1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2016-2017 Intel Corporation
5 #include <rte_atomic.h>
6 #include <rte_branch_prediction.h>
7 #include <rte_byteorder.h>
8 #include <rte_common.h>
10 #include <rte_ethdev_driver.h>
11 #include <rte_ethdev_vdev.h>
12 #include <rte_malloc.h>
13 #include <rte_bus_vdev.h>
14 #include <rte_kvargs.h>
16 #include <rte_debug.h>
18 #include <rte_string_fns.h>
20 #include <sys/types.h>
22 #include <sys/socket.h>
23 #include <sys/ioctl.h>
24 #include <sys/utsname.h>
32 #include <arpa/inet.h>
34 #include <linux/if_tun.h>
35 #include <linux/if_ether.h>
39 #include <rte_eth_tap.h>
41 #include <tap_netlink.h>
42 #include <tap_tcmsgs.h>
44 /* Linux based path to the TUN device */
45 #define TUN_TAP_DEV_PATH "/dev/net/tun"
46 #define DEFAULT_TAP_NAME "dtap"
47 #define DEFAULT_TUN_NAME "dtun"
49 #define ETH_TAP_IFACE_ARG "iface"
50 #define ETH_TAP_REMOTE_ARG "remote"
51 #define ETH_TAP_MAC_ARG "mac"
52 #define ETH_TAP_MAC_FIXED "fixed"
54 #define ETH_TAP_USR_MAC_FMT "xx:xx:xx:xx:xx:xx"
55 #define ETH_TAP_CMP_MAC_FMT "0123456789ABCDEFabcdef"
56 #define ETH_TAP_MAC_ARG_FMT ETH_TAP_MAC_FIXED "|" ETH_TAP_USR_MAC_FMT
58 static struct rte_vdev_driver pmd_tap_drv;
59 static struct rte_vdev_driver pmd_tun_drv;
61 static const char *valid_arguments[] = {
68 static unsigned int tap_unit;
69 static unsigned int tun_unit;
71 static char tuntap_name[8];
73 static volatile uint32_t tap_trigger; /* Rx trigger */
75 static struct rte_eth_link pmd_link = {
76 .link_speed = ETH_SPEED_NUM_10G,
77 .link_duplex = ETH_LINK_FULL_DUPLEX,
78 .link_status = ETH_LINK_DOWN,
79 .link_autoneg = ETH_LINK_FIXED,
83 tap_trigger_cb(int sig __rte_unused)
85 /* Valid trigger values are nonzero */
86 tap_trigger = (tap_trigger + 1) | 0x80000000;
89 /* Specifies on what netdevices the ioctl should be applied */
96 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
99 * Tun/Tap allocation routine
102 * Pointer to private structure.
104 * @param[in] is_keepalive
108 * -1 on failure, fd on success
111 tun_alloc(struct pmd_internals *pmd, int is_keepalive)
114 #ifdef IFF_MULTI_QUEUE
115 unsigned int features;
119 memset(&ifr, 0, sizeof(struct ifreq));
122 * Do not set IFF_NO_PI as packet information header will be needed
123 * to check if a received packet has been truncated.
125 ifr.ifr_flags = (pmd->type == ETH_TUNTAP_TYPE_TAP) ?
126 IFF_TAP : IFF_TUN | IFF_POINTOPOINT;
127 snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
129 TAP_LOG(DEBUG, "ifr_name '%s'", ifr.ifr_name);
131 fd = open(TUN_TAP_DEV_PATH, O_RDWR);
133 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
137 #ifdef IFF_MULTI_QUEUE
138 /* Grab the TUN features to verify we can work multi-queue */
139 if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
140 TAP_LOG(ERR, "%s unable to get TUN/TAP features",
144 TAP_LOG(DEBUG, "%s Features %08x", tuntap_name, features);
146 if (features & IFF_MULTI_QUEUE) {
147 TAP_LOG(DEBUG, " Multi-queue support for %d queues",
148 RTE_PMD_TAP_MAX_QUEUES);
149 ifr.ifr_flags |= IFF_MULTI_QUEUE;
153 ifr.ifr_flags |= IFF_ONE_QUEUE;
154 TAP_LOG(DEBUG, " Single queue only support");
157 /* Set the TUN/TAP configuration and set the name if needed */
158 if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
159 TAP_LOG(WARNING, "Unable to set TUNSETIFF for %s: %s",
160 ifr.ifr_name, strerror(errno));
166 * Detach the TUN/TAP keep-alive queue
167 * to avoid traffic through it
169 ifr.ifr_flags = IFF_DETACH_QUEUE;
170 if (ioctl(fd, TUNSETQUEUE, (void *)&ifr) < 0) {
172 "Unable to detach keep-alive queue for %s: %s",
173 ifr.ifr_name, strerror(errno));
178 /* Always set the file descriptor to non-blocking */
179 if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
181 "Unable to set %s to nonblocking: %s",
182 ifr.ifr_name, strerror(errno));
186 /* Set up trigger to optimize empty Rx bursts */
190 int flags = fcntl(fd, F_GETFL);
192 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
194 if (sa.sa_handler != tap_trigger_cb) {
196 * Make sure SIGIO is not already taken. This is done
197 * as late as possible to leave the application a
198 * chance to set up its own signal handler first.
