4 * Copyright(c) 2016 Intel Corporation. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * * Neither the name of Intel Corporation nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 #include <rte_atomic.h>
35 #include <rte_branch_prediction.h>
36 #include <rte_common.h>
38 #include <rte_ethdev.h>
39 #include <rte_malloc.h>
41 #include <rte_kvargs.h>
44 #include <sys/types.h>
46 #include <sys/socket.h>
47 #include <sys/ioctl.h>
48 #include <sys/utsname.h>
55 #include <arpa/inet.h>
57 #include <linux/if_tun.h>
58 #include <linux/if_ether.h>
59 #include <linux/version.h>
62 #include <rte_eth_tap.h>
64 #include <tap_netlink.h>
65 #include <tap_tcmsgs.h>
67 /* Linux based path to the TUN device */
68 #define TUN_TAP_DEV_PATH "/dev/net/tun"
69 #define DEFAULT_TAP_NAME "dtap"
71 #define ETH_TAP_IFACE_ARG "iface"
72 #define ETH_TAP_SPEED_ARG "speed"
73 #define ETH_TAP_REMOTE_ARG "remote"
75 #define FLOWER_KERNEL_VERSION KERNEL_VERSION(4, 2, 0)
76 #define FLOWER_VLAN_KERNEL_VERSION KERNEL_VERSION(4, 9, 0)
78 static struct rte_vdev_driver pmd_tap_drv;
80 static const char *valid_arguments[] = {
89 static volatile uint32_t tap_trigger; /* Rx trigger */
91 static struct rte_eth_link pmd_link = {
92 .link_speed = ETH_SPEED_NUM_10G,
93 .link_duplex = ETH_LINK_FULL_DUPLEX,
94 .link_status = ETH_LINK_DOWN,
95 .link_autoneg = ETH_LINK_SPEED_AUTONEG
99 tap_trigger_cb(int sig __rte_unused)
101 /* Valid trigger values are nonzero */
102 tap_trigger = (tap_trigger + 1) | 0x80000000;
105 /* Specifies on what netdevices the ioctl should be applied */
113 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
114 struct ifreq *ifr, int set, enum ioctl_mode mode);
116 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
118 /* Tun/Tap allocation routine
120 * name is the number of the interface to use, unless NULL to take the host
124 tun_alloc(struct pmd_internals *pmd, uint16_t qid)
127 #ifdef IFF_MULTI_QUEUE
128 unsigned int features;
132 memset(&ifr, 0, sizeof(struct ifreq));
135 * Do not set IFF_NO_PI as packet information header will be needed
136 * to check if a received packet has been truncated.
138 ifr.ifr_flags = IFF_TAP;
139 snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
141 RTE_LOG(DEBUG, PMD, "ifr_name '%s'\n", ifr.ifr_name);
143 fd = open(TUN_TAP_DEV_PATH, O_RDWR);
145 RTE_LOG(ERR, PMD, "Unable to create TAP interface");
149 #ifdef IFF_MULTI_QUEUE
150 /* Grab the TUN features to verify we can work multi-queue */
151 if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
152 RTE_LOG(ERR, PMD, "TAP unable to get TUN/TAP features\n");
155 RTE_LOG(DEBUG, PMD, " TAP Features %08x\n", features);
157 if (features & IFF_MULTI_QUEUE) {
158 RTE_LOG(DEBUG, PMD, " Multi-queue support for %d queues\n",
159 RTE_PMD_TAP_MAX_QUEUES);
160 ifr.ifr_flags |= IFF_MULTI_QUEUE;
164 ifr.ifr_flags |= IFF_ONE_QUEUE;
165 RTE_LOG(DEBUG, PMD, " Single queue only support\n");
168 /* Set the TUN/TAP configuration and set the name if needed */
169 if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
170 RTE_LOG(WARNING, PMD,
171 "Unable to set TUNSETIFF for %s\n",
177 /* Always set the file descriptor to non-blocking */
178 if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
179 RTE_LOG(WARNING, PMD,
180 "Unable to set %s to nonblocking\n",
182 perror("F_SETFL, NONBLOCK");
186 /* Set up trigger to optimize empty Rx bursts */
190 int flags = fcntl(fd, F_GETFL);
192 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
194 if (sa.sa_handler != tap_trigger_cb) {
196 * Make sure SIGIO is not already taken. This is done
197 * as late as possible to leave the application a
198 * chance to set up its own signal handler first.
