++) {
471 sa_idx = ip->res[i] & PROTECT_MASK;
472 if ((ip->res[i] == 0) || (ip->res[i] & DISCARD))
474 else if (sa_idx != 0) {
475 ipsec->res[ipsec->num] = sa_idx;
476 ipsec->pkts[ipsec->num++] = m;
484 process_pkts_outbound(struct ipsec_ctx *ipsec_ctx,
485 struct ipsec_traffic *traffic)
488 uint16_t idx, nb_pkts_out, i;
490 /* Drop any IPsec traffic from protected ports */
491 for (i = 0; i < traffic->ipsec.num; i++)
492 rte_pktmbuf_free(traffic->ipsec.pkts[i]);
494 traffic->ipsec.num = 0;
496 outbound_sp(ipsec_ctx->sp4_ctx, &traffic->ip4, &traffic->ipsec);
498 outbound_sp(ipsec_ctx->sp6_ctx, &traffic->ip6, &traffic->ipsec);
500 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
501 traffic->ipsec.res, traffic->ipsec.num,
504 for (i = 0; i < nb_pkts_out; i++) {
505 m = traffic->ipsec.pkts[i];
506 struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
507 if (ip->ip_v == IPVERSION) {
508 idx = traffic->ip4.num++;
509 traffic->ip4.pkts[idx] = m;
511 idx = traffic->ip6.num++;
512 traffic->ip6.pkts[idx] = m;
518 process_pkts_inbound_nosp(struct ipsec_ctx *ipsec_ctx,
519 struct ipsec_traffic *traffic)
522 uint32_t nb_pkts_in, i, idx;
524 /* Drop any IPv4 traffic from unprotected ports */
525 for (i = 0; i < traffic->ip4.num; i++)
526 rte_pktmbuf_free(traffic->ip4.pkts[i]);
528 traffic->ip4.num = 0;
530 /* Drop any IPv6 traffic from unprotected ports */
531 for (i = 0; i < traffic->ip6.num; i++)
532 rte_pktmbuf_free(traffic->ip6.pkts[i]);
534 traffic->ip6.num = 0;
536 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
537 traffic->ipsec.num, MAX_PKT_BURST);
539 for (i = 0; i < nb_pkts_in; i++) {
540 m = traffic->ipsec.pkts[i];
541 struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
542 if (ip->ip_v == IPVERSION) {
543 idx = traffic->ip4.num++;
544 traffic->ip4.pkts[idx] = m;
546 idx = traffic->ip6.num++;
547 traffic->ip6.pkts[idx] = m;
553 process_pkts_outbound_nosp(struct ipsec_ctx *ipsec_ctx,
554 struct ipsec_traffic *traffic)
557 uint32_t nb_pkts_out, i;
560 /* Drop any IPsec traffic from protected ports */
561 for (i = 0; i < traffic->ipsec.num; i++)
562 rte_pktmbuf_free(traffic->ipsec.pkts[i]);
564 traffic->ipsec.num = 0;
566 for (i = 0; i < traffic->ip4.num; i++)
567 traffic->ip4.res[i] = single_sa_idx;
569 for (i = 0; i < traffic->ip6.num; i++)
570 traffic->ip6.res[i] = single_sa_idx;
572 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ip4.pkts,
573 traffic->ip4.res, traffic->ip4.num,
576 /* They all sue the same SA (ip4 or ip6 tunnel) */
577 m = traffic->ipsec.pkts[i];
578 ip = rte_pktmbuf_mtod(m, struct ip *);
579 if (ip->ip_v == IPVERSION)
580 traffic->ip4.num = nb_pkts_out;
582 traffic->ip6.num = nb_pkts_out;
586 route4_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
588 uint32_t hop[MAX_PKT_BURST * 2];
589 uint32_t dst_ip[MAX_PKT_BURST * 2];
595 for (i = 0; i < nb_pkts; i++) {
596 offset = offsetof(struct ip, ip_dst);
597 dst_ip[i] = *rte_pktmbuf_mtod_offset(pkts[i],
599 dst_ip[i] = rte_be_to_cpu_32(dst_ip[i]);
602 rte_lpm_lookup_bulk((struct rte_lpm *)rt_ctx, dst_ip, hop, nb_pkts);
604 for (i = 0; i < nb_pkts; i++) {
