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;
74 static struct rte_vdev_driver pmd_tap_drv;
75 static struct rte_vdev_driver pmd_tun_drv;
77 static const char *valid_arguments[] = {
84 static volatile uint32_t tap_trigger; /* Rx trigger */
86 static struct rte_eth_link pmd_link = {
87 .link_speed = ETH_SPEED_NUM_10G,
88 .link_duplex = ETH_LINK_FULL_DUPLEX,
89 .link_status = ETH_LINK_DOWN,
90 .link_autoneg = ETH_LINK_FIXED,
94 tap_trigger_cb(int sig __rte_unused)
96 /* Valid trigger values are nonzero */
97 tap_trigger = (tap_trigger + 1) | 0x80000000;
100 /* Specifies on what netdevices the ioctl should be applied */
107 /* Message header to synchronize queues via IPC */
109 char port_name[RTE_DEV_NAME_MAX_LEN];
113 * The file descriptors are in the dedicated part
114 * of the Unix message to be translated by the kernel.
118 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
121 * Tun/Tap allocation routine
124 * Pointer to private structure.
126 * @param[in] is_keepalive
130 * -1 on failure, fd on success
133 tun_alloc(struct pmd_internals *pmd, int is_keepalive)
136 #ifdef IFF_MULTI_QUEUE
137 unsigned int features;
141 memset(&ifr, 0, sizeof(struct ifreq));
144 * Do not set IFF_NO_PI as packet information header will be needed
145 * to check if a received packet has been truncated.
147 ifr.ifr_flags = (pmd->type == ETH_TUNTAP_TYPE_TAP) ?
148 IFF_TAP : IFF_TUN | IFF_POINTOPOINT;
149 strlcpy(ifr.ifr_name, pmd->name, IFNAMSIZ);
151 fd = open(TUN_TAP_DEV_PATH, O_RDWR);
153 TAP_LOG(ERR, "Unable to open %s interface", TUN_TAP_DEV_PATH);
157 #ifdef IFF_MULTI_QUEUE
158 /* Grab the TUN features to verify we can work multi-queue */
159 if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
160 TAP_LOG(ERR, "unable to get TUN/TAP features");
163 TAP_LOG(DEBUG, "%s Features %08x", TUN_TAP_DEV_PATH, features);
165 if (features & IFF_MULTI_QUEUE) {
166 TAP_LOG(DEBUG, " Multi-queue support for %d queues",
167 RTE_PMD_TAP_MAX_QUEUES);
168 ifr.ifr_flags |= IFF_MULTI_QUEUE;
172 ifr.ifr_flags |= IFF_ONE_QUEUE;
173 TAP_LOG(DEBUG, " Single queue only support");
176 /* Set the TUN/TAP configuration and set the name if needed */
177 if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
178 TAP_LOG(WARNING, "Unable to set TUNSETIFF for %s: %s",
179 ifr.ifr_name, strerror(errno));
184 * Name passed to kernel might be wildcard like dtun%d
185 * and need to find the resulting device.
187 TAP_LOG(DEBUG, "Device name is '%s'", ifr.ifr_name);
188 strlcpy(pmd->name, ifr.ifr_name, RTE_ETH_NAME_MAX_LEN);
192 * Detach the TUN/TAP keep-alive queue
193 * to avoid traffic through it
195 ifr.ifr_flags = IFF_DETACH_QUEUE;
196 if (ioctl(fd, TUNSETQUEUE, (void *)&ifr) < 0) {
198 "Unable to detach keep-alive queue for %s: %s",
199 ifr.ifr_name, strerror(errno));
204 /* Always set the file descriptor to non-blocking */
205 if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
207 "Unable to set %s to nonblocking: %s",
208 ifr.ifr_name, strerror(errno));
212 /* Set up trigger to optimize empty Rx bursts */
216 int flags = fcntl(fd, F_GETFL);
218 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
220 if (sa.sa_handler != tap_trigger_cb) {
222 * Make sure SIGIO is not already taken. This is done
223 * as late as possible to leave the application a
224 * chance to set up its own signal handler first.
226 if (sa.sa_handler != SIG_IGN &&
227 sa.sa_handler != SIG_DFL) {
231 sa = (struct sigaction){
232 .sa_flags = SA_RESTART,
233 .sa_handler = tap_trigger_cb,
235 if (sigaction(SIGIO, &sa, NULL) == -1)
238 /* Enable SIGIO on file descriptor */
239 fcntl(fd, F_SETFL, flags | O_ASYNC);
240 fcntl(fd, F_SETOWN, getpid());
244 /* Disable trigger globally in case of error */
246 TAP_LOG(WARNING, "Rx trigger disabled: %s",
259 tap_verify_csum(struct rte_mbuf *mbuf)
261 uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
262 uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
263 uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
264 unsigned int l2_len = sizeof(struct rte_ether_hdr);
270 if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
272 else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
274 /* Don't verify checksum for packets with discontinuous L2 header */
275 if (unlikely(l2_len + sizeof(struct rte_ipv4_hdr) >
276 rte_pktmbuf_data_len(mbuf)))
278 l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
279 if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
280 struct rte_ipv4_hdr *iph = l3_hdr;
282 /* ihl contains the number of 4-byte words in the header */
283 l3_len = 4 * (iph->version_ihl & 0xf);
284 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
286 /* check that the total length reported by header is not
287 * greater than the total received size
289 if (l2_len + rte_be_to_cpu_16(iph->total_length) >
290 rte_pktmbuf_data_len(mbuf))
293 cksum = ~rte_raw_cksum(iph, l3_len);
294 mbuf->ol_flags |= cksum ?
295 PKT_RX_IP_CKSUM_BAD :
296 PKT_RX_IP_CKSUM_GOOD;
297 } else if (l3 == RTE_PTYPE_L3_IPV6) {
298 struct rte_ipv6_hdr *iph = l3_hdr;
300 l3_len = sizeof(struct rte_ipv6_hdr);
301 /* check that the total length reported by header is not
302 * greater than the total received size
304 if (l2_len + l3_len + rte_be_to_cpu_16(iph->payload_len) >
305 rte_pktmbuf_data_len(mbuf))
308 /* IPv6 extensions are not supported */
311 if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
312 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
313 /* Don't verify checksum for multi-segment packets. */
314 if (mbuf->nb_segs > 1)
316 if (l3 == RTE_PTYPE_L3_IPV4)
317 cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
318 else if (l3 == RTE_PTYPE_L3_IPV6)
319 cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
320 mbuf->ol_flags |= cksum ?
321 PKT_RX_L4_CKSUM_BAD :
322 PKT_RX_L4_CKSUM_GOOD;
327 tap_rx_offload_get_port_capa(void)
330 * No specific port Rx offload capabilities.
336 tap_rx_offload_get_queue_capa(void)
338 return DEV_RX_OFFLOAD_SCATTER |
339 DEV_RX_OFFLOAD_IPV4_CKSUM |
340 DEV_RX_OFFLOAD_UDP_CKSUM |
341 DEV_RX_OFFLOAD_TCP_CKSUM;
344 /* Callback to handle the rx burst of packets to the correct interface and
345 * file descriptor(s) in a multi-queue setup.
