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_ethdev_vdev.h>
40 #include <rte_malloc.h>
42 #include <rte_kvargs.h>
45 #include <sys/types.h>
47 #include <sys/socket.h>
48 #include <sys/ioctl.h>
49 #include <sys/utsname.h>
56 #include <arpa/inet.h>
58 #include <linux/if_tun.h>
59 #include <linux/if_ether.h>
60 #include <linux/version.h>
63 #include <rte_eth_tap.h>
65 #include <tap_netlink.h>
66 #include <tap_tcmsgs.h>
68 /* Linux based path to the TUN device */
69 #define TUN_TAP_DEV_PATH "/dev/net/tun"
70 #define DEFAULT_TAP_NAME "dtap"
72 #define ETH_TAP_IFACE_ARG "iface"
73 #define ETH_TAP_SPEED_ARG "speed"
74 #define ETH_TAP_REMOTE_ARG "remote"
76 #define FLOWER_KERNEL_VERSION KERNEL_VERSION(4, 2, 0)
77 #define FLOWER_VLAN_KERNEL_VERSION KERNEL_VERSION(4, 9, 0)
79 static struct rte_vdev_driver pmd_tap_drv;
81 static const char *valid_arguments[] = {
90 static volatile uint32_t tap_trigger; /* Rx trigger */
92 static struct rte_eth_link pmd_link = {
93 .link_speed = ETH_SPEED_NUM_10G,
94 .link_duplex = ETH_LINK_FULL_DUPLEX,
95 .link_status = ETH_LINK_DOWN,
96 .link_autoneg = ETH_LINK_SPEED_AUTONEG
100 tap_trigger_cb(int sig __rte_unused)
102 /* Valid trigger values are nonzero */
103 tap_trigger = (tap_trigger + 1) | 0x80000000;
106 /* Specifies on what netdevices the ioctl should be applied */
114 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
115 struct ifreq *ifr, int set, enum ioctl_mode mode);
117 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
119 /* Tun/Tap allocation routine
121 * name is the number of the interface to use, unless NULL to take the host
125 tun_alloc(struct pmd_internals *pmd, uint16_t qid)
128 #ifdef IFF_MULTI_QUEUE
129 unsigned int features;
133 memset(&ifr, 0, sizeof(struct ifreq));
136 * Do not set IFF_NO_PI as packet information header will be needed
137 * to check if a received packet has been truncated.
139 ifr.ifr_flags = IFF_TAP;
140 snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
142 RTE_LOG(DEBUG, PMD, "ifr_name '%s'\n", ifr.ifr_name);
144 fd = open(TUN_TAP_DEV_PATH, O_RDWR);
146 RTE_LOG(ERR, PMD, "Unable to create TAP interface");
150 #ifdef IFF_MULTI_QUEUE
151 /* Grab the TUN features to verify we can work multi-queue */
152 if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
153 RTE_LOG(ERR, PMD, "TAP unable to get TUN/TAP features\n");
156 RTE_LOG(DEBUG, PMD, " TAP Features %08x\n", features);
158 if (features & IFF_MULTI_QUEUE) {
159 RTE_LOG(DEBUG, PMD, " Multi-queue support for %d queues\n",
160 RTE_PMD_TAP_MAX_QUEUES);
161 ifr.ifr_flags |= IFF_MULTI_QUEUE;
165 ifr.ifr_flags |= IFF_ONE_QUEUE;
166 RTE_LOG(DEBUG, PMD, " Single queue only support\n");
169 /* Set the TUN/TAP configuration and set the name if needed */
170 if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
171 RTE_LOG(WARNING, PMD,
172 "Unable to set TUNSETIFF for %s\n",
178 /* Always set the file descriptor to non-blocking */
179 if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
180 RTE_LOG(WARNING, PMD,
181 "Unable to set %s to nonblocking\n",
183 perror("F_SETFL, NONBLOCK");
187 /* Set up trigger to optimize empty Rx bursts */
191 int flags = fcntl(fd, F_GETFL);
193 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
195 if (sa.sa_handler != tap_trigger_cb) {
197 * Make sure SIGIO is not already taken. This is done
198 * as late as possible to leave the application a
199 * chance to set up its own signal handler first.
