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>
19 #include <rte_ethdev.h>
20 #include <rte_errno.h>
22 #include <sys/types.h>
24 #include <sys/socket.h>
25 #include <sys/ioctl.h>
26 #include <sys/utsname.h>
34 #include <arpa/inet.h>
36 #include <linux/if_tun.h>
37 #include <linux/if_ether.h>
42 #include <rte_eth_tap.h>
44 #include <tap_netlink.h>
45 #include <tap_tcmsgs.h>
47 /* Linux based path to the TUN device */
48 #define TUN_TAP_DEV_PATH "/dev/net/tun"
49 #define DEFAULT_TAP_NAME "dtap"
50 #define DEFAULT_TUN_NAME "dtun"
52 #define ETH_TAP_IFACE_ARG "iface"
53 #define ETH_TAP_REMOTE_ARG "remote"
54 #define ETH_TAP_MAC_ARG "mac"
55 #define ETH_TAP_MAC_FIXED "fixed"
57 #define ETH_TAP_USR_MAC_FMT "xx:xx:xx:xx:xx:xx"
58 #define ETH_TAP_CMP_MAC_FMT "0123456789ABCDEFabcdef"
59 #define ETH_TAP_MAC_ARG_FMT ETH_TAP_MAC_FIXED "|" ETH_TAP_USR_MAC_FMT
61 #define TAP_GSO_MBUFS_PER_CORE 128
62 #define TAP_GSO_MBUF_SEG_SIZE 128
63 #define TAP_GSO_MBUF_CACHE_SIZE 4
64 #define TAP_GSO_MBUFS_NUM \
65 (TAP_GSO_MBUFS_PER_CORE * TAP_GSO_MBUF_CACHE_SIZE)
67 /* IPC key for queue fds sync */
68 #define TAP_MP_KEY "tap_mp_sync_queues"
70 #define TAP_IOV_DEFAULT_MAX 1024
72 static int tap_devices_count;
74 static const char *valid_arguments[] = {
81 static volatile uint32_t tap_trigger; /* Rx trigger */
83 static struct rte_eth_link pmd_link = {
84 .link_speed = ETH_SPEED_NUM_10G,
85 .link_duplex = ETH_LINK_FULL_DUPLEX,
86 .link_status = ETH_LINK_DOWN,
87 .link_autoneg = ETH_LINK_FIXED,
91 tap_trigger_cb(int sig __rte_unused)
93 /* Valid trigger values are nonzero */
94 tap_trigger = (tap_trigger + 1) | 0x80000000;
97 /* Specifies on what netdevices the ioctl should be applied */
104 /* Message header to synchronize queues via IPC */
106 char port_name[RTE_DEV_NAME_MAX_LEN];
110 * The file descriptors are in the dedicated part
111 * of the Unix message to be translated by the kernel.
115 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
118 * Tun/Tap allocation routine
121 * Pointer to private structure.
123 * @param[in] is_keepalive
127 * -1 on failure, fd on success
130 tun_alloc(struct pmd_internals *pmd, int is_keepalive)
133 #ifdef IFF_MULTI_QUEUE
134 unsigned int features;
138 memset(&ifr, 0, sizeof(struct ifreq));
141 * Do not set IFF_NO_PI as packet information header will be needed
142 * to check if a received packet has been truncated.
144 ifr.ifr_flags = (pmd->type == ETH_TUNTAP_TYPE_TAP) ?
145 IFF_TAP : IFF_TUN | IFF_POINTOPOINT;
146 strlcpy(ifr.ifr_name, pmd->name, IFNAMSIZ);
148 fd = open(TUN_TAP_DEV_PATH, O_RDWR);
150 TAP_LOG(ERR, "Unable to open %s interface", TUN_TAP_DEV_PATH);
154 #ifdef IFF_MULTI_QUEUE
155 /* Grab the TUN features to verify we can work multi-queue */
156 if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
157 TAP_LOG(ERR, "unable to get TUN/TAP features");
160 TAP_LOG(DEBUG, "%s Features %08x", TUN_TAP_DEV_PATH, features);
162 if (features & IFF_MULTI_QUEUE) {
163 TAP_LOG(DEBUG, " Multi-queue support for %d queues",
164 RTE_PMD_TAP_MAX_QUEUES);
165 ifr.ifr_flags |= IFF_MULTI_QUEUE;
169 ifr.ifr_flags |= IFF_ONE_QUEUE;
170 TAP_LOG(DEBUG, " Single queue only support");
173 /* Set the TUN/TAP configuration and set the name if needed */
174 if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
175 TAP_LOG(WARNING, "Unable to set TUNSETIFF for %s: %s",
176 ifr.ifr_name, strerror(errno));
181 * Name passed to kernel might be wildcard like dtun%d
182 * and need to find the resulting device.
184 TAP_LOG(DEBUG, "Device name is '%s'", ifr.ifr_name);
185 strlcpy(pmd->name, ifr.ifr_name, RTE_ETH_NAME_MAX_LEN);
189 * Detach the TUN/TAP keep-alive queue
190 * to avoid traffic through it
192 ifr.ifr_flags = IFF_DETACH_QUEUE;
193 if (ioctl(fd, TUNSETQUEUE, (void *)&ifr) < 0) {
195 "Unable to detach keep-alive queue for %s: %s",
196 ifr.ifr_name, strerror(errno));
201 /* Always set the file descriptor to non-blocking */
202 if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
204 "Unable to set %s to nonblocking: %s",
205 ifr.ifr_name, strerror(errno));
209 /* Set up trigger to optimize empty Rx bursts */
213 int flags = fcntl(fd, F_GETFL);
215 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
217 if (sa.sa_handler != tap_trigger_cb) {
219 * Make sure SIGIO is not already taken. This is done
220 * as late as possible to leave the application a
221 * chance to set up its own signal handler first.
223 if (sa.sa_handler != SIG_IGN &&
224 sa.sa_handler != SIG_DFL) {
228 sa = (struct sigaction){
229 .sa_flags = SA_RESTART,
230 .sa_handler = tap_trigger_cb,
232 if (sigaction(SIGIO, &sa, NULL) == -1)
235 /* Enable SIGIO on file descriptor */
236 fcntl(fd, F_SETFL, flags | O_ASYNC);
237 fcntl(fd, F_SETOWN, getpid());
241 /* Disable trigger globally in case of error */
243 TAP_LOG(WARNING, "Rx trigger disabled: %s",
256 tap_verify_csum(struct rte_mbuf *mbuf)
258 uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
259 uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
260 uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
261 unsigned int l2_len = sizeof(struct rte_ether_hdr);
267 if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
269 else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
271 /* Don't verify checksum for packets with discontinuous L2 header */
272 if (unlikely(l2_len + sizeof(struct rte_ipv4_hdr) >
273 rte_pktmbuf_data_len(mbuf)))
275 l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
276 if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
277 struct rte_ipv4_hdr *iph = l3_hdr;
279 /* ihl contains the number of 4-byte words in the header */
280 l3_len = 4 * (iph->version_ihl & 0xf);
281 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
283 /* check that the total length reported by header is not
284 * greater than the total received size
286 if (l2_len + rte_be_to_cpu_16(iph->total_length) >
287 rte_pktmbuf_data_len(mbuf))
290 cksum = ~rte_raw_cksum(iph, l3_len);
291 mbuf->ol_flags |= cksum ?
292 PKT_RX_IP_CKSUM_BAD :
293 PKT_RX_IP_CKSUM_GOOD;
294 } else if (l3 == RTE_PTYPE_L3_IPV6) {
295 struct rte_ipv6_hdr *iph = l3_hdr;
297 l3_len = sizeof(struct rte_ipv6_hdr);
298 /* check that the total length reported by header is not
299 * greater than the total received size
301 if (l2_len + l3_len + rte_be_to_cpu_16(iph->payload_len) >
302 rte_pktmbuf_data_len(mbuf))
305 /* IPv6 extensions are not supported */
308 if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
309 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
310 /* Don't verify checksum for multi-segment packets. */
311 if (mbuf->nb_segs > 1)
313 if (l3 == RTE_PTYPE_L3_IPV4)
314 cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
315 else if (l3 == RTE_PTYPE_L3_IPV6)
316 cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
317 mbuf->ol_flags |= cksum ?
