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>
19 #include <sys/types.h>
21 #include <sys/socket.h>
22 #include <sys/ioctl.h>
23 #include <sys/utsname.h>
31 #include <arpa/inet.h>
33 #include <linux/if_tun.h>
34 #include <linux/if_ether.h>
37 #include <rte_eth_tap.h>
39 #include <tap_netlink.h>
40 #include <tap_tcmsgs.h>
42 /* Linux based path to the TUN device */
43 #define TUN_TAP_DEV_PATH "/dev/net/tun"
44 #define DEFAULT_TAP_NAME "dtap"
46 #define ETH_TAP_IFACE_ARG "iface"
47 #define ETH_TAP_REMOTE_ARG "remote"
48 #define ETH_TAP_MAC_ARG "mac"
49 #define ETH_TAP_MAC_FIXED "fixed"
51 static struct rte_vdev_driver pmd_tap_drv;
53 static const char *valid_arguments[] = {
62 static volatile uint32_t tap_trigger; /* Rx trigger */
64 static struct rte_eth_link pmd_link = {
65 .link_speed = ETH_SPEED_NUM_10G,
66 .link_duplex = ETH_LINK_FULL_DUPLEX,
67 .link_status = ETH_LINK_DOWN,
68 .link_autoneg = ETH_LINK_AUTONEG
72 tap_trigger_cb(int sig __rte_unused)
74 /* Valid trigger values are nonzero */
75 tap_trigger = (tap_trigger + 1) | 0x80000000;
78 /* Specifies on what netdevices the ioctl should be applied */
85 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
87 /* Tun/Tap allocation routine
89 * name is the number of the interface to use, unless NULL to take the host
93 tun_alloc(struct pmd_internals *pmd)
96 #ifdef IFF_MULTI_QUEUE
97 unsigned int features;
101 memset(&ifr, 0, sizeof(struct ifreq));
104 * Do not set IFF_NO_PI as packet information header will be needed
105 * to check if a received packet has been truncated.
107 ifr.ifr_flags = IFF_TAP;
108 snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
110 RTE_LOG(DEBUG, PMD, "ifr_name '%s'\n", ifr.ifr_name);
112 fd = open(TUN_TAP_DEV_PATH, O_RDWR);
114 RTE_LOG(ERR, PMD, "Unable to create TAP interface\n");
118 #ifdef IFF_MULTI_QUEUE
119 /* Grab the TUN features to verify we can work multi-queue */
120 if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
121 RTE_LOG(ERR, PMD, "TAP unable to get TUN/TAP features\n");
124 RTE_LOG(DEBUG, PMD, " TAP Features %08x\n", features);
126 if (features & IFF_MULTI_QUEUE) {
127 RTE_LOG(DEBUG, PMD, " Multi-queue support for %d queues\n",
128 RTE_PMD_TAP_MAX_QUEUES);
129 ifr.ifr_flags |= IFF_MULTI_QUEUE;
133 ifr.ifr_flags |= IFF_ONE_QUEUE;
134 RTE_LOG(DEBUG, PMD, " Single queue only support\n");
137 /* Set the TUN/TAP configuration and set the name if needed */
138 if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
139 RTE_LOG(WARNING, PMD,
140 "Unable to set TUNSETIFF for %s\n",
146 /* Always set the file descriptor to non-blocking */
147 if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
148 RTE_LOG(WARNING, PMD,
149 "Unable to set %s to nonblocking\n",
151 perror("F_SETFL, NONBLOCK");
155 /* Set up trigger to optimize empty Rx bursts */
159 int flags = fcntl(fd, F_GETFL);
161 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
163 if (sa.sa_handler != tap_trigger_cb) {
165 * Make sure SIGIO is not already taken. This is done
166 * as late as possible to leave the application a
167 * chance to set up its own signal handler first.
169 if (sa.sa_handler != SIG_IGN &&
170 sa.sa_handler != SIG_DFL) {
174 sa = (struct sigaction){
175 .sa_flags = SA_RESTART,
176 .sa_handler = tap_trigger_cb,
178 if (sigaction(SIGIO, &sa, NULL) == -1)
181 /* Enable SIGIO on file descriptor */
182 fcntl(fd, F_SETFL, flags | O_ASYNC);
183 fcntl(fd, F_SETOWN, getpid());
186 /* Disable trigger globally in case of error */
188 RTE_LOG(WARNING, PMD, "Rx trigger disabled: %s\n",
201 tap_verify_csum(struct rte_mbuf *mbuf)
203 uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
204 uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
205 uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
206 unsigned int l2_len = sizeof(struct ether_hdr);
212 if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
214 else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
216 /* Don't verify checksum for packets with discontinuous L2 header */
217 if (unlikely(l2_len + sizeof(struct ipv4_hdr) >
218 rte_pktmbuf_data_len(mbuf)))
220 l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
221 if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
222 struct ipv4_hdr *iph = l3_hdr;
224 /* ihl contains the number of 4-byte words in the header */
225 l3_len = 4 * (iph->version_ihl & 0xf);
226 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
229 cksum = ~rte_raw_cksum(iph, l3_len);
230 mbuf->ol_flags |= cksum ?
