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"
72 #define ETH_TAP_REMOTE_ARG "remote"
74 #ifdef IFF_MULTI_QUEUE
75 #define RTE_PMD_TAP_MAX_QUEUES 16
77 #define RTE_PMD_TAP_MAX_QUEUES 1
80 #define FLOWER_KERNEL_VERSION KERNEL_VERSION(4, 2, 0)
81 #define FLOWER_VLAN_KERNEL_VERSION KERNEL_VERSION(4, 9, 0)
83 static struct rte_vdev_driver pmd_tap_drv;
85 static const char *valid_arguments[] = {
94 static volatile uint32_t tap_trigger; /* Rx trigger */
96 static struct rte_eth_link pmd_link = {
97 .link_speed = ETH_SPEED_NUM_10G,
98 .link_duplex = ETH_LINK_FULL_DUPLEX,
99 .link_status = ETH_LINK_DOWN,
100 .link_autoneg = ETH_LINK_SPEED_AUTONEG
104 tap_trigger_cb(int sig __rte_unused)
106 /* Valid trigger values are nonzero */
107 tap_trigger = (tap_trigger + 1) | 0x80000000;
111 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
112 struct ifreq *ifr, int set);
114 /* Tun/Tap allocation routine
116 * name is the number of the interface to use, unless NULL to take the host
120 tun_alloc(struct pmd_internals *pmd, uint16_t qid)
123 #ifdef IFF_MULTI_QUEUE
124 unsigned int features;
128 memset(&ifr, 0, sizeof(struct ifreq));
131 * Do not set IFF_NO_PI as packet information header will be needed
132 * to check if a received packet has been truncated.
134 ifr.ifr_flags = IFF_TAP;
135 snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
137 RTE_LOG(DEBUG, PMD, "ifr_name '%s'\n", ifr.ifr_name);
139 fd = open(TUN_TAP_DEV_PATH, O_RDWR);
141 RTE_LOG(ERR, PMD, "Unable to create TAP interface");
145 #ifdef IFF_MULTI_QUEUE
146 /* Grab the TUN features to verify we can work multi-queue */
147 if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
148 RTE_LOG(ERR, PMD, "TAP unable to get TUN/TAP features\n");
151 RTE_LOG(DEBUG, PMD, " TAP Features %08x\n", features);
153 if (features & IFF_MULTI_QUEUE) {
154 RTE_LOG(DEBUG, PMD, " Multi-queue support for %d queues\n",
155 RTE_PMD_TAP_MAX_QUEUES);
156 ifr.ifr_flags |= IFF_MULTI_QUEUE;
160 ifr.ifr_flags |= IFF_ONE_QUEUE;
161 RTE_LOG(DEBUG, PMD, " Single queue only support\n");
164 /* Set the TUN/TAP configuration and set the name if needed */
165 if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
166 RTE_LOG(WARNING, PMD,
167 "Unable to set TUNSETIFF for %s\n",
173 /* Always set the file descriptor to non-blocking */
174 if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
175 RTE_LOG(WARNING, PMD,
176 "Unable to set %s to nonblocking\n",
178 perror("F_SETFL, NONBLOCK");
182 /* Set up trigger to optimize empty Rx bursts */
186 int flags = fcntl(fd, F_GETFL);
188 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
190 if (sa.sa_handler != tap_trigger_cb) {
192 * Make sure SIGIO is not already taken. This is done
193 * as late as possible to leave the application a
194 * chance to set up its own signal handler first.
