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_tcmsgs.h>
66 /* Linux based path to the TUN device */
67 #define TUN_TAP_DEV_PATH "/dev/net/tun"
68 #define DEFAULT_TAP_NAME "dtap"
70 #define ETH_TAP_IFACE_ARG "iface"
71 #define ETH_TAP_SPEED_ARG "speed"
73 #ifdef IFF_MULTI_QUEUE
74 #define RTE_PMD_TAP_MAX_QUEUES 16
76 #define RTE_PMD_TAP_MAX_QUEUES 1
79 #define FLOWER_KERNEL_VERSION KERNEL_VERSION(4, 2, 0)
80 #define FLOWER_VLAN_KERNEL_VERSION KERNEL_VERSION(4, 9, 0)
82 static struct rte_vdev_driver pmd_tap_drv;
84 static const char *valid_arguments[] = {
92 static volatile uint32_t tap_trigger; /* Rx trigger */
94 static struct rte_eth_link pmd_link = {
95 .link_speed = ETH_SPEED_NUM_10G,
96 .link_duplex = ETH_LINK_FULL_DUPLEX,
97 .link_status = ETH_LINK_DOWN,
98 .link_autoneg = ETH_LINK_SPEED_AUTONEG
102 tap_trigger_cb(int sig __rte_unused)
104 /* Valid trigger values are nonzero */
105 tap_trigger = (tap_trigger + 1) | 0x80000000;
109 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
110 struct ifreq *ifr, int set);
112 /* Tun/Tap allocation routine
114 * name is the number of the interface to use, unless NULL to take the host
118 tun_alloc(struct pmd_internals *pmd, uint16_t qid)
121 #ifdef IFF_MULTI_QUEUE
122 unsigned int features;
126 memset(&ifr, 0, sizeof(struct ifreq));
129 * Do not set IFF_NO_PI as packet information header will be needed
130 * to check if a received packet has been truncated.
132 ifr.ifr_flags = IFF_TAP;
133 snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
135 RTE_LOG(DEBUG, PMD, "ifr_name '%s'\n", ifr.ifr_name);
137 fd = open(TUN_TAP_DEV_PATH, O_RDWR);
139 RTE_LOG(ERR, PMD, "Unable to create TAP interface");
143 #ifdef IFF_MULTI_QUEUE
144 /* Grab the TUN features to verify we can work multi-queue */
145 if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
146 RTE_LOG(ERR, PMD, "TAP unable to get TUN/TAP features\n");
149 RTE_LOG(DEBUG, PMD, " TAP Features %08x\n", features);
151 if (features & IFF_MULTI_QUEUE) {
152 RTE_LOG(DEBUG, PMD, " Multi-queue support for %d queues\n",
153 RTE_PMD_TAP_MAX_QUEUES);
154 ifr.ifr_flags |= IFF_MULTI_QUEUE;
158 ifr.ifr_flags |= IFF_ONE_QUEUE;
159 RTE_LOG(DEBUG, PMD, " Single queue only support\n");
162 /* Set the TUN/TAP configuration and set the name if needed */
163 if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
164 RTE_LOG(WARNING, PMD,
165 "Unable to set TUNSETIFF for %s\n",
171 /* Always set the file descriptor to non-blocking */
172 if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
173 RTE_LOG(WARNING, PMD,
174 "Unable to set %s to nonblocking\n",
176 perror("F_SETFL, NONBLOCK");
180 /* Set up trigger to optimize empty Rx bursts */
184 int flags = fcntl(fd, F_GETFL);
186 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
188 if (sa.sa_handler != tap_trigger_cb) {
190 * Make sure SIGIO is not already taken. This is done
191 * as late as possible to leave the application a
192 * chance to set up its own signal handler first.