231 PKT_RX_IP_CKSUM_BAD :
232 PKT_RX_IP_CKSUM_GOOD;
233 } else if (l3 == RTE_PTYPE_L3_IPV6) {
234 l3_len = sizeof(struct ipv6_hdr);
236 /* IPv6 extensions are not supported */
239 if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
240 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
241 /* Don't verify checksum for multi-segment packets. */
242 if (mbuf->nb_segs > 1)
244 if (l3 == RTE_PTYPE_L3_IPV4)
245 cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
246 else if (l3 == RTE_PTYPE_L3_IPV6)
247 cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
248 mbuf->ol_flags |= cksum ?
249 PKT_RX_L4_CKSUM_BAD :
250 PKT_RX_L4_CKSUM_GOOD;
255 tap_rx_offload_get_port_capa(void)
258 * In order to support legacy apps,
259 * report capabilities also as port capabilities.
261 return DEV_RX_OFFLOAD_SCATTER |
262 DEV_RX_OFFLOAD_IPV4_CKSUM |
263 DEV_RX_OFFLOAD_UDP_CKSUM |
264 DEV_RX_OFFLOAD_TCP_CKSUM |
265 DEV_RX_OFFLOAD_CRC_STRIP;
269 tap_rx_offload_get_queue_capa(void)
271 return DEV_RX_OFFLOAD_SCATTER |
272 DEV_RX_OFFLOAD_IPV4_CKSUM |
273 DEV_RX_OFFLOAD_UDP_CKSUM |
274 DEV_RX_OFFLOAD_TCP_CKSUM |
275 DEV_RX_OFFLOAD_CRC_STRIP;
279 tap_rxq_are_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
281 uint64_t port_offloads = dev->data->dev_conf.rxmode.offloads;
282 uint64_t queue_supp_offloads = tap_rx_offload_get_queue_capa();
283 uint64_t port_supp_offloads = tap_rx_offload_get_port_capa();
285 if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
288 if ((port_offloads ^ offloads) & port_supp_offloads)
293 /* Callback to handle the rx burst of packets to the correct interface and
294 * file descriptor(s) in a multi-queue setup.
297 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
299 struct rx_queue *rxq = queue;
301 unsigned long num_rx_bytes = 0;
302 uint32_t trigger = tap_trigger;
304 if (trigger == rxq->trigger_seen)
307 rxq->trigger_seen = trigger;
308 rte_compiler_barrier();
309 for (num_rx = 0; num_rx < nb_pkts; ) {
310 struct rte_mbuf *mbuf = rxq->pool;
311 struct rte_mbuf *seg = NULL;
312 struct rte_mbuf *new_tail = NULL;
313 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
316 len = readv(rxq->fd, *rxq->iovecs,
318 (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
319 rxq->nb_rx_desc : 1));
320 if (len < (int)sizeof(struct tun_pi))
323 /* Packet couldn't fit in the provided mbuf */
324 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
325 rxq->stats.ierrors++;
329 len -= sizeof(struct tun_pi);
332 mbuf->port = rxq->in_port;
334 struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
336 if (unlikely(!buf)) {
337 rxq->stats.rx_nombuf++;
338 /* No new buf has been allocated: do nothing */
339 if (!new_tail || !seg)
343 rte_pktmbuf_free(mbuf);
347 seg = seg ? seg->next : mbuf;
348 if (rxq->pool == mbuf)
351 new_tail->next = buf;
353 new_tail->next = seg->next;
355 /* iovecs[0] is reserved for packet info (pi) */
356 (*rxq->iovecs)[mbuf->nb_segs].iov_len =
357 buf->buf_len - data_off;
358 (*rxq->iovecs)[mbuf->nb_segs].iov_base =
359 (char *)buf->buf_addr + data_off;
361 seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
362 seg->data_off = data_off;
364 len -= seg->data_len;
368 /* First segment has headroom, not the others */
372 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
374 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
375 tap_verify_csum(mbuf);
377 /* account for the receive frame */
378 bufs[num_rx++] = mbuf;
379 num_rx_bytes += mbuf->pkt_len;
382 rxq->stats.ipackets += num_rx;
383 rxq->stats.ibytes += num_rx_bytes;
389 tap_tx_offload_get_port_capa(void)
392 * In order to support legacy apps,
393 * report capabilities also as port capabilities.