201 if (sa.sa_handler != SIG_IGN &&
202 sa.sa_handler != SIG_DFL) {
206 sa = (struct sigaction){
207 .sa_flags = SA_RESTART,
208 .sa_handler = tap_trigger_cb,
210 if (sigaction(SIGIO, &sa, NULL) == -1)
213 /* Enable SIGIO on file descriptor */
214 fcntl(fd, F_SETFL, flags | O_ASYNC);
215 fcntl(fd, F_SETOWN, getpid());
218 /* Disable trigger globally in case of error */
220 RTE_LOG(WARNING, PMD, "Rx trigger disabled: %s\n",
228 * pmd->eth_addr contains the desired MAC, either from remote
229 * or from a random assignment. Sync it with the tap netdevice.
231 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
232 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
234 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
237 pmd->if_index = if_nametoindex(pmd->name);
238 if (!pmd->if_index) {
240 "Could not find ifindex for %s: rte_flow won't be usable.\n",
244 if (!pmd->flower_support)
246 if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
248 "Could not create multiq qdisc for %s: rte_flow won't be usable.\n",
252 if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
254 "Could not create multiq qdisc for %s: rte_flow won't be usable.\n",
258 if (pmd->remote_if_index) {
260 * Flush usually returns negative value because it tries
261 * to delete every QDISC (and on a running device, one
262 * QDISC at least is needed). Ignore negative return
265 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
266 if (qdisc_create_ingress(pmd->nlsk_fd,
267 pmd->remote_if_index) < 0)
269 LIST_INIT(&pmd->implicit_flows);
270 if (tap_flow_implicit_create(
271 pmd, TAP_REMOTE_LOCAL_MAC) < 0)
273 if (tap_flow_implicit_create(
274 pmd, TAP_REMOTE_BROADCAST) < 0)
276 if (tap_flow_implicit_create(
277 pmd, TAP_REMOTE_BROADCASTV6) < 0)
279 if (tap_flow_implicit_create(
280 pmd, TAP_REMOTE_TX) < 0)
289 "Could not set up remote flow rules for %s: remote disabled.\n",
291 pmd->remote_if_index = 0;
292 tap_flow_implicit_flush(pmd, NULL);
301 /* Callback to handle the rx burst of packets to the correct interface and
302 * file descriptor(s) in a multi-queue setup.
305 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
307 struct rx_queue *rxq = queue;
309 unsigned long num_rx_bytes = 0;
310 uint32_t trigger = tap_trigger;
312 if (trigger == rxq->trigger_seen)
315 rxq->trigger_seen = trigger;
316 rte_compiler_barrier();
317 for (num_rx = 0; num_rx < nb_pkts; ) {
318 struct rte_mbuf *mbuf = rxq->pool;
319 struct rte_mbuf *seg = NULL;
320 struct rte_mbuf *new_tail = NULL;
321 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
324 len = readv(rxq->fd, *rxq->iovecs,
325 1 + (rxq->rxmode->enable_scatter ?
326 rxq->nb_rx_desc : 1));
327 if (len < (int)sizeof(struct tun_pi))
330 /* Packet couldn't fit in the provided mbuf */
331 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
332 rxq->stats.ierrors++;
336 len -= sizeof(struct tun_pi);
339 mbuf->port = rxq->in_port;
341 struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
343 if (unlikely(!buf)) {
344 rxq->stats.rx_nombuf++;
345 /* No new buf has been allocated: do nothing */
346 if (!new_tail || !seg)
350 rte_pktmbuf_free(mbuf);
354 seg = seg ? seg->next : mbuf;
355 if (rxq->pool == mbuf)
358 new_tail->next = buf;
360 new_tail->next = seg->next;
362 /* iovecs[0] is reserved for packet info (pi) */
363 (*rxq->iovecs)[mbuf->nb_segs].iov_len =
364 buf->buf_len - data_off;
365 (*rxq->iovecs)[mbuf->nb_segs].iov_base =
366 (char *)buf->buf_addr + data_off;
368 seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
369 seg->data_off = data_off;
371 len -= seg->data_len;
375 /* First segment has headroom, not the others */
379 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
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;