605 if ((hop[i] & RTE_LPM_LOOKUP_SUCCESS) == 0) {
606 rte_pktmbuf_free(pkts[i]);
609 send_single_packet(pkts[i], hop[i] & 0xff);
614 route6_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
616 int16_t hop[MAX_PKT_BURST * 2];
617 uint8_t dst_ip[MAX_PKT_BURST * 2][16];
624 for (i = 0; i < nb_pkts; i++) {
625 offset = offsetof(struct ip6_hdr, ip6_dst);
626 ip6_dst = rte_pktmbuf_mtod_offset(pkts[i], uint8_t *, offset);
627 memcpy(&dst_ip[i][0], ip6_dst, 16);
630 rte_lpm6_lookup_bulk_func((struct rte_lpm6 *)rt_ctx, dst_ip,
633 for (i = 0; i < nb_pkts; i++) {
635 rte_pktmbuf_free(pkts[i]);
638 send_single_packet(pkts[i], hop[i] & 0xff);
643 process_pkts(struct lcore_conf *qconf, struct rte_mbuf **pkts,
644 uint8_t nb_pkts, uint8_t portid)
646 struct ipsec_traffic traffic;
648 prepare_traffic(pkts, &traffic, nb_pkts);
650 if (unlikely(single_sa)) {
651 if (UNPROTECTED_PORT(portid))
652 process_pkts_inbound_nosp(&qconf->inbound, &traffic);
654 process_pkts_outbound_nosp(&qconf->outbound, &traffic);
656 if (UNPROTECTED_PORT(portid))
657 process_pkts_inbound(&qconf->inbound, &traffic);
659 process_pkts_outbound(&qconf->outbound, &traffic);
662 route4_pkts(qconf->rt4_ctx, traffic.ip4.pkts, traffic.ip4.num);
663 route6_pkts(qconf->rt6_ctx, traffic.ip6.pkts, traffic.ip6.num);
667 drain_buffers(struct lcore_conf *qconf)
672 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
673 buf = &qconf->tx_mbufs[portid];
676 send_burst(qconf, buf->len, portid);
681 /* main processing loop */
683 main_loop(__attribute__((unused)) void *dummy)
685 struct rte_mbuf *pkts[MAX_PKT_BURST];
687 uint64_t prev_tsc, diff_tsc, cur_tsc;
689 uint8_t portid, queueid;
690 struct lcore_conf *qconf;
692 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
693 / US_PER_S * BURST_TX_DRAIN_US;
694 struct lcore_rx_queue *rxql;
697 lcore_id = rte_lcore_id();
698 qconf = &lcore_conf[lcore_id];
699 rxql = qconf->rx_queue_list;
700 socket_id = rte_lcore_to_socket_id(lcore_id);
702 qconf->rt4_ctx = socket_ctx[socket_id].rt_ip4;
703 qconf->rt6_ctx = socket_ctx[socket_id].rt_ip6;
704 qconf->inbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_in;
705 qconf->inbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_in;
706 qconf->inbound.sa_ctx = socket_ctx[socket_id].sa_in;
707 qconf->inbound.cdev_map = cdev_map_in;
708 qconf->outbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_out;
709 qconf->outbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_out;
710 qconf->outbound.sa_ctx = socket_ctx[socket_id].sa_out;
711 qconf->outbound.cdev_map = cdev_map_out;
713 if (qconf->nb_rx_queue == 0) {
714 RTE_LOG(INFO, IPSEC, "lcore %u has nothing to do\n", lcore_id);
718 RTE_LOG(INFO, IPSEC, "entering main loop on lcore %u\n", lcore_id);
720 for (i = 0; i < qconf->nb_rx_queue; i++) {
721 portid = rxql[i].port_id;
722 queueid = rxql[i].queue_id;
724 " -- lcoreid=%u portid=%hhu rxqueueid=%hhu\n",
725 lcore_id, portid, queueid);
729 cur_tsc = rte_rdtsc();
731 /* TX queue buffer drain */
732 diff_tsc = cur_tsc - prev_tsc;
734 if (unlikely(diff_tsc > drain_tsc)) {
735 drain_buffers(qconf);
739 /* Read packet from RX queues */