200 if (sa.sa_handler != SIG_IGN &&
201 sa.sa_handler != SIG_DFL) {
205 sa = (struct sigaction){
206 .sa_flags = SA_RESTART,
207 .sa_handler = tap_trigger_cb,
209 if (sigaction(SIGIO, &sa, NULL) == -1)
212 /* Enable SIGIO on file descriptor */
213 fcntl(fd, F_SETFL, flags | O_ASYNC);
214 fcntl(fd, F_SETOWN, getpid());
217 /* Disable trigger globally in case of error */
219 RTE_LOG(WARNING, PMD, "Rx trigger disabled: %s\n",
227 * pmd->eth_addr contains the desired MAC, either from remote
228 * or from a random assignment. Sync it with the tap netdevice.
230 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
231 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
233 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
236 pmd->if_index = if_nametoindex(pmd->name);
237 if (!pmd->if_index) {
239 "Could not find ifindex for %s: rte_flow won't be usable.\n",
243 if (!pmd->flower_support)
245 if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
247 "Could not create multiq qdisc for %s: rte_flow won't be usable.\n",
251 if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
253 "Could not create multiq qdisc for %s: rte_flow won't be usable.\n",
257 if (pmd->remote_if_index) {
259 * Flush usually returns negative value because it tries
260 * to delete every QDISC (and on a running device, one
261 * QDISC at least is needed). Ignore negative return
264 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
265 if (qdisc_create_ingress(pmd->nlsk_fd,
266 pmd->remote_if_index) < 0)
268 LIST_INIT(&pmd->implicit_flows);
269 if (tap_flow_implicit_create(
270 pmd, TAP_REMOTE_LOCAL_MAC) < 0)
272 if (tap_flow_implicit_create(
273 pmd, TAP_REMOTE_BROADCAST) < 0)
275 if (tap_flow_implicit_create(
276 pmd, TAP_REMOTE_BROADCASTV6) < 0)
278 if (tap_flow_implicit_create(
279 pmd, TAP_REMOTE_TX) < 0)
288 "Could not set up remote flow rules for %s: remote disabled.\n",
290 pmd->remote_if_index = 0;
291 tap_flow_implicit_flush(pmd, NULL);
300 /* Callback to handle the rx burst of packets to the correct interface and
301 * file descriptor(s) in a multi-queue setup.
304 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
306 struct rx_queue *rxq = queue;
308 unsigned long num_rx_bytes = 0;
309 uint32_t trigger = tap_trigger;
311 if (trigger == rxq->trigger_seen)
314 rxq->trigger_seen = trigger;
315 rte_compiler_barrier();
316 for (num_rx = 0; num_rx < nb_pkts; ) {
317 struct rte_mbuf *mbuf = rxq->pool;
318 struct rte_mbuf *seg = NULL;
319 struct rte_mbuf *new_tail = NULL;
320 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
323 len = readv(rxq->fd, *rxq->iovecs,
324 1 + (rxq->rxmode->enable_scatter ?
325 rxq->nb_rx_desc : 1));
326 if (len < (int)sizeof(struct tun_pi))
329 /* Packet couldn't fit in the provided mbuf */
330 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
331 rxq->stats.ierrors++;
335 len -= sizeof(struct tun_pi);
338 mbuf->port = rxq->in_port;
340 struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
342 if (unlikely(!buf)) {
343 rxq->stats.rx_nombuf++;
344 /* No new buf has been allocated: do nothing */
345 if (!new_tail || !seg)
349 rte_pktmbuf_free(mbuf);
353 seg = seg ? seg->next : mbuf;
354 if (rxq->pool == mbuf)
357 new_tail->next = buf;
359 new_tail->next = seg->next;
361 /* iovecs[0] is reserved for packet info (pi) */
362 (*rxq->iovecs)[mbuf->nb_segs].iov_len =
363 buf->buf_len - data_off;
364 (*rxq->iovecs)[mbuf->nb_segs].iov_base =
365 (char *)buf->buf_addr + data_off;
367 seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
368 seg->data_off = data_off;
370 len -= seg->data_len;
374 /* First segment has headroom, not the others */
378 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
381 /* account for the receive frame */
382 bufs[num_rx++] = mbuf;
383 num_rx_bytes += mbuf->pkt_len;