200 if (sa.sa_handler != SIG_IGN &&
201 sa.sa_handler != SIG_DFL) {
205 sa = (struct sigaction){
206 .sa_flags = SA_RESTART,
207 .sa_handler = tap_trigger_cb,
209 if (sigaction(SIGIO, &sa, NULL) == -1)
212 /* Enable SIGIO on file descriptor */
213 fcntl(fd, F_SETFL, flags | O_ASYNC);
214 fcntl(fd, F_SETOWN, getpid());
218 /* Disable trigger globally in case of error */
220 TAP_LOG(WARNING, "Rx trigger disabled: %s",
233 tap_verify_csum(struct rte_mbuf *mbuf)
235 uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
236 uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
237 uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
238 unsigned int l2_len = sizeof(struct ether_hdr);
244 if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
246 else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
248 /* Don't verify checksum for packets with discontinuous L2 header */
249 if (unlikely(l2_len + sizeof(struct ipv4_hdr) >
250 rte_pktmbuf_data_len(mbuf)))
252 l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
253 if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
254 struct ipv4_hdr *iph = l3_hdr;
256 /* ihl contains the number of 4-byte words in the header */
257 l3_len = 4 * (iph->version_ihl & 0xf);
258 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
261 cksum = ~rte_raw_cksum(iph, l3_len);
262 mbuf->ol_flags |= cksum ?
263 PKT_RX_IP_CKSUM_BAD :
264 PKT_RX_IP_CKSUM_GOOD;
265 } else if (l3 == RTE_PTYPE_L3_IPV6) {
266 l3_len = sizeof(struct ipv6_hdr);
268 /* IPv6 extensions are not supported */
271 if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
272 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
273 /* Don't verify checksum for multi-segment packets. */
274 if (mbuf->nb_segs > 1)
276 if (l3 == RTE_PTYPE_L3_IPV4)
277 cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
278 else if (l3 == RTE_PTYPE_L3_IPV6)
279 cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
280 mbuf->ol_flags |= cksum ?
281 PKT_RX_L4_CKSUM_BAD :
282 PKT_RX_L4_CKSUM_GOOD;
287 tap_rx_offload_get_port_capa(void)
290 * No specific port Rx offload capabilities.
296 tap_rx_offload_get_queue_capa(void)
298 return DEV_RX_OFFLOAD_SCATTER |
299 DEV_RX_OFFLOAD_IPV4_CKSUM |
300 DEV_RX_OFFLOAD_UDP_CKSUM |
301 DEV_RX_OFFLOAD_TCP_CKSUM |
302 DEV_RX_OFFLOAD_CRC_STRIP;
305 /* Callback to handle the rx burst of packets to the correct interface and
306 * file descriptor(s) in a multi-queue setup.
309 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
311 struct rx_queue *rxq = queue;
313 unsigned long num_rx_bytes = 0;
314 uint32_t trigger = tap_trigger;
316 if (trigger == rxq->trigger_seen)
319 rxq->trigger_seen = trigger;
320 rte_compiler_barrier();
321 for (num_rx = 0; num_rx < nb_pkts; ) {
322 struct rte_mbuf *mbuf = rxq->pool;
323 struct rte_mbuf *seg = NULL;
324 struct rte_mbuf *new_tail = NULL;
325 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
328 len = readv(rxq->fd, *rxq->iovecs,
330 (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
331 rxq->nb_rx_desc : 1));
332 if (len < (int)sizeof(struct tun_pi))
335 /* Packet couldn't fit in the provided mbuf */
336 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
337 rxq->stats.ierrors++;
341 len -= sizeof(struct tun_pi);
344 mbuf->port = rxq->in_port;
346 struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
348 if (unlikely(!buf)) {
349 rxq->stats.rx_nombuf++;
350 /* No new buf has been allocated: do nothing */
351 if (!new_tail || !seg)
355 rte_pktmbuf_free(mbuf);
359 seg = seg ? seg->next : mbuf;
360 if (rxq->pool == mbuf)
363 new_tail->next = buf;
365 new_tail->next = seg->next;
367 /* iovecs[0] is reserved for packet info (pi) */
368 (*rxq->iovecs)[mbuf->nb_segs].iov_len =
369 buf->buf_len - data_off;
370 (*rxq->iovecs)[mbuf->nb_segs].iov_base =
371 (char *)buf->buf_addr + data_off;
373 seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
374 seg->data_off = data_off;
376 len -= seg->data_len;
380 /* First segment has headroom, not the others */
384 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
386 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
387 tap_verify_csum(mbuf);
389 /* account for the receive frame */
390 bufs[num_rx++] = mbuf;
391 num_rx_bytes += mbuf->pkt_len;
394 rxq->stats.ipackets += num_rx;
395 rxq->stats.ibytes += num_rx_bytes;
401 tap_tx_offload_get_port_capa(void)
404 * No specific port Tx offload capabilities.
410 tap_tx_offload_get_queue_capa(void)
412 return DEV_TX_OFFLOAD_MULTI_SEGS |
413 DEV_TX_OFFLOAD_IPV4_CKSUM |
414 DEV_TX_OFFLOAD_UDP_CKSUM |
415 DEV_TX_OFFLOAD_TCP_CKSUM;
419 tap_tx_offload(char *packet, uint64_t ol_flags, unsigned int l2_len,
422 void *l3_hdr = packet + l2_len;
424 if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
425 struct ipv4_hdr *iph = l3_hdr;
428 iph->hdr_checksum = 0;
429 cksum = rte_raw_cksum(iph, l3_len);
430 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
432 if (ol_flags & PKT_TX_L4_MASK) {
438 l4_hdr = packet + l2_len + l3_len;
439 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
440 l4_cksum = &((struct udp_hdr *)l4_hdr)->dgram_cksum;
441 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
442 l4_cksum = &((struct tcp_hdr *)l4_hdr)->cksum;
446 if (ol_flags & PKT_TX_IPV4) {
447 struct ipv4_hdr *iph = l3_hdr;
449 l4_len = rte_be_to_cpu_16(iph->total_length) - l3_len;
450 cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
452 struct ipv6_hdr *ip6h = l3_hdr;
454 /* payload_len does not include ext headers */
455 l4_len = rte_be_to_cpu_16(ip6h->payload_len) -
456 l3_len + sizeof(struct ipv6_hdr);
457 cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
459 cksum += rte_raw_cksum(l4_hdr, l4_len);
460 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
461 cksum = (~cksum) & 0xffff;
468 /* Callback to handle sending packets from the tap interface
471 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
473 struct tx_queue *txq = queue;
475 unsigned long num_tx_bytes = 0;
479 if (unlikely(nb_pkts == 0))
482 max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
483 for (i = 0; i < nb_pkts; i++) {
484 struct rte_mbuf *mbuf = bufs[num_tx];
485 struct iovec iovecs[mbuf->nb_segs + 1];
486 struct tun_pi pi = { .flags = 0, .proto = 0x00 };
487 struct rte_mbuf *seg = mbuf;
488 char m_copy[mbuf->data_len];
492 /* stats.errs will be incremented */
493 if (rte_pktmbuf_pkt_len(mbuf) > max_size)
496 if (txq->type == ETH_TUNTAP_TYPE_TUN) {
498 * TUN and TAP are created with IFF_NO_PI disabled.