348 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
350 struct rx_queue *rxq = queue;
351 struct pmd_process_private *process_private;
353 unsigned long num_rx_bytes = 0;
354 uint32_t trigger = tap_trigger;
356 if (trigger == rxq->trigger_seen)
359 rxq->trigger_seen = trigger;
360 process_private = rte_eth_devices[rxq->in_port].process_private;
361 rte_compiler_barrier();
362 for (num_rx = 0; num_rx < nb_pkts; ) {
363 struct rte_mbuf *mbuf = rxq->pool;
364 struct rte_mbuf *seg = NULL;
365 struct rte_mbuf *new_tail = NULL;
366 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
369 len = readv(process_private->rxq_fds[rxq->queue_id],
371 1 + (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
372 rxq->nb_rx_desc : 1));
373 if (len < (int)sizeof(struct tun_pi))
376 /* Packet couldn't fit in the provided mbuf */
377 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
378 rxq->stats.ierrors++;
382 len -= sizeof(struct tun_pi);
385 mbuf->port = rxq->in_port;
387 struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
389 if (unlikely(!buf)) {
390 rxq->stats.rx_nombuf++;
391 /* No new buf has been allocated: do nothing */
392 if (!new_tail || !seg)
396 rte_pktmbuf_free(mbuf);
400 seg = seg ? seg->next : mbuf;
401 if (rxq->pool == mbuf)
404 new_tail->next = buf;
406 new_tail->next = seg->next;
408 /* iovecs[0] is reserved for packet info (pi) */
409 (*rxq->iovecs)[mbuf->nb_segs].iov_len =
410 buf->buf_len - data_off;
411 (*rxq->iovecs)[mbuf->nb_segs].iov_base =
412 (char *)buf->buf_addr + data_off;
414 seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
415 seg->data_off = data_off;
417 len -= seg->data_len;
421 /* First segment has headroom, not the others */
425 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
427 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
428 tap_verify_csum(mbuf);
430 /* account for the receive frame */
431 bufs[num_rx++] = mbuf;
432 num_rx_bytes += mbuf->pkt_len;
435 rxq->stats.ipackets += num_rx;
436 rxq->stats.ibytes += num_rx_bytes;
442 tap_tx_offload_get_port_capa(void)
445 * No specific port Tx offload capabilities.
451 tap_tx_offload_get_queue_capa(void)
453 return DEV_TX_OFFLOAD_MULTI_SEGS |
454 DEV_TX_OFFLOAD_IPV4_CKSUM |
455 DEV_TX_OFFLOAD_UDP_CKSUM |
456 DEV_TX_OFFLOAD_TCP_CKSUM |
457 DEV_TX_OFFLOAD_TCP_TSO;
460 /* Finalize l4 checksum calculation */
462 tap_tx_l4_cksum(uint16_t *l4_cksum, uint16_t l4_phdr_cksum,
463 uint32_t l4_raw_cksum)
468 cksum = __rte_raw_cksum_reduce(l4_raw_cksum);
469 cksum += l4_phdr_cksum;
471 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
472 cksum = (~cksum) & 0xffff;
479 /* Accumaulate L4 raw checksums */
481 tap_tx_l4_add_rcksum(char *l4_data, unsigned int l4_len, uint16_t *l4_cksum,
482 uint32_t *l4_raw_cksum)
484 if (l4_cksum == NULL)
487 *l4_raw_cksum = __rte_raw_cksum(l4_data, l4_len, *l4_raw_cksum);
490 /* L3 and L4 pseudo headers checksum offloads */
492 tap_tx_l3_cksum(char *packet, uint64_t ol_flags, unsigned int l2_len,
493 unsigned int l3_len, unsigned int l4_len, uint16_t **l4_cksum,
494 uint16_t *l4_phdr_cksum, uint32_t *l4_raw_cksum)
496 void *l3_hdr = packet + l2_len;
498 if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
499 struct rte_ipv4_hdr *iph = l3_hdr;
502 iph->hdr_checksum = 0;
503 cksum = rte_raw_cksum(iph, l3_len);
504 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
506 if (ol_flags & PKT_TX_L4_MASK) {
509 l4_hdr = packet + l2_len + l3_len;
510 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
511 *l4_cksum = &((struct rte_udp_hdr *)l4_hdr)->dgram_cksum;
512 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
513 *l4_cksum = &((struct rte_tcp_hdr *)l4_hdr)->cksum;
517 if (ol_flags & PKT_TX_IPV4)
518 *l4_phdr_cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
520 *l4_phdr_cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
521 *l4_raw_cksum = __rte_raw_cksum(l4_hdr, l4_len, 0);
526 tap_write_mbufs(struct tx_queue *txq, uint16_t num_mbufs,
527 struct rte_mbuf **pmbufs,
528 uint16_t *num_packets, unsigned long *num_tx_bytes)
532 struct pmd_process_private *process_private;
534 process_private = rte_eth_devices[txq->out_port].process_private;
536 for (i = 0; i < num_mbufs; i++) {
537 struct rte_mbuf *mbuf = pmbufs[i];
538 struct iovec iovecs[mbuf->nb_segs + 2];
539 struct tun_pi pi = { .flags = 0, .proto = 0x00 };
540 struct rte_mbuf *seg = mbuf;
541 char m_copy[mbuf->data_len];
545 int k; /* current index in iovecs for copying segments */
546 uint16_t seg_len; /* length of first segment */
548 uint16_t *l4_cksum; /* l4 checksum (pseudo header + payload) */
549 uint32_t l4_raw_cksum = 0; /* TCP/UDP payload raw checksum */
550 uint16_t l4_phdr_cksum = 0; /* TCP/UDP pseudo header checksum */
551 uint16_t is_cksum = 0; /* in case cksum should be offloaded */
554 if (txq->type == ETH_TUNTAP_TYPE_TUN) {
556 * TUN and TAP are created with IFF_NO_PI disabled.
557 * For TUN PMD this mandatory as fields are used by
558 * Kernel tun.c to determine whether its IP or non IP
561 * The logic fetches the first byte of data from mbuf
562 * then compares whether its v4 or v6. If first byte
563 * is 4 or 6, then protocol field is updated.
565 char *buff_data = rte_pktmbuf_mtod(seg, void *);
566 proto = (*buff_data & 0xf0);
567 pi.proto = (proto == 0x40) ?
