1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2016-2017 Intel Corporation
5 #include <rte_atomic.h>
6 #include <rte_branch_prediction.h>
7 #include <rte_byteorder.h>
8 #include <rte_common.h>
10 #include <rte_ethdev_driver.h>
11 #include <rte_ethdev_vdev.h>
12 #include <rte_malloc.h>
13 #include <rte_bus_vdev.h>
14 #include <rte_kvargs.h>
16 #include <rte_debug.h>
18 #include <rte_string_fns.h>
20 #include <sys/types.h>
22 #include <sys/socket.h>
23 #include <sys/ioctl.h>
24 #include <sys/utsname.h>
32 #include <arpa/inet.h>
34 #include <linux/if_tun.h>
35 #include <linux/if_ether.h>
39 #include <rte_eth_tap.h>
41 #include <tap_netlink.h>
42 #include <tap_tcmsgs.h>
44 /* Linux based path to the TUN device */
45 #define TUN_TAP_DEV_PATH "/dev/net/tun"
46 #define DEFAULT_TAP_NAME "dtap"
47 #define DEFAULT_TUN_NAME "dtun"
49 #define ETH_TAP_IFACE_ARG "iface"
50 #define ETH_TAP_REMOTE_ARG "remote"
51 #define ETH_TAP_MAC_ARG "mac"
52 #define ETH_TAP_MAC_FIXED "fixed"
54 #define ETH_TAP_USR_MAC_FMT "xx:xx:xx:xx:xx:xx"
55 #define ETH_TAP_CMP_MAC_FMT "0123456789ABCDEFabcdef"
56 #define ETH_TAP_MAC_ARG_FMT ETH_TAP_MAC_FIXED "|" ETH_TAP_USR_MAC_FMT
58 static struct rte_vdev_driver pmd_tap_drv;
59 static struct rte_vdev_driver pmd_tun_drv;
61 static const char *valid_arguments[] = {
72 static char tuntap_name[8];
74 static volatile uint32_t tap_trigger; /* Rx trigger */
76 static struct rte_eth_link pmd_link = {
77 .link_speed = ETH_SPEED_NUM_10G,
78 .link_duplex = ETH_LINK_FULL_DUPLEX,
79 .link_status = ETH_LINK_DOWN,
80 .link_autoneg = ETH_LINK_FIXED,
84 tap_trigger_cb(int sig __rte_unused)
86 /* Valid trigger values are nonzero */
87 tap_trigger = (tap_trigger + 1) | 0x80000000;
90 /* Specifies on what netdevices the ioctl should be applied */
97 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
99 /* Tun/Tap allocation routine
101 * name is the number of the interface to use, unless NULL to take the host
105 tun_alloc(struct pmd_internals *pmd)
108 #ifdef IFF_MULTI_QUEUE
109 unsigned int features;
113 memset(&ifr, 0, sizeof(struct ifreq));
116 * Do not set IFF_NO_PI as packet information header will be needed
117 * to check if a received packet has been truncated.
119 ifr.ifr_flags = (tap_type) ? IFF_TAP : IFF_TUN | IFF_POINTOPOINT;
120 snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
122 TAP_LOG(DEBUG, "ifr_name '%s'", ifr.ifr_name);
124 fd = open(TUN_TAP_DEV_PATH, O_RDWR);
126 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
130 #ifdef IFF_MULTI_QUEUE
131 /* Grab the TUN features to verify we can work multi-queue */
132 if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
133 TAP_LOG(ERR, "%s unable to get TUN/TAP features",
137 TAP_LOG(DEBUG, "%s Features %08x", tuntap_name, features);
139 if (features & IFF_MULTI_QUEUE) {
140 TAP_LOG(DEBUG, " Multi-queue support for %d queues",
141 RTE_PMD_TAP_MAX_QUEUES);
142 ifr.ifr_flags |= IFF_MULTI_QUEUE;
146 ifr.ifr_flags |= IFF_ONE_QUEUE;
147 TAP_LOG(DEBUG, " Single queue only support");
150 /* Set the TUN/TAP configuration and set the name if needed */
151 if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
152 TAP_LOG(WARNING, "Unable to set TUNSETIFF for %s: %s",
153 ifr.ifr_name, strerror(errno));
157 /* Always set the file descriptor to non-blocking */
158 if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
160 "Unable to set %s to nonblocking: %s",
161 ifr.ifr_name, strerror(errno));
165 /* Set up trigger to optimize empty Rx bursts */
169 int flags = fcntl(fd, F_GETFL);
171 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
173 if (sa.sa_handler != tap_trigger_cb) {
175 * Make sure SIGIO is not already taken. This is done
176 * as late as possible to leave the application a
177 * chance to set up its own signal handler first.
179 if (sa.sa_handler != SIG_IGN &&
180 sa.sa_handler != SIG_DFL) {
184 sa = (struct sigaction){
185 .sa_flags = SA_RESTART,
186 .sa_handler = tap_trigger_cb,
188 if (sigaction(SIGIO, &sa, NULL) == -1)
191 /* Enable SIGIO on file descriptor */
192 fcntl(fd, F_SETFL, flags | O_ASYNC);
193 fcntl(fd, F_SETOWN, getpid());
197 /* Disable trigger globally in case of error */
199 TAP_LOG(WARNING, "Rx trigger disabled: %s",
212 tap_verify_csum(struct rte_mbuf *mbuf)
214 uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
215 uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
216 uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
217 unsigned int l2_len = sizeof(struct ether_hdr);
223 if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
225 else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
227 /* Don't verify checksum for packets with discontinuous L2 header */
228 if (unlikely(l2_len + sizeof(struct ipv4_hdr) >
229 rte_pktmbuf_data_len(mbuf)))
231 l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
232 if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
233 struct ipv4_hdr *iph = l3_hdr;
235 /* ihl contains the number of 4-byte words in the header */
236 l3_len = 4 * (iph->version_ihl & 0xf);
237 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
240 cksum = ~rte_raw_cksum(iph, l3_len);
241 mbuf->ol_flags |= cksum ?
