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.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>
30 #include <arpa/inet.h>
32 #include <linux/if_tun.h>
33 #include <linux/if_ether.h>
36 #include <rte_eth_tap.h>
38 #include <tap_netlink.h>
39 #include <tap_tcmsgs.h>
41 /* Linux based path to the TUN device */
42 #define TUN_TAP_DEV_PATH "/dev/net/tun"
43 #define DEFAULT_TAP_NAME "dtap"
45 #define ETH_TAP_IFACE_ARG "iface"
46 #define ETH_TAP_SPEED_ARG "speed"
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[] = {
63 static volatile uint32_t tap_trigger; /* Rx trigger */
65 static struct rte_eth_link pmd_link = {
66 .link_speed = ETH_SPEED_NUM_10G,
67 .link_duplex = ETH_LINK_FULL_DUPLEX,
68 .link_status = ETH_LINK_DOWN,
69 .link_autoneg = ETH_LINK_AUTONEG
73 tap_trigger_cb(int sig __rte_unused)
75 /* Valid trigger values are nonzero */
76 tap_trigger = (tap_trigger + 1) | 0x80000000;
79 /* Specifies on what netdevices the ioctl should be applied */
86 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
88 /* Tun/Tap allocation routine
90 * name is the number of the interface to use, unless NULL to take the host
94 tun_alloc(struct pmd_internals *pmd)
97 #ifdef IFF_MULTI_QUEUE
98 unsigned int features;
102 memset(&ifr, 0, sizeof(struct ifreq));
105 * Do not set IFF_NO_PI as packet information header will be needed
106 * to check if a received packet has been truncated.
108 ifr.ifr_flags = IFF_TAP;
109 snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
111 RTE_LOG(DEBUG, PMD, "ifr_name '%s'\n", ifr.ifr_name);
113 fd = open(TUN_TAP_DEV_PATH, O_RDWR);
115 RTE_LOG(ERR, PMD, "Unable to create TAP interface\n");
119 #ifdef IFF_MULTI_QUEUE
120 /* Grab the TUN features to verify we can work multi-queue */
121 if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
122 RTE_LOG(ERR, PMD, "TAP unable to get TUN/TAP features\n");
125 RTE_LOG(DEBUG, PMD, " TAP Features %08x\n", features);
127 if (features & IFF_MULTI_QUEUE) {
128 RTE_LOG(DEBUG, PMD, " Multi-queue support for %d queues\n",
129 RTE_PMD_TAP_MAX_QUEUES);
130 ifr.ifr_flags |= IFF_MULTI_QUEUE;
134 ifr.ifr_flags |= IFF_ONE_QUEUE;
135 RTE_LOG(DEBUG, PMD, " Single queue only support\n");
138 /* Set the TUN/TAP configuration and set the name if needed */
139 if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
140 RTE_LOG(WARNING, PMD,
141 "Unable to set TUNSETIFF for %s\n",
147 /* Always set the file descriptor to non-blocking */
148 if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
149 RTE_LOG(WARNING, PMD,
150 "Unable to set %s to nonblocking\n",
152 perror("F_SETFL, NONBLOCK");
156 /* Set up trigger to optimize empty Rx bursts */
160 int flags = fcntl(fd, F_GETFL);
162 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
164 if (sa.sa_handler != tap_trigger_cb) {
166 * Make sure SIGIO is not already taken. This is done
167 * as late as possible to leave the application a
168 * chance to set up its own signal handler first.
170 if (sa.sa_handler != SIG_IGN &&
171 sa.sa_handler != SIG_DFL) {
175 sa = (struct sigaction){
176 .sa_flags = SA_RESTART,
177 .sa_handler = tap_trigger_cb,
179 if (sigaction(SIGIO, &sa, NULL) == -1)
182 /* Enable SIGIO on file descriptor */
183 fcntl(fd, F_SETFL, flags | O_ASYNC);
184 fcntl(fd, F_SETOWN, getpid());
187 /* Disable trigger globally in case of error */
189 RTE_LOG(WARNING, PMD, "Rx trigger disabled: %s\n",
202 tap_verify_csum(struct rte_mbuf *mbuf)
204 uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
205 uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
206 uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
207 unsigned int l2_len = sizeof(struct ether_hdr);
213 if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
215 else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
217 /* Don't verify checksum for packets with discontinuous L2 header */
218 if (unlikely(l2_len + sizeof(struct ipv4_hdr) >
219 rte_pktmbuf_data_len(mbuf)))
221 l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
222 if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
223 struct ipv4_hdr *iph = l3_hdr;
225 /* ihl contains the number of 4-byte words in the header */
226 l3_len = 4 * (iph->version_ihl & 0xf);
227 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
230 cksum = ~rte_raw_cksum(iph, l3_len);
231 mbuf->ol_flags |= cksum ?
