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>
38 #include <rte_eth_tap.h>
40 #include <tap_netlink.h>
41 #include <tap_tcmsgs.h>
43 /* Linux based path to the TUN device */
44 #define TUN_TAP_DEV_PATH "/dev/net/tun"
45 #define DEFAULT_TAP_NAME "dtap"
46 #define DEFAULT_TUN_NAME "dtun"
48 #define ETH_TAP_IFACE_ARG "iface"
49 #define ETH_TAP_REMOTE_ARG "remote"
50 #define ETH_TAP_MAC_ARG "mac"
51 #define ETH_TAP_MAC_FIXED "fixed"
53 #define ETH_TAP_USR_MAC_FMT "xx:xx:xx:xx:xx:xx"
54 #define ETH_TAP_CMP_MAC_FMT "0123456789ABCDEFabcdef"
55 #define ETH_TAP_MAC_ARG_FMT ETH_TAP_MAC_FIXED "|" ETH_TAP_USR_MAC_FMT
57 static struct rte_vdev_driver pmd_tap_drv;
58 static struct rte_vdev_driver pmd_tun_drv;
60 static const char *valid_arguments[] = {
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_AUTONEG
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);
98 /* Tun/Tap allocation routine
100 * name is the number of the interface to use, unless NULL to take the host
104 tun_alloc(struct pmd_internals *pmd)
107 #ifdef IFF_MULTI_QUEUE
108 unsigned int features;
112 memset(&ifr, 0, sizeof(struct ifreq));
115 * Do not set IFF_NO_PI as packet information header will be needed
116 * to check if a received packet has been truncated.
118 ifr.ifr_flags = (tap_type) ? IFF_TAP : IFF_TUN | IFF_POINTOPOINT;
119 snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
121 RTE_LOG(DEBUG, PMD, "ifr_name '%s'\n", ifr.ifr_name);
123 fd = open(TUN_TAP_DEV_PATH, O_RDWR);
125 RTE_LOG(ERR, PMD, "Unable to create TAP interface\n");
129 #ifdef IFF_MULTI_QUEUE
130 /* Grab the TUN features to verify we can work multi-queue */
131 if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
132 RTE_LOG(ERR, PMD, "TAP unable to get TUN/TAP features\n");
135 RTE_LOG(DEBUG, PMD, " TAP Features %08x\n", features);
137 if (features & IFF_MULTI_QUEUE) {
138 RTE_LOG(DEBUG, PMD, " Multi-queue support for %d queues\n",
139 RTE_PMD_TAP_MAX_QUEUES);
140 ifr.ifr_flags |= IFF_MULTI_QUEUE;
144 ifr.ifr_flags |= IFF_ONE_QUEUE;
145 RTE_LOG(DEBUG, PMD, " Single queue only support\n");
148 /* Set the TUN/TAP configuration and set the name if needed */
149 if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
150 RTE_LOG(WARNING, PMD,
151 "Unable to set TUNSETIFF for %s\n",
157 /* Always set the file descriptor to non-blocking */
158 if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
159 RTE_LOG(WARNING, PMD,
160 "Unable to set %s to nonblocking\n",
162 perror("F_SETFL, NONBLOCK");
166 /* Set up trigger to optimize empty Rx bursts */
170 int flags = fcntl(fd, F_GETFL);
172 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
174 if (sa.sa_handler != tap_trigger_cb) {
176 * Make sure SIGIO is not already taken. This is done
177 * as late as possible to leave the application a
178 * chance to set up its own signal handler first.
180 if (sa.sa_handler != SIG_IGN &&
181 sa.sa_handler != SIG_DFL) {
185 sa = (struct sigaction){
186 .sa_flags = SA_RESTART,
187 .sa_handler = tap_trigger_cb,
189 if (sigaction(SIGIO, &sa, NULL) == -1)
192 /* Enable SIGIO on file descriptor */
193 fcntl(fd, F_SETFL, flags | O_ASYNC);
194 fcntl(fd, F_SETOWN, getpid());
197 /* Disable trigger globally in case of error */
199 RTE_LOG(WARNING, PMD, "Rx trigger disabled: %s\n",
212 tap_verify_csum(struct rte_mbuf *mbuf)
214 uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
215 uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
216 uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
217 unsigned int l2_len = sizeof(struct ether_hdr);
223 if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
225 else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
227 /* Don't verify checksum for packets with discontinuous L2 header */
228 if (unlikely(l2_len + sizeof(struct ipv4_hdr) >
229 rte_pktmbuf_data_len(mbuf)))
231 l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
232 if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
233 struct ipv4_hdr *iph = l3_hdr;
235 /* ihl contains the number of 4-byte words in the header */
236 l3_len = 4 * (iph->version_ihl & 0xf);
237 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
240 cksum = ~rte_raw_cksum(iph, l3_len);
241 mbuf->ol_flags |= cksum ?
242 PKT_RX_IP_CKSUM_BAD :
243 PKT_RX_IP_CKSUM_GOOD;
244 } else if (l3 == RTE_PTYPE_L3_IPV6) {
245 l3_len = sizeof(struct ipv6_hdr);
247 /* IPv6 extensions are not supported */
250 if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
251 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
252 /* Don't verify checksum for multi-segment packets. */
253 if (mbuf->nb_segs > 1)
255 if (l3 == RTE_PTYPE_L3_IPV4)
256 cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
257 else if (l3 == RTE_PTYPE_L3_IPV6)
258 cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
259 mbuf->ol_flags |= cksum ?
260 PKT_RX_L4_CKSUM_BAD :
261 PKT_RX_L4_CKSUM_GOOD;
266 tap_rx_offload_get_port_capa(void)
269 * In order to support legacy apps,
270 * report capabilities also as port capabilities.