499 * For TUN PMD this mandatory as fields are used by
500 * Kernel tun.c to determine whether its IP or non IP
503 * The logic fetches the first byte of data from mbuf
504 * then compares whether its v4 or v6. If first byte
505 * is 4 or 6, then protocol field is updated.
507 char *buff_data = rte_pktmbuf_mtod(seg, void *);
508 j = (*buff_data & 0xf0);
509 pi.proto = (j == 0x40) ? rte_cpu_to_be_16(ETHER_TYPE_IPv4) :
510 (j == 0x60) ? rte_cpu_to_be_16(ETHER_TYPE_IPv6) : 0x00;
513 iovecs[0].iov_base = π
514 iovecs[0].iov_len = sizeof(pi);
515 for (j = 1; j <= mbuf->nb_segs; j++) {
516 iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
518 rte_pktmbuf_mtod(seg, void *);
522 ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
523 (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
524 (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
525 /* Support only packets with all data in the same seg */
526 if (mbuf->nb_segs > 1)
528 /* To change checksums, work on a copy of data. */
529 rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
530 rte_pktmbuf_data_len(mbuf));
531 tap_tx_offload(m_copy, mbuf->ol_flags,
532 mbuf->l2_len, mbuf->l3_len);
533 iovecs[1].iov_base = m_copy;
535 /* copy the tx frame data */
536 n = writev(txq->fd, iovecs, mbuf->nb_segs + 1);
541 num_tx_bytes += mbuf->pkt_len;
542 rte_pktmbuf_free(mbuf);
545 txq->stats.opackets += num_tx;
546 txq->stats.errs += nb_pkts - num_tx;
547 txq->stats.obytes += num_tx_bytes;
553 tap_ioctl_req2str(unsigned long request)
557 return "SIOCSIFFLAGS";
559 return "SIOCGIFFLAGS";
561 return "SIOCGIFHWADDR";
563 return "SIOCSIFHWADDR";
571 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
572 struct ifreq *ifr, int set, enum ioctl_mode mode)
574 short req_flags = ifr->ifr_flags;
575 int remote = pmd->remote_if_index &&
576 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
578 if (!pmd->remote_if_index && mode == REMOTE_ONLY)
581 * If there is a remote netdevice, apply ioctl on it, then apply it on
586 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
587 else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
588 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
591 /* fetch current flags to leave other flags untouched */
592 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
595 ifr->ifr_flags |= req_flags;
597 ifr->ifr_flags &= ~req_flags;
605 RTE_LOG(WARNING, PMD, "%s: ioctl() called with wrong arg\n",
609 if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
611 if (remote-- && mode == LOCAL_AND_REMOTE)
616 TAP_LOG(DEBUG, "%s(%s) failed: %s(%d)", ifr->ifr_name,
617 tap_ioctl_req2str(request), strerror(errno), errno);
622 tap_link_set_down(struct rte_eth_dev *dev)
624 struct pmd_internals *pmd = dev->data->dev_private;
625 struct ifreq ifr = { .ifr_flags = IFF_UP };
627 dev->data->dev_link.link_status = ETH_LINK_DOWN;
628 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
632 tap_link_set_up(struct rte_eth_dev *dev)
634 struct pmd_internals *pmd = dev->data->dev_private;
635 struct ifreq ifr = { .ifr_flags = IFF_UP };
637 dev->data->dev_link.link_status = ETH_LINK_UP;
638 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
642 tap_dev_start(struct rte_eth_dev *dev)
646 err = tap_intr_handle_set(dev, 1);
649 return tap_link_set_up(dev);
652 /* This function gets called when the current port gets stopped.
