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_SPEED_ARG "speed"
48 #define ETH_TAP_REMOTE_ARG "remote"
49 #define ETH_TAP_MAC_ARG "mac"
50 #define ETH_TAP_MAC_FIXED "fixed"
52 static struct rte_vdev_driver pmd_tap_drv;
54 static const char *valid_arguments[] = {
64 static volatile uint32_t tap_trigger; /* Rx trigger */
66 static struct rte_eth_link pmd_link = {
67 .link_speed = ETH_SPEED_NUM_10G,
68 .link_duplex = ETH_LINK_FULL_DUPLEX,
69 .link_status = ETH_LINK_DOWN,
70 .link_autoneg = ETH_LINK_AUTONEG
74 tap_trigger_cb(int sig __rte_unused)
76 /* Valid trigger values are nonzero */
77 tap_trigger = (tap_trigger + 1) | 0x80000000;
80 /* Specifies on what netdevices the ioctl should be applied */
87 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
89 /* Tun/Tap allocation routine
91 * name is the number of the interface to use, unless NULL to take the host
95 tun_alloc(struct pmd_internals *pmd)
98 #ifdef IFF_MULTI_QUEUE
99 unsigned int features;
103 memset(&ifr, 0, sizeof(struct ifreq));
106 * Do not set IFF_NO_PI as packet information header will be needed
107 * to check if a received packet has been truncated.
109 ifr.ifr_flags = IFF_TAP;
110 snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
112 RTE_LOG(DEBUG, PMD, "ifr_name '%s'\n", ifr.ifr_name);
114 fd = open(TUN_TAP_DEV_PATH, O_RDWR);
116 RTE_LOG(ERR, PMD, "Unable to create TAP interface\n");
120 #ifdef IFF_MULTI_QUEUE
121 /* Grab the TUN features to verify we can work multi-queue */
122 if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
123 RTE_LOG(ERR, PMD, "TAP unable to get TUN/TAP features\n");
126 RTE_LOG(DEBUG, PMD, " TAP Features %08x\n", features);
128 if (features & IFF_MULTI_QUEUE) {
129 RTE_LOG(DEBUG, PMD, " Multi-queue support for %d queues\n",
130 RTE_PMD_TAP_MAX_QUEUES);
131 ifr.ifr_flags |= IFF_MULTI_QUEUE;
135 ifr.ifr_flags |= IFF_ONE_QUEUE;
136 RTE_LOG(DEBUG, PMD, " Single queue only support\n");
139 /* Set the TUN/TAP configuration and set the name if needed */
140 if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
141 RTE_LOG(WARNING, PMD,
142 "Unable to set TUNSETIFF for %s\n",
148 /* Always set the file descriptor to non-blocking */
149 if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
150 RTE_LOG(WARNING, PMD,
151 "Unable to set %s to nonblocking\n",
153 perror("F_SETFL, NONBLOCK");
157 /* Set up trigger to optimize empty Rx bursts */
161 int flags = fcntl(fd, F_GETFL);
163 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
165 if (sa.sa_handler != tap_trigger_cb) {
167 * Make sure SIGIO is not already taken. This is done
168 * as late as possible to leave the application a
169 * chance to set up its own signal handler first.
171 if (sa.sa_handler != SIG_IGN &&
172 sa.sa_handler != SIG_DFL) {
176 sa = (struct sigaction){
177 .sa_flags = SA_RESTART,
178 .sa_handler = tap_trigger_cb,
180 if (sigaction(SIGIO, &sa, NULL) == -1)
183 /* Enable SIGIO on file descriptor */
184 fcntl(fd, F_SETFL, flags | O_ASYNC);
185 fcntl(fd, F_SETOWN, getpid());
188 /* Disable trigger globally in case of error */
190 RTE_LOG(WARNING, PMD, "Rx trigger disabled: %s\n",
203 tap_verify_csum(struct rte_mbuf *mbuf)
205 uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
206 uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
207 uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
208 unsigned int l2_len = sizeof(struct ether_hdr);
214 if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
216 else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
218 /* Don't verify checksum for packets with discontinuous L2 header */
219 if (unlikely(l2_len + sizeof(struct ipv4_hdr) >
220 rte_pktmbuf_data_len(mbuf)))
222 l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
223 if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
224 struct ipv4_hdr *iph = l3_hdr;
226 /* ihl contains the number of 4-byte words in the header */
227 l3_len = 4 * (iph->version_ihl & 0xf);
228 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
231 cksum = ~rte_raw_cksum(iph, l3_len);
232 mbuf->ol_flags |= cksum ?
233 PKT_RX_IP_CKSUM_BAD :
234 PKT_RX_IP_CKSUM_GOOD;
235 } else if (l3 == RTE_PTYPE_L3_IPV6) {
236 l3_len = sizeof(struct ipv6_hdr);
238 /* IPv6 extensions are not supported */
241 if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
242 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
243 /* Don't verify checksum for multi-segment packets. */
244 if (mbuf->nb_segs > 1)
246 if (l3 == RTE_PTYPE_L3_IPV4)
247 cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
248 else if (l3 == RTE_PTYPE_L3_IPV6)
249 cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
250 mbuf->ol_flags |= cksum ?