386 rxq->stats.ipackets += num_rx;
387 rxq->stats.ibytes += num_rx_bytes;
392 /* Callback to handle sending packets from the tap interface
395 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
397 struct tx_queue *txq = queue;
399 unsigned long num_tx_bytes = 0;
403 if (unlikely(nb_pkts == 0))
406 max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
407 for (i = 0; i < nb_pkts; i++) {
408 struct rte_mbuf *mbuf = bufs[num_tx];
409 struct iovec iovecs[mbuf->nb_segs + 1];
410 struct tun_pi pi = { .flags = 0 };
411 struct rte_mbuf *seg = mbuf;
415 /* stats.errs will be incremented */
416 if (rte_pktmbuf_pkt_len(mbuf) > max_size)
419 iovecs[0].iov_base = π
420 iovecs[0].iov_len = sizeof(pi);
421 for (j = 1; j <= mbuf->nb_segs; j++) {
422 iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
424 rte_pktmbuf_mtod(seg, void *);
427 /* copy the tx frame data */
428 n = writev(txq->fd, iovecs, mbuf->nb_segs + 1);
433 num_tx_bytes += mbuf->pkt_len;
434 rte_pktmbuf_free(mbuf);
437 txq->stats.opackets += num_tx;
438 txq->stats.errs += nb_pkts - num_tx;
439 txq->stats.obytes += num_tx_bytes;
445 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
446 struct ifreq *ifr, int set, enum ioctl_mode mode)
448 short req_flags = ifr->ifr_flags;
449 int remote = pmd->remote_if_index &&
450 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
452 if (!pmd->remote_if_index && mode == REMOTE_ONLY)
455 * If there is a remote netdevice, apply ioctl on it, then apply it on
460 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
461 else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
462 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
465 /* fetch current flags to leave other flags untouched */
466 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
469 ifr->ifr_flags |= req_flags;
471 ifr->ifr_flags &= ~req_flags;
479 RTE_LOG(WARNING, PMD, "%s: ioctl() called with wrong arg\n",
483 if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
485 if (remote-- && mode == LOCAL_AND_REMOTE)
490 RTE_LOG(ERR, PMD, "%s: ioctl(%lu) failed with error: %s\n",
491 ifr->ifr_name, request, strerror(errno));
496 tap_link_set_down(struct rte_eth_dev *dev)
498 struct pmd_internals *pmd = dev->data->dev_private;
499 struct ifreq ifr = { .ifr_flags = IFF_UP };
501 dev->data->dev_link.link_status = ETH_LINK_DOWN;
502 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
506 tap_link_set_up(struct rte_eth_dev *dev)
508 struct pmd_internals *pmd = dev->data->dev_private;
509 struct ifreq ifr = { .ifr_flags = IFF_UP };
511 dev->data->dev_link.link_status = ETH_LINK_UP;
512 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
516 tap_dev_start(struct rte_eth_dev *dev)
520 err = tap_intr_handle_set(dev, 1);
523 return tap_link_set_up(dev);
526 /* This function gets called when the current port gets stopped.
529 tap_dev_stop(struct rte_eth_dev *dev)
531 tap_intr_handle_set(dev, 0);
532 tap_link_set_down(dev);
536 tap_dev_configure(struct rte_eth_dev *dev __rte_unused)
542 tap_dev_speed_capa(void)
544 uint32_t speed = pmd_link.link_speed;
547 if (speed >= ETH_SPEED_NUM_10M)
548 capa |= ETH_LINK_SPEED_10M;
549 if (speed >= ETH_SPEED_NUM_100M)
550 capa |= ETH_LINK_SPEED_100M;
551 if (speed >= ETH_SPEED_NUM_1G)
552 capa |= ETH_LINK_SPEED_1G;
553 if (speed >= ETH_SPEED_NUM_5G)
554 capa |= ETH_LINK_SPEED_2_5G;
555 if (speed >= ETH_SPEED_NUM_5G)
556 capa |= ETH_LINK_SPEED_5G;
557 if (speed >= ETH_SPEED_NUM_10G)
558 capa |= ETH_LINK_SPEED_10G;
559 if (speed >= ETH_SPEED_NUM_20G)
560 capa |= ETH_LINK_SPEED_20G;
561 if (speed >= ETH_SPEED_NUM_25G)
562 capa |= ETH_LINK_SPEED_25G;
563 if (speed >= ETH_SPEED_NUM_40G)
564 capa |= ETH_LINK_SPEED_40G;
565 if (speed >= ETH_SPEED_NUM_50G)
566 capa |= ETH_LINK_SPEED_50G;
567 if (speed >= ETH_SPEED_NUM_56G)
568 capa |= ETH_LINK_SPEED_56G;
