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>
21 #include <rte_cycles.h>
23 #include <sys/types.h>
25 #include <sys/socket.h>
26 #include <sys/ioctl.h>
27 #include <sys/utsname.h>
35 #include <arpa/inet.h>
37 #include <linux/if_tun.h>
38 #include <linux/if_ether.h>
43 #include <rte_eth_tap.h>
45 #include <tap_netlink.h>
46 #include <tap_tcmsgs.h>
48 /* Linux based path to the TUN device */
49 #define TUN_TAP_DEV_PATH "/dev/net/tun"
50 #define DEFAULT_TAP_NAME "dtap"
51 #define DEFAULT_TUN_NAME "dtun"
53 #define ETH_TAP_IFACE_ARG "iface"
54 #define ETH_TAP_REMOTE_ARG "remote"
55 #define ETH_TAP_MAC_ARG "mac"
56 #define ETH_TAP_MAC_FIXED "fixed"
58 #define ETH_TAP_USR_MAC_FMT "xx:xx:xx:xx:xx:xx"
59 #define ETH_TAP_CMP_MAC_FMT "0123456789ABCDEFabcdef"
60 #define ETH_TAP_MAC_ARG_FMT ETH_TAP_MAC_FIXED "|" ETH_TAP_USR_MAC_FMT
62 #define TAP_GSO_MBUFS_PER_CORE 128
63 #define TAP_GSO_MBUF_SEG_SIZE 128
64 #define TAP_GSO_MBUF_CACHE_SIZE 4
65 #define TAP_GSO_MBUFS_NUM \
66 (TAP_GSO_MBUFS_PER_CORE * TAP_GSO_MBUF_CACHE_SIZE)
68 /* IPC key for queue fds sync */
69 #define TAP_MP_KEY "tap_mp_sync_queues"
71 #define TAP_IOV_DEFAULT_MAX 1024
73 static int tap_devices_count;
75 static const char *tuntap_types[ETH_TUNTAP_TYPE_MAX] = {
76 "UNKNOWN", "TUN", "TAP"
79 static const char *valid_arguments[] = {
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;
141 int fd, signo, flags;
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 strlcpy(ifr.ifr_name, pmd->name, IFNAMSIZ);
153 fd = open(TUN_TAP_DEV_PATH, O_RDWR);
155 TAP_LOG(ERR, "Unable to open %s interface", TUN_TAP_DEV_PATH);
159 #ifdef IFF_MULTI_QUEUE
160 /* Grab the TUN features to verify we can work multi-queue */
161 if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
162 TAP_LOG(ERR, "unable to get TUN/TAP features");
165 TAP_LOG(DEBUG, "%s Features %08x", TUN_TAP_DEV_PATH, features);
167 if (features & IFF_MULTI_QUEUE) {
168 TAP_LOG(DEBUG, " Multi-queue support for %d queues",
169 RTE_PMD_TAP_MAX_QUEUES);
170 ifr.ifr_flags |= IFF_MULTI_QUEUE;
174 ifr.ifr_flags |= IFF_ONE_QUEUE;
175 TAP_LOG(DEBUG, " Single queue only support");
178 /* Set the TUN/TAP configuration and set the name if needed */
179 if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
180 TAP_LOG(WARNING, "Unable to set TUNSETIFF for %s: %s",
181 ifr.ifr_name, strerror(errno));
186 * Name passed to kernel might be wildcard like dtun%d
187 * and need to find the resulting device.
189 TAP_LOG(DEBUG, "Device name is '%s'", ifr.ifr_name);
190 strlcpy(pmd->name, ifr.ifr_name, RTE_ETH_NAME_MAX_LEN);
194 * Detach the TUN/TAP keep-alive queue
195 * to avoid traffic through it
197 ifr.ifr_flags = IFF_DETACH_QUEUE;
198 if (ioctl(fd, TUNSETQUEUE, (void *)&ifr) < 0) {
200 "Unable to detach keep-alive queue for %s: %s",
201 ifr.ifr_name, strerror(errno));
206 flags = fcntl(fd, F_GETFL);
209 "Unable to get %s current flags\n",
214 /* Always set the file descriptor to non-blocking */
216 if (fcntl(fd, F_SETFL, flags) < 0) {
218 "Unable to set %s to nonblocking: %s",
219 ifr.ifr_name, strerror(errno));
223 /* Find a free realtime signal */
224 for (signo = SIGRTMIN + 1; signo < SIGRTMAX; signo++) {
227 if (sigaction(signo, NULL, &sa) == -1) {
229 "Unable to get current rt-signal %d handler",
234 /* Already have the handler we want on this signal */
235 if (sa.sa_handler == tap_trigger_cb)
238 /* Is handler in use by application */
239 if (sa.sa_handler != SIG_DFL) {
241 "Skipping used rt-signal %d", signo);
245 sa = (struct sigaction) {
246 .sa_flags = SA_RESTART,
247 .sa_handler = tap_trigger_cb,
250 if (sigaction(signo, &sa, NULL) == -1) {
252 "Unable to set rt-signal %d handler\n", signo);
256 /* Found a good signal to use */
258 "Using rt-signal %d", signo);
262 if (signo == SIGRTMAX) {
263 TAP_LOG(WARNING, "All rt-signals are in use\n");
265 /* Disable trigger globally in case of error */
267 TAP_LOG(NOTICE, "No Rx trigger signal available\n");
269 /* Enable signal on file descriptor */
270 if (fcntl(fd, F_SETSIG, signo) < 0) {
271 TAP_LOG(WARNING, "Unable to set signo %d for fd %d: %s",
272 signo, fd, strerror(errno));
275 if (fcntl(fd, F_SETFL, flags | O_ASYNC) < 0) {
276 TAP_LOG(WARNING, "Unable to set fcntl flags: %s",
281 if (fcntl(fd, F_SETOWN, getpid()) < 0) {
282 TAP_LOG(WARNING, "Unable to set fcntl owner: %s",
296 tap_verify_csum(struct rte_mbuf *mbuf)
298 uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
299 uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
300 uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
301 unsigned int l2_len = sizeof(struct rte_ether_hdr);
307 if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
309 else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
311 /* Don't verify checksum for packets with discontinuous L2 header */
312 if (unlikely(l2_len + sizeof(struct rte_ipv4_hdr) >
313 rte_pktmbuf_data_len(mbuf)))
315 l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
316 if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
317 struct rte_ipv4_hdr *iph = l3_hdr;
319 l3_len = rte_ipv4_hdr_len(iph);
320 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
322 /* check that the total length reported by header is not
323 * greater than the total received size
325 if (l2_len + rte_be_to_cpu_16(iph->total_length) >
326 rte_pktmbuf_data_len(mbuf))
329 cksum = ~rte_raw_cksum(iph, l3_len);
330 mbuf->ol_flags |= cksum ?