232 PKT_RX_IP_CKSUM_BAD :
233 PKT_RX_IP_CKSUM_GOOD;
234 } else if (l3 == RTE_PTYPE_L3_IPV6) {
235 l3_len = sizeof(struct ipv6_hdr);
237 /* IPv6 extensions are not supported */
240 if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
241 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
242 /* Don't verify checksum for multi-segment packets. */
243 if (mbuf->nb_segs > 1)
245 if (l3 == RTE_PTYPE_L3_IPV4)
246 cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
247 else if (l3 == RTE_PTYPE_L3_IPV6)
248 cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
249 mbuf->ol_flags |= cksum ?
250 PKT_RX_L4_CKSUM_BAD :
251 PKT_RX_L4_CKSUM_GOOD;
255 /* Callback to handle the rx burst of packets to the correct interface and
256 * file descriptor(s) in a multi-queue setup.
259 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
261 struct rx_queue *rxq = queue;
263 unsigned long num_rx_bytes = 0;
264 uint32_t trigger = tap_trigger;
266 if (trigger == rxq->trigger_seen)
269 rxq->trigger_seen = trigger;
270 rte_compiler_barrier();
271 for (num_rx = 0; num_rx < nb_pkts; ) {
272 struct rte_mbuf *mbuf = rxq->pool;
273 struct rte_mbuf *seg = NULL;
274 struct rte_mbuf *new_tail = NULL;
275 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
278 len = readv(rxq->fd, *rxq->iovecs,
279 1 + (rxq->rxmode->enable_scatter ?
280 rxq->nb_rx_desc : 1));
281 if (len < (int)sizeof(struct tun_pi))
284 /* Packet couldn't fit in the provided mbuf */
285 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
286 rxq->stats.ierrors++;
290 len -= sizeof(struct tun_pi);
293 mbuf->port = rxq->in_port;
295 struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
297 if (unlikely(!buf)) {
298 rxq->stats.rx_nombuf++;
299 /* No new buf has been allocated: do nothing */
300 if (!new_tail || !seg)
304 rte_pktmbuf_free(mbuf);
308 seg = seg ? seg->next : mbuf;
309 if (rxq->pool == mbuf)
312 new_tail->next = buf;
314 new_tail->next = seg->next;
316 /* iovecs[0] is reserved for packet info (pi) */
317 (*rxq->iovecs)[mbuf->nb_segs].iov_len =
318 buf->buf_len - data_off;
319 (*rxq->iovecs)[mbuf->nb_segs].iov_base =
320 (char *)buf->buf_addr + data_off;
322 seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
323 seg->data_off = data_off;
325 len -= seg->data_len;
329 /* First segment has headroom, not the others */
333 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
335 if (rxq->rxmode->hw_ip_checksum)
336 tap_verify_csum(mbuf);
338 /* account for the receive frame */
339 bufs[num_rx++] = mbuf;
340 num_rx_bytes += mbuf->pkt_len;
343 rxq->stats.ipackets += num_rx;
344 rxq->stats.ibytes += num_rx_bytes;
350 tap_tx_offload(char *packet, uint64_t ol_flags, unsigned int l2_len,
353 void *l3_hdr = packet + l2_len;
355 if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
356 struct ipv4_hdr *iph = l3_hdr;
359 iph->hdr_checksum = 0;
360 cksum = rte_raw_cksum(iph, l3_len);
361 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
363 if (ol_flags & PKT_TX_L4_MASK) {
369 l4_hdr = packet + l2_len + l3_len;
370 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
371 l4_cksum = &((struct udp_hdr *)l4_hdr)->dgram_cksum;
372 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
373 l4_cksum = &((struct tcp_hdr *)l4_hdr)->cksum;
377 if (ol_flags & PKT_TX_IPV4) {
378 struct ipv4_hdr *iph = l3_hdr;
380 l4_len = rte_be_to_cpu_16(iph->total_length) - l3_len;
381 cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
383 struct ipv6_hdr *ip6h = l3_hdr;
385 /* payload_len does not include ext headers */
386 l4_len = rte_be_to_cpu_16(ip6h->payload_len) -
387 l3_len + sizeof(struct ipv6_hdr);
388 cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
390 cksum += rte_raw_cksum(l4_hdr, l4_len);
391 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
392 cksum = (~cksum) & 0xffff;
399 /* Callback to handle sending packets from the tap interface
402 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
404 struct tx_queue *txq = queue;
406 unsigned long num_tx_bytes = 0;
410 if (unlikely(nb_pkts == 0))
413 max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
414 for (i = 0; i < nb_pkts; i++) {
415 struct rte_mbuf *mbuf = bufs[num_tx];
416 struct iovec iovecs[mbuf->nb_segs + 1];
417 struct tun_pi pi = { .flags = 0 };
418 struct rte_mbuf *seg = mbuf;
419 char m_copy[mbuf->data_len];
423 /* stats.errs will be incremented */
424 if (rte_pktmbuf_pkt_len(mbuf) > max_size)
427 iovecs[0].iov_base = π
428 iovecs[0].iov_len = sizeof(pi);
429 for (j = 1; j <= mbuf->nb_segs; j++) {
430 iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
432 rte_pktmbuf_mtod(seg, void *);
435 if (mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
436 (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
437 (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM) {
438 /* Support only packets with all data in the same seg */
439 if (mbuf->nb_segs > 1)
441 /* To change checksums, work on a copy of data. */
