1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2016 Intel Corporation
10 #include <netinet/in.h>
11 #include <netinet/ip.h>
12 #include <netinet/ip6.h>
14 #include <sys/queue.h>
19 #include <rte_common.h>
20 #include <rte_byteorder.h>
23 #include <rte_launch.h>
24 #include <rte_atomic.h>
25 #include <rte_cycles.h>
26 #include <rte_prefetch.h>
27 #include <rte_lcore.h>
28 #include <rte_per_lcore.h>
29 #include <rte_branch_prediction.h>
30 #include <rte_interrupts.h>
31 #include <rte_random.h>
32 #include <rte_debug.h>
33 #include <rte_ether.h>
34 #include <rte_ethdev.h>
35 #include <rte_mempool.h>
41 #include <rte_jhash.h>
42 #include <rte_cryptodev.h>
43 #include <rte_security.h>
45 #include <rte_ip_frag.h>
50 #define RTE_LOGTYPE_IPSEC RTE_LOGTYPE_USER1
52 #define MAX_JUMBO_PKT_LEN 9600
54 #define MEMPOOL_CACHE_SIZE 256
56 #define NB_MBUF (32000)
58 #define CDEV_QUEUE_DESC 2048
59 #define CDEV_MAP_ENTRIES 16384
60 #define CDEV_MP_NB_OBJS 1024
61 #define CDEV_MP_CACHE_SZ 64
62 #define MAX_QUEUE_PAIRS 1
64 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
68 /* Configure how many packets ahead to prefetch, when reading packets */
69 #define PREFETCH_OFFSET 3
71 #define MAX_RX_QUEUE_PER_LCORE 16
73 #define MAX_LCORE_PARAMS 1024
75 #define UNPROTECTED_PORT(port) (unprotected_port_mask & (1 << portid))
78 * Configurable number of RX/TX ring descriptors
80 #define IPSEC_SECGW_RX_DESC_DEFAULT 1024
81 #define IPSEC_SECGW_TX_DESC_DEFAULT 1024
82 static uint16_t nb_rxd = IPSEC_SECGW_RX_DESC_DEFAULT;
83 static uint16_t nb_txd = IPSEC_SECGW_TX_DESC_DEFAULT;
85 #if RTE_BYTE_ORDER != RTE_LITTLE_ENDIAN
86 #define __BYTES_TO_UINT64(a, b, c, d, e, f, g, h) \
87 (((uint64_t)((a) & 0xff) << 56) | \
88 ((uint64_t)((b) & 0xff) << 48) | \
89 ((uint64_t)((c) & 0xff) << 40) | \
90 ((uint64_t)((d) & 0xff) << 32) | \
91 ((uint64_t)((e) & 0xff) << 24) | \
92 ((uint64_t)((f) & 0xff) << 16) | \
93 ((uint64_t)((g) & 0xff) << 8) | \
94 ((uint64_t)(h) & 0xff))
96 #define __BYTES_TO_UINT64(a, b, c, d, e, f, g, h) \
97 (((uint64_t)((h) & 0xff) << 56) | \
98 ((uint64_t)((g) & 0xff) << 48) | \
99 ((uint64_t)((f) & 0xff) << 40) | \
100 ((uint64_t)((e) & 0xff) << 32) | \
101 ((uint64_t)((d) & 0xff) << 24) | \
102 ((uint64_t)((c) & 0xff) << 16) | \
103 ((uint64_t)((b) & 0xff) << 8) | \
104 ((uint64_t)(a) & 0xff))
106 #define ETHADDR(a, b, c, d, e, f) (__BYTES_TO_UINT64(a, b, c, d, e, f, 0, 0))
108 #define ETHADDR_TO_UINT64(addr) __BYTES_TO_UINT64( \
109 (addr)->addr_bytes[0], (addr)->addr_bytes[1], \
110 (addr)->addr_bytes[2], (addr)->addr_bytes[3], \
111 (addr)->addr_bytes[4], (addr)->addr_bytes[5], \
114 #define FRAG_TBL_BUCKET_ENTRIES 4
115 #define FRAG_TTL_MS (10 * MS_PER_S)
117 #define MTU_TO_FRAMELEN(x) ((x) + RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN)
119 /* port/source ethernet addr and destination ethernet addr */
120 struct ethaddr_info {
124 struct ethaddr_info ethaddr_tbl[RTE_MAX_ETHPORTS] = {
125 { 0, ETHADDR(0x00, 0x16, 0x3e, 0x7e, 0x94, 0x9a) },
126 { 0, ETHADDR(0x00, 0x16, 0x3e, 0x22, 0xa1, 0xd9) },
127 { 0, ETHADDR(0x00, 0x16, 0x3e, 0x08, 0x69, 0x26) },
128 { 0, ETHADDR(0x00, 0x16, 0x3e, 0x49, 0x9e, 0xdd) }
131 #define CMD_LINE_OPT_CONFIG "config"
132 #define CMD_LINE_OPT_SINGLE_SA "single-sa"
133 #define CMD_LINE_OPT_CRYPTODEV_MASK "cryptodev_mask"
134 #define CMD_LINE_OPT_RX_OFFLOAD "rxoffload"
135 #define CMD_LINE_OPT_TX_OFFLOAD "txoffload"
136 #define CMD_LINE_OPT_REASSEMBLE "reassemble"
137 #define CMD_LINE_OPT_MTU "mtu"
140 /* long options mapped to a short option */
142 /* first long only option value must be >= 256, so that we won't
143 * conflict with short options
145 CMD_LINE_OPT_MIN_NUM = 256,
146 CMD_LINE_OPT_CONFIG_NUM,
147 CMD_LINE_OPT_SINGLE_SA_NUM,
148 CMD_LINE_OPT_CRYPTODEV_MASK_NUM,
149 CMD_LINE_OPT_RX_OFFLOAD_NUM,
150 CMD_LINE_OPT_TX_OFFLOAD_NUM,
151 CMD_LINE_OPT_REASSEMBLE_NUM,
152 CMD_LINE_OPT_MTU_NUM,
155 static const struct option lgopts[] = {
156 {CMD_LINE_OPT_CONFIG, 1, 0, CMD_LINE_OPT_CONFIG_NUM},
157 {CMD_LINE_OPT_SINGLE_SA, 1, 0, CMD_LINE_OPT_SINGLE_SA_NUM},
158 {CMD_LINE_OPT_CRYPTODEV_MASK, 1, 0, CMD_LINE_OPT_CRYPTODEV_MASK_NUM},
159 {CMD_LINE_OPT_RX_OFFLOAD, 1, 0, CMD_LINE_OPT_RX_OFFLOAD_NUM},
160 {CMD_LINE_OPT_TX_OFFLOAD, 1, 0, CMD_LINE_OPT_TX_OFFLOAD_NUM},
161 {CMD_LINE_OPT_REASSEMBLE, 1, 0, CMD_LINE_OPT_REASSEMBLE_NUM},
162 {CMD_LINE_OPT_MTU, 1, 0, CMD_LINE_OPT_MTU_NUM},
166 /* mask of enabled ports */
167 static uint32_t enabled_port_mask;
168 static uint64_t enabled_cryptodev_mask = UINT64_MAX;
169 static uint32_t unprotected_port_mask;
170 static int32_t promiscuous_on = 1;
171 static int32_t numa_on = 1; /**< NUMA is enabled by default. */
172 static uint32_t nb_lcores;
173 static uint32_t single_sa;
174 static uint32_t single_sa_idx;
177 * RX/TX HW offload capabilities to enable/use on ethernet ports.
178 * By default all capabilities are enabled.
