1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2016-2017 Intel Corporation
5 #include <rte_atomic.h>
6 #include <rte_branch_prediction.h>
7 #include <rte_byteorder.h>
8 #include <rte_common.h>
10 #include <rte_ethdev_driver.h>
11 #include <rte_ethdev_vdev.h>
12 #include <rte_malloc.h>
13 #include <rte_bus_vdev.h>
14 #include <rte_kvargs.h>
16 #include <rte_debug.h>
18 #include <rte_string_fns.h>
20 #include <sys/types.h>
22 #include <sys/socket.h>
23 #include <sys/ioctl.h>
24 #include <sys/utsname.h>
32 #include <arpa/inet.h>
34 #include <linux/if_tun.h>
35 #include <linux/if_ether.h>
39 #include <rte_eth_tap.h>
41 #include <tap_netlink.h>
42 #include <tap_tcmsgs.h>
44 /* Linux based path to the TUN device */
45 #define TUN_TAP_DEV_PATH "/dev/net/tun"
46 #define DEFAULT_TAP_NAME "dtap"
47 #define DEFAULT_TUN_NAME "dtun"
49 #define ETH_TAP_IFACE_ARG "iface"
50 #define ETH_TAP_REMOTE_ARG "remote"
51 #define ETH_TAP_MAC_ARG "mac"
52 #define ETH_TAP_MAC_FIXED "fixed"
54 #define ETH_TAP_USR_MAC_FMT "xx:xx:xx:xx:xx:xx"
55 #define ETH_TAP_CMP_MAC_FMT "0123456789ABCDEFabcdef"
56 #define ETH_TAP_MAC_ARG_FMT ETH_TAP_MAC_FIXED "|" ETH_TAP_USR_MAC_FMT
58 static struct rte_vdev_driver pmd_tap_drv;
59 static struct rte_vdev_driver pmd_tun_drv;
61 static const char *valid_arguments[] = {
68 static unsigned int tap_unit;
69 static unsigned int tun_unit;
71 static char tuntap_name[8];
73 static volatile uint32_t tap_trigger; /* Rx trigger */
75 static struct rte_eth_link pmd_link = {
76 .link_speed = ETH_SPEED_NUM_10G,
77 .link_duplex = ETH_LINK_FULL_DUPLEX,
78 .link_status = ETH_LINK_DOWN,
79 .link_autoneg = ETH_LINK_FIXED,
83 tap_trigger_cb(int sig __rte_unused)
85 /* Valid trigger values are nonzero */
86 tap_trigger = (tap_trigger + 1) | 0x80000000;
89 /* Specifies on what netdevices the ioctl should be applied */
96 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
99 * Tun/Tap allocation routine
102 * Pointer to private structure.
104 * @param[in] is_keepalive
108 * -1 on failure, fd on success
111 tun_alloc(struct pmd_internals *pmd, int is_keepalive)
114 #ifdef IFF_MULTI_QUEUE
115 unsigned int features;
119 memset(&ifr, 0, sizeof(struct ifreq));
122 * Do not set IFF_NO_PI as packet information header will be needed
123 * to check if a received packet has been truncated.
125 ifr.ifr_flags = (pmd->type == ETH_TUNTAP_TYPE_TAP) ?
126 IFF_TAP : IFF_TUN | IFF_POINTOPOINT;
127 snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
129 TAP_LOG(DEBUG, "ifr_name '%s'", ifr.ifr_name);
131 fd = open(TUN_TAP_DEV_PATH, O_RDWR);
133 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
137 #ifdef IFF_MULTI_QUEUE
138 /* Grab the TUN features to verify we can work multi-queue */
139 if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
140 TAP_LOG(ERR, "%s unable to get TUN/TAP features",
144 TAP_LOG(DEBUG, "%s Features %08x", tuntap_name, features);
146 if (features & IFF_MULTI_QUEUE) {
147 TAP_LOG(DEBUG, " Multi-queue support for %d queues",
148 RTE_PMD_TAP_MAX_QUEUES);
149 ifr.ifr_flags |= IFF_MULTI_QUEUE;
153 ifr.ifr_flags |= IFF_ONE_QUEUE;
154 TAP_LOG(DEBUG, " Single queue only support");
157 /* Set the TUN/TAP configuration and set the name if needed */
158 if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
159 TAP_LOG(WARNING, "Unable to set TUNSETIFF for %s: %s",
160 ifr.ifr_name, strerror(errno));
166 * Detach the TUN/TAP keep-alive queue
167 * to avoid traffic through it
169 ifr.ifr_flags = IFF_DETACH_QUEUE;
170 if (ioctl(fd, TUNSETQUEUE, (void *)&ifr) < 0) {
172 "Unable to detach keep-alive queue for %s: %s",
173 ifr.ifr_name, strerror(errno));
178 /* Always set the file descriptor to non-blocking */
179 if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
181 "Unable to set %s to nonblocking: %s",
182 ifr.ifr_name, strerror(errno));
186 /* Set up trigger to optimize empty Rx bursts */
190 int flags = fcntl(fd, F_GETFL);
192 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
194 if (sa.sa_handler != tap_trigger_cb) {
196 * Make sure SIGIO is not already taken. This is done
197 * as late as possible to leave the application a
198 * chance to set up its own signal handler first.
200 if (sa.sa_handler != SIG_IGN &&
201 sa.sa_handler != SIG_DFL) {
205 sa = (struct sigaction){
206 .sa_flags = SA_RESTART,
207 .sa_handler = tap_trigger_cb,
209 if (sigaction(SIGIO, &sa, NULL) == -1)
212 /* Enable SIGIO on file descriptor */
213 fcntl(fd, F_SETFL, flags | O_ASYNC);
214 fcntl(fd, F_SETOWN, getpid());
218 /* Disable trigger globally in case of error */
220 TAP_LOG(WARNING, "Rx trigger disabled: %s",
233 tap_verify_csum(struct rte_mbuf *mbuf)
235 uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
236 uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
237 uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
238 unsigned int l2_len = sizeof(struct ether_hdr);
244 if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
246 else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
248 /* Don't verify checksum for packets with discontinuous L2 header */
249 if (unlikely(l2_len + sizeof(struct ipv4_hdr) >
250 rte_pktmbuf_data_len(mbuf)))
252 l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
253 if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
254 struct ipv4_hdr *iph = l3_hdr;
256 /* ihl contains the number of 4-byte words in the header */
257 l3_len = 4 * (iph->version_ihl & 0xf);
258 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
261 cksum = ~rte_raw_cksum(iph, l3_len);
262 mbuf->ol_flags |= cksum ?