331 PKT_RX_IP_CKSUM_BAD :
332 PKT_RX_IP_CKSUM_GOOD;
333 } else if (l3 == RTE_PTYPE_L3_IPV6) {
334 struct rte_ipv6_hdr *iph = l3_hdr;
336 l3_len = sizeof(struct rte_ipv6_hdr);
337 /* check that the total length reported by header is not
338 * greater than the total received size
340 if (l2_len + l3_len + rte_be_to_cpu_16(iph->payload_len) >
341 rte_pktmbuf_data_len(mbuf))
344 /* IPv6 extensions are not supported */
347 if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
348 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
349 /* Don't verify checksum for multi-segment packets. */
350 if (mbuf->nb_segs > 1)
352 if (l3 == RTE_PTYPE_L3_IPV4)
353 cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
354 else if (l3 == RTE_PTYPE_L3_IPV6)
355 cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
356 mbuf->ol_flags |= cksum ?
357 PKT_RX_L4_CKSUM_BAD :
358 PKT_RX_L4_CKSUM_GOOD;
363 tap_rx_offload_get_port_capa(void)
366 * No specific port Rx offload capabilities.
372 tap_rx_offload_get_queue_capa(void)
374 return DEV_RX_OFFLOAD_SCATTER |
375 DEV_RX_OFFLOAD_IPV4_CKSUM |
376 DEV_RX_OFFLOAD_UDP_CKSUM |
377 DEV_RX_OFFLOAD_TCP_CKSUM;
381 tap_rxq_pool_free(struct rte_mbuf *pool)
383 struct rte_mbuf *mbuf = pool;
384 uint16_t nb_segs = 1;
393 pool->nb_segs = nb_segs;
394 rte_pktmbuf_free(pool);
397 /* Callback to handle the rx burst of packets to the correct interface and
398 * file descriptor(s) in a multi-queue setup.
401 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
403 struct rx_queue *rxq = queue;
404 struct pmd_process_private *process_private;
406 unsigned long num_rx_bytes = 0;
407 uint32_t trigger = tap_trigger;
409 if (trigger == rxq->trigger_seen)
412 process_private = rte_eth_devices[rxq->in_port].process_private;
413 for (num_rx = 0; num_rx < nb_pkts; ) {
414 struct rte_mbuf *mbuf = rxq->pool;
415 struct rte_mbuf *seg = NULL;
416 struct rte_mbuf *new_tail = NULL;
417 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
420 len = readv(process_private->rxq_fds[rxq->queue_id],
422 1 + (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
423 rxq->nb_rx_desc : 1));
424 if (len < (int)sizeof(struct tun_pi))
427 /* Packet couldn't fit in the provided mbuf */
428 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
429 rxq->stats.ierrors++;
433 len -= sizeof(struct tun_pi);
436 mbuf->port = rxq->in_port;
438 struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
440 if (unlikely(!buf)) {
441 rxq->stats.rx_nombuf++;
442 /* No new buf has been allocated: do nothing */
443 if (!new_tail || !seg)
447 tap_rxq_pool_free(mbuf);
451 seg = seg ? seg->next : mbuf;
452 if (rxq->pool == mbuf)
455 new_tail->next = buf;
457 new_tail->next = seg->next;
459 /* iovecs[0] is reserved for packet info (pi) */
460 (*rxq->iovecs)[mbuf->nb_segs].iov_len =
461 buf->buf_len - data_off;
462 (*rxq->iovecs)[mbuf->nb_segs].iov_base =
463 (char *)buf->buf_addr + data_off;
465 seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
466 seg->data_off = data_off;
468 len -= seg->data_len;
472 /* First segment has headroom, not the others */
476 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
478 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
479 tap_verify_csum(mbuf);
481 /* account for the receive frame */
482 bufs[num_rx++] = mbuf;
483 num_rx_bytes += mbuf->pkt_len;
486 rxq->stats.ipackets += num_rx;
487 rxq->stats.ibytes += num_rx_bytes;
489 if (trigger && num_rx < nb_pkts)
490 rxq->trigger_seen = trigger;
496 tap_tx_offload_get_port_capa(void)
499 * No specific port Tx offload capabilities.
505 tap_tx_offload_get_queue_capa(void)
507 return DEV_TX_OFFLOAD_MULTI_SEGS |
508 DEV_TX_OFFLOAD_IPV4_CKSUM |
509 DEV_TX_OFFLOAD_UDP_CKSUM |
510 DEV_TX_OFFLOAD_TCP_CKSUM |
511 DEV_TX_OFFLOAD_TCP_TSO;
514 /* Finalize l4 checksum calculation */
516 tap_tx_l4_cksum(uint16_t *l4_cksum, uint16_t l4_phdr_cksum,
517 uint32_t l4_raw_cksum)
522 cksum = __rte_raw_cksum_reduce(l4_raw_cksum);
523 cksum += l4_phdr_cksum;
525 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
526 cksum = (~cksum) & 0xffff;
533 /* Accumaulate L4 raw checksums */
535 tap_tx_l4_add_rcksum(char *l4_data, unsigned int l4_len, uint16_t *l4_cksum,
536 uint32_t *l4_raw_cksum)
538 if (l4_cksum == NULL)
541 *l4_raw_cksum = __rte_raw_cksum(l4_data, l4_len, *l4_raw_cksum);
544 /* L3 and L4 pseudo headers checksum offloads */
546 tap_tx_l3_cksum(char *packet, uint64_t ol_flags, unsigned int l2_len,
547 unsigned int l3_len, unsigned int l4_len, uint16_t **l4_cksum,
548 uint16_t *l4_phdr_cksum, uint32_t *l4_raw_cksum)
550 void *l3_hdr = packet + l2_len;
552 if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
553 struct rte_ipv4_hdr *iph = l3_hdr;
556 iph->hdr_checksum = 0;
557 cksum = rte_raw_cksum(iph, l3_len);
558 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
560 if (ol_flags & PKT_TX_L4_MASK) {
563 l4_hdr = packet + l2_len + l3_len;
564 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
565 *l4_cksum = &((struct rte_udp_hdr *)l4_hdr)->dgram_cksum;
566 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
567 *l4_cksum = &((struct rte_tcp_hdr *)l4_hdr)->cksum;
571 if (ol_flags & PKT_TX_IPV4)
572 *l4_phdr_cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
574 *l4_phdr_cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
575 *l4_raw_cksum = __rte_raw_cksum(l4_hdr, l4_len, 0);
580 tap_write_mbufs(struct tx_queue *txq, uint16_t num_mbufs,
581 struct rte_mbuf **pmbufs,
582 uint16_t *num_packets, unsigned long *num_tx_bytes)
586 struct pmd_process_private *process_private;
588 process_private = rte_eth_devices[txq->out_port].process_private;
590 for (i = 0; i < num_mbufs; i++) {
591 struct rte_mbuf *mbuf = pmbufs[i];
592 struct iovec iovecs[mbuf->nb_segs + 2];
593 struct tun_pi pi = { .flags = 0, .proto = 0x00 };
594 struct rte_mbuf *seg = mbuf;
595 char m_copy[mbuf->data_len];
599 int k; /* current index in iovecs for copying segments */
600 uint16_t seg_len; /* length of first segment */
602 uint16_t *l4_cksum; /* l4 checksum (pseudo header + payload) */
603 uint32_t l4_raw_cksum = 0; /* TCP/UDP payload raw checksum */
604 uint16_t l4_phdr_cksum = 0; /* TCP/UDP pseudo header checksum */
605 uint16_t is_cksum = 0; /* in case cksum should be offloaded */
608 if (txq->type == ETH_TUNTAP_TYPE_TUN) {
610 * TUN and TAP are created with IFF_NO_PI disabled.