196 if (sa.sa_handler != SIG_IGN &&
197 sa.sa_handler != SIG_DFL) {
201 sa = (struct sigaction){
202 .sa_flags = SA_RESTART,
203 .sa_handler = tap_trigger_cb,
205 if (sigaction(SIGIO, &sa, NULL) == -1)
208 /* Enable SIGIO on file descriptor */
209 fcntl(fd, F_SETFL, flags | O_ASYNC);
210 fcntl(fd, F_SETOWN, getpid());
213 /* Disable trigger globally in case of error */
215 RTE_LOG(WARNING, PMD, "Rx trigger disabled: %s\n",
222 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0) < 0)
224 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
227 pmd->if_index = if_nametoindex(pmd->name);
228 if (!pmd->if_index) {
230 "Could not find ifindex for %s: rte_flow won't be usable.\n",
234 if (!pmd->flower_support)
236 if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
238 "Could not create multiq qdisc for %s: rte_flow won't be usable.\n",
242 if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
244 "Could not create multiq qdisc for %s: rte_flow won't be usable.\n",
248 if (pmd->remote_if_index) {
250 * Flush usually returns negative value because it tries
251 * to delete every QDISC (and on a running device, one
252 * QDISC at least is needed). Ignore negative return
255 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
256 if (qdisc_create_ingress(pmd->nlsk_fd,
257 pmd->remote_if_index) < 0)
259 LIST_INIT(&pmd->implicit_flows);
260 if (tap_flow_implicit_create(
261 pmd, TAP_REMOTE_LOCAL_MAC) < 0)
263 if (tap_flow_implicit_create(
264 pmd, TAP_REMOTE_BROADCAST) < 0)
266 if (tap_flow_implicit_create(
267 pmd, TAP_REMOTE_BROADCASTV6) < 0)
269 if (tap_flow_implicit_create(
270 pmd, TAP_REMOTE_TX) < 0)
279 "Could not set up remote flow rules for %s: remote disabled.\n",
281 pmd->remote_if_index = 0;
282 tap_flow_implicit_flush(pmd, NULL);
291 /* Callback to handle the rx burst of packets to the correct interface and
292 * file descriptor(s) in a multi-queue setup.
295 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
297 struct rx_queue *rxq = queue;
299 unsigned long num_rx_bytes = 0;
300 uint32_t trigger = tap_trigger;
302 if (trigger == rxq->trigger_seen)
305 rxq->trigger_seen = trigger;
306 rte_compiler_barrier();
307 for (num_rx = 0; num_rx < nb_pkts; ) {
308 struct rte_mbuf *mbuf = rxq->pool;
309 struct rte_mbuf *seg = NULL;
310 struct rte_mbuf *new_tail = NULL;
311 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
314 len = readv(rxq->fd, *rxq->iovecs,
315 1 + (rxq->rxmode->enable_scatter ?
316 rxq->nb_rx_desc : 1));
317 if (len < (int)sizeof(struct tun_pi))
320 /* Packet couldn't fit in the provided mbuf */
321 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
322 rxq->stats.ierrors++;
326 len -= sizeof(struct tun_pi);
329 mbuf->port = rxq->in_port;
331 struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
333 if (unlikely(!buf)) {
334 rxq->stats.rx_nombuf++;
335 /* No new buf has been allocated: do nothing */
336 if (!new_tail || !seg)
340 rte_pktmbuf_free(mbuf);
344 seg = seg ? seg->next : mbuf;
345 if (rxq->pool == mbuf)
348 new_tail->next = buf;
350 new_tail->next = seg->next;
352 /* iovecs[0] is reserved for packet info (pi) */
353 (*rxq->iovecs)[mbuf->nb_segs].iov_len =
354 buf->buf_len - data_off;
355 (*rxq->iovecs)[mbuf->nb_segs].iov_base =
356 (char *)buf->buf_addr + data_off;
358 seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
359 seg->data_off = data_off;