272 return DEV_RX_OFFLOAD_SCATTER |
273 DEV_RX_OFFLOAD_IPV4_CKSUM |
274 DEV_RX_OFFLOAD_UDP_CKSUM |
275 DEV_RX_OFFLOAD_TCP_CKSUM |
276 DEV_RX_OFFLOAD_CRC_STRIP;
280 tap_rx_offload_get_queue_capa(void)
282 return DEV_RX_OFFLOAD_SCATTER |
283 DEV_RX_OFFLOAD_IPV4_CKSUM |
284 DEV_RX_OFFLOAD_UDP_CKSUM |
285 DEV_RX_OFFLOAD_TCP_CKSUM |
286 DEV_RX_OFFLOAD_CRC_STRIP;
290 tap_rxq_are_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
292 uint64_t port_offloads = dev->data->dev_conf.rxmode.offloads;
293 uint64_t queue_supp_offloads = tap_rx_offload_get_queue_capa();
294 uint64_t port_supp_offloads = tap_rx_offload_get_port_capa();
296 if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
299 if ((port_offloads ^ offloads) & port_supp_offloads)
304 /* Callback to handle the rx burst of packets to the correct interface and
305 * file descriptor(s) in a multi-queue setup.
308 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
310 struct rx_queue *rxq = queue;
312 unsigned long num_rx_bytes = 0;
313 uint32_t trigger = tap_trigger;
315 if (trigger == rxq->trigger_seen)
318 rxq->trigger_seen = trigger;
319 rte_compiler_barrier();
320 for (num_rx = 0; num_rx < nb_pkts; ) {
321 struct rte_mbuf *mbuf = rxq->pool;
322 struct rte_mbuf *seg = NULL;
323 struct rte_mbuf *new_tail = NULL;
324 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
327 len = readv(rxq->fd, *rxq->iovecs,
329 (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
330 rxq->nb_rx_desc : 1));
331 if (len < (int)sizeof(struct tun_pi))
334 /* Packet couldn't fit in the provided mbuf */
335 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
336 rxq->stats.ierrors++;
340 len -= sizeof(struct tun_pi);
343 mbuf->port = rxq->in_port;
345 struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
347 if (unlikely(!buf)) {
348 rxq->stats.rx_nombuf++;
349 /* No new buf has been allocated: do nothing */
350 if (!new_tail || !seg)
354 rte_pktmbuf_free(mbuf);
358 seg = seg ? seg->next : mbuf;
359 if (rxq->pool == mbuf)
362 new_tail->next = buf;
364 new_tail->next = seg->next;
366 /* iovecs[0] is reserved for packet info (pi) */
367 (*rxq->iovecs)[mbuf->nb_segs].iov_len =
368 buf->buf_len - data_off;
369 (*rxq->iovecs)[mbuf->nb_segs].iov_base =
370 (char *)buf->buf_addr + data_off;
372 seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
373 seg->data_off = data_off;
375 len -= seg->data_len;
379 /* First segment has headroom, not the others */
383 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
385 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
386 tap_verify_csum(mbuf);
388 /* account for the receive frame */
389 bufs[num_rx++] = mbuf;
390 num_rx_bytes += mbuf->pkt_len;
393 rxq->stats.ipackets += num_rx;
394 rxq->stats.ibytes += num_rx_bytes;
400 tap_tx_offload_get_port_capa(void)
403 * In order to support legacy apps,
404 * report capabilities also as port capabilities.
406 return DEV_TX_OFFLOAD_MULTI_SEGS |
407 DEV_TX_OFFLOAD_IPV4_CKSUM |
408 DEV_TX_OFFLOAD_UDP_CKSUM |
409 DEV_TX_OFFLOAD_TCP_CKSUM;
413 tap_tx_offload_get_queue_capa(void)
415 return DEV_TX_OFFLOAD_MULTI_SEGS |
416 DEV_TX_OFFLOAD_IPV4_CKSUM |
417 DEV_TX_OFFLOAD_UDP_CKSUM |
418 DEV_TX_OFFLOAD_TCP_CKSUM;
422 tap_txq_are_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
424 uint64_t port_offloads = dev->data->dev_conf.txmode.offloads;
425 uint64_t queue_supp_offloads = tap_tx_offload_get_queue_capa();
426 uint64_t port_supp_offloads = tap_tx_offload_get_port_capa();
428 if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
431 /* Verify we have no conflict with port offloads */
432 if ((port_offloads ^ offloads) & port_supp_offloads)
438 tap_tx_offload(char *packet, uint64_t ol_flags, unsigned int l2_len,
441 void *l3_hdr = packet + l2_len;
443 if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
444 struct ipv4_hdr *iph = l3_hdr;
447 iph->hdr_checksum = 0;
448 cksum = rte_raw_cksum(iph, l3_len);
449 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
451 if (ol_flags & PKT_TX_L4_MASK) {
457 l4_hdr = packet + l2_len + l3_len;
458 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
459 l4_cksum = &((struct udp_hdr *)l4_hdr)->dgram_cksum;
460 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
461 l4_cksum = &((struct tcp_hdr *)l4_hdr)->cksum;
465 if (ol_flags & PKT_TX_IPV4) {
466 struct ipv4_hdr *iph = l3_hdr;
468 l4_len = rte_be_to_cpu_16(iph->total_length) - l3_len;
469 cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
471 struct ipv6_hdr *ip6h = l3_hdr;
473 /* payload_len does not include ext headers */
474 l4_len = rte_be_to_cpu_16(ip6h->payload_len) -
475 l3_len + sizeof(struct ipv6_hdr);
476 cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
478 cksum += rte_raw_cksum(l4_hdr, l4_len);
479 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
480 cksum = (~cksum) & 0xffff;
487 /* Callback to handle sending packets from the tap interface
490 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
492 struct tx_queue *txq = queue;
494 unsigned long num_tx_bytes = 0;
498 if (unlikely(nb_pkts == 0))
501 max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
502 for (i = 0; i < nb_pkts; i++) {
503 struct rte_mbuf *mbuf = bufs[num_tx];
504 struct iovec iovecs[mbuf->nb_segs + 1];
505 struct tun_pi pi = { .flags = 0, .proto = 0x00 };
506 struct rte_mbuf *seg = mbuf;
507 char m_copy[mbuf->data_len];
511 /* stats.errs will be incremented */
512 if (rte_pktmbuf_pkt_len(mbuf) > max_size)
516 * TUN and TAP are created with IFF_NO_PI disabled.