242 PKT_RX_IP_CKSUM_BAD :
243 PKT_RX_IP_CKSUM_GOOD;
244 } else if (l3 == RTE_PTYPE_L3_IPV6) {
245 l3_len = sizeof(struct ipv6_hdr);
247 /* IPv6 extensions are not supported */
250 if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
251 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
252 /* Don't verify checksum for multi-segment packets. */
253 if (mbuf->nb_segs > 1)
255 if (l3 == RTE_PTYPE_L3_IPV4)
256 cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
257 else if (l3 == RTE_PTYPE_L3_IPV6)
258 cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
259 mbuf->ol_flags |= cksum ?
260 PKT_RX_L4_CKSUM_BAD :
261 PKT_RX_L4_CKSUM_GOOD;
266 tap_rx_offload_get_port_capa(void)
269 * No specific port Rx offload capabilities.
275 tap_rx_offload_get_queue_capa(void)
277 return DEV_RX_OFFLOAD_SCATTER |
278 DEV_RX_OFFLOAD_IPV4_CKSUM |
279 DEV_RX_OFFLOAD_UDP_CKSUM |
280 DEV_RX_OFFLOAD_TCP_CKSUM |
281 DEV_RX_OFFLOAD_CRC_STRIP;
284 /* Callback to handle the rx burst of packets to the correct interface and
285 * file descriptor(s) in a multi-queue setup.
288 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
290 struct rx_queue *rxq = queue;
292 unsigned long num_rx_bytes = 0;
293 uint32_t trigger = tap_trigger;
295 if (trigger == rxq->trigger_seen)
298 rxq->trigger_seen = trigger;
299 rte_compiler_barrier();
300 for (num_rx = 0; num_rx < nb_pkts; ) {
301 struct rte_mbuf *mbuf = rxq->pool;
302 struct rte_mbuf *seg = NULL;
303 struct rte_mbuf *new_tail = NULL;
304 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
307 len = readv(rxq->fd, *rxq->iovecs,
309 (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
310 rxq->nb_rx_desc : 1));
311 if (len < (int)sizeof(struct tun_pi))
314 /* Packet couldn't fit in the provided mbuf */
315 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
316 rxq->stats.ierrors++;
320 len -= sizeof(struct tun_pi);
323 mbuf->port = rxq->in_port;
325 struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
327 if (unlikely(!buf)) {
328 rxq->stats.rx_nombuf++;
329 /* No new buf has been allocated: do nothing */
330 if (!new_tail || !seg)
334 rte_pktmbuf_free(mbuf);
338 seg = seg ? seg->next : mbuf;
339 if (rxq->pool == mbuf)
342 new_tail->next = buf;
344 new_tail->next = seg->next;
346 /* iovecs[0] is reserved for packet info (pi) */
347 (*rxq->iovecs)[mbuf->nb_segs].iov_len =
348 buf->buf_len - data_off;
349 (*rxq->iovecs)[mbuf->nb_segs].iov_base =
350 (char *)buf->buf_addr + data_off;
352 seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
353 seg->data_off = data_off;
355 len -= seg->data_len;
359 /* First segment has headroom, not the others */
363 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
365 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
366 tap_verify_csum(mbuf);
368 /* account for the receive frame */
369 bufs[num_rx++] = mbuf;
370 num_rx_bytes += mbuf->pkt_len;
373 rxq->stats.ipackets += num_rx;
374 rxq->stats.ibytes += num_rx_bytes;
380 tap_tx_offload_get_port_capa(void)
383 * No specific port Tx offload capabilities.
389 tap_tx_offload_get_queue_capa(void)
391 return DEV_TX_OFFLOAD_MULTI_SEGS |
392 DEV_TX_OFFLOAD_IPV4_CKSUM |
393 DEV_TX_OFFLOAD_UDP_CKSUM |
394 DEV_TX_OFFLOAD_TCP_CKSUM;
398 tap_tx_offload(char *packet, uint64_t ol_flags, unsigned int l2_len,
401 void *l3_hdr = packet + l2_len;
403 if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
404 struct ipv4_hdr *iph = l3_hdr;
407 iph->hdr_checksum = 0;
408 cksum = rte_raw_cksum(iph, l3_len);
409 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
411 if (ol_flags & PKT_TX_L4_MASK) {
417 l4_hdr = packet + l2_len + l3_len;
418 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
419 l4_cksum = &((struct udp_hdr *)l4_hdr)->dgram_cksum;
420 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
421 l4_cksum = &((struct tcp_hdr *)l4_hdr)->cksum;
425 if (ol_flags & PKT_TX_IPV4) {
426 struct ipv4_hdr *iph = l3_hdr;
428 l4_len = rte_be_to_cpu_16(iph->total_length) - l3_len;
429 cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
431 struct ipv6_hdr *ip6h = l3_hdr;
433 /* payload_len does not include ext headers */
434 l4_len = rte_be_to_cpu_16(ip6h->payload_len) -
435 l3_len + sizeof(struct ipv6_hdr);
436 cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
438 cksum += rte_raw_cksum(l4_hdr, l4_len);
439 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
440 cksum = (~cksum) & 0xffff;
447 /* Callback to handle sending packets from the tap interface
450 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
452 struct tx_queue *txq = queue;
454 unsigned long num_tx_bytes = 0;
458 if (unlikely(nb_pkts == 0))
461 max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
462 for (i = 0; i < nb_pkts; i++) {
463 struct rte_mbuf *mbuf = bufs[num_tx];
464 struct iovec iovecs[mbuf->nb_segs + 1];
465 struct tun_pi pi = { .flags = 0, .proto = 0x00 };
466 struct rte_mbuf *seg = mbuf;
467 char m_copy[mbuf->data_len];
471 /* stats.errs will be incremented */
472 if (rte_pktmbuf_pkt_len(mbuf) > max_size)
476 * TUN and TAP are created with IFF_NO_PI disabled.