393 /* Callback to handle sending packets from the tap interface
396 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
398 struct tx_queue *txq = queue;
400 unsigned long num_tx_bytes = 0;
404 if (unlikely(nb_pkts == 0))
407 max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
408 for (i = 0; i < nb_pkts; i++) {
409 struct rte_mbuf *mbuf = bufs[num_tx];
410 struct iovec iovecs[mbuf->nb_segs + 1];
411 struct tun_pi pi = { .flags = 0 };
412 struct rte_mbuf *seg = mbuf;
416 /* stats.errs will be incremented */
417 if (rte_pktmbuf_pkt_len(mbuf) > max_size)
420 iovecs[0].iov_base = π
421 iovecs[0].iov_len = sizeof(pi);
422 for (j = 1; j <= mbuf->nb_segs; j++) {
423 iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
425 rte_pktmbuf_mtod(seg, void *);
428 /* copy the tx frame data */
429 n = writev(txq->fd, iovecs, mbuf->nb_segs + 1);
434 num_tx_bytes += mbuf->pkt_len;
435 rte_pktmbuf_free(mbuf);
438 txq->stats.opackets += num_tx;
439 txq->stats.errs += nb_pkts - num_tx;
440 txq->stats.obytes += num_tx_bytes;
446 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
447 struct ifreq *ifr, int set, enum ioctl_mode mode)
449 short req_flags = ifr->ifr_flags;
450 int remote = pmd->remote_if_index &&
451 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
453 if (!pmd->remote_if_index && mode == REMOTE_ONLY)
456 * If there is a remote netdevice, apply ioctl on it, then apply it on
461 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
462 else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
463 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
466 /* fetch current flags to leave other flags untouched */
467 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
470 ifr->ifr_flags |= req_flags;
472 ifr->ifr_flags &= ~req_flags;
480 RTE_LOG(WARNING, PMD, "%s: ioctl() called with wrong arg\n",
484 if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
486 if (remote-- && mode == LOCAL_AND_REMOTE)
491 RTE_LOG(ERR, PMD, "%s: ioctl(%lu) failed with error: %s\n",
492 ifr->ifr_name, request, strerror(errno));
497 tap_link_set_down(struct rte_eth_dev *dev)
499 struct pmd_internals *pmd = dev->data->dev_private;
500 struct ifreq ifr = { .ifr_flags = IFF_UP };
502 dev->data->dev_link.link_status = ETH_LINK_DOWN;
503 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
507 tap_link_set_up(struct rte_eth_dev *dev)
509 struct pmd_internals *pmd = dev->data->dev_private;
510 struct ifreq ifr = { .ifr_flags = IFF_UP };
512 dev->data->dev_link.link_status = ETH_LINK_UP;
513 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
517 tap_dev_start(struct rte_eth_dev *dev)
521 err = tap_intr_handle_set(dev, 1);
524 return tap_link_set_up(dev);
527 /* This function gets called when the current port gets stopped.
530 tap_dev_stop(struct rte_eth_dev *dev)
532 tap_intr_handle_set(dev, 0);
533 tap_link_set_down(dev);
537 tap_dev_configure(struct rte_eth_dev *dev __rte_unused)
543 tap_dev_speed_capa(void)
545 uint32_t speed = pmd_link.link_speed;
548 if (speed >= ETH_SPEED_NUM_10M)
549 capa |= ETH_LINK_SPEED_10M;
550 if (speed >= ETH_SPEED_NUM_100M)
551 capa |= ETH_LINK_SPEED_100M;
552 if (speed >= ETH_SPEED_NUM_1G)
553 capa |= ETH_LINK_SPEED_1G;
554 if (speed >= ETH_SPEED_NUM_5G)
555 capa |= ETH_LINK_SPEED_2_5G;
556 if (speed >= ETH_SPEED_NUM_5G)
557 capa |= ETH_LINK_SPEED_5G;
558 if (speed >= ETH_SPEED_NUM_10G)
559 capa |= ETH_LINK_SPEED_10G;
560 if (speed >= ETH_SPEED_NUM_20G)
561 capa |= ETH_LINK_SPEED_20G;
562 if (speed >= ETH_SPEED_NUM_25G)
563 capa |= ETH_LINK_SPEED_25G;
564 if (speed >= ETH_SPEED_NUM_40G)
565 capa |= ETH_LINK_SPEED_40G;
566 if (speed >= ETH_SPEED_NUM_50G)
567 capa |= ETH_LINK_SPEED_50G;
568 if (speed >= ETH_SPEED_NUM_56G)
569 capa |= ETH_LINK_SPEED_56G;
