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>
60 #include <rte_random.h>
61 #include <rte_debug.h>
62 #include <rte_ether.h>
63 #include <rte_ethdev.h>
64 #include <rte_mempool.h>
70 #include <rte_jhash.h>
71 #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_QUEUE_DESC 2048
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;
163 static uint32_t frame_size;
165 struct lcore_rx_queue {
168 } __rte_cache_aligned;
170 struct lcore_params {
174 } __rte_cache_aligned;
176 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
178 static struct lcore_params *lcore_params;
179 static uint16_t nb_lcore_params;
181 static struct rte_hash *cdev_map_in;
182 static struct rte_hash *cdev_map_out;
186 struct rte_mbuf *m_table[MAX_PKT_BURST] __rte_aligned(sizeof(void *));
190 uint16_t nb_rx_queue;
191 struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
192 uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
193 struct buffer tx_mbufs[RTE_MAX_ETHPORTS];
194 struct ipsec_ctx inbound;
195 struct ipsec_ctx outbound;
196 struct rt_ctx *rt4_ctx;
197 struct rt_ctx *rt6_ctx;
198 } __rte_cache_aligned;
200 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
202 static struct rte_eth_conf port_conf = {
204 .mq_mode = ETH_MQ_RX_RSS,
205 .max_rx_pkt_len = ETHER_MAX_LEN,
207 .offloads = DEV_RX_OFFLOAD_CHECKSUM |
208 DEV_RX_OFFLOAD_CRC_STRIP,
209 .ignore_offload_bitfield = 1,
214 .rss_hf = ETH_RSS_IP | ETH_RSS_UDP |
215 ETH_RSS_TCP | ETH_RSS_SCTP,
219 .mq_mode = ETH_MQ_TX_NONE,
223 static struct socket_ctx socket_ctx[NB_SOCKETS];
225 struct traffic_type {
226 const uint8_t *data[MAX_PKT_BURST * 2];
227 struct rte_mbuf *pkts[MAX_PKT_BURST * 2];
228 uint32_t res[MAX_PKT_BURST * 2];
232 struct ipsec_traffic {
233 struct traffic_type ipsec;
234 struct traffic_type ip4;
235 struct traffic_type ip6;
239 prepare_one_packet(struct rte_mbuf *pkt, struct ipsec_traffic *t)
242 struct ether_hdr *eth;
244 eth = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
245 if (eth->ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4)) {
246 nlp = (uint8_t *)rte_pktmbuf_adj(pkt, ETHER_HDR_LEN);
247 nlp = RTE_PTR_ADD(nlp, offsetof(struct ip, ip_p));
248 if (*nlp == IPPROTO_ESP)
249 t->ipsec.pkts[(t->ipsec.num)++] = pkt;
251 t->ip4.data[t->ip4.num] = nlp;
252 t->ip4.pkts[(t->ip4.num)++] = pkt;
254 } else if (eth->ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6)) {
255 nlp = (uint8_t *)rte_pktmbuf_adj(pkt, ETHER_HDR_LEN);
256 nlp = RTE_PTR_ADD(nlp, offsetof(struct ip6_hdr, ip6_nxt));
257 if (*nlp == IPPROTO_ESP)
258 t->ipsec.pkts[(t->ipsec.num)++] = pkt;
260 t->ip6.data[t->ip6.num] = nlp;
261 t->ip6.pkts[(t->ip6.num)++] = pkt;
264 /* Unknown/Unsupported type, drop the packet */
265 RTE_LOG(ERR, IPSEC, "Unsupported packet type\n");
266 rte_pktmbuf_free(pkt);
271 prepare_traffic(struct rte_mbuf **pkts, struct ipsec_traffic *t,
280 for (i = 0; i < (nb_pkts - PREFETCH_OFFSET); i++) {
281 rte_prefetch0(rte_pktmbuf_mtod(pkts[i + PREFETCH_OFFSET],
283 prepare_one_packet(pkts[i], t);
285 /* Process left packets */
286 for (; i < nb_pkts; i++)
287 prepare_one_packet(pkts[i], t);
291 prepare_tx_pkt(struct rte_mbuf *pkt, uint16_t port)
294 struct ether_hdr *ethhdr;
296 ip = rte_pktmbuf_mtod(pkt, struct ip *);
298 ethhdr = (struct ether_hdr *)rte_pktmbuf_prepend(pkt, ETHER_HDR_LEN);
300 if (ip->ip_v == IPVERSION) {
301 pkt->ol_flags |= PKT_TX_IP_CKSUM | PKT_TX_IPV4;
302 pkt->l3_len = sizeof(struct ip);
303 pkt->l2_len = ETHER_HDR_LEN;
305 ethhdr->ether_type = rte_cpu_to_be_16(ETHER_TYPE_IPv4);
307 pkt->ol_flags |= PKT_TX_IPV6;
308 pkt->l3_len = sizeof(struct ip6_hdr);
309 pkt->l2_len = ETHER_HDR_LEN;
311 ethhdr->ether_type = rte_cpu_to_be_16(ETHER_TYPE_IPv6);
314 memcpy(ðhdr->s_addr, ðaddr_tbl[port].src,
315 sizeof(struct ether_addr));