180 static uint64_t dev_rx_offload = UINT64_MAX;
181 static uint64_t dev_tx_offload = UINT64_MAX;
184 * global values that determine multi-seg policy
186 static uint32_t frag_tbl_sz;
187 static uint32_t frame_buf_size = RTE_MBUF_DEFAULT_BUF_SIZE;
188 static uint32_t mtu_size = RTE_ETHER_MTU;
190 /* application wide librte_ipsec/SA parameters */
191 struct app_sa_prm app_sa_prm = {.enable = 0};
193 struct lcore_rx_queue {
196 } __rte_cache_aligned;
198 struct lcore_params {
202 } __rte_cache_aligned;
204 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
206 static struct lcore_params *lcore_params;
207 static uint16_t nb_lcore_params;
209 static struct rte_hash *cdev_map_in;
210 static struct rte_hash *cdev_map_out;
214 struct rte_mbuf *m_table[MAX_PKT_BURST] __rte_aligned(sizeof(void *));
218 uint16_t nb_rx_queue;
219 struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
220 uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
221 struct buffer tx_mbufs[RTE_MAX_ETHPORTS];
222 struct ipsec_ctx inbound;
223 struct ipsec_ctx outbound;
224 struct rt_ctx *rt4_ctx;
225 struct rt_ctx *rt6_ctx;
227 struct rte_ip_frag_tbl *tbl;
228 struct rte_mempool *pool_dir;
229 struct rte_mempool *pool_indir;
230 struct rte_ip_frag_death_row dr;
232 } __rte_cache_aligned;
234 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
236 static struct rte_eth_conf port_conf = {
238 .mq_mode = ETH_MQ_RX_RSS,
239 .max_rx_pkt_len = RTE_ETHER_MAX_LEN,
241 .offloads = DEV_RX_OFFLOAD_CHECKSUM,
246 .rss_hf = ETH_RSS_IP | ETH_RSS_UDP |
247 ETH_RSS_TCP | ETH_RSS_SCTP,
251 .mq_mode = ETH_MQ_TX_NONE,
255 static struct socket_ctx socket_ctx[NB_SOCKETS];
258 * Determine is multi-segment support required:
259 * - either frame buffer size is smaller then mtu
260 * - or reassmeble support is requested
263 multi_seg_required(void)
265 return (MTU_TO_FRAMELEN(mtu_size) + RTE_PKTMBUF_HEADROOM >
266 frame_buf_size || frag_tbl_sz != 0);
270 adjust_ipv4_pktlen(struct rte_mbuf *m, const struct rte_ipv4_hdr *iph,
275 plen = rte_be_to_cpu_16(iph->total_length) + l2_len;
276 if (plen < m->pkt_len) {
277 trim = m->pkt_len - plen;
278 rte_pktmbuf_trim(m, trim);
283 adjust_ipv6_pktlen(struct rte_mbuf *m, const struct rte_ipv6_hdr *iph,
288 plen = rte_be_to_cpu_16(iph->payload_len) + sizeof(*iph) + l2_len;
289 if (plen < m->pkt_len) {
290 trim = m->pkt_len - plen;
291 rte_pktmbuf_trim(m, trim);
296 prepare_one_packet(struct rte_mbuf *pkt, struct ipsec_traffic *t)
298 const struct rte_ether_hdr *eth;
299 const struct rte_ipv4_hdr *iph4;
300 const struct rte_ipv6_hdr *iph6;
302 eth = rte_pktmbuf_mtod(pkt, const struct rte_ether_hdr *);
303 if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
305 iph4 = (const struct rte_ipv4_hdr *)rte_pktmbuf_adj(pkt,
307 adjust_ipv4_pktlen(pkt, iph4, 0);
309 if (iph4->next_proto_id == IPPROTO_ESP)
310 t->ipsec.pkts[(t->ipsec.num)++] = pkt;
312 t->ip4.data[t->ip4.num] = &iph4->next_proto_id;
313 t->ip4.pkts[(t->ip4.num)++] = pkt;
316 pkt->l3_len = sizeof(*iph4);
317 } else if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6)) {
319 size_t l3len, ext_len;
322 /* get protocol type */
323 iph6 = (const struct rte_ipv6_hdr *)rte_pktmbuf_adj(pkt,
325 adjust_ipv6_pktlen(pkt, iph6, 0);
327 next_proto = iph6->proto;
329 /* determine l3 header size up to ESP extension */
330 l3len = sizeof(struct ip6_hdr);
331 p = rte_pktmbuf_mtod(pkt, uint8_t *);
332 while (next_proto != IPPROTO_ESP && l3len < pkt->data_len &&
333 (next_proto = rte_ipv6_get_next_ext(p + l3len,
334 next_proto, &ext_len)) >= 0)
337 /* drop packet when IPv6 header exceeds first segment length */
338 if (unlikely(l3len > pkt->data_len)) {
339 rte_pktmbuf_free(pkt);
343 if (next_proto == IPPROTO_ESP)
344 t->ipsec.pkts[(t->ipsec.num)++] = pkt;
346 t->ip6.data[t->ip6.num] = &iph6->proto;
347 t->ip6.pkts[(t->ip6.num)++] = pkt;
352 /* Unknown/Unsupported type, drop the packet */
353 RTE_LOG(ERR, IPSEC, "Unsupported packet type 0x%x\n",
354 rte_be_to_cpu_16(eth->ether_type));
355 rte_pktmbuf_free(pkt);
359 /* Check if the packet has been processed inline. For inline protocol
360 * processed packets, the metadata in the mbuf can be used to identify
361 * the security processing done on the packet. The metadata will be
362 * used to retrieve the application registered userdata associated
363 * with the security session.
366 if (pkt->ol_flags & PKT_RX_SEC_OFFLOAD) {
368 struct ipsec_mbuf_metadata *priv;
369 struct rte_security_ctx *ctx = (struct rte_security_ctx *)
370 rte_eth_dev_get_sec_ctx(
373 /* Retrieve the userdata registered. Here, the userdata
374 * registered is the SA pointer.
377 sa = (struct ipsec_sa *)
378 rte_security_get_userdata(ctx, pkt->udata64);
381 /* userdata could not be retrieved */
385 /* Save SA as priv member in mbuf. This will be used in the
386 * IPsec selector(SP-SA) check.
389 priv = get_priv(pkt);
395 prepare_traffic(struct rte_mbuf **pkts, struct ipsec_traffic *t,
404 for (i = 0; i < (nb_pkts - PREFETCH_OFFSET); i++) {
405 rte_prefetch0(rte_pktmbuf_mtod(pkts[i + PREFETCH_OFFSET],
407 prepare_one_packet(pkts[i], t);
409 /* Process left packets */
410 for (; i < nb_pkts; i++)
411 prepare_one_packet(pkts[i], t);
415 prepare_tx_pkt(struct rte_mbuf *pkt, uint16_t port,
416 const struct lcore_conf *qconf)
419 struct rte_ether_hdr *ethhdr;
421 ip = rte_pktmbuf_mtod(pkt, struct ip *);
423 ethhdr = (struct rte_ether_hdr *)
424 rte_pktmbuf_prepend(pkt, RTE_ETHER_HDR_LEN);
426 if (ip->ip_v == IPVERSION) {
427 pkt->ol_flags |= qconf->outbound.ipv4_offloads;
428 pkt->l3_len = sizeof(struct ip);
429 pkt->l2_len = RTE_ETHER_HDR_LEN;
433 /* calculate IPv4 cksum in SW */
434 if ((pkt->ol_flags & PKT_TX_IP_CKSUM) == 0)
435 ip->ip_sum = rte_ipv4_cksum((struct rte_ipv4_hdr *)ip);
437 ethhdr->ether_type = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4);
439 pkt->ol_flags |= qconf->outbound.ipv6_offloads;
440 pkt->l3_len = sizeof(struct ip6_hdr);
441 pkt->l2_len = RTE_ETHER_HDR_LEN;
443 ethhdr->ether_type = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6);
446 memcpy(ðhdr->s_addr, ðaddr_tbl[port].src,
447 sizeof(struct rte_ether_addr));
448 memcpy(ðhdr->d_addr, ðaddr_tbl[port].dst,
449 sizeof(struct rte_ether_addr));
453 prepare_tx_burst(struct rte_mbuf *pkts[], uint16_t nb_pkts, uint16_t port,
454 const struct lcore_conf *qconf)
457 const int32_t prefetch_offset = 2;
459 for (i = 0; i < (nb_pkts - prefetch_offset); i++) {
460 rte_mbuf_prefetch_part2(pkts[i + prefetch_offset]);
461 prepare_tx_pkt(pkts[i], port, qconf);
463 /* Process left packets */
464 for (; i < nb_pkts; i++)
465 prepare_tx_pkt(pkts[i], port, qconf);
468 /* Send burst of packets on an output interface */
469 static inline int32_t
470 send_burst(struct lcore_conf *qconf, uint16_t n, uint16_t port)
472 struct rte_mbuf **m_table;
476 queueid = qconf->tx_queue_id[port];
477 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
479 prepare_tx_burst(m_table, n, port, qconf);
481 ret = rte_eth_tx_burst(port, queueid, m_table, n);
482 if (unlikely(ret < n)) {
484 rte_pktmbuf_free(m_table[ret]);
492 * Helper function to fragment and queue for TX one packet.