568 rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4) :
570 rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6) :
575 iovecs[k].iov_base = π
576 iovecs[k].iov_len = sizeof(pi);
579 nb_segs = mbuf->nb_segs;
581 ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
582 (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
583 (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
586 /* Support only packets with at least layer 4
587 * header included in the first segment
589 seg_len = rte_pktmbuf_data_len(mbuf);
590 l234_hlen = mbuf->l2_len + mbuf->l3_len + mbuf->l4_len;
591 if (seg_len < l234_hlen)
594 /* To change checksums, work on a * copy of l2, l3
595 * headers + l4 pseudo header
597 rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
599 tap_tx_l3_cksum(m_copy, mbuf->ol_flags,
600 mbuf->l2_len, mbuf->l3_len, mbuf->l4_len,
601 &l4_cksum, &l4_phdr_cksum,
603 iovecs[k].iov_base = m_copy;
604 iovecs[k].iov_len = l234_hlen;
607 /* Update next iovecs[] beyond l2, l3, l4 headers */
608 if (seg_len > l234_hlen) {
609 iovecs[k].iov_len = seg_len - l234_hlen;
611 rte_pktmbuf_mtod(seg, char *) +
613 tap_tx_l4_add_rcksum(iovecs[k].iov_base,
614 iovecs[k].iov_len, l4_cksum,
622 for (j = k; j <= nb_segs; j++) {
623 iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
624 iovecs[j].iov_base = rte_pktmbuf_mtod(seg, void *);
626 tap_tx_l4_add_rcksum(iovecs[j].iov_base,
627 iovecs[j].iov_len, l4_cksum,
633 tap_tx_l4_cksum(l4_cksum, l4_phdr_cksum, l4_raw_cksum);
635 /* copy the tx frame data */
636 n = writev(process_private->txq_fds[txq->queue_id], iovecs, j);
640 (*num_tx_bytes) += rte_pktmbuf_pkt_len(mbuf);
644 /* Callback to handle sending packets from the tap interface
647 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
649 struct tx_queue *txq = queue;
651 uint16_t num_packets = 0;
652 unsigned long num_tx_bytes = 0;
656 if (unlikely(nb_pkts == 0))
659 struct rte_mbuf *gso_mbufs[MAX_GSO_MBUFS];
660 max_size = *txq->mtu + (RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN + 4);
661 for (i = 0; i < nb_pkts; i++) {
662 struct rte_mbuf *mbuf_in = bufs[num_tx];
663 struct rte_mbuf **mbuf;
664 uint16_t num_mbufs = 0;
665 uint16_t tso_segsz = 0;
671 tso = mbuf_in->ol_flags & PKT_TX_TCP_SEG;
673 struct rte_gso_ctx *gso_ctx = &txq->gso_ctx;
675 assert(gso_ctx != NULL);
677 /* TCP segmentation implies TCP checksum offload */
678 mbuf_in->ol_flags |= PKT_TX_TCP_CKSUM;
680 /* gso size is calculated without RTE_ETHER_CRC_LEN */
681 hdrs_len = mbuf_in->l2_len + mbuf_in->l3_len +
683 tso_segsz = mbuf_in->tso_segsz + hdrs_len;
684 if (unlikely(tso_segsz == hdrs_len) ||
685 tso_segsz > *txq->mtu) {
689 gso_ctx->gso_size = tso_segsz;
690 ret = rte_gso_segment(mbuf_in, /* packet to segment */
691 gso_ctx, /* gso control block */
692 (struct rte_mbuf **)&gso_mbufs, /* out mbufs */
693 RTE_DIM(gso_mbufs)); /* max tso mbufs */
695 /* ret contains the number of new created mbufs */
702 /* stats.errs will be incremented */
703 if (rte_pktmbuf_pkt_len(mbuf_in) > max_size)
706 /* ret 0 indicates no new mbufs were created */
712 tap_write_mbufs(txq, num_mbufs, mbuf,
713 &num_packets, &num_tx_bytes);
715 /* free original mbuf */
716 rte_pktmbuf_free(mbuf_in);
718 for (j = 0; j < ret; j++)
719 rte_pktmbuf_free(mbuf[j]);
722 txq->stats.opackets += num_packets;
723 txq->stats.errs += nb_pkts - num_tx;
724 txq->stats.obytes += num_tx_bytes;
730 tap_ioctl_req2str(unsigned long request)
734 return "SIOCSIFFLAGS";
736 return "SIOCGIFFLAGS";
738 return "SIOCGIFHWADDR";
740 return "SIOCSIFHWADDR";
748 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
749 struct ifreq *ifr, int set, enum ioctl_mode mode)
751 short req_flags = ifr->ifr_flags;
752 int remote = pmd->remote_if_index &&
753 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
755 if (!pmd->remote_if_index && mode == REMOTE_ONLY)
758 * If there is a remote netdevice, apply ioctl on it, then apply it on
763 strlcpy(ifr->ifr_name, pmd->remote_iface, IFNAMSIZ);
764 else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
765 strlcpy(ifr->ifr_name, pmd->name, IFNAMSIZ);
768 /* fetch current flags to leave other flags untouched */
769 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
772 ifr->ifr_flags |= req_flags;
774 ifr->ifr_flags &= ~req_flags;
782 RTE_LOG(WARNING, PMD, "%s: ioctl() called with wrong arg\n",
786 if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
788 if (remote-- && mode == LOCAL_AND_REMOTE)
793 TAP_LOG(DEBUG, "%s(%s) failed: %s(%d)", ifr->ifr_name,
794 tap_ioctl_req2str(request), strerror(errno), errno);
799 tap_link_set_down(struct rte_eth_dev *dev)
801 struct pmd_internals *pmd = dev->data->dev_private;
802 struct ifreq ifr = { .ifr_flags = IFF_UP };
804 dev->data->dev_link.link_status = ETH_LINK_DOWN;
805 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
809 tap_link_set_up(struct rte_eth_dev *dev)
811 struct pmd_internals *pmd = dev->data->dev_private;
812 struct ifreq ifr = { .ifr_flags = IFF_UP };
814 dev->data->dev_link.link_status = ETH_LINK_UP;
815 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
819 tap_dev_start(struct rte_eth_dev *dev)
823 err = tap_intr_handle_set(dev, 1);
827 err = tap_link_set_up(dev);
831 for (i = 0; i < dev->data->nb_tx_queues; i++)
832 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
833 for (i = 0; i < dev->data->nb_rx_queues; i++)
834 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
839 /* This function gets called when the current port gets stopped.
842 tap_dev_stop(struct rte_eth_dev *dev)
846 for (i = 0; i < dev->data->nb_tx_queues; i++)
847 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
848 for (i = 0; i < dev->data->nb_rx_queues; i++)
849 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
851 tap_intr_handle_set(dev, 0);
852 tap_link_set_down(dev);
856 tap_dev_configure(struct rte_eth_dev *dev)
858 struct pmd_internals *pmd = dev->data->dev_private;
860 if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
862 "%s: number of rx queues %d exceeds max num of queues %d",
864 dev->data->nb_rx_queues,
865 RTE_PMD_TAP_MAX_QUEUES);
868 if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
870 "%s: number of tx queues %d exceeds max num of queues %d",
872 dev->data->nb_tx_queues,
873 RTE_PMD_TAP_MAX_QUEUES);
877 TAP_LOG(INFO, "%s: %s: TX configured queues number: %u",
878 dev->device->name, pmd->name, dev->data->nb_tx_queues);
880 TAP_LOG(INFO, "%s: %s: RX configured queues number: %u",
881 dev->device->name, pmd->name, dev->data->nb_rx_queues);
887 tap_dev_speed_capa(void)
889 uint32_t speed = pmd_link.link_speed;
892 if (speed >= ETH_SPEED_NUM_10M)
893 capa |= ETH_LINK_SPEED_10M;
894 if (speed >= ETH_SPEED_NUM_100M)
895 capa |= ETH_LINK_SPEED_100M;
896 if (speed >= ETH_SPEED_NUM_1G)
897 capa |= ETH_LINK_SPEED_1G;
898 if (speed >= ETH_SPEED_NUM_5G)
899 capa |= ETH_LINK_SPEED_2_5G;
900 if (speed >= ETH_SPEED_NUM_5G)
901 capa |= ETH_LINK_SPEED_5G;
902 if (speed >= ETH_SPEED_NUM_10G)
903 capa |= ETH_LINK_SPEED_10G;
904 if (speed >= ETH_SPEED_NUM_20G)
905 capa |= ETH_LINK_SPEED_20G;
906 if (speed >= ETH_SPEED_NUM_25G)
907 capa |= ETH_LINK_SPEED_25G;
908 if (speed >= ETH_SPEED_NUM_40G)
909 capa |= ETH_LINK_SPEED_40G;
910 if (speed >= ETH_SPEED_NUM_50G)
911 capa |= ETH_LINK_SPEED_50G;
912 if (speed >= ETH_SPEED_NUM_56G)
913 capa |= ETH_LINK_SPEED_56G;
914 if (speed >= ETH_SPEED_NUM_100G)
915 capa |= ETH_LINK_SPEED_100G;
921 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
923 struct pmd_internals *internals = dev->data->dev_private;
925 dev_info->if_index = internals->if_index;
926 dev_info->max_mac_addrs = 1;
927 dev_info->max_rx_pktlen = (uint32_t)RTE_ETHER_MAX_VLAN_FRAME_LEN;
928 dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