569 if (speed >= ETH_SPEED_NUM_100G)
570 capa |= ETH_LINK_SPEED_100G;
576 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
578 struct pmd_internals *internals = dev->data->dev_private;
580 dev_info->if_index = internals->if_index;
581 dev_info->max_mac_addrs = 1;
582 dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
583 dev_info->max_rx_queues = internals->nb_queues;
584 dev_info->max_tx_queues = internals->nb_queues;
585 dev_info->min_rx_bufsize = 0;
586 dev_info->pci_dev = NULL;
587 dev_info->speed_capa = tap_dev_speed_capa();
591 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
593 unsigned int i, imax;
594 unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
595 unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
596 unsigned long rx_nombuf = 0, ierrors = 0;
597 const struct pmd_internals *pmd = dev->data->dev_private;
599 imax = (pmd->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
600 pmd->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
602 for (i = 0; i < imax; i++) {
603 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
604 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
605 rx_total += tap_stats->q_ipackets[i];
606 rx_bytes_total += tap_stats->q_ibytes[i];
607 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
608 ierrors += pmd->rxq[i].stats.ierrors;
610 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
611 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
612 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
613 tx_total += tap_stats->q_opackets[i];
614 tx_err_total += tap_stats->q_errors[i];
615 tx_bytes_total += tap_stats->q_obytes[i];
618 tap_stats->ipackets = rx_total;
619 tap_stats->ibytes = rx_bytes_total;
620 tap_stats->ierrors = ierrors;
621 tap_stats->rx_nombuf = rx_nombuf;
622 tap_stats->opackets = tx_total;
623 tap_stats->oerrors = tx_err_total;
624 tap_stats->obytes = tx_bytes_total;
628 tap_stats_reset(struct rte_eth_dev *dev)
631 struct pmd_internals *pmd = dev->data->dev_private;
633 for (i = 0; i < pmd->nb_queues; i++) {
634 pmd->rxq[i].stats.ipackets = 0;
635 pmd->rxq[i].stats.ibytes = 0;
636 pmd->rxq[i].stats.ierrors = 0;
637 pmd->rxq[i].stats.rx_nombuf = 0;
639 pmd->txq[i].stats.opackets = 0;
640 pmd->txq[i].stats.errs = 0;
641 pmd->txq[i].stats.obytes = 0;
646 tap_dev_close(struct rte_eth_dev *dev __rte_unused)
649 struct pmd_internals *internals = dev->data->dev_private;
651 tap_link_set_down(dev);
652 tap_flow_flush(dev, NULL);
653 tap_flow_implicit_flush(internals, NULL);
655 for (i = 0; i < internals->nb_queues; i++) {
656 if (internals->rxq[i].fd != -1)
657 close(internals->rxq[i].fd);
658 internals->rxq[i].fd = -1;
659 internals->txq[i].fd = -1;
664 tap_rx_queue_release(void *queue)
666 struct rx_queue *rxq = queue;
668 if (rxq && (rxq->fd > 0)) {
671 rte_pktmbuf_free(rxq->pool);
672 rte_free(rxq->iovecs);
679 tap_tx_queue_release(void *queue)
681 struct tx_queue *txq = queue;
683 if (txq && (txq->fd > 0)) {
690 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
692 struct rte_eth_link *dev_link = &dev->data->dev_link;
693 struct pmd_internals *pmd = dev->data->dev_private;
694 struct ifreq ifr = { .ifr_flags = 0 };
696 if (pmd->remote_if_index) {
697 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
698 if (!(ifr.ifr_flags & IFF_UP) ||
699 !(ifr.ifr_flags & IFF_RUNNING)) {
700 dev_link->link_status = ETH_LINK_DOWN;
704 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
705 dev_link->link_status =
706 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
713 tap_promisc_enable(struct rte_eth_dev *dev)
715 struct pmd_internals *pmd = dev->data->dev_private;
716 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
718 dev->data->promiscuous = 1;
719 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
720 if (pmd->remote_if_index)