318 PKT_RX_L4_CKSUM_BAD :
319 PKT_RX_L4_CKSUM_GOOD;
324 tap_rx_offload_get_port_capa(void)
327 * No specific port Rx offload capabilities.
333 tap_rx_offload_get_queue_capa(void)
335 return DEV_RX_OFFLOAD_SCATTER |
336 DEV_RX_OFFLOAD_IPV4_CKSUM |
337 DEV_RX_OFFLOAD_UDP_CKSUM |
338 DEV_RX_OFFLOAD_TCP_CKSUM;
341 /* Callback to handle the rx burst of packets to the correct interface and
342 * file descriptor(s) in a multi-queue setup.
345 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
347 struct rx_queue *rxq = queue;
348 struct pmd_process_private *process_private;
350 unsigned long num_rx_bytes = 0;
351 uint32_t trigger = tap_trigger;
353 if (trigger == rxq->trigger_seen)
356 process_private = rte_eth_devices[rxq->in_port].process_private;
357 for (num_rx = 0; num_rx < nb_pkts; ) {
358 struct rte_mbuf *mbuf = rxq->pool;
359 struct rte_mbuf *seg = NULL;
360 struct rte_mbuf *new_tail = NULL;
361 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
364 len = readv(process_private->rxq_fds[rxq->queue_id],
366 1 + (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
367 rxq->nb_rx_desc : 1));
368 if (len < (int)sizeof(struct tun_pi))
371 /* Packet couldn't fit in the provided mbuf */
372 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
373 rxq->stats.ierrors++;
377 len -= sizeof(struct tun_pi);
380 mbuf->port = rxq->in_port;
382 struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
384 if (unlikely(!buf)) {
385 rxq->stats.rx_nombuf++;
386 /* No new buf has been allocated: do nothing */
387 if (!new_tail || !seg)
391 rte_pktmbuf_free(mbuf);
395 seg = seg ? seg->next : mbuf;
396 if (rxq->pool == mbuf)
399 new_tail->next = buf;
401 new_tail->next = seg->next;
403 /* iovecs[0] is reserved for packet info (pi) */
404 (*rxq->iovecs)[mbuf->nb_segs].iov_len =
405 buf->buf_len - data_off;
406 (*rxq->iovecs)[mbuf->nb_segs].iov_base =
407 (char *)buf->buf_addr + data_off;
409 seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
410 seg->data_off = data_off;
412 len -= seg->data_len;
416 /* First segment has headroom, not the others */
420 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
422 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
423 tap_verify_csum(mbuf);
425 /* account for the receive frame */
426 bufs[num_rx++] = mbuf;
427 num_rx_bytes += mbuf->pkt_len;
430 rxq->stats.ipackets += num_rx;
431 rxq->stats.ibytes += num_rx_bytes;
433 if (trigger && num_rx < nb_pkts)
434 rxq->trigger_seen = trigger;
440 tap_tx_offload_get_port_capa(void)
443 * No specific port Tx offload capabilities.
449 tap_tx_offload_get_queue_capa(void)
451 return DEV_TX_OFFLOAD_MULTI_SEGS |
452 DEV_TX_OFFLOAD_IPV4_CKSUM |
453 DEV_TX_OFFLOAD_UDP_CKSUM |
454 DEV_TX_OFFLOAD_TCP_CKSUM |
455 DEV_TX_OFFLOAD_TCP_TSO;
458 /* Finalize l4 checksum calculation */
460 tap_tx_l4_cksum(uint16_t *l4_cksum, uint16_t l4_phdr_cksum,
461 uint32_t l4_raw_cksum)
466 cksum = __rte_raw_cksum_reduce(l4_raw_cksum);
467 cksum += l4_phdr_cksum;
469 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
470 cksum = (~cksum) & 0xffff;
477 /* Accumaulate L4 raw checksums */
479 tap_tx_l4_add_rcksum(char *l4_data, unsigned int l4_len, uint16_t *l4_cksum,
480 uint32_t *l4_raw_cksum)
482 if (l4_cksum == NULL)
485 *l4_raw_cksum = __rte_raw_cksum(l4_data, l4_len, *l4_raw_cksum);
488 /* L3 and L4 pseudo headers checksum offloads */
490 tap_tx_l3_cksum(char *packet, uint64_t ol_flags, unsigned int l2_len,
491 unsigned int l3_len, unsigned int l4_len, uint16_t **l4_cksum,
492 uint16_t *l4_phdr_cksum, uint32_t *l4_raw_cksum)
494 void *l3_hdr = packet + l2_len;
496 if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
497 struct rte_ipv4_hdr *iph = l3_hdr;
500 iph->hdr_checksum = 0;
501 cksum = rte_raw_cksum(iph, l3_len);
502 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
504 if (ol_flags & PKT_TX_L4_MASK) {
507 l4_hdr = packet + l2_len + l3_len;
508 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
509 *l4_cksum = &((struct rte_udp_hdr *)l4_hdr)->dgram_cksum;
510 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
511 *l4_cksum = &((struct rte_tcp_hdr *)l4_hdr)->cksum;
515 if (ol_flags & PKT_TX_IPV4)
516 *l4_phdr_cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
518 *l4_phdr_cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
519 *l4_raw_cksum = __rte_raw_cksum(l4_hdr, l4_len, 0);
524 tap_write_mbufs(struct tx_queue *txq, uint16_t num_mbufs,
525 struct rte_mbuf **pmbufs,
526 uint16_t *num_packets, unsigned long *num_tx_bytes)
530 struct pmd_process_private *process_private;
532 process_private = rte_eth_devices[txq->out_port].process_private;
534 for (i = 0; i < num_mbufs; i++) {
535 struct rte_mbuf *mbuf = pmbufs[i];
536 struct iovec iovecs[mbuf->nb_segs + 2];
537 struct tun_pi pi = { .flags = 0, .proto = 0x00 };
538 struct rte_mbuf *seg = mbuf;
539 char m_copy[mbuf->data_len];
543 int k; /* current index in iovecs for copying segments */
544 uint16_t seg_len; /* length of first segment */
546 uint16_t *l4_cksum; /* l4 checksum (pseudo header + payload) */
547 uint32_t l4_raw_cksum = 0; /* TCP/UDP payload raw checksum */
548 uint16_t l4_phdr_cksum = 0; /* TCP/UDP pseudo header checksum */
549 uint16_t is_cksum = 0; /* in case cksum should be offloaded */
552 if (txq->type == ETH_TUNTAP_TYPE_TUN) {
554 * TUN and TAP are created with IFF_NO_PI disabled.
555 * For TUN PMD this mandatory as fields are used by
556 * Kernel tun.c to determine whether its IP or non IP
559 * The logic fetches the first byte of data from mbuf
560 * then compares whether its v4 or v6. If first byte
561 * is 4 or 6, then protocol field is updated.
563 char *buff_data = rte_pktmbuf_mtod(seg, void *);
564 proto = (*buff_data & 0xf0);
565 pi.proto = (proto == 0x40) ?