929 dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
930 dev_info->min_rx_bufsize = 0;
931 dev_info->speed_capa = tap_dev_speed_capa();
932 dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
933 dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
934 dev_info->rx_queue_offload_capa;
935 dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
936 dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
937 dev_info->tx_queue_offload_capa;
938 dev_info->hash_key_size = TAP_RSS_HASH_KEY_SIZE;
940 * limitation: TAP supports all of IP, UDP and TCP hash
941 * functions together and not in partial combinations
943 dev_info->flow_type_rss_offloads = ~TAP_RSS_HF_MASK;
947 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
949 unsigned int i, imax;
950 unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
951 unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
952 unsigned long rx_nombuf = 0, ierrors = 0;
953 const struct pmd_internals *pmd = dev->data->dev_private;
955 /* rx queue statistics */
956 imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
957 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
958 for (i = 0; i < imax; i++) {
959 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
960 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
961 rx_total += tap_stats->q_ipackets[i];
962 rx_bytes_total += tap_stats->q_ibytes[i];
963 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
964 ierrors += pmd->rxq[i].stats.ierrors;
967 /* tx queue statistics */
968 imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
969 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
971 for (i = 0; i < imax; i++) {
972 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
973 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
974 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
975 tx_total += tap_stats->q_opackets[i];
976 tx_err_total += tap_stats->q_errors[i];
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;
1009 tap_dev_close(struct rte_eth_dev *dev)
1012 struct pmd_internals *internals = dev->data->dev_private;
1013 struct pmd_process_private *process_private = dev->process_private;
1015 tap_link_set_down(dev);
1016 tap_flow_flush(dev, NULL);
1017 tap_flow_implicit_flush(internals, NULL);
1019 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1020 if (process_private->rxq_fds[i] != -1) {
1021 close(process_private->rxq_fds[i]);
1022 process_private->rxq_fds[i] = -1;
1024 if (process_private->txq_fds[i] != -1) {
1025 close(process_private->txq_fds[i]);
1026 process_private->txq_fds[i] = -1;
1030 if (internals->remote_if_index) {
1031 /* Restore initial remote state */
1032 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
1033 &internals->remote_initial_flags);
1036 if (internals->ka_fd != -1) {
1037 close(internals->ka_fd);
1038 internals->ka_fd = -1;
1041 * Since TUN device has no more opened file descriptors
1042 * it will be removed from kernel
1047 tap_rx_queue_release(void *queue)
1049 struct rx_queue *rxq = queue;
1050 struct pmd_process_private *process_private;
1054 process_private = rte_eth_devices[rxq->in_port].process_private;
1055 if (process_private->rxq_fds[rxq->queue_id] > 0) {
1056 close(process_private->rxq_fds[rxq->queue_id]);
1057 process_private->rxq_fds[rxq->queue_id] = -1;
1058 rte_pktmbuf_free(rxq->pool);
1059 rte_free(rxq->iovecs);
1066 tap_tx_queue_release(void *queue)
1068 struct tx_queue *txq = queue;
1069 struct pmd_process_private *process_private;
1073 process_private = rte_eth_devices[txq->out_port].process_private;
1075 if (process_private->txq_fds[txq->queue_id] > 0) {
1076 close(process_private->txq_fds[txq->queue_id]);
1077 process_private->txq_fds[txq->queue_id] = -1;
1082 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
1084 struct rte_eth_link *dev_link = &dev->data->dev_link;
1085 struct pmd_internals *pmd = dev->data->dev_private;
1086 struct ifreq ifr = { .ifr_flags = 0 };
1088 if (pmd->remote_if_index) {
1089 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
1090 if (!(ifr.ifr_flags & IFF_UP) ||
1091 !(ifr.ifr_flags & IFF_RUNNING)) {
1092 dev_link->link_status = ETH_LINK_DOWN;
1096 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
1097 dev_link->link_status =
1098 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
1105 tap_promisc_enable(struct rte_eth_dev *dev)
1107 struct pmd_internals *pmd = dev->data->dev_private;
1108 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1110 dev->data->promiscuous = 1;
1111 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1112 if (pmd->remote_if_index && !pmd->flow_isolate)
1113 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
1117 tap_promisc_disable(struct rte_eth_dev *dev)
1119 struct pmd_internals *pmd = dev->data->dev_private;
1120 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1122 dev->data->promiscuous = 0;
1123 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1124 if (pmd->remote_if_index && !pmd->flow_isolate)
1125 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
1129 tap_allmulti_enable(struct rte_eth_dev *dev)
1131 struct pmd_internals *pmd = dev->data->dev_private;
1132 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1134 dev->data->all_multicast = 1;
1135 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1136 if (pmd->remote_if_index && !pmd->flow_isolate)
1137 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
1141 tap_allmulti_disable(struct rte_eth_dev *dev)
1143 struct pmd_internals *pmd = dev->data->dev_private;
1144 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1146 dev->data->all_multicast = 0;
1147 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1148 if (pmd->remote_if_index && !pmd->flow_isolate)
1149 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
1153 tap_mac_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1155 struct pmd_internals *pmd = dev->data->dev_private;
1156 enum ioctl_mode mode = LOCAL_ONLY;
1160 if (pmd->type == ETH_TUNTAP_TYPE_TUN) {
1161 TAP_LOG(ERR, "%s: can't MAC address for TUN",
1166 if (rte_is_zero_ether_addr(mac_addr)) {
1167 TAP_LOG(ERR, "%s: can't set an empty MAC address",
1171 /* Check the actual current MAC address on the tap netdevice */
1172 ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY);
1175 if (rte_is_same_ether_addr(
1176 (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1179 /* Check the current MAC address on the remote */
1180 ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY);
1183 if (!rte_is_same_ether_addr(
1184 (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1186 mode = LOCAL_AND_REMOTE;
1187 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1188 rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, RTE_ETHER_ADDR_LEN);
1189 ret = tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode);
1192 rte_memcpy(&pmd->eth_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1193 if (pmd->remote_if_index && !pmd->flow_isolate) {
1194 /* Replace MAC redirection rule after a MAC change */
1195 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC);
1198 "%s: Couldn't delete MAC redirection rule",
1202 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC);
1205 "%s: Couldn't add MAC redirection rule",
1215 tap_gso_ctx_setup(struct rte_gso_ctx *gso_ctx, struct rte_eth_dev *dev)
1221 * Create private mbuf pool with TAP_GSO_MBUF_SEG_SIZE bytes
1222 * size per mbuf use this pool for both direct and indirect mbufs
1225 struct rte_mempool *mp; /* Mempool for GSO packets */
1227 /* initialize GSO context */
1228 gso_types = DEV_TX_OFFLOAD_TCP_TSO;
1229 snprintf(pool_name, sizeof(pool_name), "mp_%s", dev->device->name);
1230 mp = rte_mempool_lookup((const char *)pool_name);
1232 mp = rte_pktmbuf_pool_create(pool_name, TAP_GSO_MBUFS_NUM,