194 if (sa.sa_handler != SIG_IGN &&
195 sa.sa_handler != SIG_DFL) {
199 sa = (struct sigaction){
200 .sa_flags = SA_RESTART,
201 .sa_handler = tap_trigger_cb,
203 if (sigaction(SIGIO, &sa, NULL) == -1)
206 /* Enable SIGIO on file descriptor */
207 fcntl(fd, F_SETFL, flags | O_ASYNC);
208 fcntl(fd, F_SETOWN, getpid());
211 /* Disable trigger globally in case of error */
213 RTE_LOG(WARNING, PMD, "Rx trigger disabled: %s\n",
220 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0) < 0)
222 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
225 pmd->if_index = if_nametoindex(pmd->name);
226 if (!pmd->if_index) {
228 "Could not find ifindex for %s: rte_flow won't be usable.\n",
232 if (!pmd->flower_support)
234 if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
236 "Could not create multiq qdisc for %s: rte_flow won't be usable.\n",
240 if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
242 "Could not create multiq qdisc for %s: rte_flow won't be usable.\n",
256 /* Callback to handle the rx burst of packets to the correct interface and
257 * file descriptor(s) in a multi-queue setup.
260 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
262 struct rx_queue *rxq = queue;
264 unsigned long num_rx_bytes = 0;
265 uint32_t trigger = tap_trigger;
267 if (trigger == rxq->trigger_seen)
270 rxq->trigger_seen = trigger;
271 rte_compiler_barrier();
272 for (num_rx = 0; num_rx < nb_pkts; ) {
273 struct rte_mbuf *mbuf = rxq->pool;
274 struct rte_mbuf *seg = NULL;
275 struct rte_mbuf *new_tail = NULL;
276 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
279 len = readv(rxq->fd, *rxq->iovecs,
280 1 + (rxq->rxmode->enable_scatter ?
281 rxq->nb_rx_desc : 1));
282 if (len < (int)sizeof(struct tun_pi))
285 /* Packet couldn't fit in the provided mbuf */
286 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
287 rxq->stats.ierrors++;
291 len -= sizeof(struct tun_pi);
294 mbuf->port = rxq->in_port;
296 struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
298 if (unlikely(!buf)) {
299 rxq->stats.rx_nombuf++;
300 /* No new buf has been allocated: do nothing */
301 if (!new_tail || !seg)
305 rte_pktmbuf_free(mbuf);
309 seg = seg ? seg->next : mbuf;
310 if (rxq->pool == mbuf)
313 new_tail->next = buf;
315 new_tail->next = seg->next;
317 /* iovecs[0] is reserved for packet info (pi) */
318 (*rxq->iovecs)[mbuf->nb_segs].iov_len =
319 buf->buf_len - data_off;
320 (*rxq->iovecs)[mbuf->nb_segs].iov_base =
321 (char *)buf->buf_addr + data_off;
323 seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
324 seg->data_off = data_off;
326 len -= seg->data_len;
330 /* First segment has headroom, not the others */
334 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
337 /* account for the receive frame */
338 bufs[num_rx++] = mbuf;
339 num_rx_bytes += mbuf->pkt_len;
342 rxq->stats.ipackets += num_rx;
343 rxq->stats.ibytes += num_rx_bytes;
348 /* Callback to handle sending packets from the tap interface
351 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
353 struct tx_queue *txq = queue;
355 unsigned long num_tx_bytes = 0;
359 if (unlikely(nb_pkts == 0))
362 max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
363 for (i = 0; i < nb_pkts; i++) {
364 struct rte_mbuf *mbuf = bufs[num_tx];
365 struct iovec iovecs[mbuf->nb_segs + 1];
366 struct tun_pi pi = { .flags = 0 };
367 struct rte_mbuf *seg = mbuf;
371 /* stats.errs will be incremented */
372 if (rte_pktmbuf_pkt_len(mbuf) > max_size)
375 iovecs[0].iov_base = π
376 iovecs[0].iov_len = sizeof(pi);
377 for (j = 1; j <= mbuf->nb_segs; j++) {
378 iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
380 rte_pktmbuf_mtod(seg, void *);
383 /* copy the tx frame data */
384 n = writev(txq->fd, iovecs, mbuf->nb_segs + 1);
389 num_tx_bytes += mbuf->pkt_len;
390 rte_pktmbuf_free(mbuf);
393 txq->stats.opackets += num_tx;
394 txq->stats.errs += nb_pkts - num_tx;
395 txq->stats.obytes += num_tx_bytes;
401 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
402 struct ifreq *ifr, int set)
404 short req_flags = ifr->ifr_flags;
406 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
409 /* fetch current flags to leave other flags untouched */
410 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
413 ifr->ifr_flags |= req_flags;
415 ifr->ifr_flags &= ~req_flags;
422 RTE_LOG(WARNING, PMD, "%s: ioctl() called with wrong arg\n",
426 if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
431 RTE_LOG(ERR, PMD, "%s: ioctl(%lu) failed with error: %s\n",
432 ifr->ifr_name, request, strerror(errno));
437 tap_link_set_down(struct rte_eth_dev *dev)
439 struct pmd_internals *pmd = dev->data->dev_private;
440 struct ifreq ifr = { .ifr_flags = IFF_UP };
442 dev->data->dev_link.link_status = ETH_LINK_DOWN;
443 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0);
447 tap_link_set_up(struct rte_eth_dev *dev)
449 struct pmd_internals *pmd = dev->data->dev_private;
450 struct ifreq ifr = { .ifr_flags = IFF_UP };
452 dev->data->dev_link.link_status = ETH_LINK_UP;
453 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1);
457 tap_dev_start(struct rte_eth_dev *dev)
459 return tap_link_set_up(dev);
462 /* This function gets called when the current port gets stopped.