570 if (speed >= ETH_SPEED_NUM_100G)
571 capa |= ETH_LINK_SPEED_100G;
577 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
579 struct pmd_internals *internals = dev->data->dev_private;
581 dev_info->if_index = internals->if_index;
582 dev_info->max_mac_addrs = 1;
583 dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
584 dev_info->max_rx_queues = internals->nb_queues;
585 dev_info->max_tx_queues = internals->nb_queues;
586 dev_info->min_rx_bufsize = 0;
587 dev_info->pci_dev = NULL;
588 dev_info->speed_capa = tap_dev_speed_capa();
592 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
594 unsigned int i, imax;
595 unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
596 unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
597 unsigned long rx_nombuf = 0, ierrors = 0;
598 const struct pmd_internals *pmd = dev->data->dev_private;
600 imax = (pmd->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
601 pmd->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
603 for (i = 0; i < imax; i++) {
604 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
605 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
606 rx_total += tap_stats->q_ipackets[i];
607 rx_bytes_total += tap_stats->q_ibytes[i];
608 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
609 ierrors += pmd->rxq[i].stats.ierrors;
611 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
612 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
613 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
614 tx_total += tap_stats->q_opackets[i];
615 tx_err_total += tap_stats->q_errors[i];
616 tx_bytes_total += tap_stats->q_obytes[i];
619 tap_stats->ipackets = rx_total;
620 tap_stats->ibytes = rx_bytes_total;
621 tap_stats->ierrors = ierrors;
622 tap_stats->rx_nombuf = rx_nombuf;
623 tap_stats->opackets = tx_total;
624 tap_stats->oerrors = tx_err_total;
625 tap_stats->obytes = tx_bytes_total;
629 tap_stats_reset(struct rte_eth_dev *dev)
632 struct pmd_internals *pmd = dev->data->dev_private;
634 for (i = 0; i < pmd->nb_queues; i++) {
635 pmd->rxq[i].stats.ipackets = 0;
636 pmd->rxq[i].stats.ibytes = 0;
637 pmd->rxq[i].stats.ierrors = 0;
638 pmd->rxq[i].stats.rx_nombuf = 0;
640 pmd->txq[i].stats.opackets = 0;
641 pmd->txq[i].stats.errs = 0;
642 pmd->txq[i].stats.obytes = 0;
647 tap_dev_close(struct rte_eth_dev *dev __rte_unused)
650 struct pmd_internals *internals = dev->data->dev_private;
652 tap_link_set_down(dev);
653 tap_flow_flush(dev, NULL);
654 tap_flow_implicit_flush(internals, NULL);
656 for (i = 0; i < internals->nb_queues; i++) {
657 if (internals->rxq[i].fd != -1)
658 close(internals->rxq[i].fd);
659 internals->rxq[i].fd = -1;
660 internals->txq[i].fd = -1;
665 tap_rx_queue_release(void *queue)
667 struct rx_queue *rxq = queue;
669 if (rxq && (rxq->fd > 0)) {
672 rte_pktmbuf_free(rxq->pool);
673 rte_free(rxq->iovecs);
680 tap_tx_queue_release(void *queue)
682 struct tx_queue *txq = queue;
684 if (txq && (txq->fd > 0)) {
691 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
693 struct rte_eth_link *dev_link = &dev->data->dev_link;
694 struct pmd_internals *pmd = dev->data->dev_private;
695 struct ifreq ifr = { .ifr_flags = 0 };
697 if (pmd->remote_if_index) {
698 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
699 if (!(ifr.ifr_flags & IFF_UP) ||
700 !(ifr.ifr_flags & IFF_RUNNING)) {
701 dev_link->link_status = ETH_LINK_DOWN;
705 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
706 dev_link->link_status =
707 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
714 tap_promisc_enable(struct rte_eth_dev *dev)
716 struct pmd_internals *pmd = dev->data->dev_private;
717 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
719 dev->data->promiscuous = 1;
720 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