395 return DEV_TX_OFFLOAD_MULTI_SEGS |
396 DEV_TX_OFFLOAD_IPV4_CKSUM |
397 DEV_TX_OFFLOAD_UDP_CKSUM |
398 DEV_TX_OFFLOAD_TCP_CKSUM;
402 tap_tx_offload_get_queue_capa(void)
404 return DEV_TX_OFFLOAD_MULTI_SEGS |
405 DEV_TX_OFFLOAD_IPV4_CKSUM |
406 DEV_TX_OFFLOAD_UDP_CKSUM |
407 DEV_TX_OFFLOAD_TCP_CKSUM;
411 tap_txq_are_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
413 uint64_t port_offloads = dev->data->dev_conf.txmode.offloads;
414 uint64_t queue_supp_offloads = tap_tx_offload_get_queue_capa();
415 uint64_t port_supp_offloads = tap_tx_offload_get_port_capa();
417 if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
420 /* Verify we have no conflict with port offloads */
421 if ((port_offloads ^ offloads) & port_supp_offloads)
427 tap_tx_offload(char *packet, uint64_t ol_flags, unsigned int l2_len,
430 void *l3_hdr = packet + l2_len;
432 if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
433 struct ipv4_hdr *iph = l3_hdr;
436 iph->hdr_checksum = 0;
437 cksum = rte_raw_cksum(iph, l3_len);
438 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
440 if (ol_flags & PKT_TX_L4_MASK) {
446 l4_hdr = packet + l2_len + l3_len;
447 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
448 l4_cksum = &((struct udp_hdr *)l4_hdr)->dgram_cksum;
449 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
450 l4_cksum = &((struct tcp_hdr *)l4_hdr)->cksum;
454 if (ol_flags & PKT_TX_IPV4) {
455 struct ipv4_hdr *iph = l3_hdr;
457 l4_len = rte_be_to_cpu_16(iph->total_length) - l3_len;
458 cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
460 struct ipv6_hdr *ip6h = l3_hdr;
462 /* payload_len does not include ext headers */
463 l4_len = rte_be_to_cpu_16(ip6h->payload_len) -
464 l3_len + sizeof(struct ipv6_hdr);
465 cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
467 cksum += rte_raw_cksum(l4_hdr, l4_len);
468 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
469 cksum = (~cksum) & 0xffff;
476 /* Callback to handle sending packets from the tap interface
479 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
481 struct tx_queue *txq = queue;
483 unsigned long num_tx_bytes = 0;
487 if (unlikely(nb_pkts == 0))
490 max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
491 for (i = 0; i < nb_pkts; i++) {
492 struct rte_mbuf *mbuf = bufs[num_tx];
493 struct iovec iovecs[mbuf->nb_segs + 1];
494 struct tun_pi pi = { .flags = 0 };
495 struct rte_mbuf *seg = mbuf;
496 char m_copy[mbuf->data_len];
500 /* stats.errs will be incremented */
501 if (rte_pktmbuf_pkt_len(mbuf) > max_size)
504 iovecs[0].iov_base = π
505 iovecs[0].iov_len = sizeof(pi);
506 for (j = 1; j <= mbuf->nb_segs; j++) {
507 iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
509 rte_pktmbuf_mtod(seg, void *);
513 ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
514 (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
515 (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
516 /* Support only packets with all data in the same seg */
517 if (mbuf->nb_segs > 1)
519 /* To change checksums, work on a copy of data. */
520 rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
521 rte_pktmbuf_data_len(mbuf));
522 tap_tx_offload(m_copy, mbuf->ol_flags,
523 mbuf->l2_len, mbuf->l3_len);
524 iovecs[1].iov_base = m_copy;
526 /* copy the tx frame data */
527 n = writev(txq->fd, iovecs, mbuf->nb_segs + 1);
532 num_tx_bytes += mbuf->pkt_len;
533 rte_pktmbuf_free(mbuf);
536 txq->stats.opackets += num_tx;
537 txq->stats.errs += nb_pkts - num_tx;
538 txq->stats.obytes += num_tx_bytes;
544 tap_ioctl_req2str(unsigned long request)
548 return "SIOCSIFFLAGS";
550 return "SIOCGIFFLAGS";
552 return "SIOCGIFHWADDR";
554 return "SIOCSIFHWADDR";
562 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
563 struct ifreq *ifr, int set, enum ioctl_mode mode)
565 short req_flags = ifr->ifr_flags;
566 int remote = pmd->remote_if_index &&
567 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
569 if (!pmd->remote_if_index && mode == REMOTE_ONLY)
572 * If there is a remote netdevice, apply ioctl on it, then apply it on
577 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
578 else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
579 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
582 /* fetch current flags to leave other flags untouched */
583 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
586 ifr->ifr_flags |= req_flags;
588 ifr->ifr_flags &= ~req_flags;
596 RTE_ASSERT(!"unsupported request type: must not happen");
598 if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
600 if (remote-- && mode == LOCAL_AND_REMOTE)
605 RTE_LOG(DEBUG, PMD, "%s: %s(%s) failed: %s(%d)\n", ifr->ifr_name,
606 __func__, tap_ioctl_req2str(request), strerror(errno), errno);
611 tap_link_set_down(struct rte_eth_dev *dev)
613 struct pmd_internals *pmd = dev->data->dev_private;
614 struct ifreq ifr = { .ifr_flags = IFF_UP };
616 dev->data->dev_link.link_status = ETH_LINK_DOWN;
617 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
621 tap_link_set_up(struct rte_eth_dev *dev)
623 struct pmd_internals *pmd = dev->data->dev_private;
624 struct ifreq ifr = { .ifr_flags = IFF_UP };
626 dev->data->dev_link.link_status = ETH_LINK_UP;
627 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
631 tap_dev_start(struct rte_eth_dev *dev)
635 err = tap_intr_handle_set(dev, 1);
638 return tap_link_set_up(dev);
641 /* This function gets called when the current port gets stopped.