465 tap_dev_stop(struct rte_eth_dev *dev)
467 tap_link_set_down(dev);
471 tap_dev_configure(struct rte_eth_dev *dev __rte_unused)
477 tap_dev_speed_capa(void)
479 uint32_t speed = pmd_link.link_speed;
482 if (speed >= ETH_SPEED_NUM_10M)
483 capa |= ETH_LINK_SPEED_10M;
484 if (speed >= ETH_SPEED_NUM_100M)
485 capa |= ETH_LINK_SPEED_100M;
486 if (speed >= ETH_SPEED_NUM_1G)
487 capa |= ETH_LINK_SPEED_1G;
488 if (speed >= ETH_SPEED_NUM_5G)
489 capa |= ETH_LINK_SPEED_2_5G;
490 if (speed >= ETH_SPEED_NUM_5G)
491 capa |= ETH_LINK_SPEED_5G;
492 if (speed >= ETH_SPEED_NUM_10G)
493 capa |= ETH_LINK_SPEED_10G;
494 if (speed >= ETH_SPEED_NUM_20G)
495 capa |= ETH_LINK_SPEED_20G;
496 if (speed >= ETH_SPEED_NUM_25G)
497 capa |= ETH_LINK_SPEED_25G;
498 if (speed >= ETH_SPEED_NUM_40G)
499 capa |= ETH_LINK_SPEED_40G;
500 if (speed >= ETH_SPEED_NUM_50G)
501 capa |= ETH_LINK_SPEED_50G;
502 if (speed >= ETH_SPEED_NUM_56G)
503 capa |= ETH_LINK_SPEED_56G;
504 if (speed >= ETH_SPEED_NUM_100G)
505 capa |= ETH_LINK_SPEED_100G;
511 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
513 struct pmd_internals *internals = dev->data->dev_private;
515 dev_info->if_index = internals->if_index;
516 dev_info->max_mac_addrs = 1;
517 dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
518 dev_info->max_rx_queues = internals->nb_queues;
519 dev_info->max_tx_queues = internals->nb_queues;
520 dev_info->min_rx_bufsize = 0;
521 dev_info->pci_dev = NULL;
522 dev_info->speed_capa = tap_dev_speed_capa();
526 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
528 unsigned int i, imax;
529 unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
530 unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
531 unsigned long rx_nombuf = 0, ierrors = 0;
532 const struct pmd_internals *pmd = dev->data->dev_private;
534 imax = (pmd->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
535 pmd->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
537 for (i = 0; i < imax; i++) {
538 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
539 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
540 rx_total += tap_stats->q_ipackets[i];
541 rx_bytes_total += tap_stats->q_ibytes[i];
542 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
543 ierrors += pmd->rxq[i].stats.ierrors;
545 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
546 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
547 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
548 tx_total += tap_stats->q_opackets[i];
549 tx_err_total += tap_stats->q_errors[i];
550 tx_bytes_total += tap_stats->q_obytes[i];
553 tap_stats->ipackets = rx_total;
554 tap_stats->ibytes = rx_bytes_total;
555 tap_stats->ierrors = ierrors;
556 tap_stats->rx_nombuf = rx_nombuf;
557 tap_stats->opackets = tx_total;
558 tap_stats->oerrors = tx_err_total;
559 tap_stats->obytes = tx_bytes_total;
563 tap_stats_reset(struct rte_eth_dev *dev)
566 struct pmd_internals *pmd = dev->data->dev_private;
568 for (i = 0; i < pmd->nb_queues; i++) {
569 pmd->rxq[i].stats.ipackets = 0;
570 pmd->rxq[i].stats.ibytes = 0;
571 pmd->rxq[i].stats.ierrors = 0;
