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 #ifdef IFF_MULTI_QUEUE
76 #define RTE_PMD_TAP_MAX_QUEUES 16
78 #define RTE_PMD_TAP_MAX_QUEUES 1
81 #define FLOWER_KERNEL_VERSION KERNEL_VERSION(4, 2, 0)
82 #define FLOWER_VLAN_KERNEL_VERSION KERNEL_VERSION(4, 9, 0)
84 static struct rte_vdev_driver pmd_tap_drv;
86 static const char *valid_arguments[] = {
95 static volatile uint32_t tap_trigger; /* Rx trigger */
97 static struct rte_eth_link pmd_link = {
98 .link_speed = ETH_SPEED_NUM_10G,
99 .link_duplex = ETH_LINK_FULL_DUPLEX,
100 .link_status = ETH_LINK_DOWN,
101 .link_autoneg = ETH_LINK_SPEED_AUTONEG
105 tap_trigger_cb(int sig __rte_unused)
107 /* Valid trigger values are nonzero */
108 tap_trigger = (tap_trigger + 1) | 0x80000000;
112 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
113 struct ifreq *ifr, int set);
115 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
117 /* Tun/Tap allocation routine
119 * name is the number of the interface to use, unless NULL to take the host
123 tun_alloc(struct pmd_internals *pmd, uint16_t qid)
126 #ifdef IFF_MULTI_QUEUE
127 unsigned int features;
131 memset(&ifr, 0, sizeof(struct ifreq));
134 * Do not set IFF_NO_PI as packet information header will be needed
135 * to check if a received packet has been truncated.
137 ifr.ifr_flags = IFF_TAP;
138 snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
140 RTE_LOG(DEBUG, PMD, "ifr_name '%s'\n", ifr.ifr_name);
142 fd = open(TUN_TAP_DEV_PATH, O_RDWR);
144 RTE_LOG(ERR, PMD, "Unable to create TAP interface");
148 #ifdef IFF_MULTI_QUEUE
149 /* Grab the TUN features to verify we can work multi-queue */
150 if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
151 RTE_LOG(ERR, PMD, "TAP unable to get TUN/TAP features\n");
154 RTE_LOG(DEBUG, PMD, " TAP Features %08x\n", features);
156 if (features & IFF_MULTI_QUEUE) {
157 RTE_LOG(DEBUG, PMD, " Multi-queue support for %d queues\n",
158 RTE_PMD_TAP_MAX_QUEUES);
159 ifr.ifr_flags |= IFF_MULTI_QUEUE;
163 ifr.ifr_flags |= IFF_ONE_QUEUE;
164 RTE_LOG(DEBUG, PMD, " Single queue only support\n");
167 /* Set the TUN/TAP configuration and set the name if needed */
168 if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
169 RTE_LOG(WARNING, PMD,
170 "Unable to set TUNSETIFF for %s\n",
176 /* Always set the file descriptor to non-blocking */
177 if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
178 RTE_LOG(WARNING, PMD,
179 "Unable to set %s to nonblocking\n",
181 perror("F_SETFL, NONBLOCK");
185 /* Set up trigger to optimize empty Rx bursts */
189 int flags = fcntl(fd, F_GETFL);
191 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
193 if (sa.sa_handler != tap_trigger_cb) {
195 * Make sure SIGIO is not already taken. This is done
196 * as late as possible to leave the application a
197 * chance to set up its own signal handler first.
