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>
23 #include <sys/types.h>
25 #include <sys/socket.h>
26 #include <sys/ioctl.h>
27 #include <sys/utsname.h>
35 #include <arpa/inet.h>
37 #include <linux/if_tun.h>
38 #include <linux/if_ether.h>
43 #include <rte_eth_tap.h>
45 #include <tap_netlink.h>
46 #include <tap_tcmsgs.h>
48 /* Linux based path to the TUN device */
49 #define TUN_TAP_DEV_PATH "/dev/net/tun"
50 #define DEFAULT_TAP_NAME "dtap"
51 #define DEFAULT_TUN_NAME "dtun"
53 #define ETH_TAP_IFACE_ARG "iface"
54 #define ETH_TAP_REMOTE_ARG "remote"
55 #define ETH_TAP_MAC_ARG "mac"
56 #define ETH_TAP_MAC_FIXED "fixed"
58 #define ETH_TAP_USR_MAC_FMT "xx:xx:xx:xx:xx:xx"
59 #define ETH_TAP_CMP_MAC_FMT "0123456789ABCDEFabcdef"
60 #define ETH_TAP_MAC_ARG_FMT ETH_TAP_MAC_FIXED "|" ETH_TAP_USR_MAC_FMT
62 #define TAP_GSO_MBUFS_PER_CORE 128
63 #define TAP_GSO_MBUF_SEG_SIZE 128
64 #define TAP_GSO_MBUF_CACHE_SIZE 4
65 #define TAP_GSO_MBUFS_NUM \
66 (TAP_GSO_MBUFS_PER_CORE * TAP_GSO_MBUF_CACHE_SIZE)
68 /* IPC key for queue fds sync */
69 #define TAP_MP_KEY "tap_mp_sync_queues"
71 #define TAP_IOV_DEFAULT_MAX 1024
73 static int tap_devices_count;
75 static const char *valid_arguments[] = {
82 static volatile uint32_t tap_trigger; /* Rx trigger */
84 static struct rte_eth_link pmd_link = {
85 .link_speed = ETH_SPEED_NUM_10G,
86 .link_duplex = ETH_LINK_FULL_DUPLEX,
87 .link_status = ETH_LINK_DOWN,
88 .link_autoneg = ETH_LINK_FIXED,
92 tap_trigger_cb(int sig __rte_unused)
94 /* Valid trigger values are nonzero */
95 tap_trigger = (tap_trigger + 1) | 0x80000000;
98 /* Specifies on what netdevices the ioctl should be applied */
105 /* Message header to synchronize queues via IPC */
107 char port_name[RTE_DEV_NAME_MAX_LEN];
111 * The file descriptors are in the dedicated part
112 * of the Unix message to be translated by the kernel.
116 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
119 * Tun/Tap allocation routine
122 * Pointer to private structure.
124 * @param[in] is_keepalive
128 * -1 on failure, fd on success
131 tun_alloc(struct pmd_internals *pmd, int is_keepalive)
134 #ifdef IFF_MULTI_QUEUE
135 unsigned int features;
139 memset(&ifr, 0, sizeof(struct ifreq));
142 * Do not set IFF_NO_PI as packet information header will be needed
143 * to check if a received packet has been truncated.
145 ifr.ifr_flags = (pmd->type == ETH_TUNTAP_TYPE_TAP) ?
146 IFF_TAP : IFF_TUN | IFF_POINTOPOINT;
147 strlcpy(ifr.ifr_name, pmd->name, IFNAMSIZ);
149 fd = open(TUN_TAP_DEV_PATH, O_RDWR);
151 TAP_LOG(ERR, "Unable to open %s interface", TUN_TAP_DEV_PATH);
155 #ifdef IFF_MULTI_QUEUE
156 /* Grab the TUN features to verify we can work multi-queue */
157 if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
158 TAP_LOG(ERR, "unable to get TUN/TAP features");
161 TAP_LOG(DEBUG, "%s Features %08x", TUN_TAP_DEV_PATH, features);
163 if (features & IFF_MULTI_QUEUE) {
164 TAP_LOG(DEBUG, " Multi-queue support for %d queues",
165 RTE_PMD_TAP_MAX_QUEUES);
166 ifr.ifr_flags |= IFF_MULTI_QUEUE;
170 ifr.ifr_flags |= IFF_ONE_QUEUE;
171 TAP_LOG(DEBUG, " Single queue only support");
174 /* Set the TUN/TAP configuration and set the name if needed */
175 if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
176 TAP_LOG(WARNING, "Unable to set TUNSETIFF for %s: %s",
177 ifr.ifr_name, strerror(errno));
182 * Name passed to kernel might be wildcard like dtun%d
183 * and need to find the resulting device.
185 TAP_LOG(DEBUG, "Device name is '%s'", ifr.ifr_name);
186 strlcpy(pmd->name, ifr.ifr_name, RTE_ETH_NAME_MAX_LEN);
190 * Detach the TUN/TAP keep-alive queue
191 * to avoid traffic through it
193 ifr.ifr_flags = IFF_DETACH_QUEUE;
194 if (ioctl(fd, TUNSETQUEUE, (void *)&ifr) < 0) {
196 "Unable to detach keep-alive queue for %s: %s",
197 ifr.ifr_name, strerror(errno));
202 /* Always set the file descriptor to non-blocking */
203 if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
205 "Unable to set %s to nonblocking: %s",
206 ifr.ifr_name, strerror(errno));
210 /* Set up trigger to optimize empty Rx bursts */
214 int flags = fcntl(fd, F_GETFL);
216 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
218 if (sa.sa_handler != tap_trigger_cb) {
220 * Make sure SIGIO is not already taken. This is done
221 * as late as possible to leave the application a
222 * chance to set up its own signal handler first.
224 if (sa.sa_handler != SIG_IGN &&
225 sa.sa_handler != SIG_DFL) {
229 sa = (struct sigaction){
230 .sa_flags = SA_RESTART,
231 .sa_handler = tap_trigger_cb,
233 if (sigaction(SIGIO, &sa, NULL) == -1)
236 /* Enable SIGIO on file descriptor */
237 fcntl(fd, F_SETFL, flags | O_ASYNC);
238 fcntl(fd, F_SETOWN, getpid());
242 /* Disable trigger globally in case of error */
244 TAP_LOG(WARNING, "Rx trigger disabled: %s",
257 tap_verify_csum(struct rte_mbuf *mbuf)
259 uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
260 uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
261 uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
262 unsigned int l2_len = sizeof(struct rte_ether_hdr);
268 if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
270 else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
272 /* Don't verify checksum for packets with discontinuous L2 header */
273 if (unlikely(l2_len + sizeof(struct rte_ipv4_hdr) >
274 rte_pktmbuf_data_len(mbuf)))
276 l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
277 if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
278 struct rte_ipv4_hdr *iph = l3_hdr;
280 /* ihl contains the number of 4-byte words in the header */
281 l3_len = 4 * (iph->version_ihl & 0xf);
282 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
284 /* check that the total length reported by header is not
285 * greater than the total received size
287 if (l2_len + rte_be_to_cpu_16(iph->total_length) >
288 rte_pktmbuf_data_len(mbuf))
291 cksum = ~rte_raw_cksum(iph, l3_len);
292 mbuf->ol_flags |= cksum ?
293 PKT_RX_IP_CKSUM_BAD :
294 PKT_RX_IP_CKSUM_GOOD;
295 } else if (l3 == RTE_PTYPE_L3_IPV6) {
296 struct rte_ipv6_hdr *iph = l3_hdr;
298 l3_len = sizeof(struct rte_ipv6_hdr);
299 /* check that the total length reported by header is not
300 * greater than the total received size
302 if (l2_len + l3_len + rte_be_to_cpu_16(iph->payload_len) >
303 rte_pktmbuf_data_len(mbuf))
306 /* IPv6 extensions are not supported */
309 if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
310 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
311 /* Don't verify checksum for multi-segment packets. */
312 if (mbuf->nb_segs > 1)
314 if (l3 == RTE_PTYPE_L3_IPV4)
315 cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
316 else if (l3 == RTE_PTYPE_L3_IPV6)
317 cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
318 mbuf->ol_flags |= cksum ?