611 * For TUN PMD this mandatory as fields are used by
612 * Kernel tun.c to determine whether its IP or non IP
615 * The logic fetches the first byte of data from mbuf
616 * then compares whether its v4 or v6. If first byte
617 * is 4 or 6, then protocol field is updated.
619 char *buff_data = rte_pktmbuf_mtod(seg, void *);
620 proto = (*buff_data & 0xf0);
621 pi.proto = (proto == 0x40) ?
622 rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4) :
624 rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6) :
629 iovecs[k].iov_base = π
630 iovecs[k].iov_len = sizeof(pi);
633 nb_segs = mbuf->nb_segs;
635 ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
636 (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
637 (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
640 /* Support only packets with at least layer 4
641 * header included in the first segment
643 seg_len = rte_pktmbuf_data_len(mbuf);
644 l234_hlen = mbuf->l2_len + mbuf->l3_len + mbuf->l4_len;
645 if (seg_len < l234_hlen)
648 /* To change checksums, work on a * copy of l2, l3
649 * headers + l4 pseudo header
651 rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
653 tap_tx_l3_cksum(m_copy, mbuf->ol_flags,
654 mbuf->l2_len, mbuf->l3_len, mbuf->l4_len,
655 &l4_cksum, &l4_phdr_cksum,
657 iovecs[k].iov_base = m_copy;
658 iovecs[k].iov_len = l234_hlen;
661 /* Update next iovecs[] beyond l2, l3, l4 headers */
662 if (seg_len > l234_hlen) {
663 iovecs[k].iov_len = seg_len - l234_hlen;
665 rte_pktmbuf_mtod(seg, char *) +
667 tap_tx_l4_add_rcksum(iovecs[k].iov_base,
668 iovecs[k].iov_len, l4_cksum,
676 for (j = k; j <= nb_segs; j++) {
677 iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
678 iovecs[j].iov_base = rte_pktmbuf_mtod(seg, void *);
680 tap_tx_l4_add_rcksum(iovecs[j].iov_base,
681 iovecs[j].iov_len, l4_cksum,
687 tap_tx_l4_cksum(l4_cksum, l4_phdr_cksum, l4_raw_cksum);
689 /* copy the tx frame data */
690 n = writev(process_private->txq_fds[txq->queue_id], iovecs, j);
695 (*num_tx_bytes) += rte_pktmbuf_pkt_len(mbuf);
700 /* Callback to handle sending packets from the tap interface
703 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
705 struct tx_queue *txq = queue;
707 uint16_t num_packets = 0;
708 unsigned long num_tx_bytes = 0;
712 if (unlikely(nb_pkts == 0))
715 struct rte_mbuf *gso_mbufs[MAX_GSO_MBUFS];
716 max_size = *txq->mtu + (RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN + 4);
717 for (i = 0; i < nb_pkts; i++) {
718 struct rte_mbuf *mbuf_in = bufs[num_tx];
719 struct rte_mbuf **mbuf;
720 uint16_t num_mbufs = 0;
721 uint16_t tso_segsz = 0;
727 tso = mbuf_in->ol_flags & PKT_TX_TCP_SEG;
729 struct rte_gso_ctx *gso_ctx = &txq->gso_ctx;
731 /* TCP segmentation implies TCP checksum offload */
732 mbuf_in->ol_flags |= PKT_TX_TCP_CKSUM;
734 /* gso size is calculated without RTE_ETHER_CRC_LEN */
735 hdrs_len = mbuf_in->l2_len + mbuf_in->l3_len +
737 tso_segsz = mbuf_in->tso_segsz + hdrs_len;
738 if (unlikely(tso_segsz == hdrs_len) ||
739 tso_segsz > *txq->mtu) {
743 gso_ctx->gso_size = tso_segsz;
744 /* 'mbuf_in' packet to segment */
745 num_tso_mbufs = rte_gso_segment(mbuf_in,
746 gso_ctx, /* gso control block */
747 (struct rte_mbuf **)&gso_mbufs, /* out mbufs */
748 RTE_DIM(gso_mbufs)); /* max tso mbufs */
750 /* ret contains the number of new created mbufs */
751 if (num_tso_mbufs < 0)
755 num_mbufs = num_tso_mbufs;
757 /* stats.errs will be incremented */
758 if (rte_pktmbuf_pkt_len(mbuf_in) > max_size)
761 /* ret 0 indicates no new mbufs were created */
767 ret = tap_write_mbufs(txq, num_mbufs, mbuf,
768 &num_packets, &num_tx_bytes);
772 if (num_tso_mbufs > 0)
773 rte_pktmbuf_free_bulk(mbuf, num_tso_mbufs);
777 /* free original mbuf */
778 rte_pktmbuf_free(mbuf_in);
780 if (num_tso_mbufs > 0)
781 rte_pktmbuf_free_bulk(mbuf, num_tso_mbufs);
784 txq->stats.opackets += num_packets;
785 txq->stats.errs += nb_pkts - num_tx;
786 txq->stats.obytes += num_tx_bytes;
792 tap_ioctl_req2str(unsigned long request)
796 return "SIOCSIFFLAGS";
798 return "SIOCGIFFLAGS";
800 return "SIOCGIFHWADDR";
802 return "SIOCSIFHWADDR";
810 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
811 struct ifreq *ifr, int set, enum ioctl_mode mode)
813 short req_flags = ifr->ifr_flags;
814 int remote = pmd->remote_if_index &&
815 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
817 if (!pmd->remote_if_index && mode == REMOTE_ONLY)
820 * If there is a remote netdevice, apply ioctl on it, then apply it on
825 strlcpy(ifr->ifr_name, pmd->remote_iface, IFNAMSIZ);
826 else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
827 strlcpy(ifr->ifr_name, pmd->name, IFNAMSIZ);
830 /* fetch current flags to leave other flags untouched */
831 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
834 ifr->ifr_flags |= req_flags;
836 ifr->ifr_flags &= ~req_flags;
844 TAP_LOG(WARNING, "%s: ioctl() called with wrong arg",
848 if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
850 if (remote-- && mode == LOCAL_AND_REMOTE)
855 TAP_LOG(DEBUG, "%s(%s) failed: %s(%d)", ifr->ifr_name,
856 tap_ioctl_req2str(request), strerror(errno), errno);
861 tap_link_set_down(struct rte_eth_dev *dev)
863 struct pmd_internals *pmd = dev->data->dev_private;
864 struct ifreq ifr = { .ifr_flags = IFF_UP };