251 PKT_RX_L4_CKSUM_BAD :
252 PKT_RX_L4_CKSUM_GOOD;
257 tap_rx_offload_get_port_capa(void)
260 * In order to support legacy apps,
261 * report capabilities also as port capabilities.
263 return DEV_RX_OFFLOAD_SCATTER |
264 DEV_RX_OFFLOAD_IPV4_CKSUM |
265 DEV_RX_OFFLOAD_UDP_CKSUM |
266 DEV_RX_OFFLOAD_TCP_CKSUM;
270 tap_rx_offload_get_queue_capa(void)
272 return DEV_RX_OFFLOAD_SCATTER |
273 DEV_RX_OFFLOAD_IPV4_CKSUM |
274 DEV_RX_OFFLOAD_UDP_CKSUM |
275 DEV_RX_OFFLOAD_TCP_CKSUM;
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_IPV4_CKSUM |
396 DEV_TX_OFFLOAD_UDP_CKSUM |
397 DEV_TX_OFFLOAD_TCP_CKSUM;
401 tap_tx_offload_get_queue_capa(void)
403 return DEV_TX_OFFLOAD_IPV4_CKSUM |
404 DEV_TX_OFFLOAD_UDP_CKSUM |
405 DEV_TX_OFFLOAD_TCP_CKSUM;
409 tap_txq_are_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
411 uint64_t port_offloads = dev->data->dev_conf.txmode.offloads;
412 uint64_t queue_supp_offloads = tap_tx_offload_get_queue_capa();
413 uint64_t port_supp_offloads = tap_tx_offload_get_port_capa();
415 if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
418 /* Verify we have no conflict with port offloads */
419 if ((port_offloads ^ offloads) & port_supp_offloads)
425 tap_tx_offload(char *packet, uint64_t ol_flags, unsigned int l2_len,
428 void *l3_hdr = packet + l2_len;
430 if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
431 struct ipv4_hdr *iph = l3_hdr;
434 iph->hdr_checksum = 0;
435 cksum = rte_raw_cksum(iph, l3_len);
436 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
438 if (ol_flags & PKT_TX_L4_MASK) {
444 l4_hdr = packet + l2_len + l3_len;
445 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
446 l4_cksum = &((struct udp_hdr *)l4_hdr)->dgram_cksum;
447 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
448 l4_cksum = &((struct tcp_hdr *)l4_hdr)->cksum;
452 if (ol_flags & PKT_TX_IPV4) {
453 struct ipv4_hdr *iph = l3_hdr;
455 l4_len = rte_be_to_cpu_16(iph->total_length) - l3_len;
456 cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
458 struct ipv6_hdr *ip6h = l3_hdr;
460 /* payload_len does not include ext headers */
461 l4_len = rte_be_to_cpu_16(ip6h->payload_len) -
462 l3_len + sizeof(struct ipv6_hdr);
463 cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
465 cksum += rte_raw_cksum(l4_hdr, l4_len);
466 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
467 cksum = (~cksum) & 0xffff;
474 /* Callback to handle sending packets from the tap interface
477 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
479 struct tx_queue *txq = queue;
481 unsigned long num_tx_bytes = 0;
485 if (unlikely(nb_pkts == 0))
488 max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
489 for (i = 0; i < nb_pkts; i++) {
490 struct rte_mbuf *mbuf = bufs[num_tx];
491 struct iovec iovecs[mbuf->nb_segs + 1];
492 struct tun_pi pi = { .flags = 0 };
493 struct rte_mbuf *seg = mbuf;
494 char m_copy[mbuf->data_len];
498 /* stats.errs will be incremented */
499 if (rte_pktmbuf_pkt_len(mbuf) > max_size)
502 iovecs[0].iov_base = π
503 iovecs[0].iov_len = sizeof(pi);
504 for (j = 1; j <= mbuf->nb_segs; j++) {
505 iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
507 rte_pktmbuf_mtod(seg, void *);
511 ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
512 (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
513 (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
514 /* Support only packets with all data in the same seg */
515 if (mbuf->nb_segs > 1)
517 /* To change checksums, work on a copy of data. */
518 rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
519 rte_pktmbuf_data_len(mbuf));
520 tap_tx_offload(m_copy, mbuf->ol_flags,
521 mbuf->l2_len, mbuf->l3_len);
522 iovecs[1].iov_base = m_copy;
524 /* copy the tx frame data */
525 n = writev(txq->fd, iovecs, mbuf->nb_segs + 1);
530 num_tx_bytes += mbuf->pkt_len;
531 rte_pktmbuf_free(mbuf);
534 txq->stats.opackets += num_tx;
535 txq->stats.errs += nb_pkts - num_tx;
536 txq->stats.obytes += num_tx_bytes;
542 tap_ioctl_req2str(unsigned long request)
546 return "SIOCSIFFLAGS";
548 return "SIOCGIFFLAGS";
550 return "SIOCGIFHWADDR";
552 return "SIOCSIFHWADDR";
560 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
561 struct ifreq *ifr, int set, enum ioctl_mode mode)
563 short req_flags = ifr->ifr_flags;
564 int remote = pmd->remote_if_index &&
565 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
567 if (!pmd->remote_if_index && mode == REMOTE_ONLY)
570 * If there is a remote netdevice, apply ioctl on it, then apply it on
575 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