655 tap_dev_stop(struct rte_eth_dev *dev)
657 tap_intr_handle_set(dev, 0);
658 tap_link_set_down(dev);
662 tap_dev_configure(struct rte_eth_dev *dev)
664 if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
666 "%s: number of rx queues %d exceeds max num of queues %d",
668 dev->data->nb_rx_queues,
669 RTE_PMD_TAP_MAX_QUEUES);
672 if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
674 "%s: number of tx queues %d exceeds max num of queues %d",
676 dev->data->nb_tx_queues,
677 RTE_PMD_TAP_MAX_QUEUES);
681 TAP_LOG(INFO, "%s: %p: TX configured queues number: %u",
682 dev->device->name, (void *)dev, dev->data->nb_tx_queues);
684 TAP_LOG(INFO, "%s: %p: RX configured queues number: %u",
685 dev->device->name, (void *)dev, dev->data->nb_rx_queues);
691 tap_dev_speed_capa(void)
693 uint32_t speed = pmd_link.link_speed;
696 if (speed >= ETH_SPEED_NUM_10M)
697 capa |= ETH_LINK_SPEED_10M;
698 if (speed >= ETH_SPEED_NUM_100M)
699 capa |= ETH_LINK_SPEED_100M;
700 if (speed >= ETH_SPEED_NUM_1G)
701 capa |= ETH_LINK_SPEED_1G;
702 if (speed >= ETH_SPEED_NUM_5G)
703 capa |= ETH_LINK_SPEED_2_5G;
704 if (speed >= ETH_SPEED_NUM_5G)
705 capa |= ETH_LINK_SPEED_5G;
706 if (speed >= ETH_SPEED_NUM_10G)
707 capa |= ETH_LINK_SPEED_10G;
708 if (speed >= ETH_SPEED_NUM_20G)
709 capa |= ETH_LINK_SPEED_20G;
710 if (speed >= ETH_SPEED_NUM_25G)
711 capa |= ETH_LINK_SPEED_25G;
712 if (speed >= ETH_SPEED_NUM_40G)
713 capa |= ETH_LINK_SPEED_40G;
714 if (speed >= ETH_SPEED_NUM_50G)
715 capa |= ETH_LINK_SPEED_50G;
716 if (speed >= ETH_SPEED_NUM_56G)
717 capa |= ETH_LINK_SPEED_56G;
718 if (speed >= ETH_SPEED_NUM_100G)
719 capa |= ETH_LINK_SPEED_100G;
725 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
727 struct pmd_internals *internals = dev->data->dev_private;
729 dev_info->if_index = internals->if_index;
730 dev_info->max_mac_addrs = 1;
731 dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
732 dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
733 dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
734 dev_info->min_rx_bufsize = 0;
735 dev_info->speed_capa = tap_dev_speed_capa();
736 dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
737 dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
738 dev_info->rx_queue_offload_capa;
739 dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
740 dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
741 dev_info->tx_queue_offload_capa;
742 dev_info->hash_key_size = TAP_RSS_HASH_KEY_SIZE;
744 * limitation: TAP supports all of IP, UDP and TCP hash
745 * functions together and not in partial combinations
747 dev_info->flow_type_rss_offloads = ~TAP_RSS_HF_MASK;
751 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
753 unsigned int i, imax;
754 unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
755 unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
756 unsigned long rx_nombuf = 0, ierrors = 0;
757 const struct pmd_internals *pmd = dev->data->dev_private;
759 /* rx queue statistics */
760 imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
761 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
762 for (i = 0; i < imax; i++) {
763 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
764 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
765 rx_total += tap_stats->q_ipackets[i];
766 rx_bytes_total += tap_stats->q_ibytes[i];
767 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
768 ierrors += pmd->rxq[i].stats.ierrors;
771 /* tx queue statistics */
772 imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
773 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
775 for (i = 0; i < imax; i++) {
776 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
777 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
778 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
779 tx_total += tap_stats->q_opackets[i];
780 tx_err_total += tap_stats->q_errors[i];
781 tx_bytes_total += tap_stats->q_obytes[i];
784 tap_stats->ipackets = rx_total;
785 tap_stats->ibytes = rx_bytes_total;
786 tap_stats->ierrors = ierrors;
787 tap_stats->rx_nombuf = rx_nombuf;
788 tap_stats->opackets = tx_total;
789 tap_stats->oerrors = tx_err_total;
790 tap_stats->obytes = tx_bytes_total;
795 tap_stats_reset(struct rte_eth_dev *dev)
798 struct pmd_internals *pmd = dev->data->dev_private;
800 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
801 pmd->rxq[i].stats.ipackets = 0;
802 pmd->rxq[i].stats.ibytes = 0;
803 pmd->rxq[i].stats.ierrors = 0;
804 pmd->rxq[i].stats.rx_nombuf = 0;
806 pmd->txq[i].stats.opackets = 0;
807 pmd->txq[i].stats.errs = 0;
808 pmd->txq[i].stats.obytes = 0;
813 tap_dev_close(struct rte_eth_dev *dev)
816 struct pmd_internals *internals = dev->data->dev_private;
818 tap_link_set_down(dev);
819 tap_flow_flush(dev, NULL);
820 tap_flow_implicit_flush(internals, NULL);
822 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
823 if (internals->rxq[i].fd != -1) {
824 close(internals->rxq[i].fd);
825 internals->rxq[i].fd = -1;
827 if (internals->txq[i].fd != -1) {
828 close(internals->txq[i].fd);
829 internals->txq[i].fd = -1;
833 if (internals->remote_if_index) {
834 /* Restore initial remote state */
835 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
836 &internals->remote_initial_flags);
839 if (internals->ka_fd != -1) {
840 close(internals->ka_fd);
841 internals->ka_fd = -1;
844 * Since TUN device has no more opened file descriptors
845 * it will be removed from kernel
850 tap_rx_queue_release(void *queue)
852 struct rx_queue *rxq = queue;
854 if (rxq && (rxq->fd > 0)) {
857 rte_pktmbuf_free(rxq->pool);
858 rte_free(rxq->iovecs);
865 tap_tx_queue_release(void *queue)
867 struct tx_queue *txq = queue;
869 if (txq && (txq->fd > 0)) {
876 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
878 struct rte_eth_link *dev_link = &dev->data->dev_link;
879 struct pmd_internals *pmd = dev->data->dev_private;
880 struct ifreq ifr = { .ifr_flags = 0 };
882 if (pmd->remote_if_index) {
883 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
884 if (!(ifr.ifr_flags & IFF_UP) ||
885 !(ifr.ifr_flags & IFF_RUNNING)) {
886 dev_link->link_status = ETH_LINK_DOWN;
890 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
891 dev_link->link_status =
892 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
899 tap_promisc_enable(struct rte_eth_dev *dev)
901 struct pmd_internals *pmd = dev->data->dev_private;
902 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
904 dev->data->promiscuous = 1;
905 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
906 if (pmd->remote_if_index && !pmd->flow_isolate)