721 if (pmd->remote_if_index)
722 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
726 tap_promisc_disable(struct rte_eth_dev *dev)
728 struct pmd_internals *pmd = dev->data->dev_private;
729 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
731 dev->data->promiscuous = 0;
732 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
733 if (pmd->remote_if_index)
734 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
738 tap_allmulti_enable(struct rte_eth_dev *dev)
740 struct pmd_internals *pmd = dev->data->dev_private;
741 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
743 dev->data->all_multicast = 1;
744 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
745 if (pmd->remote_if_index)
746 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
750 tap_allmulti_disable(struct rte_eth_dev *dev)
752 struct pmd_internals *pmd = dev->data->dev_private;
753 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
755 dev->data->all_multicast = 0;
756 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
757 if (pmd->remote_if_index)
758 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
763 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
765 struct pmd_internals *pmd = dev->data->dev_private;
768 if (is_zero_ether_addr(mac_addr)) {
769 RTE_LOG(ERR, PMD, "%s: can't set an empty MAC address\n",
773 /* Check the actual current MAC address on the tap netdevice */
774 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY) != 0) {
776 "%s: couldn't check current tap MAC address\n",
780 if (is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
784 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
785 rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
786 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, LOCAL_AND_REMOTE) < 0)
788 rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
789 if (pmd->remote_if_index) {
790 /* Replace MAC redirection rule after a MAC change */
791 if (tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC) < 0) {
793 "%s: Couldn't delete MAC redirection rule\n",
797 if (tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0)
799 "%s: Couldn't add MAC redirection rule\n",
805 tap_setup_queue(struct rte_eth_dev *dev,
806 struct pmd_internals *internals,
809 struct pmd_internals *pmd = dev->data->dev_private;
810 struct rx_queue *rx = &internals->rxq[qid];
811 struct tx_queue *tx = &internals->txq[qid];
818 RTE_LOG(INFO, PMD, "Add queue to TAP %s for qid %d\n",
820 fd = tun_alloc(pmd, qid);
822 RTE_LOG(ERR, PMD, "tun_alloc(%s, %d) failed\n",
829 ifr.ifr_mtu = dev->data->mtu;
830 if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1,
831 LOCAL_AND_REMOTE) < 0) {
841 tx->mtu = &dev->data->mtu;
842 rx->rxmode = &dev->data->dev_conf.rxmode;
848 rx_setup_queue(struct rte_eth_dev *dev,
849 struct pmd_internals *internals,
852 dev->data->rx_queues[qid] = &internals->rxq[qid];
854 return tap_setup_queue(dev, internals, qid);
858 tx_setup_queue(struct rte_eth_dev *dev,
859 struct pmd_internals *internals,
862 dev->data->tx_queues[qid] = &internals->txq[qid];
864 return tap_setup_queue(dev, internals, qid);
868 tap_rx_queue_setup(struct rte_eth_dev *dev,
869 uint16_t rx_queue_id,
871 unsigned int socket_id,
872 const struct rte_eth_rxconf *rx_conf __rte_unused,
873 struct rte_mempool *mp)
875 struct pmd_internals *internals = dev->data->dev_private;
876 struct rx_queue *rxq = &internals->rxq[rx_queue_id];
877 struct rte_mbuf **tmp = &rxq->pool;
878 long iov_max = sysconf(_SC_IOV_MAX);
879 uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
880 struct iovec (*iovecs)[nb_desc + 1];
881 int data_off = RTE_PKTMBUF_HEADROOM;
886 if ((rx_queue_id >= internals->nb_queues) || !mp) {
887 RTE_LOG(WARNING, PMD,