494 static inline uint32_t
495 send_fragment_packet(struct lcore_conf *qconf, struct rte_mbuf *m,
496 uint16_t port, uint8_t proto)
502 tbl = qconf->tx_mbufs + port;
505 /* free space for new fragments */
506 if (len + RTE_LIBRTE_IP_FRAG_MAX_FRAG >= RTE_DIM(tbl->m_table)) {
507 send_burst(qconf, len, port);
511 n = RTE_DIM(tbl->m_table) - len;
513 if (proto == IPPROTO_IP)
514 rc = rte_ipv4_fragment_packet(m, tbl->m_table + len,
515 n, mtu_size, qconf->frag.pool_dir,
516 qconf->frag.pool_indir);
518 rc = rte_ipv6_fragment_packet(m, tbl->m_table + len,
519 n, mtu_size, qconf->frag.pool_dir,
520 qconf->frag.pool_indir);
526 "%s: failed to fragment packet with size %u, "
528 __func__, m->pkt_len, rte_errno);
534 /* Enqueue a single packet, and send burst if queue is filled */
535 static inline int32_t
536 send_single_packet(struct rte_mbuf *m, uint16_t port, uint8_t proto)
540 struct lcore_conf *qconf;
542 lcore_id = rte_lcore_id();
544 qconf = &lcore_conf[lcore_id];
545 len = qconf->tx_mbufs[port].len;
547 if (m->pkt_len <= mtu_size) {
548 qconf->tx_mbufs[port].m_table[len] = m;
551 /* need to fragment the packet */
552 } else if (frag_tbl_sz > 0)
553 len = send_fragment_packet(qconf, m, port, proto);
557 /* enough pkts to be sent */
558 if (unlikely(len == MAX_PKT_BURST)) {
559 send_burst(qconf, MAX_PKT_BURST, port);
563 qconf->tx_mbufs[port].len = len;
568 inbound_sp_sa(struct sp_ctx *sp, struct sa_ctx *sa, struct traffic_type *ip,
572 uint32_t i, j, res, sa_idx;
574 if (ip->num == 0 || sp == NULL)
577 rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
578 ip->num, DEFAULT_MAX_CATEGORIES);
581 for (i = 0; i < ip->num; i++) {
588 if (res == DISCARD) {
593 /* Only check SPI match for processed IPSec packets */
594 if (i < lim && ((m->ol_flags & PKT_RX_SEC_OFFLOAD) == 0)) {
599 sa_idx = SPI2IDX(res);
600 if (!inbound_sa_check(sa, m, sa_idx)) {
610 split46_traffic(struct ipsec_traffic *trf, struct rte_mbuf *mb[], uint32_t num)
619 for (i = 0; i < num; i++) {
622 ip = rte_pktmbuf_mtod(m, struct ip *);
624 if (ip->ip_v == IPVERSION) {
625 trf->ip4.pkts[n4] = m;
626 trf->ip4.data[n4] = rte_pktmbuf_mtod_offset(m,
627 uint8_t *, offsetof(struct ip, ip_p));
629 } else if (ip->ip_v == IP6_VERSION) {
630 trf->ip6.pkts[n6] = m;
631 trf->ip6.data[n6] = rte_pktmbuf_mtod_offset(m,
633 offsetof(struct ip6_hdr, ip6_nxt));
645 process_pkts_inbound(struct ipsec_ctx *ipsec_ctx,
646 struct ipsec_traffic *traffic)
648 uint16_t nb_pkts_in, n_ip4, n_ip6;
650 n_ip4 = traffic->ip4.num;
651 n_ip6 = traffic->ip6.num;
653 if (app_sa_prm.enable == 0) {
654 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
655 traffic->ipsec.num, MAX_PKT_BURST);
656 split46_traffic(traffic, traffic->ipsec.pkts, nb_pkts_in);
658 inbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.pkts,
659 traffic->ipsec.saptr, traffic->ipsec.num);
660 ipsec_process(ipsec_ctx, traffic);
663 inbound_sp_sa(ipsec_ctx->sp4_ctx, ipsec_ctx->sa_ctx, &traffic->ip4,
666 inbound_sp_sa(ipsec_ctx->sp6_ctx, ipsec_ctx->sa_ctx, &traffic->ip6,
671 outbound_sp(struct sp_ctx *sp, struct traffic_type *ip,
672 struct traffic_type *ipsec)
675 uint32_t i, j, sa_idx;
677 if (ip->num == 0 || sp == NULL)
680 rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
681 ip->num, DEFAULT_MAX_CATEGORIES);
684 for (i = 0; i < ip->num; i++) {
686 sa_idx = SPI2IDX(ip->res[i]);
687 if (ip->res[i] == DISCARD)
689 else if (ip->res[i] == BYPASS)
692 ipsec->res[ipsec->num] = sa_idx;
693 ipsec->pkts[ipsec->num++] = m;
700 process_pkts_outbound(struct ipsec_ctx *ipsec_ctx,
701 struct ipsec_traffic *traffic)
704 uint16_t idx, nb_pkts_out, i;
706 /* Drop any IPsec traffic from protected ports */
707 for (i = 0; i < traffic->ipsec.num; i++)
708 rte_pktmbuf_free(traffic->ipsec.pkts[i]);
710 traffic->ipsec.num = 0;
712 outbound_sp(ipsec_ctx->sp4_ctx, &traffic->ip4, &traffic->ipsec);
714 outbound_sp(ipsec_ctx->sp6_ctx, &traffic->ip6, &traffic->ipsec);
716 if (app_sa_prm.enable == 0) {
718 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
719 traffic->ipsec.res, traffic->ipsec.num,
722 for (i = 0; i < nb_pkts_out; i++) {
723 m = traffic->ipsec.pkts[i];
724 struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
725 if (ip->ip_v == IPVERSION) {
726 idx = traffic->ip4.num++;
727 traffic->ip4.pkts[idx] = m;
729 idx = traffic->ip6.num++;
730 traffic->ip6.pkts[idx] = m;
734 outbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.res,
735 traffic->ipsec.saptr, traffic->ipsec.num);
736 ipsec_process(ipsec_ctx, traffic);
741 process_pkts_inbound_nosp(struct ipsec_ctx *ipsec_ctx,
742 struct ipsec_traffic *traffic)
745 uint32_t nb_pkts_in, i, idx;
747 /* Drop any IPv4 traffic from unprotected ports */
748 for (i = 0; i < traffic->ip4.num; i++)
749 rte_pktmbuf_free(traffic->ip4.pkts[i]);
751 traffic->ip4.num = 0;
753 /* Drop any IPv6 traffic from unprotected ports */
754 for (i = 0; i < traffic->ip6.num; i++)
755 rte_pktmbuf_free(traffic->ip6.pkts[i]);
757 traffic->ip6.num = 0;
759 if (app_sa_prm.enable == 0) {
761 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
762 traffic->ipsec.num, MAX_PKT_BURST);
764 for (i = 0; i < nb_pkts_in; i++) {
765 m = traffic->ipsec.pkts[i];
766 struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
767 if (ip->ip_v == IPVERSION) {
768 idx = traffic->ip4.num++;
769 traffic->ip4.pkts[idx] = m;
771 idx = traffic->ip6.num++;
772 traffic->ip6.pkts[idx] = m;
776 inbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.pkts,
777 traffic->ipsec.saptr, traffic->ipsec.num);
778 ipsec_process(ipsec_ctx, traffic);
783 process_pkts_outbound_nosp(struct ipsec_ctx *ipsec_ctx,
784 struct ipsec_traffic *traffic)
787 uint32_t nb_pkts_out, i, n;
790 /* Drop any IPsec traffic from protected ports */
791 for (i = 0; i < traffic->ipsec.num; i++)
792 rte_pktmbuf_free(traffic->ipsec.pkts[i]);
796 for (i = 0; i < traffic->ip4.num; i++) {
797 traffic->ipsec.pkts[n] = traffic->ip4.pkts[i];
798 traffic->ipsec.res[n++] = single_sa_idx;
801 for (i = 0; i < traffic->ip6.num; i++) {
802 traffic->ipsec.pkts[n] = traffic->ip6.pkts[i];
803 traffic->ipsec.res[n++] = single_sa_idx;
806 traffic->ip4.num = 0;
807 traffic->ip6.num = 0;
808 traffic->ipsec.num = n;
810 if (app_sa_prm.enable == 0) {
812 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
813 traffic->ipsec.res, traffic->ipsec.num,
816 /* They all sue the same SA (ip4 or ip6 tunnel) */
817 m = traffic->ipsec.pkts[0];
818 ip = rte_pktmbuf_mtod(m, struct ip *);
819 if (ip->ip_v == IPVERSION) {
820 traffic->ip4.num = nb_pkts_out;
821 for (i = 0; i < nb_pkts_out; i++)
822 traffic->ip4.pkts[i] = traffic->ipsec.pkts[i];
824 traffic->ip6.num = nb_pkts_out;
825 for (i = 0; i < nb_pkts_out; i++)
826 traffic->ip6.pkts[i] = traffic->ipsec.pkts[i];
829 outbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.res,
830 traffic->ipsec.saptr, traffic->ipsec.num);
831 ipsec_process(ipsec_ctx, traffic);
835 static inline int32_t
836 get_hop_for_offload_pkt(struct rte_mbuf *pkt, int is_ipv6)
838 struct ipsec_mbuf_metadata *priv;
841 priv = get_priv(pkt);
844 if (unlikely(sa == NULL)) {
845 RTE_LOG(ERR, IPSEC, "SA not saved in private data\n");
853 return (sa->portid | RTE_LPM_LOOKUP_SUCCESS);
864 route4_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
866 uint32_t hop[MAX_PKT_BURST * 2];
867 uint32_t dst_ip[MAX_PKT_BURST * 2];
870 uint16_t lpm_pkts = 0;
875 /* Need to do an LPM lookup for non-inline packets. Inline packets will
876 * have port ID in the SA
879 for (i = 0; i < nb_pkts; i++) {
880 if (!(pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD)) {
881 /* Security offload not enabled. So an LPM lookup is
882 * required to get the hop
884 offset = offsetof(struct ip, ip_dst);