199 if (sa.sa_handler != SIG_IGN &&
200 sa.sa_handler != SIG_DFL) {
204 sa = (struct sigaction){
205 .sa_flags = SA_RESTART,
206 .sa_handler = tap_trigger_cb,
208 if (sigaction(SIGIO, &sa, NULL) == -1)
211 /* Enable SIGIO on file descriptor */
212 fcntl(fd, F_SETFL, flags | O_ASYNC);
213 fcntl(fd, F_SETOWN, getpid());
216 /* Disable trigger globally in case of error */
218 RTE_LOG(WARNING, PMD, "Rx trigger disabled: %s\n",
225 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0) < 0)
227 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
230 pmd->if_index = if_nametoindex(pmd->name);
231 if (!pmd->if_index) {
233 "Could not find ifindex for %s: rte_flow won't be usable.\n",
237 if (!pmd->flower_support)
239 if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
241 "Could not create multiq qdisc for %s: rte_flow won't be usable.\n",
245 if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
247 "Could not create multiq qdisc for %s: rte_flow won't be usable.\n",
251 if (pmd->remote_if_index) {
253 * Flush usually returns negative value because it tries
254 * to delete every QDISC (and on a running device, one
255 * QDISC at least is needed). Ignore negative return
258 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
259 if (qdisc_create_ingress(pmd->nlsk_fd,
260 pmd->remote_if_index) < 0)
262 LIST_INIT(&pmd->implicit_flows);
263 if (tap_flow_implicit_create(
264 pmd, TAP_REMOTE_LOCAL_MAC) < 0)
266 if (tap_flow_implicit_create(
267 pmd, TAP_REMOTE_BROADCAST) < 0)
269 if (tap_flow_implicit_create(
270 pmd, TAP_REMOTE_BROADCASTV6) < 0)
272 if (tap_flow_implicit_create(
273 pmd, TAP_REMOTE_TX) < 0)
282 "Could not set up remote flow rules for %s: remote disabled.\n",
284 pmd->remote_if_index = 0;
285 tap_flow_implicit_flush(pmd, NULL);
294 /* Callback to handle the rx burst of packets to the correct interface and
295 * file descriptor(s) in a multi-queue setup.
298 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
300 struct rx_queue *rxq = queue;
302 unsigned long num_rx_bytes = 0;
303 uint32_t trigger = tap_trigger;
305 if (trigger == rxq->trigger_seen)
308 rxq->trigger_seen = trigger;
309 rte_compiler_barrier();
310 for (num_rx = 0; num_rx < nb_pkts; ) {
311 struct rte_mbuf *mbuf = rxq->pool;
312 struct rte_mbuf *seg = NULL;
313 struct rte_mbuf *new_tail = NULL;
314 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
317 len = readv(rxq->fd, *rxq->iovecs,
318 1 + (rxq->rxmode->enable_scatter ?
319 rxq->nb_rx_desc : 1));
320 if (len < (int)sizeof(struct tun_pi))
323 /* Packet couldn't fit in the provided mbuf */
324 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
325 rxq->stats.ierrors++;
329 len -= sizeof(struct tun_pi);
332 mbuf->port = rxq->in_port;
334 struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
336 if (unlikely(!buf)) {
337 rxq->stats.rx_nombuf++;
338 /* No new buf has been allocated: do nothing */
339 if (!new_tail || !seg)
343 rte_pktmbuf_free(mbuf);
347 seg = seg ? seg->next : mbuf;
348 if (rxq->pool == mbuf)
351 new_tail->next = buf;
353 new_tail->next = seg->next;
355 /* iovecs[0] is reserved for packet info (pi) */
356 (*rxq->iovecs)[mbuf->nb_segs].iov_len =
357 buf->buf_len - data_off;
358 (*rxq->iovecs)[mbuf->nb_segs].iov_base =
359 (char *)buf->buf_addr + data_off;
361 seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
362 seg->data_off = data_off;
364 len -= seg->data_len;
368 /* First segment has headroom, not the others */
372 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
375 /* account for the receive frame */
376 bufs[num_rx++] = mbuf;
377 num_rx_bytes += mbuf->pkt_len;
380 rxq->stats.ipackets += num_rx;
381 rxq->stats.ibytes += num_rx_bytes;
386 /* Callback to handle sending packets from the tap interface
389 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
391 struct tx_queue *txq = queue;
393 unsigned long num_tx_bytes = 0;
397 if (unlikely(nb_pkts == 0))
400 max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
401 for (i = 0; i < nb_pkts; i++) {
402 struct rte_mbuf *mbuf = bufs[num_tx];
403 struct iovec iovecs[mbuf->nb_segs + 1];
404 struct tun_pi pi = { .flags = 0 };
405 struct rte_mbuf *seg = mbuf;
409 /* stats.errs will be incremented */
410 if (rte_pktmbuf_pkt_len(mbuf) > max_size)
413 iovecs[0].iov_base = π
414 iovecs[0].iov_len = sizeof(pi);
415 for (j = 1; j <= mbuf->nb_segs; j++) {
416 iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
418 rte_pktmbuf_mtod(seg, void *);
421 /* copy the tx frame data */
422 n = writev(txq->fd, iovecs, mbuf->nb_segs + 1);
427 num_tx_bytes += mbuf->pkt_len;
428 rte_pktmbuf_free(mbuf);
431 txq->stats.opackets += num_tx;
432 txq->stats.errs += nb_pkts - num_tx;
433 txq->stats.obytes += num_tx_bytes;
439 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
440 struct ifreq *ifr, int set)
442 short req_flags = ifr->ifr_flags;
443 int remote = !!pmd->remote_if_index;
446 * If there is a remote netdevice, apply ioctl on it, then apply it on
449 if (request == SIOCGIFFLAGS && !set) {
451 * Special case for getting flags. If set is given,
452 * then return the flags from the remote netdevice only.