740 for (i = 0; i < qconf->nb_rx_queue; ++i) {
741 portid = rxql[i].port_id;
742 queueid = rxql[i].queue_id;
743 nb_rx = rte_eth_rx_burst(portid, queueid,
744 pkts, MAX_PKT_BURST);
747 process_pkts(qconf, pkts, nb_rx, portid);
755 uint8_t lcore, portid, nb_ports;
759 if (lcore_params == NULL) {
760 printf("Error: No port/queue/core mappings\n");
764 nb_ports = rte_eth_dev_count();
766 for (i = 0; i < nb_lcore_params; ++i) {
767 lcore = lcore_params[i].lcore_id;
768 if (!rte_lcore_is_enabled(lcore)) {
769 printf("error: lcore %hhu is not enabled in "
770 "lcore mask\n", lcore);
773 socket_id = rte_lcore_to_socket_id(lcore);
774 if (socket_id != 0 && numa_on == 0) {
775 printf("warning: lcore %hhu is on socket %d "
779 portid = lcore_params[i].port_id;
780 if ((enabled_port_mask & (1 << portid)) == 0) {
781 printf("port %u is not enabled in port mask\n", portid);
784 if (portid >= nb_ports) {
785 printf("port %u is not present on the board\n", portid);
793 get_port_nb_rx_queues(const uint8_t port)
798 for (i = 0; i < nb_lcore_params; ++i) {
799 if (lcore_params[i].port_id == port &&
800 lcore_params[i].queue_id > queue)
801 queue = lcore_params[i].queue_id;
803 return (uint8_t)(++queue);
807 init_lcore_rx_queues(void)
809 uint16_t i, nb_rx_queue;
812 for (i = 0; i < nb_lcore_params; ++i) {
813 lcore = lcore_params[i].lcore_id;
814 nb_rx_queue = lcore_conf[lcore].nb_rx_queue;
815 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
816 printf("error: too many queues (%u) for lcore: %u\n",
817 nb_rx_queue + 1, lcore);
820 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
821 lcore_params[i].port_id;
822 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
823 lcore_params[i].queue_id;
824 lcore_conf[lcore].nb_rx_queue++;
831 print_usage(const char *prgname)
833 printf("%s [EAL options] -- -p PORTMASK -P -u PORTMASK"
834 " --"OPTION_CONFIG" (port,queue,lcore)[,(port,queue,lcore]"
835 " --single-sa SAIDX --ep0|--ep1\n"
836 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
837 " -P : enable promiscuous mode\n"
838 " -u PORTMASK: hexadecimal bitmask of unprotected ports\n"
839 " --"OPTION_CONFIG": (port,queue,lcore): "
840 "rx queues configuration\n"
841 " --single-sa SAIDX: use single SA index for outbound, "
843 " --ep0: Configure as Endpoint 0\n"
844 " --ep1: Configure as Endpoint 1\n", prgname);
848 parse_portmask(const char *portmask)
853 /* parse hexadecimal string */
854 pm = strtoul(portmask, &end, 16);
855 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
858 if ((pm == 0) && errno)
865 parse_decimal(const char *str)
870 num = strtoul(str, &end, 10);
871 if ((str[0] == '\0') || (end == NULL) || (*end != '\0'))
878 parse_config(const char *q_arg)
881 const char *p, *p0 = q_arg;
889 unsigned long int_fld[_NUM_FLD];
890 char *str_fld[_NUM_FLD];
896 while ((p = strchr(p0, '(')) != NULL) {
903 if (size >= sizeof(s))
906 snprintf(s, sizeof(s), "%.*s", size, p);
907 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
910 for (i = 0; i < _NUM_FLD; i++) {
912 int_fld[i] = strtoul(str_fld[i], &end, 0);
913 if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