907 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
911 tap_promisc_disable(struct rte_eth_dev *dev)
913 struct pmd_internals *pmd = dev->data->dev_private;
914 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
916 dev->data->promiscuous = 0;
917 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
918 if (pmd->remote_if_index && !pmd->flow_isolate)
919 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
923 tap_allmulti_enable(struct rte_eth_dev *dev)
925 struct pmd_internals *pmd = dev->data->dev_private;
926 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
928 dev->data->all_multicast = 1;
929 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
930 if (pmd->remote_if_index && !pmd->flow_isolate)
931 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
935 tap_allmulti_disable(struct rte_eth_dev *dev)
937 struct pmd_internals *pmd = dev->data->dev_private;
938 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
940 dev->data->all_multicast = 0;
941 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
942 if (pmd->remote_if_index && !pmd->flow_isolate)
943 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
947 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
949 struct pmd_internals *pmd = dev->data->dev_private;
950 enum ioctl_mode mode = LOCAL_ONLY;
954 if (pmd->type == ETH_TUNTAP_TYPE_TUN) {
955 TAP_LOG(ERR, "%s: can't MAC address for TUN",
960 if (is_zero_ether_addr(mac_addr)) {
961 TAP_LOG(ERR, "%s: can't set an empty MAC address",
965 /* Check the actual current MAC address on the tap netdevice */
966 ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY);
969 if (is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
972 /* Check the current MAC address on the remote */
973 ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY);
976 if (!is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
978 mode = LOCAL_AND_REMOTE;
979 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
980 rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
981 ret = tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode);
984 rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
985 if (pmd->remote_if_index && !pmd->flow_isolate) {
986 /* Replace MAC redirection rule after a MAC change */
987 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC);
990 "%s: Couldn't delete MAC redirection rule",
994 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC);
997 "%s: Couldn't add MAC redirection rule",
1007 tap_setup_queue(struct rte_eth_dev *dev,
1008 struct pmd_internals *internals,
1015 struct pmd_internals *pmd = dev->data->dev_private;
1016 struct rx_queue *rx = &internals->rxq[qid];
1017 struct tx_queue *tx = &internals->txq[qid];
1029 /* fd for this queue already exists */
1030 TAP_LOG(DEBUG, "%s: fd %d for %s queue qid %d exists",
1031 pmd->name, *fd, dir, qid);
1032 } else if (*other_fd != -1) {
1033 /* Only other_fd exists. dup it */
1034 *fd = dup(*other_fd);
1037 TAP_LOG(ERR, "%s: dup() failed.", pmd->name);
1040 TAP_LOG(DEBUG, "%s: dup fd %d for %s queue qid %d (%d)",
1041 pmd->name, *other_fd, dir, qid, *fd);
1043 /* Both RX and TX fds do not exist (equal -1). Create fd */
1044 *fd = tun_alloc(pmd, 0);
1046 *fd = -1; /* restore original value */
1047 TAP_LOG(ERR, "%s: tun_alloc() failed.", pmd->name);
1050 TAP_LOG(DEBUG, "%s: add %s queue for qid %d fd %d",
1051 pmd->name, dir, qid, *fd);
1054 tx->mtu = &dev->data->mtu;
1055 rx->rxmode = &dev->data->dev_conf.rxmode;
1057 tx->type = pmd->type;
1063 tap_rx_queue_setup(struct rte_eth_dev *dev,
1064 uint16_t rx_queue_id,
1065 uint16_t nb_rx_desc,
1066 unsigned int socket_id,
1067 const struct rte_eth_rxconf *rx_conf __rte_unused,
1068 struct rte_mempool *mp)
1070 struct pmd_internals *internals = dev->data->dev_private;
1071 struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1072 struct rte_mbuf **tmp = &rxq->pool;
1073 long iov_max = sysconf(_SC_IOV_MAX);
1074 uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1075 struct iovec (*iovecs)[nb_desc + 1];
1076 int data_off = RTE_PKTMBUF_HEADROOM;
1081 if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1083 "nb_rx_queues %d too small or mempool NULL",
1084 dev->data->nb_rx_queues);
1089 rxq->trigger_seen = 1; /* force initial burst */
1090 rxq->in_port = dev->data->port_id;
1091 rxq->nb_rx_desc = nb_desc;
1092 iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1096 "%s: Couldn't allocate %d RX descriptors",
1097 dev->device->name, nb_desc);
1100 rxq->iovecs = iovecs;
1102 dev->data->rx_queues[rx_queue_id] = rxq;
1103 fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1109 (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1110 (*rxq->iovecs)[0].iov_base = &rxq->pi;
1112 for (i = 1; i <= nb_desc; i++) {
1113 *tmp = rte_pktmbuf_alloc(rxq->mp);
1116 "%s: couldn't allocate memory for queue %d",
1117 dev->device->name, rx_queue_id);
1121 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1122 (*rxq->iovecs)[i].iov_base =
1123 (char *)(*tmp)->buf_addr + data_off;
1125 tmp = &(*tmp)->next;
1128 TAP_LOG(DEBUG, " RX TUNTAP device name %s, qid %d on fd %d",
1129 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
1134 rte_pktmbuf_free(rxq->pool);
1136 rte_free(rxq->iovecs);
1142 tap_tx_queue_setup(struct rte_eth_dev *dev,
1143 uint16_t tx_queue_id,
1144 uint16_t nb_tx_desc __rte_unused,
1145 unsigned int socket_id __rte_unused,
1146 const struct rte_eth_txconf *tx_conf)
1148 struct pmd_internals *internals = dev->data->dev_private;
1149 struct tx_queue *txq;
1153 if (tx_queue_id >= dev->data->nb_tx_queues)
1155 dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1156 txq = dev->data->tx_queues[tx_queue_id];
1158 offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1159 txq->csum = !!(offloads &
1160 (DEV_TX_OFFLOAD_IPV4_CKSUM |
1161 DEV_TX_OFFLOAD_UDP_CKSUM |
1162 DEV_TX_OFFLOAD_TCP_CKSUM));
1164 ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1168 " TX TUNTAP device name %s, qid %d on fd %d csum %s",
1169 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd,
1170 txq->csum ? "on" : "off");
1176 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1178 struct pmd_internals *pmd = dev->data->dev_private;
1179 struct ifreq ifr = { .ifr_mtu = mtu };
1182 err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1184 dev->data->mtu = mtu;
1190 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1191 struct ether_addr *mc_addr_set __rte_unused,
1192 uint32_t nb_mc_addr __rte_unused)
1195 * Nothing to do actually: the tap has no filtering whatsoever, every
1196 * packet is received.