566 rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4) :
568 rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6) :
573 iovecs[k].iov_base = π
574 iovecs[k].iov_len = sizeof(pi);
577 nb_segs = mbuf->nb_segs;
579 ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
580 (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
581 (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
584 /* Support only packets with at least layer 4
585 * header included in the first segment
587 seg_len = rte_pktmbuf_data_len(mbuf);
588 l234_hlen = mbuf->l2_len + mbuf->l3_len + mbuf->l4_len;
589 if (seg_len < l234_hlen)
592 /* To change checksums, work on a * copy of l2, l3
593 * headers + l4 pseudo header
595 rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
597 tap_tx_l3_cksum(m_copy, mbuf->ol_flags,
598 mbuf->l2_len, mbuf->l3_len, mbuf->l4_len,
599 &l4_cksum, &l4_phdr_cksum,
601 iovecs[k].iov_base = m_copy;
602 iovecs[k].iov_len = l234_hlen;
605 /* Update next iovecs[] beyond l2, l3, l4 headers */
606 if (seg_len > l234_hlen) {
607 iovecs[k].iov_len = seg_len - l234_hlen;
609 rte_pktmbuf_mtod(seg, char *) +
611 tap_tx_l4_add_rcksum(iovecs[k].iov_base,
612 iovecs[k].iov_len, l4_cksum,
620 for (j = k; j <= nb_segs; j++) {
621 iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
622 iovecs[j].iov_base = rte_pktmbuf_mtod(seg, void *);
624 tap_tx_l4_add_rcksum(iovecs[j].iov_base,
625 iovecs[j].iov_len, l4_cksum,
631 tap_tx_l4_cksum(l4_cksum, l4_phdr_cksum, l4_raw_cksum);
633 /* copy the tx frame data */
634 n = writev(process_private->txq_fds[txq->queue_id], iovecs, j);
639 (*num_tx_bytes) += rte_pktmbuf_pkt_len(mbuf);
644 /* Callback to handle sending packets from the tap interface
647 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
649 struct tx_queue *txq = queue;
651 uint16_t num_packets = 0;
652 unsigned long num_tx_bytes = 0;
656 if (unlikely(nb_pkts == 0))
659 struct rte_mbuf *gso_mbufs[MAX_GSO_MBUFS];
660 max_size = *txq->mtu + (RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN + 4);
661 for (i = 0; i < nb_pkts; i++) {
662 struct rte_mbuf *mbuf_in = bufs[num_tx];
663 struct rte_mbuf **mbuf;
664 uint16_t num_mbufs = 0;
665 uint16_t tso_segsz = 0;
671 tso = mbuf_in->ol_flags & PKT_TX_TCP_SEG;
673 struct rte_gso_ctx *gso_ctx = &txq->gso_ctx;
675 /* TCP segmentation implies TCP checksum offload */
676 mbuf_in->ol_flags |= PKT_TX_TCP_CKSUM;
678 /* gso size is calculated without RTE_ETHER_CRC_LEN */
679 hdrs_len = mbuf_in->l2_len + mbuf_in->l3_len +
681 tso_segsz = mbuf_in->tso_segsz + hdrs_len;
682 if (unlikely(tso_segsz == hdrs_len) ||
683 tso_segsz > *txq->mtu) {
687 gso_ctx->gso_size = tso_segsz;
688 /* 'mbuf_in' packet to segment */
689 num_tso_mbufs = rte_gso_segment(mbuf_in,
690 gso_ctx, /* gso control block */
691 (struct rte_mbuf **)&gso_mbufs, /* out mbufs */
692 RTE_DIM(gso_mbufs)); /* max tso mbufs */
694 /* ret contains the number of new created mbufs */
695 if (num_tso_mbufs < 0)
699 num_mbufs = num_tso_mbufs;
701 /* stats.errs will be incremented */
702 if (rte_pktmbuf_pkt_len(mbuf_in) > max_size)
705 /* ret 0 indicates no new mbufs were created */
711 ret = tap_write_mbufs(txq, num_mbufs, mbuf,
712 &num_packets, &num_tx_bytes);
716 if (num_tso_mbufs > 0)
717 rte_pktmbuf_free_bulk(mbuf, num_tso_mbufs);
721 /* free original mbuf */
722 rte_pktmbuf_free(mbuf_in);
724 if (num_tso_mbufs > 0)
725 rte_pktmbuf_free_bulk(mbuf, num_tso_mbufs);
728 txq->stats.opackets += num_packets;
729 txq->stats.errs += nb_pkts - num_tx;
730 txq->stats.obytes += num_tx_bytes;
736 tap_ioctl_req2str(unsigned long request)
740 return "SIOCSIFFLAGS";
742 return "SIOCGIFFLAGS";
744 return "SIOCGIFHWADDR";
746 return "SIOCSIFHWADDR";
754 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
755 struct ifreq *ifr, int set, enum ioctl_mode mode)
757 short req_flags = ifr->ifr_flags;
758 int remote = pmd->remote_if_index &&
759 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
761 if (!pmd->remote_if_index && mode == REMOTE_ONLY)
764 * If there is a remote netdevice, apply ioctl on it, then apply it on
769 strlcpy(ifr->ifr_name, pmd->remote_iface, IFNAMSIZ);
770 else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
771 strlcpy(ifr->ifr_name, pmd->name, IFNAMSIZ);
774 /* fetch current flags to leave other flags untouched */
775 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
778 ifr->ifr_flags |= req_flags;
780 ifr->ifr_flags &= ~req_flags;
788 TAP_LOG(WARNING, "%s: ioctl() called with wrong arg",
792 if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
794 if (remote-- && mode == LOCAL_AND_REMOTE)
799 TAP_LOG(DEBUG, "%s(%s) failed: %s(%d)", ifr->ifr_name,
800 tap_ioctl_req2str(request), strerror(errno), errno);
805 tap_link_set_down(struct rte_eth_dev *dev)
807 struct pmd_internals *pmd = dev->data->dev_private;
808 struct ifreq ifr = { .ifr_flags = IFF_UP };
810 dev->data->dev_link.link_status = ETH_LINK_DOWN;
811 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
815 tap_link_set_up(struct rte_eth_dev *dev)
817 struct pmd_internals *pmd = dev->data->dev_private;
818 struct ifreq ifr = { .ifr_flags = IFF_UP };
820 dev->data->dev_link.link_status = ETH_LINK_UP;
821 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
825 tap_dev_start(struct rte_eth_dev *dev)
829 err = tap_intr_handle_set(dev, 1);
833 err = tap_link_set_up(dev);
837 for (i = 0; i < dev->data->nb_tx_queues; i++)
838 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
839 for (i = 0; i < dev->data->nb_rx_queues; i++)
840 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
845 /* This function gets called when the current port gets stopped.
848 tap_dev_stop(struct rte_eth_dev *dev)
852 for (i = 0; i < dev->data->nb_tx_queues; i++)
853 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
854 for (i = 0; i < dev->data->nb_rx_queues; i++)
855 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
857 tap_intr_handle_set(dev, 0);
858 tap_link_set_down(dev);
862 tap_dev_configure(struct rte_eth_dev *dev)
864 struct pmd_internals *pmd = dev->data->dev_private;
866 if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
868 "%s: number of rx queues %d exceeds max num of queues %d",
870 dev->data->nb_rx_queues,
871 RTE_PMD_TAP_MAX_QUEUES);
874 if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
876 "%s: number of tx queues %d exceeds max num of queues %d",
878 dev->data->nb_tx_queues,
879 RTE_PMD_TAP_MAX_QUEUES);
883 TAP_LOG(INFO, "%s: %s: TX configured queues number: %u",
884 dev->device->name, pmd->name, dev->data->nb_tx_queues);
886 TAP_LOG(INFO, "%s: %s: RX configured queues number: %u",
887 dev->device->name, pmd->name, dev->data->nb_rx_queues);
893 tap_dev_speed_capa(void)
895 uint32_t speed = pmd_link.link_speed;
898 if (speed >= ETH_SPEED_NUM_10M)
899 capa |= ETH_LINK_SPEED_10M;
900 if (speed >= ETH_SPEED_NUM_100M)
901 capa |= ETH_LINK_SPEED_100M;
902 if (speed >= ETH_SPEED_NUM_1G)
903 capa |= ETH_LINK_SPEED_1G;
904 if (speed >= ETH_SPEED_NUM_5G)
905 capa |= ETH_LINK_SPEED_2_5G;
906 if (speed >= ETH_SPEED_NUM_5G)
907 capa |= ETH_LINK_SPEED_5G;
908 if (speed >= ETH_SPEED_NUM_10G)
909 capa |= ETH_LINK_SPEED_10G;
910 if (speed >= ETH_SPEED_NUM_20G)
911 capa |= ETH_LINK_SPEED_20G;
912 if (speed >= ETH_SPEED_NUM_25G)
913 capa |= ETH_LINK_SPEED_25G;
914 if (speed >= ETH_SPEED_NUM_40G)
915 capa |= ETH_LINK_SPEED_40G;
916 if (speed >= ETH_SPEED_NUM_50G)
917 capa |= ETH_LINK_SPEED_50G;
918 if (speed >= ETH_SPEED_NUM_56G)
919 capa |= ETH_LINK_SPEED_56G;
920 if (speed >= ETH_SPEED_NUM_100G)