319 PKT_RX_L4_CKSUM_BAD :
320 PKT_RX_L4_CKSUM_GOOD;
325 tap_rx_offload_get_port_capa(void)
328 * No specific port Rx offload capabilities.
334 tap_rx_offload_get_queue_capa(void)
336 return DEV_RX_OFFLOAD_SCATTER |
337 DEV_RX_OFFLOAD_IPV4_CKSUM |
338 DEV_RX_OFFLOAD_UDP_CKSUM |
339 DEV_RX_OFFLOAD_TCP_CKSUM;
342 /* Callback to handle the rx burst of packets to the correct interface and
343 * file descriptor(s) in a multi-queue setup.
346 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
348 struct rx_queue *rxq = queue;
349 struct pmd_process_private *process_private;
351 unsigned long num_rx_bytes = 0;
352 uint32_t trigger = tap_trigger;
354 if (trigger == rxq->trigger_seen)
357 process_private = rte_eth_devices[rxq->in_port].process_private;
358 for (num_rx = 0; num_rx < nb_pkts; ) {
359 struct rte_mbuf *mbuf = rxq->pool;
360 struct rte_mbuf *seg = NULL;
361 struct rte_mbuf *new_tail = NULL;
362 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
365 len = readv(process_private->rxq_fds[rxq->queue_id],
367 1 + (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
368 rxq->nb_rx_desc : 1));
369 if (len < (int)sizeof(struct tun_pi))
372 /* Packet couldn't fit in the provided mbuf */
373 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
374 rxq->stats.ierrors++;
378 len -= sizeof(struct tun_pi);
381 mbuf->port = rxq->in_port;
383 struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
385 if (unlikely(!buf)) {
386 rxq->stats.rx_nombuf++;
387 /* No new buf has been allocated: do nothing */
388 if (!new_tail || !seg)
392 rte_pktmbuf_free(mbuf);
396 seg = seg ? seg->next : mbuf;
397 if (rxq->pool == mbuf)
400 new_tail->next = buf;
402 new_tail->next = seg->next;
404 /* iovecs[0] is reserved for packet info (pi) */
405 (*rxq->iovecs)[mbuf->nb_segs].iov_len =
406 buf->buf_len - data_off;
407 (*rxq->iovecs)[mbuf->nb_segs].iov_base =
408 (char *)buf->buf_addr + data_off;
410 seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
411 seg->data_off = data_off;
413 len -= seg->data_len;
417 /* First segment has headroom, not the others */
421 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
423 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
424 tap_verify_csum(mbuf);
426 /* account for the receive frame */
427 bufs[num_rx++] = mbuf;
428 num_rx_bytes += mbuf->pkt_len;
431 rxq->stats.ipackets += num_rx;
432 rxq->stats.ibytes += num_rx_bytes;
434 if (trigger && num_rx < nb_pkts)
435 rxq->trigger_seen = trigger;
441 tap_tx_offload_get_port_capa(void)
444 * No specific port Tx offload capabilities.
450 tap_tx_offload_get_queue_capa(void)
452 return DEV_TX_OFFLOAD_MULTI_SEGS |
453 DEV_TX_OFFLOAD_IPV4_CKSUM |
454 DEV_TX_OFFLOAD_UDP_CKSUM |
455 DEV_TX_OFFLOAD_TCP_CKSUM |
456 DEV_TX_OFFLOAD_TCP_TSO;
459 /* Finalize l4 checksum calculation */
461 tap_tx_l4_cksum(uint16_t *l4_cksum, uint16_t l4_phdr_cksum,
462 uint32_t l4_raw_cksum)
467 cksum = __rte_raw_cksum_reduce(l4_raw_cksum);
468 cksum += l4_phdr_cksum;
470 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
471 cksum = (~cksum) & 0xffff;
478 /* Accumaulate L4 raw checksums */
480 tap_tx_l4_add_rcksum(char *l4_data, unsigned int l4_len, uint16_t *l4_cksum,
481 uint32_t *l4_raw_cksum)
483 if (l4_cksum == NULL)
486 *l4_raw_cksum = __rte_raw_cksum(l4_data, l4_len, *l4_raw_cksum);
489 /* L3 and L4 pseudo headers checksum offloads */
491 tap_tx_l3_cksum(char *packet, uint64_t ol_flags, unsigned int l2_len,
492 unsigned int l3_len, unsigned int l4_len, uint16_t **l4_cksum,
493 uint16_t *l4_phdr_cksum, uint32_t *l4_raw_cksum)
495 void *l3_hdr = packet + l2_len;
497 if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
498 struct rte_ipv4_hdr *iph = l3_hdr;
501 iph->hdr_checksum = 0;
502 cksum = rte_raw_cksum(iph, l3_len);
503 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
505 if (ol_flags & PKT_TX_L4_MASK) {
508 l4_hdr = packet + l2_len + l3_len;
509 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
510 *l4_cksum = &((struct rte_udp_hdr *)l4_hdr)->dgram_cksum;
511 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
512 *l4_cksum = &((struct rte_tcp_hdr *)l4_hdr)->cksum;
516 if (ol_flags & PKT_TX_IPV4)
517 *l4_phdr_cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
519 *l4_phdr_cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
520 *l4_raw_cksum = __rte_raw_cksum(l4_hdr, l4_len, 0);
525 tap_write_mbufs(struct tx_queue *txq, uint16_t num_mbufs,
526 struct rte_mbuf **pmbufs,
527 uint16_t *num_packets, unsigned long *num_tx_bytes)
531 struct pmd_process_private *process_private;
533 process_private = rte_eth_devices[txq->out_port].process_private;
535 for (i = 0; i < num_mbufs; i++) {
536 struct rte_mbuf *mbuf = pmbufs[i];
537 struct iovec iovecs[mbuf->nb_segs + 2];
538 struct tun_pi pi = { .flags = 0, .proto = 0x00 };
539 struct rte_mbuf *seg = mbuf;
540 char m_copy[mbuf->data_len];
544 int k; /* current index in iovecs for copying segments */
545 uint16_t seg_len; /* length of first segment */
547 uint16_t *l4_cksum; /* l4 checksum (pseudo header + payload) */
548 uint32_t l4_raw_cksum = 0; /* TCP/UDP payload raw checksum */
549 uint16_t l4_phdr_cksum = 0; /* TCP/UDP pseudo header checksum */
550 uint16_t is_cksum = 0; /* in case cksum should be offloaded */
553 if (txq->type == ETH_TUNTAP_TYPE_TUN) {
555 * TUN and TAP are created with IFF_NO_PI disabled.
556 * For TUN PMD this mandatory as fields are used by
557 * Kernel tun.c to determine whether its IP or non IP
560 * The logic fetches the first byte of data from mbuf
561 * then compares whether its v4 or v6. If first byte
562 * is 4 or 6, then protocol field is updated.
564 char *buff_data = rte_pktmbuf_mtod(seg, void *);
565 proto = (*buff_data & 0xf0);
566 pi.proto = (proto == 0x40) ?