263 PKT_RX_IP_CKSUM_BAD :
264 PKT_RX_IP_CKSUM_GOOD;
265 } else if (l3 == RTE_PTYPE_L3_IPV6) {
266 l3_len = sizeof(struct ipv6_hdr);
268 /* IPv6 extensions are not supported */
271 if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
272 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
273 /* Don't verify checksum for multi-segment packets. */
274 if (mbuf->nb_segs > 1)
276 if (l3 == RTE_PTYPE_L3_IPV4)
277 cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
278 else if (l3 == RTE_PTYPE_L3_IPV6)
279 cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
280 mbuf->ol_flags |= cksum ?
281 PKT_RX_L4_CKSUM_BAD :
282 PKT_RX_L4_CKSUM_GOOD;
287 tap_rx_offload_get_port_capa(void)
290 * No specific port Rx offload capabilities.
296 tap_rx_offload_get_queue_capa(void)
298 return DEV_RX_OFFLOAD_SCATTER |
299 DEV_RX_OFFLOAD_IPV4_CKSUM |
300 DEV_RX_OFFLOAD_UDP_CKSUM |
301 DEV_RX_OFFLOAD_TCP_CKSUM |
302 DEV_RX_OFFLOAD_CRC_STRIP;
305 /* Callback to handle the rx burst of packets to the correct interface and
306 * file descriptor(s) in a multi-queue setup.
309 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
311 struct rx_queue *rxq = queue;
313 unsigned long num_rx_bytes = 0;
314 uint32_t trigger = tap_trigger;
316 if (trigger == rxq->trigger_seen)
319 rxq->trigger_seen = trigger;
320 rte_compiler_barrier();
321 for (num_rx = 0; num_rx < nb_pkts; ) {
322 struct rte_mbuf *mbuf = rxq->pool;
323 struct rte_mbuf *seg = NULL;
324 struct rte_mbuf *new_tail = NULL;
325 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
328 len = readv(rxq->fd, *rxq->iovecs,
330 (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
331 rxq->nb_rx_desc : 1));
332 if (len < (int)sizeof(struct tun_pi))
335 /* Packet couldn't fit in the provided mbuf */
336 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
337 rxq->stats.ierrors++;
341 len -= sizeof(struct tun_pi);
344 mbuf->port = rxq->in_port;
346 struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
348 if (unlikely(!buf)) {
349 rxq->stats.rx_nombuf++;
350 /* No new buf has been allocated: do nothing */
351 if (!new_tail || !seg)
355 rte_pktmbuf_free(mbuf);
359 seg = seg ? seg->next : mbuf;
360 if (rxq->pool == mbuf)
363 new_tail->next = buf;
365 new_tail->next = seg->next;
367 /* iovecs[0] is reserved for packet info (pi) */
368 (*rxq->iovecs)[mbuf->nb_segs].iov_len =
369 buf->buf_len - data_off;
370 (*rxq->iovecs)[mbuf->nb_segs].iov_base =
371 (char *)buf->buf_addr + data_off;
373 seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
374 seg->data_off = data_off;
376 len -= seg->data_len;
380 /* First segment has headroom, not the others */
384 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
386 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
387 tap_verify_csum(mbuf);
389 /* account for the receive frame */
390 bufs[num_rx++] = mbuf;
391 num_rx_bytes += mbuf->pkt_len;
394 rxq->stats.ipackets += num_rx;
395 rxq->stats.ibytes += num_rx_bytes;
401 tap_tx_offload_get_port_capa(void)
404 * No specific port Tx offload capabilities.
410 tap_tx_offload_get_queue_capa(void)
412 return DEV_TX_OFFLOAD_MULTI_SEGS |
413 DEV_TX_OFFLOAD_IPV4_CKSUM |
414 DEV_TX_OFFLOAD_UDP_CKSUM |
415 DEV_TX_OFFLOAD_TCP_CKSUM;
418 /* Finalize l4 checksum calculation */
420 tap_tx_l4_cksum(uint16_t *l4_cksum, uint16_t l4_phdr_cksum,
421 uint32_t l4_raw_cksum)
426 cksum = __rte_raw_cksum_reduce(l4_raw_cksum);
427 cksum += l4_phdr_cksum;
429 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
430 cksum = (~cksum) & 0xffff;
437 /* Accumaulate L4 raw checksums */
439 tap_tx_l4_add_rcksum(char *l4_data, unsigned int l4_len, uint16_t *l4_cksum,
440 uint32_t *l4_raw_cksum)
442 if (l4_cksum == NULL)
445 *l4_raw_cksum = __rte_raw_cksum(l4_data, l4_len, *l4_raw_cksum);
448 /* L3 and L4 pseudo headers checksum offloads */
450 tap_tx_l3_cksum(char *packet, uint64_t ol_flags, unsigned int l2_len,
451 unsigned int l3_len, unsigned int l4_len, uint16_t **l4_cksum,
452 uint16_t *l4_phdr_cksum, uint32_t *l4_raw_cksum)
454 void *l3_hdr = packet + l2_len;
456 if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
457 struct ipv4_hdr *iph = l3_hdr;
460 iph->hdr_checksum = 0;
461 cksum = rte_raw_cksum(iph, l3_len);
462 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
464 if (ol_flags & PKT_TX_L4_MASK) {
467 l4_hdr = packet + l2_len + l3_len;
468 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
469 *l4_cksum = &((struct udp_hdr *)l4_hdr)->dgram_cksum;
470 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
471 *l4_cksum = &((struct tcp_hdr *)l4_hdr)->cksum;
475 if (ol_flags & PKT_TX_IPV4)
476 *l4_phdr_cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
478 *l4_phdr_cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
479 *l4_raw_cksum = __rte_raw_cksum(l4_hdr, l4_len, 0);
483 /* Callback to handle sending packets from the tap interface
486 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
488 struct tx_queue *txq = queue;
490 unsigned long num_tx_bytes = 0;
494 if (unlikely(nb_pkts == 0))
497 max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
498 for (i = 0; i < nb_pkts; i++) {
499 struct rte_mbuf *mbuf = bufs[num_tx];
500 struct iovec iovecs[mbuf->nb_segs + 2];
501 struct tun_pi pi = { .flags = 0, .proto = 0x00 };
502 struct rte_mbuf *seg = mbuf;
503 char m_copy[mbuf->data_len];
507 int k; /* first index in iovecs for copying segments */
508 uint16_t l234_hlen; /* length of layers 2,3,4 headers */
509 uint16_t seg_len; /* length of first segment */
511 uint16_t *l4_cksum; /* l4 checksum (pseudo header + payload) */
512 uint32_t l4_raw_cksum = 0; /* TCP/UDP payload raw checksum */
513 uint16_t l4_phdr_cksum = 0; /* TCP/UDP pseudo header checksum */
514 uint16_t is_cksum = 0; /* in case cksum should be offloaded */
516 /* stats.errs will be incremented */
517 if (rte_pktmbuf_pkt_len(mbuf) > max_size)
521 if (txq->type == ETH_TUNTAP_TYPE_TUN) {
523 * TUN and TAP are created with IFF_NO_PI disabled.