916 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
917 printf("exceeded max number of lcore params: %hu\n",
921 lcore_params_array[nb_lcore_params].port_id =
922 (uint8_t)int_fld[FLD_PORT];
923 lcore_params_array[nb_lcore_params].queue_id =
924 (uint8_t)int_fld[FLD_QUEUE];
925 lcore_params_array[nb_lcore_params].lcore_id =
926 (uint8_t)int_fld[FLD_LCORE];
929 lcore_params = lcore_params_array;
933 #define __STRNCMP(name, opt) (!strncmp(name, opt, sizeof(opt)))
935 parse_args_long_options(struct option *lgopts, int32_t option_index)
938 const char *optname = lgopts[option_index].name;
940 if (__STRNCMP(optname, OPTION_CONFIG)) {
941 ret = parse_config(optarg);
943 printf("invalid config\n");
946 if (__STRNCMP(optname, OPTION_SINGLE_SA)) {
947 ret = parse_decimal(optarg);
951 printf("Configured with single SA index %u\n",
957 if (__STRNCMP(optname, OPTION_EP0)) {
958 printf("endpoint 0\n");
963 if (__STRNCMP(optname, OPTION_EP1)) {
964 printf("endpoint 1\n");
974 parse_args(int32_t argc, char **argv)
978 int32_t option_index;
979 char *prgname = argv[0];
980 static struct option lgopts[] = {
981 {OPTION_CONFIG, 1, 0, 0},
982 {OPTION_SINGLE_SA, 1, 0, 0},
983 {OPTION_EP0, 0, 0, 0},
984 {OPTION_EP1, 0, 0, 0},
990 while ((opt = getopt_long(argc, argvopt, "p:Pu:",
991 lgopts, &option_index)) != EOF) {
995 enabled_port_mask = parse_portmask(optarg);
996 if (enabled_port_mask == 0) {
997 printf("invalid portmask\n");
998 print_usage(prgname);
1003 printf("Promiscuous mode selected\n");
1007 unprotected_port_mask = parse_portmask(optarg);
1008 if (unprotected_port_mask == 0) {
1009 printf("invalid unprotected portmask\n");
1010 print_usage(prgname);
1015 if (parse_args_long_options(lgopts, option_index)) {
1016 print_usage(prgname);
1021 print_usage(prgname);
1027 argv[optind-1] = prgname;
1030 optind = 0; /* reset getopt lib */
1035 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1037 char buf[ETHER_ADDR_FMT_SIZE];
1038 ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
1039 printf("%s%s", name, buf);
1042 /* Check the link status of all ports in up to 9s, and print them finally */
1044 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
1046 #define CHECK_INTERVAL 100 /* 100ms */
1047 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1048 uint8_t portid, count, all_ports_up, print_flag = 0;
1049 struct rte_eth_link link;
1051 printf("\nChecking link status");
1053 for (count = 0; count <= MAX_CHECK_TIME; count++) {
1055 for (portid = 0; portid < port_num; portid++) {
1056 if ((port_mask & (1 << portid)) == 0)
1058 memset(&link, 0, sizeof(link));
1059 rte_eth_link_get_nowait(portid, &link);
1060 /* print link status if flag set */
1061 if (print_flag == 1) {
1062 if (link.link_status)
1063 printf("Port %d Link Up - speed %u "
1064 "Mbps - %s\n", (uint8_t)portid,
1065 (uint32_t)link.link_speed,
1066 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1067 ("full-duplex") : ("half-duplex\n"));
1069 printf("Port %d Link Down\n",
1073 /* clear all_ports_up flag if any link down */
1074 if (link.link_status == ETH_LINK_DOWN) {