477 * For TUN PMD this mandatory as fields are used by
478 * Kernel tun.c to determine whether its IP or non IP
481 * The logic fetches the first byte of data from mbuf.
482 * compares whether its v4 or v6. If none matches default
483 * value 0x00 is taken for protocol field.
485 char *buff_data = rte_pktmbuf_mtod(seg, void *);
486 j = (*buff_data & 0xf0);
487 pi.proto = (j == 0x40) ? 0x0008 :
488 (j == 0x60) ? 0xdd86 : 0x00;
490 iovecs[0].iov_base = π
491 iovecs[0].iov_len = sizeof(pi);
492 for (j = 1; j <= mbuf->nb_segs; j++) {
493 iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
495 rte_pktmbuf_mtod(seg, void *);
499 ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
500 (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
501 (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
502 /* Support only packets with all data in the same seg */
503 if (mbuf->nb_segs > 1)
505 /* To change checksums, work on a copy of data. */
506 rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
507 rte_pktmbuf_data_len(mbuf));
508 tap_tx_offload(m_copy, mbuf->ol_flags,
509 mbuf->l2_len, mbuf->l3_len);
510 iovecs[1].iov_base = m_copy;
512 /* copy the tx frame data */
513 n = writev(txq->fd, iovecs, mbuf->nb_segs + 1);
518 num_tx_bytes += mbuf->pkt_len;
519 rte_pktmbuf_free(mbuf);
522 txq->stats.opackets += num_tx;
523 txq->stats.errs += nb_pkts - num_tx;
524 txq->stats.obytes += num_tx_bytes;
530 tap_ioctl_req2str(unsigned long request)
534 return "SIOCSIFFLAGS";
536 return "SIOCGIFFLAGS";
538 return "SIOCGIFHWADDR";
540 return "SIOCSIFHWADDR";
548 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
549 struct ifreq *ifr, int set, enum ioctl_mode mode)
551 short req_flags = ifr->ifr_flags;
552 int remote = pmd->remote_if_index &&
553 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
555 if (!pmd->remote_if_index && mode == REMOTE_ONLY)
558 * If there is a remote netdevice, apply ioctl on it, then apply it on
563 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
564 else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
565 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
568 /* fetch current flags to leave other flags untouched */
569 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
572 ifr->ifr_flags |= req_flags;
574 ifr->ifr_flags &= ~req_flags;
582 RTE_LOG(WARNING, PMD, "%s: ioctl() called with wrong arg\n",
586 if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
588 if (remote-- && mode == LOCAL_AND_REMOTE)
593 TAP_LOG(DEBUG, "%s(%s) failed: %s(%d)", ifr->ifr_name,
594 tap_ioctl_req2str(request), strerror(errno), errno);
599 tap_link_set_down(struct rte_eth_dev *dev)
601 struct pmd_internals *pmd = dev->data->dev_private;
602 struct ifreq ifr = { .ifr_flags = IFF_UP };
604 dev->data->dev_link.link_status = ETH_LINK_DOWN;
605 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
609 tap_link_set_up(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_UP;
615 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
619 tap_dev_start(struct rte_eth_dev *dev)
623 err = tap_intr_handle_set(dev, 1);
626 return tap_link_set_up(dev);
629 /* This function gets called when the current port gets stopped.
