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>
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 static int tap_devices_count;
71 static struct rte_vdev_driver pmd_tap_drv;
72 static struct rte_vdev_driver pmd_tun_drv;
74 static const char *valid_arguments[] = {
81 static unsigned int tap_unit;
82 static unsigned int tun_unit;
84 static char tuntap_name[8];
86 static volatile uint32_t tap_trigger; /* Rx trigger */
88 static struct rte_eth_link pmd_link = {
89 .link_speed = ETH_SPEED_NUM_10G,
90 .link_duplex = ETH_LINK_FULL_DUPLEX,
91 .link_status = ETH_LINK_DOWN,
92 .link_autoneg = ETH_LINK_FIXED,
96 tap_trigger_cb(int sig __rte_unused)
98 /* Valid trigger values are nonzero */
99 tap_trigger = (tap_trigger + 1) | 0x80000000;
102 /* Specifies on what netdevices the ioctl should be applied */
109 /* Message header to synchronize queues via IPC */
111 char port_name[RTE_DEV_NAME_MAX_LEN];
115 * The file descriptors are in the dedicated part
116 * of the Unix message to be translated by the kernel.
120 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
123 * Tun/Tap allocation routine
126 * Pointer to private structure.
128 * @param[in] is_keepalive
132 * -1 on failure, fd on success
135 tun_alloc(struct pmd_internals *pmd, int is_keepalive)
138 #ifdef IFF_MULTI_QUEUE
139 unsigned int features;
143 memset(&ifr, 0, sizeof(struct ifreq));
146 * Do not set IFF_NO_PI as packet information header will be needed
147 * to check if a received packet has been truncated.
149 ifr.ifr_flags = (pmd->type == ETH_TUNTAP_TYPE_TAP) ?
150 IFF_TAP : IFF_TUN | IFF_POINTOPOINT;
151 snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
153 TAP_LOG(DEBUG, "ifr_name '%s'", ifr.ifr_name);
155 fd = open(TUN_TAP_DEV_PATH, O_RDWR);
157 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
161 #ifdef IFF_MULTI_QUEUE
162 /* Grab the TUN features to verify we can work multi-queue */
163 if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
164 TAP_LOG(ERR, "%s unable to get TUN/TAP features",
168 TAP_LOG(DEBUG, "%s Features %08x", tuntap_name, features);
170 if (features & IFF_MULTI_QUEUE) {
171 TAP_LOG(DEBUG, " Multi-queue support for %d queues",
172 RTE_PMD_TAP_MAX_QUEUES);
173 ifr.ifr_flags |= IFF_MULTI_QUEUE;
177 ifr.ifr_flags |= IFF_ONE_QUEUE;
178 TAP_LOG(DEBUG, " Single queue only support");
181 /* Set the TUN/TAP configuration and set the name if needed */
182 if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
183 TAP_LOG(WARNING, "Unable to set TUNSETIFF for %s: %s",
184 ifr.ifr_name, strerror(errno));
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 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 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 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)))
285 cksum = ~rte_raw_cksum(iph, l3_len);
286 mbuf->ol_flags |= cksum ?
287 PKT_RX_IP_CKSUM_BAD :
288 PKT_RX_IP_CKSUM_GOOD;
289 } else if (l3 == RTE_PTYPE_L3_IPV6) {
290 l3_len = sizeof(struct ipv6_hdr);
292 /* IPv6 extensions are not supported */
295 if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
296 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
297 /* Don't verify checksum for multi-segment packets. */
298 if (mbuf->nb_segs > 1)
300 if (l3 == RTE_PTYPE_L3_IPV4)
301 cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
302 else if (l3 == RTE_PTYPE_L3_IPV6)
303 cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
304 mbuf->ol_flags |= cksum ?
305 PKT_RX_L4_CKSUM_BAD :
306 PKT_RX_L4_CKSUM_GOOD;
311 tap_rx_offload_get_port_capa(void)
314 * No specific port Rx offload capabilities.
320 tap_rx_offload_get_queue_capa(void)
322 return DEV_RX_OFFLOAD_SCATTER |
323 DEV_RX_OFFLOAD_IPV4_CKSUM |
324 DEV_RX_OFFLOAD_UDP_CKSUM |
325 DEV_RX_OFFLOAD_TCP_CKSUM;
328 /* Callback to handle the rx burst of packets to the correct interface and
329 * file descriptor(s) in a multi-queue setup.
332 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
334 struct rx_queue *rxq = queue;
335 struct pmd_process_private *process_private;
337 unsigned long num_rx_bytes = 0;
338 uint32_t trigger = tap_trigger;
340 if (trigger == rxq->trigger_seen)
343 rxq->trigger_seen = trigger;
344 process_private = rte_eth_devices[rxq->in_port].process_private;
345 rte_compiler_barrier();
346 for (num_rx = 0; num_rx < nb_pkts; ) {
347 struct rte_mbuf *mbuf = rxq->pool;
348 struct rte_mbuf *seg = NULL;
349 struct rte_mbuf *new_tail = NULL;
350 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
353 len = readv(process_private->rxq_fds[rxq->queue_id],
355 1 + (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
356 rxq->nb_rx_desc : 1));
357 if (len < (int)sizeof(struct tun_pi))
360 /* Packet couldn't fit in the provided mbuf */
361 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
362 rxq->stats.ierrors++;
366 len -= sizeof(struct tun_pi);
369 mbuf->port = rxq->in_port;
371 struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
373 if (unlikely(!buf)) {
374 rxq->stats.rx_nombuf++;
375 /* No new buf has been allocated: do nothing */
376 if (!new_tail || !seg)
380 rte_pktmbuf_free(mbuf);
384 seg = seg ? seg->next : mbuf;
385 if (rxq->pool == mbuf)
388 new_tail->next = buf;
390 new_tail->next = seg->next;
392 /* iovecs[0] is reserved for packet info (pi) */
393 (*rxq->iovecs)[mbuf->nb_segs].iov_len =
394 buf->buf_len - data_off;
395 (*rxq->iovecs)[mbuf->nb_segs].iov_base =
396 (char *)buf->buf_addr + data_off;
398 seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
399 seg->data_off = data_off;
401 len -= seg->data_len;
405 /* First segment has headroom, not the others */
409 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
411 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
412 tap_verify_csum(mbuf);
414 /* account for the receive frame */
415 bufs[num_rx++] = mbuf;
416 num_rx_bytes += mbuf->pkt_len;
419 rxq->stats.ipackets += num_rx;
420 rxq->stats.ibytes += num_rx_bytes;
426 tap_tx_offload_get_port_capa(void)
429 * No specific port Tx offload capabilities.