524 * For TUN PMD this mandatory as fields are used by
525 * Kernel tun.c to determine whether its IP or non IP
528 * The logic fetches the first byte of data from mbuf
529 * then compares whether its v4 or v6. If first byte
530 * is 4 or 6, then protocol field is updated.
532 char *buff_data = rte_pktmbuf_mtod(seg, void *);
533 proto = (*buff_data & 0xf0);
534 pi.proto = (proto == 0x40) ?
535 rte_cpu_to_be_16(ETHER_TYPE_IPv4) :
537 rte_cpu_to_be_16(ETHER_TYPE_IPv6) :
542 iovecs[k].iov_base = π
543 iovecs[k].iov_len = sizeof(pi);
546 nb_segs = mbuf->nb_segs;
548 ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
549 (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
550 (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
553 /* Support only packets with at least layer 4
554 * header included in the first segment
556 seg_len = rte_pktmbuf_data_len(mbuf);
557 l234_hlen = mbuf->l2_len + mbuf->l3_len + mbuf->l4_len;
558 if (seg_len < l234_hlen)
561 /* To change checksums, work on a
562 * copy of l2, l3 l4 headers.
564 rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
566 tap_tx_l3_cksum(m_copy, mbuf->ol_flags,
567 mbuf->l2_len, mbuf->l3_len, mbuf->l4_len,
568 &l4_cksum, &l4_phdr_cksum,
570 iovecs[k].iov_base = m_copy;
571 iovecs[k].iov_len = l234_hlen;
574 /* Update next iovecs[] beyond l2, l3, l4 headers */
575 if (seg_len > l234_hlen) {
576 iovecs[k].iov_len = seg_len - l234_hlen;
578 rte_pktmbuf_mtod(seg, char *) +
580 tap_tx_l4_add_rcksum(iovecs[k].iov_base,
581 iovecs[k].iov_len, l4_cksum,
589 for (j = k; j <= nb_segs; j++) {
590 iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
591 iovecs[j].iov_base = rte_pktmbuf_mtod(seg, void *);
593 tap_tx_l4_add_rcksum(iovecs[j].iov_base,
594 iovecs[j].iov_len, l4_cksum,
600 tap_tx_l4_cksum(l4_cksum, l4_phdr_cksum, l4_raw_cksum);
602 /* copy the tx frame data */
603 n = writev(txq->fd, iovecs, j);
608 num_tx_bytes += mbuf->pkt_len;
609 rte_pktmbuf_free(mbuf);
612 txq->stats.opackets += num_tx;
613 txq->stats.errs += nb_pkts - num_tx;
614 txq->stats.obytes += num_tx_bytes;
620 tap_ioctl_req2str(unsigned long request)
624 return "SIOCSIFFLAGS";
626 return "SIOCGIFFLAGS";
628 return "SIOCGIFHWADDR";
630 return "SIOCSIFHWADDR";
638 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
639 struct ifreq *ifr, int set, enum ioctl_mode mode)
641 short req_flags = ifr->ifr_flags;
642 int remote = pmd->remote_if_index &&
643 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
645 if (!pmd->remote_if_index && mode == REMOTE_ONLY)
648 * If there is a remote netdevice, apply ioctl on it, then apply it on
653 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
654 else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
655 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
658 /* fetch current flags to leave other flags untouched */
659 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
662 ifr->ifr_flags |= req_flags;
664 ifr->ifr_flags &= ~req_flags;
672 RTE_LOG(WARNING, PMD, "%s: ioctl() called with wrong arg\n",
676 if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
678 if (remote-- && mode == LOCAL_AND_REMOTE)
683 TAP_LOG(DEBUG, "%s(%s) failed: %s(%d)", ifr->ifr_name,
684 tap_ioctl_req2str(request), strerror(errno), errno);
689 tap_link_set_down(struct rte_eth_dev *dev)
691 struct pmd_internals *pmd = dev->data->dev_private;
692 struct ifreq ifr = { .ifr_flags = IFF_UP };
694 dev->data->dev_link.link_status = ETH_LINK_DOWN;
695 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
699 tap_link_set_up(struct rte_eth_dev *dev)
701 struct pmd_internals *pmd = dev->data->dev_private;
702 struct ifreq ifr = { .ifr_flags = IFF_UP };
704 dev->data->dev_link.link_status = ETH_LINK_UP;
705 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
709 tap_dev_start(struct rte_eth_dev *dev)
713 err = tap_intr_handle_set(dev, 1);
716 return tap_link_set_up(dev);
719 /* This function gets called when the current port gets stopped.
