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>
77 #define RTE_LOGTYPE_IPSEC RTE_LOGTYPE_USER1
79 #define MAX_JUMBO_PKT_LEN 9600
81 #define MEMPOOL_CACHE_SIZE 256
83 #define NB_MBUF (32000)
85 #define CDEV_MAP_ENTRIES 1024
86 #define CDEV_MP_NB_OBJS 2048
87 #define CDEV_MP_CACHE_SZ 64
88 #define MAX_QUEUE_PAIRS 1
90 #define OPTION_CONFIG "config"
91 #define OPTION_SINGLE_SA "single-sa"
93 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
97 /* Configure how many packets ahead to prefetch, when reading packets */
98 #define PREFETCH_OFFSET 3
100 #define MAX_RX_QUEUE_PER_LCORE 16
102 #define MAX_LCORE_PARAMS 1024
104 #define UNPROTECTED_PORT(port) (unprotected_port_mask & (1 << portid))
107 * Configurable number of RX/TX ring descriptors
109 #define IPSEC_SECGW_RX_DESC_DEFAULT 128
110 #define IPSEC_SECGW_TX_DESC_DEFAULT 512
111 static uint16_t nb_rxd = IPSEC_SECGW_RX_DESC_DEFAULT;
112 static uint16_t nb_txd = IPSEC_SECGW_TX_DESC_DEFAULT;
114 #if RTE_BYTE_ORDER != RTE_LITTLE_ENDIAN
115 #define __BYTES_TO_UINT64(a, b, c, d, e, f, g, h) \
116 (((uint64_t)((a) & 0xff) << 56) | \
117 ((uint64_t)((b) & 0xff) << 48) | \
118 ((uint64_t)((c) & 0xff) << 40) | \
119 ((uint64_t)((d) & 0xff) << 32) | \
120 ((uint64_t)((e) & 0xff) << 24) | \
121 ((uint64_t)((f) & 0xff) << 16) | \
122 ((uint64_t)((g) & 0xff) << 8) | \
123 ((uint64_t)(h) & 0xff))
125 #define __BYTES_TO_UINT64(a, b, c, d, e, f, g, h) \
126 (((uint64_t)((h) & 0xff) << 56) | \
127 ((uint64_t)((g) & 0xff) << 48) | \
128 ((uint64_t)((f) & 0xff) << 40) | \
129 ((uint64_t)((e) & 0xff) << 32) | \
130 ((uint64_t)((d) & 0xff) << 24) | \
131 ((uint64_t)((c) & 0xff) << 16) | \
132 ((uint64_t)((b) & 0xff) << 8) | \
133 ((uint64_t)(a) & 0xff))
135 #define ETHADDR(a, b, c, d, e, f) (__BYTES_TO_UINT64(a, b, c, d, e, f, 0, 0))
137 #define ETHADDR_TO_UINT64(addr) __BYTES_TO_UINT64( \
138 addr.addr_bytes[0], addr.addr_bytes[1], \
139 addr.addr_bytes[2], addr.addr_bytes[3], \
140 addr.addr_bytes[4], addr.addr_bytes[5], \
143 /* port/source ethernet addr and destination ethernet addr */
144 struct ethaddr_info {
148 struct ethaddr_info ethaddr_tbl[RTE_MAX_ETHPORTS] = {
149 { 0, ETHADDR(0x00, 0x16, 0x3e, 0x7e, 0x94, 0x9a) },
150 { 0, ETHADDR(0x00, 0x16, 0x3e, 0x22, 0xa1, 0xd9) },
151 { 0, ETHADDR(0x00, 0x16, 0x3e, 0x08, 0x69, 0x26) },
152 { 0, ETHADDR(0x00, 0x16, 0x3e, 0x49, 0x9e, 0xdd) }
155 /* mask of enabled ports */
156 static uint32_t enabled_port_mask;
157 static uint32_t unprotected_port_mask;
158 static int32_t promiscuous_on = 1;
159 static int32_t numa_on = 1; /**< NUMA is enabled by default. */
160 static uint32_t nb_lcores;
161 static uint32_t single_sa;
162 static uint32_t single_sa_idx;
164 struct lcore_rx_queue {
167 } __rte_cache_aligned;
169 struct lcore_params {
173 } __rte_cache_aligned;
175 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
177 static struct lcore_params *lcore_params;
178 static uint16_t nb_lcore_params;
180 static struct rte_hash *cdev_map_in;
181 static struct rte_hash *cdev_map_out;
185 struct rte_mbuf *m_table[MAX_PKT_BURST] __rte_aligned(sizeof(void *));
189 uint16_t nb_rx_queue;
190 struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
191 uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
192 struct buffer tx_mbufs[RTE_MAX_ETHPORTS];
193 struct ipsec_ctx inbound;
194 struct ipsec_ctx outbound;
195 struct rt_ctx *rt4_ctx;
196 struct rt_ctx *rt6_ctx;
197 } __rte_cache_aligned;
199 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
201 static struct rte_eth_conf port_conf = {
203 .mq_mode = ETH_MQ_RX_RSS,
204 .max_rx_pkt_len = ETHER_MAX_LEN,
206 .header_split = 0, /**< Header Split disabled */
207 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
208 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
209 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */
210 .hw_strip_crc = 0, /**< CRC stripped by hardware */
215 .rss_hf = ETH_RSS_IP | ETH_RSS_UDP |