921 capa |= ETH_LINK_SPEED_100G;
927 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
929 struct pmd_internals *internals = dev->data->dev_private;
931 dev_info->if_index = internals->if_index;
932 dev_info->max_mac_addrs = 1;
933 dev_info->max_rx_pktlen = (uint32_t)RTE_ETHER_MAX_VLAN_FRAME_LEN;
934 dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
935 dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
936 dev_info->min_rx_bufsize = 0;
937 dev_info->speed_capa = tap_dev_speed_capa();
938 dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
939 dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
940 dev_info->rx_queue_offload_capa;
941 dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
942 dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
943 dev_info->tx_queue_offload_capa;
944 dev_info->hash_key_size = TAP_RSS_HASH_KEY_SIZE;
946 * limitation: TAP supports all of IP, UDP and TCP hash
947 * functions together and not in partial combinations
949 dev_info->flow_type_rss_offloads = ~TAP_RSS_HF_MASK;
955 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
957 unsigned int i, imax;
958 unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
959 unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
960 unsigned long rx_nombuf = 0, ierrors = 0;
961 const struct pmd_internals *pmd = dev->data->dev_private;
963 /* rx queue statistics */
964 imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
965 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
966 for (i = 0; i < imax; i++) {
967 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
968 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
969 rx_total += tap_stats->q_ipackets[i];
970 rx_bytes_total += tap_stats->q_ibytes[i];
971 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
972 ierrors += pmd->rxq[i].stats.ierrors;
975 /* tx queue statistics */
976 imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
977 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
979 for (i = 0; i < imax; i++) {
980 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
981 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
982 tx_total += tap_stats->q_opackets[i];
983 tx_err_total += pmd->txq[i].stats.errs;
984 tx_bytes_total += tap_stats->q_obytes[i];
987 tap_stats->ipackets = rx_total;
988 tap_stats->ibytes = rx_bytes_total;
989 tap_stats->ierrors = ierrors;
990 tap_stats->rx_nombuf = rx_nombuf;
991 tap_stats->opackets = tx_total;
992 tap_stats->oerrors = tx_err_total;
993 tap_stats->obytes = tx_bytes_total;
998 tap_stats_reset(struct rte_eth_dev *dev)
1001 struct pmd_internals *pmd = dev->data->dev_private;
1003 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1004 pmd->rxq[i].stats.ipackets = 0;
1005 pmd->rxq[i].stats.ibytes = 0;
1006 pmd->rxq[i].stats.ierrors = 0;
1007 pmd->rxq[i].stats.rx_nombuf = 0;
1009 pmd->txq[i].stats.opackets = 0;
1010 pmd->txq[i].stats.errs = 0;
1011 pmd->txq[i].stats.obytes = 0;
1018 tap_dev_close(struct rte_eth_dev *dev)
1021 struct pmd_internals *internals = dev->data->dev_private;
1022 struct pmd_process_private *process_private = dev->process_private;
1024 tap_link_set_down(dev);
1025 tap_flow_flush(dev, NULL);
1026 tap_flow_implicit_flush(internals, NULL);
1028 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1029 if (process_private->rxq_fds[i] != -1) {
1030 close(process_private->rxq_fds[i]);
1031 process_private->rxq_fds[i] = -1;
1033 if (process_private->txq_fds[i] != -1) {
1034 close(process_private->txq_fds[i]);
1035 process_private->txq_fds[i] = -1;
1039 if (internals->remote_if_index) {
1040 /* Restore initial remote state */
1041 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
1042 &internals->remote_initial_flags);
1045 if (internals->ka_fd != -1) {
1046 close(internals->ka_fd);
1047 internals->ka_fd = -1;
1050 * Since TUN device has no more opened file descriptors
1051 * it will be removed from kernel
1056 tap_rx_queue_release(void *queue)
1058 struct rx_queue *rxq = queue;
1059 struct pmd_process_private *process_private;
1063 process_private = rte_eth_devices[rxq->in_port].process_private;
1064 if (process_private->rxq_fds[rxq->queue_id] > 0) {
1065 close(process_private->rxq_fds[rxq->queue_id]);
1066 process_private->rxq_fds[rxq->queue_id] = -1;
1067 rte_pktmbuf_free(rxq->pool);
1068 rte_free(rxq->iovecs);
1075 tap_tx_queue_release(void *queue)
1077 struct tx_queue *txq = queue;
1078 struct pmd_process_private *process_private;
1082 process_private = rte_eth_devices[txq->out_port].process_private;
1084 if (process_private->txq_fds[txq->queue_id] > 0) {
1085 close(process_private->txq_fds[txq->queue_id]);
1086 process_private->txq_fds[txq->queue_id] = -1;
1091 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
1093 struct rte_eth_link *dev_link = &dev->data->dev_link;
1094 struct pmd_internals *pmd = dev->data->dev_private;
1095 struct ifreq ifr = { .ifr_flags = 0 };
1097 if (pmd->remote_if_index) {
1098 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
1099 if (!(ifr.ifr_flags & IFF_UP) ||
1100 !(ifr.ifr_flags & IFF_RUNNING)) {
1101 dev_link->link_status = ETH_LINK_DOWN;
1105 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
1106 dev_link->link_status =
1107 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
1114 tap_promisc_enable(struct rte_eth_dev *dev)
1116 struct pmd_internals *pmd = dev->data->dev_private;
1117 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1120 ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1124 if (pmd->remote_if_index && !pmd->flow_isolate) {
1125 dev->data->promiscuous = 1;
1126 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
1128 /* Rollback promisc flag */
1129 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1131 * rte_eth_dev_promiscuous_enable() rollback
1132 * dev->data->promiscuous in the case of failure.
1142 tap_promisc_disable(struct rte_eth_dev *dev)
1144 struct pmd_internals *pmd = dev->data->dev_private;
1145 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1148 ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1152 if (pmd->remote_if_index && !pmd->flow_isolate) {
1153 dev->data->promiscuous = 0;
1154 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
1156 /* Rollback promisc flag */
1157 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1159 * rte_eth_dev_promiscuous_disable() rollback
1160 * dev->data->promiscuous in the case of failure.
1170 tap_allmulti_enable(struct rte_eth_dev *dev)
1172 struct pmd_internals *pmd = dev->data->dev_private;
1173 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1176 ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1180 if (pmd->remote_if_index && !pmd->flow_isolate) {
1181 dev->data->all_multicast = 1;
1182 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
1184 /* Rollback allmulti flag */
1185 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1187 * rte_eth_dev_allmulticast_enable() rollback
1188 * dev->data->all_multicast in the case of failure.
1198 tap_allmulti_disable(struct rte_eth_dev *dev)
1200 struct pmd_internals *pmd = dev->data->dev_private;
1201 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1204 ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1208 if (pmd->remote_if_index && !pmd->flow_isolate) {
1209 dev->data->all_multicast = 0;
1210 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
1212 /* Rollback allmulti flag */
1213 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1215 * rte_eth_dev_allmulticast_disable() rollback
1216 * dev->data->all_multicast in the case of failure.