722 tap_dev_stop(struct rte_eth_dev *dev)
724 tap_intr_handle_set(dev, 0);
725 tap_link_set_down(dev);
729 tap_dev_configure(struct rte_eth_dev *dev)
731 if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
733 "%s: number of rx queues %d exceeds max num of queues %d",
735 dev->data->nb_rx_queues,
736 RTE_PMD_TAP_MAX_QUEUES);
739 if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
741 "%s: number of tx queues %d exceeds max num of queues %d",
743 dev->data->nb_tx_queues,
744 RTE_PMD_TAP_MAX_QUEUES);
748 TAP_LOG(INFO, "%s: %p: TX configured queues number: %u",
749 dev->device->name, (void *)dev, dev->data->nb_tx_queues);
751 TAP_LOG(INFO, "%s: %p: RX configured queues number: %u",
752 dev->device->name, (void *)dev, dev->data->nb_rx_queues);
758 tap_dev_speed_capa(void)
760 uint32_t speed = pmd_link.link_speed;
763 if (speed >= ETH_SPEED_NUM_10M)
764 capa |= ETH_LINK_SPEED_10M;
765 if (speed >= ETH_SPEED_NUM_100M)
766 capa |= ETH_LINK_SPEED_100M;
767 if (speed >= ETH_SPEED_NUM_1G)
768 capa |= ETH_LINK_SPEED_1G;
769 if (speed >= ETH_SPEED_NUM_5G)
770 capa |= ETH_LINK_SPEED_2_5G;
771 if (speed >= ETH_SPEED_NUM_5G)
772 capa |= ETH_LINK_SPEED_5G;
773 if (speed >= ETH_SPEED_NUM_10G)
774 capa |= ETH_LINK_SPEED_10G;
775 if (speed >= ETH_SPEED_NUM_20G)
776 capa |= ETH_LINK_SPEED_20G;
777 if (speed >= ETH_SPEED_NUM_25G)
778 capa |= ETH_LINK_SPEED_25G;
779 if (speed >= ETH_SPEED_NUM_40G)
780 capa |= ETH_LINK_SPEED_40G;
781 if (speed >= ETH_SPEED_NUM_50G)
782 capa |= ETH_LINK_SPEED_50G;
783 if (speed >= ETH_SPEED_NUM_56G)
784 capa |= ETH_LINK_SPEED_56G;
785 if (speed >= ETH_SPEED_NUM_100G)
786 capa |= ETH_LINK_SPEED_100G;
792 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
794 struct pmd_internals *internals = dev->data->dev_private;
796 dev_info->if_index = internals->if_index;
797 dev_info->max_mac_addrs = 1;
798 dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
799 dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
800 dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
801 dev_info->min_rx_bufsize = 0;
802 dev_info->speed_capa = tap_dev_speed_capa();
803 dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
804 dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
805 dev_info->rx_queue_offload_capa;
806 dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
807 dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
808 dev_info->tx_queue_offload_capa;
809 dev_info->hash_key_size = TAP_RSS_HASH_KEY_SIZE;
811 * limitation: TAP supports all of IP, UDP and TCP hash
812 * functions together and not in partial combinations
814 dev_info->flow_type_rss_offloads = ~TAP_RSS_HF_MASK;
818 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
820 unsigned int i, imax;
821 unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
822 unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
823 unsigned long rx_nombuf = 0, ierrors = 0;
824 const struct pmd_internals *pmd = dev->data->dev_private;
826 /* rx queue statistics */
827 imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
828 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
829 for (i = 0; i < imax; i++) {
830 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
831 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
832 rx_total += tap_stats->q_ipackets[i];
833 rx_bytes_total += tap_stats->q_ibytes[i];
834 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
835 ierrors += pmd->rxq[i].stats.ierrors;
838 /* tx queue statistics */
839 imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
840 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
842 for (i = 0; i < imax; i++) {
843 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
844 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
845 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
846 tx_total += tap_stats->q_opackets[i];
847 tx_err_total += tap_stats->q_errors[i];
848 tx_bytes_total += tap_stats->q_obytes[i];
851 tap_stats->ipackets = rx_total;
852 tap_stats->ibytes = rx_bytes_total;
853 tap_stats->ierrors = ierrors;
854 tap_stats->rx_nombuf = rx_nombuf;
855 tap_stats->opackets = tx_total;
856 tap_stats->oerrors = tx_err_total;
857 tap_stats->obytes = tx_bytes_total;
862 tap_stats_reset(struct rte_eth_dev *dev)
865 struct pmd_internals *pmd = dev->data->dev_private;
867 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
868 pmd->rxq[i].stats.ipackets = 0;
869 pmd->rxq[i].stats.ibytes = 0;
870 pmd->rxq[i].stats.ierrors = 0;
871 pmd->rxq[i].stats.rx_nombuf = 0;
873 pmd->txq[i].stats.opackets = 0;
874 pmd->txq[i].stats.errs = 0;
875 pmd->txq[i].stats.obytes = 0;
880 tap_dev_close(struct rte_eth_dev *dev)
883 struct pmd_internals *internals = dev->data->dev_private;
885 tap_link_set_down(dev);
886 tap_flow_flush(dev, NULL);
887 tap_flow_implicit_flush(internals, NULL);
889 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
890 if (internals->rxq[i].fd != -1) {
891 close(internals->rxq[i].fd);
892 internals->rxq[i].fd = -1;
894 if (internals->txq[i].fd != -1) {
895 close(internals->txq[i].fd);
896 internals->txq[i].fd = -1;
900 if (internals->remote_if_index) {
901 /* Restore initial remote state */
902 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
903 &internals->remote_initial_flags);
906 if (internals->ka_fd != -1) {
907 close(internals->ka_fd);
908 internals->ka_fd = -1;
911 * Since TUN device has no more opened file descriptors
912 * it will be removed from kernel
917 tap_rx_queue_release(void *queue)
919 struct rx_queue *rxq = queue;
921 if (rxq && (rxq->fd > 0)) {
924 rte_pktmbuf_free(rxq->pool);
925 rte_free(rxq->iovecs);
932 tap_tx_queue_release(void *queue)
934 struct tx_queue *txq = queue;
936 if (txq && (txq->fd > 0)) {
943 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
945 struct rte_eth_link *dev_link = &dev->data->dev_link;
946 struct pmd_internals *pmd = dev->data->dev_private;
947 struct ifreq ifr = { .ifr_flags = 0 };
949 if (pmd->remote_if_index) {
950 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
951 if (!(ifr.ifr_flags & IFF_UP) ||
952 !(ifr.ifr_flags & IFF_RUNNING)) {
953 dev_link->link_status = ETH_LINK_DOWN;
957 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
958 dev_link->link_status =
959 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
966 tap_promisc_enable(struct rte_eth_dev *dev)
968 struct pmd_internals *pmd = dev->data->dev_private;
969 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
971 dev->data->promiscuous = 1;
972 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
973 if (pmd->remote_if_index && !pmd->flow_isolate)