888 "nb_queues %d too small or mempool NULL\n",
889 internals->nb_queues);
894 rxq->trigger_seen = 1; /* force initial burst */
895 rxq->in_port = dev->data->port_id;
896 rxq->nb_rx_desc = nb_desc;
897 iovecs = rte_zmalloc_socket(dev->data->name, sizeof(*iovecs), 0,
900 RTE_LOG(WARNING, PMD,
901 "%s: Couldn't allocate %d RX descriptors\n",
902 dev->data->name, nb_desc);
905 rxq->iovecs = iovecs;
907 fd = rx_setup_queue(dev, internals, rx_queue_id);
913 (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
914 (*rxq->iovecs)[0].iov_base = &rxq->pi;
916 for (i = 1; i <= nb_desc; i++) {
917 *tmp = rte_pktmbuf_alloc(rxq->mp);
919 RTE_LOG(WARNING, PMD,
920 "%s: couldn't allocate memory for queue %d\n",
921 dev->data->name, rx_queue_id);
925 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
926 (*rxq->iovecs)[i].iov_base =
927 (char *)(*tmp)->buf_addr + data_off;
932 RTE_LOG(DEBUG, PMD, " RX TAP device name %s, qid %d on fd %d\n",
933 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
938 rte_pktmbuf_free(rxq->pool);
940 rte_free(rxq->iovecs);
946 tap_tx_queue_setup(struct rte_eth_dev *dev,
947 uint16_t tx_queue_id,
948 uint16_t nb_tx_desc __rte_unused,
949 unsigned int socket_id __rte_unused,
950 const struct rte_eth_txconf *tx_conf __rte_unused)
952 struct pmd_internals *internals = dev->data->dev_private;
955 if (tx_queue_id >= internals->nb_queues)
958 ret = tx_setup_queue(dev, internals, tx_queue_id);
962 RTE_LOG(DEBUG, PMD, " TX TAP device name %s, qid %d on fd %d\n",
963 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd);
969 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
971 struct pmd_internals *pmd = dev->data->dev_private;
972 struct ifreq ifr = { .ifr_mtu = mtu };
975 err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
977 dev->data->mtu = mtu;
983 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
984 struct ether_addr *mc_addr_set __rte_unused,
985 uint32_t nb_mc_addr __rte_unused)
988 * Nothing to do actually: the tap has no filtering whatsoever, every
989 * packet is received.
995 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
997 struct rte_eth_dev *dev = arg;
998 struct pmd_internals *pmd = dev->data->dev_private;
999 struct ifinfomsg *info = NLMSG_DATA(nh);
1001 if (nh->nlmsg_type != RTM_NEWLINK ||
1002 (info->ifi_index != pmd->if_index &&
1003 info->ifi_index != pmd->remote_if_index))
1005 return tap_link_update(dev, 0);
1009 tap_dev_intr_handler(void *cb_arg)
1011 struct rte_eth_dev *dev = cb_arg;
1012 struct pmd_internals *pmd = dev->data->dev_private;
1014 nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1018 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1020 struct pmd_internals *pmd = dev->data->dev_private;
1022 /* In any case, disable interrupt if the conf is no longer there. */
1023 if (!dev->data->dev_conf.intr_conf.lsc) {
1024 if (pmd->intr_handle.fd != -1)
1025 nl_final(pmd->intr_handle.fd);
1026 rte_intr_callback_unregister(
1027 &pmd->intr_handle, tap_dev_intr_handler, dev);
1031 pmd->intr_handle.fd = nl_init(RTMGRP_LINK);
1032 if (unlikely(pmd->intr_handle.fd == -1))
1034 return rte_intr_callback_register(
1035 &pmd->intr_handle, tap_dev_intr_handler, dev);
1037 nl_final(pmd->intr_handle.fd);
1038 return rte_intr_callback_unregister(&pmd->intr_handle,
1039 tap_dev_intr_handler, dev);
1042 static const uint32_t*
1043 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1045 static const uint32_t ptypes[] = {
1046 RTE_PTYPE_INNER_L2_ETHER,
1047 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1048 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1049 RTE_PTYPE_INNER_L3_IPV4,
1050 RTE_PTYPE_INNER_L3_IPV4_EXT,
1051 RTE_PTYPE_INNER_L3_IPV6,
1052 RTE_PTYPE_INNER_L3_IPV6_EXT,
1053 RTE_PTYPE_INNER_L4_FRAG,