866 dev->data->dev_link.link_status = ETH_LINK_DOWN;
867 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
871 tap_link_set_up(struct rte_eth_dev *dev)
873 struct pmd_internals *pmd = dev->data->dev_private;
874 struct ifreq ifr = { .ifr_flags = IFF_UP };
876 dev->data->dev_link.link_status = ETH_LINK_UP;
877 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
881 tap_dev_start(struct rte_eth_dev *dev)
885 err = tap_intr_handle_set(dev, 1);
889 err = tap_link_set_up(dev);
893 for (i = 0; i < dev->data->nb_tx_queues; i++)
894 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
895 for (i = 0; i < dev->data->nb_rx_queues; i++)
896 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
901 /* This function gets called when the current port gets stopped.
904 tap_dev_stop(struct rte_eth_dev *dev)
908 for (i = 0; i < dev->data->nb_tx_queues; i++)
909 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
910 for (i = 0; i < dev->data->nb_rx_queues; i++)
911 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
913 tap_intr_handle_set(dev, 0);
914 tap_link_set_down(dev);
920 tap_dev_configure(struct rte_eth_dev *dev)
922 struct pmd_internals *pmd = dev->data->dev_private;
924 if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
926 "%s: number of rx queues %d exceeds max num of queues %d",
928 dev->data->nb_rx_queues,
929 RTE_PMD_TAP_MAX_QUEUES);
932 if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
934 "%s: number of tx queues %d exceeds max num of queues %d",
936 dev->data->nb_tx_queues,
937 RTE_PMD_TAP_MAX_QUEUES);
941 TAP_LOG(INFO, "%s: %s: TX configured queues number: %u",
942 dev->device->name, pmd->name, dev->data->nb_tx_queues);
944 TAP_LOG(INFO, "%s: %s: RX configured queues number: %u",
945 dev->device->name, pmd->name, dev->data->nb_rx_queues);
951 tap_dev_speed_capa(void)
953 uint32_t speed = pmd_link.link_speed;
956 if (speed >= ETH_SPEED_NUM_10M)
957 capa |= ETH_LINK_SPEED_10M;
958 if (speed >= ETH_SPEED_NUM_100M)
959 capa |= ETH_LINK_SPEED_100M;
960 if (speed >= ETH_SPEED_NUM_1G)
961 capa |= ETH_LINK_SPEED_1G;
962 if (speed >= ETH_SPEED_NUM_5G)
963 capa |= ETH_LINK_SPEED_2_5G;
964 if (speed >= ETH_SPEED_NUM_5G)
965 capa |= ETH_LINK_SPEED_5G;
966 if (speed >= ETH_SPEED_NUM_10G)
967 capa |= ETH_LINK_SPEED_10G;
968 if (speed >= ETH_SPEED_NUM_20G)
969 capa |= ETH_LINK_SPEED_20G;
970 if (speed >= ETH_SPEED_NUM_25G)
971 capa |= ETH_LINK_SPEED_25G;
972 if (speed >= ETH_SPEED_NUM_40G)
973 capa |= ETH_LINK_SPEED_40G;
974 if (speed >= ETH_SPEED_NUM_50G)
975 capa |= ETH_LINK_SPEED_50G;
976 if (speed >= ETH_SPEED_NUM_56G)
977 capa |= ETH_LINK_SPEED_56G;
978 if (speed >= ETH_SPEED_NUM_100G)
979 capa |= ETH_LINK_SPEED_100G;
985 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
987 struct pmd_internals *internals = dev->data->dev_private;
989 dev_info->if_index = internals->if_index;
990 dev_info->max_mac_addrs = 1;
991 dev_info->max_rx_pktlen = (uint32_t)RTE_ETHER_MAX_VLAN_FRAME_LEN;
992 dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
993 dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
994 dev_info->min_rx_bufsize = 0;
995 dev_info->speed_capa = tap_dev_speed_capa();
996 dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
997 dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
998 dev_info->rx_queue_offload_capa;
999 dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
1000 dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
1001 dev_info->tx_queue_offload_capa;
1002 dev_info->hash_key_size = TAP_RSS_HASH_KEY_SIZE;
1004 * limitation: TAP supports all of IP, UDP and TCP hash
1005 * functions together and not in partial combinations
1007 dev_info->flow_type_rss_offloads = ~TAP_RSS_HF_MASK;
1013 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
1015 unsigned int i, imax;
1016 unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
1017 unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
1018 unsigned long rx_nombuf = 0, ierrors = 0;
1019 const struct pmd_internals *pmd = dev->data->dev_private;
1021 /* rx queue statistics */
1022 imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
1023 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
1024 for (i = 0; i < imax; i++) {
1025 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
1026 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
1027 rx_total += tap_stats->q_ipackets[i];
1028 rx_bytes_total += tap_stats->q_ibytes[i];
1029 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
1030 ierrors += pmd->rxq[i].stats.ierrors;
1033 /* tx queue statistics */
1034 imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
1035 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
1037 for (i = 0; i < imax; i++) {
1038 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
1039 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
1040 tx_total += tap_stats->q_opackets[i];
1041 tx_err_total += pmd->txq[i].stats.errs;
1042 tx_bytes_total += tap_stats->q_obytes[i];
1045 tap_stats->ipackets = rx_total;
1046 tap_stats->ibytes = rx_bytes_total;
1047 tap_stats->ierrors = ierrors;
1048 tap_stats->rx_nombuf = rx_nombuf;
1049 tap_stats->opackets = tx_total;
1050 tap_stats->oerrors = tx_err_total;
1051 tap_stats->obytes = tx_bytes_total;
1056 tap_stats_reset(struct rte_eth_dev *dev)
1059 struct pmd_internals *pmd = dev->data->dev_private;