517 * For TUN PMD this mandatory as fields are used by
518 * Kernel tun.c to determine whether its IP or non IP
521 * The logic fetches the first byte of data from mbuf.
522 * compares whether its v4 or v6. If none matches default
523 * value 0x00 is taken for protocol field.
525 char *buff_data = rte_pktmbuf_mtod(seg, void *);
526 j = (*buff_data & 0xf0);
527 if (j & (0x40 | 0x60))
528 pi.proto = (j == 0x40) ? 0x0008 : 0xdd86;
530 iovecs[0].iov_base = π
531 iovecs[0].iov_len = sizeof(pi);
532 for (j = 1; j <= mbuf->nb_segs; j++) {
533 iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
535 rte_pktmbuf_mtod(seg, void *);
539 ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
540 (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
541 (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
542 /* Support only packets with all data in the same seg */
543 if (mbuf->nb_segs > 1)
545 /* To change checksums, work on a copy of data. */
546 rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
547 rte_pktmbuf_data_len(mbuf));
548 tap_tx_offload(m_copy, mbuf->ol_flags,
549 mbuf->l2_len, mbuf->l3_len);
550 iovecs[1].iov_base = m_copy;
552 /* copy the tx frame data */
553 n = writev(txq->fd, iovecs, mbuf->nb_segs + 1);
558 num_tx_bytes += mbuf->pkt_len;
559 rte_pktmbuf_free(mbuf);
562 txq->stats.opackets += num_tx;
563 txq->stats.errs += nb_pkts - num_tx;
564 txq->stats.obytes += num_tx_bytes;
570 tap_ioctl_req2str(unsigned long request)
574 return "SIOCSIFFLAGS";
576 return "SIOCGIFFLAGS";
578 return "SIOCGIFHWADDR";
580 return "SIOCSIFHWADDR";
588 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
589 struct ifreq *ifr, int set, enum ioctl_mode mode)
591 short req_flags = ifr->ifr_flags;
592 int remote = pmd->remote_if_index &&
593 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
595 if (!pmd->remote_if_index && mode == REMOTE_ONLY)
598 * If there is a remote netdevice, apply ioctl on it, then apply it on
603 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
604 else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
605 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
608 /* fetch current flags to leave other flags untouched */
609 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
612 ifr->ifr_flags |= req_flags;
614 ifr->ifr_flags &= ~req_flags;
622 RTE_ASSERT(!"unsupported request type: must not happen");
624 if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
626 if (remote-- && mode == LOCAL_AND_REMOTE)
631 RTE_LOG(DEBUG, PMD, "%s: %s(%s) failed: %s(%d)\n", ifr->ifr_name,
632 __func__, tap_ioctl_req2str(request), strerror(errno), errno);
637 tap_link_set_down(struct rte_eth_dev *dev)
639 struct pmd_internals *pmd = dev->data->dev_private;
640 struct ifreq ifr = { .ifr_flags = IFF_UP };
642 dev->data->dev_link.link_status = ETH_LINK_DOWN;
643 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
647 tap_link_set_up(struct rte_eth_dev *dev)
649 struct pmd_internals *pmd = dev->data->dev_private;
650 struct ifreq ifr = { .ifr_flags = IFF_UP };
652 dev->data->dev_link.link_status = ETH_LINK_UP;
653 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
657 tap_dev_start(struct rte_eth_dev *dev)
661 err = tap_intr_handle_set(dev, 1);
664 return tap_link_set_up(dev);
667 /* This function gets called when the current port gets stopped.