1233 TAP_GSO_MBUF_CACHE_SIZE, 0,
1234 RTE_PKTMBUF_HEADROOM + TAP_GSO_MBUF_SEG_SIZE,
1237 struct pmd_internals *pmd = dev->data->dev_private;
1238 RTE_LOG(DEBUG, PMD, "%s: failed to create mbuf pool for device %s\n",
1239 pmd->name, dev->device->name);
1244 gso_ctx->direct_pool = mp;
1245 gso_ctx->indirect_pool = mp;
1246 gso_ctx->gso_types = gso_types;
1247 gso_ctx->gso_size = 0; /* gso_size is set in tx_burst() per packet */
1254 tap_setup_queue(struct rte_eth_dev *dev,
1255 struct pmd_internals *internals,
1263 struct pmd_internals *pmd = dev->data->dev_private;
1264 struct pmd_process_private *process_private = dev->process_private;
1265 struct rx_queue *rx = &internals->rxq[qid];
1266 struct tx_queue *tx = &internals->txq[qid];
1267 struct rte_gso_ctx *gso_ctx;
1270 fd = &process_private->rxq_fds[qid];
1271 other_fd = &process_private->txq_fds[qid];
1275 fd = &process_private->txq_fds[qid];
1276 other_fd = &process_private->rxq_fds[qid];
1278 gso_ctx = &tx->gso_ctx;
1281 /* fd for this queue already exists */
1282 TAP_LOG(DEBUG, "%s: fd %d for %s queue qid %d exists",
1283 pmd->name, *fd, dir, qid);
1285 } else if (*other_fd != -1) {
1286 /* Only other_fd exists. dup it */
1287 *fd = dup(*other_fd);
1290 TAP_LOG(ERR, "%s: dup() failed.", pmd->name);
1293 TAP_LOG(DEBUG, "%s: dup fd %d for %s queue qid %d (%d)",
1294 pmd->name, *other_fd, dir, qid, *fd);
1296 /* Both RX and TX fds do not exist (equal -1). Create fd */
1297 *fd = tun_alloc(pmd, 0);
1299 *fd = -1; /* restore original value */
1300 TAP_LOG(ERR, "%s: tun_alloc() failed.", pmd->name);
1303 TAP_LOG(DEBUG, "%s: add %s queue for qid %d fd %d",
1304 pmd->name, dir, qid, *fd);
1307 tx->mtu = &dev->data->mtu;
1308 rx->rxmode = &dev->data->dev_conf.rxmode;
1310 ret = tap_gso_ctx_setup(gso_ctx, dev);
1315 tx->type = pmd->type;
1321 tap_rx_queue_setup(struct rte_eth_dev *dev,
1322 uint16_t rx_queue_id,
1323 uint16_t nb_rx_desc,
1324 unsigned int socket_id,
1325 const struct rte_eth_rxconf *rx_conf __rte_unused,
1326 struct rte_mempool *mp)
1328 struct pmd_internals *internals = dev->data->dev_private;
1329 struct pmd_process_private *process_private = dev->process_private;
1330 struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1331 struct rte_mbuf **tmp = &rxq->pool;
1332 long iov_max = sysconf(_SC_IOV_MAX);
1336 "_SC_IOV_MAX is not defined. Using %d as default",
1337 TAP_IOV_DEFAULT_MAX);
1338 iov_max = TAP_IOV_DEFAULT_MAX;
1340 uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1341 struct iovec (*iovecs)[nb_desc + 1];
1342 int data_off = RTE_PKTMBUF_HEADROOM;
1347 if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1349 "nb_rx_queues %d too small or mempool NULL",
1350 dev->data->nb_rx_queues);
1355 rxq->trigger_seen = 1; /* force initial burst */
1356 rxq->in_port = dev->data->port_id;
1357 rxq->queue_id = rx_queue_id;
1358 rxq->nb_rx_desc = nb_desc;
1359 iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1363 "%s: Couldn't allocate %d RX descriptors",
1364 dev->device->name, nb_desc);
1367 rxq->iovecs = iovecs;
1369 dev->data->rx_queues[rx_queue_id] = rxq;
1370 fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1376 (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1377 (*rxq->iovecs)[0].iov_base = &rxq->pi;
1379 for (i = 1; i <= nb_desc; i++) {
1380 *tmp = rte_pktmbuf_alloc(rxq->mp);
1383 "%s: couldn't allocate memory for queue %d",
1384 dev->device->name, rx_queue_id);
1388 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1389 (*rxq->iovecs)[i].iov_base =
1390 (char *)(*tmp)->buf_addr + data_off;
1392 tmp = &(*tmp)->next;
1395 TAP_LOG(DEBUG, " RX TUNTAP device name %s, qid %d on fd %d",
1396 internals->name, rx_queue_id,
1397 process_private->rxq_fds[rx_queue_id]);
1402 rte_pktmbuf_free(rxq->pool);
1404 rte_free(rxq->iovecs);
1410 tap_tx_queue_setup(struct rte_eth_dev *dev,
1411 uint16_t tx_queue_id,
1412 uint16_t nb_tx_desc __rte_unused,
1413 unsigned int socket_id __rte_unused,
1414 const struct rte_eth_txconf *tx_conf)
1416 struct pmd_internals *internals = dev->data->dev_private;
1417 struct pmd_process_private *process_private = dev->process_private;
1418 struct tx_queue *txq;
1422 if (tx_queue_id >= dev->data->nb_tx_queues)
1424 dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1425 txq = dev->data->tx_queues[tx_queue_id];
1426 txq->out_port = dev->data->port_id;
1427 txq->queue_id = tx_queue_id;
1429 offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1430 txq->csum = !!(offloads &
1431 (DEV_TX_OFFLOAD_IPV4_CKSUM |
1432 DEV_TX_OFFLOAD_UDP_CKSUM |
1433 DEV_TX_OFFLOAD_TCP_CKSUM));
1435 ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1439 " TX TUNTAP device name %s, qid %d on fd %d csum %s",
1440 internals->name, tx_queue_id,
1441 process_private->txq_fds[tx_queue_id],
1442 txq->csum ? "on" : "off");
1448 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1450 struct pmd_internals *pmd = dev->data->dev_private;
1451 struct ifreq ifr = { .ifr_mtu = mtu };
1454 err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1456 dev->data->mtu = mtu;
1462 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1463 struct rte_ether_addr *mc_addr_set __rte_unused,
1464 uint32_t nb_mc_addr __rte_unused)
1467 * Nothing to do actually: the tap has no filtering whatsoever, every
1468 * packet is received.
1474 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1476 struct rte_eth_dev *dev = arg;
1477 struct pmd_internals *pmd = dev->data->dev_private;
1478 struct ifinfomsg *info = NLMSG_DATA(nh);
1480 if (nh->nlmsg_type != RTM_NEWLINK ||
1481 (info->ifi_index != pmd->if_index &&
1482 info->ifi_index != pmd->remote_if_index))
1484 return tap_link_update(dev, 0);
1488 tap_dev_intr_handler(void *cb_arg)
1490 struct rte_eth_dev *dev = cb_arg;
1491 struct pmd_internals *pmd = dev->data->dev_private;
1493 tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1497 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1499 struct pmd_internals *pmd = dev->data->dev_private;
1501 /* In any case, disable interrupt if the conf is no longer there. */
1502 if (!dev->data->dev_conf.intr_conf.lsc) {
1503 if (pmd->intr_handle.fd != -1) {
1504 tap_nl_final(pmd->intr_handle.fd);
1505 rte_intr_callback_unregister(&pmd->intr_handle,
1506 tap_dev_intr_handler, dev);
1511 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1512 if (unlikely(pmd->intr_handle.fd == -1))
1514 return rte_intr_callback_register(
1515 &pmd->intr_handle, tap_dev_intr_handler, dev);
1517 tap_nl_final(pmd->intr_handle.fd);
1518 return rte_intr_callback_unregister(&pmd->intr_handle,
1519 tap_dev_intr_handler, dev);
1523 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1527 err = tap_lsc_intr_handle_set(dev, set);
1530 err = tap_rx_intr_vec_set(dev, set);
1532 tap_lsc_intr_handle_set(dev, 0);
1536 static const uint32_t*
1537 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1539 static const uint32_t ptypes[] = {
1540 RTE_PTYPE_INNER_L2_ETHER,
1541 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1542 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1543 RTE_PTYPE_INNER_L3_IPV4,
1544 RTE_PTYPE_INNER_L3_IPV4_EXT,
1545 RTE_PTYPE_INNER_L3_IPV6,
1546 RTE_PTYPE_INNER_L3_IPV6_EXT,
1547 RTE_PTYPE_INNER_L4_FRAG,
1548 RTE_PTYPE_INNER_L4_UDP,
1549 RTE_PTYPE_INNER_L4_TCP,
1550 RTE_PTYPE_INNER_L4_SCTP,
1552 RTE_PTYPE_L2_ETHER_VLAN,
1553 RTE_PTYPE_L2_ETHER_QINQ,
1555 RTE_PTYPE_L3_IPV4_EXT,
1556 RTE_PTYPE_L3_IPV6_EXT,
1568 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1569 struct rte_eth_fc_conf *fc_conf)
1571 fc_conf->mode = RTE_FC_NONE;
1576 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1577 struct rte_eth_fc_conf *fc_conf)
1579 if (fc_conf->mode != RTE_FC_NONE)
1585 * DPDK callback to update the RSS hash configuration.