453 * Otherwise return the flags from the tap netdevice.
459 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
461 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
464 /* fetch current flags to leave other flags untouched */
465 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
468 ifr->ifr_flags |= req_flags;
470 ifr->ifr_flags &= ~req_flags;
474 remote = 0; /* don't loop */
477 /* Set remote MAC on the tap netdevice */
478 if (!remote && pmd->remote_if_index) {
479 request = SIOCSIFHWADDR;
487 RTE_LOG(WARNING, PMD, "%s: ioctl() called with wrong arg\n",
491 if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
498 RTE_LOG(ERR, PMD, "%s: ioctl(%lu) failed with error: %s\n",
499 ifr->ifr_name, request, strerror(errno));
504 tap_link_set_down(struct rte_eth_dev *dev)
506 struct pmd_internals *pmd = dev->data->dev_private;
507 struct ifreq ifr = { .ifr_flags = IFF_UP };
509 dev->data->dev_link.link_status = ETH_LINK_DOWN;
510 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0);
514 tap_link_set_up(struct rte_eth_dev *dev)
516 struct pmd_internals *pmd = dev->data->dev_private;
517 struct ifreq ifr = { .ifr_flags = IFF_UP };
519 dev->data->dev_link.link_status = ETH_LINK_UP;
520 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1);
524 tap_dev_start(struct rte_eth_dev *dev)
528 err = tap_intr_handle_set(dev, 1);
531 return tap_link_set_up(dev);
534 /* This function gets called when the current port gets stopped.
537 tap_dev_stop(struct rte_eth_dev *dev)
539 tap_intr_handle_set(dev, 0);
540 tap_link_set_down(dev);
544 tap_dev_configure(struct rte_eth_dev *dev __rte_unused)
550 tap_dev_speed_capa(void)
552 uint32_t speed = pmd_link.link_speed;
555 if (speed >= ETH_SPEED_NUM_10M)
556 capa |= ETH_LINK_SPEED_10M;
557 if (speed >= ETH_SPEED_NUM_100M)
558 capa |= ETH_LINK_SPEED_100M;
559 if (speed >= ETH_SPEED_NUM_1G)
560 capa |= ETH_LINK_SPEED_1G;
561 if (speed >= ETH_SPEED_NUM_5G)
562 capa |= ETH_LINK_SPEED_2_5G;
563 if (speed >= ETH_SPEED_NUM_5G)
564 capa |= ETH_LINK_SPEED_5G;
565 if (speed >= ETH_SPEED_NUM_10G)
566 capa |= ETH_LINK_SPEED_10G;
567 if (speed >= ETH_SPEED_NUM_20G)
568 capa |= ETH_LINK_SPEED_20G;
569 if (speed >= ETH_SPEED_NUM_25G)
570 capa |= ETH_LINK_SPEED_25G;
571 if (speed >= ETH_SPEED_NUM_40G)
572 capa |= ETH_LINK_SPEED_40G;
573 if (speed >= ETH_SPEED_NUM_50G)
574 capa |= ETH_LINK_SPEED_50G;
575 if (speed >= ETH_SPEED_NUM_56G)
576 capa |= ETH_LINK_SPEED_56G;
577 if (speed >= ETH_SPEED_NUM_100G)
578 capa |= ETH_LINK_SPEED_100G;
584 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
586 struct pmd_internals *internals = dev->data->dev_private;
588 dev_info->if_index = internals->if_index;
589 dev_info->max_mac_addrs = 1;
590 dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
591 dev_info->max_rx_queues = internals->nb_queues;
592 dev_info->max_tx_queues = internals->nb_queues;
593 dev_info->min_rx_bufsize = 0;
594 dev_info->pci_dev = NULL;
595 dev_info->speed_capa = tap_dev_speed_capa();
599 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
601 unsigned int i, imax;
602 unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
603 unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
604 unsigned long rx_nombuf = 0, ierrors = 0;
605 const struct pmd_internals *pmd = dev->data->dev_private;
607 imax = (pmd->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
608 pmd->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
610 for (i = 0; i < imax; i++) {
611 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
612 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
613 rx_total += tap_stats->q_ipackets[i];
614 rx_bytes_total += tap_stats->q_ibytes[i];
615 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
616 ierrors += pmd->rxq[i].stats.ierrors;
618 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
619 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