316 memcpy(ðhdr->d_addr, ðaddr_tbl[port].dst,
317 sizeof(struct ether_addr));
321 prepare_tx_burst(struct rte_mbuf *pkts[], uint16_t nb_pkts, uint16_t port)
324 const int32_t prefetch_offset = 2;
326 for (i = 0; i < (nb_pkts - prefetch_offset); i++) {
327 rte_mbuf_prefetch_part2(pkts[i + prefetch_offset]);
328 prepare_tx_pkt(pkts[i], port);
330 /* Process left packets */
331 for (; i < nb_pkts; i++)
332 prepare_tx_pkt(pkts[i], port);
335 /* Send burst of packets on an output interface */
336 static inline int32_t
337 send_burst(struct lcore_conf *qconf, uint16_t n, uint16_t port)
339 struct rte_mbuf **m_table;
343 queueid = qconf->tx_queue_id[port];
344 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
346 prepare_tx_burst(m_table, n, port);
348 ret = rte_eth_tx_burst(port, queueid, m_table, n);
349 if (unlikely(ret < n)) {
351 rte_pktmbuf_free(m_table[ret]);
358 /* Enqueue a single packet, and send burst if queue is filled */
359 static inline int32_t
360 send_single_packet(struct rte_mbuf *m, uint16_t port)
364 struct lcore_conf *qconf;
366 lcore_id = rte_lcore_id();
368 qconf = &lcore_conf[lcore_id];
369 len = qconf->tx_mbufs[port].len;
370 qconf->tx_mbufs[port].m_table[len] = m;
373 /* enough pkts to be sent */
374 if (unlikely(len == MAX_PKT_BURST)) {
375 send_burst(qconf, MAX_PKT_BURST, port);
379 qconf->tx_mbufs[port].len = len;
384 inbound_sp_sa(struct sp_ctx *sp, struct sa_ctx *sa, struct traffic_type *ip,
388 uint32_t i, j, res, sa_idx;
390 if (ip->num == 0 || sp == NULL)
393 rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
394 ip->num, DEFAULT_MAX_CATEGORIES);
397 for (i = 0; i < ip->num; i++) {
404 if (res & DISCARD || i < lim) {
408 /* Only check SPI match for processed IPSec packets */
409 sa_idx = ip->res[i] & PROTECT_MASK;
410 if (sa_idx == 0 || !inbound_sa_check(sa, m, sa_idx)) {
420 process_pkts_inbound(struct ipsec_ctx *ipsec_ctx,
421 struct ipsec_traffic *traffic)
424 uint16_t idx, nb_pkts_in, i, n_ip4, n_ip6;
426 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
427 traffic->ipsec.num, MAX_PKT_BURST);
429 n_ip4 = traffic->ip4.num;
430 n_ip6 = traffic->ip6.num;
432 /* SP/ACL Inbound check ipsec and ip4 */
433 for (i = 0; i < nb_pkts_in; i++) {
434 m = traffic->ipsec.pkts[i];
435 struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
436 if (ip->ip_v == IPVERSION) {
437 idx = traffic->ip4.num++;
438 traffic->ip4.pkts[idx] = m;
439 traffic->ip4.data[idx] = rte_pktmbuf_mtod_offset(m,
440 uint8_t *, offsetof(struct ip, ip_p));
441 } else if (ip->ip_v == IP6_VERSION) {
442 idx = traffic->ip6.num++;
443 traffic->ip6.pkts[idx] = m;
444 traffic->ip6.data[idx] = rte_pktmbuf_mtod_offset(m,
446 offsetof(struct ip6_hdr, ip6_nxt));
451 inbound_sp_sa(ipsec_ctx->sp4_ctx, ipsec_ctx->sa_ctx, &traffic->ip4,
454 inbound_sp_sa(ipsec_ctx->sp6_ctx, ipsec_ctx->sa_ctx, &traffic->ip6,
459 outbound_sp(struct sp_ctx *sp, struct traffic_type *ip,
460 struct traffic_type *ipsec)
463 uint32_t i, j, sa_idx;
465 if (ip->num == 0 || sp == NULL)
468 rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
469 ip->num, DEFAULT_MAX_CATEGORIES);
472 for (i = 0; i < ip->num; i++) {
474 sa_idx = ip->res[i] & PROTECT_MASK;
475 if ((ip->res[i] == 0) || (ip->res[i] & DISCARD))
477 else if (sa_idx != 0) {
478 ipsec->res[ipsec->num] = sa_idx;
479 ipsec->pkts[ipsec->num++] = m;
487 process_pkts_outbound(struct ipsec_ctx *ipsec_ctx,
488 struct ipsec_traffic *traffic)
491 uint16_t idx, nb_pkts_out, i;
493 /* Drop any IPsec traffic from protected ports */
494 for (i = 0; i < traffic->ipsec.num; i++)
495 rte_pktmbuf_free(traffic->ipsec.pkts[i]);
497 traffic->ipsec.num = 0;
499 outbound_sp(ipsec_ctx->sp4_ctx, &traffic->ip4, &traffic->ipsec);
501 outbound_sp(ipsec_ctx->sp6_ctx, &traffic->ip6, &traffic->ipsec);