632 tap_dev_stop(struct rte_eth_dev *dev)
634 tap_intr_handle_set(dev, 0);
635 tap_link_set_down(dev);
639 tap_dev_configure(struct rte_eth_dev *dev)
641 if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
643 "%s: number of rx queues %d exceeds max num of queues %d",
645 dev->data->nb_rx_queues,
646 RTE_PMD_TAP_MAX_QUEUES);
649 if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
651 "%s: number of tx queues %d exceeds max num of queues %d",
653 dev->data->nb_tx_queues,
654 RTE_PMD_TAP_MAX_QUEUES);
658 TAP_LOG(INFO, "%s: %p: TX configured queues number: %u",
659 dev->device->name, (void *)dev, dev->data->nb_tx_queues);
661 TAP_LOG(INFO, "%s: %p: RX configured queues number: %u",
662 dev->device->name, (void *)dev, dev->data->nb_rx_queues);
668 tap_dev_speed_capa(void)
670 uint32_t speed = pmd_link.link_speed;
673 if (speed >= ETH_SPEED_NUM_10M)
674 capa |= ETH_LINK_SPEED_10M;
675 if (speed >= ETH_SPEED_NUM_100M)
676 capa |= ETH_LINK_SPEED_100M;
677 if (speed >= ETH_SPEED_NUM_1G)
678 capa |= ETH_LINK_SPEED_1G;
679 if (speed >= ETH_SPEED_NUM_5G)
680 capa |= ETH_LINK_SPEED_2_5G;
681 if (speed >= ETH_SPEED_NUM_5G)
682 capa |= ETH_LINK_SPEED_5G;
683 if (speed >= ETH_SPEED_NUM_10G)
684 capa |= ETH_LINK_SPEED_10G;
685 if (speed >= ETH_SPEED_NUM_20G)
686 capa |= ETH_LINK_SPEED_20G;
687 if (speed >= ETH_SPEED_NUM_25G)
688 capa |= ETH_LINK_SPEED_25G;
689 if (speed >= ETH_SPEED_NUM_40G)
690 capa |= ETH_LINK_SPEED_40G;
691 if (speed >= ETH_SPEED_NUM_50G)
692 capa |= ETH_LINK_SPEED_50G;
693 if (speed >= ETH_SPEED_NUM_56G)
694 capa |= ETH_LINK_SPEED_56G;
695 if (speed >= ETH_SPEED_NUM_100G)
696 capa |= ETH_LINK_SPEED_100G;
702 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
704 struct pmd_internals *internals = dev->data->dev_private;
706 dev_info->if_index = internals->if_index;
707 dev_info->max_mac_addrs = 1;
708 dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
709 dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
710 dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
711 dev_info->min_rx_bufsize = 0;
712 dev_info->speed_capa = tap_dev_speed_capa();
713 dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
714 dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
715 dev_info->rx_queue_offload_capa;
716 dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
717 dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
718 dev_info->tx_queue_offload_capa;
719 dev_info->hash_key_size = TAP_RSS_HASH_KEY_SIZE;
721 * limitation: TAP suppors all of the following hash
722 * functions together and not in partial combinations
724 dev_info->flow_type_rss_offloads =
725 ETH_RSS_IP | ETH_RSS_UDP | ETH_RSS_TCP;
729 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
731 unsigned int i, imax;
732 unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
733 unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
734 unsigned long rx_nombuf = 0, ierrors = 0;
735 const struct pmd_internals *pmd = dev->data->dev_private;
737 /* rx queue statistics */
738 imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
739 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
740 for (i = 0; i < imax; i++) {
741 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
742 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
743 rx_total += tap_stats->q_ipackets[i];
744 rx_bytes_total += tap_stats->q_ibytes[i];
745 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
746 ierrors += pmd->rxq[i].stats.ierrors;
749 /* tx queue statistics */
750 imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
751 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
753 for (i = 0; i < imax; i++) {
754 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
755 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
756 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
757 tx_total += tap_stats->q_opackets[i];
758 tx_err_total += tap_stats->q_errors[i];
759 tx_bytes_total += tap_stats->q_obytes[i];
762 tap_stats->ipackets = rx_total;
763 tap_stats->ibytes = rx_bytes_total;
764 tap_stats->ierrors = ierrors;
765 tap_stats->rx_nombuf = rx_nombuf;
766 tap_stats->opackets = tx_total;
767 tap_stats->oerrors = tx_err_total;
768 tap_stats->obytes = tx_bytes_total;
773 tap_stats_reset(struct rte_eth_dev *dev)
776 struct pmd_internals *pmd = dev->data->dev_private;
778 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
779 pmd->rxq[i].stats.ipackets = 0;
780 pmd->rxq[i].stats.ibytes = 0;
781 pmd->rxq[i].stats.ierrors = 0;
782 pmd->rxq[i].stats.rx_nombuf = 0;
784 pmd->txq[i].stats.opackets = 0;
785 pmd->txq[i].stats.errs = 0;
786 pmd->txq[i].stats.obytes = 0;
791 tap_dev_close(struct rte_eth_dev *dev)
794 struct pmd_internals *internals = dev->data->dev_private;
796 tap_link_set_down(dev);
797 tap_flow_flush(dev, NULL);
798 tap_flow_implicit_flush(internals, NULL);
800 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
801 if (internals->rxq[i].fd != -1) {
802 close(internals->rxq[i].fd);
803 internals->rxq[i].fd = -1;
805 if (internals->txq[i].fd != -1) {
806 close(internals->txq[i].fd);
807 internals->txq[i].fd = -1;
811 if (internals->remote_if_index) {
812 /* Restore initial remote state */
813 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
814 &internals->remote_initial_flags);
819 tap_rx_queue_release(void *queue)
821 struct rx_queue *rxq = queue;
823 if (rxq && (rxq->fd > 0)) {
826 rte_pktmbuf_free(rxq->pool);
827 rte_free(rxq->iovecs);
834 tap_tx_queue_release(void *queue)
836 struct tx_queue *txq = queue;
838 if (txq && (txq->fd > 0)) {
845 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
847 struct rte_eth_link *dev_link = &dev->data->dev_link;
848 struct pmd_internals *pmd = dev->data->dev_private;
849 struct ifreq ifr = { .ifr_flags = 0 };
851 if (pmd->remote_if_index) {
852 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
853 if (!(ifr.ifr_flags & IFF_UP) ||
854 !(ifr.ifr_flags & IFF_RUNNING)) {
855 dev_link->link_status = ETH_LINK_DOWN;
859 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
860 dev_link->link_status =
861 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
868 tap_promisc_enable(struct rte_eth_dev *dev)
870 struct pmd_internals *pmd = dev->data->dev_private;
871 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
873 dev->data->promiscuous = 1;
874 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
875 if (pmd->remote_if_index && !pmd->flow_isolate)