1054 RTE_PTYPE_INNER_L4_UDP,
1055 RTE_PTYPE_INNER_L4_TCP,
1056 RTE_PTYPE_INNER_L4_SCTP,
1058 RTE_PTYPE_L2_ETHER_VLAN,
1059 RTE_PTYPE_L2_ETHER_QINQ,
1061 RTE_PTYPE_L3_IPV4_EXT,
1062 RTE_PTYPE_L3_IPV6_EXT,
1074 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1075 struct rte_eth_fc_conf *fc_conf)
1077 fc_conf->mode = RTE_FC_NONE;
1082 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1083 struct rte_eth_fc_conf *fc_conf)
1085 if (fc_conf->mode != RTE_FC_NONE)
1090 static const struct eth_dev_ops ops = {
1091 .dev_start = tap_dev_start,
1092 .dev_stop = tap_dev_stop,
1093 .dev_close = tap_dev_close,
1094 .dev_configure = tap_dev_configure,
1095 .dev_infos_get = tap_dev_info,
1096 .rx_queue_setup = tap_rx_queue_setup,
1097 .tx_queue_setup = tap_tx_queue_setup,
1098 .rx_queue_release = tap_rx_queue_release,
1099 .tx_queue_release = tap_tx_queue_release,
1100 .flow_ctrl_get = tap_flow_ctrl_get,
1101 .flow_ctrl_set = tap_flow_ctrl_set,
1102 .link_update = tap_link_update,
1103 .dev_set_link_up = tap_link_set_up,
1104 .dev_set_link_down = tap_link_set_down,
1105 .promiscuous_enable = tap_promisc_enable,
1106 .promiscuous_disable = tap_promisc_disable,
1107 .allmulticast_enable = tap_allmulti_enable,
1108 .allmulticast_disable = tap_allmulti_disable,
1109 .mac_addr_set = tap_mac_set,
1110 .mtu_set = tap_mtu_set,
1111 .set_mc_addr_list = tap_set_mc_addr_list,
1112 .stats_get = tap_stats_get,
1113 .stats_reset = tap_stats_reset,
1114 .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1115 .filter_ctrl = tap_dev_filter_ctrl,
1119 tap_kernel_support(struct pmd_internals *pmd)
1121 struct utsname utsname;
1124 if (uname(&utsname) == -1 ||
1125 sscanf(utsname.release, "%d.%d.%d",
1126 &ver[0], &ver[1], &ver[2]) != 3)
1128 if (KERNEL_VERSION(ver[0], ver[1], ver[2]) >= FLOWER_KERNEL_VERSION)
1129 pmd->flower_support = 1;
1130 if (KERNEL_VERSION(ver[0], ver[1], ver[2]) >=
1131 FLOWER_VLAN_KERNEL_VERSION)
1132 pmd->flower_vlan_support = 1;
1137 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
1140 int numa_node = rte_socket_id();
1141 struct rte_eth_dev *dev;
1142 struct pmd_internals *pmd;
1143 struct rte_eth_dev_data *data;
1146 RTE_LOG(DEBUG, PMD, " TAP device on numa %u\n", rte_socket_id());
1148 data = rte_zmalloc_socket(tap_name, sizeof(*data), 0, numa_node);
1150 RTE_LOG(ERR, PMD, "TAP Failed to allocate data\n");
1154 dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1156 RTE_LOG(ERR, PMD, "TAP Unable to allocate device struct\n");
1160 pmd = dev->data->dev_private;
1161 snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1162 pmd->nb_queues = RTE_PMD_TAP_MAX_QUEUES;
1164 pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1165 if (pmd->ioctl_sock == -1) {
1167 "TAP Unable to get a socket for management: %s\n",
1172 /* Setup some default values */
1173 rte_memcpy(data, dev->data, sizeof(*data));
1174 data->dev_private = pmd;
1175 data->dev_flags = RTE_ETH_DEV_DETACHABLE | RTE_ETH_DEV_INTR_LSC;
1176 data->numa_node = numa_node;
1177 data->drv_name = pmd_tap_drv.driver.name;
1179 data->dev_link = pmd_link;
1180 data->mac_addrs = &pmd->eth_addr;
1181 data->nb_rx_queues = pmd->nb_queues;
1182 data->nb_tx_queues = pmd->nb_queues;
1185 dev->dev_ops = &ops;
1186 dev->rx_pkt_burst = pmd_rx_burst;
1187 dev->tx_pkt_burst = pmd_tx_burst;
1189 pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1190 pmd->intr_handle.fd = -1;
1192 /* Presetup the fds to -1 as being not valid */
1193 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1194 pmd->rxq[i].fd = -1;
1195 pmd->txq[i].fd = -1;
1198 tap_kernel_support(pmd);
1199 if (!pmd->flower_support)
1201 LIST_INIT(&pmd->flows);
1203 * If no netlink socket can be created, then it will fail when
1204 * creating/destroying flow rules.