503 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
504 traffic->ipsec.res, traffic->ipsec.num,
507 for (i = 0; i < nb_pkts_out; i++) {
508 m = traffic->ipsec.pkts[i];
509 struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
510 if (ip->ip_v == IPVERSION) {
511 idx = traffic->ip4.num++;
512 traffic->ip4.pkts[idx] = m;
514 idx = traffic->ip6.num++;
515 traffic->ip6.pkts[idx] = m;
521 process_pkts_inbound_nosp(struct ipsec_ctx *ipsec_ctx,
522 struct ipsec_traffic *traffic)
525 uint32_t nb_pkts_in, i, idx;
527 /* Drop any IPv4 traffic from unprotected ports */
528 for (i = 0; i < traffic->ip4.num; i++)
529 rte_pktmbuf_free(traffic->ip4.pkts[i]);
531 traffic->ip4.num = 0;
533 /* Drop any IPv6 traffic from unprotected ports */
534 for (i = 0; i < traffic->ip6.num; i++)
535 rte_pktmbuf_free(traffic->ip6.pkts[i]);
537 traffic->ip6.num = 0;
539 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
540 traffic->ipsec.num, MAX_PKT_BURST);
542 for (i = 0; i < nb_pkts_in; i++) {
543 m = traffic->ipsec.pkts[i];
544 struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
545 if (ip->ip_v == IPVERSION) {
546 idx = traffic->ip4.num++;
547 traffic->ip4.pkts[idx] = m;
549 idx = traffic->ip6.num++;
550 traffic->ip6.pkts[idx] = m;
556 process_pkts_outbound_nosp(struct ipsec_ctx *ipsec_ctx,
557 struct ipsec_traffic *traffic)
560 uint32_t nb_pkts_out, i;
563 /* Drop any IPsec traffic from protected ports */
564 for (i = 0; i < traffic->ipsec.num; i++)
565 rte_pktmbuf_free(traffic->ipsec.pkts[i]);
567 traffic->ipsec.num = 0;
569 for (i = 0; i < traffic->ip4.num; i++)
570 traffic->ip4.res[i] = single_sa_idx;
572 for (i = 0; i < traffic->ip6.num; i++)
573 traffic->ip6.res[i] = single_sa_idx;
575 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ip4.pkts,
576 traffic->ip4.res, traffic->ip4.num,
579 /* They all sue the same SA (ip4 or ip6 tunnel) */
580 m = traffic->ipsec.pkts[i];
581 ip = rte_pktmbuf_mtod(m, struct ip *);
582 if (ip->ip_v == IPVERSION)
583 traffic->ip4.num = nb_pkts_out;
585 traffic->ip6.num = nb_pkts_out;
589 route4_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
591 uint32_t hop[MAX_PKT_BURST * 2];
592 uint32_t dst_ip[MAX_PKT_BURST * 2];
598 for (i = 0; i < nb_pkts; i++) {
599 offset = offsetof(struct ip, ip_dst);
600 dst_ip[i] = *rte_pktmbuf_mtod_offset(pkts[i],
602 dst_ip[i] = rte_be_to_cpu_32(dst_ip[i]);
605 rte_lpm_lookup_bulk((struct rte_lpm *)rt_ctx, dst_ip, hop, nb_pkts);
607 for (i = 0; i < nb_pkts; i++) {
608 if ((hop[i] & RTE_LPM_LOOKUP_SUCCESS) == 0) {
609 rte_pktmbuf_free(pkts[i]);
612 send_single_packet(pkts[i], hop[i] & 0xff);
617 route6_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
619 int32_t hop[MAX_PKT_BURST * 2];
620 uint8_t dst_ip[MAX_PKT_BURST * 2][16];
627 for (i = 0; i < nb_pkts; i++) {
628 offset = offsetof(struct ip6_hdr, ip6_dst);
629 ip6_dst = rte_pktmbuf_mtod_offset(pkts[i], uint8_t *, offset);
630 memcpy(&dst_ip[i][0], ip6_dst, 16);
633 rte_lpm6_lookup_bulk_func((struct rte_lpm6 *)rt_ctx, dst_ip,
636 for (i = 0; i < nb_pkts; i++) {
638 rte_pktmbuf_free(pkts[i]);
641 send_single_packet(pkts[i], hop[i] & 0xff);
646 process_pkts(struct lcore_conf *qconf, struct rte_mbuf **pkts,
647 uint8_t nb_pkts, uint16_t portid)
649 struct ipsec_traffic traffic;
651 prepare_traffic(pkts, &traffic, nb_pkts);
653 if (unlikely(single_sa)) {
654 if (UNPROTECTED_PORT(portid))
655 process_pkts_inbound_nosp(&qconf->inbound, &traffic);
657 process_pkts_outbound_nosp(&qconf->outbound, &traffic);
659 if (UNPROTECTED_PORT(portid))
660 process_pkts_inbound(&qconf->inbound, &traffic);
662 process_pkts_outbound(&qconf->outbound, &traffic);
665 route4_pkts(qconf->rt4_ctx, traffic.ip4.pkts, traffic.ip4.num);
666 route6_pkts(qconf->rt6_ctx, traffic.ip6.pkts, traffic.ip6.num);
670 drain_buffers(struct lcore_conf *qconf)