361 len -= seg->data_len;
365 /* First segment has headroom, not the others */
369 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
372 /* account for the receive frame */
373 bufs[num_rx++] = mbuf;
374 num_rx_bytes += mbuf->pkt_len;
377 rxq->stats.ipackets += num_rx;
378 rxq->stats.ibytes += num_rx_bytes;
383 /* Callback to handle sending packets from the tap interface
386 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
388 struct tx_queue *txq = queue;
390 unsigned long num_tx_bytes = 0;
394 if (unlikely(nb_pkts == 0))
397 max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
398 for (i = 0; i < nb_pkts; i++) {
399 struct rte_mbuf *mbuf = bufs[num_tx];
400 struct iovec iovecs[mbuf->nb_segs + 1];
401 struct tun_pi pi = { .flags = 0 };
402 struct rte_mbuf *seg = mbuf;
406 /* stats.errs will be incremented */
407 if (rte_pktmbuf_pkt_len(mbuf) > max_size)
410 iovecs[0].iov_base = π
411 iovecs[0].iov_len = sizeof(pi);
412 for (j = 1; j <= mbuf->nb_segs; j++) {
413 iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
415 rte_pktmbuf_mtod(seg, void *);
418 /* copy the tx frame data */
419 n = writev(txq->fd, iovecs, mbuf->nb_segs + 1);
424 num_tx_bytes += mbuf->pkt_len;
425 rte_pktmbuf_free(mbuf);
428 txq->stats.opackets += num_tx;
429 txq->stats.errs += nb_pkts - num_tx;
430 txq->stats.obytes += num_tx_bytes;
436 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
437 struct ifreq *ifr, int set)
439 short req_flags = ifr->ifr_flags;
440 int remote = !!pmd->remote_if_index;
443 * If there is a remote netdevice, apply ioctl on it, then apply it on
448 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
450 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
453 /* fetch current flags to leave other flags untouched */
454 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
457 ifr->ifr_flags |= req_flags;
459 ifr->ifr_flags &= ~req_flags;
462 /* Set remote MAC on the tap netdevice */
463 if (!remote && pmd->remote_if_index) {
464 request = SIOCSIFHWADDR;
472 RTE_LOG(WARNING, PMD, "%s: ioctl() called with wrong arg\n",
476 if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
483 RTE_LOG(ERR, PMD, "%s: ioctl(%lu) failed with error: %s\n",
484 ifr->ifr_name, request, strerror(errno));
489 tap_link_set_down(struct rte_eth_dev *dev)
491 struct pmd_internals *pmd = dev->data->dev_private;
492 struct ifreq ifr = { .ifr_flags = IFF_UP };
494 dev->data->dev_link.link_status = ETH_LINK_DOWN;
495 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0);
499 tap_link_set_up(struct rte_eth_dev *dev)
501 struct pmd_internals *pmd = dev->data->dev_private;
502 struct ifreq ifr = { .ifr_flags = IFF_UP };
504 dev->data->dev_link.link_status = ETH_LINK_UP;
505 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1);
509 tap_dev_start(struct rte_eth_dev *dev)
511 return tap_link_set_up(dev);
514 /* This function gets called when the current port gets stopped.
517 tap_dev_stop(struct rte_eth_dev *dev)
519 tap_link_set_down(dev);
523 tap_dev_configure(struct rte_eth_dev *dev __rte_unused)
529 tap_dev_speed_capa(void)
531 uint32_t speed = pmd_link.link_speed;
534 if (speed >= ETH_SPEED_NUM_10M)
535 capa |= ETH_LINK_SPEED_10M;
536 if (speed >= ETH_SPEED_NUM_100M)
537 capa |= ETH_LINK_SPEED_100M;
538 if (speed >= ETH_SPEED_NUM_1G)
539 capa |= ETH_LINK_SPEED_1G;
540 if (speed >= ETH_SPEED_NUM_5G)
541 capa |= ETH_LINK_SPEED_2_5G;