1061 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1062 pmd->rxq[i].stats.ipackets = 0;
1063 pmd->rxq[i].stats.ibytes = 0;
1064 pmd->rxq[i].stats.ierrors = 0;
1065 pmd->rxq[i].stats.rx_nombuf = 0;
1067 pmd->txq[i].stats.opackets = 0;
1068 pmd->txq[i].stats.errs = 0;
1069 pmd->txq[i].stats.obytes = 0;
1076 tap_dev_close(struct rte_eth_dev *dev)
1079 struct pmd_internals *internals = dev->data->dev_private;
1080 struct pmd_process_private *process_private = dev->process_private;
1081 struct rx_queue *rxq;
1083 if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
1084 rte_free(dev->process_private);
1088 tap_link_set_down(dev);
1089 if (internals->nlsk_fd != -1) {
1090 tap_flow_flush(dev, NULL);
1091 tap_flow_implicit_flush(internals, NULL);
1092 tap_nl_final(internals->nlsk_fd);
1093 internals->nlsk_fd = -1;
1096 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1097 if (process_private->rxq_fds[i] != -1) {
1098 rxq = &internals->rxq[i];
1099 close(process_private->rxq_fds[i]);
1100 process_private->rxq_fds[i] = -1;
1101 tap_rxq_pool_free(rxq->pool);
1102 rte_free(rxq->iovecs);
1106 if (process_private->txq_fds[i] != -1) {
1107 close(process_private->txq_fds[i]);
1108 process_private->txq_fds[i] = -1;
1112 if (internals->remote_if_index) {
1113 /* Restore initial remote state */
1114 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
1115 &internals->remote_initial_flags);
1118 rte_mempool_free(internals->gso_ctx_mp);
1119 internals->gso_ctx_mp = NULL;
1121 if (internals->ka_fd != -1) {
1122 close(internals->ka_fd);
1123 internals->ka_fd = -1;
1126 /* mac_addrs must not be freed alone because part of dev_private */
1127 dev->data->mac_addrs = NULL;
1129 internals = dev->data->dev_private;
1130 TAP_LOG(DEBUG, "Closing %s Ethernet device on numa %u",
1131 tuntap_types[internals->type], rte_socket_id());
1133 if (internals->ioctl_sock != -1) {
1134 close(internals->ioctl_sock);
1135 internals->ioctl_sock = -1;
1137 rte_free(dev->process_private);
1138 if (tap_devices_count == 1)
1139 rte_mp_action_unregister(TAP_MP_KEY);
1140 tap_devices_count--;
1142 * Since TUN device has no more opened file descriptors
1143 * it will be removed from kernel
1150 tap_rx_queue_release(void *queue)
1152 struct rx_queue *rxq = queue;
1153 struct pmd_process_private *process_private;
1157 process_private = rte_eth_devices[rxq->in_port].process_private;
1158 if (process_private->rxq_fds[rxq->queue_id] != -1) {
1159 close(process_private->rxq_fds[rxq->queue_id]);
1160 process_private->rxq_fds[rxq->queue_id] = -1;
1161 tap_rxq_pool_free(rxq->pool);
1162 rte_free(rxq->iovecs);
1169 tap_tx_queue_release(void *queue)
1171 struct tx_queue *txq = queue;
1172 struct pmd_process_private *process_private;
1176 process_private = rte_eth_devices[txq->out_port].process_private;
1178 if (process_private->txq_fds[txq->queue_id] != -1) {
1179 close(process_private->txq_fds[txq->queue_id]);
1180 process_private->txq_fds[txq->queue_id] = -1;
1185 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
1187 struct rte_eth_link *dev_link = &dev->data->dev_link;
1188 struct pmd_internals *pmd = dev->data->dev_private;
1189 struct ifreq ifr = { .ifr_flags = 0 };
1191 if (pmd->remote_if_index) {
1192 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
1193 if (!(ifr.ifr_flags & IFF_UP) ||
1194 !(ifr.ifr_flags & IFF_RUNNING)) {
1195 dev_link->link_status = ETH_LINK_DOWN;
1199 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
1200 dev_link->link_status =
1201 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
1208 tap_promisc_enable(struct rte_eth_dev *dev)
1210 struct pmd_internals *pmd = dev->data->dev_private;
1211 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1214 ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1218 if (pmd->remote_if_index && !pmd->flow_isolate) {
1219 dev->data->promiscuous = 1;
1220 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
1222 /* Rollback promisc flag */
1223 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1225 * rte_eth_dev_promiscuous_enable() rollback
1226 * dev->data->promiscuous in the case of failure.
1236 tap_promisc_disable(struct rte_eth_dev *dev)
1238 struct pmd_internals *pmd = dev->data->dev_private;
1239 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1242 ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1246 if (pmd->remote_if_index && !pmd->flow_isolate) {
1247 dev->data->promiscuous = 0;
1248 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
1250 /* Rollback promisc flag */
1251 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1253 * rte_eth_dev_promiscuous_disable() rollback
1254 * dev->data->promiscuous in the case of failure.
1264 tap_allmulti_enable(struct rte_eth_dev *dev)
1266 struct pmd_internals *pmd = dev->data->dev_private;
1267 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1270 ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1274 if (pmd->remote_if_index && !pmd->flow_isolate) {
1275 dev->data->all_multicast = 1;
1276 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
1278 /* Rollback allmulti flag */
1279 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1281 * rte_eth_dev_allmulticast_enable() rollback
1282 * dev->data->all_multicast in the case of failure.