675 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
676 buf = &qconf->tx_mbufs[portid];
679 send_burst(qconf, buf->len, portid);
684 /* main processing loop */
686 main_loop(__attribute__((unused)) void *dummy)
688 struct rte_mbuf *pkts[MAX_PKT_BURST];
690 uint64_t prev_tsc, diff_tsc, cur_tsc;
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->inbound.session_pool = socket_ctx[socket_id].session_pool;
713 qconf->outbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_out;
714 qconf->outbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_out;
715 qconf->outbound.sa_ctx = socket_ctx[socket_id].sa_out;
716 qconf->outbound.cdev_map = cdev_map_out;
717 qconf->outbound.session_pool = socket_ctx[socket_id].session_pool;
719 if (qconf->nb_rx_queue == 0) {
720 RTE_LOG(INFO, IPSEC, "lcore %u has nothing to do\n", lcore_id);
724 RTE_LOG(INFO, IPSEC, "entering main loop on lcore %u\n", lcore_id);
726 for (i = 0; i < qconf->nb_rx_queue; i++) {
727 portid = rxql[i].port_id;
728 queueid = rxql[i].queue_id;
730 " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
731 lcore_id, portid, queueid);
735 cur_tsc = rte_rdtsc();
737 /* TX queue buffer drain */
738 diff_tsc = cur_tsc - prev_tsc;
740 if (unlikely(diff_tsc > drain_tsc)) {
741 drain_buffers(qconf);
745 /* Read packet from RX queues */
746 for (i = 0; i < qconf->nb_rx_queue; ++i) {
747 portid = rxql[i].port_id;
748 queueid = rxql[i].queue_id;
749 nb_rx = rte_eth_rx_burst(portid, queueid,
750 pkts, MAX_PKT_BURST);
753 process_pkts(qconf, pkts, nb_rx, portid);
762 uint16_t portid, nb_ports;
766 if (lcore_params == NULL) {
767 printf("Error: No port/queue/core mappings\n");
771 nb_ports = rte_eth_dev_count();
773 for (i = 0; i < nb_lcore_params; ++i) {
774 lcore = lcore_params[i].lcore_id;
775 if (!rte_lcore_is_enabled(lcore)) {
776 printf("error: lcore %hhu is not enabled in "
777 "lcore mask\n", lcore);
780 socket_id = rte_lcore_to_socket_id(lcore);
781 if (socket_id != 0 && numa_on == 0) {
782 printf("warning: lcore %hhu is on socket %d "
786 portid = lcore_params[i].port_id;
787 if ((enabled_port_mask & (1 << portid)) == 0) {
788 printf("port %u is not enabled in port mask\n", portid);
791 if (portid >= nb_ports) {
792 printf("port %u is not present on the board\n", portid);
800 get_port_nb_rx_queues(const uint16_t port)
805 for (i = 0; i < nb_lcore_params; ++i) {
806 if (lcore_params[i].port_id == port &&
807 lcore_params[i].queue_id > queue)
808 queue = lcore_params[i].queue_id;
810 return (uint8_t)(++queue);
814 init_lcore_rx_queues(void)
816 uint16_t i, nb_rx_queue;
819 for (i = 0; i < nb_lcore_params; ++i) {
820 lcore = lcore_params[i].lcore_id;
821 nb_rx_queue = lcore_conf[lcore].nb_rx_queue;
822 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
823 printf("error: too many queues (%u) for lcore: %u\n",
824 nb_rx_queue + 1, lcore);
827 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
828 lcore_params[i].port_id;
829 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
830 lcore_params[i].queue_id;
831 lcore_conf[lcore].nb_rx_queue++;
838 print_usage(const char *prgname)
840 printf("%s [EAL options] -- -p PORTMASK -P -u PORTMASK"
841 " --"OPTION_CONFIG" (port,queue,lcore)[,(port,queue,lcore]"
842 " --single-sa SAIDX -f CONFIG_FILE\n"
843 " -p PORTMASK: hexadecimal bitmask of ports to configure\n"
844 " -P : enable promiscuous mode\n"
845 " -u PORTMASK: hexadecimal bitmask of unprotected ports\n"
846 " -j FRAMESIZE: jumbo frame maximum size\n"
847 " --"OPTION_CONFIG": (port,queue,lcore): "
848 "rx queues configuration\n"
849 " --single-sa SAIDX: use single SA index for outbound, "
851 " -f CONFIG_FILE: Configuration file path\n",
856 parse_portmask(const char *portmask)
861 /* parse hexadecimal string */