670 tap_dev_stop(struct rte_eth_dev *dev)
672 tap_intr_handle_set(dev, 0);
673 tap_link_set_down(dev);
677 tap_dev_configure(struct rte_eth_dev *dev)
679 uint64_t supp_tx_offloads = tap_tx_offload_get_port_capa();
680 uint64_t tx_offloads = dev->data->dev_conf.txmode.offloads;
682 if ((tx_offloads & supp_tx_offloads) != tx_offloads) {
685 "Some Tx offloads are not supported "
686 "requested 0x%" PRIx64 " supported 0x%" PRIx64 "\n",
687 tx_offloads, supp_tx_offloads);
690 if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
692 "%s: number of rx queues %d exceeds max num of queues %d\n",
694 dev->data->nb_rx_queues,
695 RTE_PMD_TAP_MAX_QUEUES);
698 if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
700 "%s: number of tx queues %d exceeds max num of queues %d\n",
702 dev->data->nb_tx_queues,
703 RTE_PMD_TAP_MAX_QUEUES);
707 RTE_LOG(INFO, PMD, "%s: %p: TX configured queues number: %u\n",
708 dev->device->name, (void *)dev, dev->data->nb_tx_queues);
710 RTE_LOG(INFO, PMD, "%s: %p: RX configured queues number: %u\n",
711 dev->device->name, (void *)dev, dev->data->nb_rx_queues);
717 tap_dev_speed_capa(void)
719 uint32_t speed = pmd_link.link_speed;
722 if (speed >= ETH_SPEED_NUM_10M)
723 capa |= ETH_LINK_SPEED_10M;
724 if (speed >= ETH_SPEED_NUM_100M)
725 capa |= ETH_LINK_SPEED_100M;
726 if (speed >= ETH_SPEED_NUM_1G)
727 capa |= ETH_LINK_SPEED_1G;
728 if (speed >= ETH_SPEED_NUM_5G)
729 capa |= ETH_LINK_SPEED_2_5G;
730 if (speed >= ETH_SPEED_NUM_5G)
731 capa |= ETH_LINK_SPEED_5G;
732 if (speed >= ETH_SPEED_NUM_10G)
733 capa |= ETH_LINK_SPEED_10G;
734 if (speed >= ETH_SPEED_NUM_20G)
735 capa |= ETH_LINK_SPEED_20G;
736 if (speed >= ETH_SPEED_NUM_25G)
737 capa |= ETH_LINK_SPEED_25G;
738 if (speed >= ETH_SPEED_NUM_40G)
739 capa |= ETH_LINK_SPEED_40G;
740 if (speed >= ETH_SPEED_NUM_50G)
741 capa |= ETH_LINK_SPEED_50G;
742 if (speed >= ETH_SPEED_NUM_56G)
743 capa |= ETH_LINK_SPEED_56G;
744 if (speed >= ETH_SPEED_NUM_100G)
745 capa |= ETH_LINK_SPEED_100G;
751 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
753 struct pmd_internals *internals = dev->data->dev_private;
755 dev_info->if_index = internals->if_index;
756 dev_info->max_mac_addrs = 1;
757 dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
758 dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
759 dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
760 dev_info->min_rx_bufsize = 0;
761 dev_info->pci_dev = NULL;
762 dev_info->speed_capa = tap_dev_speed_capa();
763 dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
764 dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
765 dev_info->rx_queue_offload_capa;
766 dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
767 dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
768 dev_info->tx_queue_offload_capa;
772 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
774 unsigned int i, imax;
775 unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
776 unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
777 unsigned long rx_nombuf = 0, ierrors = 0;
778 const struct pmd_internals *pmd = dev->data->dev_private;
780 /* rx queue statistics */
781 imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
782 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
783 for (i = 0; i < imax; i++) {
784 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
785 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
786 rx_total += tap_stats->q_ipackets[i];
787 rx_bytes_total += tap_stats->q_ibytes[i];
788 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
789 ierrors += pmd->rxq[i].stats.ierrors;
792 /* tx queue statistics */
793 imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
794 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
796 for (i = 0; i < imax; i++) {
797 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
798 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
799 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
800 tx_total += tap_stats->q_opackets[i];
801 tx_err_total += tap_stats->q_errors[i];
802 tx_bytes_total += tap_stats->q_obytes[i];
805 tap_stats->ipackets = rx_total;
806 tap_stats->ibytes = rx_bytes_total;
807 tap_stats->ierrors = ierrors;
808 tap_stats->rx_nombuf = rx_nombuf;
809 tap_stats->opackets = tx_total;
810 tap_stats->oerrors = tx_err_total;
811 tap_stats->obytes = tx_bytes_total;
816 tap_stats_reset(struct rte_eth_dev *dev)
819 struct pmd_internals *pmd = dev->data->dev_private;
821 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
822 pmd->rxq[i].stats.ipackets = 0;
823 pmd->rxq[i].stats.ibytes = 0;
824 pmd->rxq[i].stats.ierrors = 0;
825 pmd->rxq[i].stats.rx_nombuf = 0;
827 pmd->txq[i].stats.opackets = 0;
828 pmd->txq[i].stats.errs = 0;
829 pmd->txq[i].stats.obytes = 0;
834 tap_dev_close(struct rte_eth_dev *dev)
837 struct pmd_internals *internals = dev->data->dev_private;
839 tap_link_set_down(dev);
840 tap_flow_flush(dev, NULL);
841 tap_flow_implicit_flush(internals, NULL);
843 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
844 if (internals->rxq[i].fd != -1) {
845 close(internals->rxq[i].fd);
846 internals->rxq[i].fd = -1;
848 if (internals->txq[i].fd != -1) {
849 close(internals->txq[i].fd);
850 internals->txq[i].fd = -1;
854 if (internals->remote_if_index) {
855 /* Restore initial remote state */
856 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
857 &internals->remote_initial_flags);
862 tap_rx_queue_release(void *queue)
864 struct rx_queue *rxq = queue;
866 if (rxq && (rxq->fd > 0)) {
869 rte_pktmbuf_free(rxq->pool);
870 rte_free(rxq->iovecs);
877 tap_tx_queue_release(void *queue)
879 struct tx_queue *txq = queue;
881 if (txq && (txq->fd > 0)) {
888 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
890 struct rte_eth_link *dev_link = &dev->data->dev_link;
891 struct pmd_internals *pmd = dev->data->dev_private;
892 struct ifreq ifr = { .ifr_flags = 0 };
894 if (pmd->remote_if_index) {
895 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
896 if (!(ifr.ifr_flags & IFF_UP) ||
897 !(ifr.ifr_flags & IFF_RUNNING)) {
898 dev_link->link_status = ETH_LINK_DOWN;