644 tap_dev_stop(struct rte_eth_dev *dev)
646 tap_intr_handle_set(dev, 0);
647 tap_link_set_down(dev);
651 tap_dev_configure(struct rte_eth_dev *dev)
653 uint64_t supp_tx_offloads = tap_tx_offload_get_port_capa();
654 uint64_t tx_offloads = dev->data->dev_conf.txmode.offloads;
656 if ((tx_offloads & supp_tx_offloads) != tx_offloads) {
659 "Some Tx offloads are not supported "
660 "requested 0x%" PRIx64 " supported 0x%" PRIx64 "\n",
661 tx_offloads, supp_tx_offloads);
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\n",
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\n",
676 dev->data->nb_tx_queues,
677 RTE_PMD_TAP_MAX_QUEUES);
681 RTE_LOG(INFO, PMD, "%s: %p: TX configured queues number: %u\n",
682 dev->device->name, (void *)dev, dev->data->nb_tx_queues);
684 RTE_LOG(INFO, PMD, "%s: %p: RX configured queues number: %u\n",
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->pci_dev = NULL;
736 dev_info->speed_capa = tap_dev_speed_capa();
737 dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
738 dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
739 dev_info->rx_queue_offload_capa;
740 dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
741 dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
742 dev_info->tx_queue_offload_capa;
746 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
748 unsigned int i, imax;
749 unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
750 unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
751 unsigned long rx_nombuf = 0, ierrors = 0;
752 const struct pmd_internals *pmd = dev->data->dev_private;
754 /* rx queue statistics */
755 imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
756 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
757 for (i = 0; i < imax; i++) {
758 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
759 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
760 rx_total += tap_stats->q_ipackets[i];
761 rx_bytes_total += tap_stats->q_ibytes[i];
762 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
763 ierrors += pmd->rxq[i].stats.ierrors;
766 /* tx queue statistics */
767 imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
768 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
770 for (i = 0; i < imax; i++) {
771 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
772 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
773 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
774 tx_total += tap_stats->q_opackets[i];
775 tx_err_total += tap_stats->q_errors[i];
776 tx_bytes_total += tap_stats->q_obytes[i];
779 tap_stats->ipackets = rx_total;
780 tap_stats->ibytes = rx_bytes_total;
781 tap_stats->ierrors = ierrors;
782 tap_stats->rx_nombuf = rx_nombuf;
783 tap_stats->opackets = tx_total;
784 tap_stats->oerrors = tx_err_total;
785 tap_stats->obytes = tx_bytes_total;
790 tap_stats_reset(struct rte_eth_dev *dev)
793 struct pmd_internals *pmd = dev->data->dev_private;
795 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
796 pmd->rxq[i].stats.ipackets = 0;
797 pmd->rxq[i].stats.ibytes = 0;
798 pmd->rxq[i].stats.ierrors = 0;
799 pmd->rxq[i].stats.rx_nombuf = 0;
801 pmd->txq[i].stats.opackets = 0;
802 pmd->txq[i].stats.errs = 0;
803 pmd->txq[i].stats.obytes = 0;
808 tap_dev_close(struct rte_eth_dev *dev)
811 struct pmd_internals *internals = dev->data->dev_private;
813 tap_link_set_down(dev);
814 tap_flow_flush(dev, NULL);
815 tap_flow_implicit_flush(internals, NULL);
817 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
818 if (internals->rxq[i].fd != -1) {
819 close(internals->rxq[i].fd);
820 internals->rxq[i].fd = -1;
822 if (internals->txq[i].fd != -1) {
823 close(internals->txq[i].fd);
824 internals->txq[i].fd = -1;
828 if (internals->remote_if_index) {
829 /* Restore initial remote state */
830 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
831 &internals->remote_initial_flags);
836 tap_rx_queue_release(void *queue)
838 struct rx_queue *rxq = queue;
840 if (rxq && (rxq->fd > 0)) {
843 rte_pktmbuf_free(rxq->pool);
844 rte_free(rxq->iovecs);
851 tap_tx_queue_release(void *queue)
853 struct tx_queue *txq = queue;
855 if (txq && (txq->fd > 0)) {
862 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
864 struct rte_eth_link *dev_link = &dev->data->dev_link;
865 struct pmd_internals *pmd = dev->data->dev_private;
866 struct ifreq ifr = { .ifr_flags = 0 };
868 if (pmd->remote_if_index) {
869 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
870 if (!(ifr.ifr_flags & IFF_UP) ||
871 !(ifr.ifr_flags & IFF_RUNNING)) {
872 dev_link->link_status = ETH_LINK_DOWN;
876 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
877 dev_link->link_status =
878 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
885 tap_promisc_enable(struct rte_eth_dev *dev)
887 struct pmd_internals *pmd = dev->data->dev_private;
888 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
890 dev->data->promiscuous = 1;
891 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
892 if (pmd->remote_if_index && !pmd->flow_isolate)
893 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