216 ETH_RSS_TCP | ETH_RSS_SCTP,
220 .mq_mode = ETH_MQ_TX_NONE,
224 static struct socket_ctx socket_ctx[NB_SOCKETS];
226 struct traffic_type {
227 const uint8_t *data[MAX_PKT_BURST * 2];
228 struct rte_mbuf *pkts[MAX_PKT_BURST * 2];
229 uint32_t res[MAX_PKT_BURST * 2];
233 struct ipsec_traffic {
234 struct traffic_type ipsec;
235 struct traffic_type ip4;
236 struct traffic_type ip6;
240 prepare_one_packet(struct rte_mbuf *pkt, struct ipsec_traffic *t)
243 struct ether_hdr *eth;
245 eth = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
246 if (eth->ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4)) {
247 nlp = (uint8_t *)rte_pktmbuf_adj(pkt, ETHER_HDR_LEN);
248 nlp = RTE_PTR_ADD(nlp, offsetof(struct ip, ip_p));
249 if (*nlp == IPPROTO_ESP)
250 t->ipsec.pkts[(t->ipsec.num)++] = pkt;
252 t->ip4.data[t->ip4.num] = nlp;
253 t->ip4.pkts[(t->ip4.num)++] = pkt;
255 } else if (eth->ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6)) {
256 nlp = (uint8_t *)rte_pktmbuf_adj(pkt, ETHER_HDR_LEN);
257 nlp = RTE_PTR_ADD(nlp, offsetof(struct ip6_hdr, ip6_nxt));
258 if (*nlp == IPPROTO_ESP)
259 t->ipsec.pkts[(t->ipsec.num)++] = pkt;
261 t->ip6.data[t->ip6.num] = nlp;
262 t->ip6.pkts[(t->ip6.num)++] = pkt;
265 /* Unknown/Unsupported type, drop the packet */
266 RTE_LOG(ERR, IPSEC, "Unsupported packet type\n");
267 rte_pktmbuf_free(pkt);
272 prepare_traffic(struct rte_mbuf **pkts, struct ipsec_traffic *t,
281 for (i = 0; i < (nb_pkts - PREFETCH_OFFSET); i++) {
282 rte_prefetch0(rte_pktmbuf_mtod(pkts[i + PREFETCH_OFFSET],
284 prepare_one_packet(pkts[i], t);
286 /* Process left packets */
287 for (; i < nb_pkts; i++)
288 prepare_one_packet(pkts[i], t);
292 prepare_tx_pkt(struct rte_mbuf *pkt, uint8_t port)
295 struct ether_hdr *ethhdr;
297 ip = rte_pktmbuf_mtod(pkt, struct ip *);
299 ethhdr = (struct ether_hdr *)rte_pktmbuf_prepend(pkt, ETHER_HDR_LEN);
301 if (ip->ip_v == IPVERSION) {
302 pkt->ol_flags |= PKT_TX_IP_CKSUM | PKT_TX_IPV4;
303 pkt->l3_len = sizeof(struct ip);
304 pkt->l2_len = ETHER_HDR_LEN;
306 ethhdr->ether_type = rte_cpu_to_be_16(ETHER_TYPE_IPv4);
308 pkt->ol_flags |= PKT_TX_IPV6;
309 pkt->l3_len = sizeof(struct ip6_hdr);
310 pkt->l2_len = ETHER_HDR_LEN;
312 ethhdr->ether_type = rte_cpu_to_be_16(ETHER_TYPE_IPv6);
315 memcpy(ðhdr->s_addr, ðaddr_tbl[port].src,
316 sizeof(struct ether_addr));
317 memcpy(ðhdr->d_addr, ðaddr_tbl[port].dst,
318 sizeof(struct ether_addr));
322 prepare_tx_burst(struct rte_mbuf *pkts[], uint16_t nb_pkts, uint8_t port)
325 const int32_t prefetch_offset = 2;
327 for (i = 0; i < (nb_pkts - prefetch_offset); i++) {
328 rte_mbuf_prefetch_part2(pkts[i + prefetch_offset]);
329 prepare_tx_pkt(pkts[i], port);
331 /* Process left packets */
332 for (; i < nb_pkts; i++)
333 prepare_tx_pkt(pkts[i], port);
336 /* Send burst of packets on an output interface */
337 static inline int32_t
338 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
340 struct rte_mbuf **m_table;
344 queueid = qconf->tx_queue_id[port];
345 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
347 prepare_tx_burst(m_table, n, port);
349 ret = rte_eth_tx_burst(port, queueid, m_table, n);
350 if (unlikely(ret < n)) {
352 rte_pktmbuf_free(m_table[ret]);
359 /* Enqueue a single packet, and send burst if queue is filled */
360 static inline int32_t
361 send_single_packet(struct rte_mbuf *m, uint8_t port)
365 struct lcore_conf *qconf;
367 lcore_id = rte_lcore_id();
369 qconf = &lcore_conf[lcore_id];
370 len = qconf->tx_mbufs[port].len;
371 qconf->tx_mbufs[port].m_table[len] = m;
374 /* enough pkts to be sent */
375 if (unlikely(len == MAX_PKT_BURST)) {
376 send_burst(qconf, MAX_PKT_BURST, port);
380 qconf->tx_mbufs[port].len = len;
385 inbound_sp_sa(struct sp_ctx *sp, struct sa_ctx *sa, struct traffic_type *ip,
389 uint32_t i, j, res, sa_idx;
391 if (ip->num == 0 || sp == NULL)
394 rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