567 rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4) :
569 rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6) :
574 iovecs[k].iov_base = π
575 iovecs[k].iov_len = sizeof(pi);
578 nb_segs = mbuf->nb_segs;
580 ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
581 (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
582 (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
585 /* Support only packets with at least layer 4
586 * header included in the first segment
588 seg_len = rte_pktmbuf_data_len(mbuf);
589 l234_hlen = mbuf->l2_len + mbuf->l3_len + mbuf->l4_len;
590 if (seg_len < l234_hlen)
593 /* To change checksums, work on a * copy of l2, l3
594 * headers + l4 pseudo header
596 rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
598 tap_tx_l3_cksum(m_copy, mbuf->ol_flags,
599 mbuf->l2_len, mbuf->l3_len, mbuf->l4_len,
600 &l4_cksum, &l4_phdr_cksum,
602 iovecs[k].iov_base = m_copy;
603 iovecs[k].iov_len = l234_hlen;
606 /* Update next iovecs[] beyond l2, l3, l4 headers */
607 if (seg_len > l234_hlen) {
608 iovecs[k].iov_len = seg_len - l234_hlen;
610 rte_pktmbuf_mtod(seg, char *) +
612 tap_tx_l4_add_rcksum(iovecs[k].iov_base,
613 iovecs[k].iov_len, l4_cksum,
621 for (j = k; j <= nb_segs; j++) {
622 iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
623 iovecs[j].iov_base = rte_pktmbuf_mtod(seg, void *);
625 tap_tx_l4_add_rcksum(iovecs[j].iov_base,
626 iovecs[j].iov_len, l4_cksum,
632 tap_tx_l4_cksum(l4_cksum, l4_phdr_cksum, l4_raw_cksum);
634 /* copy the tx frame data */
635 n = writev(process_private->txq_fds[txq->queue_id], iovecs, j);
639 (*num_tx_bytes) += rte_pktmbuf_pkt_len(mbuf);
643 /* Callback to handle sending packets from the tap interface
646 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
648 struct tx_queue *txq = queue;
650 uint16_t num_packets = 0;
651 unsigned long num_tx_bytes = 0;
655 if (unlikely(nb_pkts == 0))
658 struct rte_mbuf *gso_mbufs[MAX_GSO_MBUFS];
659 max_size = *txq->mtu + (RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN + 4);
660 for (i = 0; i < nb_pkts; i++) {
661 struct rte_mbuf *mbuf_in = bufs[num_tx];
662 struct rte_mbuf **mbuf;
663 uint16_t num_mbufs = 0;
664 uint16_t tso_segsz = 0;
670 tso = mbuf_in->ol_flags & PKT_TX_TCP_SEG;
672 struct rte_gso_ctx *gso_ctx = &txq->gso_ctx;
674 assert(gso_ctx != NULL);
676 /* TCP segmentation implies TCP checksum offload */
677 mbuf_in->ol_flags |= PKT_TX_TCP_CKSUM;
679 /* gso size is calculated without RTE_ETHER_CRC_LEN */
680 hdrs_len = mbuf_in->l2_len + mbuf_in->l3_len +
682 tso_segsz = mbuf_in->tso_segsz + hdrs_len;
683 if (unlikely(tso_segsz == hdrs_len) ||
684 tso_segsz > *txq->mtu) {
688 gso_ctx->gso_size = tso_segsz;
689 ret = rte_gso_segment(mbuf_in, /* packet to segment */
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 */
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 tap_write_mbufs(txq, num_mbufs, mbuf,
712 &num_packets, &num_tx_bytes);
714 /* free original mbuf */
715 rte_pktmbuf_free(mbuf_in);
717 for (j = 0; j < ret; j++)
718 rte_pktmbuf_free(mbuf[j]);
721 txq->stats.opackets += num_packets;
722 txq->stats.errs += nb_pkts - num_tx;
723 txq->stats.obytes += num_tx_bytes;
729 tap_ioctl_req2str(unsigned long request)
733 return "SIOCSIFFLAGS";
735 return "SIOCGIFFLAGS";
737 return "SIOCGIFHWADDR";
739 return "SIOCSIFHWADDR";
747 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
748 struct ifreq *ifr, int set, enum ioctl_mode mode)
750 short req_flags = ifr->ifr_flags;
751 int remote = pmd->remote_if_index &&
752 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
754 if (!pmd->remote_if_index && mode == REMOTE_ONLY)
757 * If there is a remote netdevice, apply ioctl on it, then apply it on
762 strlcpy(ifr->ifr_name, pmd->remote_iface, IFNAMSIZ);
763 else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
764 strlcpy(ifr->ifr_name, pmd->name, IFNAMSIZ);
767 /* fetch current flags to leave other flags untouched */
768 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
771 ifr->ifr_flags |= req_flags;
773 ifr->ifr_flags &= ~req_flags;
781 RTE_LOG(WARNING, PMD, "%s: ioctl() called with wrong arg\n",
785 if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
787 if (remote-- && mode == LOCAL_AND_REMOTE)
792 TAP_LOG(DEBUG, "%s(%s) failed: %s(%d)", ifr->ifr_name,
793 tap_ioctl_req2str(request), strerror(errno), errno);
798 tap_link_set_down(struct rte_eth_dev *dev)
800 struct pmd_internals *pmd = dev->data->dev_private;
801 struct ifreq ifr = { .ifr_flags = IFF_UP };
803 dev->data->dev_link.link_status = ETH_LINK_DOWN;
804 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
808 tap_link_set_up(struct rte_eth_dev *dev)
810 struct pmd_internals *pmd = dev->data->dev_private;
811 struct ifreq ifr = { .ifr_flags = IFF_UP };
813 dev->data->dev_link.link_status = ETH_LINK_UP;
814 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
818 tap_dev_start(struct rte_eth_dev *dev)
822 err = tap_intr_handle_set(dev, 1);
826 err = tap_link_set_up(dev);
830 for (i = 0; i < dev->data->nb_tx_queues; i++)
831 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
832 for (i = 0; i < dev->data->nb_rx_queues; i++)
833 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
838 /* This function gets called when the current port gets stopped.
841 tap_dev_stop(struct rte_eth_dev *dev)
845 for (i = 0; i < dev->data->nb_tx_queues; i++)
846 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
847 for (i = 0; i < dev->data->nb_rx_queues; i++)
848 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
850 tap_intr_handle_set(dev, 0);
851 tap_link_set_down(dev);
855 tap_dev_configure(struct rte_eth_dev *dev)
857 struct pmd_internals *pmd = dev->data->dev_private;
859 if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
861 "%s: number of rx queues %d exceeds max num of queues %d",
863 dev->data->nb_rx_queues,
864 RTE_PMD_TAP_MAX_QUEUES);
867 if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
869 "%s: number of tx queues %d exceeds max num of queues %d",
871 dev->data->nb_tx_queues,
872 RTE_PMD_TAP_MAX_QUEUES);
876 TAP_LOG(INFO, "%s: %s: TX configured queues number: %u",
877 dev->device->name, pmd->name, dev->data->nb_tx_queues);
879 TAP_LOG(INFO, "%s: %s: RX configured queues number: %u",
880 dev->device->name, pmd->name, dev->data->nb_rx_queues);
886 tap_dev_speed_capa(void)
888 uint32_t speed = pmd_link.link_speed;
891 if (speed >= ETH_SPEED_NUM_10M)
892 capa |= ETH_LINK_SPEED_10M;
893 if (speed >= ETH_SPEED_NUM_100M)
894 capa |= ETH_LINK_SPEED_100M;
895 if (speed >= ETH_SPEED_NUM_1G)
896 capa |= ETH_LINK_SPEED_1G;
897 if (speed >= ETH_SPEED_NUM_5G)
898 capa |= ETH_LINK_SPEED_2_5G;
899 if (speed >= ETH_SPEED_NUM_5G)
900 capa |= ETH_LINK_SPEED_5G;
901 if (speed >= ETH_SPEED_NUM_10G)
902 capa |= ETH_LINK_SPEED_10G;
903 if (speed >= ETH_SPEED_NUM_20G)
904 capa |= ETH_LINK_SPEED_20G;
905 if (speed >= ETH_SPEED_NUM_25G)
906 capa |= ETH_LINK_SPEED_25G;
907 if (speed >= ETH_SPEED_NUM_40G)
908 capa |= ETH_LINK_SPEED_40G;
909 if (speed >= ETH_SPEED_NUM_50G)
910 capa |= ETH_LINK_SPEED_50G;
911 if (speed >= ETH_SPEED_NUM_56G)
912 capa |= ETH_LINK_SPEED_56G;
913 if (speed >= ETH_SPEED_NUM_100G)
914 capa |= ETH_LINK_SPEED_100G;
920 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
922 struct pmd_internals *internals = dev->data->dev_private;
924 dev_info->if_index = internals->if_index;