576 else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
577 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
580 /* fetch current flags to leave other flags untouched */
581 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
584 ifr->ifr_flags |= req_flags;
586 ifr->ifr_flags &= ~req_flags;
594 RTE_ASSERT(!"unsupported request type: must not happen");
596 if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
598 if (remote-- && mode == LOCAL_AND_REMOTE)
603 RTE_LOG(DEBUG, PMD, "%s: %s(%s) failed: %s(%d)\n", ifr->ifr_name,
604 __func__, tap_ioctl_req2str(request), strerror(errno), errno);
609 tap_link_set_down(struct rte_eth_dev *dev)
611 struct pmd_internals *pmd = dev->data->dev_private;
612 struct ifreq ifr = { .ifr_flags = IFF_UP };
614 dev->data->dev_link.link_status = ETH_LINK_DOWN;
615 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
619 tap_link_set_up(struct rte_eth_dev *dev)
621 struct pmd_internals *pmd = dev->data->dev_private;
622 struct ifreq ifr = { .ifr_flags = IFF_UP };
624 dev->data->dev_link.link_status = ETH_LINK_UP;
625 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
629 tap_dev_start(struct rte_eth_dev *dev)
633 err = tap_intr_handle_set(dev, 1);
636 return tap_link_set_up(dev);
639 /* This function gets called when the current port gets stopped.
642 tap_dev_stop(struct rte_eth_dev *dev)
644 tap_intr_handle_set(dev, 0);
645 tap_link_set_down(dev);
649 tap_dev_configure(struct rte_eth_dev *dev)
651 uint64_t supp_tx_offloads = tap_tx_offload_get_port_capa();
652 uint64_t tx_offloads = dev->data->dev_conf.txmode.offloads;
654 if ((tx_offloads & supp_tx_offloads) != tx_offloads) {
657 "Some Tx offloads are not supported "
658 "requested 0x%" PRIx64 " supported 0x%" PRIx64 "\n",
659 tx_offloads, supp_tx_offloads);
662 if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
664 "%s: number of rx queues %d exceeds max num of queues %d\n",
666 dev->data->nb_rx_queues,
667 RTE_PMD_TAP_MAX_QUEUES);
670 if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
672 "%s: number of tx queues %d exceeds max num of queues %d\n",
674 dev->data->nb_tx_queues,
675 RTE_PMD_TAP_MAX_QUEUES);
679 RTE_LOG(INFO, PMD, "%s: %p: TX configured queues number: %u\n",
680 dev->device->name, (void *)dev, dev->data->nb_tx_queues);
682 RTE_LOG(INFO, PMD, "%s: %p: RX configured queues number: %u\n",
683 dev->device->name, (void *)dev, dev->data->nb_rx_queues);
689 tap_dev_speed_capa(void)
691 uint32_t speed = pmd_link.link_speed;
694 if (speed >= ETH_SPEED_NUM_10M)
695 capa |= ETH_LINK_SPEED_10M;
696 if (speed >= ETH_SPEED_NUM_100M)
697 capa |= ETH_LINK_SPEED_100M;
698 if (speed >= ETH_SPEED_NUM_1G)
699 capa |= ETH_LINK_SPEED_1G;
700 if (speed >= ETH_SPEED_NUM_5G)
701 capa |= ETH_LINK_SPEED_2_5G;
702 if (speed >= ETH_SPEED_NUM_5G)
703 capa |= ETH_LINK_SPEED_5G;
704 if (speed >= ETH_SPEED_NUM_10G)
705 capa |= ETH_LINK_SPEED_10G;
706 if (speed >= ETH_SPEED_NUM_20G)
707 capa |= ETH_LINK_SPEED_20G;
708 if (speed >= ETH_SPEED_NUM_25G)
709 capa |= ETH_LINK_SPEED_25G;
710 if (speed >= ETH_SPEED_NUM_40G)
711 capa |= ETH_LINK_SPEED_40G;
712 if (speed >= ETH_SPEED_NUM_50G)
713 capa |= ETH_LINK_SPEED_50G;
714 if (speed >= ETH_SPEED_NUM_56G)
715 capa |= ETH_LINK_SPEED_56G;
716 if (speed >= ETH_SPEED_NUM_100G)
717 capa |= ETH_LINK_SPEED_100G;
723 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
725 struct pmd_internals *internals = dev->data->dev_private;
727 dev_info->if_index = internals->if_index;
728 dev_info->max_mac_addrs = 1;
729 dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
730 dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
731 dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
732 dev_info->min_rx_bufsize = 0;
733 dev_info->pci_dev = NULL;
734 dev_info->speed_capa = tap_dev_speed_capa();
735 dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
736 dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
737 dev_info->rx_queue_offload_capa;
738 dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
739 dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
740 dev_info->tx_queue_offload_capa;
744 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
746 unsigned int i, imax;
747 unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
748 unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
749 unsigned long rx_nombuf = 0, ierrors = 0;
750 const struct pmd_internals *pmd = dev->data->dev_private;
752 /* rx queue statistics */
753 imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
754 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