1226 tap_mac_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1228 struct pmd_internals *pmd = dev->data->dev_private;
1229 enum ioctl_mode mode = LOCAL_ONLY;
1233 if (pmd->type == ETH_TUNTAP_TYPE_TUN) {
1234 TAP_LOG(ERR, "%s: can't MAC address for TUN",
1239 if (rte_is_zero_ether_addr(mac_addr)) {
1240 TAP_LOG(ERR, "%s: can't set an empty MAC address",
1244 /* Check the actual current MAC address on the tap netdevice */
1245 ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY);
1248 if (rte_is_same_ether_addr(
1249 (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1252 /* Check the current MAC address on the remote */
1253 ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY);
1256 if (!rte_is_same_ether_addr(
1257 (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1259 mode = LOCAL_AND_REMOTE;
1260 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1261 rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, RTE_ETHER_ADDR_LEN);
1262 ret = tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode);
1265 rte_memcpy(&pmd->eth_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1266 if (pmd->remote_if_index && !pmd->flow_isolate) {
1267 /* Replace MAC redirection rule after a MAC change */
1268 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC);
1271 "%s: Couldn't delete MAC redirection rule",
1275 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC);
1278 "%s: Couldn't add MAC redirection rule",
1288 tap_gso_ctx_setup(struct rte_gso_ctx *gso_ctx, struct rte_eth_dev *dev)
1294 * Create private mbuf pool with TAP_GSO_MBUF_SEG_SIZE bytes
1295 * size per mbuf use this pool for both direct and indirect mbufs
1298 struct rte_mempool *mp; /* Mempool for GSO packets */
1300 /* initialize GSO context */
1301 gso_types = DEV_TX_OFFLOAD_TCP_TSO;
1302 snprintf(pool_name, sizeof(pool_name), "mp_%s", dev->device->name);
1303 mp = rte_mempool_lookup((const char *)pool_name);
1305 mp = rte_pktmbuf_pool_create(pool_name, TAP_GSO_MBUFS_NUM,
1306 TAP_GSO_MBUF_CACHE_SIZE, 0,
1307 RTE_PKTMBUF_HEADROOM + TAP_GSO_MBUF_SEG_SIZE,
1310 struct pmd_internals *pmd = dev->data->dev_private;
1313 "%s: failed to create mbuf pool for device %s\n",
1314 pmd->name, dev->device->name);
1319 gso_ctx->direct_pool = mp;
1320 gso_ctx->indirect_pool = mp;
1321 gso_ctx->gso_types = gso_types;
1322 gso_ctx->gso_size = 0; /* gso_size is set in tx_burst() per packet */
1329 tap_setup_queue(struct rte_eth_dev *dev,
1330 struct pmd_internals *internals,
1338 struct pmd_internals *pmd = dev->data->dev_private;
1339 struct pmd_process_private *process_private = dev->process_private;
1340 struct rx_queue *rx = &internals->rxq[qid];
1341 struct tx_queue *tx = &internals->txq[qid];
1342 struct rte_gso_ctx *gso_ctx;
1345 fd = &process_private->rxq_fds[qid];
1346 other_fd = &process_private->txq_fds[qid];
1350 fd = &process_private->txq_fds[qid];
1351 other_fd = &process_private->rxq_fds[qid];
1353 gso_ctx = &tx->gso_ctx;
1356 /* fd for this queue already exists */
1357 TAP_LOG(DEBUG, "%s: fd %d for %s queue qid %d exists",
1358 pmd->name, *fd, dir, qid);
1360 } else if (*other_fd != -1) {
1361 /* Only other_fd exists. dup it */
1362 *fd = dup(*other_fd);
1365 TAP_LOG(ERR, "%s: dup() failed.", pmd->name);
1368 TAP_LOG(DEBUG, "%s: dup fd %d for %s queue qid %d (%d)",
1369 pmd->name, *other_fd, dir, qid, *fd);
1371 /* Both RX and TX fds do not exist (equal -1). Create fd */
1372 *fd = tun_alloc(pmd, 0);
1374 *fd = -1; /* restore original value */
1375 TAP_LOG(ERR, "%s: tun_alloc() failed.", pmd->name);
1378 TAP_LOG(DEBUG, "%s: add %s queue for qid %d fd %d",
1379 pmd->name, dir, qid, *fd);
1382 tx->mtu = &dev->data->mtu;
1383 rx->rxmode = &dev->data->dev_conf.rxmode;
1385 ret = tap_gso_ctx_setup(gso_ctx, dev);
1390 tx->type = pmd->type;
1396 tap_rx_queue_setup(struct rte_eth_dev *dev,
1397 uint16_t rx_queue_id,
1398 uint16_t nb_rx_desc,
1399 unsigned int socket_id,
1400 const struct rte_eth_rxconf *rx_conf __rte_unused,
1401 struct rte_mempool *mp)
1403 struct pmd_internals *internals = dev->data->dev_private;
1404 struct pmd_process_private *process_private = dev->process_private;
1405 struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1406 struct rte_mbuf **tmp = &rxq->pool;
1407 long iov_max = sysconf(_SC_IOV_MAX);
1411 "_SC_IOV_MAX is not defined. Using %d as default",
1412 TAP_IOV_DEFAULT_MAX);
1413 iov_max = TAP_IOV_DEFAULT_MAX;
1415 uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1416 struct iovec (*iovecs)[nb_desc + 1];
1417 int data_off = RTE_PKTMBUF_HEADROOM;
1422 if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1424 "nb_rx_queues %d too small or mempool NULL",
1425 dev->data->nb_rx_queues);
1430 rxq->trigger_seen = 1; /* force initial burst */
1431 rxq->in_port = dev->data->port_id;
1432 rxq->queue_id = rx_queue_id;
1433 rxq->nb_rx_desc = nb_desc;
1434 iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1438 "%s: Couldn't allocate %d RX descriptors",
1439 dev->device->name, nb_desc);
1442 rxq->iovecs = iovecs;
1444 dev->data->rx_queues[rx_queue_id] = rxq;
1445 fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1451 (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1452 (*rxq->iovecs)[0].iov_base = &rxq->pi;
1454 for (i = 1; i <= nb_desc; i++) {
1455 *tmp = rte_pktmbuf_alloc(rxq->mp);
1458 "%s: couldn't allocate memory for queue %d",
1459 dev->device->name, rx_queue_id);
1463 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1464 (*rxq->iovecs)[i].iov_base =
1465 (char *)(*tmp)->buf_addr + data_off;
1467 tmp = &(*tmp)->next;
1470 TAP_LOG(DEBUG, " RX TUNTAP device name %s, qid %d on fd %d",
1471 internals->name, rx_queue_id,
1472 process_private->rxq_fds[rx_queue_id]);
1477 rte_pktmbuf_free(rxq->pool);
1479 rte_free(rxq->iovecs);
1485 tap_tx_queue_setup(struct rte_eth_dev *dev,
1486 uint16_t tx_queue_id,
1487 uint16_t nb_tx_desc __rte_unused,
1488 unsigned int socket_id __rte_unused,
1489 const struct rte_eth_txconf *tx_conf)
1491 struct pmd_internals *internals = dev->data->dev_private;
1492 struct pmd_process_private *process_private = dev->process_private;
1493 struct tx_queue *txq;
1497 if (tx_queue_id >= dev->data->nb_tx_queues)
1499 dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1500 txq = dev->data->tx_queues[tx_queue_id];
1501 txq->out_port = dev->data->port_id;
1502 txq->queue_id = tx_queue_id;
1504 offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1505 txq->csum = !!(offloads &
1506 (DEV_TX_OFFLOAD_IPV4_CKSUM |
1507 DEV_TX_OFFLOAD_UDP_CKSUM |
1508 DEV_TX_OFFLOAD_TCP_CKSUM));
1510 ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1514 " TX TUNTAP device name %s, qid %d on fd %d csum %s",
1515 internals->name, tx_queue_id,
1516 process_private->txq_fds[tx_queue_id],
1517 txq->csum ? "on" : "off");
1523 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1525 struct pmd_internals *pmd = dev->data->dev_private;
1526 struct ifreq ifr = { .ifr_mtu = mtu };
1529 err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1531 dev->data->mtu = mtu;
1537 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1538 struct rte_ether_addr *mc_addr_set __rte_unused,
1539 uint32_t nb_mc_addr __rte_unused)
1542 * Nothing to do actually: the tap has no filtering whatsoever, every
1543 * packet is received.