572 pmd->rxq[i].stats.rx_nombuf = 0;
574 pmd->txq[i].stats.opackets = 0;
575 pmd->txq[i].stats.errs = 0;
576 pmd->txq[i].stats.obytes = 0;
581 tap_dev_close(struct rte_eth_dev *dev __rte_unused)
584 struct pmd_internals *internals = dev->data->dev_private;
586 tap_link_set_down(dev);
587 tap_flow_flush(dev, NULL);
589 for (i = 0; i < internals->nb_queues; i++) {
590 if (internals->rxq[i].fd != -1)
591 close(internals->rxq[i].fd);
592 internals->rxq[i].fd = -1;
593 internals->txq[i].fd = -1;
598 tap_rx_queue_release(void *queue)
600 struct rx_queue *rxq = queue;
602 if (rxq && (rxq->fd > 0)) {
605 rte_pktmbuf_free(rxq->pool);
606 rte_free(rxq->iovecs);
613 tap_tx_queue_release(void *queue)
615 struct tx_queue *txq = queue;
617 if (txq && (txq->fd > 0)) {
624 tap_link_update(struct rte_eth_dev *dev __rte_unused,
625 int wait_to_complete __rte_unused)
631 tap_promisc_enable(struct rte_eth_dev *dev)
633 struct pmd_internals *pmd = dev->data->dev_private;
634 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
636 dev->data->promiscuous = 1;
637 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1);
641 tap_promisc_disable(struct rte_eth_dev *dev)
643 struct pmd_internals *pmd = dev->data->dev_private;
644 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
646 dev->data->promiscuous = 0;
647 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0);
651 tap_allmulti_enable(struct rte_eth_dev *dev)
653 struct pmd_internals *pmd = dev->data->dev_private;
654 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
656 dev->data->all_multicast = 1;
657 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1);
661 tap_allmulti_disable(struct rte_eth_dev *dev)
663 struct pmd_internals *pmd = dev->data->dev_private;
664 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
666 dev->data->all_multicast = 0;
667 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0);
672 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
674 struct pmd_internals *pmd = dev->data->dev_private;
677 if (is_zero_ether_addr(mac_addr)) {
678 RTE_LOG(ERR, PMD, "%s: can't set an empty MAC address\n",
683 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
684 rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
685 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1) < 0)
687 rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
691 tap_setup_queue(struct rte_eth_dev *dev,
692 struct pmd_internals *internals,
695 struct pmd_internals *pmd = dev->data->dev_private;
696 struct rx_queue *rx = &internals->rxq[qid];
697 struct tx_queue *tx = &internals->txq[qid];
704 RTE_LOG(INFO, PMD, "Add queue to TAP %s for qid %d\n",
706 fd = tun_alloc(pmd, qid);
708 RTE_LOG(ERR, PMD, "tun_alloc(%s, %d) failed\n",
715 ifr.ifr_mtu = dev->data->mtu;
716 if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1) < 0) {
726 tx->mtu = &dev->data->mtu;
727 rx->rxmode = &dev->data->dev_conf.rxmode;
733 rx_setup_queue(struct rte_eth_dev *dev,
734 struct pmd_internals *internals,
737 dev->data->rx_queues[qid] = &internals->rxq[qid];
739 return tap_setup_queue(dev, internals, qid);