876 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
880 tap_promisc_disable(struct rte_eth_dev *dev)
882 struct pmd_internals *pmd = dev->data->dev_private;
883 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
885 dev->data->promiscuous = 0;
886 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
887 if (pmd->remote_if_index && !pmd->flow_isolate)
888 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
892 tap_allmulti_enable(struct rte_eth_dev *dev)
894 struct pmd_internals *pmd = dev->data->dev_private;
895 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
897 dev->data->all_multicast = 1;
898 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
899 if (pmd->remote_if_index && !pmd->flow_isolate)
900 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
904 tap_allmulti_disable(struct rte_eth_dev *dev)
906 struct pmd_internals *pmd = dev->data->dev_private;
907 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
909 dev->data->all_multicast = 0;
910 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
911 if (pmd->remote_if_index && !pmd->flow_isolate)
912 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
916 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
918 struct pmd_internals *pmd = dev->data->dev_private;
919 enum ioctl_mode mode = LOCAL_ONLY;
923 if (is_zero_ether_addr(mac_addr)) {
924 TAP_LOG(ERR, "%s: can't set an empty MAC address",
928 /* Check the actual current MAC address on the tap netdevice */
929 ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY);
932 if (is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
935 /* Check the current MAC address on the remote */
936 ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY);
939 if (!is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
941 mode = LOCAL_AND_REMOTE;
942 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
943 rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
944 ret = tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode);
947 rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
948 if (pmd->remote_if_index && !pmd->flow_isolate) {
949 /* Replace MAC redirection rule after a MAC change */
950 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC);
953 "%s: Couldn't delete MAC redirection rule",
957 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC);
960 "%s: Couldn't add MAC redirection rule",
970 tap_setup_queue(struct rte_eth_dev *dev,
971 struct pmd_internals *internals,
978 struct pmd_internals *pmd = dev->data->dev_private;
979 struct rx_queue *rx = &internals->rxq[qid];
980 struct tx_queue *tx = &internals->txq[qid];
992 /* fd for this queue already exists */
993 TAP_LOG(DEBUG, "%s: fd %d for %s queue qid %d exists",
994 pmd->name, *fd, dir, qid);
995 } else if (*other_fd != -1) {
996 /* Only other_fd exists. dup it */
997 *fd = dup(*other_fd);
1000 TAP_LOG(ERR, "%s: dup() failed.", pmd->name);
1003 TAP_LOG(DEBUG, "%s: dup fd %d for %s queue qid %d (%d)",
1004 pmd->name, *other_fd, dir, qid, *fd);
1006 /* Both RX and TX fds do not exist (equal -1). Create fd */
1007 *fd = tun_alloc(pmd);
1009 *fd = -1; /* restore original value */
1010 TAP_LOG(ERR, "%s: tun_alloc() failed.", pmd->name);
1013 TAP_LOG(DEBUG, "%s: add %s queue for qid %d fd %d",
1014 pmd->name, dir, qid, *fd);
1017 tx->mtu = &dev->data->mtu;
1018 rx->rxmode = &dev->data->dev_conf.rxmode;
1024 tap_rx_queue_setup(struct rte_eth_dev *dev,
1025 uint16_t rx_queue_id,
1026 uint16_t nb_rx_desc,
1027 unsigned int socket_id,
1028 const struct rte_eth_rxconf *rx_conf __rte_unused,
1029 struct rte_mempool *mp)
1031 struct pmd_internals *internals = dev->data->dev_private;
1032 struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1033 struct rte_mbuf **tmp = &rxq->pool;
1034 long iov_max = sysconf(_SC_IOV_MAX);
1035 uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1036 struct iovec (*iovecs)[nb_desc + 1];
1037 int data_off = RTE_PKTMBUF_HEADROOM;
1042 if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1044 "nb_rx_queues %d too small or mempool NULL",
1045 dev->data->nb_rx_queues);
1050 rxq->trigger_seen = 1; /* force initial burst */
1051 rxq->in_port = dev->data->port_id;
1052 rxq->nb_rx_desc = nb_desc;
1053 iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1057 "%s: Couldn't allocate %d RX descriptors",
1058 dev->device->name, nb_desc);
1061 rxq->iovecs = iovecs;
1063 dev->data->rx_queues[rx_queue_id] = rxq;
1064 fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1070 (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1071 (*rxq->iovecs)[0].iov_base = &rxq->pi;
1073 for (i = 1; i <= nb_desc; i++) {
1074 *tmp = rte_pktmbuf_alloc(rxq->mp);
1077 "%s: couldn't allocate memory for queue %d",
1078 dev->device->name, rx_queue_id);
1082 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1083 (*rxq->iovecs)[i].iov_base =
1084 (char *)(*tmp)->buf_addr + data_off;
1086 tmp = &(*tmp)->next;
1089 TAP_LOG(DEBUG, " RX TUNTAP device name %s, qid %d on fd %d",
1090 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
1095 rte_pktmbuf_free(rxq->pool);
1097 rte_free(rxq->iovecs);
1103 tap_tx_queue_setup(struct rte_eth_dev *dev,
1104 uint16_t tx_queue_id,
1105 uint16_t nb_tx_desc __rte_unused,
1106 unsigned int socket_id __rte_unused,
1107 const struct rte_eth_txconf *tx_conf)
1109 struct pmd_internals *internals = dev->data->dev_private;
1110 struct tx_queue *txq;
1114 if (tx_queue_id >= dev->data->nb_tx_queues)
1116 dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1117 txq = dev->data->tx_queues[tx_queue_id];
1119 offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1120 txq->csum = !!(offloads &
1121 (DEV_TX_OFFLOAD_IPV4_CKSUM |
1122 DEV_TX_OFFLOAD_UDP_CKSUM |
1123 DEV_TX_OFFLOAD_TCP_CKSUM));
1125 ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1129 " TX TUNTAP device name %s, qid %d on fd %d csum %s",
1130 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd,
1131 txq->csum ? "on" : "off");
1137 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1139 struct pmd_internals *pmd = dev->data->dev_private;
1140 struct ifreq ifr = { .ifr_mtu = mtu };
1143 err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1145 dev->data->mtu = mtu;
1151 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1152 struct ether_addr *mc_addr_set __rte_unused,
1153 uint32_t nb_mc_addr __rte_unused)
1156 * Nothing to do actually: the tap has no filtering whatsoever, every
1157 * packet is received.