755 for (i = 0; i < imax; i++) {
756 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
757 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
758 rx_total += tap_stats->q_ipackets[i];
759 rx_bytes_total += tap_stats->q_ibytes[i];
760 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
761 ierrors += pmd->rxq[i].stats.ierrors;
764 /* tx queue statistics */
765 imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
766 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
768 for (i = 0; i < imax; i++) {
769 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
770 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
771 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
772 tx_total += tap_stats->q_opackets[i];
773 tx_err_total += tap_stats->q_errors[i];
774 tx_bytes_total += tap_stats->q_obytes[i];
777 tap_stats->ipackets = rx_total;
778 tap_stats->ibytes = rx_bytes_total;
779 tap_stats->ierrors = ierrors;
780 tap_stats->rx_nombuf = rx_nombuf;
781 tap_stats->opackets = tx_total;
782 tap_stats->oerrors = tx_err_total;
783 tap_stats->obytes = tx_bytes_total;
788 tap_stats_reset(struct rte_eth_dev *dev)
791 struct pmd_internals *pmd = dev->data->dev_private;
793 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
794 pmd->rxq[i].stats.ipackets = 0;
795 pmd->rxq[i].stats.ibytes = 0;
796 pmd->rxq[i].stats.ierrors = 0;
797 pmd->rxq[i].stats.rx_nombuf = 0;
799 pmd->txq[i].stats.opackets = 0;
800 pmd->txq[i].stats.errs = 0;
801 pmd->txq[i].stats.obytes = 0;
806 tap_dev_close(struct rte_eth_dev *dev)
809 struct pmd_internals *internals = dev->data->dev_private;
811 tap_link_set_down(dev);
812 tap_flow_flush(dev, NULL);
813 tap_flow_implicit_flush(internals, NULL);
815 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
816 if (internals->rxq[i].fd != -1) {
817 close(internals->rxq[i].fd);
818 internals->rxq[i].fd = -1;
820 if (internals->txq[i].fd != -1) {
821 close(internals->txq[i].fd);
822 internals->txq[i].fd = -1;
826 if (internals->remote_if_index) {
827 /* Restore initial remote state */
828 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
829 &internals->remote_initial_flags);
834 tap_rx_queue_release(void *queue)
836 struct rx_queue *rxq = queue;
838 if (rxq && (rxq->fd > 0)) {
841 rte_pktmbuf_free(rxq->pool);
842 rte_free(rxq->iovecs);
849 tap_tx_queue_release(void *queue)
851 struct tx_queue *txq = queue;
853 if (txq && (txq->fd > 0)) {
860 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
862 struct rte_eth_link *dev_link = &dev->data->dev_link;
863 struct pmd_internals *pmd = dev->data->dev_private;
864 struct ifreq ifr = { .ifr_flags = 0 };
866 if (pmd->remote_if_index) {
867 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
868 if (!(ifr.ifr_flags & IFF_UP) ||
869 !(ifr.ifr_flags & IFF_RUNNING)) {
870 dev_link->link_status = ETH_LINK_DOWN;
874 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
875 dev_link->link_status =
876 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
883 tap_promisc_enable(struct rte_eth_dev *dev)
885 struct pmd_internals *pmd = dev->data->dev_private;
886 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
888 dev->data->promiscuous = 1;
889 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
890 if (pmd->remote_if_index && !pmd->flow_isolate)
891 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
895 tap_promisc_disable(struct rte_eth_dev *dev)
897 struct pmd_internals *pmd = dev->data->dev_private;
898 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
900 dev->data->promiscuous = 0;
901 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
902 if (pmd->remote_if_index && !pmd->flow_isolate)
903 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
907 tap_allmulti_enable(struct rte_eth_dev *dev)
909 struct pmd_internals *pmd = dev->data->dev_private;
910 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
912 dev->data->all_multicast = 1;
913 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
914 if (pmd->remote_if_index && !pmd->flow_isolate)
915 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
919 tap_allmulti_disable(struct rte_eth_dev *dev)
921 struct pmd_internals *pmd = dev->data->dev_private;
922 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
924 dev->data->all_multicast = 0;
925 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
926 if (pmd->remote_if_index && !pmd->flow_isolate)
927 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
931 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
933 struct pmd_internals *pmd = dev->data->dev_private;