902 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
903 dev_link->link_status =
904 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
911 tap_promisc_enable(struct rte_eth_dev *dev)
913 struct pmd_internals *pmd = dev->data->dev_private;
914 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
916 dev->data->promiscuous = 1;
917 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
918 if (pmd->remote_if_index && !pmd->flow_isolate)
919 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
923 tap_promisc_disable(struct rte_eth_dev *dev)
925 struct pmd_internals *pmd = dev->data->dev_private;
926 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
928 dev->data->promiscuous = 0;
929 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
930 if (pmd->remote_if_index && !pmd->flow_isolate)
931 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
935 tap_allmulti_enable(struct rte_eth_dev *dev)
937 struct pmd_internals *pmd = dev->data->dev_private;
938 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
940 dev->data->all_multicast = 1;
941 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
942 if (pmd->remote_if_index && !pmd->flow_isolate)
943 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
947 tap_allmulti_disable(struct rte_eth_dev *dev)
949 struct pmd_internals *pmd = dev->data->dev_private;
950 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
952 dev->data->all_multicast = 0;
953 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
954 if (pmd->remote_if_index && !pmd->flow_isolate)
955 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
959 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
961 struct pmd_internals *pmd = dev->data->dev_private;
962 enum ioctl_mode mode = LOCAL_ONLY;
965 if (is_zero_ether_addr(mac_addr)) {
966 RTE_LOG(ERR, PMD, "%s: can't set an empty MAC address\n",
970 /* Check the actual current MAC address on the tap netdevice */
971 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
973 if (is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
976 /* Check the current MAC address on the remote */
977 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0)
979 if (!is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
981 mode = LOCAL_AND_REMOTE;
982 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
983 rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
984 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode) < 0)
986 rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
987 if (pmd->remote_if_index && !pmd->flow_isolate) {
988 /* Replace MAC redirection rule after a MAC change */
989 if (tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC) < 0) {
991 "%s: Couldn't delete MAC redirection rule\n",
995 if (tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0)
997 "%s: Couldn't add MAC redirection rule\n",
1003 tap_setup_queue(struct rte_eth_dev *dev,
1004 struct pmd_internals *internals,
1011 struct pmd_internals *pmd = dev->data->dev_private;
1012 struct rx_queue *rx = &internals->rxq[qid];
1013 struct tx_queue *tx = &internals->txq[qid];
1025 /* fd for this queue already exists */
1026 RTE_LOG(DEBUG, PMD, "%s: fd %d for %s queue qid %d exists\n",
1027 pmd->name, *fd, dir, qid);
1028 } else if (*other_fd != -1) {
1029 /* Only other_fd exists. dup it */
1030 *fd = dup(*other_fd);
1033 RTE_LOG(ERR, PMD, "%s: dup() failed.\n",
1037 RTE_LOG(DEBUG, PMD, "%s: dup fd %d for %s queue qid %d (%d)\n",
1038 pmd->name, *other_fd, dir, qid, *fd);
1040 /* Both RX and TX fds do not exist (equal -1). Create fd */
1041 *fd = tun_alloc(pmd);
1043 *fd = -1; /* restore original value */
1044 RTE_LOG(ERR, PMD, "%s: tun_alloc() failed.\n",
1048 RTE_LOG(DEBUG, PMD, "%s: add %s queue for qid %d fd %d\n",
1049 pmd->name, dir, qid, *fd);
1052 tx->mtu = &dev->data->mtu;
1053 rx->rxmode = &dev->data->dev_conf.rxmode;
1059 tap_rx_queue_setup(struct rte_eth_dev *dev,
1060 uint16_t rx_queue_id,
1061 uint16_t nb_rx_desc,
1062 unsigned int socket_id,
1063 const struct rte_eth_rxconf *rx_conf __rte_unused,
1064 struct rte_mempool *mp)
1066 struct pmd_internals *internals = dev->data->dev_private;
1067 struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1068 struct rte_mbuf **tmp = &rxq->pool;
1069 long iov_max = sysconf(_SC_IOV_MAX);
1070 uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1071 struct iovec (*iovecs)[nb_desc + 1];
1072 int data_off = RTE_PKTMBUF_HEADROOM;
1077 if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1078 RTE_LOG(WARNING, PMD,
1079 "nb_rx_queues %d too small or mempool NULL\n",
1080 dev->data->nb_rx_queues);
1084 /* Verify application offloads are valid for our port and queue. */
1085 if (!tap_rxq_are_offloads_valid(dev, rx_conf->offloads)) {
1086 rte_errno = ENOTSUP;
1088 "%p: Rx queue offloads 0x%" PRIx64
1089 " don't match port offloads 0x%" PRIx64
1090 " or supported offloads 0x%" PRIx64 "\n",
1091 (void *)dev, rx_conf->offloads,
1092 dev->data->dev_conf.rxmode.offloads,
1093 (tap_rx_offload_get_port_capa() |
1094 tap_rx_offload_get_queue_capa()));
1098 rxq->trigger_seen = 1; /* force initial burst */
1099 rxq->in_port = dev->data->port_id;
1100 rxq->nb_rx_desc = nb_desc;
1101 iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1104 RTE_LOG(WARNING, PMD,
1105 "%s: Couldn't allocate %d RX descriptors\n",
1106 dev->device->name, nb_desc);
1109 rxq->iovecs = iovecs;
1111 dev->data->rx_queues[rx_queue_id] = rxq;
1112 fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1118 (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1119 (*rxq->iovecs)[0].iov_base = &rxq->pi;
1121 for (i = 1; i <= nb_desc; i++) {
1122 *tmp = rte_pktmbuf_alloc(rxq->mp);
1124 RTE_LOG(WARNING, PMD,
1125 "%s: couldn't allocate memory for queue %d\n",
1126 dev->device->name, rx_queue_id);
1130 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1131 (*rxq->iovecs)[i].iov_base =
1132 (char *)(*tmp)->buf_addr + data_off;
1134 tmp = &(*tmp)->next;
1137 RTE_LOG(DEBUG, PMD, " RX TAP device name %s, qid %d on fd %d\n",
1138 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