1163 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1165 struct rte_eth_dev *dev = arg;
1166 struct pmd_internals *pmd = dev->data->dev_private;
1167 struct ifinfomsg *info = NLMSG_DATA(nh);
1169 if (nh->nlmsg_type != RTM_NEWLINK ||
1170 (info->ifi_index != pmd->if_index &&
1171 info->ifi_index != pmd->remote_if_index))
1173 return tap_link_update(dev, 0);
1177 tap_dev_intr_handler(void *cb_arg)
1179 struct rte_eth_dev *dev = cb_arg;
1180 struct pmd_internals *pmd = dev->data->dev_private;
1182 tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1186 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1188 struct pmd_internals *pmd = dev->data->dev_private;
1190 /* In any case, disable interrupt if the conf is no longer there. */
1191 if (!dev->data->dev_conf.intr_conf.lsc) {
1192 if (pmd->intr_handle.fd != -1) {
1193 tap_nl_final(pmd->intr_handle.fd);
1194 rte_intr_callback_unregister(&pmd->intr_handle,
1195 tap_dev_intr_handler, dev);
1200 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1201 if (unlikely(pmd->intr_handle.fd == -1))
1203 return rte_intr_callback_register(
1204 &pmd->intr_handle, tap_dev_intr_handler, dev);
1206 tap_nl_final(pmd->intr_handle.fd);
1207 return rte_intr_callback_unregister(&pmd->intr_handle,
1208 tap_dev_intr_handler, dev);
1212 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1216 err = tap_lsc_intr_handle_set(dev, set);
1219 err = tap_rx_intr_vec_set(dev, set);
1221 tap_lsc_intr_handle_set(dev, 0);
1225 static const uint32_t*
1226 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1228 static const uint32_t ptypes[] = {
1229 RTE_PTYPE_INNER_L2_ETHER,
1230 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1231 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1232 RTE_PTYPE_INNER_L3_IPV4,
1233 RTE_PTYPE_INNER_L3_IPV4_EXT,
1234 RTE_PTYPE_INNER_L3_IPV6,
1235 RTE_PTYPE_INNER_L3_IPV6_EXT,
1236 RTE_PTYPE_INNER_L4_FRAG,
1237 RTE_PTYPE_INNER_L4_UDP,
1238 RTE_PTYPE_INNER_L4_TCP,
1239 RTE_PTYPE_INNER_L4_SCTP,
1241 RTE_PTYPE_L2_ETHER_VLAN,
1242 RTE_PTYPE_L2_ETHER_QINQ,
1244 RTE_PTYPE_L3_IPV4_EXT,
1245 RTE_PTYPE_L3_IPV6_EXT,
1257 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1258 struct rte_eth_fc_conf *fc_conf)
1260 fc_conf->mode = RTE_FC_NONE;
1265 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1266 struct rte_eth_fc_conf *fc_conf)
1268 if (fc_conf->mode != RTE_FC_NONE)
1273 static const struct eth_dev_ops ops = {
1274 .dev_start = tap_dev_start,
1275 .dev_stop = tap_dev_stop,
1276 .dev_close = tap_dev_close,
1277 .dev_configure = tap_dev_configure,
1278 .dev_infos_get = tap_dev_info,
1279 .rx_queue_setup = tap_rx_queue_setup,
1280 .tx_queue_setup = tap_tx_queue_setup,
1281 .rx_queue_release = tap_rx_queue_release,
1282 .tx_queue_release = tap_tx_queue_release,
1283 .flow_ctrl_get = tap_flow_ctrl_get,
1284 .flow_ctrl_set = tap_flow_ctrl_set,
1285 .link_update = tap_link_update,
1286 .dev_set_link_up = tap_link_set_up,
1287 .dev_set_link_down = tap_link_set_down,
1288 .promiscuous_enable = tap_promisc_enable,
1289 .promiscuous_disable = tap_promisc_disable,
1290 .allmulticast_enable = tap_allmulti_enable,
1291 .allmulticast_disable = tap_allmulti_disable,
1292 .mac_addr_set = tap_mac_set,
1293 .mtu_set = tap_mtu_set,
1294 .set_mc_addr_list = tap_set_mc_addr_list,
1295 .stats_get = tap_stats_get,
1296 .stats_reset = tap_stats_reset,
1297 .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1298 .filter_ctrl = tap_dev_filter_ctrl,
1302 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
1303 char *remote_iface, struct ether_addr *mac_addr)
1305 int numa_node = rte_socket_id();
1306 struct rte_eth_dev *dev;
1307 struct pmd_internals *pmd;
1308 struct rte_eth_dev_data *data;
1312 TAP_LOG(DEBUG, "%s device on numa %u",
1313 tuntap_name, rte_socket_id());
1315 dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1317 TAP_LOG(ERR, "%s Unable to allocate device struct",
1319 goto error_exit_nodev;
1322 pmd = dev->data->dev_private;
1324 snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1326 pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1327 if (pmd->ioctl_sock == -1) {
1329 "%s Unable to get a socket for management: %s",
1330 tuntap_name, strerror(errno));
1334 /* Setup some default values */
1336 data->dev_private = pmd;
1337 data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1338 data->numa_node = numa_node;
1340 data->dev_link = pmd_link;
1341 data->mac_addrs = &pmd->eth_addr;
1342 /* Set the number of RX and TX queues */
1343 data->nb_rx_queues = 0;
1344 data->nb_tx_queues = 0;
1346 dev->dev_ops = &ops;
1347 dev->rx_pkt_burst = pmd_rx_burst;
1348 dev->tx_pkt_burst = pmd_tx_burst;
1350 pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1351 pmd->intr_handle.fd = -1;
1352 dev->intr_handle = &pmd->intr_handle;
1354 /* Presetup the fds to -1 as being not valid */
1355 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1356 pmd->rxq[i].fd = -1;
1357 pmd->txq[i].fd = -1;
1361 if (is_zero_ether_addr(mac_addr))
1362 eth_random_addr((uint8_t *)&pmd->eth_addr);