897 tap_promisc_disable(struct rte_eth_dev *dev)
899 struct pmd_internals *pmd = dev->data->dev_private;
900 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
902 dev->data->promiscuous = 0;
903 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
904 if (pmd->remote_if_index && !pmd->flow_isolate)
905 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
909 tap_allmulti_enable(struct rte_eth_dev *dev)
911 struct pmd_internals *pmd = dev->data->dev_private;
912 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
914 dev->data->all_multicast = 1;
915 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
916 if (pmd->remote_if_index && !pmd->flow_isolate)
917 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
921 tap_allmulti_disable(struct rte_eth_dev *dev)
923 struct pmd_internals *pmd = dev->data->dev_private;
924 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
926 dev->data->all_multicast = 0;
927 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
928 if (pmd->remote_if_index && !pmd->flow_isolate)
929 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
933 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
935 struct pmd_internals *pmd = dev->data->dev_private;
936 enum ioctl_mode mode = LOCAL_ONLY;
939 if (is_zero_ether_addr(mac_addr)) {
940 RTE_LOG(ERR, PMD, "%s: can't set an empty MAC address\n",
944 /* Check the actual current MAC address on the tap netdevice */
945 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
947 if (is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
950 /* Check the current MAC address on the remote */
951 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0)
953 if (!is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
955 mode = LOCAL_AND_REMOTE;
956 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
957 rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
958 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode) < 0)
960 rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
961 if (pmd->remote_if_index && !pmd->flow_isolate) {
962 /* Replace MAC redirection rule after a MAC change */
963 if (tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC) < 0) {
965 "%s: Couldn't delete MAC redirection rule\n",
969 if (tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0)
971 "%s: Couldn't add MAC redirection rule\n",
977 tap_setup_queue(struct rte_eth_dev *dev,
978 struct pmd_internals *internals,
985 struct pmd_internals *pmd = dev->data->dev_private;
986 struct rx_queue *rx = &internals->rxq[qid];
987 struct tx_queue *tx = &internals->txq[qid];
999 /* fd for this queue already exists */
1000 RTE_LOG(DEBUG, PMD, "%s: fd %d for %s queue qid %d exists\n",
1001 pmd->name, *fd, dir, qid);
1002 } else if (*other_fd != -1) {
1003 /* Only other_fd exists. dup it */
1004 *fd = dup(*other_fd);
1007 RTE_LOG(ERR, PMD, "%s: dup() failed.\n",
1011 RTE_LOG(DEBUG, PMD, "%s: dup fd %d for %s queue qid %d (%d)\n",
1012 pmd->name, *other_fd, dir, qid, *fd);
1014 /* Both RX and TX fds do not exist (equal -1). Create fd */
1015 *fd = tun_alloc(pmd);
1017 *fd = -1; /* restore original value */
1018 RTE_LOG(ERR, PMD, "%s: tun_alloc() failed.\n",
1022 RTE_LOG(DEBUG, PMD, "%s: add %s queue for qid %d fd %d\n",
1023 pmd->name, dir, qid, *fd);
1026 tx->mtu = &dev->data->mtu;
1027 rx->rxmode = &dev->data->dev_conf.rxmode;
1033 tap_rx_queue_setup(struct rte_eth_dev *dev,
1034 uint16_t rx_queue_id,
1035 uint16_t nb_rx_desc,
1036 unsigned int socket_id,
1037 const struct rte_eth_rxconf *rx_conf __rte_unused,
1038 struct rte_mempool *mp)
1040 struct pmd_internals *internals = dev->data->dev_private;
1041 struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1042 struct rte_mbuf **tmp = &rxq->pool;
1043 long iov_max = sysconf(_SC_IOV_MAX);
1044 uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1045 struct iovec (*iovecs)[nb_desc + 1];
1046 int data_off = RTE_PKTMBUF_HEADROOM;
1051 if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1052 RTE_LOG(WARNING, PMD,
1053 "nb_rx_queues %d too small or mempool NULL\n",
1054 dev->data->nb_rx_queues);
1058 /* Verify application offloads are valid for our port and queue. */
1059 if (!tap_rxq_are_offloads_valid(dev, rx_conf->offloads)) {
1060 rte_errno = ENOTSUP;
1062 "%p: Rx queue offloads 0x%" PRIx64
1063 " don't match port offloads 0x%" PRIx64
1064 " or supported offloads 0x%" PRIx64 "\n",
1065 (void *)dev, rx_conf->offloads,
1066 dev->data->dev_conf.rxmode.offloads,
1067 (tap_rx_offload_get_port_capa() |
1068 tap_rx_offload_get_queue_capa()));
1072 rxq->trigger_seen = 1; /* force initial burst */
1073 rxq->in_port = dev->data->port_id;
1074 rxq->nb_rx_desc = nb_desc;
1075 iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1078 RTE_LOG(WARNING, PMD,
1079 "%s: Couldn't allocate %d RX descriptors\n",
1080 dev->device->name, nb_desc);
1083 rxq->iovecs = iovecs;
1085 dev->data->rx_queues[rx_queue_id] = rxq;
1086 fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1092 (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1093 (*rxq->iovecs)[0].iov_base = &rxq->pi;