542 if (speed >= ETH_SPEED_NUM_5G)
543 capa |= ETH_LINK_SPEED_5G;
544 if (speed >= ETH_SPEED_NUM_10G)
545 capa |= ETH_LINK_SPEED_10G;
546 if (speed >= ETH_SPEED_NUM_20G)
547 capa |= ETH_LINK_SPEED_20G;
548 if (speed >= ETH_SPEED_NUM_25G)
549 capa |= ETH_LINK_SPEED_25G;
550 if (speed >= ETH_SPEED_NUM_40G)
551 capa |= ETH_LINK_SPEED_40G;
552 if (speed >= ETH_SPEED_NUM_50G)
553 capa |= ETH_LINK_SPEED_50G;
554 if (speed >= ETH_SPEED_NUM_56G)
555 capa |= ETH_LINK_SPEED_56G;
556 if (speed >= ETH_SPEED_NUM_100G)
557 capa |= ETH_LINK_SPEED_100G;
563 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
565 struct pmd_internals *internals = dev->data->dev_private;
567 dev_info->if_index = internals->if_index;
568 dev_info->max_mac_addrs = 1;
569 dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
570 dev_info->max_rx_queues = internals->nb_queues;
571 dev_info->max_tx_queues = internals->nb_queues;
572 dev_info->min_rx_bufsize = 0;
573 dev_info->pci_dev = NULL;
574 dev_info->speed_capa = tap_dev_speed_capa();
578 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
580 unsigned int i, imax;
581 unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
582 unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
583 unsigned long rx_nombuf = 0, ierrors = 0;
584 const struct pmd_internals *pmd = dev->data->dev_private;
586 imax = (pmd->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
587 pmd->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
589 for (i = 0; i < imax; i++) {
590 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
591 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
592 rx_total += tap_stats->q_ipackets[i];
593 rx_bytes_total += tap_stats->q_ibytes[i];
594 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
595 ierrors += pmd->rxq[i].stats.ierrors;
597 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
598 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
599 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
600 tx_total += tap_stats->q_opackets[i];
601 tx_err_total += tap_stats->q_errors[i];
602 tx_bytes_total += tap_stats->q_obytes[i];
605 tap_stats->ipackets = rx_total;
606 tap_stats->ibytes = rx_bytes_total;
607 tap_stats->ierrors = ierrors;
608 tap_stats->rx_nombuf = rx_nombuf;
609 tap_stats->opackets = tx_total;
610 tap_stats->oerrors = tx_err_total;
611 tap_stats->obytes = tx_bytes_total;
615 tap_stats_reset(struct rte_eth_dev *dev)
618 struct pmd_internals *pmd = dev->data->dev_private;
620 for (i = 0; i < pmd->nb_queues; i++) {
621 pmd->rxq[i].stats.ipackets = 0;
622 pmd->rxq[i].stats.ibytes = 0;
623 pmd->rxq[i].stats.ierrors = 0;
624 pmd->rxq[i].stats.rx_nombuf = 0;
626 pmd->txq[i].stats.opackets = 0;
627 pmd->txq[i].stats.errs = 0;
628 pmd->txq[i].stats.obytes = 0;
633 tap_dev_close(struct rte_eth_dev *dev __rte_unused)
636 struct pmd_internals *internals = dev->data->dev_private;
638 tap_link_set_down(dev);
639 tap_flow_flush(dev, NULL);
640 tap_flow_implicit_flush(internals, NULL);
642 for (i = 0; i < internals->nb_queues; i++) {
643 if (internals->rxq[i].fd != -1)
644 close(internals->rxq[i].fd);
645 internals->rxq[i].fd = -1;
646 internals->txq[i].fd = -1;
651 tap_rx_queue_release(void *queue)
653 struct rx_queue *rxq = queue;
655 if (rxq && (rxq->fd > 0)) {
658 rte_pktmbuf_free(rxq->pool);
659 rte_free(rxq->iovecs);
666 tap_tx_queue_release(void *queue)