442 rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
443 rte_pktmbuf_data_len(mbuf));
444 tap_tx_offload(m_copy, mbuf->ol_flags,
445 mbuf->l2_len, mbuf->l3_len);
446 iovecs[1].iov_base = m_copy;
448 /* copy the tx frame data */
449 n = writev(txq->fd, iovecs, mbuf->nb_segs + 1);
454 num_tx_bytes += mbuf->pkt_len;
455 rte_pktmbuf_free(mbuf);
458 txq->stats.opackets += num_tx;
459 txq->stats.errs += nb_pkts - num_tx;
460 txq->stats.obytes += num_tx_bytes;
466 tap_ioctl_req2str(unsigned long request)
470 return "SIOCSIFFLAGS";
472 return "SIOCGIFFLAGS";
474 return "SIOCGIFHWADDR";
476 return "SIOCSIFHWADDR";
484 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
485 struct ifreq *ifr, int set, enum ioctl_mode mode)
487 short req_flags = ifr->ifr_flags;
488 int remote = pmd->remote_if_index &&
489 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
491 if (!pmd->remote_if_index && mode == REMOTE_ONLY)
494 * If there is a remote netdevice, apply ioctl on it, then apply it on
499 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
500 else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
501 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
504 /* fetch current flags to leave other flags untouched */
505 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
508 ifr->ifr_flags |= req_flags;
510 ifr->ifr_flags &= ~req_flags;
518 RTE_ASSERT(!"unsupported request type: must not happen");
520 if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
522 if (remote-- && mode == LOCAL_AND_REMOTE)
527 RTE_LOG(DEBUG, PMD, "%s: %s(%s) failed: %s(%d)\n", ifr->ifr_name,
528 __func__, tap_ioctl_req2str(request), strerror(errno), errno);
533 tap_link_set_down(struct rte_eth_dev *dev)
535 struct pmd_internals *pmd = dev->data->dev_private;
536 struct ifreq ifr = { .ifr_flags = IFF_UP };
538 dev->data->dev_link.link_status = ETH_LINK_DOWN;
539 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
543 tap_link_set_up(struct rte_eth_dev *dev)
545 struct pmd_internals *pmd = dev->data->dev_private;
546 struct ifreq ifr = { .ifr_flags = IFF_UP };
548 dev->data->dev_link.link_status = ETH_LINK_UP;
549 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
553 tap_dev_start(struct rte_eth_dev *dev)
557 err = tap_intr_handle_set(dev, 1);
560 return tap_link_set_up(dev);
563 /* This function gets called when the current port gets stopped.
566 tap_dev_stop(struct rte_eth_dev *dev)
568 tap_intr_handle_set(dev, 0);
569 tap_link_set_down(dev);
573 tap_dev_configure(struct rte_eth_dev *dev)
575 if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
577 "%s: number of rx queues %d exceeds max num of queues %d\n",
579 dev->data->nb_rx_queues,
580 RTE_PMD_TAP_MAX_QUEUES);
583 if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
585 "%s: number of tx queues %d exceeds max num of queues %d\n",
587 dev->data->nb_tx_queues,
588 RTE_PMD_TAP_MAX_QUEUES);
592 RTE_LOG(INFO, PMD, "%s: %p: TX configured queues number: %u\n",
593 dev->device->name, (void *)dev, dev->data->nb_tx_queues);
595 RTE_LOG(INFO, PMD, "%s: %p: RX configured queues number: %u\n",
596 dev->device->name, (void *)dev, dev->data->nb_rx_queues);
602 tap_dev_speed_capa(void)
604 uint32_t speed = pmd_link.link_speed;
607 if (speed >= ETH_SPEED_NUM_10M)
608 capa |= ETH_LINK_SPEED_10M;
609 if (speed >= ETH_SPEED_NUM_100M)
610 capa |= ETH_LINK_SPEED_100M;
611 if (speed >= ETH_SPEED_NUM_1G)
612 capa |= ETH_LINK_SPEED_1G;
613 if (speed >= ETH_SPEED_NUM_5G)
614 capa |= ETH_LINK_SPEED_2_5G;
615 if (speed >= ETH_SPEED_NUM_5G)
616 capa |= ETH_LINK_SPEED_5G;
617 if (speed >= ETH_SPEED_NUM_10G)
618 capa |= ETH_LINK_SPEED_10G;
619 if (speed >= ETH_SPEED_NUM_20G)
620 capa |= ETH_LINK_SPEED_20G;
621 if (speed >= ETH_SPEED_NUM_25G)
622 capa |= ETH_LINK_SPEED_25G;
623 if (speed >= ETH_SPEED_NUM_40G)
624 capa |= ETH_LINK_SPEED_40G;
625 if (speed >= ETH_SPEED_NUM_50G)
626 capa |= ETH_LINK_SPEED_50G;
627 if (speed >= ETH_SPEED_NUM_56G)
628 capa |= ETH_LINK_SPEED_56G;
629 if (speed >= ETH_SPEED_NUM_100G)
630 capa |= ETH_LINK_SPEED_100G;
636 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
638 struct pmd_internals *internals = dev->data->dev_private;
640 dev_info->if_index = internals->if_index;
641 dev_info->max_mac_addrs = 1;
642 dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
643 dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
644 dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
645 dev_info->min_rx_bufsize = 0;
646 dev_info->pci_dev = NULL;
647 dev_info->speed_capa = tap_dev_speed_capa();
648 dev_info->rx_offload_capa = (DEV_RX_OFFLOAD_IPV4_CKSUM |
649 DEV_RX_OFFLOAD_UDP_CKSUM |
650 DEV_RX_OFFLOAD_TCP_CKSUM);
651 dev_info->tx_offload_capa =
652 (DEV_TX_OFFLOAD_IPV4_CKSUM |
653 DEV_TX_OFFLOAD_UDP_CKSUM |
654 DEV_TX_OFFLOAD_TCP_CKSUM);
658 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
660 unsigned int i, imax;
661 unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