395 ip->num, DEFAULT_MAX_CATEGORIES);
398 for (i = 0; i < ip->num; i++) {
405 if (res & DISCARD || i < lim) {
409 /* Only check SPI match for processed IPSec packets */
410 sa_idx = ip->res[i] & PROTECT_MASK;
411 if (sa_idx == 0 || !inbound_sa_check(sa, m, sa_idx)) {
421 process_pkts_inbound(struct ipsec_ctx *ipsec_ctx,
422 struct ipsec_traffic *traffic)
425 uint16_t idx, nb_pkts_in, i, n_ip4, n_ip6;
427 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
428 traffic->ipsec.num, MAX_PKT_BURST);
430 n_ip4 = traffic->ip4.num;
431 n_ip6 = traffic->ip6.num;
433 /* SP/ACL Inbound check ipsec and ip4 */
434 for (i = 0; i < nb_pkts_in; i++) {
435 m = traffic->ipsec.pkts[i];
436 struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
437 if (ip->ip_v == IPVERSION) {
438 idx = traffic->ip4.num++;
439 traffic->ip4.pkts[idx] = m;
440 traffic->ip4.data[idx] = rte_pktmbuf_mtod_offset(m,
441 uint8_t *, offsetof(struct ip, ip_p));
442 } else if (ip->ip_v == IP6_VERSION) {
443 idx = traffic->ip6.num++;
444 traffic->ip6.pkts[idx] = m;
445 traffic->ip6.data[idx] = rte_pktmbuf_mtod_offset(m,
447 offsetof(struct ip6_hdr, ip6_nxt));
452 inbound_sp_sa(ipsec_ctx->sp4_ctx, ipsec_ctx->sa_ctx, &traffic->ip4,
455 inbound_sp_sa(ipsec_ctx->sp6_ctx, ipsec_ctx->sa_ctx, &traffic->ip6,
460 outbound_sp(struct sp_ctx *sp, struct traffic_type *ip,
461 struct traffic_type *ipsec)
464 uint32_t i, j, sa_idx;
466 if (ip->num == 0 || sp == NULL)
469 rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
470 ip->num, DEFAULT_MAX_CATEGORIES);
473 for (i = 0; i < ip->num; i++) {
475 sa_idx = ip->res[i] & PROTECT_MASK;
476 if ((ip->res[i] == 0) || (ip->res[i] & DISCARD))
478 else if (sa_idx != 0) {
479 ipsec->res[ipsec->num] = sa_idx;
480 ipsec->pkts[ipsec->num++] = m;
488 process_pkts_outbound(struct ipsec_ctx *ipsec_ctx,
489 struct ipsec_traffic *traffic)
492 uint16_t idx, nb_pkts_out, i;
494 /* Drop any IPsec traffic from protected ports */
495 for (i = 0; i < traffic->ipsec.num; i++)
496 rte_pktmbuf_free(traffic->ipsec.pkts[i]);
498 traffic->ipsec.num = 0;
500 outbound_sp(ipsec_ctx->sp4_ctx, &traffic->ip4, &traffic->ipsec);
502 outbound_sp(ipsec_ctx->sp6_ctx, &traffic->ip6, &traffic->ipsec);
504 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
505 traffic->ipsec.res, traffic->ipsec.num,
508 for (i = 0; i < nb_pkts_out; i++) {
509 m = traffic->ipsec.pkts[i];
510 struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
511 if (ip->ip_v == IPVERSION) {
512 idx = traffic->ip4.num++;
513 traffic->ip4.pkts[idx] = m;
515 idx = traffic->ip6.num++;
516 traffic->ip6.pkts[idx] = m;
522 process_pkts_inbound_nosp(struct ipsec_ctx *ipsec_ctx,
523 struct ipsec_traffic *traffic)
526 uint32_t nb_pkts_in, i, idx;
528 /* Drop any IPv4 traffic from unprotected ports */
529 for (i = 0; i < traffic->ip4.num; i++)
530 rte_pktmbuf_free(traffic->ip4.pkts[i]);
532 traffic->ip4.num = 0;
534 /* Drop any IPv6 traffic from unprotected ports */
535 for (i = 0; i < traffic->ip6.num; i++)
536 rte_pktmbuf_free(traffic->ip6.pkts[i]);
538 traffic->ip6.num = 0;
540 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
541 traffic->ipsec.num, MAX_PKT_BURST);
543 for (i = 0; i < nb_pkts_in; i++) {
544 m = traffic->ipsec.pkts[i];
545 struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
546 if (ip->ip_v == IPVERSION) {
547 idx = traffic->ip4.num++;
548 traffic->ip4.pkts[idx] = m;
550 idx = traffic->ip6.num++;
551 traffic->ip6.pkts[idx] = m;
557 process_pkts_outbound_nosp(struct ipsec_ctx *ipsec_ctx,
558 struct ipsec_traffic *traffic)
561 uint32_t nb_pkts_out, i;
564 /* Drop any IPsec traffic from protected ports */
565 for (i = 0; i < traffic->ipsec.num; i++)
566 rte_pktmbuf_free(traffic->ipsec.pkts[i]);
568 traffic->ipsec.num = 0;
570 for (i = 0; i < traffic->ip4.num; i++)
571 traffic->ip4.res[i] = single_sa_idx;
573 for (i = 0; i < traffic->ip6.num; i++)
574 traffic->ip6.res[i] = single_sa_idx;