668 struct tx_queue *txq = queue;
670 if (txq && (txq->fd > 0)) {
677 tap_link_update(struct rte_eth_dev *dev __rte_unused,
678 int wait_to_complete __rte_unused)
684 tap_promisc_enable(struct rte_eth_dev *dev)
686 struct pmd_internals *pmd = dev->data->dev_private;
687 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
689 dev->data->promiscuous = 1;
690 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1);
691 if (pmd->remote_if_index)
692 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
696 tap_promisc_disable(struct rte_eth_dev *dev)
698 struct pmd_internals *pmd = dev->data->dev_private;
699 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
701 dev->data->promiscuous = 0;
702 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0);
703 if (pmd->remote_if_index)
704 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
708 tap_allmulti_enable(struct rte_eth_dev *dev)
710 struct pmd_internals *pmd = dev->data->dev_private;
711 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
713 dev->data->all_multicast = 1;
714 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1);
715 if (pmd->remote_if_index)
716 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
720 tap_allmulti_disable(struct rte_eth_dev *dev)
722 struct pmd_internals *pmd = dev->data->dev_private;
723 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
725 dev->data->all_multicast = 0;
726 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0);
727 if (pmd->remote_if_index)
728 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
733 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
735 struct pmd_internals *pmd = dev->data->dev_private;
738 if (is_zero_ether_addr(mac_addr)) {
739 RTE_LOG(ERR, PMD, "%s: can't set an empty MAC address\n",
744 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
745 rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
746 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1) < 0)
748 rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
752 tap_setup_queue(struct rte_eth_dev *dev,
753 struct pmd_internals *internals,
756 struct pmd_internals *pmd = dev->data->dev_private;
757 struct rx_queue *rx = &internals->rxq[qid];
758 struct tx_queue *tx = &internals->txq[qid];
765 RTE_LOG(INFO, PMD, "Add queue to TAP %s for qid %d\n",
767 fd = tun_alloc(pmd, qid);
769 RTE_LOG(ERR, PMD, "tun_alloc(%s, %d) failed\n",
776 ifr.ifr_mtu = dev->data->mtu;
777 if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1) < 0) {
787 tx->mtu = &dev->data->mtu;
788 rx->rxmode = &dev->data->dev_conf.rxmode;
794 rx_setup_queue(struct rte_eth_dev *dev,
795 struct pmd_internals *internals,
798 dev->data->rx_queues[qid] = &internals->rxq[qid];
800 return tap_setup_queue(dev, internals, qid);
804 tx_setup_queue(struct rte_eth_dev *dev,
805 struct pmd_internals *internals,
808 dev->data->tx_queues[qid] = &internals->txq[qid];
810 return tap_setup_queue(dev, internals, qid);
814 tap_rx_queue_setup(struct rte_eth_dev *dev,
815 uint16_t rx_queue_id,
817 unsigned int socket_id,
818 const struct rte_eth_rxconf *rx_conf __rte_unused,
819 struct rte_mempool *mp)
821 struct pmd_internals *internals = dev->data->dev_private;
822 struct rx_queue *rxq = &internals->rxq[rx_queue_id];
823 struct rte_mbuf **tmp = &rxq->pool;
824 struct iovec (*iovecs)[nb_rx_desc + 1];
825 int data_off = RTE_PKTMBUF_HEADROOM;