885 dst_ip[lpm_pkts] = *rte_pktmbuf_mtod_offset(pkts[i],
887 dst_ip[lpm_pkts] = rte_be_to_cpu_32(dst_ip[lpm_pkts]);
892 rte_lpm_lookup_bulk((struct rte_lpm *)rt_ctx, dst_ip, hop, lpm_pkts);
896 for (i = 0; i < nb_pkts; i++) {
897 if (pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD) {
898 /* Read hop from the SA */
899 pkt_hop = get_hop_for_offload_pkt(pkts[i], 0);
901 /* Need to use hop returned by lookup */
902 pkt_hop = hop[lpm_pkts++];
905 if ((pkt_hop & RTE_LPM_LOOKUP_SUCCESS) == 0) {
906 rte_pktmbuf_free(pkts[i]);
909 send_single_packet(pkts[i], pkt_hop & 0xff, IPPROTO_IP);
914 route6_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
916 int32_t hop[MAX_PKT_BURST * 2];
917 uint8_t dst_ip[MAX_PKT_BURST * 2][16];
921 uint16_t lpm_pkts = 0;
926 /* Need to do an LPM lookup for non-inline packets. Inline packets will
927 * have port ID in the SA
930 for (i = 0; i < nb_pkts; i++) {
931 if (!(pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD)) {
932 /* Security offload not enabled. So an LPM lookup is
933 * required to get the hop
935 offset = offsetof(struct ip6_hdr, ip6_dst);
936 ip6_dst = rte_pktmbuf_mtod_offset(pkts[i], uint8_t *,
938 memcpy(&dst_ip[lpm_pkts][0], ip6_dst, 16);
943 rte_lpm6_lookup_bulk_func((struct rte_lpm6 *)rt_ctx, dst_ip, hop,
948 for (i = 0; i < nb_pkts; i++) {
949 if (pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD) {
950 /* Read hop from the SA */
951 pkt_hop = get_hop_for_offload_pkt(pkts[i], 1);
953 /* Need to use hop returned by lookup */
954 pkt_hop = hop[lpm_pkts++];
958 rte_pktmbuf_free(pkts[i]);
961 send_single_packet(pkts[i], pkt_hop & 0xff, IPPROTO_IPV6);
966 process_pkts(struct lcore_conf *qconf, struct rte_mbuf **pkts,
967 uint8_t nb_pkts, uint16_t portid)
969 struct ipsec_traffic traffic;
971 prepare_traffic(pkts, &traffic, nb_pkts);
973 if (unlikely(single_sa)) {
974 if (UNPROTECTED_PORT(portid))
975 process_pkts_inbound_nosp(&qconf->inbound, &traffic);
977 process_pkts_outbound_nosp(&qconf->outbound, &traffic);
979 if (UNPROTECTED_PORT(portid))
980 process_pkts_inbound(&qconf->inbound, &traffic);
982 process_pkts_outbound(&qconf->outbound, &traffic);
985 route4_pkts(qconf->rt4_ctx, traffic.ip4.pkts, traffic.ip4.num);
986 route6_pkts(qconf->rt6_ctx, traffic.ip6.pkts, traffic.ip6.num);
990 drain_tx_buffers(struct lcore_conf *qconf)
995 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
996 buf = &qconf->tx_mbufs[portid];
999 send_burst(qconf, buf->len, portid);
1005 drain_crypto_buffers(struct lcore_conf *qconf)
1008 struct ipsec_ctx *ctx;
1010 /* drain inbound buffers*/
1011 ctx = &qconf->inbound;
1012 for (i = 0; i != ctx->nb_qps; i++) {
1013 if (ctx->tbl[i].len != 0)
1014 enqueue_cop_burst(ctx->tbl + i);
1017 /* drain outbound buffers*/
1018 ctx = &qconf->outbound;
1019 for (i = 0; i != ctx->nb_qps; i++) {
1020 if (ctx->tbl[i].len != 0)
1021 enqueue_cop_burst(ctx->tbl + i);
1026 drain_inbound_crypto_queues(const struct lcore_conf *qconf,
1027 struct ipsec_ctx *ctx)
1030 struct ipsec_traffic trf;
1032 if (app_sa_prm.enable == 0) {
1034 /* dequeue packets from crypto-queue */
1035 n = ipsec_inbound_cqp_dequeue(ctx, trf.ipsec.pkts,
1036 RTE_DIM(trf.ipsec.pkts));
1041 /* split traffic by ipv4-ipv6 */
1042 split46_traffic(&trf, trf.ipsec.pkts, n);
1044 ipsec_cqp_process(ctx, &trf);
1046 /* process ipv4 packets */
1047 if (trf.ip4.num != 0) {
1048 inbound_sp_sa(ctx->sp4_ctx, ctx->sa_ctx, &trf.ip4, 0);
1049 route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
1052 /* process ipv6 packets */
1053 if (trf.ip6.num != 0) {
1054 inbound_sp_sa(ctx->sp6_ctx, ctx->sa_ctx, &trf.ip6, 0);
1055 route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
1060 drain_outbound_crypto_queues(const struct lcore_conf *qconf,
1061 struct ipsec_ctx *ctx)
1064 struct ipsec_traffic trf;
1066 if (app_sa_prm.enable == 0) {
1068 /* dequeue packets from crypto-queue */
1069 n = ipsec_outbound_cqp_dequeue(ctx, trf.ipsec.pkts,
1070 RTE_DIM(trf.ipsec.pkts));
1075 /* split traffic by ipv4-ipv6 */
1076 split46_traffic(&trf, trf.ipsec.pkts, n);
1078 ipsec_cqp_process(ctx, &trf);
1080 /* process ipv4 packets */
1081 if (trf.ip4.num != 0)
1082 route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
1084 /* process ipv6 packets */
1085 if (trf.ip6.num != 0)
1086 route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
1089 /* main processing loop */
1091 main_loop(__attribute__((unused)) void *dummy)
1093 struct rte_mbuf *pkts[MAX_PKT_BURST];
1095 uint64_t prev_tsc, diff_tsc, cur_tsc;
1099 struct lcore_conf *qconf;
1101 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
1102 / US_PER_S * BURST_TX_DRAIN_US;
1103 struct lcore_rx_queue *rxql;
1106 lcore_id = rte_lcore_id();
1107 qconf = &lcore_conf[lcore_id];
1108 rxql = qconf->rx_queue_list;
1109 socket_id = rte_lcore_to_socket_id(lcore_id);
1111 qconf->rt4_ctx = socket_ctx[socket_id].rt_ip4;
1112 qconf->rt6_ctx = socket_ctx[socket_id].rt_ip6;
1113 qconf->inbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_in;
1114 qconf->inbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_in;
1115 qconf->inbound.sa_ctx = socket_ctx[socket_id].sa_in;
1116 qconf->inbound.cdev_map = cdev_map_in;
1117 qconf->inbound.session_pool = socket_ctx[socket_id].session_pool;
1118 qconf->inbound.session_priv_pool =
1119 socket_ctx[socket_id].session_priv_pool;
1120 qconf->outbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_out;
1121 qconf->outbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_out;
1122 qconf->outbound.sa_ctx = socket_ctx[socket_id].sa_out;
1123 qconf->outbound.cdev_map = cdev_map_out;
1124 qconf->outbound.session_pool = socket_ctx[socket_id].session_pool;
1125 qconf->outbound.session_priv_pool =
1126 socket_ctx[socket_id].session_priv_pool;
1127 qconf->frag.pool_dir = socket_ctx[socket_id].mbuf_pool;
1128 qconf->frag.pool_indir = socket_ctx[socket_id].mbuf_pool_indir;
1130 if (qconf->nb_rx_queue == 0) {
1131 RTE_LOG(DEBUG, IPSEC, "lcore %u has nothing to do\n",
1136 RTE_LOG(INFO, IPSEC, "entering main loop on lcore %u\n", lcore_id);
1138 for (i = 0; i < qconf->nb_rx_queue; i++) {
1139 portid = rxql[i].port_id;
1140 queueid = rxql[i].queue_id;
1141 RTE_LOG(INFO, IPSEC,
1142 " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
1143 lcore_id, portid, queueid);
1147 cur_tsc = rte_rdtsc();
1149 /* TX queue buffer drain */
1150 diff_tsc = cur_tsc - prev_tsc;
1152 if (unlikely(diff_tsc > drain_tsc)) {
1153 drain_tx_buffers(qconf);
1154 drain_crypto_buffers(qconf);
1158 for (i = 0; i < qconf->nb_rx_queue; ++i) {
1160 /* Read packets from RX queues */
1161 portid = rxql[i].port_id;
1162 queueid = rxql[i].queue_id;
1163 nb_rx = rte_eth_rx_burst(portid, queueid,
1164 pkts, MAX_PKT_BURST);
1167 process_pkts(qconf, pkts, nb_rx, portid);
1169 /* dequeue and process completed crypto-ops */
1170 if (UNPROTECTED_PORT(portid))
1171 drain_inbound_crypto_queues(qconf,
1174 drain_outbound_crypto_queues(qconf,
1188 if (lcore_params == NULL) {
1189 printf("Error: No port/queue/core mappings\n");
1193 for (i = 0; i < nb_lcore_params; ++i) {
1194 lcore = lcore_params[i].lcore_id;
1195 if (!rte_lcore_is_enabled(lcore)) {
1196 printf("error: lcore %hhu is not enabled in "
1197 "lcore mask\n", lcore);
1200 socket_id = rte_lcore_to_socket_id(lcore);
1201 if (socket_id != 0 && numa_on == 0) {
1202 printf("warning: lcore %hhu is on socket %d "
1206 portid = lcore_params[i].port_id;
1207 if ((enabled_port_mask & (1 << portid)) == 0) {
1208 printf("port %u is not enabled in port mask\n", portid);
1211 if (!rte_eth_dev_is_valid_port(portid)) {
1212 printf("port %u is not present on the board\n", portid);
1220 get_port_nb_rx_queues(const uint16_t port)
1225 for (i = 0; i < nb_lcore_params; ++i) {
1226 if (lcore_params[i].port_id == port &&