721 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
725 tap_promisc_disable(struct rte_eth_dev *dev)
727 struct pmd_internals *pmd = dev->data->dev_private;
728 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
730 dev->data->promiscuous = 0;
731 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
732 if (pmd->remote_if_index)
733 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
737 tap_allmulti_enable(struct rte_eth_dev *dev)
739 struct pmd_internals *pmd = dev->data->dev_private;
740 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
742 dev->data->all_multicast = 1;
743 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
744 if (pmd->remote_if_index)
745 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
749 tap_allmulti_disable(struct rte_eth_dev *dev)
751 struct pmd_internals *pmd = dev->data->dev_private;
752 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
754 dev->data->all_multicast = 0;
755 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
756 if (pmd->remote_if_index)
757 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
762 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
764 struct pmd_internals *pmd = dev->data->dev_private;
767 if (is_zero_ether_addr(mac_addr)) {
768 RTE_LOG(ERR, PMD, "%s: can't set an empty MAC address\n",
772 /* Check the actual current MAC address on the tap netdevice */
773 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY) != 0) {
775 "%s: couldn't check current tap MAC address\n",
779 if (is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
783 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
784 rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
785 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, LOCAL_AND_REMOTE) < 0)
787 rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
788 if (pmd->remote_if_index) {
789 /* Replace MAC redirection rule after a MAC change */
790 if (tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC) < 0) {
792 "%s: Couldn't delete MAC redirection rule\n",
796 if (tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0)
798 "%s: Couldn't add MAC redirection rule\n",
804 tap_setup_queue(struct rte_eth_dev *dev,
805 struct pmd_internals *internals,
808 struct pmd_internals *pmd = dev->data->dev_private;
809 struct rx_queue *rx = &internals->rxq[qid];
810 struct tx_queue *tx = &internals->txq[qid];
817 RTE_LOG(INFO, PMD, "Add queue to TAP %s for qid %d\n",
819 fd = tun_alloc(pmd, qid);
821 RTE_LOG(ERR, PMD, "tun_alloc(%s, %d) failed\n",
828 ifr.ifr_mtu = dev->data->mtu;
829 if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1,
830 LOCAL_AND_REMOTE) < 0) {
840 tx->mtu = &dev->data->mtu;
841 rx->rxmode = &dev->data->dev_conf.rxmode;
847 rx_setup_queue(struct rte_eth_dev *dev,
848 struct pmd_internals *internals,
851 dev->data->rx_queues[qid] = &internals->rxq[qid];
853 return tap_setup_queue(dev, internals, qid);
857 tx_setup_queue(struct rte_eth_dev *dev,
858 struct pmd_internals *internals,
861 dev->data->tx_queues[qid] = &internals->txq[qid];
863 return tap_setup_queue(dev, internals, qid);
867 tap_rx_queue_setup(struct rte_eth_dev *dev,
868 uint16_t rx_queue_id,
870 unsigned int socket_id,
871 const struct rte_eth_rxconf *rx_conf __rte_unused,
872 struct rte_mempool *mp)
874 struct pmd_internals *internals = dev->data->dev_private;
875 struct rx_queue *rxq = &internals->rxq[rx_queue_id];
876 struct rte_mbuf **tmp = &rxq->pool;
877 struct iovec (*iovecs)[nb_rx_desc + 1];
878 int data_off = RTE_PKTMBUF_HEADROOM;
883 if ((rx_queue_id >= internals->nb_queues) || !mp) {
884 RTE_LOG(WARNING, PMD,
885 "nb_queues %d too small or mempool NULL\n",
886 internals->nb_queues);
891 rxq->trigger_seen = 1; /* force initial burst */
892 rxq->in_port = dev->data->port_id;
893 rxq->nb_rx_desc = nb_rx_desc;
