1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2016-2017 Intel Corporation
5 #include <rte_atomic.h>
6 #include <rte_branch_prediction.h>
7 #include <rte_byteorder.h>
8 #include <rte_common.h>
10 #include <rte_ethdev_driver.h>
11 #include <rte_ethdev_vdev.h>
12 #include <rte_malloc.h>
13 #include <rte_bus_vdev.h>
14 #include <rte_kvargs.h>
16 #include <rte_debug.h>
18 #include <rte_string_fns.h>
20 #include <sys/types.h>
22 #include <sys/socket.h>
23 #include <sys/ioctl.h>
24 #include <sys/utsname.h>
32 #include <arpa/inet.h>
34 #include <linux/if_tun.h>
35 #include <linux/if_ether.h>
38 #include <rte_eth_tap.h>
40 #include <tap_netlink.h>
41 #include <tap_tcmsgs.h>
43 /* Linux based path to the TUN device */
44 #define TUN_TAP_DEV_PATH "/dev/net/tun"
45 #define DEFAULT_TAP_NAME "dtap"
47 #define ETH_TAP_IFACE_ARG "iface"
48 #define ETH_TAP_REMOTE_ARG "remote"
49 #define ETH_TAP_MAC_ARG "mac"
50 #define ETH_TAP_MAC_FIXED "fixed"
52 #define ETH_TAP_USR_MAC_FMT "xx:xx:xx:xx:xx:xx"
53 #define ETH_TAP_CMP_MAC_FMT "0123456789ABCDEFabcdef"
54 #define ETH_TAP_MAC_ARG_FMT ETH_TAP_MAC_FIXED "|" ETH_TAP_USR_MAC_FMT
56 static struct rte_vdev_driver pmd_tap_drv;
58 static const char *valid_arguments[] = {
67 static volatile uint32_t tap_trigger; /* Rx trigger */
69 static struct rte_eth_link pmd_link = {
70 .link_speed = ETH_SPEED_NUM_10G,
71 .link_duplex = ETH_LINK_FULL_DUPLEX,
72 .link_status = ETH_LINK_DOWN,
73 .link_autoneg = ETH_LINK_AUTONEG
77 tap_trigger_cb(int sig __rte_unused)
79 /* Valid trigger values are nonzero */
80 tap_trigger = (tap_trigger + 1) | 0x80000000;
83 /* Specifies on what netdevices the ioctl should be applied */
90 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
92 /* Tun/Tap allocation routine
94 * name is the number of the interface to use, unless NULL to take the host
98 tun_alloc(struct pmd_internals *pmd)
101 #ifdef IFF_MULTI_QUEUE
102 unsigned int features;
106 memset(&ifr, 0, sizeof(struct ifreq));
109 * Do not set IFF_NO_PI as packet information header will be needed
110 * to check if a received packet has been truncated.
112 ifr.ifr_flags = IFF_TAP;
113 snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
115 RTE_LOG(DEBUG, PMD, "ifr_name '%s'\n", ifr.ifr_name);
117 fd = open(TUN_TAP_DEV_PATH, O_RDWR);
119 RTE_LOG(ERR, PMD, "Unable to create TAP interface\n");
123 #ifdef IFF_MULTI_QUEUE
124 /* Grab the TUN features to verify we can work multi-queue */
125 if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
126 RTE_LOG(ERR, PMD, "TAP unable to get TUN/TAP features\n");
129 RTE_LOG(DEBUG, PMD, " TAP Features %08x\n", features);
131 if (features & IFF_MULTI_QUEUE) {
132 RTE_LOG(DEBUG, PMD, " Multi-queue support for %d queues\n",
133 RTE_PMD_TAP_MAX_QUEUES);
134 ifr.ifr_flags |= IFF_MULTI_QUEUE;
138 ifr.ifr_flags |= IFF_ONE_QUEUE;
139 RTE_LOG(DEBUG, PMD, " Single queue only support\n");
142 /* Set the TUN/TAP configuration and set the name if needed */
143 if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
144 RTE_LOG(WARNING, PMD,
145 "Unable to set TUNSETIFF for %s\n",
151 /* Always set the file descriptor to non-blocking */
152 if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
153 RTE_LOG(WARNING, PMD,
154 "Unable to set %s to nonblocking\n",
156 perror("F_SETFL, NONBLOCK");
160 /* Set up trigger to optimize empty Rx bursts */
164 int flags = fcntl(fd, F_GETFL);
166 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
168 if (sa.sa_handler != tap_trigger_cb) {
170 * Make sure SIGIO is not already taken. This is done
171 * as late as possible to leave the application a
172 * chance to set up its own signal handler first.
174 if (sa.sa_handler != SIG_IGN &&
175 sa.sa_handler != SIG_DFL) {
179 sa = (struct sigaction){
180 .sa_flags = SA_RESTART,
181 .sa_handler = tap_trigger_cb,
183 if (sigaction(SIGIO, &sa, NULL) == -1)
186 /* Enable SIGIO on file descriptor */
187 fcntl(fd, F_SETFL, flags | O_ASYNC);
188 fcntl(fd, F_SETOWN, getpid());
191 /* Disable trigger globally in case of error */
193 RTE_LOG(WARNING, PMD, "Rx trigger disabled: %s\n",
206 tap_verify_csum(struct rte_mbuf *mbuf)
208 uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
209 uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
210 uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
211 unsigned int l2_len = sizeof(struct ether_hdr);
217 if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
219 else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
221 /* Don't verify checksum for packets with discontinuous L2 header */
222 if (unlikely(l2_len + sizeof(struct ipv4_hdr) >
223 rte_pktmbuf_data_len(mbuf)))
225 l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
226 if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
227 struct ipv4_hdr *iph = l3_hdr;
229 /* ihl contains the number of 4-byte words in the header */
230 l3_len = 4 * (iph->version_ihl & 0xf);
231 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
234 cksum = ~rte_raw_cksum(iph, l3_len);
235 mbuf->ol_flags |= cksum ?