1227 lcore_params[i].queue_id > queue)
1228 queue = lcore_params[i].queue_id;
1230 return (uint8_t)(++queue);
1234 init_lcore_rx_queues(void)
1236 uint16_t i, nb_rx_queue;
1239 for (i = 0; i < nb_lcore_params; ++i) {
1240 lcore = lcore_params[i].lcore_id;
1241 nb_rx_queue = lcore_conf[lcore].nb_rx_queue;
1242 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1243 printf("error: too many queues (%u) for lcore: %u\n",
1244 nb_rx_queue + 1, lcore);
1247 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1248 lcore_params[i].port_id;
1249 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1250 lcore_params[i].queue_id;
1251 lcore_conf[lcore].nb_rx_queue++;
1258 print_usage(const char *prgname)
1260 fprintf(stderr, "%s [EAL options] --"
1266 " [-w REPLAY_WINDOW_SIZE]"
1270 " --config (port,queue,lcore)[,(port,queue,lcore)]"
1271 " [--single-sa SAIDX]"
1272 " [--cryptodev_mask MASK]"
1273 " [--" CMD_LINE_OPT_RX_OFFLOAD " RX_OFFLOAD_MASK]"
1274 " [--" CMD_LINE_OPT_TX_OFFLOAD " TX_OFFLOAD_MASK]"
1275 " [--" CMD_LINE_OPT_REASSEMBLE " REASSEMBLE_TABLE_SIZE]"
1276 " [--" CMD_LINE_OPT_MTU " MTU]"
1278 " -p PORTMASK: Hexadecimal bitmask of ports to configure\n"
1279 " -P : Enable promiscuous mode\n"
1280 " -u PORTMASK: Hexadecimal bitmask of unprotected ports\n"
1281 " -j FRAMESIZE: Data buffer size, minimum (and default)\n"
1282 " value: RTE_MBUF_DEFAULT_BUF_SIZE\n"
1283 " -l enables code-path that uses librte_ipsec\n"
1284 " -w REPLAY_WINDOW_SIZE specifies IPsec SQN replay window\n"
1285 " size for each SA\n"
1287 " -a enables SA SQN atomic behaviour\n"
1288 " -f CONFIG_FILE: Configuration file\n"
1289 " --config (port,queue,lcore): Rx queue configuration\n"
1290 " --single-sa SAIDX: Use single SA index for outbound traffic,\n"
1291 " bypassing the SP\n"
1292 " --cryptodev_mask MASK: Hexadecimal bitmask of the crypto\n"
1293 " devices to configure\n"
1294 " --" CMD_LINE_OPT_RX_OFFLOAD
1295 ": bitmask of the RX HW offload capabilities to enable/use\n"
1296 " (DEV_RX_OFFLOAD_*)\n"
1297 " --" CMD_LINE_OPT_TX_OFFLOAD
1298 ": bitmask of the TX HW offload capabilities to enable/use\n"
1299 " (DEV_TX_OFFLOAD_*)\n"
1300 " --" CMD_LINE_OPT_REASSEMBLE " NUM"
1301 ": max number of entries in reassemble(fragment) table\n"
1302 " (zero (default value) disables reassembly)\n"
1303 " --" CMD_LINE_OPT_MTU " MTU"
1304 ": MTU value on all ports (default value: 1500)\n"
1305 " outgoing packets with bigger size will be fragmented\n"
1306 " incoming packets with bigger size will be discarded\n"
1312 parse_mask(const char *str, uint64_t *val)
1318 t = strtoul(str, &end, 0);
1319 if (errno != 0 || end[0] != 0)
1327 parse_portmask(const char *portmask)
1332 /* parse hexadecimal string */
1333 pm = strtoul(portmask, &end, 16);
1334 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1337 if ((pm == 0) && errno)
1344 parse_decimal(const char *str)
1349 num = strtoul(str, &end, 10);
1350 if ((str[0] == '\0') || (end == NULL) || (*end != '\0'))
1357 parse_config(const char *q_arg)
1360 const char *p, *p0 = q_arg;
1368 unsigned long int_fld[_NUM_FLD];
1369 char *str_fld[_NUM_FLD];
1373 nb_lcore_params = 0;
1375 while ((p = strchr(p0, '(')) != NULL) {
1377 p0 = strchr(p, ')');
1382 if (size >= sizeof(s))
1385 snprintf(s, sizeof(s), "%.*s", size, p);
1386 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1389 for (i = 0; i < _NUM_FLD; i++) {
1391 int_fld[i] = strtoul(str_fld[i], &end, 0);
1392 if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
1395 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1396 printf("exceeded max number of lcore params: %hu\n",
1400 lcore_params_array[nb_lcore_params].port_id =
1401 (uint8_t)int_fld[FLD_PORT];
1402 lcore_params_array[nb_lcore_params].queue_id =
1403 (uint8_t)int_fld[FLD_QUEUE];
1404 lcore_params_array[nb_lcore_params].lcore_id =
1405 (uint8_t)int_fld[FLD_LCORE];
1408 lcore_params = lcore_params_array;
1413 print_app_sa_prm(const struct app_sa_prm *prm)
1415 printf("librte_ipsec usage: %s\n",
1416 (prm->enable == 0) ? "disabled" : "enabled");
1418 if (prm->enable == 0)
1421 printf("replay window size: %u\n", prm->window_size);
1422 printf("ESN: %s\n", (prm->enable_esn == 0) ? "disabled" : "enabled");
1423 printf("SA flags: %#" PRIx64 "\n", prm->flags);
1427 parse_args(int32_t argc, char **argv)
1431 int32_t option_index;
1432 char *prgname = argv[0];
1433 int32_t f_present = 0;
1437 while ((opt = getopt_long(argc, argvopt, "aelp:Pu:f:j:w:",
1438 lgopts, &option_index)) != EOF) {
1442 enabled_port_mask = parse_portmask(optarg);
1443 if (enabled_port_mask == 0) {
1444 printf("invalid portmask\n");
1445 print_usage(prgname);
1450 printf("Promiscuous mode selected\n");
1454 unprotected_port_mask = parse_portmask(optarg);
1455 if (unprotected_port_mask == 0) {
1456 printf("invalid unprotected portmask\n");
1457 print_usage(prgname);
1462 if (f_present == 1) {
1463 printf("\"-f\" option present more than "
1465 print_usage(prgname);
1468 if (parse_cfg_file(optarg) < 0) {
1469 printf("parsing file \"%s\" failed\n",
1471 print_usage(prgname);
1477 ret = parse_decimal(optarg);
1478 if (ret < RTE_MBUF_DEFAULT_BUF_SIZE ||
1480 printf("Invalid frame buffer size value: %s\n",
1482 print_usage(prgname);
1485 frame_buf_size = ret;
1486 printf("Custom frame buffer size %u\n", frame_buf_size);
1489 app_sa_prm.enable = 1;
1492 app_sa_prm.enable = 1;
1493 app_sa_prm.window_size = parse_decimal(optarg);
1496 app_sa_prm.enable = 1;
1497 app_sa_prm.enable_esn = 1;
1500 app_sa_prm.enable = 1;
1501 app_sa_prm.flags |= RTE_IPSEC_SAFLAG_SQN_ATOM;
1503 case CMD_LINE_OPT_CONFIG_NUM:
1504 ret = parse_config(optarg);
1506 printf("Invalid config\n");
1507 print_usage(prgname);
1511 case CMD_LINE_OPT_SINGLE_SA_NUM:
1512 ret = parse_decimal(optarg);
1514 printf("Invalid argument[sa_idx]\n");
1515 print_usage(prgname);
1521 single_sa_idx = ret;
1522 printf("Configured with single SA index %u\n",
1525 case CMD_LINE_OPT_CRYPTODEV_MASK_NUM:
1526 ret = parse_portmask(optarg);
1528 printf("Invalid argument[portmask]\n");
1529 print_usage(prgname);
1534 enabled_cryptodev_mask = ret;
1536 case CMD_LINE_OPT_RX_OFFLOAD_NUM:
1537 ret = parse_mask(optarg, &dev_rx_offload);
1539 printf("Invalid argument for \'%s\': %s\n",
1540 CMD_LINE_OPT_RX_OFFLOAD, optarg);
1541 print_usage(prgname);
1545 case CMD_LINE_OPT_TX_OFFLOAD_NUM:
1546 ret = parse_mask(optarg, &dev_tx_offload);
1548 printf("Invalid argument for \'%s\': %s\n",
1549 CMD_LINE_OPT_TX_OFFLOAD, optarg);
1550 print_usage(prgname);
1554 case CMD_LINE_OPT_REASSEMBLE_NUM:
1555 ret = parse_decimal(optarg);
1557 printf("Invalid argument for \'%s\': %s\n",
1558 CMD_LINE_OPT_REASSEMBLE, optarg);
1559 print_usage(prgname);
1564 case CMD_LINE_OPT_MTU_NUM:
1565 ret = parse_decimal(optarg);
1566 if (ret < 0 || ret > RTE_IPV4_MAX_PKT_LEN) {
1567 printf("Invalid argument for \'%s\': %s\n",
1568 CMD_LINE_OPT_MTU, optarg);
1569 print_usage(prgname);
1575 print_usage(prgname);
1580 if (f_present == 0) {
1581 printf("Mandatory option \"-f\" not present\n");
1585 /* check do we need to enable multi-seg support */
1586 if (multi_seg_required()) {
1587 /* legacy mode doesn't support multi-seg */
1588 app_sa_prm.enable = 1;
1589 printf("frame buf size: %u, mtu: %u, "
1590 "number of reassemble entries: %u\n"
1591 "multi-segment support is required\n",
1592 frame_buf_size, mtu_size, frag_tbl_sz);
1595 print_app_sa_prm(&app_sa_prm);
1598 argv[optind-1] = prgname;
1601 optind = 1; /* reset getopt lib */
1606 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
1608 char buf[RTE_ETHER_ADDR_FMT_SIZE];
1609 rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
1610 printf("%s%s", name, buf);
1614 * Update destination ethaddr for the port.