1364 rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1367 /* Immediately create the netdevice (this will create the 1st queue). */
1369 if (tap_setup_queue(dev, pmd, 0, 1) == -1)
1372 if (tap_setup_queue(dev, pmd, 0, 0) == -1)
1375 ifr.ifr_mtu = dev->data->mtu;
1376 if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1380 memset(&ifr, 0, sizeof(struct ifreq));
1381 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1382 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
1384 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1389 * Set up everything related to rte_flow:
1391 * - tap / remote if_index
1392 * - mandatory QDISCs
1393 * - rte_flow actual/implicit lists
1396 pmd->nlsk_fd = tap_nl_init(0);
1397 if (pmd->nlsk_fd == -1) {
1398 TAP_LOG(WARNING, "%s: failed to create netlink socket.",
1400 goto disable_rte_flow;
1402 pmd->if_index = if_nametoindex(pmd->name);
1403 if (!pmd->if_index) {
1404 TAP_LOG(ERR, "%s: failed to get if_index.", pmd->name);
1405 goto disable_rte_flow;
1407 if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1408 TAP_LOG(ERR, "%s: failed to create multiq qdisc.",
1410 goto disable_rte_flow;
1412 if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1413 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1415 goto disable_rte_flow;
1417 LIST_INIT(&pmd->flows);
1419 if (strlen(remote_iface)) {
1420 pmd->remote_if_index = if_nametoindex(remote_iface);
1421 if (!pmd->remote_if_index) {
1422 TAP_LOG(ERR, "%s: failed to get %s if_index.",
1423 pmd->name, remote_iface);
1426 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1427 "%s", remote_iface);
1429 /* Save state of remote device */
1430 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1432 /* Replicate remote MAC address */
1433 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1434 TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1435 pmd->name, pmd->remote_iface);
1438 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1440 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1441 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1442 TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1443 pmd->name, remote_iface);
1448 * Flush usually returns negative value because it tries to
1449 * delete every QDISC (and on a running device, one QDISC at
1450 * least is needed). Ignore negative return value.
1452 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1453 if (qdisc_create_ingress(pmd->nlsk_fd,
1454 pmd->remote_if_index) < 0) {
1455 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1459 LIST_INIT(&pmd->implicit_flows);
1460 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1461 tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1462 tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1463 tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1465 "%s: failed to create implicit rules.",
1474 TAP_LOG(ERR, " Disabling rte flow support: %s(%d)",
1475 strerror(errno), errno);
1476 if (strlen(remote_iface)) {
1477 TAP_LOG(ERR, "Remote feature requires flow support.");
1483 TAP_LOG(ERR, " Can't set up remote feature: %s(%d)",
1484 strerror(errno), errno);
1485 tap_flow_implicit_flush(pmd, NULL);
1488 if (pmd->ioctl_sock > 0)
1489 close(pmd->ioctl_sock);
1490 rte_eth_dev_release_port(dev);
1493 TAP_LOG(ERR, "%s Unable to initialize %s",
1494 tuntap_name, rte_vdev_device_name(vdev));
1500 set_interface_name(const char *key __rte_unused,
1504 char *name = (char *)extra_args;
1507 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN - 1);
1509 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1510 DEFAULT_TAP_NAME, (tap_unit - 1));
1516 set_remote_iface(const char *key __rte_unused,
1520 char *name = (char *)extra_args;
1523 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
1528 static int parse_user_mac(struct ether_addr *user_mac,
1531 unsigned int index = 0;
1532 char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
1534 if (user_mac == NULL || value == NULL)
1537 strlcpy(mac_temp, value, sizeof(mac_temp));
1538 mac_byte = strtok(mac_temp, ":");
1540 while ((mac_byte != NULL) &&
1541 (strlen(mac_byte) <= 2) &&
1542 (strlen(mac_byte) == strspn(mac_byte,
1543 ETH_TAP_CMP_MAC_FMT))) {
1544 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
1545 mac_byte = strtok(NULL, ":");
1552 set_mac_type(const char *key __rte_unused,
1556 struct ether_addr *user_mac = extra_args;
1561 if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
1562 static int iface_idx;
1564 /* fixed mac = 00:64:74:61:70:<iface_idx> */
1565 memcpy((char *)user_mac->addr_bytes, "\0dtap", ETHER_ADDR_LEN);
1566 user_mac->addr_bytes[ETHER_ADDR_LEN - 1] = iface_idx++ + '0';
1570 if (parse_user_mac(user_mac, value) != 6)
1573 TAP_LOG(DEBUG, "TAP user MAC param (%s)", value);
1577 TAP_LOG(ERR, "TAP user MAC (%s) is not in format (%s|%s)",
1578 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
1583 * Open a TUN interface device. TUN PMD
1584 * 1) sets tap_type as false
1585 * 2) intakes iface as argument.