974 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
978 tap_promisc_disable(struct rte_eth_dev *dev)
980 struct pmd_internals *pmd = dev->data->dev_private;
981 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
983 dev->data->promiscuous = 0;
984 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
985 if (pmd->remote_if_index && !pmd->flow_isolate)
986 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
990 tap_allmulti_enable(struct rte_eth_dev *dev)
992 struct pmd_internals *pmd = dev->data->dev_private;
993 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
995 dev->data->all_multicast = 1;
996 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
997 if (pmd->remote_if_index && !pmd->flow_isolate)
998 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
1002 tap_allmulti_disable(struct rte_eth_dev *dev)
1004 struct pmd_internals *pmd = dev->data->dev_private;
1005 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1007 dev->data->all_multicast = 0;
1008 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1009 if (pmd->remote_if_index && !pmd->flow_isolate)
1010 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
1014 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1016 struct pmd_internals *pmd = dev->data->dev_private;
1017 enum ioctl_mode mode = LOCAL_ONLY;
1021 if (pmd->type == ETH_TUNTAP_TYPE_TUN) {
1022 TAP_LOG(ERR, "%s: can't MAC address for TUN",
1027 if (is_zero_ether_addr(mac_addr)) {
1028 TAP_LOG(ERR, "%s: can't set an empty MAC address",
1032 /* Check the actual current MAC address on the tap netdevice */
1033 ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY);
1036 if (is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
1039 /* Check the current MAC address on the remote */
1040 ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY);
1043 if (!is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
1045 mode = LOCAL_AND_REMOTE;
1046 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1047 rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
1048 ret = tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode);
1051 rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
1052 if (pmd->remote_if_index && !pmd->flow_isolate) {
1053 /* Replace MAC redirection rule after a MAC change */
1054 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC);
1057 "%s: Couldn't delete MAC redirection rule",
1061 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC);
1064 "%s: Couldn't add MAC redirection rule",
1074 tap_setup_queue(struct rte_eth_dev *dev,
1075 struct pmd_internals *internals,
1082 struct pmd_internals *pmd = dev->data->dev_private;
1083 struct rx_queue *rx = &internals->rxq[qid];
1084 struct tx_queue *tx = &internals->txq[qid];
1096 /* fd for this queue already exists */
1097 TAP_LOG(DEBUG, "%s: fd %d for %s queue qid %d exists",
1098 pmd->name, *fd, dir, qid);
1099 } else if (*other_fd != -1) {
1100 /* Only other_fd exists. dup it */
1101 *fd = dup(*other_fd);
1104 TAP_LOG(ERR, "%s: dup() failed.", pmd->name);
1107 TAP_LOG(DEBUG, "%s: dup fd %d for %s queue qid %d (%d)",
1108 pmd->name, *other_fd, dir, qid, *fd);
1110 /* Both RX and TX fds do not exist (equal -1). Create fd */
1111 *fd = tun_alloc(pmd, 0);
1113 *fd = -1; /* restore original value */
1114 TAP_LOG(ERR, "%s: tun_alloc() failed.", pmd->name);
1117 TAP_LOG(DEBUG, "%s: add %s queue for qid %d fd %d",
1118 pmd->name, dir, qid, *fd);
1121 tx->mtu = &dev->data->mtu;
1122 rx->rxmode = &dev->data->dev_conf.rxmode;
1124 tx->type = pmd->type;
1130 tap_rx_queue_setup(struct rte_eth_dev *dev,
1131 uint16_t rx_queue_id,
1132 uint16_t nb_rx_desc,
1133 unsigned int socket_id,
1134 const struct rte_eth_rxconf *rx_conf __rte_unused,
1135 struct rte_mempool *mp)
1137 struct pmd_internals *internals = dev->data->dev_private;
1138 struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1139 struct rte_mbuf **tmp = &rxq->pool;
1140 long iov_max = sysconf(_SC_IOV_MAX);
1141 uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1142 struct iovec (*iovecs)[nb_desc + 1];
1143 int data_off = RTE_PKTMBUF_HEADROOM;
1148 if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1150 "nb_rx_queues %d too small or mempool NULL",
1151 dev->data->nb_rx_queues);
1156 rxq->trigger_seen = 1; /* force initial burst */
1157 rxq->in_port = dev->data->port_id;
1158 rxq->nb_rx_desc = nb_desc;
1159 iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1163 "%s: Couldn't allocate %d RX descriptors",
1164 dev->device->name, nb_desc);
1167 rxq->iovecs = iovecs;
1169 dev->data->rx_queues[rx_queue_id] = rxq;
1170 fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1176 (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1177 (*rxq->iovecs)[0].iov_base = &rxq->pi;
1179 for (i = 1; i <= nb_desc; i++) {
1180 *tmp = rte_pktmbuf_alloc(rxq->mp);
1183 "%s: couldn't allocate memory for queue %d",
1184 dev->device->name, rx_queue_id);
1188 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1189 (*rxq->iovecs)[i].iov_base =
1190 (char *)(*tmp)->buf_addr + data_off;
1192 tmp = &(*tmp)->next;
1195 TAP_LOG(DEBUG, " RX TUNTAP device name %s, qid %d on fd %d",
1196 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
1201 rte_pktmbuf_free(rxq->pool);
1203 rte_free(rxq->iovecs);
1209 tap_tx_queue_setup(struct rte_eth_dev *dev,
1210 uint16_t tx_queue_id,
1211 uint16_t nb_tx_desc __rte_unused,
1212 unsigned int socket_id __rte_unused,
1213 const struct rte_eth_txconf *tx_conf)
1215 struct pmd_internals *internals = dev->data->dev_private;
1216 struct tx_queue *txq;
1220 if (tx_queue_id >= dev->data->nb_tx_queues)
1222 dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1223 txq = dev->data->tx_queues[tx_queue_id];
1225 offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1226 txq->csum = !!(offloads &
1227 (DEV_TX_OFFLOAD_IPV4_CKSUM |
1228 DEV_TX_OFFLOAD_UDP_CKSUM |
1229 DEV_TX_OFFLOAD_TCP_CKSUM));
1231 ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1235 " TX TUNTAP device name %s, qid %d on fd %d csum %s",
1236 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd,
1237 txq->csum ? "on" : "off");
1243 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1245 struct pmd_internals *pmd = dev->data->dev_private;
1246 struct ifreq ifr = { .ifr_mtu = mtu };
1249 err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1251 dev->data->mtu = mtu;
1257 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1258 struct ether_addr *mc_addr_set __rte_unused,
1259 uint32_t nb_mc_addr __rte_unused)
1262 * Nothing to do actually: the tap has no filtering whatsoever, every
1263 * packet is received.