894 iovecs = rte_zmalloc_socket(dev->data->name, sizeof(*iovecs), 0,
897 RTE_LOG(WARNING, PMD,
898 "%s: Couldn't allocate %d RX descriptors\n",
899 dev->data->name, nb_rx_desc);
902 rxq->iovecs = iovecs;
904 fd = rx_setup_queue(dev, internals, rx_queue_id);
910 (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
911 (*rxq->iovecs)[0].iov_base = &rxq->pi;
913 for (i = 1; i <= nb_rx_desc; i++) {
914 *tmp = rte_pktmbuf_alloc(rxq->mp);
916 RTE_LOG(WARNING, PMD,
917 "%s: couldn't allocate memory for queue %d\n",
918 dev->data->name, rx_queue_id);
922 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
923 (*rxq->iovecs)[i].iov_base =
924 (char *)(*tmp)->buf_addr + data_off;
929 RTE_LOG(DEBUG, PMD, " RX TAP device name %s, qid %d on fd %d\n",
930 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
935 rte_pktmbuf_free(rxq->pool);
937 rte_free(rxq->iovecs);
943 tap_tx_queue_setup(struct rte_eth_dev *dev,
944 uint16_t tx_queue_id,
945 uint16_t nb_tx_desc __rte_unused,
946 unsigned int socket_id __rte_unused,
947 const struct rte_eth_txconf *tx_conf __rte_unused)
949 struct pmd_internals *internals = dev->data->dev_private;
952 if (tx_queue_id >= internals->nb_queues)
955 ret = tx_setup_queue(dev, internals, tx_queue_id);
959 RTE_LOG(DEBUG, PMD, " TX TAP device name %s, qid %d on fd %d\n",
960 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd);
966 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
968 struct pmd_internals *pmd = dev->data->dev_private;
969 struct ifreq ifr = { .ifr_mtu = mtu };
972 err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
974 dev->data->mtu = mtu;
980 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
981 struct ether_addr *mc_addr_set __rte_unused,
982 uint32_t nb_mc_addr __rte_unused)
985 * Nothing to do actually: the tap has no filtering whatsoever, every
986 * packet is received.
992 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
994 struct rte_eth_dev *dev = arg;
995 struct pmd_internals *pmd = dev->data->dev_private;
996 struct ifinfomsg *info = NLMSG_DATA(nh);
998 if (nh->nlmsg_type != RTM_NEWLINK ||
999 (info->ifi_index != pmd->if_index &&
1000 info->ifi_index != pmd->remote_if_index))
1002 return tap_link_update(dev, 0);
1006 tap_dev_intr_handler(void *cb_arg)
1008 struct rte_eth_dev *dev = cb_arg;
1009 struct pmd_internals *pmd = dev->data->dev_private;
1011 nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1015 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1017 struct pmd_internals *pmd = dev->data->dev_private;
1019 /* In any case, disable interrupt if the conf is no longer there. */
1020 if (!dev->data->dev_conf.intr_conf.lsc) {
1021 if (pmd->intr_handle.fd != -1)
1022 nl_final(pmd->intr_handle.fd);
1023 rte_intr_callback_unregister(
1024 &pmd->intr_handle, tap_dev_intr_handler, dev);
1028 pmd->intr_handle.fd = nl_init(RTMGRP_LINK);
1029 if (unlikely(pmd->intr_handle.fd == -1))
1031 return rte_intr_callback_register(
1032 &pmd->intr_handle, tap_dev_intr_handler, dev);
1034 nl_final(pmd->intr_handle.fd);
1035 return rte_intr_callback_unregister(&pmd->intr_handle,
1036 tap_dev_intr_handler, dev);
1039 static const uint32_t*
1040 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1042 static const uint32_t ptypes[] = {
1043 RTE_PTYPE_INNER_L2_ETHER,
1044 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1045 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1046 RTE_PTYPE_INNER_L3_IPV4,
1047 RTE_PTYPE_INNER_L3_IPV4_EXT,
1048 RTE_PTYPE_INNER_L3_IPV6,
1049 RTE_PTYPE_INNER_L3_IPV6_EXT,
1050 RTE_PTYPE_INNER_L4_FRAG,
1051 RTE_PTYPE_INNER_L4_UDP,
1052 RTE_PTYPE_INNER_L4_TCP,
1053 RTE_PTYPE_INNER_L4_SCTP,
1055 RTE_PTYPE_L2_ETHER_VLAN,
1056 RTE_PTYPE_L2_ETHER_QINQ,
1058 RTE_PTYPE_L3_IPV4_EXT,
1059 RTE_PTYPE_L3_IPV6_EXT,