576 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ip4.pkts,
577 traffic->ip4.res, traffic->ip4.num,
580 /* They all sue the same SA (ip4 or ip6 tunnel) */
581 m = traffic->ipsec.pkts[i];
582 ip = rte_pktmbuf_mtod(m, struct ip *);
583 if (ip->ip_v == IPVERSION)
584 traffic->ip4.num = nb_pkts_out;
586 traffic->ip6.num = nb_pkts_out;
590 route4_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
592 uint32_t hop[MAX_PKT_BURST * 2];
593 uint32_t dst_ip[MAX_PKT_BURST * 2];
599 for (i = 0; i < nb_pkts; i++) {
600 offset = offsetof(struct ip, ip_dst);
601 dst_ip[i] = *rte_pktmbuf_mtod_offset(pkts[i],
603 dst_ip[i] = rte_be_to_cpu_32(dst_ip[i]);
606 rte_lpm_lookup_bulk((struct rte_lpm *)rt_ctx, dst_ip, hop, nb_pkts);
608 for (i = 0; i < nb_pkts; i++) {
609 if ((hop[i] & RTE_LPM_LOOKUP_SUCCESS) == 0) {
610 rte_pktmbuf_free(pkts[i]);
613 send_single_packet(pkts[i], hop[i] & 0xff);
618 route6_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
620 int16_t hop[MAX_PKT_BURST * 2];
621 uint8_t dst_ip[MAX_PKT_BURST * 2][16];
628 for (i = 0; i < nb_pkts; i++) {
629 offset = offsetof(struct ip6_hdr, ip6_dst);
630 ip6_dst = rte_pktmbuf_mtod_offset(pkts[i], uint8_t *, offset);
631 memcpy(&dst_ip[i][0], ip6_dst, 16);
634 rte_lpm6_lookup_bulk_func((struct rte_lpm6 *)rt_ctx, dst_ip,
637 for (i = 0; i < nb_pkts; i++) {
639 rte_pktmbuf_free(pkts[i]);
642 send_single_packet(pkts[i], hop[i] & 0xff);
647 process_pkts(struct lcore_conf *qconf, struct rte_mbuf **pkts,
648 uint8_t nb_pkts, uint8_t portid)
650 struct ipsec_traffic traffic;
652 prepare_traffic(pkts, &traffic, nb_pkts);
654 if (unlikely(single_sa)) {
655 if (UNPROTECTED_PORT(portid))
656 process_pkts_inbound_nosp(&qconf->inbound, &traffic);
658 process_pkts_outbound_nosp(&qconf->outbound, &traffic);
660 if (UNPROTECTED_PORT(portid))
661 process_pkts_inbound(&qconf->inbound, &traffic);
663 process_pkts_outbound(&qconf->outbound, &traffic);
666 route4_pkts(qconf->rt4_ctx, traffic.ip4.pkts, traffic.ip4.num);
667 route6_pkts(qconf->rt6_ctx, traffic.ip6.pkts, traffic.ip6.num);
671 drain_buffers(struct lcore_conf *qconf)
676 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
677 buf = &qconf->tx_mbufs[portid];
680 send_burst(qconf, buf->len, portid);
685 /* main processing loop */
687 main_loop(__attribute__((unused)) void *dummy)
689 struct rte_mbuf *pkts[MAX_PKT_BURST];
691 uint64_t prev_tsc, diff_tsc, cur_tsc;
693 uint8_t portid, queueid;
694 struct lcore_conf *qconf;
696 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
697 / US_PER_S * BURST_TX_DRAIN_US;
698 struct lcore_rx_queue *rxql;
701 lcore_id = rte_lcore_id();
702 qconf = &lcore_conf[lcore_id];
703 rxql = qconf->rx_queue_list;
704 socket_id = rte_lcore_to_socket_id(lcore_id);
706 qconf->rt4_ctx = socket_ctx[socket_id].rt_ip4;
707 qconf->rt6_ctx = socket_ctx[socket_id].rt_ip6;
708 qconf->inbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_in;
709 qconf->inbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_in;
710 qconf->inbound.sa_ctx = socket_ctx[socket_id].sa_in;
711 qconf->inbound.cdev_map = cdev_map_in;
712 qconf->outbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_out;
713 qconf->outbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_out;
714 qconf->outbound.sa_ctx = socket_ctx[socket_id].sa_out;
715 qconf->outbound.cdev_map = cdev_map_out;
717 if (qconf->nb_rx_queue == 0) {
718 RTE_LOG(INFO, IPSEC, "lcore %u has nothing to do\n", lcore_id);
722 RTE_LOG(INFO, IPSEC, "entering main loop on lcore %u\n", lcore_id);
724 for (i = 0; i < qconf->nb_rx_queue; i++) {
725 portid = rxql[i].port_id;
726 queueid = rxql[i].queue_id;
728 " -- lcoreid=%u portid=%hhu rxqueueid=%hhu\n",
729 lcore_id, portid, queueid);
733 cur_tsc = rte_rdtsc();
735 /* TX queue buffer drain */
736 diff_tsc = cur_tsc - prev_tsc;
738 if (unlikely(diff_tsc > drain_tsc)) {
739 drain_buffers(qconf);
743 /* Read packet from RX queues */