925 dev_info->max_mac_addrs = 1;
926 dev_info->max_rx_pktlen = (uint32_t)RTE_ETHER_MAX_VLAN_FRAME_LEN;
927 dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
928 dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
929 dev_info->min_rx_bufsize = 0;
930 dev_info->speed_capa = tap_dev_speed_capa();
931 dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
932 dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
933 dev_info->rx_queue_offload_capa;
934 dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
935 dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
936 dev_info->tx_queue_offload_capa;
937 dev_info->hash_key_size = TAP_RSS_HASH_KEY_SIZE;
939 * limitation: TAP supports all of IP, UDP and TCP hash
940 * functions together and not in partial combinations
942 dev_info->flow_type_rss_offloads = ~TAP_RSS_HF_MASK;
948 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
950 unsigned int i, imax;
951 unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
952 unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
953 unsigned long rx_nombuf = 0, ierrors = 0;
954 const struct pmd_internals *pmd = dev->data->dev_private;
956 /* rx queue statistics */
957 imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
958 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
959 for (i = 0; i < imax; i++) {
960 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
961 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
962 rx_total += tap_stats->q_ipackets[i];
963 rx_bytes_total += tap_stats->q_ibytes[i];
964 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
965 ierrors += pmd->rxq[i].stats.ierrors;
968 /* tx queue statistics */
969 imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
970 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
972 for (i = 0; i < imax; i++) {
973 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
974 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
975 tx_total += tap_stats->q_opackets[i];
976 tx_err_total += pmd->txq[i].stats.errs;
977 tx_bytes_total += tap_stats->q_obytes[i];
980 tap_stats->ipackets = rx_total;
981 tap_stats->ibytes = rx_bytes_total;
982 tap_stats->ierrors = ierrors;
983 tap_stats->rx_nombuf = rx_nombuf;
984 tap_stats->opackets = tx_total;
985 tap_stats->oerrors = tx_err_total;
986 tap_stats->obytes = tx_bytes_total;
991 tap_stats_reset(struct rte_eth_dev *dev)
994 struct pmd_internals *pmd = dev->data->dev_private;
996 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
997 pmd->rxq[i].stats.ipackets = 0;
998 pmd->rxq[i].stats.ibytes = 0;
999 pmd->rxq[i].stats.ierrors = 0;
1000 pmd->rxq[i].stats.rx_nombuf = 0;
1002 pmd->txq[i].stats.opackets = 0;
1003 pmd->txq[i].stats.errs = 0;
1004 pmd->txq[i].stats.obytes = 0;
1011 tap_dev_close(struct rte_eth_dev *dev)
1014 struct pmd_internals *internals = dev->data->dev_private;
1015 struct pmd_process_private *process_private = dev->process_private;
1017 tap_link_set_down(dev);
1018 tap_flow_flush(dev, NULL);
1019 tap_flow_implicit_flush(internals, NULL);
1021 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1022 if (process_private->rxq_fds[i] != -1) {
1023 close(process_private->rxq_fds[i]);
1024 process_private->rxq_fds[i] = -1;
1026 if (process_private->txq_fds[i] != -1) {
1027 close(process_private->txq_fds[i]);
1028 process_private->txq_fds[i] = -1;
1032 if (internals->remote_if_index) {
1033 /* Restore initial remote state */
1034 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
1035 &internals->remote_initial_flags);
1038 if (internals->ka_fd != -1) {
1039 close(internals->ka_fd);
1040 internals->ka_fd = -1;
1043 * Since TUN device has no more opened file descriptors
1044 * it will be removed from kernel
1049 tap_rx_queue_release(void *queue)
1051 struct rx_queue *rxq = queue;
1052 struct pmd_process_private *process_private;
1056 process_private = rte_eth_devices[rxq->in_port].process_private;
1057 if (process_private->rxq_fds[rxq->queue_id] > 0) {
1058 close(process_private->rxq_fds[rxq->queue_id]);
1059 process_private->rxq_fds[rxq->queue_id] = -1;
1060 rte_pktmbuf_free(rxq->pool);
1061 rte_free(rxq->iovecs);
1068 tap_tx_queue_release(void *queue)
1070 struct tx_queue *txq = queue;
1071 struct pmd_process_private *process_private;
1075 process_private = rte_eth_devices[txq->out_port].process_private;
1077 if (process_private->txq_fds[txq->queue_id] > 0) {
1078 close(process_private->txq_fds[txq->queue_id]);
1079 process_private->txq_fds[txq->queue_id] = -1;
1084 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
1086 struct rte_eth_link *dev_link = &dev->data->dev_link;
1087 struct pmd_internals *pmd = dev->data->dev_private;
1088 struct ifreq ifr = { .ifr_flags = 0 };
1090 if (pmd->remote_if_index) {
1091 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
1092 if (!(ifr.ifr_flags & IFF_UP) ||
1093 !(ifr.ifr_flags & IFF_RUNNING)) {
1094 dev_link->link_status = ETH_LINK_DOWN;
1098 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
1099 dev_link->link_status =
1100 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
1107 tap_promisc_enable(struct rte_eth_dev *dev)
1109 struct pmd_internals *pmd = dev->data->dev_private;
1110 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1113 ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1117 if (pmd->remote_if_index && !pmd->flow_isolate) {
1118 dev->data->promiscuous = 1;
1119 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
1121 /* Rollback promisc flag */
1122 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1124 * rte_eth_dev_promiscuous_enable() rollback
1125 * dev->data->promiscuous in the case of failure.
1135 tap_promisc_disable(struct rte_eth_dev *dev)
1137 struct pmd_internals *pmd = dev->data->dev_private;
1138 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1141 ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1145 if (pmd->remote_if_index && !pmd->flow_isolate) {
1146 dev->data->promiscuous = 0;
1147 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
1149 /* Rollback promisc flag */
1150 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1152 * rte_eth_dev_promiscuous_disable() rollback
1153 * dev->data->promiscuous in the case of failure.
1163 tap_allmulti_enable(struct rte_eth_dev *dev)
1165 struct pmd_internals *pmd = dev->data->dev_private;
1166 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1169 ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1173 if (pmd->remote_if_index && !pmd->flow_isolate) {
1174 dev->data->all_multicast = 1;
1175 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
1177 /* Rollback allmulti flag */
1178 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1180 * rte_eth_dev_allmulticast_enable() rollback
1181 * dev->data->all_multicast in the case of failure.
1191 tap_allmulti_disable(struct rte_eth_dev *dev)
1193 struct pmd_internals *pmd = dev->data->dev_private;
1194 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1197 ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1201 if (pmd->remote_if_index && !pmd->flow_isolate) {
1202 dev->data->all_multicast = 0;
1203 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
1205 /* Rollback allmulti flag */
1206 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1208 * rte_eth_dev_allmulticast_disable() rollback
1209 * dev->data->all_multicast in the case of failure.