435 tap_tx_offload_get_queue_capa(void)
437 return DEV_TX_OFFLOAD_MULTI_SEGS |
438 DEV_TX_OFFLOAD_IPV4_CKSUM |
439 DEV_TX_OFFLOAD_UDP_CKSUM |
440 DEV_TX_OFFLOAD_TCP_CKSUM |
441 DEV_TX_OFFLOAD_TCP_TSO;
444 /* Finalize l4 checksum calculation */
446 tap_tx_l4_cksum(uint16_t *l4_cksum, uint16_t l4_phdr_cksum,
447 uint32_t l4_raw_cksum)
452 cksum = __rte_raw_cksum_reduce(l4_raw_cksum);
453 cksum += l4_phdr_cksum;
455 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
456 cksum = (~cksum) & 0xffff;
463 /* Accumaulate L4 raw checksums */
465 tap_tx_l4_add_rcksum(char *l4_data, unsigned int l4_len, uint16_t *l4_cksum,
466 uint32_t *l4_raw_cksum)
468 if (l4_cksum == NULL)
471 *l4_raw_cksum = __rte_raw_cksum(l4_data, l4_len, *l4_raw_cksum);
474 /* L3 and L4 pseudo headers checksum offloads */
476 tap_tx_l3_cksum(char *packet, uint64_t ol_flags, unsigned int l2_len,
477 unsigned int l3_len, unsigned int l4_len, uint16_t **l4_cksum,
478 uint16_t *l4_phdr_cksum, uint32_t *l4_raw_cksum)
480 void *l3_hdr = packet + l2_len;
482 if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
483 struct ipv4_hdr *iph = l3_hdr;
486 iph->hdr_checksum = 0;
487 cksum = rte_raw_cksum(iph, l3_len);
488 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
490 if (ol_flags & PKT_TX_L4_MASK) {
493 l4_hdr = packet + l2_len + l3_len;
494 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
495 *l4_cksum = &((struct udp_hdr *)l4_hdr)->dgram_cksum;
496 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
497 *l4_cksum = &((struct tcp_hdr *)l4_hdr)->cksum;
501 if (ol_flags & PKT_TX_IPV4)
502 *l4_phdr_cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
504 *l4_phdr_cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
505 *l4_raw_cksum = __rte_raw_cksum(l4_hdr, l4_len, 0);
510 tap_write_mbufs(struct tx_queue *txq, uint16_t num_mbufs,
511 struct rte_mbuf **pmbufs,
512 uint16_t *num_packets, unsigned long *num_tx_bytes)
516 struct pmd_process_private *process_private;
518 process_private = rte_eth_devices[txq->out_port].process_private;
520 for (i = 0; i < num_mbufs; i++) {
521 struct rte_mbuf *mbuf = pmbufs[i];
522 struct iovec iovecs[mbuf->nb_segs + 2];
523 struct tun_pi pi = { .flags = 0, .proto = 0x00 };
524 struct rte_mbuf *seg = mbuf;
525 char m_copy[mbuf->data_len];
529 int k; /* current index in iovecs for copying segments */
530 uint16_t seg_len; /* length of first segment */
532 uint16_t *l4_cksum; /* l4 checksum (pseudo header + payload) */
533 uint32_t l4_raw_cksum = 0; /* TCP/UDP payload raw checksum */
534 uint16_t l4_phdr_cksum = 0; /* TCP/UDP pseudo header checksum */
535 uint16_t is_cksum = 0; /* in case cksum should be offloaded */
538 if (txq->type == ETH_TUNTAP_TYPE_TUN) {
540 * TUN and TAP are created with IFF_NO_PI disabled.
541 * For TUN PMD this mandatory as fields are used by
542 * Kernel tun.c to determine whether its IP or non IP
545 * The logic fetches the first byte of data from mbuf
546 * then compares whether its v4 or v6. If first byte
547 * is 4 or 6, then protocol field is updated.
549 char *buff_data = rte_pktmbuf_mtod(seg, void *);
550 proto = (*buff_data & 0xf0);
551 pi.proto = (proto == 0x40) ?
552 rte_cpu_to_be_16(ETHER_TYPE_IPv4) :
554 rte_cpu_to_be_16(ETHER_TYPE_IPv6) :
559 iovecs[k].iov_base = π
560 iovecs[k].iov_len = sizeof(pi);
563 nb_segs = mbuf->nb_segs;
565 ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
566 (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
567 (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
570 /* Support only packets with at least layer 4
571 * header included in the first segment
573 seg_len = rte_pktmbuf_data_len(mbuf);
574 l234_hlen = mbuf->l2_len + mbuf->l3_len + mbuf->l4_len;
575 if (seg_len < l234_hlen)
578 /* To change checksums, work on a * copy of l2, l3
579 * headers + l4 pseudo header
581 rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
583 tap_tx_l3_cksum(m_copy, mbuf->ol_flags,
584 mbuf->l2_len, mbuf->l3_len, mbuf->l4_len,
585 &l4_cksum, &l4_phdr_cksum,
587 iovecs[k].iov_base = m_copy;
588 iovecs[k].iov_len = l234_hlen;
591 /* Update next iovecs[] beyond l2, l3, l4 headers */
592 if (seg_len > l234_hlen) {
593 iovecs[k].iov_len = seg_len - l234_hlen;
595 rte_pktmbuf_mtod(seg, char *) +
597 tap_tx_l4_add_rcksum(iovecs[k].iov_base,
598 iovecs[k].iov_len, l4_cksum,
606 for (j = k; j <= nb_segs; j++) {
607 iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
608 iovecs[j].iov_base = rte_pktmbuf_mtod(seg, void *);
610 tap_tx_l4_add_rcksum(iovecs[j].iov_base,
611 iovecs[j].iov_len, l4_cksum,
617 tap_tx_l4_cksum(l4_cksum, l4_phdr_cksum, l4_raw_cksum);
619 /* copy the tx frame data */
620 n = writev(process_private->txq_fds[txq->queue_id], iovecs, j);
624 (*num_tx_bytes) += rte_pktmbuf_pkt_len(mbuf);
628 /* Callback to handle sending packets from the tap interface
631 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
633 struct tx_queue *txq = queue;
635 uint16_t num_packets = 0;
636 unsigned long num_tx_bytes = 0;
640 if (unlikely(nb_pkts == 0))
643 struct rte_mbuf *gso_mbufs[MAX_GSO_MBUFS];
644 max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
645 for (i = 0; i < nb_pkts; i++) {
646 struct rte_mbuf *mbuf_in = bufs[num_tx];
647 struct rte_mbuf **mbuf;
648 uint16_t num_mbufs = 0;
649 uint16_t tso_segsz = 0;
655 tso = mbuf_in->ol_flags & PKT_TX_TCP_SEG;
657 struct rte_gso_ctx *gso_ctx = &txq->gso_ctx;
659 assert(gso_ctx != NULL);
661 /* TCP segmentation implies TCP checksum offload */
662 mbuf_in->ol_flags |= PKT_TX_TCP_CKSUM;
664 /* gso size is calculated without ETHER_CRC_LEN */
665 hdrs_len = mbuf_in->l2_len + mbuf_in->l3_len +
667 tso_segsz = mbuf_in->tso_segsz + hdrs_len;
668 if (unlikely(tso_segsz == hdrs_len) ||
669 tso_segsz > *txq->mtu) {
673 gso_ctx->gso_size = tso_segsz;
674 ret = rte_gso_segment(mbuf_in, /* packet to segment */
675 gso_ctx, /* gso control block */
676 (struct rte_mbuf **)&gso_mbufs, /* out mbufs */
677 RTE_DIM(gso_mbufs)); /* max tso mbufs */
679 /* ret contains the number of new created mbufs */
686 /* stats.errs will be incremented */
687 if (rte_pktmbuf_pkt_len(mbuf_in) > max_size)
690 /* ret 0 indicates no new mbufs were created */
696 tap_write_mbufs(txq, num_mbufs, mbuf,
697 &num_packets, &num_tx_bytes);
699 /* free original mbuf */
700 rte_pktmbuf_free(mbuf_in);
702 for (j = 0; j < ret; j++)
703 rte_pktmbuf_free(mbuf[j]);
706 txq->stats.opackets += num_packets;
707 txq->stats.errs += nb_pkts - num_tx;
708 txq->stats.obytes += num_tx_bytes;
714 tap_ioctl_req2str(unsigned long request)
718 return "SIOCSIFFLAGS";
720 return "SIOCGIFFLAGS";
722 return "SIOCGIFHWADDR";
724 return "SIOCSIFHWADDR";
732 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