236 PKT_RX_IP_CKSUM_BAD :
237 PKT_RX_IP_CKSUM_GOOD;
238 } else if (l3 == RTE_PTYPE_L3_IPV6) {
239 l3_len = sizeof(struct ipv6_hdr);
241 /* IPv6 extensions are not supported */
244 if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
245 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
246 /* Don't verify checksum for multi-segment packets. */
247 if (mbuf->nb_segs > 1)
249 if (l3 == RTE_PTYPE_L3_IPV4)
250 cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
251 else if (l3 == RTE_PTYPE_L3_IPV6)
252 cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
253 mbuf->ol_flags |= cksum ?
254 PKT_RX_L4_CKSUM_BAD :
255 PKT_RX_L4_CKSUM_GOOD;
260 tap_rx_offload_get_port_capa(void)
263 * In order to support legacy apps,
264 * report capabilities also as port capabilities.
266 return DEV_RX_OFFLOAD_SCATTER |
267 DEV_RX_OFFLOAD_IPV4_CKSUM |
268 DEV_RX_OFFLOAD_UDP_CKSUM |
269 DEV_RX_OFFLOAD_TCP_CKSUM |
270 DEV_RX_OFFLOAD_CRC_STRIP;
274 tap_rx_offload_get_queue_capa(void)
276 return DEV_RX_OFFLOAD_SCATTER |
277 DEV_RX_OFFLOAD_IPV4_CKSUM |
278 DEV_RX_OFFLOAD_UDP_CKSUM |
279 DEV_RX_OFFLOAD_TCP_CKSUM |
280 DEV_RX_OFFLOAD_CRC_STRIP;
284 tap_rxq_are_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
286 uint64_t port_offloads = dev->data->dev_conf.rxmode.offloads;
287 uint64_t queue_supp_offloads = tap_rx_offload_get_queue_capa();
288 uint64_t port_supp_offloads = tap_rx_offload_get_port_capa();
290 if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
293 if ((port_offloads ^ offloads) & port_supp_offloads)
298 /* Callback to handle the rx burst of packets to the correct interface and
299 * file descriptor(s) in a multi-queue setup.
302 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
304 struct rx_queue *rxq = queue;
306 unsigned long num_rx_bytes = 0;
307 uint32_t trigger = tap_trigger;
309 if (trigger == rxq->trigger_seen)
312 rxq->trigger_seen = trigger;
313 rte_compiler_barrier();
314 for (num_rx = 0; num_rx < nb_pkts; ) {
315 struct rte_mbuf *mbuf = rxq->pool;
316 struct rte_mbuf *seg = NULL;
317 struct rte_mbuf *new_tail = NULL;
318 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
321 len = readv(rxq->fd, *rxq->iovecs,
323 (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
324 rxq->nb_rx_desc : 1));
325 if (len < (int)sizeof(struct tun_pi))
328 /* Packet couldn't fit in the provided mbuf */
329 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
330 rxq->stats.ierrors++;
334 len -= sizeof(struct tun_pi);
337 mbuf->port = rxq->in_port;
339 struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
341 if (unlikely(!buf)) {
342 rxq->stats.rx_nombuf++;
343 /* No new buf has been allocated: do nothing */
344 if (!new_tail || !seg)
348 rte_pktmbuf_free(mbuf);
352 seg = seg ? seg->next : mbuf;
353 if (rxq->pool == mbuf)
356 new_tail->next = buf;
358 new_tail->next = seg->next;
360 /* iovecs[0] is reserved for packet info (pi) */
361 (*rxq->iovecs)[mbuf->nb_segs].iov_len =
362 buf->buf_len - data_off;
363 (*rxq->iovecs)[mbuf->nb_segs].iov_base =
364 (char *)buf->buf_addr + data_off;
366 seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
367 seg->data_off = data_off;
369 len -= seg->data_len;
373 /* First segment has headroom, not the others */
377 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
379 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
380 tap_verify_csum(mbuf);
382 /* account for the receive frame */
383 bufs[num_rx++] = mbuf;
384 num_rx_bytes += mbuf->pkt_len;
387 rxq->stats.ipackets += num_rx;
388 rxq->stats.ibytes += num_rx_bytes;
394 tap_tx_offload_get_port_capa(void)
397 * In order to support legacy apps,
398 * report capabilities also as port capabilities.