662 unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
663 unsigned long rx_nombuf = 0, ierrors = 0;
664 const struct pmd_internals *pmd = dev->data->dev_private;
666 /* rx queue statistics */
667 imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
668 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
669 for (i = 0; i < imax; i++) {
670 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
671 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
672 rx_total += tap_stats->q_ipackets[i];
673 rx_bytes_total += tap_stats->q_ibytes[i];
674 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
675 ierrors += pmd->rxq[i].stats.ierrors;
678 /* tx queue statistics */
679 imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
680 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
682 for (i = 0; i < imax; i++) {
683 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
684 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
685 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
686 tx_total += tap_stats->q_opackets[i];
687 tx_err_total += tap_stats->q_errors[i];
688 tx_bytes_total += tap_stats->q_obytes[i];
691 tap_stats->ipackets = rx_total;
692 tap_stats->ibytes = rx_bytes_total;
693 tap_stats->ierrors = ierrors;
694 tap_stats->rx_nombuf = rx_nombuf;
695 tap_stats->opackets = tx_total;
696 tap_stats->oerrors = tx_err_total;
697 tap_stats->obytes = tx_bytes_total;
702 tap_stats_reset(struct rte_eth_dev *dev)
705 struct pmd_internals *pmd = dev->data->dev_private;
707 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
708 pmd->rxq[i].stats.ipackets = 0;
709 pmd->rxq[i].stats.ibytes = 0;
710 pmd->rxq[i].stats.ierrors = 0;
711 pmd->rxq[i].stats.rx_nombuf = 0;
713 pmd->txq[i].stats.opackets = 0;
714 pmd->txq[i].stats.errs = 0;
715 pmd->txq[i].stats.obytes = 0;
720 tap_dev_close(struct rte_eth_dev *dev)
723 struct pmd_internals *internals = dev->data->dev_private;
725 tap_link_set_down(dev);
726 tap_flow_flush(dev, NULL);
727 tap_flow_implicit_flush(internals, NULL);
729 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
730 if (internals->rxq[i].fd != -1) {
731 close(internals->rxq[i].fd);
732 internals->rxq[i].fd = -1;
734 if (internals->txq[i].fd != -1) {
735 close(internals->txq[i].fd);
736 internals->txq[i].fd = -1;
740 if (internals->remote_if_index) {
741 /* Restore initial remote state */
742 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
743 &internals->remote_initial_flags);
748 tap_rx_queue_release(void *queue)
750 struct rx_queue *rxq = queue;
752 if (rxq && (rxq->fd > 0)) {
755 rte_pktmbuf_free(rxq->pool);
756 rte_free(rxq->iovecs);
763 tap_tx_queue_release(void *queue)
765 struct tx_queue *txq = queue;
767 if (txq && (txq->fd > 0)) {
774 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
776 struct rte_eth_link *dev_link = &dev->data->dev_link;
777 struct pmd_internals *pmd = dev->data->dev_private;
778 struct ifreq ifr = { .ifr_flags = 0 };
780 if (pmd->remote_if_index) {
781 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
782 if (!(ifr.ifr_flags & IFF_UP) ||
783 !(ifr.ifr_flags & IFF_RUNNING)) {
784 dev_link->link_status = ETH_LINK_DOWN;
788 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
789 dev_link->link_status =
790 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
797 tap_promisc_enable(struct rte_eth_dev *dev)
799 struct pmd_internals *pmd = dev->data->dev_private;
800 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
802 dev->data->promiscuous = 1;
803 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
804 if (pmd->remote_if_index && !pmd->flow_isolate)
805 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
809 tap_promisc_disable(struct rte_eth_dev *dev)
811 struct pmd_internals *pmd = dev->data->dev_private;
812 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
814 dev->data->promiscuous = 0;
815 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
816 if (pmd->remote_if_index && !pmd->flow_isolate)
817 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
821 tap_allmulti_enable(struct rte_eth_dev *dev)
823 struct pmd_internals *pmd = dev->data->dev_private;
824 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
826 dev->data->all_multicast = 1;
827 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
828 if (pmd->remote_if_index && !pmd->flow_isolate)
829 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
833 tap_allmulti_disable(struct rte_eth_dev *dev)
835 struct pmd_internals *pmd = dev->data->dev_private;
836 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
838 dev->data->all_multicast = 0;
839 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
840 if (pmd->remote_if_index && !pmd->flow_isolate)
841 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
845 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
847 struct pmd_internals *pmd = dev->data->dev_private;
848 enum ioctl_mode mode = LOCAL_ONLY;