1079 /* after finally printing all link status, get out */
1080 if (print_flag == 1)
1083 if (all_ports_up == 0) {
1086 rte_delay_ms(CHECK_INTERVAL);
1089 /* set the print_flag if all ports up or timeout */
1090 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1098 add_mapping(struct rte_hash *map, const char *str, uint16_t cdev_id,
1099 uint16_t qp, struct lcore_params *params,
1100 struct ipsec_ctx *ipsec_ctx,
1101 const struct rte_cryptodev_capabilities *cipher,
1102 const struct rte_cryptodev_capabilities *auth)
1106 struct cdev_key key = { 0 };
1108 key.lcore_id = params->lcore_id;
1110 key.cipher_algo = cipher->sym.cipher.algo;
1112 key.auth_algo = auth->sym.auth.algo;
1114 ret = rte_hash_lookup(map, &key);
1118 for (i = 0; i < ipsec_ctx->nb_qps; i++)
1119 if (ipsec_ctx->tbl[i].id == cdev_id)
1122 if (i == ipsec_ctx->nb_qps) {
1123 if (ipsec_ctx->nb_qps == MAX_QP_PER_LCORE) {
1124 printf("Maximum number of crypto devices assigned to "
1125 "a core, increase MAX_QP_PER_LCORE value\n");
1128 ipsec_ctx->tbl[i].id = cdev_id;
1129 ipsec_ctx->tbl[i].qp = qp;
1130 ipsec_ctx->nb_qps++;
1131 printf("%s cdev mapping: lcore %u using cdev %u qp %u "
1132 "(cdev_id_qp %lu)\n", str, key.lcore_id,
1136 ret = rte_hash_add_key_data(map, &key, (void *)i);
1138 printf("Faled to insert cdev mapping for (lcore %u, "
1139 "cdev %u, qp %u), errno %d\n",
1140 key.lcore_id, ipsec_ctx->tbl[i].id,
1141 ipsec_ctx->tbl[i].qp, ret);
1149 add_cdev_mapping(struct rte_cryptodev_info *dev_info, uint16_t cdev_id,
1150 uint16_t qp, struct lcore_params *params)
1153 const struct rte_cryptodev_capabilities *i, *j;
1154 struct rte_hash *map;
1155 struct lcore_conf *qconf;
1156 struct ipsec_ctx *ipsec_ctx;
1159 qconf = &lcore_conf[params->lcore_id];
1161 if ((unprotected_port_mask & (1 << params->port_id)) == 0) {
1163 ipsec_ctx = &qconf->outbound;
1167 ipsec_ctx = &qconf->inbound;
1171 /* Required cryptodevs with operation chainning */
1172 if (!(dev_info->feature_flags &
1173 RTE_CRYPTODEV_FF_SYM_OPERATION_CHAINING))
1176 for (i = dev_info->capabilities;
1177 i->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; i++) {
1178 if (i->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1181 if (i->sym.xform_type != RTE_CRYPTO_SYM_XFORM_CIPHER)
1184 for (j = dev_info->capabilities;
1185 j->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; j++) {
1186 if (j->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1189 if (j->sym.xform_type != RTE_CRYPTO_SYM_XFORM_AUTH)
1192 ret |= add_mapping(map, str, cdev_id, qp, params,
1201 cryptodevs_init(void)
1203 struct rte_cryptodev_config dev_conf;
1204 struct rte_cryptodev_qp_conf qp_conf;
1205 uint16_t idx, max_nb_qps, qp, i;
1207 struct rte_hash_parameters params = { 0 };
1209 params.entries = CDEV_MAP_ENTRIES;
1210 params.key_len = sizeof(struct cdev_key);
1211 params.hash_func = rte_jhash;
1212 params.hash_func_init_val = 0;
1213 params.socket_id = rte_socket_id();
1215 params.name = "cdev_map_in";
1216 cdev_map_in = rte_hash_create(¶ms);
1217 if (cdev_map_in == NULL)
1218 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1221 params.name = "cdev_map_out";