1095 for (i = 1; i <= nb_desc; i++) {
1096 *tmp = rte_pktmbuf_alloc(rxq->mp);
1098 RTE_LOG(WARNING, PMD,
1099 "%s: couldn't allocate memory for queue %d\n",
1100 dev->device->name, rx_queue_id);
1104 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1105 (*rxq->iovecs)[i].iov_base =
1106 (char *)(*tmp)->buf_addr + data_off;
1108 tmp = &(*tmp)->next;
1111 RTE_LOG(DEBUG, PMD, " RX TAP device name %s, qid %d on fd %d\n",
1112 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
1117 rte_pktmbuf_free(rxq->pool);
1119 rte_free(rxq->iovecs);
1125 tap_tx_queue_setup(struct rte_eth_dev *dev,
1126 uint16_t tx_queue_id,
1127 uint16_t nb_tx_desc __rte_unused,
1128 unsigned int socket_id __rte_unused,
1129 const struct rte_eth_txconf *tx_conf)
1131 struct pmd_internals *internals = dev->data->dev_private;
1132 struct tx_queue *txq;
1135 if (tx_queue_id >= dev->data->nb_tx_queues)
1137 dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1138 txq = dev->data->tx_queues[tx_queue_id];
1140 * Don't verify port offloads for application which
1143 if (tx_conf != NULL &&
1144 !!(tx_conf->txq_flags & ETH_TXQ_FLAGS_IGNORE)) {
1145 if (tap_txq_are_offloads_valid(dev, tx_conf->offloads)) {
1146 txq->csum = !!(tx_conf->offloads &
1147 (DEV_TX_OFFLOAD_IPV4_CKSUM |
1148 DEV_TX_OFFLOAD_UDP_CKSUM |
1149 DEV_TX_OFFLOAD_TCP_CKSUM));
1151 rte_errno = ENOTSUP;
1153 "%p: Tx queue offloads 0x%" PRIx64
1154 " don't match port offloads 0x%" PRIx64
1155 " or supported offloads 0x%" PRIx64,
1156 (void *)dev, tx_conf->offloads,
1157 dev->data->dev_conf.txmode.offloads,
1158 tap_tx_offload_get_port_capa());
1162 ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1166 " TX TAP device name %s, qid %d on fd %d csum %s\n",
1167 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd,
1168 txq->csum ? "on" : "off");
1174 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1176 struct pmd_internals *pmd = dev->data->dev_private;
1177 struct ifreq ifr = { .ifr_mtu = mtu };
1180 err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1182 dev->data->mtu = mtu;
1188 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1189 struct ether_addr *mc_addr_set __rte_unused,
1190 uint32_t nb_mc_addr __rte_unused)
1193 * Nothing to do actually: the tap has no filtering whatsoever, every
1194 * packet is received.
1200 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1202 struct rte_eth_dev *dev = arg;
1203 struct pmd_internals *pmd = dev->data->dev_private;
1204 struct ifinfomsg *info = NLMSG_DATA(nh);
1206 if (nh->nlmsg_type != RTM_NEWLINK ||
1207 (info->ifi_index != pmd->if_index &&
1208 info->ifi_index != pmd->remote_if_index))
1210 return tap_link_update(dev, 0);
1214 tap_dev_intr_handler(void *cb_arg)
1216 struct rte_eth_dev *dev = cb_arg;
1217 struct pmd_internals *pmd = dev->data->dev_private;
1219 tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1223 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1225 struct pmd_internals *pmd = dev->data->dev_private;
1227 /* In any case, disable interrupt if the conf is no longer there. */
1228 if (!dev->data->dev_conf.intr_conf.lsc) {
1229 if (pmd->intr_handle.fd != -1) {
1230 tap_nl_final(pmd->intr_handle.fd);
1231 rte_intr_callback_unregister(&pmd->intr_handle,
1232 tap_dev_intr_handler, dev);
1237 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1238 if (unlikely(pmd->intr_handle.fd == -1))
1240 return rte_intr_callback_register(
1241 &pmd->intr_handle, tap_dev_intr_handler, dev);
1243 tap_nl_final(pmd->intr_handle.fd);
1244 return rte_intr_callback_unregister(&pmd->intr_handle,
1245 tap_dev_intr_handler, dev);
1249 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1253 err = tap_lsc_intr_handle_set(dev, set);
1256 err = tap_rx_intr_vec_set(dev, set);
1258 tap_lsc_intr_handle_set(dev, 0);
1262 static const uint32_t*
1263 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1265 static const uint32_t ptypes[] = {
1266 RTE_PTYPE_INNER_L2_ETHER,
1267 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1268 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1269 RTE_PTYPE_INNER_L3_IPV4,
1270 RTE_PTYPE_INNER_L3_IPV4_EXT,
1271 RTE_PTYPE_INNER_L3_IPV6,
1272 RTE_PTYPE_INNER_L3_IPV6_EXT,
1273 RTE_PTYPE_INNER_L4_FRAG,
1274 RTE_PTYPE_INNER_L4_UDP,
1275 RTE_PTYPE_INNER_L4_TCP,
1276 RTE_PTYPE_INNER_L4_SCTP,
1278 RTE_PTYPE_L2_ETHER_VLAN,
1279 RTE_PTYPE_L2_ETHER_QINQ,
1281 RTE_PTYPE_L3_IPV4_EXT,
1282 RTE_PTYPE_L3_IPV6_EXT,
1294 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1295 struct rte_eth_fc_conf *fc_conf)
1297 fc_conf->mode = RTE_FC_NONE;
1302 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1303 struct rte_eth_fc_conf *fc_conf)
1305 if (fc_conf->mode != RTE_FC_NONE)
1310 static const struct eth_dev_ops ops = {
1311 .dev_start = tap_dev_start,
1312 .dev_stop = tap_dev_stop,
1313 .dev_close = tap_dev_close,
1314 .dev_configure = tap_dev_configure,
1315 .dev_infos_get = tap_dev_info,
1316 .rx_queue_setup = tap_rx_queue_setup,
1317 .tx_queue_setup = tap_tx_queue_setup,
1318 .rx_queue_release = tap_rx_queue_release,