620 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
621 tx_total += tap_stats->q_opackets[i];
622 tx_err_total += tap_stats->q_errors[i];
623 tx_bytes_total += tap_stats->q_obytes[i];
626 tap_stats->ipackets = rx_total;
627 tap_stats->ibytes = rx_bytes_total;
628 tap_stats->ierrors = ierrors;
629 tap_stats->rx_nombuf = rx_nombuf;
630 tap_stats->opackets = tx_total;
631 tap_stats->oerrors = tx_err_total;
632 tap_stats->obytes = tx_bytes_total;
636 tap_stats_reset(struct rte_eth_dev *dev)
639 struct pmd_internals *pmd = dev->data->dev_private;
641 for (i = 0; i < pmd->nb_queues; i++) {
642 pmd->rxq[i].stats.ipackets = 0;
643 pmd->rxq[i].stats.ibytes = 0;
644 pmd->rxq[i].stats.ierrors = 0;
645 pmd->rxq[i].stats.rx_nombuf = 0;
647 pmd->txq[i].stats.opackets = 0;
648 pmd->txq[i].stats.errs = 0;
649 pmd->txq[i].stats.obytes = 0;
654 tap_dev_close(struct rte_eth_dev *dev __rte_unused)
657 struct pmd_internals *internals = dev->data->dev_private;
659 tap_link_set_down(dev);
660 tap_flow_flush(dev, NULL);
661 tap_flow_implicit_flush(internals, NULL);
663 for (i = 0; i < internals->nb_queues; i++) {
664 if (internals->rxq[i].fd != -1)
665 close(internals->rxq[i].fd);
666 internals->rxq[i].fd = -1;
667 internals->txq[i].fd = -1;
672 tap_rx_queue_release(void *queue)
674 struct rx_queue *rxq = queue;
676 if (rxq && (rxq->fd > 0)) {
679 rte_pktmbuf_free(rxq->pool);
680 rte_free(rxq->iovecs);
687 tap_tx_queue_release(void *queue)
689 struct tx_queue *txq = queue;
691 if (txq && (txq->fd > 0)) {
698 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
700 struct rte_eth_link *dev_link = &dev->data->dev_link;
701 struct pmd_internals *pmd = dev->data->dev_private;
702 struct ifreq ifr = { .ifr_flags = 0 };
704 if (pmd->remote_if_index) {
705 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 1);
706 if (!(ifr.ifr_flags & IFF_UP) ||
707 !(ifr.ifr_flags & IFF_RUNNING)) {
708 dev_link->link_status = ETH_LINK_DOWN;
712 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0);
713 dev_link->link_status =
714 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
721 tap_promisc_enable(struct rte_eth_dev *dev)
723 struct pmd_internals *pmd = dev->data->dev_private;
724 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
726 dev->data->promiscuous = 1;
727 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1);
728 if (pmd->remote_if_index)
729 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
733 tap_promisc_disable(struct rte_eth_dev *dev)
735 struct pmd_internals *pmd = dev->data->dev_private;
736 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
738 dev->data->promiscuous = 0;
739 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0);
740 if (pmd->remote_if_index)
741 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
745 tap_allmulti_enable(struct rte_eth_dev *dev)
747 struct pmd_internals *pmd = dev->data->dev_private;
748 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
750 dev->data->all_multicast = 1;
751 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1);
752 if (pmd->remote_if_index)
753 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
757 tap_allmulti_disable(struct rte_eth_dev *dev)
759 struct pmd_internals *pmd = dev->data->dev_private;
760 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
762 dev->data->all_multicast = 0;
763 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0);
764 if (pmd->remote_if_index)
765 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
770 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
772 struct pmd_internals *pmd = dev->data->dev_private;
775 if (is_zero_ether_addr(mac_addr)) {
776 RTE_LOG(ERR, PMD, "%s: can't set an empty MAC address\n",
781 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
782 rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
783 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1) < 0)