1549 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1551 struct rte_eth_dev *dev = arg;
1552 struct pmd_internals *pmd = dev->data->dev_private;
1553 struct ifinfomsg *info = NLMSG_DATA(nh);
1555 if (nh->nlmsg_type != RTM_NEWLINK ||
1556 (info->ifi_index != pmd->if_index &&
1557 info->ifi_index != pmd->remote_if_index))
1559 return tap_link_update(dev, 0);
1563 tap_dev_intr_handler(void *cb_arg)
1565 struct rte_eth_dev *dev = cb_arg;
1566 struct pmd_internals *pmd = dev->data->dev_private;
1568 tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1572 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1574 struct pmd_internals *pmd = dev->data->dev_private;
1576 /* In any case, disable interrupt if the conf is no longer there. */
1577 if (!dev->data->dev_conf.intr_conf.lsc) {
1578 if (pmd->intr_handle.fd != -1) {
1579 tap_nl_final(pmd->intr_handle.fd);
1580 rte_intr_callback_unregister(&pmd->intr_handle,
1581 tap_dev_intr_handler, dev);
1586 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1587 if (unlikely(pmd->intr_handle.fd == -1))
1589 return rte_intr_callback_register(
1590 &pmd->intr_handle, tap_dev_intr_handler, dev);
1592 tap_nl_final(pmd->intr_handle.fd);
1593 return rte_intr_callback_unregister(&pmd->intr_handle,
1594 tap_dev_intr_handler, dev);
1598 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1602 err = tap_lsc_intr_handle_set(dev, set);
1605 tap_rx_intr_vec_set(dev, 0);
1608 err = tap_rx_intr_vec_set(dev, set);
1610 tap_lsc_intr_handle_set(dev, 0);
1614 static const uint32_t*
1615 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1617 static const uint32_t ptypes[] = {
1618 RTE_PTYPE_INNER_L2_ETHER,
1619 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1620 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1621 RTE_PTYPE_INNER_L3_IPV4,
1622 RTE_PTYPE_INNER_L3_IPV4_EXT,
1623 RTE_PTYPE_INNER_L3_IPV6,
1624 RTE_PTYPE_INNER_L3_IPV6_EXT,
1625 RTE_PTYPE_INNER_L4_FRAG,
1626 RTE_PTYPE_INNER_L4_UDP,
1627 RTE_PTYPE_INNER_L4_TCP,
1628 RTE_PTYPE_INNER_L4_SCTP,
1630 RTE_PTYPE_L2_ETHER_VLAN,
1631 RTE_PTYPE_L2_ETHER_QINQ,
1633 RTE_PTYPE_L3_IPV4_EXT,
1634 RTE_PTYPE_L3_IPV6_EXT,
1646 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1647 struct rte_eth_fc_conf *fc_conf)
1649 fc_conf->mode = RTE_FC_NONE;
1654 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1655 struct rte_eth_fc_conf *fc_conf)
1657 if (fc_conf->mode != RTE_FC_NONE)
1663 * DPDK callback to update the RSS hash configuration.
1666 * Pointer to Ethernet device structure.
1667 * @param[in] rss_conf
1668 * RSS configuration data.
1671 * 0 on success, a negative errno value otherwise and rte_errno is set.
1674 tap_rss_hash_update(struct rte_eth_dev *dev,
1675 struct rte_eth_rss_conf *rss_conf)
1677 if (rss_conf->rss_hf & TAP_RSS_HF_MASK) {
1681 if (rss_conf->rss_key && rss_conf->rss_key_len) {
1683 * Currently TAP RSS key is hard coded
1684 * and cannot be updated
1687 "port %u RSS key cannot be updated",
1688 dev->data->port_id);
1696 tap_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1698 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1704 tap_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1706 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1712 tap_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1714 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1720 tap_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1722 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1726 static const struct eth_dev_ops ops = {
1727 .dev_start = tap_dev_start,
1728 .dev_stop = tap_dev_stop,
1729 .dev_close = tap_dev_close,
1730 .dev_configure = tap_dev_configure,
1731 .dev_infos_get = tap_dev_info,
1732 .rx_queue_setup = tap_rx_queue_setup,
1733 .tx_queue_setup = tap_tx_queue_setup,
1734 .rx_queue_start = tap_rx_queue_start,
1735 .tx_queue_start = tap_tx_queue_start,
1736 .rx_queue_stop = tap_rx_queue_stop,
1737 .tx_queue_stop = tap_tx_queue_stop,
1738 .rx_queue_release = tap_rx_queue_release,
1739 .tx_queue_release = tap_tx_queue_release,
1740 .flow_ctrl_get = tap_flow_ctrl_get,
1741 .flow_ctrl_set = tap_flow_ctrl_set,
1742 .link_update = tap_link_update,
1743 .dev_set_link_up = tap_link_set_up,
1744 .dev_set_link_down = tap_link_set_down,
1745 .promiscuous_enable = tap_promisc_enable,
1746 .promiscuous_disable = tap_promisc_disable,
1747 .allmulticast_enable = tap_allmulti_enable,
1748 .allmulticast_disable = tap_allmulti_disable,
1749 .mac_addr_set = tap_mac_set,
1750 .mtu_set = tap_mtu_set,
1751 .set_mc_addr_list = tap_set_mc_addr_list,
1752 .stats_get = tap_stats_get,
1753 .stats_reset = tap_stats_reset,
1754 .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1755 .rss_hash_update = tap_rss_hash_update,
1756 .filter_ctrl = tap_dev_filter_ctrl,
1759 static const char *tuntap_types[ETH_TUNTAP_TYPE_MAX] = {
1760 "UNKNOWN", "TUN", "TAP"
1764 eth_dev_tap_create(struct rte_vdev_device *vdev, const char *tap_name,
1765 char *remote_iface, struct rte_ether_addr *mac_addr,
1766 enum rte_tuntap_type type)
1768 int numa_node = rte_socket_id();
1769 struct rte_eth_dev *dev;
1770 struct pmd_internals *pmd;
1771 struct pmd_process_private *process_private;
1772 const char *tuntap_name = tuntap_types[type];
1773 struct rte_eth_dev_data *data;
1777 TAP_LOG(DEBUG, "%s device on numa %u", tuntap_name, rte_socket_id());
1779 dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1781 TAP_LOG(ERR, "%s Unable to allocate device struct",
1783 goto error_exit_nodev;
1786 process_private = (struct pmd_process_private *)
1787 rte_zmalloc_socket(tap_name, sizeof(struct pmd_process_private),
1788 RTE_CACHE_LINE_SIZE, dev->device->numa_node);
1790 if (process_private == NULL) {
1791 TAP_LOG(ERR, "Failed to alloc memory for process private");
1794 pmd = dev->data->dev_private;
1795 dev->process_private = process_private;
1797 strlcpy(pmd->name, tap_name, sizeof(pmd->name));
1800 pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1801 if (pmd->ioctl_sock == -1) {
1803 "%s Unable to get a socket for management: %s",
1804 tuntap_name, strerror(errno));
1808 /* Setup some default values */
1810 data->dev_private = pmd;
1811 data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1812 data->numa_node = numa_node;
1814 data->dev_link = pmd_link;
1815 data->mac_addrs = &pmd->eth_addr;
1816 /* Set the number of RX and TX queues */
1817 data->nb_rx_queues = 0;
1818 data->nb_tx_queues = 0;
1820 dev->dev_ops = &ops;
1821 dev->rx_pkt_burst = pmd_rx_burst;
1822 dev->tx_pkt_burst = pmd_tx_burst;
1824 pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1825 pmd->intr_handle.fd = -1;
1826 dev->intr_handle = &pmd->intr_handle;
1828 /* Presetup the fds to -1 as being not valid */
1830 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1831 process_private->rxq_fds[i] = -1;
1832 process_private->txq_fds[i] = -1;
1835 if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1836 if (rte_is_zero_ether_addr(mac_addr))
1837 rte_eth_random_addr((uint8_t *)&pmd->eth_addr);
1839 rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1843 * Allocate a TUN device keep-alive file descriptor that will only be
1844 * closed when the TUN device itself is closed or removed.