1269 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1271 struct rte_eth_dev *dev = arg;
1272 struct pmd_internals *pmd = dev->data->dev_private;
1273 struct ifinfomsg *info = NLMSG_DATA(nh);
1275 if (nh->nlmsg_type != RTM_NEWLINK ||
1276 (info->ifi_index != pmd->if_index &&
1277 info->ifi_index != pmd->remote_if_index))
1279 return tap_link_update(dev, 0);
1283 tap_dev_intr_handler(void *cb_arg)
1285 struct rte_eth_dev *dev = cb_arg;
1286 struct pmd_internals *pmd = dev->data->dev_private;
1288 tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1292 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1294 struct pmd_internals *pmd = dev->data->dev_private;
1296 /* In any case, disable interrupt if the conf is no longer there. */
1297 if (!dev->data->dev_conf.intr_conf.lsc) {
1298 if (pmd->intr_handle.fd != -1) {
1299 tap_nl_final(pmd->intr_handle.fd);
1300 rte_intr_callback_unregister(&pmd->intr_handle,
1301 tap_dev_intr_handler, dev);
1306 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1307 if (unlikely(pmd->intr_handle.fd == -1))
1309 return rte_intr_callback_register(
1310 &pmd->intr_handle, tap_dev_intr_handler, dev);
1312 tap_nl_final(pmd->intr_handle.fd);
1313 return rte_intr_callback_unregister(&pmd->intr_handle,
1314 tap_dev_intr_handler, dev);
1318 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1322 err = tap_lsc_intr_handle_set(dev, set);
1325 err = tap_rx_intr_vec_set(dev, set);
1327 tap_lsc_intr_handle_set(dev, 0);
1331 static const uint32_t*
1332 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1334 static const uint32_t ptypes[] = {
1335 RTE_PTYPE_INNER_L2_ETHER,
1336 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1337 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1338 RTE_PTYPE_INNER_L3_IPV4,
1339 RTE_PTYPE_INNER_L3_IPV4_EXT,
1340 RTE_PTYPE_INNER_L3_IPV6,
1341 RTE_PTYPE_INNER_L3_IPV6_EXT,
1342 RTE_PTYPE_INNER_L4_FRAG,
1343 RTE_PTYPE_INNER_L4_UDP,
1344 RTE_PTYPE_INNER_L4_TCP,
1345 RTE_PTYPE_INNER_L4_SCTP,
1347 RTE_PTYPE_L2_ETHER_VLAN,
1348 RTE_PTYPE_L2_ETHER_QINQ,
1350 RTE_PTYPE_L3_IPV4_EXT,
1351 RTE_PTYPE_L3_IPV6_EXT,
1363 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1364 struct rte_eth_fc_conf *fc_conf)
1366 fc_conf->mode = RTE_FC_NONE;
1371 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1372 struct rte_eth_fc_conf *fc_conf)
1374 if (fc_conf->mode != RTE_FC_NONE)
1380 * DPDK callback to update the RSS hash configuration.
1383 * Pointer to Ethernet device structure.
1384 * @param[in] rss_conf
1385 * RSS configuration data.
1388 * 0 on success, a negative errno value otherwise and rte_errno is set.
1391 tap_rss_hash_update(struct rte_eth_dev *dev,
1392 struct rte_eth_rss_conf *rss_conf)
1394 if (rss_conf->rss_hf & TAP_RSS_HF_MASK) {
1398 if (rss_conf->rss_key && rss_conf->rss_key_len) {
1400 * Currently TAP RSS key is hard coded
1401 * and cannot be updated
1404 "port %u RSS key cannot be updated",
1405 dev->data->port_id);
1412 static const struct eth_dev_ops ops = {
1413 .dev_start = tap_dev_start,
1414 .dev_stop = tap_dev_stop,
1415 .dev_close = tap_dev_close,
1416 .dev_configure = tap_dev_configure,
1417 .dev_infos_get = tap_dev_info,
1418 .rx_queue_setup = tap_rx_queue_setup,
1419 .tx_queue_setup = tap_tx_queue_setup,
1420 .rx_queue_release = tap_rx_queue_release,
1421 .tx_queue_release = tap_tx_queue_release,
1422 .flow_ctrl_get = tap_flow_ctrl_get,
1423 .flow_ctrl_set = tap_flow_ctrl_set,
1424 .link_update = tap_link_update,
1425 .dev_set_link_up = tap_link_set_up,
1426 .dev_set_link_down = tap_link_set_down,
1427 .promiscuous_enable = tap_promisc_enable,
1428 .promiscuous_disable = tap_promisc_disable,
1429 .allmulticast_enable = tap_allmulti_enable,
1430 .allmulticast_disable = tap_allmulti_disable,
1431 .mac_addr_set = tap_mac_set,
1432 .mtu_set = tap_mtu_set,
1433 .set_mc_addr_list = tap_set_mc_addr_list,
1434 .stats_get = tap_stats_get,
1435 .stats_reset = tap_stats_reset,
1436 .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1437 .rss_hash_update = tap_rss_hash_update,
1438 .filter_ctrl = tap_dev_filter_ctrl,
1442 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
1443 char *remote_iface, struct ether_addr *mac_addr,
1444 enum rte_tuntap_type type)
1446 int numa_node = rte_socket_id();
1447 struct rte_eth_dev *dev;
1448 struct pmd_internals *pmd;
1449 struct rte_eth_dev_data *data;
1453 TAP_LOG(DEBUG, "%s device on numa %u",
1454 tuntap_name, rte_socket_id());
1456 dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1458 TAP_LOG(ERR, "%s Unable to allocate device struct",
1460 goto error_exit_nodev;
1463 pmd = dev->data->dev_private;
1465 snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1468 pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1469 if (pmd->ioctl_sock == -1) {
1471 "%s Unable to get a socket for management: %s",
1472 tuntap_name, strerror(errno));
1476 /* Setup some default values */
1478 data->dev_private = pmd;
1479 data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1480 data->numa_node = numa_node;
1482 data->dev_link = pmd_link;
1483 data->mac_addrs = &pmd->eth_addr;
1484 /* Set the number of RX and TX queues */
1485 data->nb_rx_queues = 0;
1486 data->nb_tx_queues = 0;
1488 dev->dev_ops = &ops;
1489 dev->rx_pkt_burst = pmd_rx_burst;
1490 dev->tx_pkt_burst = pmd_tx_burst;
1492 pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1493 pmd->intr_handle.fd = -1;
1494 dev->intr_handle = &pmd->intr_handle;
1496 /* Presetup the fds to -1 as being not valid */
1498 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1499 pmd->rxq[i].fd = -1;
1500 pmd->txq[i].fd = -1;
1503 if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1504 if (is_zero_ether_addr(mac_addr))
1505 eth_random_addr((uint8_t *)&pmd->eth_addr);
1507 rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1511 * Allocate a TUN device keep-alive file descriptor that will only be
1512 * closed when the TUN device itself is closed or removed.