1219 tap_mac_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1221 struct pmd_internals *pmd = dev->data->dev_private;
1222 enum ioctl_mode mode = LOCAL_ONLY;
1226 if (pmd->type == ETH_TUNTAP_TYPE_TUN) {
1227 TAP_LOG(ERR, "%s: can't MAC address for TUN",
1232 if (rte_is_zero_ether_addr(mac_addr)) {
1233 TAP_LOG(ERR, "%s: can't set an empty MAC address",
1237 /* Check the actual current MAC address on the tap netdevice */
1238 ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY);
1241 if (rte_is_same_ether_addr(
1242 (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1245 /* Check the current MAC address on the remote */
1246 ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY);
1249 if (!rte_is_same_ether_addr(
1250 (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1252 mode = LOCAL_AND_REMOTE;
1253 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1254 rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, RTE_ETHER_ADDR_LEN);
1255 ret = tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode);
1258 rte_memcpy(&pmd->eth_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1259 if (pmd->remote_if_index && !pmd->flow_isolate) {
1260 /* Replace MAC redirection rule after a MAC change */
1261 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC);
1264 "%s: Couldn't delete MAC redirection rule",
1268 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC);
1271 "%s: Couldn't add MAC redirection rule",
1281 tap_gso_ctx_setup(struct rte_gso_ctx *gso_ctx, struct rte_eth_dev *dev)
1287 * Create private mbuf pool with TAP_GSO_MBUF_SEG_SIZE bytes
1288 * size per mbuf use this pool for both direct and indirect mbufs
1291 struct rte_mempool *mp; /* Mempool for GSO packets */
1293 /* initialize GSO context */
1294 gso_types = DEV_TX_OFFLOAD_TCP_TSO;
1295 snprintf(pool_name, sizeof(pool_name), "mp_%s", dev->device->name);
1296 mp = rte_mempool_lookup((const char *)pool_name);
1298 mp = rte_pktmbuf_pool_create(pool_name, TAP_GSO_MBUFS_NUM,
1299 TAP_GSO_MBUF_CACHE_SIZE, 0,
1300 RTE_PKTMBUF_HEADROOM + TAP_GSO_MBUF_SEG_SIZE,
1303 struct pmd_internals *pmd = dev->data->dev_private;
1304 RTE_LOG(DEBUG, PMD, "%s: failed to create mbuf pool for device %s\n",
1305 pmd->name, dev->device->name);
1310 gso_ctx->direct_pool = mp;
1311 gso_ctx->indirect_pool = mp;
1312 gso_ctx->gso_types = gso_types;
1313 gso_ctx->gso_size = 0; /* gso_size is set in tx_burst() per packet */
1320 tap_setup_queue(struct rte_eth_dev *dev,
1321 struct pmd_internals *internals,
1329 struct pmd_internals *pmd = dev->data->dev_private;
1330 struct pmd_process_private *process_private = dev->process_private;
1331 struct rx_queue *rx = &internals->rxq[qid];
1332 struct tx_queue *tx = &internals->txq[qid];
1333 struct rte_gso_ctx *gso_ctx;
1336 fd = &process_private->rxq_fds[qid];
1337 other_fd = &process_private->txq_fds[qid];
1341 fd = &process_private->txq_fds[qid];
1342 other_fd = &process_private->rxq_fds[qid];
1344 gso_ctx = &tx->gso_ctx;
1347 /* fd for this queue already exists */
1348 TAP_LOG(DEBUG, "%s: fd %d for %s queue qid %d exists",
1349 pmd->name, *fd, dir, qid);
1351 } else if (*other_fd != -1) {
1352 /* Only other_fd exists. dup it */
1353 *fd = dup(*other_fd);
1356 TAP_LOG(ERR, "%s: dup() failed.", pmd->name);
1359 TAP_LOG(DEBUG, "%s: dup fd %d for %s queue qid %d (%d)",
1360 pmd->name, *other_fd, dir, qid, *fd);
1362 /* Both RX and TX fds do not exist (equal -1). Create fd */
1363 *fd = tun_alloc(pmd, 0);
1365 *fd = -1; /* restore original value */
1366 TAP_LOG(ERR, "%s: tun_alloc() failed.", pmd->name);
1369 TAP_LOG(DEBUG, "%s: add %s queue for qid %d fd %d",
1370 pmd->name, dir, qid, *fd);
1373 tx->mtu = &dev->data->mtu;
1374 rx->rxmode = &dev->data->dev_conf.rxmode;
1376 ret = tap_gso_ctx_setup(gso_ctx, dev);
1381 tx->type = pmd->type;
1387 tap_rx_queue_setup(struct rte_eth_dev *dev,
1388 uint16_t rx_queue_id,
1389 uint16_t nb_rx_desc,
1390 unsigned int socket_id,
1391 const struct rte_eth_rxconf *rx_conf __rte_unused,
1392 struct rte_mempool *mp)
1394 struct pmd_internals *internals = dev->data->dev_private;
1395 struct pmd_process_private *process_private = dev->process_private;
1396 struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1397 struct rte_mbuf **tmp = &rxq->pool;
1398 long iov_max = sysconf(_SC_IOV_MAX);
1402 "_SC_IOV_MAX is not defined. Using %d as default",
1403 TAP_IOV_DEFAULT_MAX);
1404 iov_max = TAP_IOV_DEFAULT_MAX;
1406 uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1407 struct iovec (*iovecs)[nb_desc + 1];
1408 int data_off = RTE_PKTMBUF_HEADROOM;
1413 if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1415 "nb_rx_queues %d too small or mempool NULL",
1416 dev->data->nb_rx_queues);
1421 rxq->trigger_seen = 1; /* force initial burst */
1422 rxq->in_port = dev->data->port_id;
1423 rxq->queue_id = rx_queue_id;
1424 rxq->nb_rx_desc = nb_desc;
1425 iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1429 "%s: Couldn't allocate %d RX descriptors",
1430 dev->device->name, nb_desc);
1433 rxq->iovecs = iovecs;
1435 dev->data->rx_queues[rx_queue_id] = rxq;
1436 fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1442 (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1443 (*rxq->iovecs)[0].iov_base = &rxq->pi;
1445 for (i = 1; i <= nb_desc; i++) {
1446 *tmp = rte_pktmbuf_alloc(rxq->mp);
1449 "%s: couldn't allocate memory for queue %d",
1450 dev->device->name, rx_queue_id);
1454 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1455 (*rxq->iovecs)[i].iov_base =
1456 (char *)(*tmp)->buf_addr + data_off;
1458 tmp = &(*tmp)->next;
1461 TAP_LOG(DEBUG, " RX TUNTAP device name %s, qid %d on fd %d",
1462 internals->name, rx_queue_id,
1463 process_private->rxq_fds[rx_queue_id]);
1468 rte_pktmbuf_free(rxq->pool);
1470 rte_free(rxq->iovecs);
1476 tap_tx_queue_setup(struct rte_eth_dev *dev,
1477 uint16_t tx_queue_id,
1478 uint16_t nb_tx_desc __rte_unused,
1479 unsigned int socket_id __rte_unused,
1480 const struct rte_eth_txconf *tx_conf)
1482 struct pmd_internals *internals = dev->data->dev_private;
1483 struct pmd_process_private *process_private = dev->process_private;
1484 struct tx_queue *txq;
1488 if (tx_queue_id >= dev->data->nb_tx_queues)
1490 dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1491 txq = dev->data->tx_queues[tx_queue_id];
1492 txq->out_port = dev->data->port_id;
1493 txq->queue_id = tx_queue_id;
1495 offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1496 txq->csum = !!(offloads &
1497 (DEV_TX_OFFLOAD_IPV4_CKSUM |
1498 DEV_TX_OFFLOAD_UDP_CKSUM |
1499 DEV_TX_OFFLOAD_TCP_CKSUM));
1501 ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1505 " TX TUNTAP device name %s, qid %d on fd %d csum %s",
1506 internals->name, tx_queue_id,
1507 process_private->txq_fds[tx_queue_id],
1508 txq->csum ? "on" : "off");
1514 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1516 struct pmd_internals *pmd = dev->data->dev_private;
1517 struct ifreq ifr = { .ifr_mtu = mtu };
1520 err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1522 dev->data->mtu = mtu;
1528 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1529 struct rte_ether_addr *mc_addr_set __rte_unused,
1530 uint32_t nb_mc_addr __rte_unused)
1533 * Nothing to do actually: the tap has no filtering whatsoever, every
1534 * packet is received.