400 return DEV_TX_OFFLOAD_MULTI_SEGS |
401 DEV_TX_OFFLOAD_IPV4_CKSUM |
402 DEV_TX_OFFLOAD_UDP_CKSUM |
403 DEV_TX_OFFLOAD_TCP_CKSUM;
407 tap_tx_offload_get_queue_capa(void)
409 return DEV_TX_OFFLOAD_MULTI_SEGS |
410 DEV_TX_OFFLOAD_IPV4_CKSUM |
411 DEV_TX_OFFLOAD_UDP_CKSUM |
412 DEV_TX_OFFLOAD_TCP_CKSUM;
416 tap_txq_are_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
418 uint64_t port_offloads = dev->data->dev_conf.txmode.offloads;
419 uint64_t queue_supp_offloads = tap_tx_offload_get_queue_capa();
420 uint64_t port_supp_offloads = tap_tx_offload_get_port_capa();
422 if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
425 /* Verify we have no conflict with port offloads */
426 if ((port_offloads ^ offloads) & port_supp_offloads)
432 tap_tx_offload(char *packet, uint64_t ol_flags, unsigned int l2_len,
435 void *l3_hdr = packet + l2_len;
437 if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
438 struct ipv4_hdr *iph = l3_hdr;
441 iph->hdr_checksum = 0;
442 cksum = rte_raw_cksum(iph, l3_len);
443 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
445 if (ol_flags & PKT_TX_L4_MASK) {
451 l4_hdr = packet + l2_len + l3_len;
452 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
453 l4_cksum = &((struct udp_hdr *)l4_hdr)->dgram_cksum;
454 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
455 l4_cksum = &((struct tcp_hdr *)l4_hdr)->cksum;
459 if (ol_flags & PKT_TX_IPV4) {
460 struct ipv4_hdr *iph = l3_hdr;
462 l4_len = rte_be_to_cpu_16(iph->total_length) - l3_len;
463 cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
465 struct ipv6_hdr *ip6h = l3_hdr;
467 /* payload_len does not include ext headers */
468 l4_len = rte_be_to_cpu_16(ip6h->payload_len) -
469 l3_len + sizeof(struct ipv6_hdr);
470 cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
472 cksum += rte_raw_cksum(l4_hdr, l4_len);
473 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
474 cksum = (~cksum) & 0xffff;
481 /* Callback to handle sending packets from the tap interface
484 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
486 struct tx_queue *txq = queue;
488 unsigned long num_tx_bytes = 0;
492 if (unlikely(nb_pkts == 0))
495 max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
496 for (i = 0; i < nb_pkts; i++) {
497 struct rte_mbuf *mbuf = bufs[num_tx];
498 struct iovec iovecs[mbuf->nb_segs + 1];
499 struct tun_pi pi = { .flags = 0 };
500 struct rte_mbuf *seg = mbuf;
501 char m_copy[mbuf->data_len];
505 /* stats.errs will be incremented */
506 if (rte_pktmbuf_pkt_len(mbuf) > max_size)
509 iovecs[0].iov_base = π
510 iovecs[0].iov_len = sizeof(pi);
511 for (j = 1; j <= mbuf->nb_segs; j++) {
512 iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
514 rte_pktmbuf_mtod(seg, void *);
518 ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
519 (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
520 (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
521 /* Support only packets with all data in the same seg */
522 if (mbuf->nb_segs > 1)
524 /* To change checksums, work on a copy of data. */
525 rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
526 rte_pktmbuf_data_len(mbuf));
527 tap_tx_offload(m_copy, mbuf->ol_flags,
528 mbuf->l2_len, mbuf->l3_len);
529 iovecs[1].iov_base = m_copy;
531 /* copy the tx frame data */
532 n = writev(txq->fd, iovecs, mbuf->nb_segs + 1);
537 num_tx_bytes += mbuf->pkt_len;
538 rte_pktmbuf_free(mbuf);
541 txq->stats.opackets += num_tx;
542 txq->stats.errs += nb_pkts - num_tx;
543 txq->stats.obytes += num_tx_bytes;
549 tap_ioctl_req2str(unsigned long request)
553 return "SIOCSIFFLAGS";
555 return "SIOCGIFFLAGS";
557 return "SIOCGIFHWADDR";
559 return "SIOCSIFHWADDR";
567 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
568 struct ifreq *ifr, int set, enum ioctl_mode mode)
570 short req_flags = ifr->ifr_flags;
571 int remote = pmd->remote_if_index &&
572 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
574 if (!pmd->remote_if_index && mode == REMOTE_ONLY)
577 * If there is a remote netdevice, apply ioctl on it, then apply it on
582 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
583 else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
584 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
587 /* fetch current flags to leave other flags untouched */
588 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
591 ifr->ifr_flags |= req_flags;
593 ifr->ifr_flags &= ~req_flags;
601 RTE_ASSERT(!"unsupported request type: must not happen");
603 if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
605 if (remote-- && mode == LOCAL_AND_REMOTE)
610 RTE_LOG(DEBUG, PMD, "%s: %s(%s) failed: %s(%d)\n", ifr->ifr_name,
611 __func__, tap_ioctl_req2str(request), strerror(errno), errno);
616 tap_link_set_down(struct rte_eth_dev *dev)
618 struct pmd_internals *pmd = dev->data->dev_private;
619 struct ifreq ifr = { .ifr_flags = IFF_UP };
621 dev->data->dev_link.link_status = ETH_LINK_DOWN;
622 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
626 tap_link_set_up(struct rte_eth_dev *dev)
628 struct pmd_internals *pmd = dev->data->dev_private;
629 struct ifreq ifr = { .ifr_flags = IFF_UP };
631 dev->data->dev_link.link_status = ETH_LINK_UP;
632 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
636 tap_dev_start(struct rte_eth_dev *dev)
640 err = tap_intr_handle_set(dev, 1);
643 return tap_link_set_up(dev);
646 /* This function gets called when the current port gets stopped.