1588 * Pointer to Ethernet device structure.
1589 * @param[in] rss_conf
1590 * RSS configuration data.
1593 * 0 on success, a negative errno value otherwise and rte_errno is set.
1596 tap_rss_hash_update(struct rte_eth_dev *dev,
1597 struct rte_eth_rss_conf *rss_conf)
1599 if (rss_conf->rss_hf & TAP_RSS_HF_MASK) {
1603 if (rss_conf->rss_key && rss_conf->rss_key_len) {
1605 * Currently TAP RSS key is hard coded
1606 * and cannot be updated
1609 "port %u RSS key cannot be updated",
1610 dev->data->port_id);
1618 tap_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1620 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1626 tap_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1628 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1634 tap_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1636 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1642 tap_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1644 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1648 static const struct eth_dev_ops ops = {
1649 .dev_start = tap_dev_start,
1650 .dev_stop = tap_dev_stop,
1651 .dev_close = tap_dev_close,
1652 .dev_configure = tap_dev_configure,
1653 .dev_infos_get = tap_dev_info,
1654 .rx_queue_setup = tap_rx_queue_setup,
1655 .tx_queue_setup = tap_tx_queue_setup,
1656 .rx_queue_start = tap_rx_queue_start,
1657 .tx_queue_start = tap_tx_queue_start,
1658 .rx_queue_stop = tap_rx_queue_stop,
1659 .tx_queue_stop = tap_tx_queue_stop,
1660 .rx_queue_release = tap_rx_queue_release,
1661 .tx_queue_release = tap_tx_queue_release,
1662 .flow_ctrl_get = tap_flow_ctrl_get,
1663 .flow_ctrl_set = tap_flow_ctrl_set,
1664 .link_update = tap_link_update,
1665 .dev_set_link_up = tap_link_set_up,
1666 .dev_set_link_down = tap_link_set_down,
1667 .promiscuous_enable = tap_promisc_enable,
1668 .promiscuous_disable = tap_promisc_disable,
1669 .allmulticast_enable = tap_allmulti_enable,
1670 .allmulticast_disable = tap_allmulti_disable,
1671 .mac_addr_set = tap_mac_set,
1672 .mtu_set = tap_mtu_set,
1673 .set_mc_addr_list = tap_set_mc_addr_list,
1674 .stats_get = tap_stats_get,
1675 .stats_reset = tap_stats_reset,
1676 .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1677 .rss_hash_update = tap_rss_hash_update,
1678 .filter_ctrl = tap_dev_filter_ctrl,
1681 static const char *tuntap_types[ETH_TUNTAP_TYPE_MAX] = {
1682 "UNKNOWN", "TUN", "TAP"
1686 eth_dev_tap_create(struct rte_vdev_device *vdev, const char *tap_name,
1687 char *remote_iface, struct rte_ether_addr *mac_addr,
1688 enum rte_tuntap_type type)
1690 int numa_node = rte_socket_id();
1691 struct rte_eth_dev *dev;
1692 struct pmd_internals *pmd;
1693 struct pmd_process_private *process_private;
1694 const char *tuntap_name = tuntap_types[type];
1695 struct rte_eth_dev_data *data;
1699 TAP_LOG(DEBUG, "%s device on numa %u", tuntap_name, rte_socket_id());
1701 dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1703 TAP_LOG(ERR, "%s Unable to allocate device struct",
1705 goto error_exit_nodev;
1708 process_private = (struct pmd_process_private *)
1709 rte_zmalloc_socket(tap_name, sizeof(struct pmd_process_private),
1710 RTE_CACHE_LINE_SIZE, dev->device->numa_node);
1712 if (process_private == NULL) {
1713 TAP_LOG(ERR, "Failed to alloc memory for process private");
1716 pmd = dev->data->dev_private;
1717 dev->process_private = process_private;
1719 strlcpy(pmd->name, tap_name, sizeof(pmd->name));
1722 pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1723 if (pmd->ioctl_sock == -1) {
1725 "%s Unable to get a socket for management: %s",
1726 tuntap_name, strerror(errno));
1730 /* Setup some default values */
1732 data->dev_private = pmd;
1733 data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1734 data->numa_node = numa_node;
1736 data->dev_link = pmd_link;
1737 data->mac_addrs = &pmd->eth_addr;
1738 /* Set the number of RX and TX queues */
1739 data->nb_rx_queues = 0;
1740 data->nb_tx_queues = 0;
1742 dev->dev_ops = &ops;
1743 dev->rx_pkt_burst = pmd_rx_burst;
1744 dev->tx_pkt_burst = pmd_tx_burst;
1746 pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1747 pmd->intr_handle.fd = -1;
1748 dev->intr_handle = &pmd->intr_handle;
1750 /* Presetup the fds to -1 as being not valid */
1752 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1753 process_private->rxq_fds[i] = -1;
1754 process_private->txq_fds[i] = -1;
1757 if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1758 if (rte_is_zero_ether_addr(mac_addr))
1759 rte_eth_random_addr((uint8_t *)&pmd->eth_addr);
1761 rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1765 * Allocate a TUN device keep-alive file descriptor that will only be
1766 * closed when the TUN device itself is closed or removed.
1767 * This keep-alive file descriptor will guarantee that the TUN device
1768 * exists even when all of its queues are closed
1770 pmd->ka_fd = tun_alloc(pmd, 1);
1771 if (pmd->ka_fd == -1) {
1772 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
1775 TAP_LOG(DEBUG, "allocated %s", pmd->name);
1777 ifr.ifr_mtu = dev->data->mtu;
1778 if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1781 if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1782 memset(&ifr, 0, sizeof(struct ifreq));
1783 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1784 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
1785 RTE_ETHER_ADDR_LEN);
1786 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1791 * Set up everything related to rte_flow:
1793 * - tap / remote if_index
1794 * - mandatory QDISCs
1795 * - rte_flow actual/implicit lists
1798 pmd->nlsk_fd = tap_nl_init(0);
1799 if (pmd->nlsk_fd == -1) {
1800 TAP_LOG(WARNING, "%s: failed to create netlink socket.",
1802 goto disable_rte_flow;
1804 pmd->if_index = if_nametoindex(pmd->name);
1805 if (!pmd->if_index) {
1806 TAP_LOG(ERR, "%s: failed to get if_index.", pmd->name);
1807 goto disable_rte_flow;
1809 if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1810 TAP_LOG(ERR, "%s: failed to create multiq qdisc.",
1812 goto disable_rte_flow;
1814 if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1815 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1817 goto disable_rte_flow;
1819 LIST_INIT(&pmd->flows);
1821 if (strlen(remote_iface)) {
1822 pmd->remote_if_index = if_nametoindex(remote_iface);
1823 if (!pmd->remote_if_index) {
1824 TAP_LOG(ERR, "%s: failed to get %s if_index.",
1825 pmd->name, remote_iface);
1828 strlcpy(pmd->remote_iface, remote_iface, RTE_ETH_NAME_MAX_LEN);
1830 /* Save state of remote device */
1831 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1833 /* Replicate remote MAC address */
1834 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1835 TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1836 pmd->name, pmd->remote_iface);
1839 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1840 RTE_ETHER_ADDR_LEN);
1841 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1842 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1843 TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1844 pmd->name, remote_iface);
1849 * Flush usually returns negative value because it tries to
1850 * delete every QDISC (and on a running device, one QDISC at
1851 * least is needed). Ignore negative return value.