785 rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
789 tap_setup_queue(struct rte_eth_dev *dev,
790 struct pmd_internals *internals,
793 struct pmd_internals *pmd = dev->data->dev_private;
794 struct rx_queue *rx = &internals->rxq[qid];
795 struct tx_queue *tx = &internals->txq[qid];
802 RTE_LOG(INFO, PMD, "Add queue to TAP %s for qid %d\n",
804 fd = tun_alloc(pmd, qid);
806 RTE_LOG(ERR, PMD, "tun_alloc(%s, %d) failed\n",
813 ifr.ifr_mtu = dev->data->mtu;
814 if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1) < 0) {
824 tx->mtu = &dev->data->mtu;
825 rx->rxmode = &dev->data->dev_conf.rxmode;
831 rx_setup_queue(struct rte_eth_dev *dev,
832 struct pmd_internals *internals,
835 dev->data->rx_queues[qid] = &internals->rxq[qid];
837 return tap_setup_queue(dev, internals, qid);
841 tx_setup_queue(struct rte_eth_dev *dev,
842 struct pmd_internals *internals,
845 dev->data->tx_queues[qid] = &internals->txq[qid];
847 return tap_setup_queue(dev, internals, qid);
851 tap_rx_queue_setup(struct rte_eth_dev *dev,
852 uint16_t rx_queue_id,
854 unsigned int socket_id,
855 const struct rte_eth_rxconf *rx_conf __rte_unused,
856 struct rte_mempool *mp)
858 struct pmd_internals *internals = dev->data->dev_private;
859 struct rx_queue *rxq = &internals->rxq[rx_queue_id];
860 struct rte_mbuf **tmp = &rxq->pool;
861 struct iovec (*iovecs)[nb_rx_desc + 1];
862 int data_off = RTE_PKTMBUF_HEADROOM;
868 if ((rx_queue_id >= internals->nb_queues) || !mp) {
869 RTE_LOG(WARNING, PMD,
870 "nb_queues %d too small or mempool NULL\n",
871 internals->nb_queues);
876 rxq->trigger_seen = 1; /* force initial burst */
877 rxq->in_port = dev->data->port_id;
878 rxq->nb_rx_desc = nb_rx_desc;
879 iovecs = rte_zmalloc_socket(dev->data->name, sizeof(*iovecs), 0,
882 RTE_LOG(WARNING, PMD,
883 "%s: Couldn't allocate %d RX descriptors\n",
884 dev->data->name, nb_rx_desc);
887 rxq->iovecs = iovecs;
889 /* Now get the space available for data in the mbuf */
890 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(mp) -
891 RTE_PKTMBUF_HEADROOM);
893 if (buf_size < ETH_FRAME_LEN) {
894 RTE_LOG(WARNING, PMD,
895 "%s: %d bytes will not fit in mbuf (%d bytes)\n",
896 dev->data->name, ETH_FRAME_LEN, buf_size);
901 fd = rx_setup_queue(dev, internals, rx_queue_id);
907 (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
908 (*rxq->iovecs)[0].iov_base = &rxq->pi;
910 for (i = 1; i <= nb_rx_desc; i++) {
911 *tmp = rte_pktmbuf_alloc(rxq->mp);
913 RTE_LOG(WARNING, PMD,
914 "%s: couldn't allocate memory for queue %d\n",
915 dev->data->name, rx_queue_id);
919 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
920 (*rxq->iovecs)[i].iov_base =
921 (char *)(*tmp)->buf_addr + data_off;
926 RTE_LOG(DEBUG, PMD, " RX TAP device name %s, qid %d on fd %d\n",
927 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
932 rte_pktmbuf_free(rxq->pool);
934 rte_free(rxq->iovecs);
940 tap_tx_queue_setup(struct rte_eth_dev *dev,
941 uint16_t tx_queue_id,
942 uint16_t nb_tx_desc __rte_unused,
943 unsigned int socket_id __rte_unused,
944 const struct rte_eth_txconf *tx_conf __rte_unused)
946 struct pmd_internals *internals = dev->data->dev_private;
949 if (tx_queue_id >= internals->nb_queues)
952 ret = tx_setup_queue(dev, internals, tx_queue_id);
956 RTE_LOG(DEBUG, PMD, " TX TAP device name %s, qid %d on fd %d\n",
957 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd);
963 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
965 struct pmd_internals *pmd = dev->data->dev_private;
966 struct ifreq ifr = { .ifr_mtu = mtu };
969 err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1);
971 dev->data->mtu = mtu;
977 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
978 struct ether_addr *mc_addr_set __rte_unused,
979 uint32_t nb_mc_addr __rte_unused)
982 * Nothing to do actually: the tap has no filtering whatsoever, every
983 * packet is received.