649 tap_dev_stop(struct rte_eth_dev *dev)
651 tap_intr_handle_set(dev, 0);
652 tap_link_set_down(dev);
656 tap_dev_configure(struct rte_eth_dev *dev)
658 uint64_t supp_tx_offloads = tap_tx_offload_get_port_capa();
659 uint64_t tx_offloads = dev->data->dev_conf.txmode.offloads;
661 if ((tx_offloads & supp_tx_offloads) != tx_offloads) {
664 "Some Tx offloads are not supported "
665 "requested 0x%" PRIx64 " supported 0x%" PRIx64 "\n",
666 tx_offloads, supp_tx_offloads);
669 if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
671 "%s: number of rx queues %d exceeds max num of queues %d\n",
673 dev->data->nb_rx_queues,
674 RTE_PMD_TAP_MAX_QUEUES);
677 if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
679 "%s: number of tx queues %d exceeds max num of queues %d\n",
681 dev->data->nb_tx_queues,
682 RTE_PMD_TAP_MAX_QUEUES);
686 RTE_LOG(INFO, PMD, "%s: %p: TX configured queues number: %u\n",
687 dev->device->name, (void *)dev, dev->data->nb_tx_queues);
689 RTE_LOG(INFO, PMD, "%s: %p: RX configured queues number: %u\n",
690 dev->device->name, (void *)dev, dev->data->nb_rx_queues);
696 tap_dev_speed_capa(void)
698 uint32_t speed = pmd_link.link_speed;
701 if (speed >= ETH_SPEED_NUM_10M)
702 capa |= ETH_LINK_SPEED_10M;
703 if (speed >= ETH_SPEED_NUM_100M)
704 capa |= ETH_LINK_SPEED_100M;
705 if (speed >= ETH_SPEED_NUM_1G)
706 capa |= ETH_LINK_SPEED_1G;
707 if (speed >= ETH_SPEED_NUM_5G)
708 capa |= ETH_LINK_SPEED_2_5G;
709 if (speed >= ETH_SPEED_NUM_5G)
710 capa |= ETH_LINK_SPEED_5G;
711 if (speed >= ETH_SPEED_NUM_10G)
712 capa |= ETH_LINK_SPEED_10G;
713 if (speed >= ETH_SPEED_NUM_20G)
714 capa |= ETH_LINK_SPEED_20G;
715 if (speed >= ETH_SPEED_NUM_25G)
716 capa |= ETH_LINK_SPEED_25G;
717 if (speed >= ETH_SPEED_NUM_40G)
718 capa |= ETH_LINK_SPEED_40G;
719 if (speed >= ETH_SPEED_NUM_50G)
720 capa |= ETH_LINK_SPEED_50G;
721 if (speed >= ETH_SPEED_NUM_56G)
722 capa |= ETH_LINK_SPEED_56G;
723 if (speed >= ETH_SPEED_NUM_100G)
724 capa |= ETH_LINK_SPEED_100G;
730 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
732 struct pmd_internals *internals = dev->data->dev_private;
734 dev_info->if_index = internals->if_index;
735 dev_info->max_mac_addrs = 1;
736 dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
737 dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
738 dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
739 dev_info->min_rx_bufsize = 0;
740 dev_info->pci_dev = NULL;
741 dev_info->speed_capa = tap_dev_speed_capa();
742 dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
743 dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
744 dev_info->rx_queue_offload_capa;
745 dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
746 dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
747 dev_info->tx_queue_offload_capa;
751 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
753 unsigned int i, imax;
754 unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
755 unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
756 unsigned long rx_nombuf = 0, ierrors = 0;
757 const struct pmd_internals *pmd = dev->data->dev_private;
759 /* rx queue statistics */
760 imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
761 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
762 for (i = 0; i < imax; i++) {
763 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
764 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
765 rx_total += tap_stats->q_ipackets[i];
766 rx_bytes_total += tap_stats->q_ibytes[i];
767 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
768 ierrors += pmd->rxq[i].stats.ierrors;
771 /* tx queue statistics */
772 imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
773 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
775 for (i = 0; i < imax; i++) {
776 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
777 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
778 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
779 tx_total += tap_stats->q_opackets[i];
780 tx_err_total += tap_stats->q_errors[i];
781 tx_bytes_total += tap_stats->q_obytes[i];
784 tap_stats->ipackets = rx_total;
785 tap_stats->ibytes = rx_bytes_total;
786 tap_stats->ierrors = ierrors;
787 tap_stats->rx_nombuf = rx_nombuf;
788 tap_stats->opackets = tx_total;
789 tap_stats->oerrors = tx_err_total;
790 tap_stats->obytes = tx_bytes_total;
795 tap_stats_reset(struct rte_eth_dev *dev)
798 struct pmd_internals *pmd = dev->data->dev_private;
800 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
801 pmd->rxq[i].stats.ipackets = 0;
802 pmd->rxq[i].stats.ibytes = 0;
803 pmd->rxq[i].stats.ierrors = 0;
804 pmd->rxq[i].stats.rx_nombuf = 0;
806 pmd->txq[i].stats.opackets = 0;
807 pmd->txq[i].stats.errs = 0;
808 pmd->txq[i].stats.obytes = 0;
813 tap_dev_close(struct rte_eth_dev *dev)
816 struct pmd_internals *internals = dev->data->dev_private;
818 tap_link_set_down(dev);
819 tap_flow_flush(dev, NULL);
820 tap_flow_implicit_flush(internals, NULL);
822 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
823 if (internals->rxq[i].fd != -1) {
824 close(internals->rxq[i].fd);
825 internals->rxq[i].fd = -1;
827 if (internals->txq[i].fd != -1) {
828 close(internals->txq[i].fd);
829 internals->txq[i].fd = -1;
833 if (internals->remote_if_index) {
834 /* Restore initial remote state */
835 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
836 &internals->remote_initial_flags);