733 struct ifreq *ifr, int set, enum ioctl_mode mode)
735 short req_flags = ifr->ifr_flags;
736 int remote = pmd->remote_if_index &&
737 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
739 if (!pmd->remote_if_index && mode == REMOTE_ONLY)
742 * If there is a remote netdevice, apply ioctl on it, then apply it on
747 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
748 else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
749 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
752 /* fetch current flags to leave other flags untouched */
753 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
756 ifr->ifr_flags |= req_flags;
758 ifr->ifr_flags &= ~req_flags;
766 RTE_LOG(WARNING, PMD, "%s: ioctl() called with wrong arg\n",
770 if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
772 if (remote-- && mode == LOCAL_AND_REMOTE)
777 TAP_LOG(DEBUG, "%s(%s) failed: %s(%d)", ifr->ifr_name,
778 tap_ioctl_req2str(request), strerror(errno), errno);
783 tap_link_set_down(struct rte_eth_dev *dev)
785 struct pmd_internals *pmd = dev->data->dev_private;
786 struct ifreq ifr = { .ifr_flags = IFF_UP };
788 dev->data->dev_link.link_status = ETH_LINK_DOWN;
789 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
793 tap_link_set_up(struct rte_eth_dev *dev)
795 struct pmd_internals *pmd = dev->data->dev_private;
796 struct ifreq ifr = { .ifr_flags = IFF_UP };
798 dev->data->dev_link.link_status = ETH_LINK_UP;
799 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
803 tap_dev_start(struct rte_eth_dev *dev)
807 err = tap_intr_handle_set(dev, 1);
811 err = tap_link_set_up(dev);
815 for (i = 0; i < dev->data->nb_tx_queues; i++)
816 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
817 for (i = 0; i < dev->data->nb_rx_queues; i++)
818 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
823 /* This function gets called when the current port gets stopped.
826 tap_dev_stop(struct rte_eth_dev *dev)
830 for (i = 0; i < dev->data->nb_tx_queues; i++)
831 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
832 for (i = 0; i < dev->data->nb_rx_queues; i++)
833 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
835 tap_intr_handle_set(dev, 0);
836 tap_link_set_down(dev);
840 tap_dev_configure(struct rte_eth_dev *dev)
842 if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
844 "%s: number of rx queues %d exceeds max num of queues %d",
846 dev->data->nb_rx_queues,
847 RTE_PMD_TAP_MAX_QUEUES);
850 if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
852 "%s: number of tx queues %d exceeds max num of queues %d",
854 dev->data->nb_tx_queues,
855 RTE_PMD_TAP_MAX_QUEUES);
859 TAP_LOG(INFO, "%s: %p: TX configured queues number: %u",
860 dev->device->name, (void *)dev, dev->data->nb_tx_queues);
862 TAP_LOG(INFO, "%s: %p: RX configured queues number: %u",
863 dev->device->name, (void *)dev, dev->data->nb_rx_queues);
869 tap_dev_speed_capa(void)
871 uint32_t speed = pmd_link.link_speed;
874 if (speed >= ETH_SPEED_NUM_10M)
875 capa |= ETH_LINK_SPEED_10M;
876 if (speed >= ETH_SPEED_NUM_100M)
877 capa |= ETH_LINK_SPEED_100M;
878 if (speed >= ETH_SPEED_NUM_1G)
879 capa |= ETH_LINK_SPEED_1G;
880 if (speed >= ETH_SPEED_NUM_5G)
881 capa |= ETH_LINK_SPEED_2_5G;
882 if (speed >= ETH_SPEED_NUM_5G)
883 capa |= ETH_LINK_SPEED_5G;
884 if (speed >= ETH_SPEED_NUM_10G)
885 capa |= ETH_LINK_SPEED_10G;
886 if (speed >= ETH_SPEED_NUM_20G)
887 capa |= ETH_LINK_SPEED_20G;
888 if (speed >= ETH_SPEED_NUM_25G)
889 capa |= ETH_LINK_SPEED_25G;
890 if (speed >= ETH_SPEED_NUM_40G)
891 capa |= ETH_LINK_SPEED_40G;
892 if (speed >= ETH_SPEED_NUM_50G)
893 capa |= ETH_LINK_SPEED_50G;
894 if (speed >= ETH_SPEED_NUM_56G)
895 capa |= ETH_LINK_SPEED_56G;
896 if (speed >= ETH_SPEED_NUM_100G)
897 capa |= ETH_LINK_SPEED_100G;
903 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
905 struct pmd_internals *internals = dev->data->dev_private;
907 dev_info->if_index = internals->if_index;
908 dev_info->max_mac_addrs = 1;
909 dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
910 dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
911 dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
912 dev_info->min_rx_bufsize = 0;
913 dev_info->speed_capa = tap_dev_speed_capa();
914 dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
915 dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
916 dev_info->rx_queue_offload_capa;
917 dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
918 dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
919 dev_info->tx_queue_offload_capa;
920 dev_info->hash_key_size = TAP_RSS_HASH_KEY_SIZE;
922 * limitation: TAP supports all of IP, UDP and TCP hash
923 * functions together and not in partial combinations
925 dev_info->flow_type_rss_offloads = ~TAP_RSS_HF_MASK;
929 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
931 unsigned int i, imax;
932 unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
933 unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
934 unsigned long rx_nombuf = 0, ierrors = 0;
935 const struct pmd_internals *pmd = dev->data->dev_private;
937 /* rx queue statistics */
938 imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
939 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
940 for (i = 0; i < imax; i++) {
941 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
942 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
943 rx_total += tap_stats->q_ipackets[i];
944 rx_bytes_total += tap_stats->q_ibytes[i];
945 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
946 ierrors += pmd->rxq[i].stats.ierrors;
949 /* tx queue statistics */
950 imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
951 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
953 for (i = 0; i < imax; i++) {
954 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
955 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
956 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
957 tx_total += tap_stats->q_opackets[i];
958 tx_err_total += tap_stats->q_errors[i];
959 tx_bytes_total += tap_stats->q_obytes[i];
962 tap_stats->ipackets = rx_total;
963 tap_stats->ibytes = rx_bytes_total;
964 tap_stats->ierrors = ierrors;
965 tap_stats->rx_nombuf = rx_nombuf;
966 tap_stats->opackets = tx_total;
967 tap_stats->oerrors = tx_err_total;
968 tap_stats->obytes = tx_bytes_total;
973 tap_stats_reset(struct rte_eth_dev *dev)
976 struct pmd_internals *pmd = dev->data->dev_private;
978 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
979 pmd->rxq[i].stats.ipackets = 0;
980 pmd->rxq[i].stats.ibytes = 0;
981 pmd->rxq[i].stats.ierrors = 0;
982 pmd->rxq[i].stats.rx_nombuf = 0;
984 pmd->txq[i].stats.opackets = 0;