1222 cdev_map_out = rte_hash_create(¶ms);
1223 if (cdev_map_out == NULL)
1224 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1227 printf("lcore/cryptodev/qp mappings:\n");
1230 /* Start from last cdev id to give HW priority */
1231 for (cdev_id = rte_cryptodev_count() - 1; cdev_id >= 0; cdev_id--) {
1232 struct rte_cryptodev_info cdev_info;
1234 rte_cryptodev_info_get(cdev_id, &cdev_info);
1236 if (nb_lcore_params > cdev_info.max_nb_queue_pairs)
1237 max_nb_qps = cdev_info.max_nb_queue_pairs;
1239 max_nb_qps = nb_lcore_params;
1243 while (qp < max_nb_qps && i < nb_lcore_params) {
1244 if (add_cdev_mapping(&cdev_info, cdev_id, qp,
1245 &lcore_params[idx]))
1248 idx = idx % nb_lcore_params;
1255 dev_conf.socket_id = rte_cryptodev_socket_id(cdev_id);
1256 dev_conf.nb_queue_pairs = qp;
1257 dev_conf.session_mp.nb_objs = CDEV_MP_NB_OBJS;
1258 dev_conf.session_mp.cache_size = CDEV_MP_CACHE_SZ;
1260 if (rte_cryptodev_configure(cdev_id, &dev_conf))
1261 rte_panic("Failed to initialize crypodev %u\n",
1264 qp_conf.nb_descriptors = CDEV_MP_NB_OBJS;
1265 for (qp = 0; qp < dev_conf.nb_queue_pairs; qp++)
1266 if (rte_cryptodev_queue_pair_setup(cdev_id, qp,
1267 &qp_conf, dev_conf.socket_id))
1268 rte_panic("Failed to setup queue %u for "
1269 "cdev_id %u\n", 0, cdev_id);
1278 port_init(uint8_t portid)
1280 struct rte_eth_dev_info dev_info;
1281 struct rte_eth_txconf *txconf;
1282 uint16_t nb_tx_queue, nb_rx_queue;
1283 uint16_t tx_queueid, rx_queueid, queue, lcore_id;
1284 int32_t ret, socket_id;
1285 struct lcore_conf *qconf;
1286 struct ether_addr ethaddr;
1288 rte_eth_dev_info_get(portid, &dev_info);
1290 printf("Configuring device port %u:\n", portid);
1292 rte_eth_macaddr_get(portid, ðaddr);
1293 ethaddr_tbl[portid].src = ETHADDR_TO_UINT64(ethaddr);
1294 print_ethaddr("Address: ", ðaddr);
1297 nb_rx_queue = get_port_nb_rx_queues(portid);
1298 nb_tx_queue = nb_lcores;
1300 if (nb_rx_queue > dev_info.max_rx_queues)
1301 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
1302 "(max rx queue is %u)\n",
1303 nb_rx_queue, dev_info.max_rx_queues);
1305 if (nb_tx_queue > dev_info.max_tx_queues)
1306 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
1307 "(max tx queue is %u)\n",
1308 nb_tx_queue, dev_info.max_tx_queues);
1310 printf("Creating queues: nb_rx_queue=%d nb_tx_queue=%u...\n",
1311 nb_rx_queue, nb_tx_queue);
1313 ret = rte_eth_dev_configure(portid, nb_rx_queue, nb_tx_queue,
1316 rte_exit(EXIT_FAILURE, "Cannot configure device: "
1317 "err=%d, port=%d\n", ret, portid);
1319 /* init one TX queue per lcore */
1321 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1322 if (rte_lcore_is_enabled(lcore_id) == 0)
1326 socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1331 printf("Setup txq=%u,%d,%d\n", lcore_id, tx_queueid, socket_id);
1333 txconf = &dev_info.default_txconf;
1334 txconf->txq_flags = 0;
1336 ret = rte_eth_tx_queue_setup(portid, tx_queueid, nb_txd,
1339 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
1340 "err=%d, port=%d\n", ret, portid);
1342 qconf = &lcore_conf[lcore_id];
1343 qconf->tx_queue_id[portid] = tx_queueid;