1319 .tx_queue_release = tap_tx_queue_release,
1320 .flow_ctrl_get = tap_flow_ctrl_get,
1321 .flow_ctrl_set = tap_flow_ctrl_set,
1322 .link_update = tap_link_update,
1323 .dev_set_link_up = tap_link_set_up,
1324 .dev_set_link_down = tap_link_set_down,
1325 .promiscuous_enable = tap_promisc_enable,
1326 .promiscuous_disable = tap_promisc_disable,
1327 .allmulticast_enable = tap_allmulti_enable,
1328 .allmulticast_disable = tap_allmulti_disable,
1329 .mac_addr_set = tap_mac_set,
1330 .mtu_set = tap_mtu_set,
1331 .set_mc_addr_list = tap_set_mc_addr_list,
1332 .stats_get = tap_stats_get,
1333 .stats_reset = tap_stats_reset,
1334 .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1335 .filter_ctrl = tap_dev_filter_ctrl,
1339 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
1340 char *remote_iface, int fixed_mac_type)
1342 int numa_node = rte_socket_id();
1343 struct rte_eth_dev *dev;
1344 struct pmd_internals *pmd;
1345 struct rte_eth_dev_data *data;
1349 RTE_LOG(DEBUG, PMD, " TAP device on numa %u\n", rte_socket_id());
1351 data = rte_zmalloc_socket(tap_name, sizeof(*data), 0, numa_node);
1353 RTE_LOG(ERR, PMD, "TAP Failed to allocate data\n");
1354 goto error_exit_nodev;
1357 dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1359 RTE_LOG(ERR, PMD, "TAP Unable to allocate device struct\n");
1360 goto error_exit_nodev;
1363 pmd = dev->data->dev_private;
1365 snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1367 pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1368 if (pmd->ioctl_sock == -1) {
1370 "TAP Unable to get a socket for management: %s\n",
1375 /* Setup some default values */
1376 rte_memcpy(data, dev->data, sizeof(*data));
1377 data->dev_private = pmd;
1378 data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1379 data->numa_node = numa_node;
1381 data->dev_link = pmd_link;
1382 data->mac_addrs = &pmd->eth_addr;
1383 /* Set the number of RX and TX queues */
1384 data->nb_rx_queues = 0;
1385 data->nb_tx_queues = 0;
1388 dev->dev_ops = &ops;
1389 dev->rx_pkt_burst = pmd_rx_burst;
1390 dev->tx_pkt_burst = pmd_tx_burst;
1392 pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1393 pmd->intr_handle.fd = -1;
1394 dev->intr_handle = &pmd->intr_handle;
1396 /* Presetup the fds to -1 as being not valid */
1397 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1398 pmd->rxq[i].fd = -1;
1399 pmd->txq[i].fd = -1;
1402 if (fixed_mac_type) {
1403 /* fixed mac = 00:64:74:61:70:<iface_idx> */
1404 static int iface_idx;
1405 char mac[ETHER_ADDR_LEN] = "\0dtap";
1407 mac[ETHER_ADDR_LEN - 1] = iface_idx++;
1408 rte_memcpy(&pmd->eth_addr, mac, ETHER_ADDR_LEN);
1410 eth_random_addr((uint8_t *)&pmd->eth_addr);
1413 /* Immediately create the netdevice (this will create the 1st queue). */
1415 if (tap_setup_queue(dev, pmd, 0, 1) == -1)
1418 if (tap_setup_queue(dev, pmd, 0, 0) == -1)
1421 ifr.ifr_mtu = dev->data->mtu;
1422 if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1425 memset(&ifr, 0, sizeof(struct ifreq));
1426 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1427 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr, ETHER_ADDR_LEN);
1428 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1432 * Set up everything related to rte_flow:
1434 * - tap / remote if_index
1435 * - mandatory QDISCs
1436 * - rte_flow actual/implicit lists
1439 pmd->nlsk_fd = tap_nl_init(0);
1440 if (pmd->nlsk_fd == -1) {
1441 RTE_LOG(WARNING, PMD, "%s: failed to create netlink socket.\n",
1443 goto disable_rte_flow;
1445 pmd->if_index = if_nametoindex(pmd->name);
1446 if (!pmd->if_index) {
1447 RTE_LOG(ERR, PMD, "%s: failed to get if_index.\n", pmd->name);
1448 goto disable_rte_flow;
1450 if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1451 RTE_LOG(ERR, PMD, "%s: failed to create multiq qdisc.\n",
1453 goto disable_rte_flow;
1455 if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1456 RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1458 goto disable_rte_flow;
1460 LIST_INIT(&pmd->flows);
1462 if (strlen(remote_iface)) {
1463 pmd->remote_if_index = if_nametoindex(remote_iface);
1464 if (!pmd->remote_if_index) {
1465 RTE_LOG(ERR, PMD, "%s: failed to get %s if_index.\n",
1466 pmd->name, remote_iface);
1469 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1470 "%s", remote_iface);
1472 /* Save state of remote device */
1473 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1475 /* Replicate remote MAC address */
1476 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1477 RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1478 pmd->name, pmd->remote_iface);
1481 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1483 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1484 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1485 RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1486 pmd->name, remote_iface);
1491 * Flush usually returns negative value because it tries to
1492 * delete every QDISC (and on a running device, one QDISC at
1493 * least is needed). Ignore negative return value.