1071 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1072 struct rte_eth_fc_conf *fc_conf)
1074 fc_conf->mode = RTE_FC_NONE;
1079 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1080 struct rte_eth_fc_conf *fc_conf)
1082 if (fc_conf->mode != RTE_FC_NONE)
1087 static const struct eth_dev_ops ops = {
1088 .dev_start = tap_dev_start,
1089 .dev_stop = tap_dev_stop,
1090 .dev_close = tap_dev_close,
1091 .dev_configure = tap_dev_configure,
1092 .dev_infos_get = tap_dev_info,
1093 .rx_queue_setup = tap_rx_queue_setup,
1094 .tx_queue_setup = tap_tx_queue_setup,
1095 .rx_queue_release = tap_rx_queue_release,
1096 .tx_queue_release = tap_tx_queue_release,
1097 .flow_ctrl_get = tap_flow_ctrl_get,
1098 .flow_ctrl_set = tap_flow_ctrl_set,
1099 .link_update = tap_link_update,
1100 .dev_set_link_up = tap_link_set_up,
1101 .dev_set_link_down = tap_link_set_down,
1102 .promiscuous_enable = tap_promisc_enable,
1103 .promiscuous_disable = tap_promisc_disable,
1104 .allmulticast_enable = tap_allmulti_enable,
1105 .allmulticast_disable = tap_allmulti_disable,
1106 .mac_addr_set = tap_mac_set,
1107 .mtu_set = tap_mtu_set,
1108 .set_mc_addr_list = tap_set_mc_addr_list,
1109 .stats_get = tap_stats_get,
1110 .stats_reset = tap_stats_reset,
1111 .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1112 .filter_ctrl = tap_dev_filter_ctrl,
1116 tap_kernel_support(struct pmd_internals *pmd)
1118 struct utsname utsname;
1121 if (uname(&utsname) == -1 ||
1122 sscanf(utsname.release, "%d.%d.%d",
1123 &ver[0], &ver[1], &ver[2]) != 3)
1125 if (KERNEL_VERSION(ver[0], ver[1], ver[2]) >= FLOWER_KERNEL_VERSION)
1126 pmd->flower_support = 1;
1127 if (KERNEL_VERSION(ver[0], ver[1], ver[2]) >=
1128 FLOWER_VLAN_KERNEL_VERSION)
1129 pmd->flower_vlan_support = 1;
1134 eth_dev_tap_create(const char *name, char *tap_name, char *remote_iface)
1136 int numa_node = rte_socket_id();
1137 struct rte_eth_dev *dev = NULL;
1138 struct pmd_internals *pmd = NULL;
1139 struct rte_eth_dev_data *data = NULL;
1142 RTE_LOG(DEBUG, PMD, " TAP device on numa %u\n", rte_socket_id());
1144 data = rte_zmalloc_socket(tap_name, sizeof(*data), 0, numa_node);
1146 RTE_LOG(ERR, PMD, "TAP Failed to allocate data\n");
1150 pmd = rte_zmalloc_socket(tap_name, sizeof(*pmd), 0, numa_node);
1152 RTE_LOG(ERR, PMD, "TAP Unable to allocate internal struct\n");
1156 /* name in allocation and data->name must be consistent */
1157 snprintf(data->name, sizeof(data->name), "%s", name);
1158 dev = rte_eth_dev_allocate(name);
1160 RTE_LOG(ERR, PMD, "TAP Unable to allocate device struct\n");
1164 snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1166 pmd->nb_queues = RTE_PMD_TAP_MAX_QUEUES;
1168 pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1169 if (pmd->ioctl_sock == -1) {
1171 "TAP Unable to get a socket for management: %s\n",
1176 /* Setup some default values */
1177 data->dev_private = pmd;
1178 data->port_id = dev->data->port_id;
1179 data->mtu = dev->data->mtu;
1180 data->dev_flags = RTE_ETH_DEV_DETACHABLE | RTE_ETH_DEV_INTR_LSC;
1181 data->kdrv = RTE_KDRV_NONE;
1182 data->drv_name = pmd_tap_drv.driver.name;
1183 data->numa_node = numa_node;
1185 data->dev_link = pmd_link;
1186 data->mac_addrs = &pmd->eth_addr;
1187 data->nb_rx_queues = pmd->nb_queues;
1188 data->nb_tx_queues = pmd->nb_queues;
1191 dev->dev_ops = &ops;
1193 dev->rx_pkt_burst = pmd_rx_burst;
1194 dev->tx_pkt_burst = pmd_tx_burst;
1196 pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1197 pmd->intr_handle.fd = -1;
1199 /* Presetup the fds to -1 as being not valid */
1200 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1201 pmd->rxq[i].fd = -1;
1202 pmd->txq[i].fd = -1;
1205 tap_kernel_support(pmd);
1206 if (!pmd->flower_support)
1208 LIST_INIT(&pmd->flows);
1210 * If no netlink socket can be created, then it will fail when
1211 * creating/destroying flow rules.