985 pmd->txq[i].stats.errs = 0;
986 pmd->txq[i].stats.obytes = 0;
991 tap_dev_close(struct rte_eth_dev *dev)
994 struct pmd_internals *internals = dev->data->dev_private;
995 struct pmd_process_private *process_private = dev->process_private;
997 tap_link_set_down(dev);
998 tap_flow_flush(dev, NULL);
999 tap_flow_implicit_flush(internals, NULL);
1001 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1002 if (process_private->rxq_fds[i] != -1) {
1003 close(process_private->rxq_fds[i]);
1004 process_private->rxq_fds[i] = -1;
1006 if (process_private->txq_fds[i] != -1) {
1007 close(process_private->txq_fds[i]);
1008 process_private->txq_fds[i] = -1;
1012 if (internals->remote_if_index) {
1013 /* Restore initial remote state */
1014 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
1015 &internals->remote_initial_flags);
1018 if (internals->ka_fd != -1) {
1019 close(internals->ka_fd);
1020 internals->ka_fd = -1;
1023 * Since TUN device has no more opened file descriptors
1024 * it will be removed from kernel
1029 tap_rx_queue_release(void *queue)
1031 struct rx_queue *rxq = queue;
1032 struct pmd_process_private *process_private;
1036 process_private = rte_eth_devices[rxq->in_port].process_private;
1037 if (process_private->rxq_fds[rxq->queue_id] > 0) {
1038 close(process_private->rxq_fds[rxq->queue_id]);
1039 process_private->rxq_fds[rxq->queue_id] = -1;
1040 rte_pktmbuf_free(rxq->pool);
1041 rte_free(rxq->iovecs);
1048 tap_tx_queue_release(void *queue)
1050 struct tx_queue *txq = queue;
1051 struct pmd_process_private *process_private;
1055 process_private = rte_eth_devices[txq->out_port].process_private;
1057 if (process_private->txq_fds[txq->queue_id] > 0) {
1058 close(process_private->txq_fds[txq->queue_id]);
1059 process_private->txq_fds[txq->queue_id] = -1;
1064 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
1066 struct rte_eth_link *dev_link = &dev->data->dev_link;
1067 struct pmd_internals *pmd = dev->data->dev_private;
1068 struct ifreq ifr = { .ifr_flags = 0 };
1070 if (pmd->remote_if_index) {
1071 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
1072 if (!(ifr.ifr_flags & IFF_UP) ||
1073 !(ifr.ifr_flags & IFF_RUNNING)) {
1074 dev_link->link_status = ETH_LINK_DOWN;
1078 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
1079 dev_link->link_status =
1080 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
1087 tap_promisc_enable(struct rte_eth_dev *dev)
1089 struct pmd_internals *pmd = dev->data->dev_private;
1090 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1092 dev->data->promiscuous = 1;
1093 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1094 if (pmd->remote_if_index && !pmd->flow_isolate)
1095 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
1099 tap_promisc_disable(struct rte_eth_dev *dev)
1101 struct pmd_internals *pmd = dev->data->dev_private;
1102 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1104 dev->data->promiscuous = 0;
1105 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1106 if (pmd->remote_if_index && !pmd->flow_isolate)
1107 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
1111 tap_allmulti_enable(struct rte_eth_dev *dev)
1113 struct pmd_internals *pmd = dev->data->dev_private;
1114 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1116 dev->data->all_multicast = 1;
1117 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1118 if (pmd->remote_if_index && !pmd->flow_isolate)
1119 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
1123 tap_allmulti_disable(struct rte_eth_dev *dev)
1125 struct pmd_internals *pmd = dev->data->dev_private;
1126 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1128 dev->data->all_multicast = 0;
1129 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1130 if (pmd->remote_if_index && !pmd->flow_isolate)
1131 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
1135 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1137 struct pmd_internals *pmd = dev->data->dev_private;
1138 enum ioctl_mode mode = LOCAL_ONLY;
1142 if (pmd->type == ETH_TUNTAP_TYPE_TUN) {
1143 TAP_LOG(ERR, "%s: can't MAC address for TUN",
1148 if (is_zero_ether_addr(mac_addr)) {
1149 TAP_LOG(ERR, "%s: can't set an empty MAC address",
1153 /* Check the actual current MAC address on the tap netdevice */
1154 ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY);
1157 if (is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
1160 /* Check the current MAC address on the remote */
1161 ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY);
1164 if (!is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
1166 mode = LOCAL_AND_REMOTE;
1167 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1168 rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
1169 ret = tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode);
1172 rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
1173 if (pmd->remote_if_index && !pmd->flow_isolate) {
1174 /* Replace MAC redirection rule after a MAC change */
1175 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC);
1178 "%s: Couldn't delete MAC redirection rule",
1182 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC);
1185 "%s: Couldn't add MAC redirection rule",
1195 tap_gso_ctx_setup(struct rte_gso_ctx *gso_ctx, struct rte_eth_dev *dev)
1201 * Create private mbuf pool with TAP_GSO_MBUF_SEG_SIZE bytes
1202 * size per mbuf use this pool for both direct and indirect mbufs
1205 struct rte_mempool *mp; /* Mempool for GSO packets */
1207 /* initialize GSO context */
1208 gso_types = DEV_TX_OFFLOAD_TCP_TSO;
1209 snprintf(pool_name, sizeof(pool_name), "mp_%s", dev->device->name);
1210 mp = rte_mempool_lookup((const char *)pool_name);
1212 mp = rte_pktmbuf_pool_create(pool_name, TAP_GSO_MBUFS_NUM,
1213 TAP_GSO_MBUF_CACHE_SIZE, 0,
1214 RTE_PKTMBUF_HEADROOM + TAP_GSO_MBUF_SEG_SIZE,
1217 struct pmd_internals *pmd = dev->data->dev_private;
1218 RTE_LOG(DEBUG, PMD, "%s: failed to create mbuf pool for device %s\n",
1219 pmd->name, dev->device->name);
1224 gso_ctx->direct_pool = mp;
1225 gso_ctx->indirect_pool = mp;
1226 gso_ctx->gso_types = gso_types;
1227 gso_ctx->gso_size = 0; /* gso_size is set in tx_burst() per packet */
1234 tap_setup_queue(struct rte_eth_dev *dev,
1235 struct pmd_internals *internals,
1243 struct pmd_internals *pmd = dev->data->dev_private;
1244 struct pmd_process_private *process_private = dev->process_private;
1245 struct rx_queue *rx = &internals->rxq[qid];
1246 struct tx_queue *tx = &internals->txq[qid];
1247 struct rte_gso_ctx *gso_ctx;
1250 fd = &process_private->rxq_fds[qid];
1251 other_fd = &process_private->txq_fds[qid];
1255 fd = &process_private->txq_fds[qid];
1256 other_fd = &process_private->rxq_fds[qid];