1617 add_dst_ethaddr(uint16_t port, const struct rte_ether_addr *addr)
1619 if (port >= RTE_DIM(ethaddr_tbl))
1622 ethaddr_tbl[port].dst = ETHADDR_TO_UINT64(addr);
1626 /* Check the link status of all ports in up to 9s, and print them finally */
1628 check_all_ports_link_status(uint32_t port_mask)
1630 #define CHECK_INTERVAL 100 /* 100ms */
1631 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1633 uint8_t count, all_ports_up, print_flag = 0;
1634 struct rte_eth_link link;
1637 printf("\nChecking link status");
1639 for (count = 0; count <= MAX_CHECK_TIME; count++) {
1641 RTE_ETH_FOREACH_DEV(portid) {
1642 if ((port_mask & (1 << portid)) == 0)
1644 memset(&link, 0, sizeof(link));
1645 ret = rte_eth_link_get_nowait(portid, &link);
1648 if (print_flag == 1)
1649 printf("Port %u link get failed: %s\n",
1650 portid, rte_strerror(-ret));
1653 /* print link status if flag set */
1654 if (print_flag == 1) {
1655 if (link.link_status)
1657 "Port%d Link Up - speed %u Mbps -%s\n",
1658 portid, link.link_speed,
1659 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1660 ("full-duplex") : ("half-duplex\n"));
1662 printf("Port %d Link Down\n", portid);
1665 /* clear all_ports_up flag if any link down */
1666 if (link.link_status == ETH_LINK_DOWN) {
1671 /* after finally printing all link status, get out */
1672 if (print_flag == 1)
1675 if (all_ports_up == 0) {
1678 rte_delay_ms(CHECK_INTERVAL);
1681 /* set the print_flag if all ports up or timeout */
1682 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1690 add_mapping(struct rte_hash *map, const char *str, uint16_t cdev_id,
1691 uint16_t qp, struct lcore_params *params,
1692 struct ipsec_ctx *ipsec_ctx,
1693 const struct rte_cryptodev_capabilities *cipher,
1694 const struct rte_cryptodev_capabilities *auth,
1695 const struct rte_cryptodev_capabilities *aead)
1699 struct cdev_key key = { 0 };
1701 key.lcore_id = params->lcore_id;
1703 key.cipher_algo = cipher->sym.cipher.algo;
1705 key.auth_algo = auth->sym.auth.algo;
1707 key.aead_algo = aead->sym.aead.algo;
1709 ret = rte_hash_lookup(map, &key);
1713 for (i = 0; i < ipsec_ctx->nb_qps; i++)
1714 if (ipsec_ctx->tbl[i].id == cdev_id)
1717 if (i == ipsec_ctx->nb_qps) {
1718 if (ipsec_ctx->nb_qps == MAX_QP_PER_LCORE) {
1719 printf("Maximum number of crypto devices assigned to "
1720 "a core, increase MAX_QP_PER_LCORE value\n");
1723 ipsec_ctx->tbl[i].id = cdev_id;
1724 ipsec_ctx->tbl[i].qp = qp;
1725 ipsec_ctx->nb_qps++;
1726 printf("%s cdev mapping: lcore %u using cdev %u qp %u "
1727 "(cdev_id_qp %lu)\n", str, key.lcore_id,
1731 ret = rte_hash_add_key_data(map, &key, (void *)i);
1733 printf("Faled to insert cdev mapping for (lcore %u, "
1734 "cdev %u, qp %u), errno %d\n",
1735 key.lcore_id, ipsec_ctx->tbl[i].id,
1736 ipsec_ctx->tbl[i].qp, ret);
1744 add_cdev_mapping(struct rte_cryptodev_info *dev_info, uint16_t cdev_id,
1745 uint16_t qp, struct lcore_params *params)
1748 const struct rte_cryptodev_capabilities *i, *j;
1749 struct rte_hash *map;
1750 struct lcore_conf *qconf;
1751 struct ipsec_ctx *ipsec_ctx;
1754 qconf = &lcore_conf[params->lcore_id];
1756 if ((unprotected_port_mask & (1 << params->port_id)) == 0) {
1758 ipsec_ctx = &qconf->outbound;
1762 ipsec_ctx = &qconf->inbound;
1766 /* Required cryptodevs with operation chainning */
1767 if (!(dev_info->feature_flags &
1768 RTE_CRYPTODEV_FF_SYM_OPERATION_CHAINING))
1771 for (i = dev_info->capabilities;
1772 i->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; i++) {
1773 if (i->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1776 if (i->sym.xform_type == RTE_CRYPTO_SYM_XFORM_AEAD) {
1777 ret |= add_mapping(map, str, cdev_id, qp, params,
1778 ipsec_ctx, NULL, NULL, i);
1782 if (i->sym.xform_type != RTE_CRYPTO_SYM_XFORM_CIPHER)
1785 for (j = dev_info->capabilities;
1786 j->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; j++) {
1787 if (j->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1790 if (j->sym.xform_type != RTE_CRYPTO_SYM_XFORM_AUTH)
1793 ret |= add_mapping(map, str, cdev_id, qp, params,
1794 ipsec_ctx, i, j, NULL);
1801 /* Check if the device is enabled by cryptodev_mask */
1803 check_cryptodev_mask(uint8_t cdev_id)
1805 if (enabled_cryptodev_mask & (1 << cdev_id))
1812 cryptodevs_init(void)
1814 struct rte_cryptodev_config dev_conf;
1815 struct rte_cryptodev_qp_conf qp_conf;
1816 uint16_t idx, max_nb_qps, qp, i;
1818 struct rte_hash_parameters params = { 0 };
1820 const uint64_t mseg_flag = multi_seg_required() ?
1821 RTE_CRYPTODEV_FF_IN_PLACE_SGL : 0;
1823 params.entries = CDEV_MAP_ENTRIES;
1824 params.key_len = sizeof(struct cdev_key);
1825 params.hash_func = rte_jhash;
1826 params.hash_func_init_val = 0;
1827 params.socket_id = rte_socket_id();
1829 params.name = "cdev_map_in";
1830 cdev_map_in = rte_hash_create(¶ms);
1831 if (cdev_map_in == NULL)
1832 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1835 params.name = "cdev_map_out";
1836 cdev_map_out = rte_hash_create(¶ms);
1837 if (cdev_map_out == NULL)
1838 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1841 printf("lcore/cryptodev/qp mappings:\n");
1844 for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
1845 struct rte_cryptodev_info cdev_info;
1847 if (check_cryptodev_mask((uint8_t)cdev_id))
1850 rte_cryptodev_info_get(cdev_id, &cdev_info);
1852 if ((mseg_flag & cdev_info.feature_flags) != mseg_flag)
1853 rte_exit(EXIT_FAILURE,
1854 "Device %hd does not support \'%s\' feature\n",
1856 rte_cryptodev_get_feature_name(mseg_flag));
1858 if (nb_lcore_params > cdev_info.max_nb_queue_pairs)
1859 max_nb_qps = cdev_info.max_nb_queue_pairs;
1861 max_nb_qps = nb_lcore_params;
1865 while (qp < max_nb_qps && i < nb_lcore_params) {
1866 if (add_cdev_mapping(&cdev_info, cdev_id, qp,
1867 &lcore_params[idx]))
1870 idx = idx % nb_lcore_params;
1877 dev_conf.socket_id = rte_cryptodev_socket_id(cdev_id);
1878 dev_conf.nb_queue_pairs = qp;
1879 dev_conf.ff_disable = RTE_CRYPTODEV_FF_ASYMMETRIC_CRYPTO;
1881 uint32_t dev_max_sess = cdev_info.sym.max_nb_sessions;
1882 if (dev_max_sess != 0 && dev_max_sess < CDEV_MP_NB_OBJS)
1883 rte_exit(EXIT_FAILURE,
1884 "Device does not support at least %u "
1885 "sessions", CDEV_MP_NB_OBJS);
1887 if (rte_cryptodev_configure(cdev_id, &dev_conf))
1888 rte_panic("Failed to initialize cryptodev %u\n",
1891 qp_conf.nb_descriptors = CDEV_QUEUE_DESC;
1892 qp_conf.mp_session =
1893 socket_ctx[dev_conf.socket_id].session_pool;
1894 qp_conf.mp_session_private =
1895 socket_ctx[dev_conf.socket_id].session_priv_pool;
1896 for (qp = 0; qp < dev_conf.nb_queue_pairs; qp++)
1897 if (rte_cryptodev_queue_pair_setup(cdev_id, qp,
1898 &qp_conf, dev_conf.socket_id))
1899 rte_panic("Failed to setup queue %u for "
1900 "cdev_id %u\n", 0, cdev_id);
1902 if (rte_cryptodev_start(cdev_id))
1903 rte_panic("Failed to start cryptodev %u\n",
1913 port_init(uint16_t portid, uint64_t req_rx_offloads, uint64_t req_tx_offloads)
1915 uint32_t frame_size;
1916 struct rte_eth_dev_info dev_info;
1917 struct rte_eth_txconf *txconf;
1918 uint16_t nb_tx_queue, nb_rx_queue;
1919 uint16_t tx_queueid, rx_queueid, queue, lcore_id;
1920 int32_t ret, socket_id;
1921 struct lcore_conf *qconf;
1922 struct rte_ether_addr ethaddr;
1923 struct rte_eth_conf local_port_conf = port_conf;
1925 ret = rte_eth_dev_info_get(portid, &dev_info);
1927 rte_exit(EXIT_FAILURE,