862 pm = strtoul(portmask, &end, 16);
863 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
866 if ((pm == 0) && errno)
873 parse_decimal(const char *str)
878 num = strtoul(str, &end, 10);
879 if ((str[0] == '\0') || (end == NULL) || (*end != '\0'))
886 parse_config(const char *q_arg)
889 const char *p, *p0 = q_arg;
897 unsigned long int_fld[_NUM_FLD];
898 char *str_fld[_NUM_FLD];
904 while ((p = strchr(p0, '(')) != NULL) {
911 if (size >= sizeof(s))
914 snprintf(s, sizeof(s), "%.*s", size, p);
915 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
918 for (i = 0; i < _NUM_FLD; i++) {
920 int_fld[i] = strtoul(str_fld[i], &end, 0);
921 if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
924 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
925 printf("exceeded max number of lcore params: %hu\n",
929 lcore_params_array[nb_lcore_params].port_id =
930 (uint8_t)int_fld[FLD_PORT];
931 lcore_params_array[nb_lcore_params].queue_id =
932 (uint8_t)int_fld[FLD_QUEUE];
933 lcore_params_array[nb_lcore_params].lcore_id =
934 (uint8_t)int_fld[FLD_LCORE];
937 lcore_params = lcore_params_array;
941 #define __STRNCMP(name, opt) (!strncmp(name, opt, sizeof(opt)))
943 parse_args_long_options(struct option *lgopts, int32_t option_index)
946 const char *optname = lgopts[option_index].name;
948 if (__STRNCMP(optname, OPTION_CONFIG)) {
949 ret = parse_config(optarg);
951 printf("invalid config\n");
954 if (__STRNCMP(optname, OPTION_SINGLE_SA)) {
955 ret = parse_decimal(optarg);
959 printf("Configured with single SA index %u\n",
970 parse_args(int32_t argc, char **argv)
974 int32_t option_index;
975 char *prgname = argv[0];
976 static struct option lgopts[] = {
977 {OPTION_CONFIG, 1, 0, 0},
978 {OPTION_SINGLE_SA, 1, 0, 0},
981 int32_t f_present = 0;
985 while ((opt = getopt_long(argc, argvopt, "p:Pu:f:j:",
986 lgopts, &option_index)) != EOF) {
990 enabled_port_mask = parse_portmask(optarg);
991 if (enabled_port_mask == 0) {
992 printf("invalid portmask\n");
993 print_usage(prgname);
998 printf("Promiscuous mode selected\n");
1002 unprotected_port_mask = parse_portmask(optarg);
1003 if (unprotected_port_mask == 0) {
1004 printf("invalid unprotected portmask\n");
1005 print_usage(prgname);
1010 if (f_present == 1) {
1011 printf("\"-f\" option present more than "
1013 print_usage(prgname);
1016 if (parse_cfg_file(optarg) < 0) {
1017 printf("parsing file \"%s\" failed\n",
1019 print_usage(prgname);
1026 int32_t size = parse_decimal(optarg);
1028 printf("Invalid jumbo frame size\n");
1030 print_usage(prgname);
1033 printf("Using default value 9000\n");
1039 printf("Enabled jumbo frames size %u\n", frame_size);
1042 if (parse_args_long_options(lgopts, option_index)) {
1043 print_usage(prgname);
1048 print_usage(prgname);
1053 if (f_present == 0) {
1054 printf("Mandatory option \"-f\" not present\n");
1059 argv[optind-1] = prgname;
1062 optind = 1; /* reset getopt lib */
1067 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1069 char buf[ETHER_ADDR_FMT_SIZE];
1070 ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
1071 printf("%s%s", name, buf);
1074 /* Check the link status of all ports in up to 9s, and print them finally */
1076 check_all_ports_link_status(uint16_t port_num, uint32_t port_mask)
1078 #define CHECK_INTERVAL 100 /* 100ms */
1079 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1081 uint8_t count, all_ports_up, print_flag = 0;
1082 struct rte_eth_link link;
1084 printf("\nChecking link status");
1086 for (count = 0; count <= MAX_CHECK_TIME; count++) {
1088 for (portid = 0; portid < port_num; portid++) {
1089 if ((port_mask & (1 << portid)) == 0)
1091 memset(&link, 0, sizeof(link));
1092 rte_eth_link_get_nowait(portid, &link);
1093 /* print link status if flag set */
1094 if (print_flag == 1) {
1095 if (link.link_status)