1258 gso_ctx = &tx->gso_ctx;
1261 /* fd for this queue already exists */
1262 TAP_LOG(DEBUG, "%s: fd %d for %s queue qid %d exists",
1263 pmd->name, *fd, dir, qid);
1265 } else if (*other_fd != -1) {
1266 /* Only other_fd exists. dup it */
1267 *fd = dup(*other_fd);
1270 TAP_LOG(ERR, "%s: dup() failed.", pmd->name);
1273 TAP_LOG(DEBUG, "%s: dup fd %d for %s queue qid %d (%d)",
1274 pmd->name, *other_fd, dir, qid, *fd);
1276 /* Both RX and TX fds do not exist (equal -1). Create fd */
1277 *fd = tun_alloc(pmd, 0);
1279 *fd = -1; /* restore original value */
1280 TAP_LOG(ERR, "%s: tun_alloc() failed.", pmd->name);
1283 TAP_LOG(DEBUG, "%s: add %s queue for qid %d fd %d",
1284 pmd->name, dir, qid, *fd);
1287 tx->mtu = &dev->data->mtu;
1288 rx->rxmode = &dev->data->dev_conf.rxmode;
1290 ret = tap_gso_ctx_setup(gso_ctx, dev);
1295 tx->type = pmd->type;
1301 tap_rx_queue_setup(struct rte_eth_dev *dev,
1302 uint16_t rx_queue_id,
1303 uint16_t nb_rx_desc,
1304 unsigned int socket_id,
1305 const struct rte_eth_rxconf *rx_conf __rte_unused,
1306 struct rte_mempool *mp)
1308 struct pmd_internals *internals = dev->data->dev_private;
1309 struct pmd_process_private *process_private = dev->process_private;
1310 struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1311 struct rte_mbuf **tmp = &rxq->pool;
1312 long iov_max = sysconf(_SC_IOV_MAX);
1313 uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1314 struct iovec (*iovecs)[nb_desc + 1];
1315 int data_off = RTE_PKTMBUF_HEADROOM;
1320 if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1322 "nb_rx_queues %d too small or mempool NULL",
1323 dev->data->nb_rx_queues);
1328 rxq->trigger_seen = 1; /* force initial burst */
1329 rxq->in_port = dev->data->port_id;
1330 rxq->queue_id = rx_queue_id;
1331 rxq->nb_rx_desc = nb_desc;
1332 iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1336 "%s: Couldn't allocate %d RX descriptors",
1337 dev->device->name, nb_desc);
1340 rxq->iovecs = iovecs;
1342 dev->data->rx_queues[rx_queue_id] = rxq;
1343 fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1349 (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1350 (*rxq->iovecs)[0].iov_base = &rxq->pi;
1352 for (i = 1; i <= nb_desc; i++) {
1353 *tmp = rte_pktmbuf_alloc(rxq->mp);
1356 "%s: couldn't allocate memory for queue %d",
1357 dev->device->name, rx_queue_id);
1361 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1362 (*rxq->iovecs)[i].iov_base =
1363 (char *)(*tmp)->buf_addr + data_off;
1365 tmp = &(*tmp)->next;
1368 TAP_LOG(DEBUG, " RX TUNTAP device name %s, qid %d on fd %d",
1369 internals->name, rx_queue_id,
1370 process_private->rxq_fds[rx_queue_id]);
1375 rte_pktmbuf_free(rxq->pool);
1377 rte_free(rxq->iovecs);
1383 tap_tx_queue_setup(struct rte_eth_dev *dev,
1384 uint16_t tx_queue_id,
1385 uint16_t nb_tx_desc __rte_unused,
1386 unsigned int socket_id __rte_unused,
1387 const struct rte_eth_txconf *tx_conf)
1389 struct pmd_internals *internals = dev->data->dev_private;
1390 struct pmd_process_private *process_private = dev->process_private;
1391 struct tx_queue *txq;
1395 if (tx_queue_id >= dev->data->nb_tx_queues)
1397 dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1398 txq = dev->data->tx_queues[tx_queue_id];
1399 txq->out_port = dev->data->port_id;
1400 txq->queue_id = tx_queue_id;
1402 offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1403 txq->csum = !!(offloads &
1404 (DEV_TX_OFFLOAD_IPV4_CKSUM |
1405 DEV_TX_OFFLOAD_UDP_CKSUM |
1406 DEV_TX_OFFLOAD_TCP_CKSUM));
1408 ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1412 " TX TUNTAP device name %s, qid %d on fd %d csum %s",
1413 internals->name, tx_queue_id,
1414 process_private->txq_fds[tx_queue_id],
1415 txq->csum ? "on" : "off");
1421 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1423 struct pmd_internals *pmd = dev->data->dev_private;
1424 struct ifreq ifr = { .ifr_mtu = mtu };
1427 err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1429 dev->data->mtu = mtu;
1435 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1436 struct ether_addr *mc_addr_set __rte_unused,
1437 uint32_t nb_mc_addr __rte_unused)
1440 * Nothing to do actually: the tap has no filtering whatsoever, every
1441 * packet is received.
1447 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1449 struct rte_eth_dev *dev = arg;
1450 struct pmd_internals *pmd = dev->data->dev_private;
1451 struct ifinfomsg *info = NLMSG_DATA(nh);
1453 if (nh->nlmsg_type != RTM_NEWLINK ||
1454 (info->ifi_index != pmd->if_index &&
1455 info->ifi_index != pmd->remote_if_index))
1457 return tap_link_update(dev, 0);
1461 tap_dev_intr_handler(void *cb_arg)
1463 struct rte_eth_dev *dev = cb_arg;
1464 struct pmd_internals *pmd = dev->data->dev_private;
1466 tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1470 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1472 struct pmd_internals *pmd = dev->data->dev_private;
1474 /* In any case, disable interrupt if the conf is no longer there. */
1475 if (!dev->data->dev_conf.intr_conf.lsc) {
1476 if (pmd->intr_handle.fd != -1) {
1477 tap_nl_final(pmd->intr_handle.fd);
1478 rte_intr_callback_unregister(&pmd->intr_handle,
1479 tap_dev_intr_handler, dev);
1484 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1485 if (unlikely(pmd->intr_handle.fd == -1))
1487 return rte_intr_callback_register(
1488 &pmd->intr_handle, tap_dev_intr_handler, dev);
1490 tap_nl_final(pmd->intr_handle.fd);
1491 return rte_intr_callback_unregister(&pmd->intr_handle,
1492 tap_dev_intr_handler, dev);
1496 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1500 err = tap_lsc_intr_handle_set(dev, set);
1503 err = tap_rx_intr_vec_set(dev, set);
1505 tap_lsc_intr_handle_set(dev, 0);
1509 static const uint32_t*
1510 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1512 static const uint32_t ptypes[] = {
1513 RTE_PTYPE_INNER_L2_ETHER,
1514 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1515 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1516 RTE_PTYPE_INNER_L3_IPV4,
1517 RTE_PTYPE_INNER_L3_IPV4_EXT,
1518 RTE_PTYPE_INNER_L3_IPV6,
1519 RTE_PTYPE_INNER_L3_IPV6_EXT,
1520 RTE_PTYPE_INNER_L4_FRAG,
1521 RTE_PTYPE_INNER_L4_UDP,
1522 RTE_PTYPE_INNER_L4_TCP,
1523 RTE_PTYPE_INNER_L4_SCTP,
1525 RTE_PTYPE_L2_ETHER_VLAN,
1526 RTE_PTYPE_L2_ETHER_QINQ,
1528 RTE_PTYPE_L3_IPV4_EXT,
1529 RTE_PTYPE_L3_IPV6_EXT,
1541 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1542 struct rte_eth_fc_conf *fc_conf)
1544 fc_conf->mode = RTE_FC_NONE;
1549 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1550 struct rte_eth_fc_conf *fc_conf)
1552 if (fc_conf->mode != RTE_FC_NONE)
1558 * DPDK callback to update the RSS hash configuration.