1586 * 3) as interface is virtual set speed to 10G
1589 rte_pmd_tun_probe(struct rte_vdev_device *dev)
1591 const char *name, *params;
1593 struct rte_kvargs *kvlist = NULL;
1594 char tun_name[RTE_ETH_NAME_MAX_LEN];
1595 char remote_iface[RTE_ETH_NAME_MAX_LEN];
1598 strcpy(tuntap_name, "TUN");
1600 name = rte_vdev_device_name(dev);
1601 params = rte_vdev_device_args(dev);
1602 memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1604 if (params && (params[0] != '\0')) {
1605 TAP_LOG(DEBUG, "parameters (%s)", params);
1607 kvlist = rte_kvargs_parse(params, valid_arguments);
1609 if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1610 ret = rte_kvargs_process(kvlist,
1612 &set_interface_name,
1620 pmd_link.link_speed = ETH_SPEED_NUM_10G;
1622 TAP_LOG(NOTICE, "Initializing pmd_tun for %s as %s",
1625 ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0);
1629 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
1631 tun_unit--; /* Restore the unit number */
1633 rte_kvargs_free(kvlist);
1638 /* Open a TAP interface device.
1641 rte_pmd_tap_probe(struct rte_vdev_device *dev)
1643 const char *name, *params;
1645 struct rte_kvargs *kvlist = NULL;
1647 char tap_name[RTE_ETH_NAME_MAX_LEN];
1648 char remote_iface[RTE_ETH_NAME_MAX_LEN];
1649 struct ether_addr user_mac = { .addr_bytes = {0} };
1650 struct rte_eth_dev *eth_dev;
1653 strcpy(tuntap_name, "TAP");
1655 name = rte_vdev_device_name(dev);
1656 params = rte_vdev_device_args(dev);
1658 if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
1659 strlen(params) == 0) {
1660 eth_dev = rte_eth_dev_attach_secondary(name);
1662 TAP_LOG(ERR, "Failed to probe %s", name);
1665 /* TODO: request info from primary to set up Rx and Tx */
1666 eth_dev->dev_ops = &ops;
1670 speed = ETH_SPEED_NUM_10G;
1671 snprintf(tap_name, sizeof(tap_name), "%s%d",
1672 DEFAULT_TAP_NAME, tap_unit++);
1673 memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1675 if (params && (params[0] != '\0')) {
1676 TAP_LOG(DEBUG, "parameters (%s)", params);
1678 kvlist = rte_kvargs_parse(params, valid_arguments);
1680 if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1681 ret = rte_kvargs_process(kvlist,
1683 &set_interface_name,
1689 if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
1690 ret = rte_kvargs_process(kvlist,
1698 if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
1699 ret = rte_kvargs_process(kvlist,
1708 pmd_link.link_speed = speed;
1710 TAP_LOG(NOTICE, "Initializing pmd_tap for %s as %s",
1713 ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac);
1717 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
1719 tap_unit--; /* Restore the unit number */
1721 rte_kvargs_free(kvlist);
1726 /* detach a TUNTAP device.
1729 rte_pmd_tap_remove(struct rte_vdev_device *dev)
1731 struct rte_eth_dev *eth_dev = NULL;
1732 struct pmd_internals *internals;
1735 TAP_LOG(DEBUG, "Closing TUN/TAP Ethernet device on numa %u",
1738 /* find the ethdev entry */
1739 eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1743 internals = eth_dev->data->dev_private;
1744 if (internals->nlsk_fd) {
1745 tap_flow_flush(eth_dev, NULL);
1746 tap_flow_implicit_flush(internals, NULL);
1747 tap_nl_final(internals->nlsk_fd);
1749 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1750 if (internals->rxq[i].fd != -1) {
1751 close(internals->rxq[i].fd);
1752 internals->rxq[i].fd = -1;
1754 if (internals->txq[i].fd != -1) {
1755 close(internals->txq[i].fd);
1756 internals->txq[i].fd = -1;
1760 close(internals->ioctl_sock);
1761 rte_free(eth_dev->data->dev_private);
1763 rte_eth_dev_release_port(eth_dev);
1768 static struct rte_vdev_driver pmd_tun_drv = {
1769 .probe = rte_pmd_tun_probe,
1770 .remove = rte_pmd_tap_remove,
1773 static struct rte_vdev_driver pmd_tap_drv = {
1774 .probe = rte_pmd_tap_probe,
1775 .remove = rte_pmd_tap_remove,
1778 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1779 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
1780 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1781 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
1782 ETH_TAP_IFACE_ARG "=<string> ");
1783 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
1784 ETH_TAP_IFACE_ARG "=<string> "
1785 ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
1786 ETH_TAP_REMOTE_ARG "=<string>");
1789 RTE_INIT(tap_init_log);
1793 tap_logtype = rte_log_register("pmd.net.tap");
1794 if (tap_logtype >= 0)
1795 rte_log_set_level(tap_logtype, RTE_LOG_NOTICE);