1292 tap_allmulti_disable(struct rte_eth_dev *dev)
1294 struct pmd_internals *pmd = dev->data->dev_private;
1295 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1298 ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1302 if (pmd->remote_if_index && !pmd->flow_isolate) {
1303 dev->data->all_multicast = 0;
1304 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
1306 /* Rollback allmulti flag */
1307 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1309 * rte_eth_dev_allmulticast_disable() rollback
1310 * dev->data->all_multicast in the case of failure.
1320 tap_mac_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1322 struct pmd_internals *pmd = dev->data->dev_private;
1323 enum ioctl_mode mode = LOCAL_ONLY;
1327 if (pmd->type == ETH_TUNTAP_TYPE_TUN) {
1328 TAP_LOG(ERR, "%s: can't MAC address for TUN",
1333 if (rte_is_zero_ether_addr(mac_addr)) {
1334 TAP_LOG(ERR, "%s: can't set an empty MAC address",
1338 /* Check the actual current MAC address on the tap netdevice */
1339 ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY);
1342 if (rte_is_same_ether_addr(
1343 (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1346 /* Check the current MAC address on the remote */
1347 ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY);
1350 if (!rte_is_same_ether_addr(
1351 (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1353 mode = LOCAL_AND_REMOTE;
1354 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1355 rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, RTE_ETHER_ADDR_LEN);
1356 ret = tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode);
1359 rte_memcpy(&pmd->eth_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1360 if (pmd->remote_if_index && !pmd->flow_isolate) {
1361 /* Replace MAC redirection rule after a MAC change */
1362 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC);
1365 "%s: Couldn't delete MAC redirection rule",
1369 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC);
1372 "%s: Couldn't add MAC redirection rule",
1382 tap_gso_ctx_setup(struct rte_gso_ctx *gso_ctx, struct rte_eth_dev *dev)
1386 struct pmd_internals *pmd = dev->data->dev_private;
1389 /* initialize GSO context */
1390 gso_types = DEV_TX_OFFLOAD_TCP_TSO;
1391 if (!pmd->gso_ctx_mp) {
1393 * Create private mbuf pool with TAP_GSO_MBUF_SEG_SIZE
1394 * bytes size per mbuf use this pool for both direct and
1397 ret = snprintf(pool_name, sizeof(pool_name), "mp_%s",
1399 if (ret < 0 || ret >= (int)sizeof(pool_name)) {
1401 "%s: failed to create mbuf pool name for device %s,"
1402 "device name too long or output error, ret: %d\n",
1403 pmd->name, dev->device->name, ret);
1404 return -ENAMETOOLONG;
1406 pmd->gso_ctx_mp = rte_pktmbuf_pool_create(pool_name,
1407 TAP_GSO_MBUFS_NUM, TAP_GSO_MBUF_CACHE_SIZE, 0,
1408 RTE_PKTMBUF_HEADROOM + TAP_GSO_MBUF_SEG_SIZE,
1410 if (!pmd->gso_ctx_mp) {
1412 "%s: failed to create mbuf pool for device %s\n",
1413 pmd->name, dev->device->name);
1418 gso_ctx->direct_pool = pmd->gso_ctx_mp;
1419 gso_ctx->indirect_pool = pmd->gso_ctx_mp;
1420 gso_ctx->gso_types = gso_types;
1421 gso_ctx->gso_size = 0; /* gso_size is set in tx_burst() per packet */
1428 tap_setup_queue(struct rte_eth_dev *dev,
1429 struct pmd_internals *internals,
1437 struct pmd_internals *pmd = dev->data->dev_private;
1438 struct pmd_process_private *process_private = dev->process_private;
1439 struct rx_queue *rx = &internals->rxq[qid];
1440 struct tx_queue *tx = &internals->txq[qid];
1441 struct rte_gso_ctx *gso_ctx;
1444 fd = &process_private->rxq_fds[qid];
1445 other_fd = &process_private->txq_fds[qid];
1449 fd = &process_private->txq_fds[qid];
1450 other_fd = &process_private->rxq_fds[qid];
1452 gso_ctx = &tx->gso_ctx;
1455 /* fd for this queue already exists */
1456 TAP_LOG(DEBUG, "%s: fd %d for %s queue qid %d exists",
1457 pmd->name, *fd, dir, qid);
1459 } else if (*other_fd != -1) {
1460 /* Only other_fd exists. dup it */
1461 *fd = dup(*other_fd);
1464 TAP_LOG(ERR, "%s: dup() failed.", pmd->name);
1467 TAP_LOG(DEBUG, "%s: dup fd %d for %s queue qid %d (%d)",
1468 pmd->name, *other_fd, dir, qid, *fd);
1470 /* Both RX and TX fds do not exist (equal -1). Create fd */
1471 *fd = tun_alloc(pmd, 0);
1473 *fd = -1; /* restore original value */
1474 TAP_LOG(ERR, "%s: tun_alloc() failed.", pmd->name);
1477 TAP_LOG(DEBUG, "%s: add %s queue for qid %d fd %d",
1478 pmd->name, dir, qid, *fd);
1481 tx->mtu = &dev->data->mtu;
1482 rx->rxmode = &dev->data->dev_conf.rxmode;
1484 ret = tap_gso_ctx_setup(gso_ctx, dev);
1489 tx->type = pmd->type;
1495 tap_rx_queue_setup(struct rte_eth_dev *dev,
1496 uint16_t rx_queue_id,
1497 uint16_t nb_rx_desc,
1498 unsigned int socket_id,
1499 const struct rte_eth_rxconf *rx_conf __rte_unused,
1500 struct rte_mempool *mp)
1502 struct pmd_internals *internals = dev->data->dev_private;
1503 struct pmd_process_private *process_private = dev->process_private;
1504 struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1505 struct rte_mbuf **tmp = &rxq->pool;
1506 long iov_max = sysconf(_SC_IOV_MAX);
1510 "_SC_IOV_MAX is not defined. Using %d as default",
1511 TAP_IOV_DEFAULT_MAX);
1512 iov_max = TAP_IOV_DEFAULT_MAX;
1514 uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1515 struct iovec (*iovecs)[nb_desc + 1];
1516 int data_off = RTE_PKTMBUF_HEADROOM;