1206 pmd->nlsk_fd = nl_init(0);
1207 if (strlen(remote_iface)) {
1210 pmd->remote_if_index = if_nametoindex(remote_iface);
1211 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1212 "%s", remote_iface);
1213 if (!pmd->remote_if_index) {
1214 RTE_LOG(ERR, PMD, "Could not find %s ifindex: "
1215 "remote interface will remain unconfigured\n",
1219 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1220 RTE_LOG(ERR, PMD, "Could not get remote MAC address\n");
1223 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1226 eth_random_addr((uint8_t *)&pmd->eth_addr);
1232 RTE_LOG(DEBUG, PMD, "TAP Unable to initialize %s\n",
1233 rte_vdev_device_name(vdev));
1240 set_interface_name(const char *key __rte_unused,
1244 char *name = (char *)extra_args;
1247 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s", value);
1249 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1250 DEFAULT_TAP_NAME, (tap_unit - 1));
1256 set_interface_speed(const char *key __rte_unused,
1260 *(int *)extra_args = (value) ? atoi(value) : ETH_SPEED_NUM_10G;
1266 set_remote_iface(const char *key __rte_unused,
1270 char *name = (char *)extra_args;
1273 snprintf(name, RTE_ETH_NAME_MAX_LEN, "%s", value);
1278 /* Open a TAP interface device.
1281 rte_pmd_tap_probe(struct rte_vdev_device *dev)
1283 const char *name, *params;
1285 struct rte_kvargs *kvlist = NULL;
1287 char tap_name[RTE_ETH_NAME_MAX_LEN];
1288 char remote_iface[RTE_ETH_NAME_MAX_LEN];
1290 name = rte_vdev_device_name(dev);
1291 params = rte_vdev_device_args(dev);
1293 speed = ETH_SPEED_NUM_10G;
1294 snprintf(tap_name, sizeof(tap_name), "%s%d",
1295 DEFAULT_TAP_NAME, tap_unit++);
1296 memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1298 if (params && (params[0] != '\0')) {
1299 RTE_LOG(DEBUG, PMD, "paramaters (%s)\n", params);
1301 kvlist = rte_kvargs_parse(params, valid_arguments);
1303 if (rte_kvargs_count(kvlist, ETH_TAP_SPEED_ARG) == 1) {
1304 ret = rte_kvargs_process(kvlist,
1306 &set_interface_speed,
1312 if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1313 ret = rte_kvargs_process(kvlist,
1315 &set_interface_name,
1321 if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
1322 ret = rte_kvargs_process(kvlist,
1331 pmd_link.link_speed = speed;
1333 RTE_LOG(NOTICE, PMD, "Initializing pmd_tap for %s as %s\n",
1336 ret = eth_dev_tap_create(dev, tap_name, remote_iface);
1340 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
1342 tap_unit--; /* Restore the unit number */
1344 rte_kvargs_free(kvlist);
1349 /* detach a TAP device.
1352 rte_pmd_tap_remove(struct rte_vdev_device *dev)
1354 struct rte_eth_dev *eth_dev = NULL;
1355 struct pmd_internals *internals;
1358 RTE_LOG(DEBUG, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
1361 /* find the ethdev entry */
1362 eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1366 internals = eth_dev->data->dev_private;
1367 if (internals->flower_support && internals->nlsk_fd) {
1368 tap_flow_flush(eth_dev, NULL);
1369 tap_flow_implicit_flush(internals, NULL);
1370 nl_final(internals->nlsk_fd);
1372 for (i = 0; i < internals->nb_queues; i++)
1373 if (internals->rxq[i].fd != -1)
1374 close(internals->rxq[i].fd);
1376 close(internals->ioctl_sock);
1377 rte_free(eth_dev->data->dev_private);
1378 rte_free(eth_dev->data);
1380 rte_eth_dev_release_port(eth_dev);
1385 static struct rte_vdev_driver pmd_tap_drv = {
1386 .probe = rte_pmd_tap_probe,
1387 .remove = rte_pmd_tap_remove,
1389 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1390 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1391 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
1392 ETH_TAP_IFACE_ARG "=<string> "
1393 ETH_TAP_SPEED_ARG "=<int> "
1394 ETH_TAP_REMOTE_ARG "=<string>");