841 tap_rx_queue_release(void *queue)
843 struct rx_queue *rxq = queue;
845 if (rxq && (rxq->fd > 0)) {
848 rte_pktmbuf_free(rxq->pool);
849 rte_free(rxq->iovecs);
856 tap_tx_queue_release(void *queue)
858 struct tx_queue *txq = queue;
860 if (txq && (txq->fd > 0)) {
867 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
869 struct rte_eth_link *dev_link = &dev->data->dev_link;
870 struct pmd_internals *pmd = dev->data->dev_private;
871 struct ifreq ifr = { .ifr_flags = 0 };
873 if (pmd->remote_if_index) {
874 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
875 if (!(ifr.ifr_flags & IFF_UP) ||
876 !(ifr.ifr_flags & IFF_RUNNING)) {
877 dev_link->link_status = ETH_LINK_DOWN;
881 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
882 dev_link->link_status =
883 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
890 tap_promisc_enable(struct rte_eth_dev *dev)
892 struct pmd_internals *pmd = dev->data->dev_private;
893 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
895 dev->data->promiscuous = 1;
896 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
897 if (pmd->remote_if_index && !pmd->flow_isolate)
898 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
902 tap_promisc_disable(struct rte_eth_dev *dev)
904 struct pmd_internals *pmd = dev->data->dev_private;
905 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
907 dev->data->promiscuous = 0;
908 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
909 if (pmd->remote_if_index && !pmd->flow_isolate)
910 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
914 tap_allmulti_enable(struct rte_eth_dev *dev)
916 struct pmd_internals *pmd = dev->data->dev_private;
917 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
919 dev->data->all_multicast = 1;
920 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
921 if (pmd->remote_if_index && !pmd->flow_isolate)
922 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
926 tap_allmulti_disable(struct rte_eth_dev *dev)
928 struct pmd_internals *pmd = dev->data->dev_private;
929 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
931 dev->data->all_multicast = 0;
932 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
933 if (pmd->remote_if_index && !pmd->flow_isolate)
934 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
938 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
940 struct pmd_internals *pmd = dev->data->dev_private;
941 enum ioctl_mode mode = LOCAL_ONLY;
944 if (is_zero_ether_addr(mac_addr)) {
945 RTE_LOG(ERR, PMD, "%s: can't set an empty MAC address\n",
949 /* Check the actual current MAC address on the tap netdevice */
950 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
952 if (is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
955 /* Check the current MAC address on the remote */
956 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0)
958 if (!is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
960 mode = LOCAL_AND_REMOTE;
961 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
962 rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
963 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode) < 0)
965 rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
966 if (pmd->remote_if_index && !pmd->flow_isolate) {
967 /* Replace MAC redirection rule after a MAC change */
968 if (tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC) < 0) {
970 "%s: Couldn't delete MAC redirection rule\n",
974 if (tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0)
976 "%s: Couldn't add MAC redirection rule\n",
982 tap_setup_queue(struct rte_eth_dev *dev,
983 struct pmd_internals *internals,
990 struct pmd_internals *pmd = dev->data->dev_private;
991 struct rx_queue *rx = &internals->rxq[qid];
992 struct tx_queue *tx = &internals->txq[qid];
1004 /* fd for this queue already exists */
1005 RTE_LOG(DEBUG, PMD, "%s: fd %d for %s queue qid %d exists\n",
1006 pmd->name, *fd, dir, qid);
1007 } else if (*other_fd != -1) {
1008 /* Only other_fd exists. dup it */
1009 *fd = dup(*other_fd);
1012 RTE_LOG(ERR, PMD, "%s: dup() failed.\n",
1016 RTE_LOG(DEBUG, PMD, "%s: dup fd %d for %s queue qid %d (%d)\n",
1017 pmd->name, *other_fd, dir, qid, *fd);
1019 /* Both RX and TX fds do not exist (equal -1). Create fd */
1020 *fd = tun_alloc(pmd);
1022 *fd = -1; /* restore original value */
1023 RTE_LOG(ERR, PMD, "%s: tun_alloc() failed.\n",
1027 RTE_LOG(DEBUG, PMD, "%s: add %s queue for qid %d fd %d\n",
1028 pmd->name, dir, qid, *fd);
1031 tx->mtu = &dev->data->mtu;
1032 rx->rxmode = &dev->data->dev_conf.rxmode;
1038 tap_rx_queue_setup(struct rte_eth_dev *dev,
1039 uint16_t rx_queue_id,
1040 uint16_t nb_rx_desc,
1041 unsigned int socket_id,
1042 const struct rte_eth_rxconf *rx_conf __rte_unused,
1043 struct rte_mempool *mp)
1045 struct pmd_internals *internals = dev->data->dev_private;
1046 struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1047 struct rte_mbuf **tmp = &rxq->pool;
1048 long iov_max = sysconf(_SC_IOV_MAX);
1049 uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1050 struct iovec (*iovecs)[nb_desc + 1];
1051 int data_off = RTE_PKTMBUF_HEADROOM;
1056 if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1057 RTE_LOG(WARNING, PMD,
1058 "nb_rx_queues %d too small or mempool NULL\n",
1059 dev->data->nb_rx_queues);
1063 /* Verify application offloads are valid for our port and queue. */
1064 if (!tap_rxq_are_offloads_valid(dev, rx_conf->offloads)) {