1513 * This keep-alive file descriptor will guarantee that the TUN device
1514 * exists even when all of its queues are closed
1516 pmd->ka_fd = tun_alloc(pmd, 1);
1517 if (pmd->ka_fd == -1) {
1518 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
1522 ifr.ifr_mtu = dev->data->mtu;
1523 if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1526 if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1527 memset(&ifr, 0, sizeof(struct ifreq));
1528 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1529 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
1531 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1536 * Set up everything related to rte_flow:
1538 * - tap / remote if_index
1539 * - mandatory QDISCs
1540 * - rte_flow actual/implicit lists
1543 pmd->nlsk_fd = tap_nl_init(0);
1544 if (pmd->nlsk_fd == -1) {
1545 TAP_LOG(WARNING, "%s: failed to create netlink socket.",
1547 goto disable_rte_flow;
1549 pmd->if_index = if_nametoindex(pmd->name);
1550 if (!pmd->if_index) {
1551 TAP_LOG(ERR, "%s: failed to get if_index.", pmd->name);
1552 goto disable_rte_flow;
1554 if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1555 TAP_LOG(ERR, "%s: failed to create multiq qdisc.",
1557 goto disable_rte_flow;
1559 if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1560 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1562 goto disable_rte_flow;
1564 LIST_INIT(&pmd->flows);
1566 if (strlen(remote_iface)) {
1567 pmd->remote_if_index = if_nametoindex(remote_iface);
1568 if (!pmd->remote_if_index) {
1569 TAP_LOG(ERR, "%s: failed to get %s if_index.",
1570 pmd->name, remote_iface);
1573 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1574 "%s", remote_iface);
1576 /* Save state of remote device */
1577 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1579 /* Replicate remote MAC address */
1580 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1581 TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1582 pmd->name, pmd->remote_iface);
1585 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1587 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1588 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1589 TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1590 pmd->name, remote_iface);
1595 * Flush usually returns negative value because it tries to
1596 * delete every QDISC (and on a running device, one QDISC at
1597 * least is needed). Ignore negative return value.
1599 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1600 if (qdisc_create_ingress(pmd->nlsk_fd,
1601 pmd->remote_if_index) < 0) {
1602 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1606 LIST_INIT(&pmd->implicit_flows);
1607 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1608 tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1609 tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1610 tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1612 "%s: failed to create implicit rules.",
1618 rte_eth_dev_probing_finish(dev);
1622 TAP_LOG(ERR, " Disabling rte flow support: %s(%d)",
1623 strerror(errno), errno);
1624 if (strlen(remote_iface)) {
1625 TAP_LOG(ERR, "Remote feature requires flow support.");
1631 TAP_LOG(ERR, " Can't set up remote feature: %s(%d)",
1632 strerror(errno), errno);
1633 tap_flow_implicit_flush(pmd, NULL);
1636 if (pmd->ioctl_sock > 0)
1637 close(pmd->ioctl_sock);
1638 rte_eth_dev_release_port(dev);
1641 TAP_LOG(ERR, "%s Unable to initialize %s",
1642 tuntap_name, rte_vdev_device_name(vdev));
1648 set_interface_name(const char *key __rte_unused,
1652 char *name = (char *)extra_args;
1655 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN - 1);
1657 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1658 DEFAULT_TAP_NAME, (tap_unit - 1));
1664 set_remote_iface(const char *key __rte_unused,
1668 char *name = (char *)extra_args;
1671 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
1676 static int parse_user_mac(struct ether_addr *user_mac,
1679 unsigned int index = 0;
1680 char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
1682 if (user_mac == NULL || value == NULL)
1685 strlcpy(mac_temp, value, sizeof(mac_temp));
1686 mac_byte = strtok(mac_temp, ":");
1688 while ((mac_byte != NULL) &&
1689 (strlen(mac_byte) <= 2) &&
1690 (strlen(mac_byte) == strspn(mac_byte,
1691 ETH_TAP_CMP_MAC_FMT))) {
1692 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
1693 mac_byte = strtok(NULL, ":");
1700 set_mac_type(const char *key __rte_unused,
1704 struct ether_addr *user_mac = extra_args;
1709 if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
1710 static int iface_idx;
1712 /* fixed mac = 00:64:74:61:70:<iface_idx> */
1713 memcpy((char *)user_mac->addr_bytes, "\0dtap", ETHER_ADDR_LEN);
1714 user_mac->addr_bytes[ETHER_ADDR_LEN - 1] = iface_idx++ + '0';
1718 if (parse_user_mac(user_mac, value) != 6)
1721 TAP_LOG(DEBUG, "TAP user MAC param (%s)", value);
1725 TAP_LOG(ERR, "TAP user MAC (%s) is not in format (%s|%s)",
1726 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
1731 * Open a TUN interface device. TUN PMD
1732 * 1) sets tap_type as false
1733 * 2) intakes iface as argument.