1202 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1204 struct rte_eth_dev *dev = arg;
1205 struct pmd_internals *pmd = dev->data->dev_private;
1206 struct ifinfomsg *info = NLMSG_DATA(nh);
1208 if (nh->nlmsg_type != RTM_NEWLINK ||
1209 (info->ifi_index != pmd->if_index &&
1210 info->ifi_index != pmd->remote_if_index))
1212 return tap_link_update(dev, 0);
1216 tap_dev_intr_handler(void *cb_arg)
1218 struct rte_eth_dev *dev = cb_arg;
1219 struct pmd_internals *pmd = dev->data->dev_private;
1221 tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1225 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1227 struct pmd_internals *pmd = dev->data->dev_private;
1229 /* In any case, disable interrupt if the conf is no longer there. */
1230 if (!dev->data->dev_conf.intr_conf.lsc) {
1231 if (pmd->intr_handle.fd != -1) {
1232 tap_nl_final(pmd->intr_handle.fd);
1233 rte_intr_callback_unregister(&pmd->intr_handle,
1234 tap_dev_intr_handler, dev);
1239 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1240 if (unlikely(pmd->intr_handle.fd == -1))
1242 return rte_intr_callback_register(
1243 &pmd->intr_handle, tap_dev_intr_handler, dev);
1245 tap_nl_final(pmd->intr_handle.fd);
1246 return rte_intr_callback_unregister(&pmd->intr_handle,
1247 tap_dev_intr_handler, dev);
1251 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1255 err = tap_lsc_intr_handle_set(dev, set);
1258 err = tap_rx_intr_vec_set(dev, set);
1260 tap_lsc_intr_handle_set(dev, 0);
1264 static const uint32_t*
1265 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1267 static const uint32_t ptypes[] = {
1268 RTE_PTYPE_INNER_L2_ETHER,
1269 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1270 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1271 RTE_PTYPE_INNER_L3_IPV4,
1272 RTE_PTYPE_INNER_L3_IPV4_EXT,
1273 RTE_PTYPE_INNER_L3_IPV6,
1274 RTE_PTYPE_INNER_L3_IPV6_EXT,
1275 RTE_PTYPE_INNER_L4_FRAG,
1276 RTE_PTYPE_INNER_L4_UDP,
1277 RTE_PTYPE_INNER_L4_TCP,
1278 RTE_PTYPE_INNER_L4_SCTP,
1280 RTE_PTYPE_L2_ETHER_VLAN,
1281 RTE_PTYPE_L2_ETHER_QINQ,
1283 RTE_PTYPE_L3_IPV4_EXT,
1284 RTE_PTYPE_L3_IPV6_EXT,
1296 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1297 struct rte_eth_fc_conf *fc_conf)
1299 fc_conf->mode = RTE_FC_NONE;
1304 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1305 struct rte_eth_fc_conf *fc_conf)
1307 if (fc_conf->mode != RTE_FC_NONE)
1313 * DPDK callback to update the RSS hash configuration.
1316 * Pointer to Ethernet device structure.
1317 * @param[in] rss_conf
1318 * RSS configuration data.
1321 * 0 on success, a negative errno value otherwise and rte_errno is set.
1324 tap_rss_hash_update(struct rte_eth_dev *dev,
1325 struct rte_eth_rss_conf *rss_conf)
1327 if (rss_conf->rss_hf & TAP_RSS_HF_MASK) {
1331 if (rss_conf->rss_key && rss_conf->rss_key_len) {
1333 * Currently TAP RSS key is hard coded
1334 * and cannot be updated
1337 "port %u RSS key cannot be updated",
1338 dev->data->port_id);
1345 static const struct eth_dev_ops ops = {
1346 .dev_start = tap_dev_start,
1347 .dev_stop = tap_dev_stop,
1348 .dev_close = tap_dev_close,
1349 .dev_configure = tap_dev_configure,
1350 .dev_infos_get = tap_dev_info,
1351 .rx_queue_setup = tap_rx_queue_setup,
1352 .tx_queue_setup = tap_tx_queue_setup,
1353 .rx_queue_release = tap_rx_queue_release,
1354 .tx_queue_release = tap_tx_queue_release,
1355 .flow_ctrl_get = tap_flow_ctrl_get,
1356 .flow_ctrl_set = tap_flow_ctrl_set,
1357 .link_update = tap_link_update,
1358 .dev_set_link_up = tap_link_set_up,
1359 .dev_set_link_down = tap_link_set_down,
1360 .promiscuous_enable = tap_promisc_enable,
1361 .promiscuous_disable = tap_promisc_disable,
1362 .allmulticast_enable = tap_allmulti_enable,
1363 .allmulticast_disable = tap_allmulti_disable,
1364 .mac_addr_set = tap_mac_set,
1365 .mtu_set = tap_mtu_set,
1366 .set_mc_addr_list = tap_set_mc_addr_list,
1367 .stats_get = tap_stats_get,
1368 .stats_reset = tap_stats_reset,
1369 .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1370 .rss_hash_update = tap_rss_hash_update,
1371 .filter_ctrl = tap_dev_filter_ctrl,
1375 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
1376 char *remote_iface, struct ether_addr *mac_addr,
1377 enum rte_tuntap_type type)
1379 int numa_node = rte_socket_id();
1380 struct rte_eth_dev *dev;
1381 struct pmd_internals *pmd;
1382 struct rte_eth_dev_data *data;
1386 TAP_LOG(DEBUG, "%s device on numa %u",
1387 tuntap_name, rte_socket_id());
1389 dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1391 TAP_LOG(ERR, "%s Unable to allocate device struct",
1393 goto error_exit_nodev;
1396 pmd = dev->data->dev_private;
1398 snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1401 pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1402 if (pmd->ioctl_sock == -1) {
1404 "%s Unable to get a socket for management: %s",
1405 tuntap_name, strerror(errno));
1409 /* Setup some default values */
1411 data->dev_private = pmd;
1412 data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1413 data->numa_node = numa_node;
1415 data->dev_link = pmd_link;
1416 data->mac_addrs = &pmd->eth_addr;
1417 /* Set the number of RX and TX queues */
1418 data->nb_rx_queues = 0;
1419 data->nb_tx_queues = 0;
1421 dev->dev_ops = &ops;
1422 dev->rx_pkt_burst = pmd_rx_burst;
1423 dev->tx_pkt_burst = pmd_tx_burst;
1425 pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1426 pmd->intr_handle.fd = -1;
1427 dev->intr_handle = &pmd->intr_handle;
1429 /* Presetup the fds to -1 as being not valid */
1431 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1432 pmd->rxq[i].fd = -1;
1433 pmd->txq[i].fd = -1;
1436 if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1437 if (is_zero_ether_addr(mac_addr))
1438 eth_random_addr((uint8_t *)&pmd->eth_addr);
1440 rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1444 * Allocate a TUN device keep-alive file descriptor that will only be
1445 * closed when the TUN device itself is closed or removed.