743 tx_setup_queue(struct rte_eth_dev *dev,
744 struct pmd_internals *internals,
747 dev->data->tx_queues[qid] = &internals->txq[qid];
749 return tap_setup_queue(dev, internals, qid);
753 tap_rx_queue_setup(struct rte_eth_dev *dev,
754 uint16_t rx_queue_id,
756 unsigned int socket_id,
757 const struct rte_eth_rxconf *rx_conf __rte_unused,
758 struct rte_mempool *mp)
760 struct pmd_internals *internals = dev->data->dev_private;
761 struct rx_queue *rxq = &internals->rxq[rx_queue_id];
762 struct rte_mbuf **tmp = &rxq->pool;
763 struct iovec (*iovecs)[nb_rx_desc + 1];
764 int data_off = RTE_PKTMBUF_HEADROOM;
770 if ((rx_queue_id >= internals->nb_queues) || !mp) {
771 RTE_LOG(WARNING, PMD,
772 "nb_queues %d too small or mempool NULL\n",
773 internals->nb_queues);
778 rxq->trigger_seen = 1; /* force initial burst */
779 rxq->in_port = dev->data->port_id;
780 rxq->nb_rx_desc = nb_rx_desc;
781 iovecs = rte_zmalloc_socket(dev->data->name, sizeof(*iovecs), 0,
784 RTE_LOG(WARNING, PMD,
785 "%s: Couldn't allocate %d RX descriptors\n",
786 dev->data->name, nb_rx_desc);
789 rxq->iovecs = iovecs;
791 /* Now get the space available for data in the mbuf */
792 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(mp) -
793 RTE_PKTMBUF_HEADROOM);
795 if (buf_size < ETH_FRAME_LEN) {
796 RTE_LOG(WARNING, PMD,
797 "%s: %d bytes will not fit in mbuf (%d bytes)\n",
798 dev->data->name, ETH_FRAME_LEN, buf_size);
803 fd = rx_setup_queue(dev, internals, rx_queue_id);
809 (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
810 (*rxq->iovecs)[0].iov_base = &rxq->pi;
812 for (i = 1; i <= nb_rx_desc; i++) {
813 *tmp = rte_pktmbuf_alloc(rxq->mp);
815 RTE_LOG(WARNING, PMD,
816 "%s: couldn't allocate memory for queue %d\n",
817 dev->data->name, rx_queue_id);
821 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
822 (*rxq->iovecs)[i].iov_base =
823 (char *)(*tmp)->buf_addr + data_off;
828 RTE_LOG(DEBUG, PMD, " RX TAP device name %s, qid %d on fd %d\n",
829 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
834 rte_pktmbuf_free(rxq->pool);
836 rte_free(rxq->iovecs);
842 tap_tx_queue_setup(struct rte_eth_dev *dev,
843 uint16_t tx_queue_id,
844 uint16_t nb_tx_desc __rte_unused,
845 unsigned int socket_id __rte_unused,
846 const struct rte_eth_txconf *tx_conf __rte_unused)
848 struct pmd_internals *internals = dev->data->dev_private;
851 if (tx_queue_id >= internals->nb_queues)
854 ret = tx_setup_queue(dev, internals, tx_queue_id);
858 RTE_LOG(DEBUG, PMD, " TX TAP device name %s, qid %d on fd %d\n",
859 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd);
865 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
867 struct pmd_internals *pmd = dev->data->dev_private;
868 struct ifreq ifr = { .ifr_mtu = mtu };
871 err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1);
873 dev->data->mtu = mtu;
879 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
880 struct ether_addr *mc_addr_set __rte_unused,
881 uint32_t nb_mc_addr __rte_unused)
884 * Nothing to do actually: the tap has no filtering whatsoever, every
885 * packet is received.