1853 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1854 if (qdisc_create_ingress(pmd->nlsk_fd,
1855 pmd->remote_if_index) < 0) {
1856 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1860 LIST_INIT(&pmd->implicit_flows);
1861 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1862 tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1863 tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1864 tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1866 "%s: failed to create implicit rules.",
1872 rte_eth_dev_probing_finish(dev);
1876 TAP_LOG(ERR, " Disabling rte flow support: %s(%d)",
1877 strerror(errno), errno);
1878 if (strlen(remote_iface)) {
1879 TAP_LOG(ERR, "Remote feature requires flow support.");
1882 rte_eth_dev_probing_finish(dev);
1886 TAP_LOG(ERR, " Can't set up remote feature: %s(%d)",
1887 strerror(errno), errno);
1888 tap_flow_implicit_flush(pmd, NULL);
1891 if (pmd->ioctl_sock > 0)
1892 close(pmd->ioctl_sock);
1893 /* mac_addrs must not be freed alone because part of dev_private */
1894 dev->data->mac_addrs = NULL;
1895 rte_eth_dev_release_port(dev);
1898 TAP_LOG(ERR, "%s Unable to initialize %s",
1899 tuntap_name, rte_vdev_device_name(vdev));
1904 /* make sure name is a possible Linux network device name */
1906 is_valid_iface(const char *name)
1911 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1915 if (*name == '/' || *name == ':' || isspace(*name))
1923 set_interface_name(const char *key __rte_unused,
1927 char *name = (char *)extra_args;
1930 if (!is_valid_iface(value)) {
1931 TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
1935 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
1937 /* use tap%d which causes kernel to choose next available */
1938 strlcpy(name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
1944 set_remote_iface(const char *key __rte_unused,
1948 char *name = (char *)extra_args;
1951 if (!is_valid_iface(value)) {
1952 TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
1956 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
1962 static int parse_user_mac(struct rte_ether_addr *user_mac,
1965 unsigned int index = 0;
1966 char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
1968 if (user_mac == NULL || value == NULL)
1971 strlcpy(mac_temp, value, sizeof(mac_temp));
1972 mac_byte = strtok(mac_temp, ":");
1974 while ((mac_byte != NULL) &&
1975 (strlen(mac_byte) <= 2) &&
1976 (strlen(mac_byte) == strspn(mac_byte,
1977 ETH_TAP_CMP_MAC_FMT))) {
1978 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
1979 mac_byte = strtok(NULL, ":");
1986 set_mac_type(const char *key __rte_unused,
1990 struct rte_ether_addr *user_mac = extra_args;
1995 if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
1996 static int iface_idx;
1998 /* fixed mac = 00:64:74:61:70:<iface_idx> */
1999 memcpy((char *)user_mac->addr_bytes, "\0dtap",
2000 RTE_ETHER_ADDR_LEN);
2001 user_mac->addr_bytes[RTE_ETHER_ADDR_LEN - 1] =
2006 if (parse_user_mac(user_mac, value) != 6)
2009 TAP_LOG(DEBUG, "TAP user MAC param (%s)", value);
2013 TAP_LOG(ERR, "TAP user MAC (%s) is not in format (%s|%s)",
2014 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
2019 * Open a TUN interface device. TUN PMD
2020 * 1) sets tap_type as false
2021 * 2) intakes iface as argument.
2022 * 3) as interface is virtual set speed to 10G
2025 rte_pmd_tun_probe(struct rte_vdev_device *dev)
2027 const char *name, *params;
2029 struct rte_kvargs *kvlist = NULL;
2030 char tun_name[RTE_ETH_NAME_MAX_LEN];
2031 char remote_iface[RTE_ETH_NAME_MAX_LEN];
2032 struct rte_eth_dev *eth_dev;
2034 name = rte_vdev_device_name(dev);
2035 params = rte_vdev_device_args(dev);
2036 memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2038 if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
2039 strlen(params) == 0) {
2040 eth_dev = rte_eth_dev_attach_secondary(name);
2042 TAP_LOG(ERR, "Failed to probe %s", name);
2045 eth_dev->dev_ops = &ops;
2046 eth_dev->device = &dev->device;
2047 rte_eth_dev_probing_finish(eth_dev);
2051 /* use tun%d which causes kernel to choose next available */
2052 strlcpy(tun_name, DEFAULT_TUN_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2054 if (params && (params[0] != '\0')) {
2055 TAP_LOG(DEBUG, "parameters (%s)", params);
2057 kvlist = rte_kvargs_parse(params, valid_arguments);
2059 if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2060 ret = rte_kvargs_process(kvlist,
2062 &set_interface_name,
2070 pmd_link.link_speed = ETH_SPEED_NUM_10G;
2072 TAP_LOG(DEBUG, "Initializing pmd_tun for %s", name);
2074 ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0,
2075 ETH_TUNTAP_TYPE_TUN);
2079 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2082 rte_kvargs_free(kvlist);
2087 /* Request queue file descriptors from secondary to primary. */
2089 tap_mp_attach_queues(const char *port_name, struct rte_eth_dev *dev)
2092 struct timespec timeout = {.tv_sec = 1, .tv_nsec = 0};
2093 struct rte_mp_msg request, *reply;
2094 struct rte_mp_reply replies;
2095 struct ipc_queues *request_param = (struct ipc_queues *)request.param;
2096 struct ipc_queues *reply_param;
2097 struct pmd_process_private *process_private = dev->process_private;
2098 int queue, fd_iterator;
2100 /* Prepare the request */
2101 memset(&request, 0, sizeof(request));
2102 strlcpy(request.name, TAP_MP_KEY, sizeof(request.name));
2103 strlcpy(request_param->port_name, port_name,
2104 sizeof(request_param->port_name));
2105 request.len_param = sizeof(*request_param);
2106 /* Send request and receive reply */
2107 ret = rte_mp_request_sync(&request, &replies, &timeout);
2108 if (ret < 0 || replies.nb_received != 1) {
2109 TAP_LOG(ERR, "Failed to request queues from primary: %d",
2113 reply = &replies.msgs[0];
2114 reply_param = (struct ipc_queues *)reply->param;
2115 TAP_LOG(DEBUG, "Received IPC reply for %s", reply_param->port_name);
2117 /* Attach the queues from received file descriptors */
2118 if (reply_param->rxq_count + reply_param->txq_count != reply->num_fds) {
2119 TAP_LOG(ERR, "Unexpected number of fds received");
2123 dev->data->nb_rx_queues = reply_param->rxq_count;
2124 dev->data->nb_tx_queues = reply_param->txq_count;
2126 for (queue = 0; queue < reply_param->rxq_count; queue++)
2127 process_private->rxq_fds[queue] = reply->fds[fd_iterator++];
2128 for (queue = 0; queue < reply_param->txq_count; queue++)
2129 process_private->txq_fds[queue] = reply->fds[fd_iterator++];
2134 /* Send the queue file descriptors from the primary process to secondary. */
2136 tap_mp_sync_queues(const struct rte_mp_msg *request, const void *peer)
2138 struct rte_eth_dev *dev;
2139 struct pmd_process_private *process_private;
2140 struct rte_mp_msg reply;
2141 const struct ipc_queues *request_param =
2142 (const struct ipc_queues *)request->param;
2143 struct ipc_queues *reply_param =
2144 (struct ipc_queues *)reply.param;
2149 /* Get requested port */
2150 TAP_LOG(DEBUG, "Received IPC request for %s", request_param->port_name);
2151 ret = rte_eth_dev_get_port_by_name(request_param->port_name, &port_id);
2153 TAP_LOG(ERR, "Failed to get port id for %s",
2154 request_param->port_name);
2157 dev = &rte_eth_devices[port_id];
2158 process_private = dev->process_private;
2160 /* Fill file descriptors for all queues */
2162 reply_param->rxq_count = 0;
2163 if (dev->data->nb_rx_queues + dev->data->nb_tx_queues >
2165 TAP_LOG(ERR, "Number of rx/tx queues exceeds max number of fds");
2169 for (queue = 0; queue < dev->data->nb_rx_queues; queue++) {
2170 reply.fds[reply.num_fds++] = process_private->rxq_fds[queue];
2171 reply_param->rxq_count++;
2173 RTE_ASSERT(reply_param->rxq_count == dev->data->nb_rx_queues);
2175 reply_param->txq_count = 0;
2176 for (queue = 0; queue < dev->data->nb_tx_queues; queue++) {
2177 reply.fds[reply.num_fds++] = process_private->txq_fds[queue];
2178 reply_param->txq_count++;
2180 RTE_ASSERT(reply_param->txq_count == dev->data->nb_tx_queues);
2183 strlcpy(reply.name, request->name, sizeof(reply.name));
2184 strlcpy(reply_param->port_name, request_param->port_name,
2185 sizeof(reply_param->port_name));
2186 reply.len_param = sizeof(*reply_param);
2187 if (rte_mp_reply(&reply, peer) < 0) {
2188 TAP_LOG(ERR, "Failed to reply an IPC request to sync queues");
2194 /* Open a TAP interface device.