1065 rte_errno = ENOTSUP;
1067 "%p: Rx queue offloads 0x%" PRIx64
1068 " don't match port offloads 0x%" PRIx64
1069 " or supported offloads 0x%" PRIx64 "\n",
1070 (void *)dev, rx_conf->offloads,
1071 dev->data->dev_conf.rxmode.offloads,
1072 (tap_rx_offload_get_port_capa() |
1073 tap_rx_offload_get_queue_capa()));
1077 rxq->trigger_seen = 1; /* force initial burst */
1078 rxq->in_port = dev->data->port_id;
1079 rxq->nb_rx_desc = nb_desc;
1080 iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1083 RTE_LOG(WARNING, PMD,
1084 "%s: Couldn't allocate %d RX descriptors\n",
1085 dev->device->name, nb_desc);
1088 rxq->iovecs = iovecs;
1090 dev->data->rx_queues[rx_queue_id] = rxq;
1091 fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1097 (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1098 (*rxq->iovecs)[0].iov_base = &rxq->pi;
1100 for (i = 1; i <= nb_desc; i++) {
1101 *tmp = rte_pktmbuf_alloc(rxq->mp);
1103 RTE_LOG(WARNING, PMD,
1104 "%s: couldn't allocate memory for queue %d\n",
1105 dev->device->name, rx_queue_id);
1109 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1110 (*rxq->iovecs)[i].iov_base =
1111 (char *)(*tmp)->buf_addr + data_off;
1113 tmp = &(*tmp)->next;
1116 RTE_LOG(DEBUG, PMD, " RX TAP device name %s, qid %d on fd %d\n",
1117 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
1122 rte_pktmbuf_free(rxq->pool);
1124 rte_free(rxq->iovecs);
1130 tap_tx_queue_setup(struct rte_eth_dev *dev,
1131 uint16_t tx_queue_id,
1132 uint16_t nb_tx_desc __rte_unused,
1133 unsigned int socket_id __rte_unused,
1134 const struct rte_eth_txconf *tx_conf)
1136 struct pmd_internals *internals = dev->data->dev_private;
1137 struct tx_queue *txq;
1140 if (tx_queue_id >= dev->data->nb_tx_queues)
1142 dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1143 txq = dev->data->tx_queues[tx_queue_id];
1145 * Don't verify port offloads for application which
1148 if (tx_conf != NULL &&
1149 !!(tx_conf->txq_flags & ETH_TXQ_FLAGS_IGNORE)) {
1150 if (tap_txq_are_offloads_valid(dev, tx_conf->offloads)) {
1151 txq->csum = !!(tx_conf->offloads &
1152 (DEV_TX_OFFLOAD_IPV4_CKSUM |
1153 DEV_TX_OFFLOAD_UDP_CKSUM |
1154 DEV_TX_OFFLOAD_TCP_CKSUM));
1156 rte_errno = ENOTSUP;
1158 "%p: Tx queue offloads 0x%" PRIx64
1159 " don't match port offloads 0x%" PRIx64
1160 " or supported offloads 0x%" PRIx64,
1161 (void *)dev, tx_conf->offloads,
1162 dev->data->dev_conf.txmode.offloads,
1163 tap_tx_offload_get_port_capa());
1167 ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1171 " TX TAP device name %s, qid %d on fd %d csum %s\n",
1172 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd,
1173 txq->csum ? "on" : "off");
1179 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1181 struct pmd_internals *pmd = dev->data->dev_private;
1182 struct ifreq ifr = { .ifr_mtu = mtu };
1185 err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1187 dev->data->mtu = mtu;
1193 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1194 struct ether_addr *mc_addr_set __rte_unused,
1195 uint32_t nb_mc_addr __rte_unused)
1198 * Nothing to do actually: the tap has no filtering whatsoever, every
1199 * packet is received.
1205 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1207 struct rte_eth_dev *dev = arg;
1208 struct pmd_internals *pmd = dev->data->dev_private;
1209 struct ifinfomsg *info = NLMSG_DATA(nh);
1211 if (nh->nlmsg_type != RTM_NEWLINK ||
1212 (info->ifi_index != pmd->if_index &&
1213 info->ifi_index != pmd->remote_if_index))
1215 return tap_link_update(dev, 0);
1219 tap_dev_intr_handler(void *cb_arg)
1221 struct rte_eth_dev *dev = cb_arg;
1222 struct pmd_internals *pmd = dev->data->dev_private;
1224 tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1228 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1230 struct pmd_internals *pmd = dev->data->dev_private;
1232 /* In any case, disable interrupt if the conf is no longer there. */
1233 if (!dev->data->dev_conf.intr_conf.lsc) {
1234 if (pmd->intr_handle.fd != -1) {
1235 tap_nl_final(pmd->intr_handle.fd);
1236 rte_intr_callback_unregister(&pmd->intr_handle,
1237 tap_dev_intr_handler, dev);
1242 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1243 if (unlikely(pmd->intr_handle.fd == -1))
1245 return rte_intr_callback_register(
1246 &pmd->intr_handle, tap_dev_intr_handler, dev);
1248 tap_nl_final(pmd->intr_handle.fd);
1249 return rte_intr_callback_unregister(&pmd->intr_handle,
1250 tap_dev_intr_handler, dev);
1254 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1258 err = tap_lsc_intr_handle_set(dev, set);
1261 err = tap_rx_intr_vec_set(dev, set);
1263 tap_lsc_intr_handle_set(dev, 0);
1267 static const uint32_t*
1268 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1270 static const uint32_t ptypes[] = {
1271 RTE_PTYPE_INNER_L2_ETHER,
1272 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1273 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1274 RTE_PTYPE_INNER_L3_IPV4,
1275 RTE_PTYPE_INNER_L3_IPV4_EXT,
1276 RTE_PTYPE_INNER_L3_IPV6,
1277 RTE_PTYPE_INNER_L3_IPV6_EXT,
1278 RTE_PTYPE_INNER_L4_FRAG,
1279 RTE_PTYPE_INNER_L4_UDP,
1280 RTE_PTYPE_INNER_L4_TCP,
1281 RTE_PTYPE_INNER_L4_SCTP,
1283 RTE_PTYPE_L2_ETHER_VLAN,
1284 RTE_PTYPE_L2_ETHER_QINQ,
1286 RTE_PTYPE_L3_IPV4_EXT,
1287 RTE_PTYPE_L3_IPV6_EXT,
1299 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1300 struct rte_eth_fc_conf *fc_conf)
1302 fc_conf->mode = RTE_FC_NONE;