1143 rte_pktmbuf_free(rxq->pool);
1145 rte_free(rxq->iovecs);
1151 tap_tx_queue_setup(struct rte_eth_dev *dev,
1152 uint16_t tx_queue_id,
1153 uint16_t nb_tx_desc __rte_unused,
1154 unsigned int socket_id __rte_unused,
1155 const struct rte_eth_txconf *tx_conf)
1157 struct pmd_internals *internals = dev->data->dev_private;
1158 struct tx_queue *txq;
1161 if (tx_queue_id >= dev->data->nb_tx_queues)
1163 dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1164 txq = dev->data->tx_queues[tx_queue_id];
1166 * Don't verify port offloads for application which
1169 if (tx_conf != NULL &&
1170 !!(tx_conf->txq_flags & ETH_TXQ_FLAGS_IGNORE)) {
1171 if (tap_txq_are_offloads_valid(dev, tx_conf->offloads)) {
1172 txq->csum = !!(tx_conf->offloads &
1173 (DEV_TX_OFFLOAD_IPV4_CKSUM |
1174 DEV_TX_OFFLOAD_UDP_CKSUM |
1175 DEV_TX_OFFLOAD_TCP_CKSUM));
1177 rte_errno = ENOTSUP;
1179 "%p: Tx queue offloads 0x%" PRIx64
1180 " don't match port offloads 0x%" PRIx64
1181 " or supported offloads 0x%" PRIx64,
1182 (void *)dev, tx_conf->offloads,
1183 dev->data->dev_conf.txmode.offloads,
1184 tap_tx_offload_get_port_capa());
1188 ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1192 " TX TAP device name %s, qid %d on fd %d csum %s\n",
1193 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd,
1194 txq->csum ? "on" : "off");
1200 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1202 struct pmd_internals *pmd = dev->data->dev_private;
1203 struct ifreq ifr = { .ifr_mtu = mtu };
1206 err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1208 dev->data->mtu = mtu;
1214 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1215 struct ether_addr *mc_addr_set __rte_unused,
1216 uint32_t nb_mc_addr __rte_unused)
1219 * Nothing to do actually: the tap has no filtering whatsoever, every
1220 * packet is received.
1226 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1228 struct rte_eth_dev *dev = arg;
1229 struct pmd_internals *pmd = dev->data->dev_private;
1230 struct ifinfomsg *info = NLMSG_DATA(nh);
1232 if (nh->nlmsg_type != RTM_NEWLINK ||
1233 (info->ifi_index != pmd->if_index &&
1234 info->ifi_index != pmd->remote_if_index))
1236 return tap_link_update(dev, 0);
1240 tap_dev_intr_handler(void *cb_arg)
1242 struct rte_eth_dev *dev = cb_arg;
1243 struct pmd_internals *pmd = dev->data->dev_private;
1245 tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1249 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1251 struct pmd_internals *pmd = dev->data->dev_private;
1253 /* In any case, disable interrupt if the conf is no longer there. */
1254 if (!dev->data->dev_conf.intr_conf.lsc) {
1255 if (pmd->intr_handle.fd != -1) {
1256 tap_nl_final(pmd->intr_handle.fd);
1257 rte_intr_callback_unregister(&pmd->intr_handle,
1258 tap_dev_intr_handler, dev);
1263 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1264 if (unlikely(pmd->intr_handle.fd == -1))
1266 return rte_intr_callback_register(
1267 &pmd->intr_handle, tap_dev_intr_handler, dev);
1269 tap_nl_final(pmd->intr_handle.fd);
1270 return rte_intr_callback_unregister(&pmd->intr_handle,
1271 tap_dev_intr_handler, dev);
1275 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1279 err = tap_lsc_intr_handle_set(dev, set);
1282 err = tap_rx_intr_vec_set(dev, set);
1284 tap_lsc_intr_handle_set(dev, 0);
1288 static const uint32_t*
1289 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1291 static const uint32_t ptypes[] = {
1292 RTE_PTYPE_INNER_L2_ETHER,
1293 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1294 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1295 RTE_PTYPE_INNER_L3_IPV4,
1296 RTE_PTYPE_INNER_L3_IPV4_EXT,
1297 RTE_PTYPE_INNER_L3_IPV6,
1298 RTE_PTYPE_INNER_L3_IPV6_EXT,
1299 RTE_PTYPE_INNER_L4_FRAG,
1300 RTE_PTYPE_INNER_L4_UDP,
1301 RTE_PTYPE_INNER_L4_TCP,
1302 RTE_PTYPE_INNER_L4_SCTP,
1304 RTE_PTYPE_L2_ETHER_VLAN,
1305 RTE_PTYPE_L2_ETHER_QINQ,
1307 RTE_PTYPE_L3_IPV4_EXT,
1308 RTE_PTYPE_L3_IPV6_EXT,
1320 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1321 struct rte_eth_fc_conf *fc_conf)
1323 fc_conf->mode = RTE_FC_NONE;
1328 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1329 struct rte_eth_fc_conf *fc_conf)
1331 if (fc_conf->mode != RTE_FC_NONE)
1336 static const struct eth_dev_ops ops = {
1337 .dev_start = tap_dev_start,
1338 .dev_stop = tap_dev_stop,
1339 .dev_close = tap_dev_close,
1340 .dev_configure = tap_dev_configure,
1341 .dev_infos_get = tap_dev_info,
1342 .rx_queue_setup = tap_rx_queue_setup,
1343 .tx_queue_setup = tap_tx_queue_setup,
1344 .rx_queue_release = tap_rx_queue_release,
1345 .tx_queue_release = tap_tx_queue_release,
1346 .flow_ctrl_get = tap_flow_ctrl_get,
1347 .flow_ctrl_set = tap_flow_ctrl_set,
1348 .link_update = tap_link_update,
1349 .dev_set_link_up = tap_link_set_up,
1350 .dev_set_link_down = tap_link_set_down,
1351 .promiscuous_enable = tap_promisc_enable,
1352 .promiscuous_disable = tap_promisc_disable,
1353 .allmulticast_enable = tap_allmulti_enable,
1354 .allmulticast_disable = tap_allmulti_disable,
1355 .mac_addr_set = tap_mac_set,
1356 .mtu_set = tap_mtu_set,
1357 .set_mc_addr_list = tap_set_mc_addr_list,
1358 .stats_get = tap_stats_get,
1359 .stats_reset = tap_stats_reset,
1360 .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1361 .filter_ctrl = tap_dev_filter_ctrl,
1365 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
1366 char *remote_iface, struct ether_addr *mac_addr)
1368 int numa_node = rte_socket_id();
1369 struct rte_eth_dev *dev;
1370 struct pmd_internals *pmd;
1371 struct rte_eth_dev_data *data;
1375 RTE_LOG(DEBUG, PMD, " TAP device on numa %u\n", rte_socket_id());
1377 data = rte_zmalloc_socket(tap_name, sizeof(*data), 0, numa_node);
1379 RTE_LOG(ERR, PMD, "TAP Failed to allocate data\n");
1380 goto error_exit_nodev;
1383 dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1385 RTE_LOG(ERR, PMD, "TAP Unable to allocate device struct\n");
1386 goto error_exit_nodev;
1389 pmd = dev->data->dev_private;
1391 snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1393 pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1394 if (pmd->ioctl_sock == -1) {