851 if (is_zero_ether_addr(mac_addr)) {
852 RTE_LOG(ERR, PMD, "%s: can't set an empty MAC address\n",
856 /* Check the actual current MAC address on the tap netdevice */
857 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
859 if (is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
862 /* Check the current MAC address on the remote */
863 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0)
865 if (!is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
867 mode = LOCAL_AND_REMOTE;
868 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
869 rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
870 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode) < 0)
872 rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
873 if (pmd->remote_if_index && !pmd->flow_isolate) {
874 /* Replace MAC redirection rule after a MAC change */
875 if (tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC) < 0) {
877 "%s: Couldn't delete MAC redirection rule\n",
881 if (tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0)
883 "%s: Couldn't add MAC redirection rule\n",
889 tap_setup_queue(struct rte_eth_dev *dev,
890 struct pmd_internals *internals,
897 struct pmd_internals *pmd = dev->data->dev_private;
898 struct rx_queue *rx = &internals->rxq[qid];
899 struct tx_queue *tx = &internals->txq[qid];
911 /* fd for this queue already exists */
912 RTE_LOG(DEBUG, PMD, "%s: fd %d for %s queue qid %d exists\n",
913 pmd->name, *fd, dir, qid);
914 } else if (*other_fd != -1) {
915 /* Only other_fd exists. dup it */
916 *fd = dup(*other_fd);
919 RTE_LOG(ERR, PMD, "%s: dup() failed.\n",
923 RTE_LOG(DEBUG, PMD, "%s: dup fd %d for %s queue qid %d (%d)\n",
924 pmd->name, *other_fd, dir, qid, *fd);
926 /* Both RX and TX fds do not exist (equal -1). Create fd */
927 *fd = tun_alloc(pmd);
929 *fd = -1; /* restore original value */
930 RTE_LOG(ERR, PMD, "%s: tun_alloc() failed.\n",
934 RTE_LOG(DEBUG, PMD, "%s: add %s queue for qid %d fd %d\n",
935 pmd->name, dir, qid, *fd);
938 tx->mtu = &dev->data->mtu;
939 rx->rxmode = &dev->data->dev_conf.rxmode;
945 tap_rx_queue_setup(struct rte_eth_dev *dev,
946 uint16_t rx_queue_id,
948 unsigned int socket_id,
949 const struct rte_eth_rxconf *rx_conf __rte_unused,
950 struct rte_mempool *mp)
952 struct pmd_internals *internals = dev->data->dev_private;
953 struct rx_queue *rxq = &internals->rxq[rx_queue_id];
954 struct rte_mbuf **tmp = &rxq->pool;
955 long iov_max = sysconf(_SC_IOV_MAX);
956 uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
957 struct iovec (*iovecs)[nb_desc + 1];
958 int data_off = RTE_PKTMBUF_HEADROOM;
963 if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
964 RTE_LOG(WARNING, PMD,
965 "nb_rx_queues %d too small or mempool NULL\n",
966 dev->data->nb_rx_queues);
971 rxq->trigger_seen = 1; /* force initial burst */
972 rxq->in_port = dev->data->port_id;
973 rxq->nb_rx_desc = nb_desc;
974 iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
977 RTE_LOG(WARNING, PMD,
978 "%s: Couldn't allocate %d RX descriptors\n",
979 dev->device->name, nb_desc);
982 rxq->iovecs = iovecs;
984 dev->data->rx_queues[rx_queue_id] = rxq;
985 fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
991 (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
992 (*rxq->iovecs)[0].iov_base = &rxq->pi;
994 for (i = 1; i <= nb_desc; i++) {
995 *tmp = rte_pktmbuf_alloc(rxq->mp);
997 RTE_LOG(WARNING, PMD,
998 "%s: couldn't allocate memory for queue %d\n",
999 dev->device->name, rx_queue_id);
1003 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1004 (*rxq->iovecs)[i].iov_base =
1005 (char *)(*tmp)->buf_addr + data_off;
1007 tmp = &(*tmp)->next;
1010 RTE_LOG(DEBUG, PMD, " RX TAP device name %s, qid %d on fd %d\n",
1011 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
1016 rte_pktmbuf_free(rxq->pool);
1018 rte_free(rxq->iovecs);
1024 tap_tx_queue_setup(struct rte_eth_dev *dev,
1025 uint16_t tx_queue_id,
1026 uint16_t nb_tx_desc __rte_unused,
1027 unsigned int socket_id __rte_unused,
1028 const struct rte_eth_txconf *tx_conf __rte_unused)
1030 struct pmd_internals *internals = dev->data->dev_private;
1033 if (tx_queue_id >= dev->data->nb_tx_queues)
1036 dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1037 ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1041 RTE_LOG(DEBUG, PMD, " TX TAP device name %s, qid %d on fd %d\n",
1042 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd);
1048 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1050 struct pmd_internals *pmd = dev->data->dev_private;
1051 struct ifreq ifr = { .ifr_mtu = mtu };
1054 err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1056 dev->data->mtu = mtu;
1062 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1063 struct ether_addr *mc_addr_set __rte_unused,
1064 uint32_t nb_mc_addr __rte_unused)
1067 * Nothing to do actually: the tap has no filtering whatsoever, every
1068 * packet is received.