1521 if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1523 "nb_rx_queues %d too small or mempool NULL",
1524 dev->data->nb_rx_queues);
1529 rxq->trigger_seen = 1; /* force initial burst */
1530 rxq->in_port = dev->data->port_id;
1531 rxq->queue_id = rx_queue_id;
1532 rxq->nb_rx_desc = nb_desc;
1533 iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1537 "%s: Couldn't allocate %d RX descriptors",
1538 dev->device->name, nb_desc);
1541 rxq->iovecs = iovecs;
1543 dev->data->rx_queues[rx_queue_id] = rxq;
1544 fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1550 (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1551 (*rxq->iovecs)[0].iov_base = &rxq->pi;
1553 for (i = 1; i <= nb_desc; i++) {
1554 *tmp = rte_pktmbuf_alloc(rxq->mp);
1557 "%s: couldn't allocate memory for queue %d",
1558 dev->device->name, rx_queue_id);
1562 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1563 (*rxq->iovecs)[i].iov_base =
1564 (char *)(*tmp)->buf_addr + data_off;
1566 tmp = &(*tmp)->next;
1569 TAP_LOG(DEBUG, " RX TUNTAP device name %s, qid %d on fd %d",
1570 internals->name, rx_queue_id,
1571 process_private->rxq_fds[rx_queue_id]);
1576 tap_rxq_pool_free(rxq->pool);
1578 rte_free(rxq->iovecs);
1584 tap_tx_queue_setup(struct rte_eth_dev *dev,
1585 uint16_t tx_queue_id,
1586 uint16_t nb_tx_desc __rte_unused,
1587 unsigned int socket_id __rte_unused,
1588 const struct rte_eth_txconf *tx_conf)
1590 struct pmd_internals *internals = dev->data->dev_private;
1591 struct pmd_process_private *process_private = dev->process_private;
1592 struct tx_queue *txq;
1596 if (tx_queue_id >= dev->data->nb_tx_queues)
1598 dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1599 txq = dev->data->tx_queues[tx_queue_id];
1600 txq->out_port = dev->data->port_id;
1601 txq->queue_id = tx_queue_id;
1603 offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1604 txq->csum = !!(offloads &
1605 (DEV_TX_OFFLOAD_IPV4_CKSUM |
1606 DEV_TX_OFFLOAD_UDP_CKSUM |
1607 DEV_TX_OFFLOAD_TCP_CKSUM));
1609 ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1613 " TX TUNTAP device name %s, qid %d on fd %d csum %s",
1614 internals->name, tx_queue_id,
1615 process_private->txq_fds[tx_queue_id],
1616 txq->csum ? "on" : "off");
1622 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1624 struct pmd_internals *pmd = dev->data->dev_private;
1625 struct ifreq ifr = { .ifr_mtu = mtu };
1628 err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1630 dev->data->mtu = mtu;
1636 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1637 struct rte_ether_addr *mc_addr_set __rte_unused,
1638 uint32_t nb_mc_addr __rte_unused)
1641 * Nothing to do actually: the tap has no filtering whatsoever, every
1642 * packet is received.
1648 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1650 struct rte_eth_dev *dev = arg;
1651 struct pmd_internals *pmd = dev->data->dev_private;
1652 struct ifinfomsg *info = NLMSG_DATA(nh);
1654 if (nh->nlmsg_type != RTM_NEWLINK ||
1655 (info->ifi_index != pmd->if_index &&
1656 info->ifi_index != pmd->remote_if_index))
1658 return tap_link_update(dev, 0);
1662 tap_dev_intr_handler(void *cb_arg)
1664 struct rte_eth_dev *dev = cb_arg;
1665 struct pmd_internals *pmd = dev->data->dev_private;
1667 tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1671 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1673 struct pmd_internals *pmd = dev->data->dev_private;
1676 /* In any case, disable interrupt if the conf is no longer there. */
1677 if (!dev->data->dev_conf.intr_conf.lsc) {
1678 if (pmd->intr_handle.fd != -1) {
1684 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1685 if (unlikely(pmd->intr_handle.fd == -1))
1687 return rte_intr_callback_register(
1688 &pmd->intr_handle, tap_dev_intr_handler, dev);
1693 ret = rte_intr_callback_unregister(&pmd->intr_handle,
1694 tap_dev_intr_handler, dev);
1697 } else if (ret == -EAGAIN) {
1700 TAP_LOG(ERR, "intr callback unregister failed: %d",
1706 tap_nl_final(pmd->intr_handle.fd);
1707 pmd->intr_handle.fd = -1;
1713 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1717 err = tap_lsc_intr_handle_set(dev, set);
1720 tap_rx_intr_vec_set(dev, 0);
1723 err = tap_rx_intr_vec_set(dev, set);
1725 tap_lsc_intr_handle_set(dev, 0);
1729 static const uint32_t*
1730 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1732 static const uint32_t ptypes[] = {
1733 RTE_PTYPE_INNER_L2_ETHER,
1734 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1735 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1736 RTE_PTYPE_INNER_L3_IPV4,
1737 RTE_PTYPE_INNER_L3_IPV4_EXT,
1738 RTE_PTYPE_INNER_L3_IPV6,
1739 RTE_PTYPE_INNER_L3_IPV6_EXT,
1740 RTE_PTYPE_INNER_L4_FRAG,
1741 RTE_PTYPE_INNER_L4_UDP,
1742 RTE_PTYPE_INNER_L4_TCP,
1743 RTE_PTYPE_INNER_L4_SCTP,
1745 RTE_PTYPE_L2_ETHER_VLAN,
1746 RTE_PTYPE_L2_ETHER_QINQ,
1748 RTE_PTYPE_L3_IPV4_EXT,
1749 RTE_PTYPE_L3_IPV6_EXT,
1761 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1762 struct rte_eth_fc_conf *fc_conf)
1764 fc_conf->mode = RTE_FC_NONE;
1769 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1770 struct rte_eth_fc_conf *fc_conf)
1772 if (fc_conf->mode != RTE_FC_NONE)
1778 * DPDK callback to update the RSS hash configuration.
1781 * Pointer to Ethernet device structure.
1782 * @param[in] rss_conf
1783 * RSS configuration data.
1786 * 0 on success, a negative errno value otherwise and rte_errno is set.