1346 /* init RX queues */
1347 for (queue = 0; queue < qconf->nb_rx_queue; ++queue) {
1348 if (portid != qconf->rx_queue_list[queue].port_id)
1351 rx_queueid = qconf->rx_queue_list[queue].queue_id;
1353 printf("Setup rxq=%d,%d,%d\n", portid, rx_queueid,
1356 ret = rte_eth_rx_queue_setup(portid, rx_queueid,
1357 nb_rxd, socket_id, NULL,
1358 socket_ctx[socket_id].mbuf_pool);
1360 rte_exit(EXIT_FAILURE,
1361 "rte_eth_rx_queue_setup: err=%d, "
1362 "port=%d\n", ret, portid);
1369 pool_init(struct socket_ctx *ctx, int32_t socket_id, uint32_t nb_mbuf)
1373 snprintf(s, sizeof(s), "mbuf_pool_%d", socket_id);
1374 ctx->mbuf_pool = rte_pktmbuf_pool_create(s, nb_mbuf,
1375 MEMPOOL_CACHE_SIZE, ipsec_metadata_size(),
1376 RTE_MBUF_DEFAULT_BUF_SIZE,
1378 if (ctx->mbuf_pool == NULL)
1379 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool on socket %d\n",
1382 printf("Allocated mbuf pool on socket %d\n", socket_id);
1386 main(int32_t argc, char **argv)
1389 uint32_t lcore_id, nb_ports;
1390 uint8_t portid, socket_id;
1393 ret = rte_eal_init(argc, argv);
1395 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1399 /* parse application arguments (after the EAL ones) */
1400 ret = parse_args(argc, argv);
1402 rte_exit(EXIT_FAILURE, "Invalid parameters\n");
1405 rte_exit(EXIT_FAILURE, "need to choose either EP0 or EP1\n");
1407 if ((unprotected_port_mask & enabled_port_mask) !=
1408 unprotected_port_mask)
1409 rte_exit(EXIT_FAILURE, "Invalid unprotected portmask 0x%x\n",
1410 unprotected_port_mask);
1412 nb_ports = rte_eth_dev_count();
1414 if (check_params() < 0)
1415 rte_exit(EXIT_FAILURE, "check_params failed\n");
1417 ret = init_lcore_rx_queues();
1419 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1421 nb_lcores = rte_lcore_count();
1423 /* Replicate each contex per socket */
1424 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1425 if (rte_lcore_is_enabled(lcore_id) == 0)
1429 socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1433 if (socket_ctx[socket_id].mbuf_pool)
1436 sa_init(&socket_ctx[socket_id], socket_id, ep);
1438 sp4_init(&socket_ctx[socket_id], socket_id, ep);
1440 sp6_init(&socket_ctx[socket_id], socket_id, ep);
1442 rt_init(&socket_ctx[socket_id], socket_id, ep);
1444 pool_init(&socket_ctx[socket_id], socket_id, NB_MBUF);
1447 for (portid = 0; portid < nb_ports; portid++) {
1448 if ((enabled_port_mask & (1 << portid)) == 0)
1457 for (portid = 0; portid < nb_ports; portid++) {
1458 if ((enabled_port_mask & (1 << portid)) == 0)
1462 ret = rte_eth_dev_start(portid);
1464 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: "
1465 "err=%d, port=%d\n", ret, portid);
1467 * If enabled, put device in promiscuous mode.
1468 * This allows IO forwarding mode to forward packets
1469 * to itself through 2 cross-connected ports of the
1473 rte_eth_promiscuous_enable(portid);
1476 check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
1478 /* launch per-lcore init on every lcore */
1479 rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1480 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1481 if (rte_eal_wait_lcore(lcore_id) < 0)