1446 * This keep-alive file descriptor will guarantee that the TUN device
1447 * exists even when all of its queues are closed
1449 pmd->ka_fd = tun_alloc(pmd, 1);
1450 if (pmd->ka_fd == -1) {
1451 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
1455 ifr.ifr_mtu = dev->data->mtu;
1456 if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1459 if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1460 memset(&ifr, 0, sizeof(struct ifreq));
1461 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1462 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
1464 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1469 * Set up everything related to rte_flow:
1471 * - tap / remote if_index
1472 * - mandatory QDISCs
1473 * - rte_flow actual/implicit lists
1476 pmd->nlsk_fd = tap_nl_init(0);
1477 if (pmd->nlsk_fd == -1) {
1478 TAP_LOG(WARNING, "%s: failed to create netlink socket.",
1480 goto disable_rte_flow;
1482 pmd->if_index = if_nametoindex(pmd->name);
1483 if (!pmd->if_index) {
1484 TAP_LOG(ERR, "%s: failed to get if_index.", pmd->name);
1485 goto disable_rte_flow;
1487 if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1488 TAP_LOG(ERR, "%s: failed to create multiq qdisc.",
1490 goto disable_rte_flow;
1492 if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1493 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1495 goto disable_rte_flow;
1497 LIST_INIT(&pmd->flows);
1499 if (strlen(remote_iface)) {
1500 pmd->remote_if_index = if_nametoindex(remote_iface);
1501 if (!pmd->remote_if_index) {
1502 TAP_LOG(ERR, "%s: failed to get %s if_index.",
1503 pmd->name, remote_iface);
1506 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1507 "%s", remote_iface);
1509 /* Save state of remote device */
1510 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1512 /* Replicate remote MAC address */
1513 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1514 TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1515 pmd->name, pmd->remote_iface);
1518 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1520 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1521 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1522 TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1523 pmd->name, remote_iface);
1528 * Flush usually returns negative value because it tries to
1529 * delete every QDISC (and on a running device, one QDISC at
1530 * least is needed). Ignore negative return value.
1532 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1533 if (qdisc_create_ingress(pmd->nlsk_fd,
1534 pmd->remote_if_index) < 0) {
1535 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1539 LIST_INIT(&pmd->implicit_flows);
1540 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1541 tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1542 tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1543 tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1545 "%s: failed to create implicit rules.",
1551 rte_eth_dev_probing_finish(dev);
1555 TAP_LOG(ERR, " Disabling rte flow support: %s(%d)",
1556 strerror(errno), errno);
1557 if (strlen(remote_iface)) {
1558 TAP_LOG(ERR, "Remote feature requires flow support.");
1564 TAP_LOG(ERR, " Can't set up remote feature: %s(%d)",
1565 strerror(errno), errno);
1566 tap_flow_implicit_flush(pmd, NULL);
1569 if (pmd->ioctl_sock > 0)
1570 close(pmd->ioctl_sock);
1571 rte_eth_dev_release_port(dev);
1574 TAP_LOG(ERR, "%s Unable to initialize %s",
1575 tuntap_name, rte_vdev_device_name(vdev));
1581 set_interface_name(const char *key __rte_unused,
1585 char *name = (char *)extra_args;
1588 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN - 1);
1590 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1591 DEFAULT_TAP_NAME, (tap_unit - 1));
1597 set_remote_iface(const char *key __rte_unused,
1601 char *name = (char *)extra_args;
1604 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
1609 static int parse_user_mac(struct ether_addr *user_mac,
1612 unsigned int index = 0;
1613 char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
1615 if (user_mac == NULL || value == NULL)
1618 strlcpy(mac_temp, value, sizeof(mac_temp));
1619 mac_byte = strtok(mac_temp, ":");
1621 while ((mac_byte != NULL) &&
1622 (strlen(mac_byte) <= 2) &&
1623 (strlen(mac_byte) == strspn(mac_byte,
1624 ETH_TAP_CMP_MAC_FMT))) {
1625 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
1626 mac_byte = strtok(NULL, ":");
1633 set_mac_type(const char *key __rte_unused,
1637 struct ether_addr *user_mac = extra_args;
1642 if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
1643 static int iface_idx;
1645 /* fixed mac = 00:64:74:61:70:<iface_idx> */
1646 memcpy((char *)user_mac->addr_bytes, "\0dtap", ETHER_ADDR_LEN);
1647 user_mac->addr_bytes[ETHER_ADDR_LEN - 1] = iface_idx++ + '0';
1651 if (parse_user_mac(user_mac, value) != 6)
1654 TAP_LOG(DEBUG, "TAP user MAC param (%s)", value);
1658 TAP_LOG(ERR, "TAP user MAC (%s) is not in format (%s|%s)",
1659 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
1664 * Open a TUN interface device. TUN PMD
1665 * 1) sets tap_type as false
1666 * 2) intakes iface as argument.