744 for (i = 0; i < qconf->nb_rx_queue; ++i) {
745 portid = rxql[i].port_id;
746 queueid = rxql[i].queue_id;
747 nb_rx = rte_eth_rx_burst(portid, queueid,
748 pkts, MAX_PKT_BURST);
751 process_pkts(qconf, pkts, nb_rx, portid);
759 uint8_t lcore, portid, nb_ports;
763 if (lcore_params == NULL) {
764 printf("Error: No port/queue/core mappings\n");
768 nb_ports = rte_eth_dev_count();
770 for (i = 0; i < nb_lcore_params; ++i) {
771 lcore = lcore_params[i].lcore_id;
772 if (!rte_lcore_is_enabled(lcore)) {
773 printf("error: lcore %hhu is not enabled in "
774 "lcore mask\n", lcore);
777 socket_id = rte_lcore_to_socket_id(lcore);
778 if (socket_id != 0 && numa_on == 0) {
779 printf("warning: lcore %hhu is on socket %d "
783 portid = lcore_params[i].port_id;
784 if ((enabled_port_mask & (1 << portid)) == 0) {
785 printf("port %u is not enabled in port mask\n", portid);
788 if (portid >= nb_ports) {
789 printf("port %u is not present on the board\n", portid);
797 get_port_nb_rx_queues(const uint8_t port)
802 for (i = 0; i < nb_lcore_params; ++i) {
803 if (lcore_params[i].port_id == port &&
804 lcore_params[i].queue_id > queue)
805 queue = lcore_params[i].queue_id;
807 return (uint8_t)(++queue);
811 init_lcore_rx_queues(void)
813 uint16_t i, nb_rx_queue;
816 for (i = 0; i < nb_lcore_params; ++i) {
817 lcore = lcore_params[i].lcore_id;
818 nb_rx_queue = lcore_conf[lcore].nb_rx_queue;
819 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
820 printf("error: too many queues (%u) for lcore: %u\n",
821 nb_rx_queue + 1, lcore);
824 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
825 lcore_params[i].port_id;
826 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
827 lcore_params[i].queue_id;
828 lcore_conf[lcore].nb_rx_queue++;
835 print_usage(const char *prgname)
837 printf("%s [EAL options] -- -p PORTMASK -P -u PORTMASK"
838 " --"OPTION_CONFIG" (port,queue,lcore)[,(port,queue,lcore]"
839 " --single-sa SAIDX -f CONFIG_FILE\n"
840 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
841 " -P : enable promiscuous mode\n"
842 " -u PORTMASK: hexadecimal bitmask of unprotected ports\n"
843 " --"OPTION_CONFIG": (port,queue,lcore): "
844 "rx queues configuration\n"
845 " --single-sa SAIDX: use single SA index for outbound, "
847 " -f CONFIG_FILE: Configuration file path\n",
852 parse_portmask(const char *portmask)
857 /* parse hexadecimal string */
858 pm = strtoul(portmask, &end, 16);
859 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
862 if ((pm == 0) && errno)
869 parse_decimal(const char *str)
874 num = strtoul(str, &end, 10);
875 if ((str[0] == '\0') || (end == NULL) || (*end != '\0'))
882 parse_config(const char *q_arg)
885 const char *p, *p0 = q_arg;
893 unsigned long int_fld[_NUM_FLD];
894 char *str_fld[_NUM_FLD];
900 while ((p = strchr(p0, '(')) != NULL) {
907 if (size >= sizeof(s))
910 snprintf(s, sizeof(s), "%.*s", size, p);
911 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
914 for (i = 0; i < _NUM_FLD; i++) {
916 int_fld[i] = strtoul(str_fld[i], &end, 0);
917 if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
920 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
921 printf("exceeded max number of lcore params: %hu\n",
925 lcore_params_array[nb_lcore_params].port_id =
926 (uint8_t)int_fld[FLD_PORT];
927 lcore_params_array[nb_lcore_params].queue_id =
928 (uint8_t)int_fld[FLD_QUEUE];
929 lcore_params_array[nb_lcore_params].lcore_id =
930 (uint8_t)int_fld[FLD_LCORE];
933 lcore_params = lcore_params_array;
937 #define __STRNCMP(name, opt) (!strncmp(name, opt, sizeof(opt)))
939 parse_args_long_options(struct option *lgopts, int32_t option_index)
942 const char *optname = lgopts[option_index].name;
944 if (__STRNCMP(optname, OPTION_CONFIG)) {
945 ret = parse_config(optarg);
947 printf("invalid config\n");
950 if (__STRNCMP(optname, OPTION_SINGLE_SA)) {
951 ret = parse_decimal(optarg);
955 printf("Configured with single SA index %u\n",