1097 "Port%d Link Up - speed %u Mbps -%s\n",
1098 portid, link.link_speed,
1099 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1100 ("full-duplex") : ("half-duplex\n"));
1102 printf("Port %d Link Down\n", portid);
1105 /* clear all_ports_up flag if any link down */
1106 if (link.link_status == ETH_LINK_DOWN) {
1111 /* after finally printing all link status, get out */
1112 if (print_flag == 1)
1115 if (all_ports_up == 0) {
1118 rte_delay_ms(CHECK_INTERVAL);
1121 /* set the print_flag if all ports up or timeout */
1122 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1130 add_mapping(struct rte_hash *map, const char *str, uint16_t cdev_id,
1131 uint16_t qp, struct lcore_params *params,
1132 struct ipsec_ctx *ipsec_ctx,
1133 const struct rte_cryptodev_capabilities *cipher,
1134 const struct rte_cryptodev_capabilities *auth,
1135 const struct rte_cryptodev_capabilities *aead)
1139 struct cdev_key key = { 0 };
1141 key.lcore_id = params->lcore_id;
1143 key.cipher_algo = cipher->sym.cipher.algo;
1145 key.auth_algo = auth->sym.auth.algo;
1147 key.aead_algo = aead->sym.aead.algo;
1149 ret = rte_hash_lookup(map, &key);
1153 for (i = 0; i < ipsec_ctx->nb_qps; i++)
1154 if (ipsec_ctx->tbl[i].id == cdev_id)
1157 if (i == ipsec_ctx->nb_qps) {
1158 if (ipsec_ctx->nb_qps == MAX_QP_PER_LCORE) {
1159 printf("Maximum number of crypto devices assigned to "
1160 "a core, increase MAX_QP_PER_LCORE value\n");
1163 ipsec_ctx->tbl[i].id = cdev_id;
1164 ipsec_ctx->tbl[i].qp = qp;
1165 ipsec_ctx->nb_qps++;
1166 printf("%s cdev mapping: lcore %u using cdev %u qp %u "
1167 "(cdev_id_qp %lu)\n", str, key.lcore_id,
1171 ret = rte_hash_add_key_data(map, &key, (void *)i);
1173 printf("Faled to insert cdev mapping for (lcore %u, "
1174 "cdev %u, qp %u), errno %d\n",
1175 key.lcore_id, ipsec_ctx->tbl[i].id,
1176 ipsec_ctx->tbl[i].qp, ret);
1184 add_cdev_mapping(struct rte_cryptodev_info *dev_info, uint16_t cdev_id,
1185 uint16_t qp, struct lcore_params *params)
1188 const struct rte_cryptodev_capabilities *i, *j;
1189 struct rte_hash *map;
1190 struct lcore_conf *qconf;
1191 struct ipsec_ctx *ipsec_ctx;
1194 qconf = &lcore_conf[params->lcore_id];
1196 if ((unprotected_port_mask & (1 << params->port_id)) == 0) {
1198 ipsec_ctx = &qconf->outbound;
1202 ipsec_ctx = &qconf->inbound;
1206 /* Required cryptodevs with operation chainning */
1207 if (!(dev_info->feature_flags &
1208 RTE_CRYPTODEV_FF_SYM_OPERATION_CHAINING))
1211 for (i = dev_info->capabilities;
1212 i->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; i++) {
1213 if (i->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1216 if (i->sym.xform_type == RTE_CRYPTO_SYM_XFORM_AEAD) {
1217 ret |= add_mapping(map, str, cdev_id, qp, params,
1218 ipsec_ctx, NULL, NULL, i);
1222 if (i->sym.xform_type != RTE_CRYPTO_SYM_XFORM_CIPHER)
1225 for (j = dev_info->capabilities;
1226 j->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; j++) {
1227 if (j->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1230 if (j->sym.xform_type != RTE_CRYPTO_SYM_XFORM_AUTH)
1233 ret |= add_mapping(map, str, cdev_id, qp, params,
1234 ipsec_ctx, i, j, NULL);
1242 cryptodevs_init(void)
1244 struct rte_cryptodev_config dev_conf;
1245 struct rte_cryptodev_qp_conf qp_conf;
1246 uint16_t idx, max_nb_qps, qp, i;
1248 struct rte_hash_parameters params = { 0 };
1250 params.entries = CDEV_MAP_ENTRIES;
1251 params.key_len = sizeof(struct cdev_key);
1252 params.hash_func = rte_jhash;
1253 params.hash_func_init_val = 0;
1254 params.socket_id = rte_socket_id();
1256 params.name = "cdev_map_in";
1257 cdev_map_in = rte_hash_create(¶ms);
1258 if (cdev_map_in == NULL)
1259 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1262 params.name = "cdev_map_out";
1263 cdev_map_out = rte_hash_create(¶ms);