1540 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1542 struct rte_eth_dev *dev = arg;
1543 struct pmd_internals *pmd = dev->data->dev_private;
1544 struct ifinfomsg *info = NLMSG_DATA(nh);
1546 if (nh->nlmsg_type != RTM_NEWLINK ||
1547 (info->ifi_index != pmd->if_index &&
1548 info->ifi_index != pmd->remote_if_index))
1550 return tap_link_update(dev, 0);
1554 tap_dev_intr_handler(void *cb_arg)
1556 struct rte_eth_dev *dev = cb_arg;
1557 struct pmd_internals *pmd = dev->data->dev_private;
1559 tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1563 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1565 struct pmd_internals *pmd = dev->data->dev_private;
1567 /* In any case, disable interrupt if the conf is no longer there. */
1568 if (!dev->data->dev_conf.intr_conf.lsc) {
1569 if (pmd->intr_handle.fd != -1) {
1570 tap_nl_final(pmd->intr_handle.fd);
1571 rte_intr_callback_unregister(&pmd->intr_handle,
1572 tap_dev_intr_handler, dev);
1577 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1578 if (unlikely(pmd->intr_handle.fd == -1))
1580 return rte_intr_callback_register(
1581 &pmd->intr_handle, tap_dev_intr_handler, dev);
1583 tap_nl_final(pmd->intr_handle.fd);
1584 return rte_intr_callback_unregister(&pmd->intr_handle,
1585 tap_dev_intr_handler, dev);
1589 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1593 err = tap_lsc_intr_handle_set(dev, set);
1596 err = tap_rx_intr_vec_set(dev, set);
1598 tap_lsc_intr_handle_set(dev, 0);
1602 static const uint32_t*
1603 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1605 static const uint32_t ptypes[] = {
1606 RTE_PTYPE_INNER_L2_ETHER,
1607 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1608 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1609 RTE_PTYPE_INNER_L3_IPV4,
1610 RTE_PTYPE_INNER_L3_IPV4_EXT,
1611 RTE_PTYPE_INNER_L3_IPV6,
1612 RTE_PTYPE_INNER_L3_IPV6_EXT,
1613 RTE_PTYPE_INNER_L4_FRAG,
1614 RTE_PTYPE_INNER_L4_UDP,
1615 RTE_PTYPE_INNER_L4_TCP,
1616 RTE_PTYPE_INNER_L4_SCTP,
1618 RTE_PTYPE_L2_ETHER_VLAN,
1619 RTE_PTYPE_L2_ETHER_QINQ,
1621 RTE_PTYPE_L3_IPV4_EXT,
1622 RTE_PTYPE_L3_IPV6_EXT,
1634 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1635 struct rte_eth_fc_conf *fc_conf)
1637 fc_conf->mode = RTE_FC_NONE;
1642 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1643 struct rte_eth_fc_conf *fc_conf)
1645 if (fc_conf->mode != RTE_FC_NONE)
1651 * DPDK callback to update the RSS hash configuration.
1654 * Pointer to Ethernet device structure.
1655 * @param[in] rss_conf
1656 * RSS configuration data.
1659 * 0 on success, a negative errno value otherwise and rte_errno is set.
1662 tap_rss_hash_update(struct rte_eth_dev *dev,
1663 struct rte_eth_rss_conf *rss_conf)
1665 if (rss_conf->rss_hf & TAP_RSS_HF_MASK) {
1669 if (rss_conf->rss_key && rss_conf->rss_key_len) {
1671 * Currently TAP RSS key is hard coded
1672 * and cannot be updated
1675 "port %u RSS key cannot be updated",
1676 dev->data->port_id);
1684 tap_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1686 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1692 tap_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1694 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1700 tap_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1702 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1708 tap_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1710 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1714 static const struct eth_dev_ops ops = {
1715 .dev_start = tap_dev_start,
1716 .dev_stop = tap_dev_stop,
1717 .dev_close = tap_dev_close,
1718 .dev_configure = tap_dev_configure,
1719 .dev_infos_get = tap_dev_info,
1720 .rx_queue_setup = tap_rx_queue_setup,
1721 .tx_queue_setup = tap_tx_queue_setup,
1722 .rx_queue_start = tap_rx_queue_start,
1723 .tx_queue_start = tap_tx_queue_start,
1724 .rx_queue_stop = tap_rx_queue_stop,
1725 .tx_queue_stop = tap_tx_queue_stop,
1726 .rx_queue_release = tap_rx_queue_release,
1727 .tx_queue_release = tap_tx_queue_release,
1728 .flow_ctrl_get = tap_flow_ctrl_get,
1729 .flow_ctrl_set = tap_flow_ctrl_set,
1730 .link_update = tap_link_update,
1731 .dev_set_link_up = tap_link_set_up,
1732 .dev_set_link_down = tap_link_set_down,
1733 .promiscuous_enable = tap_promisc_enable,
1734 .promiscuous_disable = tap_promisc_disable,
1735 .allmulticast_enable = tap_allmulti_enable,
1736 .allmulticast_disable = tap_allmulti_disable,
1737 .mac_addr_set = tap_mac_set,
1738 .mtu_set = tap_mtu_set,
1739 .set_mc_addr_list = tap_set_mc_addr_list,
1740 .stats_get = tap_stats_get,
1741 .stats_reset = tap_stats_reset,
1742 .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1743 .rss_hash_update = tap_rss_hash_update,
1744 .filter_ctrl = tap_dev_filter_ctrl,
1747 static const char *tuntap_types[ETH_TUNTAP_TYPE_MAX] = {
1748 "UNKNOWN", "TUN", "TAP"
1752 eth_dev_tap_create(struct rte_vdev_device *vdev, const char *tap_name,
1753 char *remote_iface, struct rte_ether_addr *mac_addr,
1754 enum rte_tuntap_type type)
1756 int numa_node = rte_socket_id();
1757 struct rte_eth_dev *dev;
1758 struct pmd_internals *pmd;
1759 struct pmd_process_private *process_private;
1760 const char *tuntap_name = tuntap_types[type];
1761 struct rte_eth_dev_data *data;
1765 TAP_LOG(DEBUG, "%s device on numa %u", tuntap_name, rte_socket_id());
1767 dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1769 TAP_LOG(ERR, "%s Unable to allocate device struct",
1771 goto error_exit_nodev;
1774 process_private = (struct pmd_process_private *)
1775 rte_zmalloc_socket(tap_name, sizeof(struct pmd_process_private),
1776 RTE_CACHE_LINE_SIZE, dev->device->numa_node);
1778 if (process_private == NULL) {
1779 TAP_LOG(ERR, "Failed to alloc memory for process private");
1782 pmd = dev->data->dev_private;
1783 dev->process_private = process_private;
1785 strlcpy(pmd->name, tap_name, sizeof(pmd->name));
1788 pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1789 if (pmd->ioctl_sock == -1) {
1791 "%s Unable to get a socket for management: %s",
1792 tuntap_name, strerror(errno));
1796 /* Setup some default values */
1798 data->dev_private = pmd;
1799 data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1800 data->numa_node = numa_node;
1802 data->dev_link = pmd_link;
1803 data->mac_addrs = &pmd->eth_addr;
1804 /* Set the number of RX and TX queues */
1805 data->nb_rx_queues = 0;
1806 data->nb_tx_queues = 0;
1808 dev->dev_ops = &ops;
1809 dev->rx_pkt_burst = pmd_rx_burst;
1810 dev->tx_pkt_burst = pmd_tx_burst;
1812 pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1813 pmd->intr_handle.fd = -1;
1814 dev->intr_handle = &pmd->intr_handle;
1816 /* Presetup the fds to -1 as being not valid */
1818 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1819 process_private->rxq_fds[i] = -1;
1820 process_private->txq_fds[i] = -1;
1823 if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1824 if (rte_is_zero_ether_addr(mac_addr))
1825 rte_eth_random_addr((uint8_t *)&pmd->eth_addr);
1827 rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1831 * Allocate a TUN device keep-alive file descriptor that will only be
1832 * closed when the TUN device itself is closed or removed.