966 parse_args(int32_t argc, char **argv)
970 int32_t option_index;
971 char *prgname = argv[0];
972 static struct option lgopts[] = {
973 {OPTION_CONFIG, 1, 0, 0},
974 {OPTION_SINGLE_SA, 1, 0, 0},
977 int32_t f_present = 0;
981 while ((opt = getopt_long(argc, argvopt, "p:Pu:f:",
982 lgopts, &option_index)) != EOF) {
986 enabled_port_mask = parse_portmask(optarg);
987 if (enabled_port_mask == 0) {
988 printf("invalid portmask\n");
989 print_usage(prgname);
994 printf("Promiscuous mode selected\n");
998 unprotected_port_mask = parse_portmask(optarg);
999 if (unprotected_port_mask == 0) {
1000 printf("invalid unprotected portmask\n");
1001 print_usage(prgname);
1006 if (f_present == 1) {
1007 printf("\"-f\" option present more than "
1009 print_usage(prgname);
1012 if (parse_cfg_file(optarg) < 0) {
1013 printf("parsing file \"%s\" failed\n",
1015 print_usage(prgname);
1021 if (parse_args_long_options(lgopts, option_index)) {
1022 print_usage(prgname);
1027 print_usage(prgname);
1032 if (f_present == 0) {
1033 printf("Mandatory option \"-f\" not present\n");
1038 argv[optind-1] = prgname;
1041 optind = 0; /* reset getopt lib */
1046 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1048 char buf[ETHER_ADDR_FMT_SIZE];
1049 ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
1050 printf("%s%s", name, buf);
1053 /* Check the link status of all ports in up to 9s, and print them finally */
1055 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
1057 #define CHECK_INTERVAL 100 /* 100ms */
1058 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1059 uint8_t portid, count, all_ports_up, print_flag = 0;
1060 struct rte_eth_link link;
1062 printf("\nChecking link status");
1064 for (count = 0; count <= MAX_CHECK_TIME; count++) {
1066 for (portid = 0; portid < port_num; portid++) {
1067 if ((port_mask & (1 << portid)) == 0)
1069 memset(&link, 0, sizeof(link));
1070 rte_eth_link_get_nowait(portid, &link);
1071 /* print link status if flag set */
1072 if (print_flag == 1) {
1073 if (link.link_status)
1074 printf("Port %d Link Up - speed %u "
1075 "Mbps - %s\n", (uint8_t)portid,
1076 (uint32_t)link.link_speed,
1077 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1078 ("full-duplex") : ("half-duplex\n"));
1080 printf("Port %d Link Down\n",
1084 /* clear all_ports_up flag if any link down */
1085 if (link.link_status == ETH_LINK_DOWN) {
1090 /* after finally printing all link status, get out */
1091 if (print_flag == 1)
1094 if (all_ports_up == 0) {
1097 rte_delay_ms(CHECK_INTERVAL);
1100 /* set the print_flag if all ports up or timeout */
1101 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1109 add_mapping(struct rte_hash *map, const char *str, uint16_t cdev_id,
1110 uint16_t qp, struct lcore_params *params,
1111 struct ipsec_ctx *ipsec_ctx,
1112 const struct rte_cryptodev_capabilities *cipher,
1113 const struct rte_cryptodev_capabilities *auth)
1117 struct cdev_key key = { 0 };
1119 key.lcore_id = params->lcore_id;
1121 key.cipher_algo = cipher->sym.cipher.algo;
1123 key.auth_algo = auth->sym.auth.algo;
1125 ret = rte_hash_lookup(map, &key);
1129 for (i = 0; i < ipsec_ctx->nb_qps; i++)
1130 if (ipsec_ctx->tbl[i].id == cdev_id)
1133 if (i == ipsec_ctx->nb_qps) {
1134 if (ipsec_ctx->nb_qps == MAX_QP_PER_LCORE) {
1135 printf("Maximum number of crypto devices assigned to "
1136 "a core, increase MAX_QP_PER_LCORE value\n");
1139 ipsec_ctx->tbl[i].id = cdev_id;
1140 ipsec_ctx->tbl[i].qp = qp;
1141 ipsec_ctx->nb_qps++;
1142 printf("%s cdev mapping: lcore %u using cdev %u qp %u "
1143 "(cdev_id_qp %lu)\n", str, key.lcore_id,
1147 ret = rte_hash_add_key_data(map, &key, (void *)i);
1149 printf("Faled to insert cdev mapping for (lcore %u, "
1150 "cdev %u, qp %u), errno %d\n",
1151 key.lcore_id, ipsec_ctx->tbl[i].id,
1152 ipsec_ctx->tbl[i].qp, ret);
1160 add_cdev_mapping(struct rte_cryptodev_info *dev_info, uint16_t cdev_id,
1161 uint16_t qp, struct lcore_params *params)