890 static const uint32_t*
891 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
893 static const uint32_t ptypes[] = {
894 RTE_PTYPE_INNER_L2_ETHER,
895 RTE_PTYPE_INNER_L2_ETHER_VLAN,
896 RTE_PTYPE_INNER_L2_ETHER_QINQ,
897 RTE_PTYPE_INNER_L3_IPV4,
898 RTE_PTYPE_INNER_L3_IPV4_EXT,
899 RTE_PTYPE_INNER_L3_IPV6,
900 RTE_PTYPE_INNER_L3_IPV6_EXT,
901 RTE_PTYPE_INNER_L4_FRAG,
902 RTE_PTYPE_INNER_L4_UDP,
903 RTE_PTYPE_INNER_L4_TCP,
904 RTE_PTYPE_INNER_L4_SCTP,
906 RTE_PTYPE_L2_ETHER_VLAN,
907 RTE_PTYPE_L2_ETHER_QINQ,
909 RTE_PTYPE_L3_IPV4_EXT,
910 RTE_PTYPE_L3_IPV6_EXT,
922 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
923 struct rte_eth_fc_conf *fc_conf)
925 fc_conf->mode = RTE_FC_NONE;
930 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
931 struct rte_eth_fc_conf *fc_conf)
933 if (fc_conf->mode != RTE_FC_NONE)
938 static const struct eth_dev_ops ops = {
939 .dev_start = tap_dev_start,
940 .dev_stop = tap_dev_stop,
941 .dev_close = tap_dev_close,
942 .dev_configure = tap_dev_configure,
943 .dev_infos_get = tap_dev_info,
944 .rx_queue_setup = tap_rx_queue_setup,
945 .tx_queue_setup = tap_tx_queue_setup,
946 .rx_queue_release = tap_rx_queue_release,
947 .tx_queue_release = tap_tx_queue_release,
948 .flow_ctrl_get = tap_flow_ctrl_get,
949 .flow_ctrl_set = tap_flow_ctrl_set,
950 .link_update = tap_link_update,
951 .dev_set_link_up = tap_link_set_up,
952 .dev_set_link_down = tap_link_set_down,
953 .promiscuous_enable = tap_promisc_enable,
954 .promiscuous_disable = tap_promisc_disable,
955 .allmulticast_enable = tap_allmulti_enable,
956 .allmulticast_disable = tap_allmulti_disable,
957 .mac_addr_set = tap_mac_set,
958 .mtu_set = tap_mtu_set,
959 .set_mc_addr_list = tap_set_mc_addr_list,
960 .stats_get = tap_stats_get,
961 .stats_reset = tap_stats_reset,
962 .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
963 .filter_ctrl = tap_dev_filter_ctrl,
967 tap_kernel_support(struct pmd_internals *pmd)
969 struct utsname utsname;
972 if (uname(&utsname) == -1 ||
973 sscanf(utsname.release, "%d.%d.%d",
974 &ver[0], &ver[1], &ver[2]) != 3)
976 if (KERNEL_VERSION(ver[0], ver[1], ver[2]) >= FLOWER_KERNEL_VERSION)
977 pmd->flower_support = 1;
978 if (KERNEL_VERSION(ver[0], ver[1], ver[2]) >=
979 FLOWER_VLAN_KERNEL_VERSION)
980 pmd->flower_vlan_support = 1;
985 eth_dev_tap_create(const char *name, char *tap_name)
987 int numa_node = rte_socket_id();
988 struct rte_eth_dev *dev = NULL;
989 struct pmd_internals *pmd = NULL;
990 struct rte_eth_dev_data *data = NULL;
993 RTE_LOG(DEBUG, PMD, " TAP device on numa %u\n", rte_socket_id());
995 data = rte_zmalloc_socket(tap_name, sizeof(*data), 0, numa_node);
997 RTE_LOG(ERR, PMD, "TAP Failed to allocate data\n");
1001 pmd = rte_zmalloc_socket(tap_name, sizeof(*pmd), 0, numa_node);
1003 RTE_LOG(ERR, PMD, "TAP Unable to allocate internal struct\n");
1007 /* name in allocation and data->name must be consistent */
1008 snprintf(data->name, sizeof(data->name), "%s", name);
1009 dev = rte_eth_dev_allocate(name);
1011 RTE_LOG(ERR, PMD, "TAP Unable to allocate device struct\n");
1015 snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1017 pmd->nb_queues = RTE_PMD_TAP_MAX_QUEUES;
1019 pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1020 if (pmd->ioctl_sock == -1) {
1022 "TAP Unable to get a socket for management: %s\n",
1027 /* Setup some default values */
1028 data->dev_private = pmd;
1029 data->port_id = dev->data->port_id;
1030 data->mtu = dev->data->mtu;
1031 data->dev_flags = RTE_ETH_DEV_DETACHABLE;
1032 data->kdrv = RTE_KDRV_NONE;
1033 data->drv_name = pmd_tap_drv.driver.name;
1034 data->numa_node = numa_node;
1036 data->dev_link = pmd_link;
1037 data->mac_addrs = &pmd->eth_addr;
1038 data->nb_rx_queues = pmd->nb_queues;
1039 data->nb_tx_queues = pmd->nb_queues;
1042 dev->dev_ops = &ops;
1044 dev->rx_pkt_burst = pmd_rx_burst;
1045 dev->tx_pkt_burst = pmd_tx_burst;
1047 /* Presetup the fds to -1 as being not valid */
1048 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1049 pmd->rxq[i].fd = -1;
1050 pmd->txq[i].fd = -1;
1053 tap_kernel_support(pmd);
1054 if (!pmd->flower_support)
1056 LIST_INIT(&pmd->flows);
1058 * If no netlink socket can be created, then it will fail when
1059 * creating/destroying flow rules.