1833 * This keep-alive file descriptor will guarantee that the TUN device
1834 * exists even when all of its queues are closed
1836 pmd->ka_fd = tun_alloc(pmd, 1);
1837 if (pmd->ka_fd == -1) {
1838 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
1841 TAP_LOG(DEBUG, "allocated %s", pmd->name);
1843 ifr.ifr_mtu = dev->data->mtu;
1844 if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1847 if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1848 memset(&ifr, 0, sizeof(struct ifreq));
1849 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1850 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
1851 RTE_ETHER_ADDR_LEN);
1852 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1857 * Set up everything related to rte_flow:
1859 * - tap / remote if_index
1860 * - mandatory QDISCs
1861 * - rte_flow actual/implicit lists
1864 pmd->nlsk_fd = tap_nl_init(0);
1865 if (pmd->nlsk_fd == -1) {
1866 TAP_LOG(WARNING, "%s: failed to create netlink socket.",
1868 goto disable_rte_flow;
1870 pmd->if_index = if_nametoindex(pmd->name);
1871 if (!pmd->if_index) {
1872 TAP_LOG(ERR, "%s: failed to get if_index.", pmd->name);
1873 goto disable_rte_flow;
1875 if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1876 TAP_LOG(ERR, "%s: failed to create multiq qdisc.",
1878 goto disable_rte_flow;
1880 if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1881 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1883 goto disable_rte_flow;
1885 LIST_INIT(&pmd->flows);
1887 if (strlen(remote_iface)) {
1888 pmd->remote_if_index = if_nametoindex(remote_iface);
1889 if (!pmd->remote_if_index) {
1890 TAP_LOG(ERR, "%s: failed to get %s if_index.",
1891 pmd->name, remote_iface);
1894 strlcpy(pmd->remote_iface, remote_iface, RTE_ETH_NAME_MAX_LEN);
1896 /* Save state of remote device */
1897 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1899 /* Replicate remote MAC address */
1900 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1901 TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1902 pmd->name, pmd->remote_iface);
1905 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1906 RTE_ETHER_ADDR_LEN);
1907 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1908 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1909 TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1910 pmd->name, remote_iface);
1915 * Flush usually returns negative value because it tries to
1916 * delete every QDISC (and on a running device, one QDISC at
1917 * least is needed). Ignore negative return value.
1919 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1920 if (qdisc_create_ingress(pmd->nlsk_fd,
1921 pmd->remote_if_index) < 0) {
1922 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1926 LIST_INIT(&pmd->implicit_flows);
1927 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1928 tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1929 tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1930 tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1932 "%s: failed to create implicit rules.",
1938 rte_eth_dev_probing_finish(dev);
1942 TAP_LOG(ERR, " Disabling rte flow support: %s(%d)",
1943 strerror(errno), errno);
1944 if (strlen(remote_iface)) {
1945 TAP_LOG(ERR, "Remote feature requires flow support.");
1948 rte_eth_dev_probing_finish(dev);
1952 TAP_LOG(ERR, " Can't set up remote feature: %s(%d)",
1953 strerror(errno), errno);
1954 tap_flow_implicit_flush(pmd, NULL);
1957 if (pmd->ioctl_sock > 0)
1958 close(pmd->ioctl_sock);
1959 /* mac_addrs must not be freed alone because part of dev_private */
1960 dev->data->mac_addrs = NULL;
1961 rte_eth_dev_release_port(dev);
1964 TAP_LOG(ERR, "%s Unable to initialize %s",
1965 tuntap_name, rte_vdev_device_name(vdev));
1970 /* make sure name is a possible Linux network device name */
1972 is_valid_iface(const char *name)
1977 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1981 if (*name == '/' || *name == ':' || isspace(*name))
1989 set_interface_name(const char *key __rte_unused,
1993 char *name = (char *)extra_args;
1996 if (!is_valid_iface(value)) {
1997 TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
2001 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
2003 /* use tap%d which causes kernel to choose next available */
2004 strlcpy(name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2010 set_remote_iface(const char *key __rte_unused,
2014 char *name = (char *)extra_args;
2017 if (!is_valid_iface(value)) {
2018 TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
2022 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
2028 static int parse_user_mac(struct rte_ether_addr *user_mac,
2031 unsigned int index = 0;
2032 char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
2034 if (user_mac == NULL || value == NULL)
2037 strlcpy(mac_temp, value, sizeof(mac_temp));
2038 mac_byte = strtok(mac_temp, ":");
2040 while ((mac_byte != NULL) &&
2041 (strlen(mac_byte) <= 2) &&
2042 (strlen(mac_byte) == strspn(mac_byte,
2043 ETH_TAP_CMP_MAC_FMT))) {
2044 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
2045 mac_byte = strtok(NULL, ":");
2052 set_mac_type(const char *key __rte_unused,
2056 struct rte_ether_addr *user_mac = extra_args;
2061 if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
2062 static int iface_idx;
2064 /* fixed mac = 00:64:74:61:70:<iface_idx> */
2065 memcpy((char *)user_mac->addr_bytes, "\0dtap",
2066 RTE_ETHER_ADDR_LEN);
2067 user_mac->addr_bytes[RTE_ETHER_ADDR_LEN - 1] =
2072 if (parse_user_mac(user_mac, value) != 6)
2075 TAP_LOG(DEBUG, "TAP user MAC param (%s)", value);
2079 TAP_LOG(ERR, "TAP user MAC (%s) is not in format (%s|%s)",
2080 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
2085 * Open a TUN interface device. TUN PMD
2086 * 1) sets tap_type as false
2087 * 2) intakes iface as argument.
2088 * 3) as interface is virtual set speed to 10G
2091 rte_pmd_tun_probe(struct rte_vdev_device *dev)
2093 const char *name, *params;
2095 struct rte_kvargs *kvlist = NULL;
2096 char tun_name[RTE_ETH_NAME_MAX_LEN];
2097 char remote_iface[RTE_ETH_NAME_MAX_LEN];
2098 struct rte_eth_dev *eth_dev;
2100 name = rte_vdev_device_name(dev);
2101 params = rte_vdev_device_args(dev);
2102 memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2104 if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
2105 strlen(params) == 0) {
2106 eth_dev = rte_eth_dev_attach_secondary(name);
2108 TAP_LOG(ERR, "Failed to probe %s", name);
2111 eth_dev->dev_ops = &ops;
2112 eth_dev->device = &dev->device;
2113 rte_eth_dev_probing_finish(eth_dev);
2117 /* use tun%d which causes kernel to choose next available */
2118 strlcpy(tun_name, DEFAULT_TUN_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2120 if (params && (params[0] != '\0')) {
2121 TAP_LOG(DEBUG, "parameters (%s)", params);
2123 kvlist = rte_kvargs_parse(params, valid_arguments);
2125 if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2126 ret = rte_kvargs_process(kvlist,
2128 &set_interface_name,
2136 pmd_link.link_speed = ETH_SPEED_NUM_10G;
2138 TAP_LOG(DEBUG, "Initializing pmd_tun for %s", name);
2140 ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0,
2141 ETH_TUNTAP_TYPE_TUN);
2145 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2148 rte_kvargs_free(kvlist);
2153 /* Request queue file descriptors from secondary to primary. */
2155 tap_mp_attach_queues(const char *port_name, struct rte_eth_dev *dev)
2158 struct timespec timeout = {.tv_sec = 1, .tv_nsec = 0};
2159 struct rte_mp_msg request, *reply;
2160 struct rte_mp_reply replies;
2161 struct ipc_queues *request_param = (struct ipc_queues *)request.param;
2162 struct ipc_queues *reply_param;
2163 struct pmd_process_private *process_private = dev->process_private;
2164 int queue, fd_iterator;
2166 /* Prepare the request */
2167 memset(&request, 0, sizeof(request));
2168 strlcpy(request.name, TAP_MP_KEY, sizeof(request.name));
2169 strlcpy(request_param->port_name, port_name,
2170 sizeof(request_param->port_name));
2171 request.len_param = sizeof(*request_param);
2172 /* Send request and receive reply */
2173 ret = rte_mp_request_sync(&request, &replies, &timeout);
2174 if (ret < 0 || replies.nb_received != 1) {
2175 TAP_LOG(ERR, "Failed to request queues from primary: %d",
2179 reply = &replies.msgs[0];
2180 reply_param = (struct ipc_queues *)reply->param;
2181 TAP_LOG(DEBUG, "Received IPC reply for %s", reply_param->port_name);
2183 /* Attach the queues from received file descriptors */
2184 if (reply_param->rxq_count + reply_param->txq_count != reply->num_fds) {
2185 TAP_LOG(ERR, "Unexpected number of fds received");
2189 dev->data->nb_rx_queues = reply_param->rxq_count;
2190 dev->data->nb_tx_queues = reply_param->txq_count;
2192 for (queue = 0; queue < reply_param->rxq_count; queue++)
2193 process_private->rxq_fds[queue] = reply->fds[fd_iterator++];
2194 for (queue = 0; queue < reply_param->txq_count; queue++)
2195 process_private->txq_fds[queue] = reply->fds[fd_iterator++];
2200 /* Send the queue file descriptors from the primary process to secondary. */
2202 tap_mp_sync_queues(const struct rte_mp_msg *request, const void *peer)
2204 struct rte_eth_dev *dev;
2205 struct pmd_process_private *process_private;
2206 struct rte_mp_msg reply;
2207 const struct ipc_queues *request_param =
2208 (const struct ipc_queues *)request->param;
2209 struct ipc_queues *reply_param =
2210 (struct ipc_queues *)reply.param;
2215 /* Get requested port */
2216 TAP_LOG(DEBUG, "Received IPC request for %s", request_param->port_name);
2217 ret = rte_eth_dev_get_port_by_name(request_param->port_name, &port_id);
2219 TAP_LOG(ERR, "Failed to get port id for %s",
2220 request_param->port_name);
2223 dev = &rte_eth_devices[port_id];
2224 process_private = dev->process_private;
2226 /* Fill file descriptors for all queues */
2228 reply_param->rxq_count = 0;
2229 if (dev->data->nb_rx_queues + dev->data->nb_tx_queues >
2231 TAP_LOG(ERR, "Number of rx/tx queues exceeds max number of fds");
2235 for (queue = 0; queue < dev->data->nb_rx_queues; queue++) {
2236 reply.fds[reply.num_fds++] = process_private->rxq_fds[queue];
2237 reply_param->rxq_count++;
2239 RTE_ASSERT(reply_param->rxq_count == dev->data->nb_rx_queues);
2241 reply_param->txq_count = 0;
2242 for (queue = 0; queue < dev->data->nb_tx_queues; queue++) {
2243 reply.fds[reply.num_fds++] = process_private->txq_fds[queue];
2244 reply_param->txq_count++;
2246 RTE_ASSERT(reply_param->txq_count == dev->data->nb_tx_queues);
2249 strlcpy(reply.name, request->name, sizeof(reply.name));
2250 strlcpy(reply_param->port_name, request_param->port_name,
2251 sizeof(reply_param->port_name));
2252 reply.len_param = sizeof(*reply_param);
2253 if (rte_mp_reply(&reply, peer) < 0) {
2254 TAP_LOG(ERR, "Failed to reply an IPC request to sync queues");
2260 /* Open a TAP interface device.