1561 * Pointer to Ethernet device structure.
1562 * @param[in] rss_conf
1563 * RSS configuration data.
1566 * 0 on success, a negative errno value otherwise and rte_errno is set.
1569 tap_rss_hash_update(struct rte_eth_dev *dev,
1570 struct rte_eth_rss_conf *rss_conf)
1572 if (rss_conf->rss_hf & TAP_RSS_HF_MASK) {
1576 if (rss_conf->rss_key && rss_conf->rss_key_len) {
1578 * Currently TAP RSS key is hard coded
1579 * and cannot be updated
1582 "port %u RSS key cannot be updated",
1583 dev->data->port_id);
1591 tap_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1593 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1599 tap_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1601 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1607 tap_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1609 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1615 tap_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1617 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1621 static const struct eth_dev_ops ops = {
1622 .dev_start = tap_dev_start,
1623 .dev_stop = tap_dev_stop,
1624 .dev_close = tap_dev_close,
1625 .dev_configure = tap_dev_configure,
1626 .dev_infos_get = tap_dev_info,
1627 .rx_queue_setup = tap_rx_queue_setup,
1628 .tx_queue_setup = tap_tx_queue_setup,
1629 .rx_queue_start = tap_rx_queue_start,
1630 .tx_queue_start = tap_tx_queue_start,
1631 .rx_queue_stop = tap_rx_queue_stop,
1632 .tx_queue_stop = tap_tx_queue_stop,
1633 .rx_queue_release = tap_rx_queue_release,
1634 .tx_queue_release = tap_tx_queue_release,
1635 .flow_ctrl_get = tap_flow_ctrl_get,
1636 .flow_ctrl_set = tap_flow_ctrl_set,
1637 .link_update = tap_link_update,
1638 .dev_set_link_up = tap_link_set_up,
1639 .dev_set_link_down = tap_link_set_down,
1640 .promiscuous_enable = tap_promisc_enable,
1641 .promiscuous_disable = tap_promisc_disable,
1642 .allmulticast_enable = tap_allmulti_enable,
1643 .allmulticast_disable = tap_allmulti_disable,
1644 .mac_addr_set = tap_mac_set,
1645 .mtu_set = tap_mtu_set,
1646 .set_mc_addr_list = tap_set_mc_addr_list,
1647 .stats_get = tap_stats_get,
1648 .stats_reset = tap_stats_reset,
1649 .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1650 .rss_hash_update = tap_rss_hash_update,
1651 .filter_ctrl = tap_dev_filter_ctrl,
1655 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
1656 char *remote_iface, struct ether_addr *mac_addr,
1657 enum rte_tuntap_type type)
1659 int numa_node = rte_socket_id();
1660 struct rte_eth_dev *dev;
1661 struct pmd_internals *pmd;
1662 struct pmd_process_private *process_private;
1663 struct rte_eth_dev_data *data;
1667 TAP_LOG(DEBUG, "%s device on numa %u",
1668 tuntap_name, rte_socket_id());
1670 dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1672 TAP_LOG(ERR, "%s Unable to allocate device struct",
1674 goto error_exit_nodev;
1677 process_private = (struct pmd_process_private *)
1678 rte_zmalloc_socket(tap_name, sizeof(struct pmd_process_private),
1679 RTE_CACHE_LINE_SIZE, dev->device->numa_node);
1681 if (process_private == NULL) {
1682 TAP_LOG(ERR, "Failed to alloc memory for process private");
1685 pmd = dev->data->dev_private;
1686 dev->process_private = process_private;
1688 snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1691 pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1692 if (pmd->ioctl_sock == -1) {
1694 "%s Unable to get a socket for management: %s",
1695 tuntap_name, strerror(errno));
1699 /* Setup some default values */
1701 data->dev_private = pmd;
1702 data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1703 data->numa_node = numa_node;
1705 data->dev_link = pmd_link;
1706 data->mac_addrs = &pmd->eth_addr;
1707 /* Set the number of RX and TX queues */
1708 data->nb_rx_queues = 0;
1709 data->nb_tx_queues = 0;
1711 dev->dev_ops = &ops;
1712 dev->rx_pkt_burst = pmd_rx_burst;
1713 dev->tx_pkt_burst = pmd_tx_burst;
1715 pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1716 pmd->intr_handle.fd = -1;
1717 dev->intr_handle = &pmd->intr_handle;
1719 /* Presetup the fds to -1 as being not valid */
1721 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1722 process_private->rxq_fds[i] = -1;
1723 process_private->txq_fds[i] = -1;
1726 if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1727 if (is_zero_ether_addr(mac_addr))
1728 eth_random_addr((uint8_t *)&pmd->eth_addr);
1730 rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1734 * Allocate a TUN device keep-alive file descriptor that will only be
1735 * closed when the TUN device itself is closed or removed.
1736 * This keep-alive file descriptor will guarantee that the TUN device
1737 * exists even when all of its queues are closed
1739 pmd->ka_fd = tun_alloc(pmd, 1);
1740 if (pmd->ka_fd == -1) {
1741 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
1745 ifr.ifr_mtu = dev->data->mtu;
1746 if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1749 if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1750 memset(&ifr, 0, sizeof(struct ifreq));
1751 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1752 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
1754 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1759 * Set up everything related to rte_flow:
1761 * - tap / remote if_index
1762 * - mandatory QDISCs
1763 * - rte_flow actual/implicit lists
1766 pmd->nlsk_fd = tap_nl_init(0);
1767 if (pmd->nlsk_fd == -1) {
1768 TAP_LOG(WARNING, "%s: failed to create netlink socket.",
1770 goto disable_rte_flow;
1772 pmd->if_index = if_nametoindex(pmd->name);
1773 if (!pmd->if_index) {
1774 TAP_LOG(ERR, "%s: failed to get if_index.", pmd->name);
1775 goto disable_rte_flow;
1777 if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1778 TAP_LOG(ERR, "%s: failed to create multiq qdisc.",
1780 goto disable_rte_flow;
1782 if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1783 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1785 goto disable_rte_flow;
1787 LIST_INIT(&pmd->flows);
1789 if (strlen(remote_iface)) {
1790 pmd->remote_if_index = if_nametoindex(remote_iface);
1791 if (!pmd->remote_if_index) {
1792 TAP_LOG(ERR, "%s: failed to get %s if_index.",
1793 pmd->name, remote_iface);
1796 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1797 "%s", remote_iface);
1799 /* Save state of remote device */
1800 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1802 /* Replicate remote MAC address */
1803 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1804 TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1805 pmd->name, pmd->remote_iface);
1808 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1810 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1811 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1812 TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1813 pmd->name, remote_iface);
1818 * Flush usually returns negative value because it tries to
1819 * delete every QDISC (and on a running device, one QDISC at
1820 * least is needed). Ignore negative return value.