1061 pmd->nlsk_fd = nl_init();
1066 RTE_LOG(DEBUG, PMD, "TAP Unable to initialize %s\n", name);
1071 rte_eth_dev_release_port(dev);
1077 set_interface_name(const char *key __rte_unused,
1081 char *name = (char *)extra_args;
1084 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s", value);
1086 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1087 DEFAULT_TAP_NAME, (tap_unit - 1));
1093 set_interface_speed(const char *key __rte_unused,
1097 *(int *)extra_args = (value) ? atoi(value) : ETH_SPEED_NUM_10G;
1102 /* Open a TAP interface device.
1105 rte_pmd_tap_probe(const char *name, const char *params)
1108 struct rte_kvargs *kvlist = NULL;
1110 char tap_name[RTE_ETH_NAME_MAX_LEN];
1112 speed = ETH_SPEED_NUM_10G;
1113 snprintf(tap_name, sizeof(tap_name), "%s%d",
1114 DEFAULT_TAP_NAME, tap_unit++);
1116 if (params && (params[0] != '\0')) {
1117 RTE_LOG(DEBUG, PMD, "paramaters (%s)\n", params);
1119 kvlist = rte_kvargs_parse(params, valid_arguments);
1121 if (rte_kvargs_count(kvlist, ETH_TAP_SPEED_ARG) == 1) {
1122 ret = rte_kvargs_process(kvlist,
1124 &set_interface_speed,
1130 if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1131 ret = rte_kvargs_process(kvlist,
1133 &set_interface_name,
1140 pmd_link.link_speed = speed;
1142 RTE_LOG(NOTICE, PMD, "Initializing pmd_tap for %s as %s\n",
1145 ret = eth_dev_tap_create(name, tap_name);
1149 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
1151 tap_unit--; /* Restore the unit number */
1153 rte_kvargs_free(kvlist);
1158 /* detach a TAP device.
1161 rte_pmd_tap_remove(const char *name)
1163 struct rte_eth_dev *eth_dev = NULL;
1164 struct pmd_internals *internals;
1167 RTE_LOG(DEBUG, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
1170 /* find the ethdev entry */
1171 eth_dev = rte_eth_dev_allocated(name);
1175 internals = eth_dev->data->dev_private;
1176 if (internals->flower_support && internals->nlsk_fd) {
1177 tap_flow_flush(eth_dev, NULL);
1178 nl_final(internals->nlsk_fd);
1180 for (i = 0; i < internals->nb_queues; i++)
1181 if (internals->rxq[i].fd != -1)
1182 close(internals->rxq[i].fd);
1184 close(internals->ioctl_sock);
1185 rte_free(eth_dev->data->dev_private);
1186 rte_free(eth_dev->data);
1188 rte_eth_dev_release_port(eth_dev);
1193 static struct rte_vdev_driver pmd_tap_drv = {
1194 .probe = rte_pmd_tap_probe,
1195 .remove = rte_pmd_tap_remove,
1197 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1198 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1199 RTE_PMD_REGISTER_PARAM_STRING(net_tap, "iface=<string>,speed=N");