2197 rte_pmd_tap_probe(struct rte_vdev_device *dev)
2199 const char *name, *params;
2201 struct rte_kvargs *kvlist = NULL;
2203 char tap_name[RTE_ETH_NAME_MAX_LEN];
2204 char remote_iface[RTE_ETH_NAME_MAX_LEN];
2205 struct rte_ether_addr user_mac = { .addr_bytes = {0} };
2206 struct rte_eth_dev *eth_dev;
2207 int tap_devices_count_increased = 0;
2209 name = rte_vdev_device_name(dev);
2210 params = rte_vdev_device_args(dev);
2212 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
2213 eth_dev = rte_eth_dev_attach_secondary(name);
2215 TAP_LOG(ERR, "Failed to probe %s", name);
2218 eth_dev->dev_ops = &ops;
2219 eth_dev->device = &dev->device;
2220 eth_dev->rx_pkt_burst = pmd_rx_burst;
2221 eth_dev->tx_pkt_burst = pmd_tx_burst;
2222 if (!rte_eal_primary_proc_alive(NULL)) {
2223 TAP_LOG(ERR, "Primary process is missing");
2226 eth_dev->process_private = (struct pmd_process_private *)
2227 rte_zmalloc_socket(name,
2228 sizeof(struct pmd_process_private),
2229 RTE_CACHE_LINE_SIZE,
2230 eth_dev->device->numa_node);
2231 if (eth_dev->process_private == NULL) {
2233 "Failed to alloc memory for process private");
2237 ret = tap_mp_attach_queues(name, eth_dev);
2240 rte_eth_dev_probing_finish(eth_dev);
2244 speed = ETH_SPEED_NUM_10G;
2246 /* use tap%d which causes kernel to choose next available */
2247 strlcpy(tap_name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2248 memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2250 if (params && (params[0] != '\0')) {
2251 TAP_LOG(DEBUG, "parameters (%s)", params);
2253 kvlist = rte_kvargs_parse(params, valid_arguments);
2255 if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2256 ret = rte_kvargs_process(kvlist,
2258 &set_interface_name,
2264 if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
2265 ret = rte_kvargs_process(kvlist,
2273 if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
2274 ret = rte_kvargs_process(kvlist,
2283 pmd_link.link_speed = speed;
2285 TAP_LOG(DEBUG, "Initializing pmd_tap for %s", name);
2287 /* Register IPC feed callback */
2288 if (!tap_devices_count) {
2289 ret = rte_mp_action_register(TAP_MP_KEY, tap_mp_sync_queues);
2291 TAP_LOG(ERR, "tap: Failed to register IPC callback: %s",
2292 strerror(rte_errno));
2296 tap_devices_count++;
2297 tap_devices_count_increased = 1;
2298 ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac,
2299 ETH_TUNTAP_TYPE_TAP);
2303 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2305 if (tap_devices_count_increased == 1) {
2306 if (tap_devices_count == 1)
2307 rte_mp_action_unregister(TAP_MP_KEY);
2308 tap_devices_count--;
2311 rte_kvargs_free(kvlist);
2316 /* detach a TUNTAP device.
2319 rte_pmd_tap_remove(struct rte_vdev_device *dev)
2321 struct rte_eth_dev *eth_dev = NULL;
2322 struct pmd_internals *internals;
2323 struct pmd_process_private *process_private;
2326 /* find the ethdev entry */
2327 eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
2331 /* mac_addrs must not be freed alone because part of dev_private */
2332 eth_dev->data->mac_addrs = NULL;
2334 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2335 return rte_eth_dev_release_port(eth_dev);
2337 internals = eth_dev->data->dev_private;
2338 process_private = eth_dev->process_private;
2340 TAP_LOG(DEBUG, "Closing %s Ethernet device on numa %u",
2341 tuntap_types[internals->type], rte_socket_id());
2343 if (internals->nlsk_fd) {
2344 tap_flow_flush(eth_dev, NULL);
2345 tap_flow_implicit_flush(internals, NULL);
2346 tap_nl_final(internals->nlsk_fd);
2348 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
2349 if (process_private->rxq_fds[i] != -1) {
2350 close(process_private->rxq_fds[i]);
2351 process_private->rxq_fds[i] = -1;
2353 if (process_private->txq_fds[i] != -1) {
2354 close(process_private->txq_fds[i]);
2355 process_private->txq_fds[i] = -1;
2359 close(internals->ioctl_sock);
2360 rte_free(eth_dev->process_private);
2361 if (tap_devices_count == 1)
2362 rte_mp_action_unregister(TAP_MP_KEY);
2363 tap_devices_count--;
2364 rte_eth_dev_release_port(eth_dev);
2366 if (internals->ka_fd != -1) {
2367 close(internals->ka_fd);
2368 internals->ka_fd = -1;
2373 static struct rte_vdev_driver pmd_tun_drv = {
2374 .probe = rte_pmd_tun_probe,
2375 .remove = rte_pmd_tap_remove,
2378 static struct rte_vdev_driver pmd_tap_drv = {
2379 .probe = rte_pmd_tap_probe,
2380 .remove = rte_pmd_tap_remove,
2383 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
2384 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
2385 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
2386 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
2387 ETH_TAP_IFACE_ARG "=<string> ");
2388 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
2389 ETH_TAP_IFACE_ARG "=<string> "
2390 ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
2391 ETH_TAP_REMOTE_ARG "=<string>");
2394 RTE_INIT(tap_init_log)
2396 tap_logtype = rte_log_register("pmd.net.tap");
2397 if (tap_logtype >= 0)
2398 rte_log_set_level(tap_logtype, RTE_LOG_NOTICE);