1822 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1823 if (qdisc_create_ingress(pmd->nlsk_fd,
1824 pmd->remote_if_index) < 0) {
1825 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1829 LIST_INIT(&pmd->implicit_flows);
1830 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1831 tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1832 tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1833 tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1835 "%s: failed to create implicit rules.",
1841 rte_eth_dev_probing_finish(dev);
1845 TAP_LOG(ERR, " Disabling rte flow support: %s(%d)",
1846 strerror(errno), errno);
1847 if (strlen(remote_iface)) {
1848 TAP_LOG(ERR, "Remote feature requires flow support.");
1851 rte_eth_dev_probing_finish(dev);
1855 TAP_LOG(ERR, " Can't set up remote feature: %s(%d)",
1856 strerror(errno), errno);
1857 tap_flow_implicit_flush(pmd, NULL);
1860 if (pmd->ioctl_sock > 0)
1861 close(pmd->ioctl_sock);
1862 /* mac_addrs must not be freed alone because part of dev_private */
1863 dev->data->mac_addrs = NULL;
1864 rte_eth_dev_release_port(dev);
1867 TAP_LOG(ERR, "%s Unable to initialize %s",
1868 tuntap_name, rte_vdev_device_name(vdev));
1874 set_interface_name(const char *key __rte_unused,
1878 char *name = (char *)extra_args;
1881 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN - 1);
1883 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1884 DEFAULT_TAP_NAME, (tap_unit - 1));
1890 set_remote_iface(const char *key __rte_unused,
1894 char *name = (char *)extra_args;
1897 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
1902 static int parse_user_mac(struct ether_addr *user_mac,
1905 unsigned int index = 0;
1906 char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
1908 if (user_mac == NULL || value == NULL)
1911 strlcpy(mac_temp, value, sizeof(mac_temp));
1912 mac_byte = strtok(mac_temp, ":");
1914 while ((mac_byte != NULL) &&
1915 (strlen(mac_byte) <= 2) &&
1916 (strlen(mac_byte) == strspn(mac_byte,
1917 ETH_TAP_CMP_MAC_FMT))) {
1918 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
1919 mac_byte = strtok(NULL, ":");
1926 set_mac_type(const char *key __rte_unused,
1930 struct ether_addr *user_mac = extra_args;
1935 if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
1936 static int iface_idx;
1938 /* fixed mac = 00:64:74:61:70:<iface_idx> */
1939 memcpy((char *)user_mac->addr_bytes, "\0dtap", ETHER_ADDR_LEN);
1940 user_mac->addr_bytes[ETHER_ADDR_LEN - 1] = iface_idx++ + '0';
1944 if (parse_user_mac(user_mac, value) != 6)
1947 TAP_LOG(DEBUG, "TAP user MAC param (%s)", value);
1951 TAP_LOG(ERR, "TAP user MAC (%s) is not in format (%s|%s)",
1952 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
1957 * Open a TUN interface device. TUN PMD
1958 * 1) sets tap_type as false
1959 * 2) intakes iface as argument.
1960 * 3) as interface is virtual set speed to 10G
1963 rte_pmd_tun_probe(struct rte_vdev_device *dev)
1965 const char *name, *params;
1967 struct rte_kvargs *kvlist = NULL;
1968 char tun_name[RTE_ETH_NAME_MAX_LEN];
1969 char remote_iface[RTE_ETH_NAME_MAX_LEN];
1970 struct rte_eth_dev *eth_dev;
1972 strcpy(tuntap_name, "TUN");
1974 name = rte_vdev_device_name(dev);
1975 params = rte_vdev_device_args(dev);
1976 memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1978 if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
1979 strlen(params) == 0) {
1980 eth_dev = rte_eth_dev_attach_secondary(name);
1982 TAP_LOG(ERR, "Failed to probe %s", name);
1985 eth_dev->dev_ops = &ops;
1986 eth_dev->device = &dev->device;
1987 rte_eth_dev_probing_finish(eth_dev);
1991 snprintf(tun_name, sizeof(tun_name), "%s%u",
1992 DEFAULT_TUN_NAME, tun_unit++);
1994 if (params && (params[0] != '\0')) {
1995 TAP_LOG(DEBUG, "parameters (%s)", params);
1997 kvlist = rte_kvargs_parse(params, valid_arguments);
1999 if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2000 ret = rte_kvargs_process(kvlist,
2002 &set_interface_name,
2010 pmd_link.link_speed = ETH_SPEED_NUM_10G;
2012 TAP_LOG(NOTICE, "Initializing pmd_tun for %s as %s",
2015 ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0,
2016 ETH_TUNTAP_TYPE_TUN);
2020 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2022 tun_unit--; /* Restore the unit number */
2024 rte_kvargs_free(kvlist);
2029 /* Request queue file descriptors from secondary to primary. */
2031 tap_mp_attach_queues(const char *port_name, struct rte_eth_dev *dev)
2034 struct timespec timeout = {.tv_sec = 1, .tv_nsec = 0};
2035 struct rte_mp_msg request, *reply;
2036 struct rte_mp_reply replies;
2037 struct ipc_queues *request_param = (struct ipc_queues *)request.param;
2038 struct ipc_queues *reply_param;
2039 struct pmd_process_private *process_private = dev->process_private;
2040 int queue, fd_iterator;
2042 /* Prepare the request */
2043 strlcpy(request.name, TAP_MP_KEY, sizeof(request.name));
2044 strlcpy(request_param->port_name, port_name,
2045 sizeof(request_param->port_name));
2046 request.len_param = sizeof(*request_param);
2047 /* Send request and receive reply */
2048 ret = rte_mp_request_sync(&request, &replies, &timeout);
2050 TAP_LOG(ERR, "Failed to request queues from primary: %d",
2054 reply = &replies.msgs[0];
2055 reply_param = (struct ipc_queues *)reply->param;
2056 TAP_LOG(DEBUG, "Received IPC reply for %s", reply_param->port_name);
2058 /* Attach the queues from received file descriptors */
2059 dev->data->nb_rx_queues = reply_param->rxq_count;
2060 dev->data->nb_tx_queues = reply_param->txq_count;
2062 for (queue = 0; queue < reply_param->rxq_count; queue++)
2063 process_private->rxq_fds[queue] = reply->fds[fd_iterator++];
2064 for (queue = 0; queue < reply_param->txq_count; queue++)
2065 process_private->txq_fds[queue] = reply->fds[fd_iterator++];
2070 /* Send the queue file descriptors from the primary process to secondary. */
2072 tap_mp_sync_queues(const struct rte_mp_msg *request, const void *peer)
2074 struct rte_eth_dev *dev;
2075 struct pmd_process_private *process_private;
2076 struct rte_mp_msg reply;
2077 const struct ipc_queues *request_param =
2078 (const struct ipc_queues *)request->param;
2079 struct ipc_queues *reply_param =
2080 (struct ipc_queues *)reply.param;
2085 /* Get requested port */
2086 TAP_LOG(DEBUG, "Received IPC request for %s", request_param->port_name);
2087 ret = rte_eth_dev_get_port_by_name(request_param->port_name, &port_id);
2089 TAP_LOG(ERR, "Failed to get port id for %s",
2090 request_param->port_name);
2093 dev = &rte_eth_devices[port_id];
2094 process_private = dev->process_private;
2096 /* Fill file descriptors for all queues */
2098 reply_param->rxq_count = 0;
2099 for (queue = 0; queue < dev->data->nb_rx_queues; queue++) {
2100 reply.fds[reply.num_fds++] = process_private->rxq_fds[queue];
2101 reply_param->rxq_count++;
2103 RTE_ASSERT(reply_param->rxq_count == dev->data->nb_rx_queues);
2104 RTE_ASSERT(reply_param->txq_count == dev->data->nb_tx_queues);
2105 RTE_ASSERT(reply.num_fds <= RTE_MP_MAX_FD_NUM);
2107 reply_param->txq_count = 0;
2108 for (queue = 0; queue < dev->data->nb_tx_queues; queue++) {
2109 reply.fds[reply.num_fds++] = process_private->txq_fds[queue];
2110 reply_param->txq_count++;
2114 strlcpy(reply.name, request->name, sizeof(reply.name));
2115 strlcpy(reply_param->port_name, request_param->port_name,
2116 sizeof(reply_param->port_name));
2117 reply.len_param = sizeof(*reply_param);
2118 if (rte_mp_reply(&reply, peer) < 0) {
2119 TAP_LOG(ERR, "Failed to reply an IPC request to sync queues");
2125 /* Open a TAP interface device.