1074 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1076 struct rte_eth_dev *dev = arg;
1077 struct pmd_internals *pmd = dev->data->dev_private;
1078 struct ifinfomsg *info = NLMSG_DATA(nh);
1080 if (nh->nlmsg_type != RTM_NEWLINK ||
1081 (info->ifi_index != pmd->if_index &&
1082 info->ifi_index != pmd->remote_if_index))
1084 return tap_link_update(dev, 0);
1088 tap_dev_intr_handler(void *cb_arg)
1090 struct rte_eth_dev *dev = cb_arg;
1091 struct pmd_internals *pmd = dev->data->dev_private;
1093 tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1097 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1099 struct pmd_internals *pmd = dev->data->dev_private;
1101 /* In any case, disable interrupt if the conf is no longer there. */
1102 if (!dev->data->dev_conf.intr_conf.lsc) {
1103 if (pmd->intr_handle.fd != -1) {
1104 tap_nl_final(pmd->intr_handle.fd);
1105 rte_intr_callback_unregister(&pmd->intr_handle,
1106 tap_dev_intr_handler, dev);
1111 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1112 if (unlikely(pmd->intr_handle.fd == -1))
1114 return rte_intr_callback_register(
1115 &pmd->intr_handle, tap_dev_intr_handler, dev);
1117 tap_nl_final(pmd->intr_handle.fd);
1118 return rte_intr_callback_unregister(&pmd->intr_handle,
1119 tap_dev_intr_handler, dev);
1122 static const uint32_t*
1123 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1125 static const uint32_t ptypes[] = {
1126 RTE_PTYPE_INNER_L2_ETHER,
1127 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1128 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1129 RTE_PTYPE_INNER_L3_IPV4,
1130 RTE_PTYPE_INNER_L3_IPV4_EXT,
1131 RTE_PTYPE_INNER_L3_IPV6,
1132 RTE_PTYPE_INNER_L3_IPV6_EXT,
1133 RTE_PTYPE_INNER_L4_FRAG,
1134 RTE_PTYPE_INNER_L4_UDP,
1135 RTE_PTYPE_INNER_L4_TCP,
1136 RTE_PTYPE_INNER_L4_SCTP,
1138 RTE_PTYPE_L2_ETHER_VLAN,
1139 RTE_PTYPE_L2_ETHER_QINQ,
1141 RTE_PTYPE_L3_IPV4_EXT,
1142 RTE_PTYPE_L3_IPV6_EXT,
1154 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1155 struct rte_eth_fc_conf *fc_conf)
1157 fc_conf->mode = RTE_FC_NONE;
1162 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1163 struct rte_eth_fc_conf *fc_conf)
1165 if (fc_conf->mode != RTE_FC_NONE)
1170 static const struct eth_dev_ops ops = {
1171 .dev_start = tap_dev_start,
1172 .dev_stop = tap_dev_stop,
1173 .dev_close = tap_dev_close,
1174 .dev_configure = tap_dev_configure,
1175 .dev_infos_get = tap_dev_info,
1176 .rx_queue_setup = tap_rx_queue_setup,
1177 .tx_queue_setup = tap_tx_queue_setup,
1178 .rx_queue_release = tap_rx_queue_release,
1179 .tx_queue_release = tap_tx_queue_release,
1180 .flow_ctrl_get = tap_flow_ctrl_get,
1181 .flow_ctrl_set = tap_flow_ctrl_set,
1182 .link_update = tap_link_update,
1183 .dev_set_link_up = tap_link_set_up,
1184 .dev_set_link_down = tap_link_set_down,
1185 .promiscuous_enable = tap_promisc_enable,
1186 .promiscuous_disable = tap_promisc_disable,
1187 .allmulticast_enable = tap_allmulti_enable,
1188 .allmulticast_disable = tap_allmulti_disable,
1189 .mac_addr_set = tap_mac_set,
1190 .mtu_set = tap_mtu_set,
1191 .set_mc_addr_list = tap_set_mc_addr_list,
1192 .stats_get = tap_stats_get,
1193 .stats_reset = tap_stats_reset,
1194 .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1195 .filter_ctrl = tap_dev_filter_ctrl,
1199 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
1200 char *remote_iface, int fixed_mac_type)
1202 int numa_node = rte_socket_id();
1203 struct rte_eth_dev *dev;
1204 struct pmd_internals *pmd;
1205 struct rte_eth_dev_data *data;
1209 RTE_LOG(DEBUG, PMD, " TAP device on numa %u\n", rte_socket_id());
1211 data = rte_zmalloc_socket(tap_name, sizeof(*data), 0, numa_node);
1213 RTE_LOG(ERR, PMD, "TAP Failed to allocate data\n");