1264 if (cdev_map_out == NULL)
1265 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1268 printf("lcore/cryptodev/qp mappings:\n");
1270 uint32_t max_sess_sz = 0, sess_sz;
1271 for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
1272 sess_sz = rte_cryptodev_get_private_session_size(cdev_id);
1273 if (sess_sz > max_sess_sz)
1274 max_sess_sz = sess_sz;
1278 /* Start from last cdev id to give HW priority */
1279 for (cdev_id = rte_cryptodev_count() - 1; cdev_id >= 0; cdev_id--) {
1280 struct rte_cryptodev_info cdev_info;
1282 rte_cryptodev_info_get(cdev_id, &cdev_info);
1284 if (nb_lcore_params > cdev_info.max_nb_queue_pairs)
1285 max_nb_qps = cdev_info.max_nb_queue_pairs;
1287 max_nb_qps = nb_lcore_params;
1291 while (qp < max_nb_qps && i < nb_lcore_params) {
1292 if (add_cdev_mapping(&cdev_info, cdev_id, qp,
1293 &lcore_params[idx]))
1296 idx = idx % nb_lcore_params;
1303 dev_conf.socket_id = rte_cryptodev_socket_id(cdev_id);
1304 dev_conf.nb_queue_pairs = qp;
1306 if (!socket_ctx[dev_conf.socket_id].session_pool) {
1307 char mp_name[RTE_MEMPOOL_NAMESIZE];
1308 struct rte_mempool *sess_mp;
1310 snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
1311 "sess_mp_%u", dev_conf.socket_id);
1312 sess_mp = rte_mempool_create(mp_name,
1316 0, NULL, NULL, NULL,
1317 NULL, dev_conf.socket_id,
1319 if (sess_mp == NULL)
1320 rte_exit(EXIT_FAILURE,
1321 "Cannot create session pool on socket %d\n",
1322 dev_conf.socket_id);
1324 printf("Allocated session pool on socket %d\n",
1325 dev_conf.socket_id);
1326 socket_ctx[dev_conf.socket_id].session_pool = sess_mp;
1329 if (rte_cryptodev_configure(cdev_id, &dev_conf))
1330 rte_panic("Failed to initialize cryptodev %u\n",
1333 qp_conf.nb_descriptors = CDEV_QUEUE_DESC;
1334 for (qp = 0; qp < dev_conf.nb_queue_pairs; qp++)
1335 if (rte_cryptodev_queue_pair_setup(cdev_id, qp,
1336 &qp_conf, dev_conf.socket_id,
1337 socket_ctx[dev_conf.socket_id].session_pool))
1338 rte_panic("Failed to setup queue %u for "
1339 "cdev_id %u\n", 0, cdev_id);
1341 if (rte_cryptodev_start(cdev_id))
1342 rte_panic("Failed to start cryptodev %u\n",
1352 port_init(uint16_t portid)
1354 struct rte_eth_dev_info dev_info;
1355 struct rte_eth_txconf *txconf;
1356 uint16_t nb_tx_queue, nb_rx_queue;
1357 uint16_t tx_queueid, rx_queueid, queue, lcore_id;
1358 int32_t ret, socket_id;
1359 struct lcore_conf *qconf;
1360 struct ether_addr ethaddr;
1362 rte_eth_dev_info_get(portid, &dev_info);
1364 printf("Configuring device port %u:\n", portid);
1366 rte_eth_macaddr_get(portid, ðaddr);
1367 ethaddr_tbl[portid].src = ETHADDR_TO_UINT64(ethaddr);
1368 print_ethaddr("Address: ", ðaddr);
1371 nb_rx_queue = get_port_nb_rx_queues(portid);
1372 nb_tx_queue = nb_lcores;
1374 if (nb_rx_queue > dev_info.max_rx_queues)
1375 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
1376 "(max rx queue is %u)\n",
1377 nb_rx_queue, dev_info.max_rx_queues);
1379 if (nb_tx_queue > dev_info.max_tx_queues)
1380 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
1381 "(max tx queue is %u)\n",
1382 nb_tx_queue, dev_info.max_tx_queues);
1384 printf("Creating queues: nb_rx_queue=%d nb_tx_queue=%u...\n",
1385 nb_rx_queue, nb_tx_queue);
1388 port_conf.rxmode.max_rx_pkt_len = frame_size;
1389 port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
1392 if (dev_info.rx_offload_capa & DEV_RX_OFFLOAD_SECURITY)
1393 port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_SECURITY;
1394 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_SECURITY)
1395 port_conf.txmode.offloads |= DEV_TX_OFFLOAD_SECURITY;
1397 ret = rte_eth_dev_configure(portid, nb_rx_queue, nb_tx_queue,
1400 rte_exit(EXIT_FAILURE, "Cannot configure device: "
1401 "err=%d, port=%d\n", ret, portid);
1403 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd, &nb_txd);