1396 "TAP Unable to get a socket for management: %s\n",
1401 /* Setup some default values */
1402 rte_memcpy(data, dev->data, sizeof(*data));
1403 data->dev_private = pmd;
1404 data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1405 data->numa_node = numa_node;
1407 data->dev_link = pmd_link;
1408 data->mac_addrs = &pmd->eth_addr;
1409 /* Set the number of RX and TX queues */
1410 data->nb_rx_queues = 0;
1411 data->nb_tx_queues = 0;
1414 dev->dev_ops = &ops;
1415 dev->rx_pkt_burst = pmd_rx_burst;
1416 dev->tx_pkt_burst = pmd_tx_burst;
1418 pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1419 pmd->intr_handle.fd = -1;
1420 dev->intr_handle = &pmd->intr_handle;
1422 /* Presetup the fds to -1 as being not valid */
1423 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1424 pmd->rxq[i].fd = -1;
1425 pmd->txq[i].fd = -1;
1429 if (is_zero_ether_addr(mac_addr))
1430 eth_random_addr((uint8_t *)&pmd->eth_addr);
1432 rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1435 /* Immediately create the netdevice (this will create the 1st queue). */
1437 if (tap_setup_queue(dev, pmd, 0, 1) == -1)
1440 if (tap_setup_queue(dev, pmd, 0, 0) == -1)
1443 ifr.ifr_mtu = dev->data->mtu;
1444 if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1448 memset(&ifr, 0, sizeof(struct ifreq));
1449 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1450 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
1452 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1457 * Set up everything related to rte_flow:
1459 * - tap / remote if_index
1460 * - mandatory QDISCs
1461 * - rte_flow actual/implicit lists
1464 pmd->nlsk_fd = tap_nl_init(0);
1465 if (pmd->nlsk_fd == -1) {
1466 RTE_LOG(WARNING, PMD, "%s: failed to create netlink socket.\n",
1468 goto disable_rte_flow;
1470 pmd->if_index = if_nametoindex(pmd->name);
1471 if (!pmd->if_index) {
1472 RTE_LOG(ERR, PMD, "%s: failed to get if_index.\n", pmd->name);
1473 goto disable_rte_flow;
1475 if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1476 RTE_LOG(ERR, PMD, "%s: failed to create multiq qdisc.\n",
1478 goto disable_rte_flow;
1480 if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1481 RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1483 goto disable_rte_flow;
1485 LIST_INIT(&pmd->flows);
1487 if (strlen(remote_iface)) {
1488 pmd->remote_if_index = if_nametoindex(remote_iface);
1489 if (!pmd->remote_if_index) {
1490 RTE_LOG(ERR, PMD, "%s: failed to get %s if_index.\n",
1491 pmd->name, remote_iface);
1494 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1495 "%s", remote_iface);
1497 /* Save state of remote device */
1498 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1500 /* Replicate remote MAC address */
1501 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1502 RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1503 pmd->name, pmd->remote_iface);
1506 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1508 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1509 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1510 RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1511 pmd->name, remote_iface);
1516 * Flush usually returns negative value because it tries to
1517 * delete every QDISC (and on a running device, one QDISC at
1518 * least is needed). Ignore negative return value.
1520 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1521 if (qdisc_create_ingress(pmd->nlsk_fd,
1522 pmd->remote_if_index) < 0) {
1523 RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1527 LIST_INIT(&pmd->implicit_flows);
1528 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1529 tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1530 tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1531 tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1533 "%s: failed to create implicit rules.\n",
1542 RTE_LOG(ERR, PMD, " Disabling rte flow support: %s(%d)\n",
1543 strerror(errno), errno);
1544 if (strlen(remote_iface)) {
1545 RTE_LOG(ERR, PMD, "Remote feature requires flow support.\n");
1551 RTE_LOG(ERR, PMD, " Can't set up remote feature: %s(%d)\n",
1552 strerror(errno), errno);
1553 tap_flow_implicit_flush(pmd, NULL);
1556 if (pmd->ioctl_sock > 0)
1557 close(pmd->ioctl_sock);
1558 rte_eth_dev_release_port(dev);
1561 RTE_LOG(ERR, PMD, "TAP Unable to initialize %s\n",
1562 rte_vdev_device_name(vdev));
1569 set_interface_name(const char *key __rte_unused,
1573 char *name = (char *)extra_args;
1576 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN - 1);
1578 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1579 DEFAULT_TAP_NAME, (tap_unit - 1));
1585 set_remote_iface(const char *key __rte_unused,
1589 char *name = (char *)extra_args;
1592 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
1597 static int parse_user_mac(struct ether_addr *user_mac,
1600 unsigned int index = 0;
1601 char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
1603 if (user_mac == NULL || value == NULL)
1606 strlcpy(mac_temp, value, sizeof(mac_temp));
1607 mac_byte = strtok(mac_temp, ":");
1609 while ((mac_byte != NULL) &&
1610 (strlen(mac_byte) <= 2) &&
1611 (strlen(mac_byte) == strspn(mac_byte,
1612 ETH_TAP_CMP_MAC_FMT))) {
1613 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
1614 mac_byte = strtok(NULL, ":");
1621 set_mac_type(const char *key __rte_unused,
1625 struct ether_addr *user_mac = extra_args;
1630 if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
1631 static int iface_idx;
1633 /* fixed mac = 00:64:74:61:70:<iface_idx> */
1634 memcpy((char *)user_mac->addr_bytes, "\0dtap", ETHER_ADDR_LEN);
1635 user_mac->addr_bytes[ETHER_ADDR_LEN - 1] = iface_idx++ + '0';
1639 if (parse_user_mac(user_mac, value) != 6)
1642 RTE_LOG(DEBUG, PMD, "TAP user MAC param (%s)\n", value);
1646 RTE_LOG(ERR, PMD, "TAP user MAC (%s) is not in format (%s|%s)\n",
1647 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
1652 * Open a TUN interface device. TUN PMD
1653 * 1) sets tap_type as false
1654 * 2) intakes iface as argument.