1307 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1308 struct rte_eth_fc_conf *fc_conf)
1310 if (fc_conf->mode != RTE_FC_NONE)
1315 static const struct eth_dev_ops ops = {
1316 .dev_start = tap_dev_start,
1317 .dev_stop = tap_dev_stop,
1318 .dev_close = tap_dev_close,
1319 .dev_configure = tap_dev_configure,
1320 .dev_infos_get = tap_dev_info,
1321 .rx_queue_setup = tap_rx_queue_setup,
1322 .tx_queue_setup = tap_tx_queue_setup,
1323 .rx_queue_release = tap_rx_queue_release,
1324 .tx_queue_release = tap_tx_queue_release,
1325 .flow_ctrl_get = tap_flow_ctrl_get,
1326 .flow_ctrl_set = tap_flow_ctrl_set,
1327 .link_update = tap_link_update,
1328 .dev_set_link_up = tap_link_set_up,
1329 .dev_set_link_down = tap_link_set_down,
1330 .promiscuous_enable = tap_promisc_enable,
1331 .promiscuous_disable = tap_promisc_disable,
1332 .allmulticast_enable = tap_allmulti_enable,
1333 .allmulticast_disable = tap_allmulti_disable,
1334 .mac_addr_set = tap_mac_set,
1335 .mtu_set = tap_mtu_set,
1336 .set_mc_addr_list = tap_set_mc_addr_list,
1337 .stats_get = tap_stats_get,
1338 .stats_reset = tap_stats_reset,
1339 .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1340 .filter_ctrl = tap_dev_filter_ctrl,
1344 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
1345 char *remote_iface, struct ether_addr *mac_addr)
1347 int numa_node = rte_socket_id();
1348 struct rte_eth_dev *dev;
1349 struct pmd_internals *pmd;
1350 struct rte_eth_dev_data *data;
1354 RTE_LOG(DEBUG, PMD, " TAP device on numa %u\n", rte_socket_id());
1356 data = rte_zmalloc_socket(tap_name, sizeof(*data), 0, numa_node);
1358 RTE_LOG(ERR, PMD, "TAP Failed to allocate data\n");
1359 goto error_exit_nodev;
1362 dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1364 RTE_LOG(ERR, PMD, "TAP Unable to allocate device struct\n");
1365 goto error_exit_nodev;
1368 pmd = dev->data->dev_private;
1370 snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1372 pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1373 if (pmd->ioctl_sock == -1) {
1375 "TAP Unable to get a socket for management: %s\n",
1380 /* Setup some default values */
1381 rte_memcpy(data, dev->data, sizeof(*data));
1382 data->dev_private = pmd;
1383 data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1384 data->numa_node = numa_node;
1386 data->dev_link = pmd_link;
1387 data->mac_addrs = &pmd->eth_addr;
1388 /* Set the number of RX and TX queues */
1389 data->nb_rx_queues = 0;
1390 data->nb_tx_queues = 0;
1393 dev->dev_ops = &ops;
1394 dev->rx_pkt_burst = pmd_rx_burst;
1395 dev->tx_pkt_burst = pmd_tx_burst;
1397 pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1398 pmd->intr_handle.fd = -1;
1399 dev->intr_handle = &pmd->intr_handle;
1401 /* Presetup the fds to -1 as being not valid */
1402 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1403 pmd->rxq[i].fd = -1;
1404 pmd->txq[i].fd = -1;
1407 if (is_zero_ether_addr(mac_addr))
1408 eth_random_addr((uint8_t *)&pmd->eth_addr);
1410 rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(mac_addr));
1412 /* Immediately create the netdevice (this will create the 1st queue). */
1414 if (tap_setup_queue(dev, pmd, 0, 1) == -1)
1417 if (tap_setup_queue(dev, pmd, 0, 0) == -1)
1420 ifr.ifr_mtu = dev->data->mtu;
1421 if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1424 memset(&ifr, 0, sizeof(struct ifreq));
1425 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1426 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr, ETHER_ADDR_LEN);
1427 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1431 * Set up everything related to rte_flow:
1433 * - tap / remote if_index
1434 * - mandatory QDISCs
1435 * - rte_flow actual/implicit lists
1438 pmd->nlsk_fd = tap_nl_init(0);
1439 if (pmd->nlsk_fd == -1) {
1440 RTE_LOG(WARNING, PMD, "%s: failed to create netlink socket.\n",
1442 goto disable_rte_flow;
1444 pmd->if_index = if_nametoindex(pmd->name);
1445 if (!pmd->if_index) {
1446 RTE_LOG(ERR, PMD, "%s: failed to get if_index.\n", pmd->name);
1447 goto disable_rte_flow;
1449 if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1450 RTE_LOG(ERR, PMD, "%s: failed to create multiq qdisc.\n",
1452 goto disable_rte_flow;
1454 if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1455 RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1457 goto disable_rte_flow;
1459 LIST_INIT(&pmd->flows);
1461 if (strlen(remote_iface)) {
1462 pmd->remote_if_index = if_nametoindex(remote_iface);
1463 if (!pmd->remote_if_index) {
1464 RTE_LOG(ERR, PMD, "%s: failed to get %s if_index.\n",
1465 pmd->name, remote_iface);
1468 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1469 "%s", remote_iface);
1471 /* Save state of remote device */
1472 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1474 /* Replicate remote MAC address */
1475 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1476 RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1477 pmd->name, pmd->remote_iface);
1480 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1482 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1483 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1484 RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1485 pmd->name, remote_iface);
1490 * Flush usually returns negative value because it tries to
1491 * delete every QDISC (and on a running device, one QDISC at
1492 * least is needed). Ignore negative return value.