1667 * 3) as interface is virtual set speed to 10G
1670 rte_pmd_tun_probe(struct rte_vdev_device *dev)
1672 const char *name, *params;
1674 struct rte_kvargs *kvlist = NULL;
1675 char tun_name[RTE_ETH_NAME_MAX_LEN];
1676 char remote_iface[RTE_ETH_NAME_MAX_LEN];
1677 struct rte_eth_dev *eth_dev;
1679 strcpy(tuntap_name, "TUN");
1681 name = rte_vdev_device_name(dev);
1682 params = rte_vdev_device_args(dev);
1683 memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1685 if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
1686 strlen(params) == 0) {
1687 eth_dev = rte_eth_dev_attach_secondary(name);
1689 TAP_LOG(ERR, "Failed to probe %s", name);
1692 eth_dev->dev_ops = &ops;
1696 snprintf(tun_name, sizeof(tun_name), "%s%u",
1697 DEFAULT_TUN_NAME, tun_unit++);
1699 if (params && (params[0] != '\0')) {
1700 TAP_LOG(DEBUG, "parameters (%s)", params);
1702 kvlist = rte_kvargs_parse(params, valid_arguments);
1704 if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1705 ret = rte_kvargs_process(kvlist,
1707 &set_interface_name,
1715 pmd_link.link_speed = ETH_SPEED_NUM_10G;
1717 TAP_LOG(NOTICE, "Initializing pmd_tun for %s as %s",
1720 ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0,
1721 ETH_TUNTAP_TYPE_TUN);
1725 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
1727 tun_unit--; /* Restore the unit number */
1729 rte_kvargs_free(kvlist);
1734 /* Open a TAP interface device.
1737 rte_pmd_tap_probe(struct rte_vdev_device *dev)
1739 const char *name, *params;
1741 struct rte_kvargs *kvlist = NULL;
1743 char tap_name[RTE_ETH_NAME_MAX_LEN];
1744 char remote_iface[RTE_ETH_NAME_MAX_LEN];
1745 struct ether_addr user_mac = { .addr_bytes = {0} };
1746 struct rte_eth_dev *eth_dev;
1748 strcpy(tuntap_name, "TAP");
1750 name = rte_vdev_device_name(dev);
1751 params = rte_vdev_device_args(dev);
1753 if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
1754 strlen(params) == 0) {
1755 eth_dev = rte_eth_dev_attach_secondary(name);
1757 TAP_LOG(ERR, "Failed to probe %s", name);
1760 /* TODO: request info from primary to set up Rx and Tx */
1761 eth_dev->dev_ops = &ops;
1762 rte_eth_dev_probing_finish(eth_dev);
1766 speed = ETH_SPEED_NUM_10G;
1767 snprintf(tap_name, sizeof(tap_name), "%s%u",
1768 DEFAULT_TAP_NAME, tap_unit++);
1769 memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1771 if (params && (params[0] != '\0')) {
1772 TAP_LOG(DEBUG, "parameters (%s)", params);
1774 kvlist = rte_kvargs_parse(params, valid_arguments);
1776 if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1777 ret = rte_kvargs_process(kvlist,
1779 &set_interface_name,
1785 if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
1786 ret = rte_kvargs_process(kvlist,
1794 if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
1795 ret = rte_kvargs_process(kvlist,
1804 pmd_link.link_speed = speed;
1806 TAP_LOG(NOTICE, "Initializing pmd_tap for %s as %s",
1809 ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac,
1810 ETH_TUNTAP_TYPE_TAP);
1814 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
1816 tap_unit--; /* Restore the unit number */
1818 rte_kvargs_free(kvlist);
1823 /* detach a TUNTAP device.
1826 rte_pmd_tap_remove(struct rte_vdev_device *dev)
1828 struct rte_eth_dev *eth_dev = NULL;
1829 struct pmd_internals *internals;
1832 /* find the ethdev entry */
1833 eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1837 internals = eth_dev->data->dev_private;
1839 TAP_LOG(DEBUG, "Closing %s Ethernet device on numa %u",
1840 (internals->type == ETH_TUNTAP_TYPE_TAP) ? "TAP" : "TUN",
1843 if (internals->nlsk_fd) {
1844 tap_flow_flush(eth_dev, NULL);
1845 tap_flow_implicit_flush(internals, NULL);
1846 tap_nl_final(internals->nlsk_fd);
1848 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1849 if (internals->rxq[i].fd != -1) {
1850 close(internals->rxq[i].fd);
1851 internals->rxq[i].fd = -1;
1853 if (internals->txq[i].fd != -1) {
1854 close(internals->txq[i].fd);
1855 internals->txq[i].fd = -1;
1859 close(internals->ioctl_sock);
1860 rte_free(eth_dev->data->dev_private);
1861 rte_eth_dev_release_port(eth_dev);
1863 if (internals->ka_fd != -1) {
1864 close(internals->ka_fd);
1865 internals->ka_fd = -1;
1870 static struct rte_vdev_driver pmd_tun_drv = {
1871 .probe = rte_pmd_tun_probe,
1872 .remove = rte_pmd_tap_remove,
1875 static struct rte_vdev_driver pmd_tap_drv = {
1876 .probe = rte_pmd_tap_probe,
1877 .remove = rte_pmd_tap_remove,
1880 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1881 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
1882 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1883 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
1884 ETH_TAP_IFACE_ARG "=<string> ");
1885 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
1886 ETH_TAP_IFACE_ARG "=<string> "
1887 ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
1888 ETH_TAP_REMOTE_ARG "=<string>");
1891 RTE_INIT(tap_init_log);
1895 tap_logtype = rte_log_register("pmd.net.tap");
1896 if (tap_logtype >= 0)
1897 rte_log_set_level(tap_logtype, RTE_LOG_NOTICE);