989 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
991 struct rte_eth_dev *dev = arg;
992 struct pmd_internals *pmd = dev->data->dev_private;
993 struct ifinfomsg *info = NLMSG_DATA(nh);
995 if (nh->nlmsg_type != RTM_NEWLINK ||
996 (info->ifi_index != pmd->if_index &&
997 info->ifi_index != pmd->remote_if_index))
999 return tap_link_update(dev, 0);
1003 tap_dev_intr_handler(struct rte_intr_handle *intr_handle __rte_unused,
1006 struct rte_eth_dev *dev = cb_arg;
1007 struct pmd_internals *pmd = dev->data->dev_private;
1009 nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1013 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1015 struct pmd_internals *pmd = dev->data->dev_private;
1017 /* In any case, disable interrupt if the conf is no longer there. */
1018 if (!dev->data->dev_conf.intr_conf.lsc) {
1019 if (pmd->intr_handle.fd != -1)
1020 nl_final(pmd->intr_handle.fd);
1021 rte_intr_callback_unregister(
1022 &pmd->intr_handle, tap_dev_intr_handler, dev);
1026 pmd->intr_handle.fd = nl_init(RTMGRP_LINK);
1027 if (unlikely(pmd->intr_handle.fd == -1))
1029 return rte_intr_callback_register(
1030 &pmd->intr_handle, tap_dev_intr_handler, dev);
1032 nl_final(pmd->intr_handle.fd);
1033 return rte_intr_callback_unregister(&pmd->intr_handle,
1034 tap_dev_intr_handler, dev);
1037 static const uint32_t*
1038 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1040 static const uint32_t ptypes[] = {
1041 RTE_PTYPE_INNER_L2_ETHER,
1042 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1043 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1044 RTE_PTYPE_INNER_L3_IPV4,
1045 RTE_PTYPE_INNER_L3_IPV4_EXT,
1046 RTE_PTYPE_INNER_L3_IPV6,
1047 RTE_PTYPE_INNER_L3_IPV6_EXT,
1048 RTE_PTYPE_INNER_L4_FRAG,
1049 RTE_PTYPE_INNER_L4_UDP,
1050 RTE_PTYPE_INNER_L4_TCP,
1051 RTE_PTYPE_INNER_L4_SCTP,
1053 RTE_PTYPE_L2_ETHER_VLAN,
1054 RTE_PTYPE_L2_ETHER_QINQ,
1056 RTE_PTYPE_L3_IPV4_EXT,
1057 RTE_PTYPE_L3_IPV6_EXT,
1069 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1070 struct rte_eth_fc_conf *fc_conf)
1072 fc_conf->mode = RTE_FC_NONE;
1077 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1078 struct rte_eth_fc_conf *fc_conf)
1080 if (fc_conf->mode != RTE_FC_NONE)
1085 static const struct eth_dev_ops ops = {
1086 .dev_start = tap_dev_start,
1087 .dev_stop = tap_dev_stop,
1088 .dev_close = tap_dev_close,
1089 .dev_configure = tap_dev_configure,
1090 .dev_infos_get = tap_dev_info,
1091 .rx_queue_setup = tap_rx_queue_setup,
1092 .tx_queue_setup = tap_tx_queue_setup,
1093 .rx_queue_release = tap_rx_queue_release,
1094 .tx_queue_release = tap_tx_queue_release,
1095 .flow_ctrl_get = tap_flow_ctrl_get,
1096 .flow_ctrl_set = tap_flow_ctrl_set,
1097 .link_update = tap_link_update,
1098 .dev_set_link_up = tap_link_set_up,
1099 .dev_set_link_down = tap_link_set_down,
1100 .promiscuous_enable = tap_promisc_enable,
1101 .promiscuous_disable = tap_promisc_disable,