1494 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1495 if (qdisc_create_ingress(pmd->nlsk_fd,
1496 pmd->remote_if_index) < 0) {
1497 RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1501 LIST_INIT(&pmd->implicit_flows);
1502 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1503 tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1504 tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1505 tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1507 "%s: failed to create implicit rules.\n",
1516 RTE_LOG(ERR, PMD, " Disabling rte flow support: %s(%d)\n",
1517 strerror(errno), errno);
1518 if (strlen(remote_iface)) {
1519 RTE_LOG(ERR, PMD, "Remote feature requires flow support.\n");
1525 RTE_LOG(ERR, PMD, " Can't set up remote feature: %s(%d)\n",
1526 strerror(errno), errno);
1527 tap_flow_implicit_flush(pmd, NULL);
1530 if (pmd->ioctl_sock > 0)
1531 close(pmd->ioctl_sock);
1532 rte_eth_dev_release_port(dev);
1535 RTE_LOG(ERR, PMD, "TAP Unable to initialize %s\n",
1536 rte_vdev_device_name(vdev));
1543 set_interface_name(const char *key __rte_unused,
1547 char *name = (char *)extra_args;
1550 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN - 1);
1552 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1553 DEFAULT_TAP_NAME, (tap_unit - 1));
1559 set_remote_iface(const char *key __rte_unused,
1563 char *name = (char *)extra_args;
1566 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
1571 static int parse_user_mac(struct ether_addr *user_mac,
1574 unsigned int index = 0;
1575 char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
1577 if (user_mac == NULL || value == NULL)
1580 strlcpy(mac_temp, value, sizeof(mac_temp));
1581 mac_byte = strtok(mac_temp, ":");
1583 while ((mac_byte != NULL) &&
1584 (strlen(mac_byte) <= 2) &&
1585 (strlen(mac_byte) == strspn(mac_byte,
1586 ETH_TAP_CMP_MAC_FMT))) {
1587 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
1588 mac_byte = strtok(NULL, ":");
1595 set_mac_type(const char *key __rte_unused,
1599 struct ether_addr *user_mac = extra_args;
1604 if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
1605 static int iface_idx;
1607 /* fixed mac = 00:64:74:61:70:<iface_idx> */
1608 memcpy((char *)user_mac->addr_bytes, "\0dtap", ETHER_ADDR_LEN);
1609 user_mac->addr_bytes[ETHER_ADDR_LEN - 1] = iface_idx++ + '0';
1613 if (parse_user_mac(user_mac, value) != 6)
1616 RTE_LOG(DEBUG, PMD, "TAP user MAC param (%s)\n", value);
1620 RTE_LOG(ERR, PMD, "TAP user MAC (%s) is not in format (%s|%s)\n",
1621 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
1625 /* Open a TAP interface device.
1628 rte_pmd_tap_probe(struct rte_vdev_device *dev)
1630 const char *name, *params;
1632 struct rte_kvargs *kvlist = NULL;
1634 char tap_name[RTE_ETH_NAME_MAX_LEN];
1635 char remote_iface[RTE_ETH_NAME_MAX_LEN];
1636 struct ether_addr user_mac = { .addr_bytes = {0} };
1638 name = rte_vdev_device_name(dev);
1639 params = rte_vdev_device_args(dev);
1641 speed = ETH_SPEED_NUM_10G;
1642 snprintf(tap_name, sizeof(tap_name), "%s%d",
1643 DEFAULT_TAP_NAME, tap_unit++);
1644 memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1646 if (params && (params[0] != '\0')) {
1647 RTE_LOG(DEBUG, PMD, "parameters (%s)\n", params);
1649 kvlist = rte_kvargs_parse(params, valid_arguments);
1651 if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1652 ret = rte_kvargs_process(kvlist,
1654 &set_interface_name,
1660 if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
1661 ret = rte_kvargs_process(kvlist,
1669 if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
1670 ret = rte_kvargs_process(kvlist,
1679 pmd_link.link_speed = speed;
1681 RTE_LOG(NOTICE, PMD, "Initializing pmd_tap for %s as %s\n",
1684 ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac);
1688 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
1690 tap_unit--; /* Restore the unit number */
1692 rte_kvargs_free(kvlist);
1697 /* detach a TAP device.
1700 rte_pmd_tap_remove(struct rte_vdev_device *dev)
1702 struct rte_eth_dev *eth_dev = NULL;
1703 struct pmd_internals *internals;
1706 RTE_LOG(DEBUG, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
1709 /* find the ethdev entry */
1710 eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1714 internals = eth_dev->data->dev_private;
1715 if (internals->nlsk_fd) {
1716 tap_flow_flush(eth_dev, NULL);
1717 tap_flow_implicit_flush(internals, NULL);
1718 tap_nl_final(internals->nlsk_fd);
1720 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1721 if (internals->rxq[i].fd != -1) {
1722 close(internals->rxq[i].fd);
1723 internals->rxq[i].fd = -1;
1725 if (internals->txq[i].fd != -1) {
1726 close(internals->txq[i].fd);
1727 internals->txq[i].fd = -1;
1731 close(internals->ioctl_sock);
1732 rte_free(eth_dev->data->dev_private);
1733 rte_free(eth_dev->data);
1735 rte_eth_dev_release_port(eth_dev);
1740 static struct rte_vdev_driver pmd_tap_drv = {
1741 .probe = rte_pmd_tap_probe,
1742 .remove = rte_pmd_tap_remove,
1744 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1745 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1746 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
1747 ETH_TAP_IFACE_ARG "=<string> "
1748 ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
1749 ETH_TAP_REMOTE_ARG "=<string>");