1845 * This keep-alive file descriptor will guarantee that the TUN device
1846 * exists even when all of its queues are closed
1848 pmd->ka_fd = tun_alloc(pmd, 1);
1849 if (pmd->ka_fd == -1) {
1850 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
1853 TAP_LOG(DEBUG, "allocated %s", pmd->name);
1855 ifr.ifr_mtu = dev->data->mtu;
1856 if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1859 if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1860 memset(&ifr, 0, sizeof(struct ifreq));
1861 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1862 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
1863 RTE_ETHER_ADDR_LEN);
1864 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1869 * Set up everything related to rte_flow:
1871 * - tap / remote if_index
1872 * - mandatory QDISCs
1873 * - rte_flow actual/implicit lists
1876 pmd->nlsk_fd = tap_nl_init(0);
1877 if (pmd->nlsk_fd == -1) {
1878 TAP_LOG(WARNING, "%s: failed to create netlink socket.",
1880 goto disable_rte_flow;
1882 pmd->if_index = if_nametoindex(pmd->name);
1883 if (!pmd->if_index) {
1884 TAP_LOG(ERR, "%s: failed to get if_index.", pmd->name);
1885 goto disable_rte_flow;
1887 if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1888 TAP_LOG(ERR, "%s: failed to create multiq qdisc.",
1890 goto disable_rte_flow;
1892 if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1893 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1895 goto disable_rte_flow;
1897 LIST_INIT(&pmd->flows);
1899 if (strlen(remote_iface)) {
1900 pmd->remote_if_index = if_nametoindex(remote_iface);
1901 if (!pmd->remote_if_index) {
1902 TAP_LOG(ERR, "%s: failed to get %s if_index.",
1903 pmd->name, remote_iface);
1906 strlcpy(pmd->remote_iface, remote_iface, RTE_ETH_NAME_MAX_LEN);
1908 /* Save state of remote device */
1909 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1911 /* Replicate remote MAC address */
1912 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1913 TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1914 pmd->name, pmd->remote_iface);
1917 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1918 RTE_ETHER_ADDR_LEN);
1919 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1920 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1921 TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1922 pmd->name, remote_iface);
1927 * Flush usually returns negative value because it tries to
1928 * delete every QDISC (and on a running device, one QDISC at
1929 * least is needed). Ignore negative return value.
1931 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1932 if (qdisc_create_ingress(pmd->nlsk_fd,
1933 pmd->remote_if_index) < 0) {
1934 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1938 LIST_INIT(&pmd->implicit_flows);
1939 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1940 tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1941 tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1942 tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1944 "%s: failed to create implicit rules.",
1950 rte_eth_dev_probing_finish(dev);
1954 TAP_LOG(ERR, " Disabling rte flow support: %s(%d)",
1955 strerror(errno), errno);
1956 if (strlen(remote_iface)) {
1957 TAP_LOG(ERR, "Remote feature requires flow support.");
1960 rte_eth_dev_probing_finish(dev);
1964 TAP_LOG(ERR, " Can't set up remote feature: %s(%d)",
1965 strerror(errno), errno);
1966 tap_flow_implicit_flush(pmd, NULL);
1969 if (pmd->ioctl_sock > 0)
1970 close(pmd->ioctl_sock);
1971 /* mac_addrs must not be freed alone because part of dev_private */
1972 dev->data->mac_addrs = NULL;
1973 rte_eth_dev_release_port(dev);
1976 TAP_LOG(ERR, "%s Unable to initialize %s",
1977 tuntap_name, rte_vdev_device_name(vdev));
1982 /* make sure name is a possible Linux network device name */
1984 is_valid_iface(const char *name)
1989 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1993 if (*name == '/' || *name == ':' || isspace(*name))
2001 set_interface_name(const char *key __rte_unused,
2005 char *name = (char *)extra_args;
2008 if (!is_valid_iface(value)) {
2009 TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
2013 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
2015 /* use tap%d which causes kernel to choose next available */
2016 strlcpy(name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2022 set_remote_iface(const char *key __rte_unused,
2026 char *name = (char *)extra_args;
2029 if (!is_valid_iface(value)) {
2030 TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
2034 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
2040 static int parse_user_mac(struct rte_ether_addr *user_mac,
2043 unsigned int index = 0;
2044 char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
2046 if (user_mac == NULL || value == NULL)
2049 strlcpy(mac_temp, value, sizeof(mac_temp));
2050 mac_byte = strtok(mac_temp, ":");
2052 while ((mac_byte != NULL) &&
2053 (strlen(mac_byte) <= 2) &&
2054 (strlen(mac_byte) == strspn(mac_byte,
2055 ETH_TAP_CMP_MAC_FMT))) {
2056 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
2057 mac_byte = strtok(NULL, ":");
2064 set_mac_type(const char *key __rte_unused,
2068 struct rte_ether_addr *user_mac = extra_args;
2073 if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
2074 static int iface_idx;
2076 /* fixed mac = 00:64:74:61:70:<iface_idx> */
2077 memcpy((char *)user_mac->addr_bytes, "\0dtap",
2078 RTE_ETHER_ADDR_LEN);
2079 user_mac->addr_bytes[RTE_ETHER_ADDR_LEN - 1] =
2084 if (parse_user_mac(user_mac, value) != 6)
2087 TAP_LOG(DEBUG, "TAP user MAC param (%s)", value);
2091 TAP_LOG(ERR, "TAP user MAC (%s) is not in format (%s|%s)",
2092 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
2097 * Open a TUN interface device. TUN PMD
2098 * 1) sets tap_type as false
2099 * 2) intakes iface as argument.
2100 * 3) as interface is virtual set speed to 10G
2103 rte_pmd_tun_probe(struct rte_vdev_device *dev)
2105 const char *name, *params;
2107 struct rte_kvargs *kvlist = NULL;
2108 char tun_name[RTE_ETH_NAME_MAX_LEN];
2109 char remote_iface[RTE_ETH_NAME_MAX_LEN];
2110 struct rte_eth_dev *eth_dev;
2112 name = rte_vdev_device_name(dev);
2113 params = rte_vdev_device_args(dev);
2114 memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2116 if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
2117 strlen(params) == 0) {
2118 eth_dev = rte_eth_dev_attach_secondary(name);
2120 TAP_LOG(ERR, "Failed to probe %s", name);
2123 eth_dev->dev_ops = &ops;
2124 eth_dev->device = &dev->device;
2125 rte_eth_dev_probing_finish(eth_dev);
2129 /* use tun%d which causes kernel to choose next available */
2130 strlcpy(tun_name, DEFAULT_TUN_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2132 if (params && (params[0] != '\0')) {
2133 TAP_LOG(DEBUG, "parameters (%s)", params);
2135 kvlist = rte_kvargs_parse(params, valid_arguments);
2137 if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2138 ret = rte_kvargs_process(kvlist,
2140 &set_interface_name,
2148 pmd_link.link_speed = ETH_SPEED_NUM_10G;
2150 TAP_LOG(DEBUG, "Initializing pmd_tun for %s", name);
2152 ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0,
2153 ETH_TUNTAP_TYPE_TUN);
2157 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2160 rte_kvargs_free(kvlist);
2165 /* Request queue file descriptors from secondary to primary. */
2167 tap_mp_attach_queues(const char *port_name, struct rte_eth_dev *dev)
2170 struct timespec timeout = {.tv_sec = 1, .tv_nsec = 0};
2171 struct rte_mp_msg request, *reply;
2172 struct rte_mp_reply replies;
2173 struct ipc_queues *request_param = (struct ipc_queues *)request.param;
2174 struct ipc_queues *reply_param;
2175 struct pmd_process_private *process_private = dev->process_private;
2176 int queue, fd_iterator;
2178 /* Prepare the request */
2179 memset(&request, 0, sizeof(request));
2180 strlcpy(request.name, TAP_MP_KEY, sizeof(request.name));
2181 strlcpy(request_param->port_name, port_name,