934 enum ioctl_mode mode = LOCAL_ONLY;
937 if (is_zero_ether_addr(mac_addr)) {
938 RTE_LOG(ERR, PMD, "%s: can't set an empty MAC address\n",
942 /* Check the actual current MAC address on the tap netdevice */
943 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
945 if (is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
948 /* Check the current MAC address on the remote */
949 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0)
951 if (!is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
953 mode = LOCAL_AND_REMOTE;
954 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
955 rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
956 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode) < 0)
958 rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
959 if (pmd->remote_if_index && !pmd->flow_isolate) {
960 /* Replace MAC redirection rule after a MAC change */
961 if (tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC) < 0) {
963 "%s: Couldn't delete MAC redirection rule\n",
967 if (tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0)
969 "%s: Couldn't add MAC redirection rule\n",
975 tap_setup_queue(struct rte_eth_dev *dev,
976 struct pmd_internals *internals,
983 struct pmd_internals *pmd = dev->data->dev_private;
984 struct rx_queue *rx = &internals->rxq[qid];
985 struct tx_queue *tx = &internals->txq[qid];
997 /* fd for this queue already exists */
998 RTE_LOG(DEBUG, PMD, "%s: fd %d for %s queue qid %d exists\n",
999 pmd->name, *fd, dir, qid);
1000 } else if (*other_fd != -1) {
1001 /* Only other_fd exists. dup it */
1002 *fd = dup(*other_fd);
1005 RTE_LOG(ERR, PMD, "%s: dup() failed.\n",
1009 RTE_LOG(DEBUG, PMD, "%s: dup fd %d for %s queue qid %d (%d)\n",
1010 pmd->name, *other_fd, dir, qid, *fd);
1012 /* Both RX and TX fds do not exist (equal -1). Create fd */
1013 *fd = tun_alloc(pmd);
1015 *fd = -1; /* restore original value */
1016 RTE_LOG(ERR, PMD, "%s: tun_alloc() failed.\n",
1020 RTE_LOG(DEBUG, PMD, "%s: add %s queue for qid %d fd %d\n",
1021 pmd->name, dir, qid, *fd);
1024 tx->mtu = &dev->data->mtu;
1025 rx->rxmode = &dev->data->dev_conf.rxmode;
1031 tap_rx_queue_setup(struct rte_eth_dev *dev,
1032 uint16_t rx_queue_id,
1033 uint16_t nb_rx_desc,
1034 unsigned int socket_id,
1035 const struct rte_eth_rxconf *rx_conf __rte_unused,
1036 struct rte_mempool *mp)
1038 struct pmd_internals *internals = dev->data->dev_private;
1039 struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1040 struct rte_mbuf **tmp = &rxq->pool;
1041 long iov_max = sysconf(_SC_IOV_MAX);
1042 uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1043 struct iovec (*iovecs)[nb_desc + 1];
1044 int data_off = RTE_PKTMBUF_HEADROOM;
1049 if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1050 RTE_LOG(WARNING, PMD,
1051 "nb_rx_queues %d too small or mempool NULL\n",
1052 dev->data->nb_rx_queues);
1056 /* Verify application offloads are valid for our port and queue. */
1057 if (!tap_rxq_are_offloads_valid(dev, rx_conf->offloads)) {
1058 rte_errno = ENOTSUP;
1060 "%p: Rx queue offloads 0x%" PRIx64
1061 " don't match port offloads 0x%" PRIx64
1062 " or supported offloads 0x%" PRIx64 "\n",
1063 (void *)dev, rx_conf->offloads,
1064 dev->data->dev_conf.rxmode.offloads,
1065 (tap_rx_offload_get_port_capa() |
1066 tap_rx_offload_get_queue_capa()));
1070 rxq->trigger_seen = 1; /* force initial burst */
1071 rxq->in_port = dev->data->port_id;
1072 rxq->nb_rx_desc = nb_desc;
1073 iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1076 RTE_LOG(WARNING, PMD,
1077 "%s: Couldn't allocate %d RX descriptors\n",
1078 dev->device->name, nb_desc);
1081 rxq->iovecs = iovecs;
1083 dev->data->rx_queues[rx_queue_id] = rxq;
1084 fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1090 (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1091 (*rxq->iovecs)[0].iov_base = &rxq->pi;
1093 for (i = 1; i <= nb_desc; i++) {
1094 *tmp = rte_pktmbuf_alloc(rxq->mp);
1096 RTE_LOG(WARNING, PMD,
1097 "%s: couldn't allocate memory for queue %d\n",
1098 dev->device->name, rx_queue_id);
1102 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1103 (*rxq->iovecs)[i].iov_base =
1104 (char *)(*tmp)->buf_addr + data_off;
1106 tmp = &(*tmp)->next;
1109 RTE_LOG(DEBUG, PMD, " RX TAP device name %s, qid %d on fd %d\n",
1110 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
1115 rte_pktmbuf_free(rxq->pool);
1117 rte_free(rxq->iovecs);
1123 tap_tx_queue_setup(struct rte_eth_dev *dev,
1124 uint16_t tx_queue_id,
1125 uint16_t nb_tx_desc __rte_unused,