1164 const struct rte_cryptodev_capabilities *i, *j;
1165 struct rte_hash *map;
1166 struct lcore_conf *qconf;
1167 struct ipsec_ctx *ipsec_ctx;
1170 qconf = &lcore_conf[params->lcore_id];
1172 if ((unprotected_port_mask & (1 << params->port_id)) == 0) {
1174 ipsec_ctx = &qconf->outbound;
1178 ipsec_ctx = &qconf->inbound;
1182 /* Required cryptodevs with operation chainning */
1183 if (!(dev_info->feature_flags &
1184 RTE_CRYPTODEV_FF_SYM_OPERATION_CHAINING))
1187 for (i = dev_info->capabilities;
1188 i->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; i++) {
1189 if (i->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1192 if (i->sym.xform_type != RTE_CRYPTO_SYM_XFORM_CIPHER)
1195 for (j = dev_info->capabilities;
1196 j->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; j++) {
1197 if (j->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1200 if (j->sym.xform_type != RTE_CRYPTO_SYM_XFORM_AUTH)
1203 ret |= add_mapping(map, str, cdev_id, qp, params,
1212 cryptodevs_init(void)
1214 struct rte_cryptodev_config dev_conf;
1215 struct rte_cryptodev_qp_conf qp_conf;
1216 uint16_t idx, max_nb_qps, qp, i;
1218 struct rte_hash_parameters params = { 0 };
1220 params.entries = CDEV_MAP_ENTRIES;
1221 params.key_len = sizeof(struct cdev_key);
1222 params.hash_func = rte_jhash;
1223 params.hash_func_init_val = 0;
1224 params.socket_id = rte_socket_id();
1226 params.name = "cdev_map_in";
1227 cdev_map_in = rte_hash_create(¶ms);
1228 if (cdev_map_in == NULL)
1229 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1232 params.name = "cdev_map_out";
1233 cdev_map_out = rte_hash_create(¶ms);
1234 if (cdev_map_out == NULL)
1235 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1238 printf("lcore/cryptodev/qp mappings:\n");
1241 /* Start from last cdev id to give HW priority */
1242 for (cdev_id = rte_cryptodev_count() - 1; cdev_id >= 0; cdev_id--) {
1243 struct rte_cryptodev_info cdev_info;
1245 rte_cryptodev_info_get(cdev_id, &cdev_info);
1247 if (nb_lcore_params > cdev_info.max_nb_queue_pairs)
1248 max_nb_qps = cdev_info.max_nb_queue_pairs;
1250 max_nb_qps = nb_lcore_params;
1254 while (qp < max_nb_qps && i < nb_lcore_params) {
1255 if (add_cdev_mapping(&cdev_info, cdev_id, qp,
1256 &lcore_params[idx]))
1259 idx = idx % nb_lcore_params;
1266 dev_conf.socket_id = rte_cryptodev_socket_id(cdev_id);
1267 dev_conf.nb_queue_pairs = qp;
1268 dev_conf.session_mp.nb_objs = CDEV_MP_NB_OBJS;
1269 dev_conf.session_mp.cache_size = CDEV_MP_CACHE_SZ;
1271 if (rte_cryptodev_configure(cdev_id, &dev_conf))
1272 rte_panic("Failed to initialize crypodev %u\n",
1275 qp_conf.nb_descriptors = CDEV_MP_NB_OBJS;
1276 for (qp = 0; qp < dev_conf.nb_queue_pairs; qp++)
1277 if (rte_cryptodev_queue_pair_setup(cdev_id, qp,
1278 &qp_conf, dev_conf.socket_id))
1279 rte_panic("Failed to setup queue %u for "
1280 "cdev_id %u\n", 0, cdev_id);
1282 if (rte_cryptodev_start(cdev_id))
1283 rte_panic("Failed to start cryptodev %u\n",
1293 port_init(uint8_t portid)
1295 struct rte_eth_dev_info dev_info;
1296 struct rte_eth_txconf *txconf;
1297 uint16_t nb_tx_queue, nb_rx_queue;
1298 uint16_t tx_queueid, rx_queueid, queue, lcore_id;
1299 int32_t ret, socket_id;
1300 struct lcore_conf *qconf;
1301 struct ether_addr ethaddr;
1303 rte_eth_dev_info_get(portid, &dev_info);
1305 printf("Configuring device port %u:\n", portid);
1307 rte_eth_macaddr_get(portid, ðaddr);
1308 ethaddr_tbl[portid].src = ETHADDR_TO_UINT64(ethaddr);
1309 print_ethaddr("Address: ", ðaddr);
1312 nb_rx_queue = get_port_nb_rx_queues(portid);
1313 nb_tx_queue = nb_lcores;
1315 if (nb_rx_queue > dev_info.max_rx_queues)
1316 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
1317 "(max rx queue is %u)\n",
1318 nb_rx_queue, dev_info.max_rx_queues);
1320 if (nb_tx_queue > dev_info.max_tx_queues)
1321 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
1322 "(max tx queue is %u)\n",