1495 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1496 if (qdisc_create_ingress(pmd->nlsk_fd,
1497 pmd->remote_if_index) < 0) {
1498 RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1502 LIST_INIT(&pmd->implicit_flows);
1503 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1504 tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1505 tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1506 tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1508 "%s: failed to create implicit rules.\n",
1517 RTE_LOG(ERR, PMD, " Disabling rte flow support: %s(%d)\n",
1518 strerror(errno), errno);
1519 if (strlen(remote_iface)) {
1520 RTE_LOG(ERR, PMD, "Remote feature requires flow support.\n");
1526 RTE_LOG(ERR, PMD, " Can't set up remote feature: %s(%d)\n",
1527 strerror(errno), errno);
1528 tap_flow_implicit_flush(pmd, NULL);
1531 if (pmd->ioctl_sock > 0)
1532 close(pmd->ioctl_sock);
1533 rte_eth_dev_release_port(dev);
1536 RTE_LOG(ERR, PMD, "TAP Unable to initialize %s\n",
1537 rte_vdev_device_name(vdev));
1544 set_interface_name(const char *key __rte_unused,
1548 char *name = (char *)extra_args;
1551 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s", value);
1553 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1554 DEFAULT_TAP_NAME, (tap_unit - 1));
1560 set_remote_iface(const char *key __rte_unused,
1564 char *name = (char *)extra_args;
1567 snprintf(name, RTE_ETH_NAME_MAX_LEN, "%s", value);
1573 set_mac_type(const char *key __rte_unused,
1578 !strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED)))
1579 *(int *)extra_args = 1;
1583 /* Open a TAP interface device.
1586 rte_pmd_tap_probe(struct rte_vdev_device *dev)
1588 const char *name, *params;
1590 struct rte_kvargs *kvlist = NULL;
1592 char tap_name[RTE_ETH_NAME_MAX_LEN];
1593 char remote_iface[RTE_ETH_NAME_MAX_LEN];
1594 int fixed_mac_type = 0;
1596 name = rte_vdev_device_name(dev);
1597 params = rte_vdev_device_args(dev);
1599 speed = ETH_SPEED_NUM_10G;
1600 snprintf(tap_name, sizeof(tap_name), "%s%d",
1601 DEFAULT_TAP_NAME, tap_unit++);
1602 memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1604 if (params && (params[0] != '\0')) {
1605 RTE_LOG(DEBUG, PMD, "parameters (%s)\n", params);
1607 kvlist = rte_kvargs_parse(params, valid_arguments);
1609 if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1610 ret = rte_kvargs_process(kvlist,
1612 &set_interface_name,
1618 if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
1619 ret = rte_kvargs_process(kvlist,
1627 if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
1628 ret = rte_kvargs_process(kvlist,
1637 pmd_link.link_speed = speed;
1639 RTE_LOG(NOTICE, PMD, "Initializing pmd_tap for %s as %s\n",
1642 ret = eth_dev_tap_create(dev, tap_name, remote_iface, fixed_mac_type);
1646 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
1648 tap_unit--; /* Restore the unit number */
1650 rte_kvargs_free(kvlist);
1655 /* detach a TAP device.
1658 rte_pmd_tap_remove(struct rte_vdev_device *dev)
1660 struct rte_eth_dev *eth_dev = NULL;
1661 struct pmd_internals *internals;
1664 RTE_LOG(DEBUG, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
1667 /* find the ethdev entry */
1668 eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1672 internals = eth_dev->data->dev_private;
1673 if (internals->nlsk_fd) {
1674 tap_flow_flush(eth_dev, NULL);
1675 tap_flow_implicit_flush(internals, NULL);
1676 tap_nl_final(internals->nlsk_fd);
1678 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1679 if (internals->rxq[i].fd != -1) {
1680 close(internals->rxq[i].fd);
1681 internals->rxq[i].fd = -1;
1683 if (internals->txq[i].fd != -1) {
1684 close(internals->txq[i].fd);
1685 internals->txq[i].fd = -1;
1689 close(internals->ioctl_sock);
1690 rte_free(eth_dev->data->dev_private);
1691 rte_free(eth_dev->data);
1693 rte_eth_dev_release_port(eth_dev);
1698 static struct rte_vdev_driver pmd_tap_drv = {
1699 .probe = rte_pmd_tap_probe,
1700 .remove = rte_pmd_tap_remove,
1702 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1703 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1704 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
1705 ETH_TAP_IFACE_ARG "=<string> "
1706 ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_FIXED " "
1707 ETH_TAP_REMOTE_ARG "=<string>");