1126 unsigned int socket_id __rte_unused,
1127 const struct rte_eth_txconf *tx_conf)
1129 struct pmd_internals *internals = dev->data->dev_private;
1130 struct tx_queue *txq;
1133 if (tx_queue_id >= dev->data->nb_tx_queues)
1135 dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1136 txq = dev->data->tx_queues[tx_queue_id];
1138 * Don't verify port offloads for application which
1141 if (tx_conf != NULL &&
1142 !!(tx_conf->txq_flags & ETH_TXQ_FLAGS_IGNORE)) {
1143 if (tap_txq_are_offloads_valid(dev, tx_conf->offloads)) {
1144 txq->csum = !!(tx_conf->offloads &
1145 (DEV_TX_OFFLOAD_IPV4_CKSUM |
1146 DEV_TX_OFFLOAD_UDP_CKSUM |
1147 DEV_TX_OFFLOAD_TCP_CKSUM));
1149 rte_errno = ENOTSUP;
1151 "%p: Tx queue offloads 0x%" PRIx64
1152 " don't match port offloads 0x%" PRIx64
1153 " or supported offloads 0x%" PRIx64,
1154 (void *)dev, tx_conf->offloads,
1155 dev->data->dev_conf.txmode.offloads,
1156 tap_tx_offload_get_port_capa());
1160 ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1164 " TX TAP device name %s, qid %d on fd %d csum %s\n",
1165 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd,
1166 txq->csum ? "on" : "off");
1172 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1174 struct pmd_internals *pmd = dev->data->dev_private;
1175 struct ifreq ifr = { .ifr_mtu = mtu };
1178 err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1180 dev->data->mtu = mtu;
1186 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1187 struct ether_addr *mc_addr_set __rte_unused,
1188 uint32_t nb_mc_addr __rte_unused)
1191 * Nothing to do actually: the tap has no filtering whatsoever, every
1192 * packet is received.
1198 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1200 struct rte_eth_dev *dev = arg;
1201 struct pmd_internals *pmd = dev->data->dev_private;
1202 struct ifinfomsg *info = NLMSG_DATA(nh);
1204 if (nh->nlmsg_type != RTM_NEWLINK ||
1205 (info->ifi_index != pmd->if_index &&
1206 info->ifi_index != pmd->remote_if_index))
1208 return tap_link_update(dev, 0);
1212 tap_dev_intr_handler(void *cb_arg)
1214 struct rte_eth_dev *dev = cb_arg;
1215 struct pmd_internals *pmd = dev->data->dev_private;
1217 tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1221 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1223 struct pmd_internals *pmd = dev->data->dev_private;
1225 /* In any case, disable interrupt if the conf is no longer there. */
1226 if (!dev->data->dev_conf.intr_conf.lsc) {
1227 if (pmd->intr_handle.fd != -1) {
1228 tap_nl_final(pmd->intr_handle.fd);
1229 rte_intr_callback_unregister(&pmd->intr_handle,
1230 tap_dev_intr_handler, dev);
1235 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1236 if (unlikely(pmd->intr_handle.fd == -1))
1238 return rte_intr_callback_register(
1239 &pmd->intr_handle, tap_dev_intr_handler, dev);
1241 tap_nl_final(pmd->intr_handle.fd);
1242 return rte_intr_callback_unregister(&pmd->intr_handle,
1243 tap_dev_intr_handler, dev);
1246 static const uint32_t*
1247 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1249 static const uint32_t ptypes[] = {
1250 RTE_PTYPE_INNER_L2_ETHER,
1251 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1252 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1253 RTE_PTYPE_INNER_L3_IPV4,
1254 RTE_PTYPE_INNER_L3_IPV4_EXT,
1255 RTE_PTYPE_INNER_L3_IPV6,
1256 RTE_PTYPE_INNER_L3_IPV6_EXT,
1257 RTE_PTYPE_INNER_L4_FRAG,
1258 RTE_PTYPE_INNER_L4_UDP,
1259 RTE_PTYPE_INNER_L4_TCP,
1260 RTE_PTYPE_INNER_L4_SCTP,
1262 RTE_PTYPE_L2_ETHER_VLAN,
1263 RTE_PTYPE_L2_ETHER_QINQ,
1265 RTE_PTYPE_L3_IPV4_EXT,
1266 RTE_PTYPE_L3_IPV6_EXT,
1278 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1279 struct rte_eth_fc_conf *fc_conf)
1281 fc_conf->mode = RTE_FC_NONE;
1286 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1287 struct rte_eth_fc_conf *fc_conf)
1289 if (fc_conf->mode != RTE_FC_NONE)
1294 static const struct eth_dev_ops ops = {
1295 .dev_start = tap_dev_start,
1296 .dev_stop = tap_dev_stop,
1297 .dev_close = tap_dev_close,
1298 .dev_configure = tap_dev_configure,
1299 .dev_infos_get = tap_dev_info,
1300 .rx_queue_setup = tap_rx_queue_setup,
1301 .tx_queue_setup = tap_tx_queue_setup,
1302 .rx_queue_release = tap_rx_queue_release,
1303 .tx_queue_release = tap_tx_queue_release,
1304 .flow_ctrl_get = tap_flow_ctrl_get,
1305 .flow_ctrl_set = tap_flow_ctrl_set,
1306 .link_update = tap_link_update,
1307 .dev_set_link_up = tap_link_set_up,
1308 .dev_set_link_down = tap_link_set_down,
1309 .promiscuous_enable = tap_promisc_enable,
1310 .promiscuous_disable = tap_promisc_disable,
1311 .allmulticast_enable = tap_allmulti_enable,