1928 "Error during getting device (port %u) info: %s\n",
1929 portid, strerror(-ret));
1931 /* limit allowed HW offloafs, as user requested */
1932 dev_info.rx_offload_capa &= dev_rx_offload;
1933 dev_info.tx_offload_capa &= dev_tx_offload;
1935 printf("Configuring device port %u:\n", portid);
1937 ret = rte_eth_macaddr_get(portid, ðaddr);
1939 rte_exit(EXIT_FAILURE,
1940 "Error getting MAC address (port %u): %s\n",
1941 portid, rte_strerror(-ret));
1943 ethaddr_tbl[portid].src = ETHADDR_TO_UINT64(ðaddr);
1944 print_ethaddr("Address: ", ðaddr);
1947 nb_rx_queue = get_port_nb_rx_queues(portid);
1948 nb_tx_queue = nb_lcores;
1950 if (nb_rx_queue > dev_info.max_rx_queues)
1951 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
1952 "(max rx queue is %u)\n",
1953 nb_rx_queue, dev_info.max_rx_queues);
1955 if (nb_tx_queue > dev_info.max_tx_queues)
1956 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
1957 "(max tx queue is %u)\n",
1958 nb_tx_queue, dev_info.max_tx_queues);
1960 printf("Creating queues: nb_rx_queue=%d nb_tx_queue=%u...\n",
1961 nb_rx_queue, nb_tx_queue);
1963 frame_size = MTU_TO_FRAMELEN(mtu_size);
1964 if (frame_size > local_port_conf.rxmode.max_rx_pkt_len)
1965 local_port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
1966 local_port_conf.rxmode.max_rx_pkt_len = frame_size;
1968 if (multi_seg_required()) {
1969 local_port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_SCATTER;
1970 local_port_conf.txmode.offloads |= DEV_TX_OFFLOAD_MULTI_SEGS;
1973 local_port_conf.rxmode.offloads |= req_rx_offloads;
1974 local_port_conf.txmode.offloads |= req_tx_offloads;
1976 /* Check that all required capabilities are supported */
1977 if ((local_port_conf.rxmode.offloads & dev_info.rx_offload_capa) !=
1978 local_port_conf.rxmode.offloads)
1979 rte_exit(EXIT_FAILURE,
1980 "Error: port %u required RX offloads: 0x%" PRIx64
1981 ", avaialbe RX offloads: 0x%" PRIx64 "\n",
1982 portid, local_port_conf.rxmode.offloads,
1983 dev_info.rx_offload_capa);
1985 if ((local_port_conf.txmode.offloads & dev_info.tx_offload_capa) !=
1986 local_port_conf.txmode.offloads)
1987 rte_exit(EXIT_FAILURE,
1988 "Error: port %u required TX offloads: 0x%" PRIx64
1989 ", avaialbe TX offloads: 0x%" PRIx64 "\n",
1990 portid, local_port_conf.txmode.offloads,
1991 dev_info.tx_offload_capa);
1993 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
1994 local_port_conf.txmode.offloads |=
1995 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
1997 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_IPV4_CKSUM)
1998 local_port_conf.txmode.offloads |= DEV_TX_OFFLOAD_IPV4_CKSUM;
2000 printf("port %u configurng rx_offloads=0x%" PRIx64
2001 ", tx_offloads=0x%" PRIx64 "\n",
2002 portid, local_port_conf.rxmode.offloads,
2003 local_port_conf.txmode.offloads);
2005 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
2006 dev_info.flow_type_rss_offloads;
2007 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
2008 port_conf.rx_adv_conf.rss_conf.rss_hf) {
2009 printf("Port %u modified RSS hash function based on hardware support,"
2010 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
2012 port_conf.rx_adv_conf.rss_conf.rss_hf,
2013 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
2016 ret = rte_eth_dev_configure(portid, nb_rx_queue, nb_tx_queue,
2019 rte_exit(EXIT_FAILURE, "Cannot configure device: "
2020 "err=%d, port=%d\n", ret, portid);
2022 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd, &nb_txd);
2024 rte_exit(EXIT_FAILURE, "Cannot adjust number of descriptors: "
2025 "err=%d, port=%d\n", ret, portid);
2027 /* init one TX queue per lcore */
2029 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2030 if (rte_lcore_is_enabled(lcore_id) == 0)
2034 socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2039 printf("Setup txq=%u,%d,%d\n", lcore_id, tx_queueid, socket_id);
2041 txconf = &dev_info.default_txconf;
2042 txconf->offloads = local_port_conf.txmode.offloads;
2044 ret = rte_eth_tx_queue_setup(portid, tx_queueid, nb_txd,
2047 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
2048 "err=%d, port=%d\n", ret, portid);
2050 qconf = &lcore_conf[lcore_id];
2051 qconf->tx_queue_id[portid] = tx_queueid;
2053 /* Pre-populate pkt offloads based on capabilities */
2054 qconf->outbound.ipv4_offloads = PKT_TX_IPV4;
2055 qconf->outbound.ipv6_offloads = PKT_TX_IPV6;
2056 if (local_port_conf.txmode.offloads & DEV_TX_OFFLOAD_IPV4_CKSUM)
2057 qconf->outbound.ipv4_offloads |= PKT_TX_IP_CKSUM;
2061 /* init RX queues */
2062 for (queue = 0; queue < qconf->nb_rx_queue; ++queue) {
2063 struct rte_eth_rxconf rxq_conf;
2065 if (portid != qconf->rx_queue_list[queue].port_id)
2068 rx_queueid = qconf->rx_queue_list[queue].queue_id;
2070 printf("Setup rxq=%d,%d,%d\n", portid, rx_queueid,
2073 rxq_conf = dev_info.default_rxconf;
2074 rxq_conf.offloads = local_port_conf.rxmode.offloads;
2075 ret = rte_eth_rx_queue_setup(portid, rx_queueid,
2076 nb_rxd, socket_id, &rxq_conf,
2077 socket_ctx[socket_id].mbuf_pool);
2079 rte_exit(EXIT_FAILURE,
2080 "rte_eth_rx_queue_setup: err=%d, "
2081 "port=%d\n", ret, portid);
2088 max_session_size(void)
2092 int16_t cdev_id, port_id, n;
2095 n = rte_cryptodev_count();
2096 for (cdev_id = 0; cdev_id != n; cdev_id++) {
2097 sz = rte_cryptodev_sym_get_private_session_size(cdev_id);
2101 * If crypto device is security capable, need to check the
2102 * size of security session as well.
2105 /* Get security context of the crypto device */
2106 sec_ctx = rte_cryptodev_get_sec_ctx(cdev_id);
2107 if (sec_ctx == NULL)
2110 /* Get size of security session */
2111 sz = rte_security_session_get_size(sec_ctx);
2116 RTE_ETH_FOREACH_DEV(port_id) {
2117 if ((enabled_port_mask & (1 << port_id)) == 0)
2120 sec_ctx = rte_eth_dev_get_sec_ctx(port_id);
2121 if (sec_ctx == NULL)
2124 sz = rte_security_session_get_size(sec_ctx);
2133 session_pool_init(struct socket_ctx *ctx, int32_t socket_id, size_t sess_sz)
2135 char mp_name[RTE_MEMPOOL_NAMESIZE];
2136 struct rte_mempool *sess_mp;
2138 snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
2139 "sess_mp_%u", socket_id);
2140 sess_mp = rte_cryptodev_sym_session_pool_create(
2141 mp_name, CDEV_MP_NB_OBJS,
2142 sess_sz, CDEV_MP_CACHE_SZ, 0,
2144 ctx->session_pool = sess_mp;
2146 if (ctx->session_pool == NULL)
2147 rte_exit(EXIT_FAILURE,
2148 "Cannot init session pool on socket %d\n", socket_id);
2150 printf("Allocated session pool on socket %d\n", socket_id);
2154 session_priv_pool_init(struct socket_ctx *ctx, int32_t socket_id,
2157 char mp_name[RTE_MEMPOOL_NAMESIZE];
2158 struct rte_mempool *sess_mp;
2160 snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
2161 "sess_mp_priv_%u", socket_id);
2162 sess_mp = rte_mempool_create(mp_name,
2166 0, NULL, NULL, NULL,
2169 ctx->session_priv_pool = sess_mp;
2171 if (ctx->session_priv_pool == NULL)
2172 rte_exit(EXIT_FAILURE,
2173 "Cannot init session priv pool on socket %d\n",
2176 printf("Allocated session priv pool on socket %d\n",
2181 pool_init(struct socket_ctx *ctx, int32_t socket_id, uint32_t nb_mbuf)
2186 snprintf(s, sizeof(s), "mbuf_pool_%d", socket_id);
2187 ctx->mbuf_pool = rte_pktmbuf_pool_create(s, nb_mbuf,
2188 MEMPOOL_CACHE_SIZE, ipsec_metadata_size(),
2189 frame_buf_size, socket_id);
2192 * if multi-segment support is enabled, then create a pool
2193 * for indirect mbufs.