1405 rte_exit(EXIT_FAILURE, "Cannot adjust number of descriptors: "
1406 "err=%d, port=%d\n", ret, portid);
1408 /* init one TX queue per lcore */
1410 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1411 if (rte_lcore_is_enabled(lcore_id) == 0)
1415 socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1420 printf("Setup txq=%u,%d,%d\n", lcore_id, tx_queueid, socket_id);
1422 txconf = &dev_info.default_txconf;
1423 txconf->txq_flags = 0;
1425 ret = rte_eth_tx_queue_setup(portid, tx_queueid, nb_txd,
1428 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
1429 "err=%d, port=%d\n", ret, portid);
1431 qconf = &lcore_conf[lcore_id];
1432 qconf->tx_queue_id[portid] = tx_queueid;
1435 /* init RX queues */
1436 for (queue = 0; queue < qconf->nb_rx_queue; ++queue) {
1437 if (portid != qconf->rx_queue_list[queue].port_id)
1440 rx_queueid = qconf->rx_queue_list[queue].queue_id;
1442 printf("Setup rxq=%d,%d,%d\n", portid, rx_queueid,
1445 ret = rte_eth_rx_queue_setup(portid, rx_queueid,
1446 nb_rxd, socket_id, NULL,
1447 socket_ctx[socket_id].mbuf_pool);
1449 rte_exit(EXIT_FAILURE,
1450 "rte_eth_rx_queue_setup: err=%d, "
1451 "port=%d\n", ret, portid);
1458 pool_init(struct socket_ctx *ctx, int32_t socket_id, uint32_t nb_mbuf)
1461 uint32_t buff_size = frame_size ? (frame_size + RTE_PKTMBUF_HEADROOM) :
1462 RTE_MBUF_DEFAULT_BUF_SIZE;
1465 snprintf(s, sizeof(s), "mbuf_pool_%d", socket_id);
1466 ctx->mbuf_pool = rte_pktmbuf_pool_create(s, nb_mbuf,
1467 MEMPOOL_CACHE_SIZE, ipsec_metadata_size(),
1470 if (ctx->mbuf_pool == NULL)
1471 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool on socket %d\n",
1474 printf("Allocated mbuf pool on socket %d\n", socket_id);
1478 main(int32_t argc, char **argv)
1483 uint16_t portid, nb_ports;
1486 ret = rte_eal_init(argc, argv);
1488 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1492 /* parse application arguments (after the EAL ones) */
1493 ret = parse_args(argc, argv);
1495 rte_exit(EXIT_FAILURE, "Invalid parameters\n");
1497 if ((unprotected_port_mask & enabled_port_mask) !=
1498 unprotected_port_mask)
1499 rte_exit(EXIT_FAILURE, "Invalid unprotected portmask 0x%x\n",
1500 unprotected_port_mask);
1502 nb_ports = rte_eth_dev_count();
1504 if (check_params() < 0)
1505 rte_exit(EXIT_FAILURE, "check_params failed\n");
1507 ret = init_lcore_rx_queues();
1509 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1511 nb_lcores = rte_lcore_count();
1513 /* Replicate each context per socket */
1514 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1515 if (rte_lcore_is_enabled(lcore_id) == 0)
1519 socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1523 if (socket_ctx[socket_id].mbuf_pool)
1526 sa_init(&socket_ctx[socket_id], socket_id);
1528 sp4_init(&socket_ctx[socket_id], socket_id);
1530 sp6_init(&socket_ctx[socket_id], socket_id);
1532 rt_init(&socket_ctx[socket_id], socket_id);
1534 pool_init(&socket_ctx[socket_id], socket_id, NB_MBUF);
1537 for (portid = 0; portid < nb_ports; portid++) {
1538 if ((enabled_port_mask & (1 << portid)) == 0)
1547 for (portid = 0; portid < nb_ports; portid++) {
1548 if ((enabled_port_mask & (1 << portid)) == 0)
1552 ret = rte_eth_dev_start(portid);
1554 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: "
1555 "err=%d, port=%d\n", ret, portid);
1557 * If enabled, put device in promiscuous mode.
1558 * This allows IO forwarding mode to forward packets
1559 * to itself through 2 cross-connected ports of the
1563 rte_eth_promiscuous_enable(portid);
1566 check_all_ports_link_status(nb_ports, enabled_port_mask);
1568 /* launch per-lcore init on every lcore */
1569 rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1570 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1571 if (rte_eal_wait_lcore(lcore_id) < 0)