831 if ((rx_queue_id >= internals->nb_queues) || !mp) {
832 RTE_LOG(WARNING, PMD,
833 "nb_queues %d too small or mempool NULL\n",
834 internals->nb_queues);
839 rxq->trigger_seen = 1; /* force initial burst */
840 rxq->in_port = dev->data->port_id;
841 rxq->nb_rx_desc = nb_rx_desc;
842 iovecs = rte_zmalloc_socket(dev->data->name, sizeof(*iovecs), 0,
845 RTE_LOG(WARNING, PMD,
846 "%s: Couldn't allocate %d RX descriptors\n",
847 dev->data->name, nb_rx_desc);
850 rxq->iovecs = iovecs;
852 /* Now get the space available for data in the mbuf */
853 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(mp) -
854 RTE_PKTMBUF_HEADROOM);
856 if (buf_size < ETH_FRAME_LEN) {
857 RTE_LOG(WARNING, PMD,
858 "%s: %d bytes will not fit in mbuf (%d bytes)\n",
859 dev->data->name, ETH_FRAME_LEN, buf_size);
864 fd = rx_setup_queue(dev, internals, rx_queue_id);
870 (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
871 (*rxq->iovecs)[0].iov_base = &rxq->pi;
873 for (i = 1; i <= nb_rx_desc; i++) {
874 *tmp = rte_pktmbuf_alloc(rxq->mp);
876 RTE_LOG(WARNING, PMD,
877 "%s: couldn't allocate memory for queue %d\n",
878 dev->data->name, rx_queue_id);
882 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
883 (*rxq->iovecs)[i].iov_base =
884 (char *)(*tmp)->buf_addr + data_off;
889 RTE_LOG(DEBUG, PMD, " RX TAP device name %s, qid %d on fd %d\n",
890 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
895 rte_pktmbuf_free(rxq->pool);
897 rte_free(rxq->iovecs);
903 tap_tx_queue_setup(struct rte_eth_dev *dev,
904 uint16_t tx_queue_id,
905 uint16_t nb_tx_desc __rte_unused,
906 unsigned int socket_id __rte_unused,
907 const struct rte_eth_txconf *tx_conf __rte_unused)
909 struct pmd_internals *internals = dev->data->dev_private;
912 if (tx_queue_id >= internals->nb_queues)
915 ret = tx_setup_queue(dev, internals, tx_queue_id);
919 RTE_LOG(DEBUG, PMD, " TX TAP device name %s, qid %d on fd %d\n",
920 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd);
926 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
928 struct pmd_internals *pmd = dev->data->dev_private;
929 struct ifreq ifr = { .ifr_mtu = mtu };
932 err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1);
934 dev->data->mtu = mtu;
940 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
941 struct ether_addr *mc_addr_set __rte_unused,
942 uint32_t nb_mc_addr __rte_unused)
945 * Nothing to do actually: the tap has no filtering whatsoever, every
946 * packet is received.
951 static const uint32_t*
952 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
954 static const uint32_t ptypes[] = {
955 RTE_PTYPE_INNER_L2_ETHER,
956 RTE_PTYPE_INNER_L2_ETHER_VLAN,
957 RTE_PTYPE_INNER_L2_ETHER_QINQ,
958 RTE_PTYPE_INNER_L3_IPV4,
959 RTE_PTYPE_INNER_L3_IPV4_EXT,
960 RTE_PTYPE_INNER_L3_IPV6,
961 RTE_PTYPE_INNER_L3_IPV6_EXT,
962 RTE_PTYPE_INNER_L4_FRAG,
963 RTE_PTYPE_INNER_L4_UDP,
964 RTE_PTYPE_INNER_L4_TCP,
965 RTE_PTYPE_INNER_L4_SCTP,
967 RTE_PTYPE_L2_ETHER_VLAN,
968 RTE_PTYPE_L2_ETHER_QINQ,
970 RTE_PTYPE_L3_IPV4_EXT,
971 RTE_PTYPE_L3_IPV6_EXT,
983 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
984 struct rte_eth_fc_conf *fc_conf)
986 fc_conf->mode = RTE_FC_NONE;
991 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
992 struct rte_eth_fc_conf *fc_conf)
994 if (fc_conf->mode != RTE_FC_NONE)
999 static const struct eth_dev_ops ops = {
1000 .dev_start = tap_dev_start,
1001 .dev_stop = tap_dev_stop,