2195 ms = multi_seg_required();
2197 snprintf(s, sizeof(s), "mbuf_pool_indir_%d", socket_id);
2198 ctx->mbuf_pool_indir = rte_pktmbuf_pool_create(s, nb_mbuf,
2199 MEMPOOL_CACHE_SIZE, 0, 0, socket_id);
2202 if (ctx->mbuf_pool == NULL || (ms != 0 && ctx->mbuf_pool_indir == NULL))
2203 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool on socket %d\n",
2206 printf("Allocated mbuf pool on socket %d\n", socket_id);
2210 inline_ipsec_event_esn_overflow(struct rte_security_ctx *ctx, uint64_t md)
2212 struct ipsec_sa *sa;
2214 /* For inline protocol processing, the metadata in the event will
2215 * uniquely identify the security session which raised the event.
2216 * Application would then need the userdata it had registered with the
2217 * security session to process the event.
2220 sa = (struct ipsec_sa *)rte_security_get_userdata(ctx, md);
2223 /* userdata could not be retrieved */
2227 /* Sequence number over flow. SA need to be re-established */
2233 inline_ipsec_event_callback(uint16_t port_id, enum rte_eth_event_type type,
2234 void *param, void *ret_param)
2237 struct rte_eth_event_ipsec_desc *event_desc = NULL;
2238 struct rte_security_ctx *ctx = (struct rte_security_ctx *)
2239 rte_eth_dev_get_sec_ctx(port_id);
2241 RTE_SET_USED(param);
2243 if (type != RTE_ETH_EVENT_IPSEC)
2246 event_desc = ret_param;
2247 if (event_desc == NULL) {
2248 printf("Event descriptor not set\n");
2252 md = event_desc->metadata;
2254 if (event_desc->subtype == RTE_ETH_EVENT_IPSEC_ESN_OVERFLOW)
2255 return inline_ipsec_event_esn_overflow(ctx, md);
2256 else if (event_desc->subtype >= RTE_ETH_EVENT_IPSEC_MAX) {
2257 printf("Invalid IPsec event reported\n");
2265 rx_callback(__rte_unused uint16_t port, __rte_unused uint16_t queue,
2266 struct rte_mbuf *pkt[], uint16_t nb_pkts,
2267 __rte_unused uint16_t max_pkts, void *user_param)
2271 struct lcore_conf *lc;
2272 struct rte_mbuf *mb;
2273 struct rte_ether_hdr *eth;
2279 for (i = 0; i != nb_pkts; i++) {
2282 eth = rte_pktmbuf_mtod(mb, struct rte_ether_hdr *);
2283 if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
2285 struct rte_ipv4_hdr *iph;
2287 iph = (struct rte_ipv4_hdr *)(eth + 1);
2288 if (rte_ipv4_frag_pkt_is_fragmented(iph)) {
2290 mb->l2_len = sizeof(*eth);
2291 mb->l3_len = sizeof(*iph);
2292 tm = (tm != 0) ? tm : rte_rdtsc();
2293 mb = rte_ipv4_frag_reassemble_packet(
2294 lc->frag.tbl, &lc->frag.dr,
2298 /* fix ip cksum after reassemble. */
2299 iph = rte_pktmbuf_mtod_offset(mb,
2300 struct rte_ipv4_hdr *,
2302 iph->hdr_checksum = 0;
2303 iph->hdr_checksum = rte_ipv4_cksum(iph);
2306 } else if (eth->ether_type ==
2307 rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6)) {
2309 struct rte_ipv6_hdr *iph;
2310 struct ipv6_extension_fragment *fh;
2312 iph = (struct rte_ipv6_hdr *)(eth + 1);
2313 fh = rte_ipv6_frag_get_ipv6_fragment_header(iph);
2315 mb->l2_len = sizeof(*eth);
2316 mb->l3_len = (uintptr_t)fh - (uintptr_t)iph +
2318 tm = (tm != 0) ? tm : rte_rdtsc();
2319 mb = rte_ipv6_frag_reassemble_packet(
2320 lc->frag.tbl, &lc->frag.dr,
2323 /* fix l3_len after reassemble. */
2324 mb->l3_len = mb->l3_len - sizeof(*fh);
2332 /* some fragments were encountered, drain death row */
2334 rte_ip_frag_free_death_row(&lc->frag.dr, 0);
2341 reassemble_lcore_init(struct lcore_conf *lc, uint32_t cid)
2345 uint64_t frag_cycles;
2346 const struct lcore_rx_queue *rxq;
2347 const struct rte_eth_rxtx_callback *cb;
2349 /* create fragment table */
2350 sid = rte_lcore_to_socket_id(cid);
2351 frag_cycles = (rte_get_tsc_hz() + MS_PER_S - 1) /
2352 MS_PER_S * FRAG_TTL_MS;
2354 lc->frag.tbl = rte_ip_frag_table_create(frag_tbl_sz,
2355 FRAG_TBL_BUCKET_ENTRIES, frag_tbl_sz, frag_cycles, sid);
2356 if (lc->frag.tbl == NULL) {
2357 printf("%s(%u): failed to create fragment table of size: %u, "
2359 __func__, cid, frag_tbl_sz, rte_errno);
2363 /* setup reassemble RX callbacks for all queues */
2364 for (i = 0; i != lc->nb_rx_queue; i++) {
2366 rxq = lc->rx_queue_list + i;
2367 cb = rte_eth_add_rx_callback(rxq->port_id, rxq->queue_id,
2370 printf("%s(%u): failed to install RX callback for "
2371 "portid=%u, queueid=%u, error code: %d\n",
2373 rxq->port_id, rxq->queue_id, rte_errno);
2382 reassemble_init(void)
2388 for (i = 0; i != nb_lcore_params; i++) {
2389 lc = lcore_params[i].lcore_id;
2390 rc = reassemble_lcore_init(lcore_conf + lc, lc);
2399 main(int32_t argc, char **argv)
2406 uint64_t req_rx_offloads, req_tx_offloads;
2410 ret = rte_eal_init(argc, argv);
2412 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
2416 /* parse application arguments (after the EAL ones) */
2417 ret = parse_args(argc, argv);
2419 rte_exit(EXIT_FAILURE, "Invalid parameters\n");
2421 if ((unprotected_port_mask & enabled_port_mask) !=
2422 unprotected_port_mask)
2423 rte_exit(EXIT_FAILURE, "Invalid unprotected portmask 0x%x\n",
2424 unprotected_port_mask);
2426 if (check_params() < 0)
2427 rte_exit(EXIT_FAILURE, "check_params failed\n");
2429 ret = init_lcore_rx_queues();
2431 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
2433 nb_lcores = rte_lcore_count();
2435 sess_sz = max_session_size();
2437 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2438 if (rte_lcore_is_enabled(lcore_id) == 0)
2442 socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2446 /* mbuf_pool is initialised by the pool_init() function*/
2447 if (socket_ctx[socket_id].mbuf_pool)
2450 pool_init(&socket_ctx[socket_id], socket_id, NB_MBUF);
2451 session_pool_init(&socket_ctx[socket_id], socket_id, sess_sz);
2452 session_priv_pool_init(&socket_ctx[socket_id], socket_id,
2456 RTE_ETH_FOREACH_DEV(portid) {
2457 if ((enabled_port_mask & (1 << portid)) == 0)
2460 sa_check_offloads(portid, &req_rx_offloads, &req_tx_offloads);
2461 port_init(portid, req_rx_offloads, req_tx_offloads);
2467 RTE_ETH_FOREACH_DEV(portid) {
2468 if ((enabled_port_mask & (1 << portid)) == 0)
2473 * note: device must be started before a flow rule
2476 ret = rte_eth_dev_start(portid);
2478 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: "
2479 "err=%d, port=%d\n", ret, portid);
2481 * If enabled, put device in promiscuous mode.
2482 * This allows IO forwarding mode to forward packets
2483 * to itself through 2 cross-connected ports of the
2486 if (promiscuous_on) {
2487 ret = rte_eth_promiscuous_enable(portid);
2489 rte_exit(EXIT_FAILURE,
2490 "rte_eth_promiscuous_enable: err=%s, port=%d\n",
2491 rte_strerror(-ret), portid);
2494 rte_eth_dev_callback_register(portid,
2495 RTE_ETH_EVENT_IPSEC, inline_ipsec_event_callback, NULL);
2498 /* fragment reassemble is enabled */
2499 if (frag_tbl_sz != 0) {
2500 ret = reassemble_init();
2502 rte_exit(EXIT_FAILURE, "failed at reassemble init");
2505 /* Replicate each context per socket */
2506 for (i = 0; i < NB_SOCKETS && i < rte_socket_count(); i++) {
2507 socket_id = rte_socket_id_by_idx(i);
2508 if ((socket_ctx[socket_id].mbuf_pool != NULL) &&
2509 (socket_ctx[socket_id].sa_in == NULL) &&
2510 (socket_ctx[socket_id].sa_out == NULL)) {
2511 sa_init(&socket_ctx[socket_id], socket_id);
2512 sp4_init(&socket_ctx[socket_id], socket_id);
2513 sp6_init(&socket_ctx[socket_id], socket_id);
2514 rt_init(&socket_ctx[socket_id], socket_id);
2518 check_all_ports_link_status(enabled_port_mask);
2520 /* launch per-lcore init on every lcore */
2521 rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
2522 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
2523 if (rte_eal_wait_lcore(lcore_id) < 0)