1734 * 3) as interface is virtual set speed to 10G
1737 rte_pmd_tun_probe(struct rte_vdev_device *dev)
1739 const char *name, *params;
1741 struct rte_kvargs *kvlist = NULL;
1742 char tun_name[RTE_ETH_NAME_MAX_LEN];
1743 char remote_iface[RTE_ETH_NAME_MAX_LEN];
1744 struct rte_eth_dev *eth_dev;
1746 strcpy(tuntap_name, "TUN");
1748 name = rte_vdev_device_name(dev);
1749 params = rte_vdev_device_args(dev);
1750 memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1752 if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
1753 strlen(params) == 0) {
1754 eth_dev = rte_eth_dev_attach_secondary(name);
1756 TAP_LOG(ERR, "Failed to probe %s", name);
1759 eth_dev->dev_ops = &ops;
1763 snprintf(tun_name, sizeof(tun_name), "%s%u",
1764 DEFAULT_TUN_NAME, tun_unit++);
1766 if (params && (params[0] != '\0')) {
1767 TAP_LOG(DEBUG, "parameters (%s)", params);
1769 kvlist = rte_kvargs_parse(params, valid_arguments);
1771 if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1772 ret = rte_kvargs_process(kvlist,
1774 &set_interface_name,
1782 pmd_link.link_speed = ETH_SPEED_NUM_10G;
1784 TAP_LOG(NOTICE, "Initializing pmd_tun for %s as %s",
1787 ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0,
1788 ETH_TUNTAP_TYPE_TUN);
1792 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
1794 tun_unit--; /* Restore the unit number */
1796 rte_kvargs_free(kvlist);
1801 /* Open a TAP interface device.
1804 rte_pmd_tap_probe(struct rte_vdev_device *dev)
1806 const char *name, *params;
1808 struct rte_kvargs *kvlist = NULL;
1810 char tap_name[RTE_ETH_NAME_MAX_LEN];
1811 char remote_iface[RTE_ETH_NAME_MAX_LEN];
1812 struct ether_addr user_mac = { .addr_bytes = {0} };
1813 struct rte_eth_dev *eth_dev;
1815 strcpy(tuntap_name, "TAP");
1817 name = rte_vdev_device_name(dev);
1818 params = rte_vdev_device_args(dev);
1820 if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
1821 strlen(params) == 0) {
1822 eth_dev = rte_eth_dev_attach_secondary(name);
1824 TAP_LOG(ERR, "Failed to probe %s", name);
1827 /* TODO: request info from primary to set up Rx and Tx */
1828 eth_dev->dev_ops = &ops;
1829 rte_eth_dev_probing_finish(eth_dev);
1833 speed = ETH_SPEED_NUM_10G;
1834 snprintf(tap_name, sizeof(tap_name), "%s%u",
1835 DEFAULT_TAP_NAME, tap_unit++);
1836 memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1838 if (params && (params[0] != '\0')) {
1839 TAP_LOG(DEBUG, "parameters (%s)", params);
1841 kvlist = rte_kvargs_parse(params, valid_arguments);
1843 if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1844 ret = rte_kvargs_process(kvlist,
1846 &set_interface_name,
1852 if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
1853 ret = rte_kvargs_process(kvlist,
1861 if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
1862 ret = rte_kvargs_process(kvlist,
1871 pmd_link.link_speed = speed;
1873 TAP_LOG(NOTICE, "Initializing pmd_tap for %s as %s",
1876 ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac,
1877 ETH_TUNTAP_TYPE_TAP);
1881 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
1883 tap_unit--; /* Restore the unit number */
1885 rte_kvargs_free(kvlist);
1890 /* detach a TUNTAP device.
1893 rte_pmd_tap_remove(struct rte_vdev_device *dev)
1895 struct rte_eth_dev *eth_dev = NULL;
1896 struct pmd_internals *internals;
1899 /* find the ethdev entry */
1900 eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1904 internals = eth_dev->data->dev_private;
1906 TAP_LOG(DEBUG, "Closing %s Ethernet device on numa %u",
1907 (internals->type == ETH_TUNTAP_TYPE_TAP) ? "TAP" : "TUN",
1910 if (internals->nlsk_fd) {
1911 tap_flow_flush(eth_dev, NULL);
1912 tap_flow_implicit_flush(internals, NULL);
1913 tap_nl_final(internals->nlsk_fd);
1915 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1916 if (internals->rxq[i].fd != -1) {
1917 close(internals->rxq[i].fd);
1918 internals->rxq[i].fd = -1;
1920 if (internals->txq[i].fd != -1) {
1921 close(internals->txq[i].fd);
1922 internals->txq[i].fd = -1;
1926 close(internals->ioctl_sock);
1927 rte_free(eth_dev->data->dev_private);
1928 rte_eth_dev_release_port(eth_dev);
1930 if (internals->ka_fd != -1) {
1931 close(internals->ka_fd);
1932 internals->ka_fd = -1;
1937 static struct rte_vdev_driver pmd_tun_drv = {
1938 .probe = rte_pmd_tun_probe,
1939 .remove = rte_pmd_tap_remove,
1942 static struct rte_vdev_driver pmd_tap_drv = {
1943 .probe = rte_pmd_tap_probe,
1944 .remove = rte_pmd_tap_remove,
1947 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1948 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
1949 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1950 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
1951 ETH_TAP_IFACE_ARG "=<string> ");
1952 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
1953 ETH_TAP_IFACE_ARG "=<string> "
1954 ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
1955 ETH_TAP_REMOTE_ARG "=<string>");
1958 RTE_INIT(tap_init_log);
1962 tap_logtype = rte_log_register("pmd.net.tap");
1963 if (tap_logtype >= 0)
1964 rte_log_set_level(tap_logtype, RTE_LOG_NOTICE);