1312 .allmulticast_disable = tap_allmulti_disable,
1313 .mac_addr_set = tap_mac_set,
1314 .mtu_set = tap_mtu_set,
1315 .set_mc_addr_list = tap_set_mc_addr_list,
1316 .stats_get = tap_stats_get,
1317 .stats_reset = tap_stats_reset,
1318 .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1319 .filter_ctrl = tap_dev_filter_ctrl,
1323 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
1324 char *remote_iface, int fixed_mac_type)
1326 int numa_node = rte_socket_id();
1327 struct rte_eth_dev *dev;
1328 struct pmd_internals *pmd;
1329 struct rte_eth_dev_data *data;
1333 RTE_LOG(DEBUG, PMD, " TAP device on numa %u\n", rte_socket_id());
1335 data = rte_zmalloc_socket(tap_name, sizeof(*data), 0, numa_node);
1337 RTE_LOG(ERR, PMD, "TAP Failed to allocate data\n");
1341 dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1343 RTE_LOG(ERR, PMD, "TAP Unable to allocate device struct\n");
1347 pmd = dev->data->dev_private;
1349 snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1351 pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1352 if (pmd->ioctl_sock == -1) {
1354 "TAP Unable to get a socket for management: %s\n",
1359 /* Setup some default values */
1360 rte_memcpy(data, dev->data, sizeof(*data));
1361 data->dev_private = pmd;
1362 data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1363 data->numa_node = numa_node;
1365 data->dev_link = pmd_link;
1366 data->mac_addrs = &pmd->eth_addr;
1367 /* Set the number of RX and TX queues */
1368 data->nb_rx_queues = 0;
1369 data->nb_tx_queues = 0;
1372 dev->dev_ops = &ops;
1373 dev->rx_pkt_burst = pmd_rx_burst;
1374 dev->tx_pkt_burst = pmd_tx_burst;
1376 pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1377 pmd->intr_handle.fd = -1;
1379 /* Presetup the fds to -1 as being not valid */
1380 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1381 pmd->rxq[i].fd = -1;
1382 pmd->txq[i].fd = -1;
1385 if (fixed_mac_type) {
1386 /* fixed mac = 00:64:74:61:70:<iface_idx> */
1387 static int iface_idx;
1388 char mac[ETHER_ADDR_LEN] = "\0dtap";
1390 mac[ETHER_ADDR_LEN - 1] = iface_idx++;
1391 rte_memcpy(&pmd->eth_addr, mac, ETHER_ADDR_LEN);
1393 eth_random_addr((uint8_t *)&pmd->eth_addr);
1396 /* Immediately create the netdevice (this will create the 1st queue). */
1398 if (tap_setup_queue(dev, pmd, 0, 1) == -1)
1401 if (tap_setup_queue(dev, pmd, 0, 0) == -1)
1404 ifr.ifr_mtu = dev->data->mtu;
1405 if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1408 memset(&ifr, 0, sizeof(struct ifreq));
1409 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1410 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr, ETHER_ADDR_LEN);
1411 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1415 * Set up everything related to rte_flow:
1417 * - tap / remote if_index
1418 * - mandatory QDISCs
1419 * - rte_flow actual/implicit lists
1422 pmd->nlsk_fd = tap_nl_init(0);
1423 if (pmd->nlsk_fd == -1) {
1424 RTE_LOG(WARNING, PMD, "%s: failed to create netlink socket.\n",
1426 goto disable_rte_flow;
1428 pmd->if_index = if_nametoindex(pmd->name);
1429 if (!pmd->if_index) {
1430 RTE_LOG(ERR, PMD, "%s: failed to get if_index.\n", pmd->name);
1431 goto disable_rte_flow;
1433 if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1434 RTE_LOG(ERR, PMD, "%s: failed to create multiq qdisc.\n",
1436 goto disable_rte_flow;
1438 if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1439 RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1441 goto disable_rte_flow;
1443 LIST_INIT(&pmd->flows);
1445 if (strlen(remote_iface)) {
1446 pmd->remote_if_index = if_nametoindex(remote_iface);
1447 if (!pmd->remote_if_index) {
1448 RTE_LOG(ERR, PMD, "%s: failed to get %s if_index.\n",
1449 pmd->name, remote_iface);
1452 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1453 "%s", remote_iface);
1455 /* Save state of remote device */
1456 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1458 /* Replicate remote MAC address */
1459 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1460 RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1461 pmd->name, pmd->remote_iface);
1464 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1466 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1467 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1468 RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1469 pmd->name, remote_iface);
1474 * Flush usually returns negative value because it tries to
1475 * delete every QDISC (and on a running device, one QDISC at
1476 * least is needed). Ignore negative return value.