2263 rte_pmd_tap_probe(struct rte_vdev_device *dev)
2265 const char *name, *params;
2267 struct rte_kvargs *kvlist = NULL;
2269 char tap_name[RTE_ETH_NAME_MAX_LEN];
2270 char remote_iface[RTE_ETH_NAME_MAX_LEN];
2271 struct rte_ether_addr user_mac = { .addr_bytes = {0} };
2272 struct rte_eth_dev *eth_dev;
2273 int tap_devices_count_increased = 0;
2275 name = rte_vdev_device_name(dev);
2276 params = rte_vdev_device_args(dev);
2278 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
2279 eth_dev = rte_eth_dev_attach_secondary(name);
2281 TAP_LOG(ERR, "Failed to probe %s", name);
2284 eth_dev->dev_ops = &ops;
2285 eth_dev->device = &dev->device;
2286 eth_dev->rx_pkt_burst = pmd_rx_burst;
2287 eth_dev->tx_pkt_burst = pmd_tx_burst;
2288 if (!rte_eal_primary_proc_alive(NULL)) {
2289 TAP_LOG(ERR, "Primary process is missing");
2292 eth_dev->process_private = (struct pmd_process_private *)
2293 rte_zmalloc_socket(name,
2294 sizeof(struct pmd_process_private),
2295 RTE_CACHE_LINE_SIZE,
2296 eth_dev->device->numa_node);
2297 if (eth_dev->process_private == NULL) {
2299 "Failed to alloc memory for process private");
2303 ret = tap_mp_attach_queues(name, eth_dev);
2306 rte_eth_dev_probing_finish(eth_dev);
2310 speed = ETH_SPEED_NUM_10G;
2312 /* use tap%d which causes kernel to choose next available */
2313 strlcpy(tap_name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2314 memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2316 if (params && (params[0] != '\0')) {
2317 TAP_LOG(DEBUG, "parameters (%s)", params);
2319 kvlist = rte_kvargs_parse(params, valid_arguments);
2321 if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2322 ret = rte_kvargs_process(kvlist,
2324 &set_interface_name,
2330 if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
2331 ret = rte_kvargs_process(kvlist,
2339 if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
2340 ret = rte_kvargs_process(kvlist,
2349 pmd_link.link_speed = speed;
2351 TAP_LOG(DEBUG, "Initializing pmd_tap for %s", name);
2353 /* Register IPC feed callback */
2354 if (!tap_devices_count) {
2355 ret = rte_mp_action_register(TAP_MP_KEY, tap_mp_sync_queues);
2356 if (ret < 0 && rte_errno != ENOTSUP) {
2357 TAP_LOG(ERR, "tap: Failed to register IPC callback: %s",
2358 strerror(rte_errno));
2362 tap_devices_count++;
2363 tap_devices_count_increased = 1;
2364 ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac,
2365 ETH_TUNTAP_TYPE_TAP);
2369 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2371 if (tap_devices_count_increased == 1) {
2372 if (tap_devices_count == 1)
2373 rte_mp_action_unregister(TAP_MP_KEY);
2374 tap_devices_count--;
2377 rte_kvargs_free(kvlist);
2382 /* detach a TUNTAP device.
2385 rte_pmd_tap_remove(struct rte_vdev_device *dev)
2387 struct rte_eth_dev *eth_dev = NULL;
2388 struct pmd_internals *internals;
2389 struct pmd_process_private *process_private;
2392 /* find the ethdev entry */
2393 eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
2397 /* mac_addrs must not be freed alone because part of dev_private */
2398 eth_dev->data->mac_addrs = NULL;
2400 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2401 return rte_eth_dev_release_port(eth_dev);
2403 internals = eth_dev->data->dev_private;
2404 process_private = eth_dev->process_private;
2406 TAP_LOG(DEBUG, "Closing %s Ethernet device on numa %u",
2407 tuntap_types[internals->type], rte_socket_id());
2409 if (internals->nlsk_fd) {
2410 tap_flow_flush(eth_dev, NULL);
2411 tap_flow_implicit_flush(internals, NULL);
2412 tap_nl_final(internals->nlsk_fd);
2414 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
2415 if (process_private->rxq_fds[i] != -1) {
2416 close(process_private->rxq_fds[i]);
2417 process_private->rxq_fds[i] = -1;
2419 if (process_private->txq_fds[i] != -1) {
2420 close(process_private->txq_fds[i]);
2421 process_private->txq_fds[i] = -1;
2425 close(internals->ioctl_sock);
2426 rte_free(eth_dev->process_private);
2427 if (tap_devices_count == 1)
2428 rte_mp_action_unregister(TAP_MP_KEY);
2429 tap_devices_count--;
2430 rte_eth_dev_release_port(eth_dev);
2432 if (internals->ka_fd != -1) {
2433 close(internals->ka_fd);
2434 internals->ka_fd = -1;
2439 static struct rte_vdev_driver pmd_tun_drv = {
2440 .probe = rte_pmd_tun_probe,
2441 .remove = rte_pmd_tap_remove,
2444 static struct rte_vdev_driver pmd_tap_drv = {
2445 .probe = rte_pmd_tap_probe,
2446 .remove = rte_pmd_tap_remove,
2449 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
2450 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
2451 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
2452 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
2453 ETH_TAP_IFACE_ARG "=<string> ");
2454 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
2455 ETH_TAP_IFACE_ARG "=<string> "
2456 ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
2457 ETH_TAP_REMOTE_ARG "=<string>");
2460 RTE_INIT(tap_init_log)
2462 tap_logtype = rte_log_register("pmd.net.tap");
2463 if (tap_logtype >= 0)
2464 rte_log_set_level(tap_logtype, RTE_LOG_NOTICE);