1102 .allmulticast_enable = tap_allmulti_enable,
1103 .allmulticast_disable = tap_allmulti_disable,
1104 .mac_addr_set = tap_mac_set,
1105 .mtu_set = tap_mtu_set,
1106 .set_mc_addr_list = tap_set_mc_addr_list,
1107 .stats_get = tap_stats_get,
1108 .stats_reset = tap_stats_reset,
1109 .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1110 .filter_ctrl = tap_dev_filter_ctrl,
1114 tap_kernel_support(struct pmd_internals *pmd)
1116 struct utsname utsname;
1119 if (uname(&utsname) == -1 ||
1120 sscanf(utsname.release, "%d.%d.%d",
1121 &ver[0], &ver[1], &ver[2]) != 3)
1123 if (KERNEL_VERSION(ver[0], ver[1], ver[2]) >= FLOWER_KERNEL_VERSION)
1124 pmd->flower_support = 1;
1125 if (KERNEL_VERSION(ver[0], ver[1], ver[2]) >=
1126 FLOWER_VLAN_KERNEL_VERSION)
1127 pmd->flower_vlan_support = 1;
1132 eth_dev_tap_create(const char *name, char *tap_name, char *remote_iface)
1134 int numa_node = rte_socket_id();
1135 struct rte_eth_dev *dev = NULL;
1136 struct pmd_internals *pmd = NULL;
1137 struct rte_eth_dev_data *data = NULL;
1140 RTE_LOG(DEBUG, PMD, " TAP device on numa %u\n", rte_socket_id());
1142 data = rte_zmalloc_socket(tap_name, sizeof(*data), 0, numa_node);
1144 RTE_LOG(ERR, PMD, "TAP Failed to allocate data\n");
1148 pmd = rte_zmalloc_socket(tap_name, sizeof(*pmd), 0, numa_node);
1150 RTE_LOG(ERR, PMD, "TAP Unable to allocate internal struct\n");
1154 /* name in allocation and data->name must be consistent */
1155 snprintf(data->name, sizeof(data->name), "%s", name);
1156 dev = rte_eth_dev_allocate(name);
1158 RTE_LOG(ERR, PMD, "TAP Unable to allocate device struct\n");
1162 snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1164 pmd->nb_queues = RTE_PMD_TAP_MAX_QUEUES;
1166 pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1167 if (pmd->ioctl_sock == -1) {
1169 "TAP Unable to get a socket for management: %s\n",
1174 /* Setup some default values */
1175 data->dev_private = pmd;
1176 data->port_id = dev->data->port_id;
1177 data->mtu = dev->data->mtu;
1178 data->dev_flags = RTE_ETH_DEV_DETACHABLE | RTE_ETH_DEV_INTR_LSC;
1179 data->kdrv = RTE_KDRV_NONE;
1180 data->drv_name = pmd_tap_drv.driver.name;
1181 data->numa_node = numa_node;
1183 data->dev_link = pmd_link;
1184 data->mac_addrs = &pmd->eth_addr;
1185 data->nb_rx_queues = pmd->nb_queues;
1186 data->nb_tx_queues = pmd->nb_queues;
1189 dev->dev_ops = &ops;
1191 dev->rx_pkt_burst = pmd_rx_burst;
1192 dev->tx_pkt_burst = pmd_tx_burst;
1194 pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1195 pmd->intr_handle.fd = -1;
1197 /* Presetup the fds to -1 as being not valid */
1198 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1199 pmd->rxq[i].fd = -1;
1200 pmd->txq[i].fd = -1;
1203 tap_kernel_support(pmd);
1204 if (!pmd->flower_support)
1206 LIST_INIT(&pmd->flows);
1208 * If no netlink socket can be created, then it will fail when
1209 * creating/destroying flow rules.