2128 rte_pmd_tap_probe(struct rte_vdev_device *dev)
2130 const char *name, *params;
2132 struct rte_kvargs *kvlist = NULL;
2134 char tap_name[RTE_ETH_NAME_MAX_LEN];
2135 char remote_iface[RTE_ETH_NAME_MAX_LEN];
2136 struct ether_addr user_mac = { .addr_bytes = {0} };
2137 struct rte_eth_dev *eth_dev;
2138 int tap_devices_count_increased = 0;
2140 strcpy(tuntap_name, "TAP");
2142 name = rte_vdev_device_name(dev);
2143 params = rte_vdev_device_args(dev);
2145 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
2146 eth_dev = rte_eth_dev_attach_secondary(name);
2148 TAP_LOG(ERR, "Failed to probe %s", name);
2151 eth_dev->dev_ops = &ops;
2152 eth_dev->device = &dev->device;
2153 eth_dev->rx_pkt_burst = pmd_rx_burst;
2154 eth_dev->tx_pkt_burst = pmd_tx_burst;
2155 if (!rte_eal_primary_proc_alive(NULL)) {
2156 TAP_LOG(ERR, "Primary process is missing");
2159 eth_dev->process_private = (struct pmd_process_private *)
2160 rte_zmalloc_socket(name,
2161 sizeof(struct pmd_process_private),
2162 RTE_CACHE_LINE_SIZE,
2163 eth_dev->device->numa_node);
2164 if (eth_dev->process_private == NULL) {
2166 "Failed to alloc memory for process private");
2170 ret = tap_mp_attach_queues(name, eth_dev);
2173 rte_eth_dev_probing_finish(eth_dev);
2177 speed = ETH_SPEED_NUM_10G;
2178 snprintf(tap_name, sizeof(tap_name), "%s%u",
2179 DEFAULT_TAP_NAME, tap_unit++);
2180 memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2182 if (params && (params[0] != '\0')) {
2183 TAP_LOG(DEBUG, "parameters (%s)", params);
2185 kvlist = rte_kvargs_parse(params, valid_arguments);
2187 if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2188 ret = rte_kvargs_process(kvlist,
2190 &set_interface_name,
2196 if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
2197 ret = rte_kvargs_process(kvlist,
2205 if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
2206 ret = rte_kvargs_process(kvlist,
2215 pmd_link.link_speed = speed;
2217 TAP_LOG(NOTICE, "Initializing pmd_tap for %s as %s",
2220 /* Register IPC feed callback */
2221 if (!tap_devices_count) {
2222 ret = rte_mp_action_register(TAP_MP_KEY, tap_mp_sync_queues);
2224 TAP_LOG(ERR, "%s: Failed to register IPC callback: %s",
2225 tuntap_name, strerror(rte_errno));
2229 tap_devices_count++;
2230 tap_devices_count_increased = 1;
2231 ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac,
2232 ETH_TUNTAP_TYPE_TAP);
2236 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2238 if (tap_devices_count_increased == 1) {
2239 if (tap_devices_count == 1)
2240 rte_mp_action_unregister(TAP_MP_KEY);
2241 tap_devices_count--;
2243 tap_unit--; /* Restore the unit number */
2245 rte_kvargs_free(kvlist);
2250 /* detach a TUNTAP device.
2253 rte_pmd_tap_remove(struct rte_vdev_device *dev)
2255 struct rte_eth_dev *eth_dev = NULL;
2256 struct pmd_internals *internals;
2257 struct pmd_process_private *process_private;
2260 /* find the ethdev entry */
2261 eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
2265 /* mac_addrs must not be freed alone because part of dev_private */
2266 eth_dev->data->mac_addrs = NULL;
2268 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2269 return rte_eth_dev_release_port(eth_dev);
2271 internals = eth_dev->data->dev_private;
2272 process_private = eth_dev->process_private;
2274 TAP_LOG(DEBUG, "Closing %s Ethernet device on numa %u",
2275 (internals->type == ETH_TUNTAP_TYPE_TAP) ? "TAP" : "TUN",
2278 if (internals->nlsk_fd) {
2279 tap_flow_flush(eth_dev, NULL);
2280 tap_flow_implicit_flush(internals, NULL);
2281 tap_nl_final(internals->nlsk_fd);
2283 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
2284 if (process_private->rxq_fds[i] != -1) {
2285 close(process_private->rxq_fds[i]);
2286 process_private->rxq_fds[i] = -1;
2288 if (process_private->txq_fds[i] != -1) {
2289 close(process_private->txq_fds[i]);
2290 process_private->txq_fds[i] = -1;
2294 close(internals->ioctl_sock);
2295 rte_free(eth_dev->process_private);
2296 if (tap_devices_count == 1)
2297 rte_mp_action_unregister(TAP_MP_KEY);
2298 tap_devices_count--;
2299 rte_eth_dev_release_port(eth_dev);
2301 if (internals->ka_fd != -1) {
2302 close(internals->ka_fd);
2303 internals->ka_fd = -1;
2308 static struct rte_vdev_driver pmd_tun_drv = {
2309 .probe = rte_pmd_tun_probe,
2310 .remove = rte_pmd_tap_remove,
2313 static struct rte_vdev_driver pmd_tap_drv = {
2314 .probe = rte_pmd_tap_probe,
2315 .remove = rte_pmd_tap_remove,
2318 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
2319 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
2320 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
2321 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
2322 ETH_TAP_IFACE_ARG "=<string> ");
2323 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
2324 ETH_TAP_IFACE_ARG "=<string> "
2325 ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
2326 ETH_TAP_REMOTE_ARG "=<string>");
2329 RTE_INIT(tap_init_log)
2331 tap_logtype = rte_log_register("pmd.net.tap");
2332 if (tap_logtype >= 0)
2333 rte_log_set_level(tap_logtype, RTE_LOG_NOTICE);