1217 dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1219 RTE_LOG(ERR, PMD, "TAP Unable to allocate device struct\n");
1223 pmd = dev->data->dev_private;
1225 snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1227 pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1228 if (pmd->ioctl_sock == -1) {
1230 "TAP Unable to get a socket for management: %s\n",
1235 /* Setup some default values */
1236 rte_memcpy(data, dev->data, sizeof(*data));
1237 data->dev_private = pmd;
1238 data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1239 data->numa_node = numa_node;
1241 data->dev_link = pmd_link;
1242 data->mac_addrs = &pmd->eth_addr;
1243 /* Set the number of RX and TX queues */
1244 data->nb_rx_queues = 0;
1245 data->nb_tx_queues = 0;
1248 dev->dev_ops = &ops;
1249 dev->rx_pkt_burst = pmd_rx_burst;
1250 dev->tx_pkt_burst = pmd_tx_burst;
1252 pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1253 pmd->intr_handle.fd = -1;
1255 /* Presetup the fds to -1 as being not valid */
1256 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1257 pmd->rxq[i].fd = -1;
1258 pmd->txq[i].fd = -1;
1261 if (fixed_mac_type) {
1262 /* fixed mac = 00:64:74:61:70:<iface_idx> */
1263 static int iface_idx;
1264 char mac[ETHER_ADDR_LEN] = "\0dtap";
1266 mac[ETHER_ADDR_LEN - 1] = iface_idx++;
1267 rte_memcpy(&pmd->eth_addr, mac, ETHER_ADDR_LEN);
1269 eth_random_addr((uint8_t *)&pmd->eth_addr);
1272 /* Immediately create the netdevice (this will create the 1st queue). */
1274 if (tap_setup_queue(dev, pmd, 0, 1) == -1)
1277 if (tap_setup_queue(dev, pmd, 0, 0) == -1)
1280 ifr.ifr_mtu = dev->data->mtu;
1281 if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1284 memset(&ifr, 0, sizeof(struct ifreq));
1285 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1286 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr, ETHER_ADDR_LEN);
1287 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1291 * Set up everything related to rte_flow:
1293 * - tap / remote if_index
1294 * - mandatory QDISCs
1295 * - rte_flow actual/implicit lists
1298 pmd->nlsk_fd = tap_nl_init(0);
1299 if (pmd->nlsk_fd == -1) {
1300 RTE_LOG(WARNING, PMD, "%s: failed to create netlink socket.\n",
1302 goto disable_rte_flow;
1304 pmd->if_index = if_nametoindex(pmd->name);
1305 if (!pmd->if_index) {
1306 RTE_LOG(ERR, PMD, "%s: failed to get if_index.\n", pmd->name);
1307 goto disable_rte_flow;
1309 if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1310 RTE_LOG(ERR, PMD, "%s: failed to create multiq qdisc.\n",
1312 goto disable_rte_flow;
1314 if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1315 RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1317 goto disable_rte_flow;
1319 LIST_INIT(&pmd->flows);
1321 if (strlen(remote_iface)) {
1322 pmd->remote_if_index = if_nametoindex(remote_iface);
1323 if (!pmd->remote_if_index) {
1324 RTE_LOG(ERR, PMD, "%s: failed to get %s if_index.\n",
1325 pmd->name, remote_iface);
1328 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1329 "%s", remote_iface);
1331 /* Save state of remote device */
1332 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1334 /* Replicate remote MAC address */
1335 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1336 RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1337 pmd->name, pmd->remote_iface);
1340 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1342 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1343 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1344 RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1345 pmd->name, remote_iface);
1350 * Flush usually returns negative value because it tries to
1351 * delete every QDISC (and on a running device, one QDISC at
1352 * least is needed). Ignore negative return value.