1211 pmd->nlsk_fd = nl_init(0);
1212 if (strlen(remote_iface)) {
1213 pmd->remote_if_index = if_nametoindex(remote_iface);
1214 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1215 "%s", remote_iface);
1216 if (!pmd->remote_if_index)
1217 RTE_LOG(ERR, PMD, "Could not find %s ifindex: "
1218 "remote interface will remain unconfigured\n",
1225 RTE_LOG(DEBUG, PMD, "TAP Unable to initialize %s\n", name);
1230 rte_eth_dev_release_port(dev);
1236 set_interface_name(const char *key __rte_unused,
1240 char *name = (char *)extra_args;
1243 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s", value);
1245 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1246 DEFAULT_TAP_NAME, (tap_unit - 1));
1252 set_interface_speed(const char *key __rte_unused,
1256 *(int *)extra_args = (value) ? atoi(value) : ETH_SPEED_NUM_10G;
1262 set_remote_iface(const char *key __rte_unused,
1266 char *name = (char *)extra_args;
1269 snprintf(name, RTE_ETH_NAME_MAX_LEN, "%s", value);
1274 /* Open a TAP interface device.
1277 rte_pmd_tap_probe(const char *name, const char *params)
1280 struct rte_kvargs *kvlist = NULL;
1282 char tap_name[RTE_ETH_NAME_MAX_LEN];
1283 char remote_iface[RTE_ETH_NAME_MAX_LEN];
1285 speed = ETH_SPEED_NUM_10G;
1286 snprintf(tap_name, sizeof(tap_name), "%s%d",
1287 DEFAULT_TAP_NAME, tap_unit++);
1288 memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1290 if (params && (params[0] != '\0')) {
1291 RTE_LOG(DEBUG, PMD, "paramaters (%s)\n", params);
1293 kvlist = rte_kvargs_parse(params, valid_arguments);
1295 if (rte_kvargs_count(kvlist, ETH_TAP_SPEED_ARG) == 1) {
1296 ret = rte_kvargs_process(kvlist,
1298 &set_interface_speed,
1304 if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1305 ret = rte_kvargs_process(kvlist,
1307 &set_interface_name,
1313 if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
1314 ret = rte_kvargs_process(kvlist,
1323 pmd_link.link_speed = speed;
1325 RTE_LOG(NOTICE, PMD, "Initializing pmd_tap for %s as %s\n",
1328 ret = eth_dev_tap_create(name, tap_name, remote_iface);
1332 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
1334 tap_unit--; /* Restore the unit number */
1336 rte_kvargs_free(kvlist);
1341 /* detach a TAP device.
1344 rte_pmd_tap_remove(const char *name)
1346 struct rte_eth_dev *eth_dev = NULL;
1347 struct pmd_internals *internals;
1350 RTE_LOG(DEBUG, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
1353 /* find the ethdev entry */
1354 eth_dev = rte_eth_dev_allocated(name);
1358 internals = eth_dev->data->dev_private;
1359 if (internals->flower_support && internals->nlsk_fd) {
1360 tap_flow_flush(eth_dev, NULL);
1361 tap_flow_implicit_flush(internals, NULL);
1362 nl_final(internals->nlsk_fd);
1364 for (i = 0; i < internals->nb_queues; i++)
1365 if (internals->rxq[i].fd != -1)
1366 close(internals->rxq[i].fd);
1368 close(internals->ioctl_sock);
1369 rte_free(eth_dev->data->dev_private);
1370 rte_free(eth_dev->data);
1372 rte_eth_dev_release_port(eth_dev);
1377 static struct rte_vdev_driver pmd_tap_drv = {
1378 .probe = rte_pmd_tap_probe,
1379 .remove = rte_pmd_tap_remove,
1381 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1382 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1383 RTE_PMD_REGISTER_PARAM_STRING(net_tap, "iface=<string>,speed=N");