1002 .dev_close = tap_dev_close,
1003 .dev_configure = tap_dev_configure,
1004 .dev_infos_get = tap_dev_info,
1005 .rx_queue_setup = tap_rx_queue_setup,
1006 .tx_queue_setup = tap_tx_queue_setup,
1007 .rx_queue_release = tap_rx_queue_release,
1008 .tx_queue_release = tap_tx_queue_release,
1009 .flow_ctrl_get = tap_flow_ctrl_get,
1010 .flow_ctrl_set = tap_flow_ctrl_set,
1011 .link_update = tap_link_update,
1012 .dev_set_link_up = tap_link_set_up,
1013 .dev_set_link_down = tap_link_set_down,
1014 .promiscuous_enable = tap_promisc_enable,
1015 .promiscuous_disable = tap_promisc_disable,
1016 .allmulticast_enable = tap_allmulti_enable,
1017 .allmulticast_disable = tap_allmulti_disable,
1018 .mac_addr_set = tap_mac_set,
1019 .mtu_set = tap_mtu_set,
1020 .set_mc_addr_list = tap_set_mc_addr_list,
1021 .stats_get = tap_stats_get,
1022 .stats_reset = tap_stats_reset,
1023 .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1024 .filter_ctrl = tap_dev_filter_ctrl,
1028 tap_kernel_support(struct pmd_internals *pmd)
1030 struct utsname utsname;
1033 if (uname(&utsname) == -1 ||
1034 sscanf(utsname.release, "%d.%d.%d",
1035 &ver[0], &ver[1], &ver[2]) != 3)
1037 if (KERNEL_VERSION(ver[0], ver[1], ver[2]) >= FLOWER_KERNEL_VERSION)
1038 pmd->flower_support = 1;
1039 if (KERNEL_VERSION(ver[0], ver[1], ver[2]) >=
1040 FLOWER_VLAN_KERNEL_VERSION)
1041 pmd->flower_vlan_support = 1;
1046 eth_dev_tap_create(const char *name, char *tap_name, char *remote_iface)
1048 int numa_node = rte_socket_id();
1049 struct rte_eth_dev *dev = NULL;
1050 struct pmd_internals *pmd = NULL;
1051 struct rte_eth_dev_data *data = NULL;
1054 RTE_LOG(DEBUG, PMD, " TAP device on numa %u\n", rte_socket_id());
1056 data = rte_zmalloc_socket(tap_name, sizeof(*data), 0, numa_node);
1058 RTE_LOG(ERR, PMD, "TAP Failed to allocate data\n");
1062 pmd = rte_zmalloc_socket(tap_name, sizeof(*pmd), 0, numa_node);
1064 RTE_LOG(ERR, PMD, "TAP Unable to allocate internal struct\n");
1068 /* name in allocation and data->name must be consistent */
1069 snprintf(data->name, sizeof(data->name), "%s", name);
1070 dev = rte_eth_dev_allocate(name);
1072 RTE_LOG(ERR, PMD, "TAP Unable to allocate device struct\n");
1076 snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1078 pmd->nb_queues = RTE_PMD_TAP_MAX_QUEUES;
1080 pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1081 if (pmd->ioctl_sock == -1) {
1083 "TAP Unable to get a socket for management: %s\n",
1088 /* Setup some default values */
1089 data->dev_private = pmd;
1090 data->port_id = dev->data->port_id;
1091 data->mtu = dev->data->mtu;
1092 data->dev_flags = RTE_ETH_DEV_DETACHABLE;
1093 data->kdrv = RTE_KDRV_NONE;
1094 data->drv_name = pmd_tap_drv.driver.name;
1095 data->numa_node = numa_node;
1097 data->dev_link = pmd_link;
1098 data->mac_addrs = &pmd->eth_addr;
1099 data->nb_rx_queues = pmd->nb_queues;
1100 data->nb_tx_queues = pmd->nb_queues;
1103 dev->dev_ops = &ops;
1105 dev->rx_pkt_burst = pmd_rx_burst;
1106 dev->tx_pkt_burst = pmd_tx_burst;
1108 /* Presetup the fds to -1 as being not valid */
1109 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1110 pmd->rxq[i].fd = -1;
1111 pmd->txq[i].fd = -1;
1114 tap_kernel_support(pmd);
1115 if (!pmd->flower_support)
1117 LIST_INIT(&pmd->flows);
1119 * If no netlink socket can be created, then it will fail when
1120 * creating/destroying flow rules.