1478 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1479 if (qdisc_create_ingress(pmd->nlsk_fd,
1480 pmd->remote_if_index) < 0) {
1481 RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1485 LIST_INIT(&pmd->implicit_flows);
1486 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1487 tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1488 tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1489 tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1491 "%s: failed to create implicit rules.\n",
1500 RTE_LOG(ERR, PMD, " Disabling rte flow support: %s(%d)\n",
1501 strerror(errno), errno);
1502 if (strlen(remote_iface)) {
1503 RTE_LOG(ERR, PMD, "Remote feature requires flow support.\n");
1509 RTE_LOG(ERR, PMD, " Can't set up remote feature: %s(%d)\n",
1510 strerror(errno), errno);
1511 tap_flow_implicit_flush(pmd, NULL);
1514 RTE_LOG(ERR, PMD, "TAP Unable to initialize %s\n",
1515 rte_vdev_device_name(vdev));
1522 set_interface_name(const char *key __rte_unused,
1526 char *name = (char *)extra_args;
1529 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s", value);
1531 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1532 DEFAULT_TAP_NAME, (tap_unit - 1));
1538 set_interface_speed(const char *key __rte_unused,
1542 *(int *)extra_args = (value) ? atoi(value) : ETH_SPEED_NUM_10G;
1548 set_remote_iface(const char *key __rte_unused,
1552 char *name = (char *)extra_args;
1555 snprintf(name, RTE_ETH_NAME_MAX_LEN, "%s", value);
1561 set_mac_type(const char *key __rte_unused,
1566 !strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED)))
1567 *(int *)extra_args = 1;
1571 /* Open a TAP interface device.
1574 rte_pmd_tap_probe(struct rte_vdev_device *dev)
1576 const char *name, *params;
1578 struct rte_kvargs *kvlist = NULL;
1580 char tap_name[RTE_ETH_NAME_MAX_LEN];
1581 char remote_iface[RTE_ETH_NAME_MAX_LEN];
1582 int fixed_mac_type = 0;
1584 name = rte_vdev_device_name(dev);
1585 params = rte_vdev_device_args(dev);
1587 speed = ETH_SPEED_NUM_10G;
1588 snprintf(tap_name, sizeof(tap_name), "%s%d",
1589 DEFAULT_TAP_NAME, tap_unit++);
1590 memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1592 if (params && (params[0] != '\0')) {
1593 RTE_LOG(DEBUG, PMD, "parameters (%s)\n", params);
1595 kvlist = rte_kvargs_parse(params, valid_arguments);
1597 if (rte_kvargs_count(kvlist, ETH_TAP_SPEED_ARG) == 1) {
1598 ret = rte_kvargs_process(kvlist,
1600 &set_interface_speed,
1606 if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1607 ret = rte_kvargs_process(kvlist,
1609 &set_interface_name,
1615 if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
1616 ret = rte_kvargs_process(kvlist,
1624 if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
1625 ret = rte_kvargs_process(kvlist,
1634 pmd_link.link_speed = speed;
1636 RTE_LOG(NOTICE, PMD, "Initializing pmd_tap for %s as %s\n",
1639 ret = eth_dev_tap_create(dev, tap_name, remote_iface, fixed_mac_type);
1643 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
1645 tap_unit--; /* Restore the unit number */
1647 rte_kvargs_free(kvlist);
1652 /* detach a TAP device.
1655 rte_pmd_tap_remove(struct rte_vdev_device *dev)
1657 struct rte_eth_dev *eth_dev = NULL;
1658 struct pmd_internals *internals;
1661 RTE_LOG(DEBUG, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
1664 /* find the ethdev entry */
1665 eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1669 internals = eth_dev->data->dev_private;
1670 if (internals->nlsk_fd) {
1671 tap_flow_flush(eth_dev, NULL);
1672 tap_flow_implicit_flush(internals, NULL);
1673 tap_nl_final(internals->nlsk_fd);
1675 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1676 if (internals->rxq[i].fd != -1) {
1677 close(internals->rxq[i].fd);
1678 internals->rxq[i].fd = -1;
1680 if (internals->txq[i].fd != -1) {
1681 close(internals->txq[i].fd);
1682 internals->txq[i].fd = -1;
1686 close(internals->ioctl_sock);
1687 rte_free(eth_dev->data->dev_private);
1688 rte_free(eth_dev->data);
1690 rte_eth_dev_release_port(eth_dev);
1695 static struct rte_vdev_driver pmd_tap_drv = {
1696 .probe = rte_pmd_tap_probe,
1697 .remove = rte_pmd_tap_remove,
1699 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1700 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1701 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
1702 ETH_TAP_IFACE_ARG "=<string> "
1703 ETH_TAP_SPEED_ARG "=<int> "
1704 ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_FIXED " "
1705 ETH_TAP_REMOTE_ARG "=<string>");