1354 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1355 if (qdisc_create_ingress(pmd->nlsk_fd,
1356 pmd->remote_if_index) < 0) {
1357 RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1361 LIST_INIT(&pmd->implicit_flows);
1362 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1363 tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1364 tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1365 tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1367 "%s: failed to create implicit rules.\n",
1376 RTE_LOG(ERR, PMD, " Disabling rte flow support: %s(%d)\n",
1377 strerror(errno), errno);
1378 if (strlen(remote_iface)) {
1379 RTE_LOG(ERR, PMD, "Remote feature requires flow support.\n");
1385 RTE_LOG(ERR, PMD, " Can't set up remote feature: %s(%d)\n",
1386 strerror(errno), errno);
1387 tap_flow_implicit_flush(pmd, NULL);
1390 RTE_LOG(ERR, PMD, "TAP Unable to initialize %s\n",
1391 rte_vdev_device_name(vdev));
1398 set_interface_name(const char *key __rte_unused,
1402 char *name = (char *)extra_args;
1405 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s", value);
1407 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1408 DEFAULT_TAP_NAME, (tap_unit - 1));
1414 set_interface_speed(const char *key __rte_unused,
1418 *(int *)extra_args = (value) ? atoi(value) : ETH_SPEED_NUM_10G;
1424 set_remote_iface(const char *key __rte_unused,
1428 char *name = (char *)extra_args;
1431 snprintf(name, RTE_ETH_NAME_MAX_LEN, "%s", value);
1437 set_mac_type(const char *key __rte_unused,
1442 !strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED)))
1443 *(int *)extra_args = 1;
1447 /* Open a TAP interface device.
1450 rte_pmd_tap_probe(struct rte_vdev_device *dev)
1452 const char *name, *params;
1454 struct rte_kvargs *kvlist = NULL;
1456 char tap_name[RTE_ETH_NAME_MAX_LEN];
1457 char remote_iface[RTE_ETH_NAME_MAX_LEN];
1458 int fixed_mac_type = 0;
1460 name = rte_vdev_device_name(dev);
1461 params = rte_vdev_device_args(dev);
1463 speed = ETH_SPEED_NUM_10G;
1464 snprintf(tap_name, sizeof(tap_name), "%s%d",
1465 DEFAULT_TAP_NAME, tap_unit++);
1466 memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1468 if (params && (params[0] != '\0')) {
1469 RTE_LOG(DEBUG, PMD, "parameters (%s)\n", params);
1471 kvlist = rte_kvargs_parse(params, valid_arguments);
1473 if (rte_kvargs_count(kvlist, ETH_TAP_SPEED_ARG) == 1) {
1474 ret = rte_kvargs_process(kvlist,
1476 &set_interface_speed,
1482 if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1483 ret = rte_kvargs_process(kvlist,
1485 &set_interface_name,
1491 if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
1492 ret = rte_kvargs_process(kvlist,
1500 if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
1501 ret = rte_kvargs_process(kvlist,
1510 pmd_link.link_speed = speed;
1512 RTE_LOG(NOTICE, PMD, "Initializing pmd_tap for %s as %s\n",
1515 ret = eth_dev_tap_create(dev, tap_name, remote_iface, fixed_mac_type);
1519 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
1521 tap_unit--; /* Restore the unit number */
1523 rte_kvargs_free(kvlist);
1528 /* detach a TAP device.
1531 rte_pmd_tap_remove(struct rte_vdev_device *dev)
1533 struct rte_eth_dev *eth_dev = NULL;
1534 struct pmd_internals *internals;
1537 RTE_LOG(DEBUG, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
1540 /* find the ethdev entry */
1541 eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1545 internals = eth_dev->data->dev_private;
1546 if (internals->nlsk_fd) {
1547 tap_flow_flush(eth_dev, NULL);
1548 tap_flow_implicit_flush(internals, NULL);
1549 tap_nl_final(internals->nlsk_fd);
1551 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1552 if (internals->rxq[i].fd != -1) {
1553 close(internals->rxq[i].fd);
1554 internals->rxq[i].fd = -1;
1556 if (internals->txq[i].fd != -1) {
1557 close(internals->txq[i].fd);
1558 internals->txq[i].fd = -1;
1562 close(internals->ioctl_sock);
1563 rte_free(eth_dev->data->dev_private);
1564 rte_free(eth_dev->data);
1566 rte_eth_dev_release_port(eth_dev);
1571 static struct rte_vdev_driver pmd_tap_drv = {
1572 .probe = rte_pmd_tap_probe,
1573 .remove = rte_pmd_tap_remove,
1575 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1576 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1577 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
1578 ETH_TAP_IFACE_ARG "=<string> "
1579 ETH_TAP_SPEED_ARG "=<int> "
1580 ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_FIXED " "
1581 ETH_TAP_REMOTE_ARG "=<string>");