1122 pmd->nlsk_fd = nl_init();
1123 if (strlen(remote_iface)) {
1124 pmd->remote_if_index = if_nametoindex(remote_iface);
1125 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1126 "%s", remote_iface);
1127 if (!pmd->remote_if_index)
1128 RTE_LOG(ERR, PMD, "Could not find %s ifindex: "
1129 "remote interface will remain unconfigured\n",
1136 RTE_LOG(DEBUG, PMD, "TAP Unable to initialize %s\n", name);
1141 rte_eth_dev_release_port(dev);
1147 set_interface_name(const char *key __rte_unused,
1151 char *name = (char *)extra_args;
1154 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s", value);
1156 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1157 DEFAULT_TAP_NAME, (tap_unit - 1));
1163 set_interface_speed(const char *key __rte_unused,
1167 *(int *)extra_args = (value) ? atoi(value) : ETH_SPEED_NUM_10G;
1173 set_remote_iface(const char *key __rte_unused,
1177 char *name = (char *)extra_args;
1180 snprintf(name, RTE_ETH_NAME_MAX_LEN, "%s", value);
1185 /* Open a TAP interface device.
1188 rte_pmd_tap_probe(const char *name, const char *params)
1191 struct rte_kvargs *kvlist = NULL;
1193 char tap_name[RTE_ETH_NAME_MAX_LEN];
1194 char remote_iface[RTE_ETH_NAME_MAX_LEN];
1196 speed = ETH_SPEED_NUM_10G;
1197 snprintf(tap_name, sizeof(tap_name), "%s%d",
1198 DEFAULT_TAP_NAME, tap_unit++);
1199 memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1201 if (params && (params[0] != '\0')) {
1202 RTE_LOG(DEBUG, PMD, "paramaters (%s)\n", params);
1204 kvlist = rte_kvargs_parse(params, valid_arguments);
1206 if (rte_kvargs_count(kvlist, ETH_TAP_SPEED_ARG) == 1) {
1207 ret = rte_kvargs_process(kvlist,
1209 &set_interface_speed,
1215 if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1216 ret = rte_kvargs_process(kvlist,
1218 &set_interface_name,
1224 if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
1225 ret = rte_kvargs_process(kvlist,
1234 pmd_link.link_speed = speed;
1236 RTE_LOG(NOTICE, PMD, "Initializing pmd_tap for %s as %s\n",
1239 ret = eth_dev_tap_create(name, tap_name, remote_iface);
1243 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
1245 tap_unit--; /* Restore the unit number */
1247 rte_kvargs_free(kvlist);
1252 /* detach a TAP device.
1255 rte_pmd_tap_remove(const char *name)
1257 struct rte_eth_dev *eth_dev = NULL;
1258 struct pmd_internals *internals;
1261 RTE_LOG(DEBUG, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
1264 /* find the ethdev entry */
1265 eth_dev = rte_eth_dev_allocated(name);
1269 internals = eth_dev->data->dev_private;
1270 if (internals->flower_support && internals->nlsk_fd) {
1271 tap_flow_flush(eth_dev, NULL);
1272 tap_flow_implicit_flush(internals, NULL);
1273 nl_final(internals->nlsk_fd);
1275 for (i = 0; i < internals->nb_queues; i++)
1276 if (internals->rxq[i].fd != -1)
1277 close(internals->rxq[i].fd);
1279 close(internals->ioctl_sock);
1280 rte_free(eth_dev->data->dev_private);
1281 rte_free(eth_dev->data);
1283 rte_eth_dev_release_port(eth_dev);
1288 static struct rte_vdev_driver pmd_tap_drv = {
1289 .probe = rte_pmd_tap_probe,
1290 .remove = rte_pmd_tap_remove,
1292 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1293 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1294 RTE_PMD_REGISTER_PARAM_STRING(net_tap, "iface=<string>,speed=N");