1213 pmd->nlsk_fd = nl_init(0);
1214 if (strlen(remote_iface)) {
1217 pmd->remote_if_index = if_nametoindex(remote_iface);
1218 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1219 "%s", remote_iface);
1220 if (!pmd->remote_if_index) {
1221 RTE_LOG(ERR, PMD, "Could not find %s ifindex: "
1222 "remote interface will remain unconfigured\n",
1226 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1227 RTE_LOG(ERR, PMD, "Could not get remote MAC address\n");
1230 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1233 eth_random_addr((uint8_t *)&pmd->eth_addr);
1239 RTE_LOG(DEBUG, PMD, "TAP Unable to initialize %s\n", name);
1244 rte_eth_dev_release_port(dev);
1250 set_interface_name(const char *key __rte_unused,
1254 char *name = (char *)extra_args;
1257 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s", value);
1259 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1260 DEFAULT_TAP_NAME, (tap_unit - 1));
1266 set_interface_speed(const char *key __rte_unused,
1270 *(int *)extra_args = (value) ? atoi(value) : ETH_SPEED_NUM_10G;
1276 set_remote_iface(const char *key __rte_unused,
1280 char *name = (char *)extra_args;
1283 snprintf(name, RTE_ETH_NAME_MAX_LEN, "%s", value);
1288 /* Open a TAP interface device.
1291 rte_pmd_tap_probe(const char *name, const char *params)
1294 struct rte_kvargs *kvlist = NULL;
1296 char tap_name[RTE_ETH_NAME_MAX_LEN];
1297 char remote_iface[RTE_ETH_NAME_MAX_LEN];
1299 speed = ETH_SPEED_NUM_10G;
1300 snprintf(tap_name, sizeof(tap_name), "%s%d",
1301 DEFAULT_TAP_NAME, tap_unit++);
1302 memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1304 if (params && (params[0] != '\0')) {
1305 RTE_LOG(DEBUG, PMD, "paramaters (%s)\n", params);
1307 kvlist = rte_kvargs_parse(params, valid_arguments);
1309 if (rte_kvargs_count(kvlist, ETH_TAP_SPEED_ARG) == 1) {
1310 ret = rte_kvargs_process(kvlist,
1312 &set_interface_speed,
1318 if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1319 ret = rte_kvargs_process(kvlist,
1321 &set_interface_name,
1327 if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
1328 ret = rte_kvargs_process(kvlist,
1337 pmd_link.link_speed = speed;
1339 RTE_LOG(NOTICE, PMD, "Initializing pmd_tap for %s as %s\n",
1342 ret = eth_dev_tap_create(name, tap_name, remote_iface);
1346 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
1348 tap_unit--; /* Restore the unit number */
1350 rte_kvargs_free(kvlist);
1355 /* detach a TAP device.
1358 rte_pmd_tap_remove(const char *name)
1360 struct rte_eth_dev *eth_dev = NULL;
1361 struct pmd_internals *internals;
1364 RTE_LOG(DEBUG, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
1367 /* find the ethdev entry */
1368 eth_dev = rte_eth_dev_allocated(name);
1372 internals = eth_dev->data->dev_private;
1373 if (internals->flower_support && internals->nlsk_fd) {
1374 tap_flow_flush(eth_dev, NULL);
1375 tap_flow_implicit_flush(internals, NULL);
1376 nl_final(internals->nlsk_fd);
1378 for (i = 0; i < internals->nb_queues; i++)
1379 if (internals->rxq[i].fd != -1)
1380 close(internals->rxq[i].fd);
1382 close(internals->ioctl_sock);
1383 rte_free(eth_dev->data->dev_private);
1384 rte_free(eth_dev->data);
1386 rte_eth_dev_release_port(eth_dev);
1391 static struct rte_vdev_driver pmd_tap_drv = {
1392 .probe = rte_pmd_tap_probe,
1393 .remove = rte_pmd_tap_remove,
1395 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1396 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1397 RTE_PMD_REGISTER_PARAM_STRING(net_tap, "iface=<string>,speed=N");