2182 sizeof(request_param->port_name));
2183 request.len_param = sizeof(*request_param);
2184 /* Send request and receive reply */
2185 ret = rte_mp_request_sync(&request, &replies, &timeout);
2186 if (ret < 0 || replies.nb_received != 1) {
2187 TAP_LOG(ERR, "Failed to request queues from primary: %d",
2191 reply = &replies.msgs[0];
2192 reply_param = (struct ipc_queues *)reply->param;
2193 TAP_LOG(DEBUG, "Received IPC reply for %s", reply_param->port_name);
2195 /* Attach the queues from received file descriptors */
2196 if (reply_param->rxq_count + reply_param->txq_count != reply->num_fds) {
2197 TAP_LOG(ERR, "Unexpected number of fds received");
2201 dev->data->nb_rx_queues = reply_param->rxq_count;
2202 dev->data->nb_tx_queues = reply_param->txq_count;
2204 for (queue = 0; queue < reply_param->rxq_count; queue++)
2205 process_private->rxq_fds[queue] = reply->fds[fd_iterator++];
2206 for (queue = 0; queue < reply_param->txq_count; queue++)
2207 process_private->txq_fds[queue] = reply->fds[fd_iterator++];
2212 /* Send the queue file descriptors from the primary process to secondary. */
2214 tap_mp_sync_queues(const struct rte_mp_msg *request, const void *peer)
2216 struct rte_eth_dev *dev;
2217 struct pmd_process_private *process_private;
2218 struct rte_mp_msg reply;
2219 const struct ipc_queues *request_param =
2220 (const struct ipc_queues *)request->param;
2221 struct ipc_queues *reply_param =
2222 (struct ipc_queues *)reply.param;
2227 /* Get requested port */
2228 TAP_LOG(DEBUG, "Received IPC request for %s", request_param->port_name);
2229 ret = rte_eth_dev_get_port_by_name(request_param->port_name, &port_id);
2231 TAP_LOG(ERR, "Failed to get port id for %s",
2232 request_param->port_name);
2235 dev = &rte_eth_devices[port_id];
2236 process_private = dev->process_private;
2238 /* Fill file descriptors for all queues */
2240 reply_param->rxq_count = 0;
2241 if (dev->data->nb_rx_queues + dev->data->nb_tx_queues >
2243 TAP_LOG(ERR, "Number of rx/tx queues exceeds max number of fds");
2247 for (queue = 0; queue < dev->data->nb_rx_queues; queue++) {
2248 reply.fds[reply.num_fds++] = process_private->rxq_fds[queue];
2249 reply_param->rxq_count++;
2251 RTE_ASSERT(reply_param->rxq_count == dev->data->nb_rx_queues);
2253 reply_param->txq_count = 0;
2254 for (queue = 0; queue < dev->data->nb_tx_queues; queue++) {
2255 reply.fds[reply.num_fds++] = process_private->txq_fds[queue];
2256 reply_param->txq_count++;
2258 RTE_ASSERT(reply_param->txq_count == dev->data->nb_tx_queues);
2261 strlcpy(reply.name, request->name, sizeof(reply.name));
2262 strlcpy(reply_param->port_name, request_param->port_name,
2263 sizeof(reply_param->port_name));
2264 reply.len_param = sizeof(*reply_param);
2265 if (rte_mp_reply(&reply, peer) < 0) {
2266 TAP_LOG(ERR, "Failed to reply an IPC request to sync queues");
2272 /* Open a TAP interface device.
2275 rte_pmd_tap_probe(struct rte_vdev_device *dev)
2277 const char *name, *params;
2279 struct rte_kvargs *kvlist = NULL;
2281 char tap_name[RTE_ETH_NAME_MAX_LEN];
2282 char remote_iface[RTE_ETH_NAME_MAX_LEN];
2283 struct rte_ether_addr user_mac = { .addr_bytes = {0} };
2284 struct rte_eth_dev *eth_dev;
2285 int tap_devices_count_increased = 0;
2287 name = rte_vdev_device_name(dev);
2288 params = rte_vdev_device_args(dev);
2290 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
2291 eth_dev = rte_eth_dev_attach_secondary(name);
2293 TAP_LOG(ERR, "Failed to probe %s", name);
2296 eth_dev->dev_ops = &ops;
2297 eth_dev->device = &dev->device;
2298 eth_dev->rx_pkt_burst = pmd_rx_burst;
2299 eth_dev->tx_pkt_burst = pmd_tx_burst;
2300 if (!rte_eal_primary_proc_alive(NULL)) {
2301 TAP_LOG(ERR, "Primary process is missing");
2304 eth_dev->process_private = (struct pmd_process_private *)
2305 rte_zmalloc_socket(name,
2306 sizeof(struct pmd_process_private),
2307 RTE_CACHE_LINE_SIZE,
2308 eth_dev->device->numa_node);
2309 if (eth_dev->process_private == NULL) {
2311 "Failed to alloc memory for process private");
2315 ret = tap_mp_attach_queues(name, eth_dev);
2318 rte_eth_dev_probing_finish(eth_dev);
2322 speed = ETH_SPEED_NUM_10G;
2324 /* use tap%d which causes kernel to choose next available */
2325 strlcpy(tap_name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2326 memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2328 if (params && (params[0] != '\0')) {
2329 TAP_LOG(DEBUG, "parameters (%s)", params);
2331 kvlist = rte_kvargs_parse(params, valid_arguments);
2333 if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2334 ret = rte_kvargs_process(kvlist,
2336 &set_interface_name,
2342 if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
2343 ret = rte_kvargs_process(kvlist,
2351 if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
2352 ret = rte_kvargs_process(kvlist,
2361 pmd_link.link_speed = speed;
2363 TAP_LOG(DEBUG, "Initializing pmd_tap for %s", name);
2365 /* Register IPC feed callback */
2366 if (!tap_devices_count) {
2367 ret = rte_mp_action_register(TAP_MP_KEY, tap_mp_sync_queues);
2368 if (ret < 0 && rte_errno != ENOTSUP) {
2369 TAP_LOG(ERR, "tap: Failed to register IPC callback: %s",
2370 strerror(rte_errno));
2374 tap_devices_count++;
2375 tap_devices_count_increased = 1;
2376 ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac,
2377 ETH_TUNTAP_TYPE_TAP);
2381 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2383 if (tap_devices_count_increased == 1) {
2384 if (tap_devices_count == 1)
2385 rte_mp_action_unregister(TAP_MP_KEY);
2386 tap_devices_count--;
2389 rte_kvargs_free(kvlist);
2394 /* detach a TUNTAP device.
2397 rte_pmd_tap_remove(struct rte_vdev_device *dev)
2399 struct rte_eth_dev *eth_dev = NULL;
2400 struct pmd_internals *internals;
2401 struct pmd_process_private *process_private;
2404 /* find the ethdev entry */
2405 eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
2409 /* mac_addrs must not be freed alone because part of dev_private */
2410 eth_dev->data->mac_addrs = NULL;
2412 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2413 return rte_eth_dev_release_port(eth_dev);
2415 internals = eth_dev->data->dev_private;
2416 process_private = eth_dev->process_private;
2418 TAP_LOG(DEBUG, "Closing %s Ethernet device on numa %u",
2419 tuntap_types[internals->type], rte_socket_id());
2421 if (internals->nlsk_fd) {
2422 tap_flow_flush(eth_dev, NULL);
2423 tap_flow_implicit_flush(internals, NULL);
2424 tap_nl_final(internals->nlsk_fd);
2426 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
2427 if (process_private->rxq_fds[i] != -1) {
2428 close(process_private->rxq_fds[i]);
2429 process_private->rxq_fds[i] = -1;
2431 if (process_private->txq_fds[i] != -1) {
2432 close(process_private->txq_fds[i]);
2433 process_private->txq_fds[i] = -1;
2437 close(internals->ioctl_sock);
2438 rte_free(eth_dev->process_private);
2439 if (tap_devices_count == 1)
2440 rte_mp_action_unregister(TAP_MP_KEY);
2441 tap_devices_count--;
2442 rte_eth_dev_release_port(eth_dev);
2444 if (internals->ka_fd != -1) {
2445 close(internals->ka_fd);
2446 internals->ka_fd = -1;
2451 static struct rte_vdev_driver pmd_tun_drv = {
2452 .probe = rte_pmd_tun_probe,
2453 .remove = rte_pmd_tap_remove,
2456 static struct rte_vdev_driver pmd_tap_drv = {
2457 .probe = rte_pmd_tap_probe,
2458 .remove = rte_pmd_tap_remove,
2461 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
2462 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
2463 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
2464 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
2465 ETH_TAP_IFACE_ARG "=<string> ");
2466 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
2467 ETH_TAP_IFACE_ARG "=<string> "
2468 ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
2469 ETH_TAP_REMOTE_ARG "=<string>");
2472 RTE_INIT(tap_init_log)
2474 tap_logtype = rte_log_register("pmd.net.tap");
2475 if (tap_logtype >= 0)
2476 rte_log_set_level(tap_logtype, RTE_LOG_NOTICE);