1655 * 3) as interface is virtual set speed to 10G
1658 rte_pmd_tun_probe(struct rte_vdev_device *dev)
1660 const char *name, *params;
1662 struct rte_kvargs *kvlist = NULL;
1663 char tun_name[RTE_ETH_NAME_MAX_LEN];
1664 char remote_iface[RTE_ETH_NAME_MAX_LEN];
1667 strcpy(tuntap_name, "TUN");
1669 name = rte_vdev_device_name(dev);
1670 params = rte_vdev_device_args(dev);
1671 memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1673 if (params && (params[0] != '\0')) {
1674 RTE_LOG(DEBUG, PMD, "parameters (%s)\n", params);
1676 kvlist = rte_kvargs_parse(params, valid_arguments);
1678 if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1679 ret = rte_kvargs_process(kvlist,
1681 &set_interface_name,
1689 pmd_link.link_speed = ETH_SPEED_NUM_10G;
1691 RTE_LOG(NOTICE, PMD, "Initializing pmd_tun for %s as %s\n",
1694 ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0);
1698 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
1700 tun_unit--; /* Restore the unit number */
1702 rte_kvargs_free(kvlist);
1707 /* Open a TAP interface device.
1710 rte_pmd_tap_probe(struct rte_vdev_device *dev)
1712 const char *name, *params;
1714 struct rte_kvargs *kvlist = NULL;
1716 char tap_name[RTE_ETH_NAME_MAX_LEN];
1717 char remote_iface[RTE_ETH_NAME_MAX_LEN];
1718 struct ether_addr user_mac = { .addr_bytes = {0} };
1721 strcpy(tuntap_name, "TAP");
1723 name = rte_vdev_device_name(dev);
1724 params = rte_vdev_device_args(dev);
1726 speed = ETH_SPEED_NUM_10G;
1727 snprintf(tap_name, sizeof(tap_name), "%s%d",
1728 DEFAULT_TAP_NAME, tap_unit++);
1729 memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1731 if (params && (params[0] != '\0')) {
1732 RTE_LOG(DEBUG, PMD, "parameters (%s)\n", params);
1734 kvlist = rte_kvargs_parse(params, valid_arguments);
1736 if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1737 ret = rte_kvargs_process(kvlist,
1739 &set_interface_name,
1745 if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
1746 ret = rte_kvargs_process(kvlist,
1754 if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
1755 ret = rte_kvargs_process(kvlist,
1764 pmd_link.link_speed = speed;
1766 RTE_LOG(NOTICE, PMD, "Initializing pmd_tap for %s as %s\n",
1769 ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac);
1773 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
1775 tap_unit--; /* Restore the unit number */
1777 rte_kvargs_free(kvlist);
1782 /* detach a TUNTAP device.
1785 rte_pmd_tap_remove(struct rte_vdev_device *dev)
1787 struct rte_eth_dev *eth_dev = NULL;
1788 struct pmd_internals *internals;
1791 RTE_LOG(DEBUG, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
1794 /* find the ethdev entry */
1795 eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1799 internals = eth_dev->data->dev_private;
1800 if (internals->nlsk_fd) {
1801 tap_flow_flush(eth_dev, NULL);
1802 tap_flow_implicit_flush(internals, NULL);
1803 tap_nl_final(internals->nlsk_fd);
1805 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1806 if (internals->rxq[i].fd != -1) {
1807 close(internals->rxq[i].fd);
1808 internals->rxq[i].fd = -1;
1810 if (internals->txq[i].fd != -1) {
1811 close(internals->txq[i].fd);
1812 internals->txq[i].fd = -1;
1816 close(internals->ioctl_sock);
1817 rte_free(eth_dev->data->dev_private);
1818 rte_free(eth_dev->data);
1820 rte_eth_dev_release_port(eth_dev);
1825 static struct rte_vdev_driver pmd_tun_drv = {
1826 .probe = rte_pmd_tun_probe,
1827 .remove = rte_pmd_tap_remove,
1830 static struct rte_vdev_driver pmd_tap_drv = {
1831 .probe = rte_pmd_tap_probe,
1832 .remove = rte_pmd_tap_remove,
1834 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1835 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
1836 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1837 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
1838 ETH_TAP_IFACE_ARG "=<string> ");
1839 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
1840 ETH_TAP_IFACE_ARG "=<string> "
1841 ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
1842 ETH_TAP_REMOTE_ARG "=<string>");