1323 nb_tx_queue, dev_info.max_tx_queues);
1325 printf("Creating queues: nb_rx_queue=%d nb_tx_queue=%u...\n",
1326 nb_rx_queue, nb_tx_queue);
1328 ret = rte_eth_dev_configure(portid, nb_rx_queue, nb_tx_queue,
1331 rte_exit(EXIT_FAILURE, "Cannot configure device: "
1332 "err=%d, port=%d\n", ret, portid);
1334 /* init one TX queue per lcore */
1336 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1337 if (rte_lcore_is_enabled(lcore_id) == 0)
1341 socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1346 printf("Setup txq=%u,%d,%d\n", lcore_id, tx_queueid, socket_id);
1348 txconf = &dev_info.default_txconf;
1349 txconf->txq_flags = 0;
1351 ret = rte_eth_tx_queue_setup(portid, tx_queueid, nb_txd,
1354 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
1355 "err=%d, port=%d\n", ret, portid);
1357 qconf = &lcore_conf[lcore_id];
1358 qconf->tx_queue_id[portid] = tx_queueid;
1361 /* init RX queues */
1362 for (queue = 0; queue < qconf->nb_rx_queue; ++queue) {
1363 if (portid != qconf->rx_queue_list[queue].port_id)
1366 rx_queueid = qconf->rx_queue_list[queue].queue_id;
1368 printf("Setup rxq=%d,%d,%d\n", portid, rx_queueid,
1371 ret = rte_eth_rx_queue_setup(portid, rx_queueid,
1372 nb_rxd, socket_id, NULL,
1373 socket_ctx[socket_id].mbuf_pool);
1375 rte_exit(EXIT_FAILURE,
1376 "rte_eth_rx_queue_setup: err=%d, "
1377 "port=%d\n", ret, portid);
1384 pool_init(struct socket_ctx *ctx, int32_t socket_id, uint32_t nb_mbuf)
1388 snprintf(s, sizeof(s), "mbuf_pool_%d", socket_id);
1389 ctx->mbuf_pool = rte_pktmbuf_pool_create(s, nb_mbuf,
1390 MEMPOOL_CACHE_SIZE, ipsec_metadata_size(),
1391 RTE_MBUF_DEFAULT_BUF_SIZE,
1393 if (ctx->mbuf_pool == NULL)
1394 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool on socket %d\n",
1397 printf("Allocated mbuf pool on socket %d\n", socket_id);
1401 main(int32_t argc, char **argv)
1404 uint32_t lcore_id, nb_ports;
1405 uint8_t portid, socket_id;
1408 ret = rte_eal_init(argc, argv);
1410 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1414 /* parse application arguments (after the EAL ones) */
1415 ret = parse_args(argc, argv);
1417 rte_exit(EXIT_FAILURE, "Invalid parameters\n");
1419 if ((unprotected_port_mask & enabled_port_mask) !=
1420 unprotected_port_mask)
1421 rte_exit(EXIT_FAILURE, "Invalid unprotected portmask 0x%x\n",
1422 unprotected_port_mask);
1424 nb_ports = rte_eth_dev_count();
1426 if (check_params() < 0)
1427 rte_exit(EXIT_FAILURE, "check_params failed\n");
1429 ret = init_lcore_rx_queues();
1431 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1433 nb_lcores = rte_lcore_count();
1435 /* Replicate each contex per socket */
1436 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1437 if (rte_lcore_is_enabled(lcore_id) == 0)
1441 socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1445 if (socket_ctx[socket_id].mbuf_pool)
1448 sa_init(&socket_ctx[socket_id], socket_id);
1450 sp4_init(&socket_ctx[socket_id], socket_id);
1452 sp6_init(&socket_ctx[socket_id], socket_id);
1454 rt_init(&socket_ctx[socket_id], socket_id);
1456 pool_init(&socket_ctx[socket_id], socket_id, NB_MBUF);
1459 for (portid = 0; portid < nb_ports; portid++) {
1460 if ((enabled_port_mask & (1 << portid)) == 0)
1469 for (portid = 0; portid < nb_ports; portid++) {
1470 if ((enabled_port_mask & (1 << portid)) == 0)
1474 ret = rte_eth_dev_start(portid);
1476 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: "
1477 "err=%d, port=%d\n", ret, portid);
1479 * If enabled, put device in promiscuous mode.
1480 * This allows IO forwarding mode to forward packets
1481 * to itself through 2 cross-connected ports of the
1485 rte_eth_promiscuous_enable(portid);
1488 check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
1490 /* launch per-lcore init on every lcore */
1491 rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1492 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1493 if (rte_eal_wait_lcore(lcore_id) < 0)