1789 tap_rss_hash_update(struct rte_eth_dev *dev,
1790 struct rte_eth_rss_conf *rss_conf)
1792 if (rss_conf->rss_hf & TAP_RSS_HF_MASK) {
1796 if (rss_conf->rss_key && rss_conf->rss_key_len) {
1798 * Currently TAP RSS key is hard coded
1799 * and cannot be updated
1802 "port %u RSS key cannot be updated",
1803 dev->data->port_id);
1811 tap_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1813 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1819 tap_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1821 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1827 tap_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1829 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1835 tap_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1837 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1841 static const struct eth_dev_ops ops = {
1842 .dev_start = tap_dev_start,
1843 .dev_stop = tap_dev_stop,
1844 .dev_close = tap_dev_close,
1845 .dev_configure = tap_dev_configure,
1846 .dev_infos_get = tap_dev_info,
1847 .rx_queue_setup = tap_rx_queue_setup,
1848 .tx_queue_setup = tap_tx_queue_setup,
1849 .rx_queue_start = tap_rx_queue_start,
1850 .tx_queue_start = tap_tx_queue_start,
1851 .rx_queue_stop = tap_rx_queue_stop,
1852 .tx_queue_stop = tap_tx_queue_stop,
1853 .rx_queue_release = tap_rx_queue_release,
1854 .tx_queue_release = tap_tx_queue_release,
1855 .flow_ctrl_get = tap_flow_ctrl_get,
1856 .flow_ctrl_set = tap_flow_ctrl_set,
1857 .link_update = tap_link_update,
1858 .dev_set_link_up = tap_link_set_up,
1859 .dev_set_link_down = tap_link_set_down,
1860 .promiscuous_enable = tap_promisc_enable,
1861 .promiscuous_disable = tap_promisc_disable,
1862 .allmulticast_enable = tap_allmulti_enable,
1863 .allmulticast_disable = tap_allmulti_disable,
1864 .mac_addr_set = tap_mac_set,
1865 .mtu_set = tap_mtu_set,
1866 .set_mc_addr_list = tap_set_mc_addr_list,
1867 .stats_get = tap_stats_get,
1868 .stats_reset = tap_stats_reset,
1869 .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1870 .rss_hash_update = tap_rss_hash_update,
1871 .filter_ctrl = tap_dev_filter_ctrl,
1875 eth_dev_tap_create(struct rte_vdev_device *vdev, const char *tap_name,
1876 char *remote_iface, struct rte_ether_addr *mac_addr,
1877 enum rte_tuntap_type type)
1879 int numa_node = rte_socket_id();
1880 struct rte_eth_dev *dev;
1881 struct pmd_internals *pmd;
1882 struct pmd_process_private *process_private;
1883 const char *tuntap_name = tuntap_types[type];
1884 struct rte_eth_dev_data *data;
1888 TAP_LOG(DEBUG, "%s device on numa %u", tuntap_name, rte_socket_id());
1890 dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1892 TAP_LOG(ERR, "%s Unable to allocate device struct",
1894 goto error_exit_nodev;
1897 process_private = (struct pmd_process_private *)
1898 rte_zmalloc_socket(tap_name, sizeof(struct pmd_process_private),
1899 RTE_CACHE_LINE_SIZE, dev->device->numa_node);
1901 if (process_private == NULL) {
1902 TAP_LOG(ERR, "Failed to alloc memory for process private");
1905 pmd = dev->data->dev_private;
1906 dev->process_private = process_private;
1908 strlcpy(pmd->name, tap_name, sizeof(pmd->name));
1912 pmd->gso_ctx_mp = NULL;
1914 pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1915 if (pmd->ioctl_sock == -1) {
1917 "%s Unable to get a socket for management: %s",
1918 tuntap_name, strerror(errno));
1922 /* Setup some default values */
1924 data->dev_private = pmd;
1925 data->dev_flags = RTE_ETH_DEV_INTR_LSC |
1926 RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
1927 data->numa_node = numa_node;
1929 data->dev_link = pmd_link;
1930 data->mac_addrs = &pmd->eth_addr;
1931 /* Set the number of RX and TX queues */
1932 data->nb_rx_queues = 0;
1933 data->nb_tx_queues = 0;
1935 dev->dev_ops = &ops;
1936 dev->rx_pkt_burst = pmd_rx_burst;
1937 dev->tx_pkt_burst = pmd_tx_burst;
1939 pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1940 pmd->intr_handle.fd = -1;
1941 dev->intr_handle = &pmd->intr_handle;
1943 /* Presetup the fds to -1 as being not valid */
1944 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1945 process_private->rxq_fds[i] = -1;
1946 process_private->txq_fds[i] = -1;
1949 if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1950 if (rte_is_zero_ether_addr(mac_addr))
1951 rte_eth_random_addr((uint8_t *)&pmd->eth_addr);
1953 rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1957 * Allocate a TUN device keep-alive file descriptor that will only be
1958 * closed when the TUN device itself is closed or removed.
1959 * This keep-alive file descriptor will guarantee that the TUN device
1960 * exists even when all of its queues are closed
1962 pmd->ka_fd = tun_alloc(pmd, 1);
1963 if (pmd->ka_fd == -1) {
1964 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
1967 TAP_LOG(DEBUG, "allocated %s", pmd->name);
1969 ifr.ifr_mtu = dev->data->mtu;
1970 if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1973 if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1974 memset(&ifr, 0, sizeof(struct ifreq));
1975 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1976 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
1977 RTE_ETHER_ADDR_LEN);
1978 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1983 * Set up everything related to rte_flow:
1985 * - tap / remote if_index
1986 * - mandatory QDISCs
1987 * - rte_flow actual/implicit lists
1990 pmd->nlsk_fd = tap_nl_init(0);
1991 if (pmd->nlsk_fd == -1) {
1992 TAP_LOG(WARNING, "%s: failed to create netlink socket.",
1994 goto disable_rte_flow;
1996 pmd->if_index = if_nametoindex(pmd->name);
1997 if (!pmd->if_index) {
1998 TAP_LOG(ERR, "%s: failed to get if_index.", pmd->name);
1999 goto disable_rte_flow;
2001 if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
2002 TAP_LOG(ERR, "%s: failed to create multiq qdisc.",
2004 goto disable_rte_flow;
2006 if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
2007 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
2009 goto disable_rte_flow;
2011 LIST_INIT(&pmd->flows);
2013 if (strlen(remote_iface)) {
2014 pmd->remote_if_index = if_nametoindex(remote_iface);
2015 if (!pmd->remote_if_index) {
2016 TAP_LOG(ERR, "%s: failed to get %s if_index.",
2017 pmd->name, remote_iface);
2020 strlcpy(pmd->remote_iface, remote_iface, RTE_ETH_NAME_MAX_LEN);
2022 /* Save state of remote device */
2023 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
2025 /* Replicate remote MAC address */
2026 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
2027 TAP_LOG(ERR, "%s: failed to get %s MAC address.",
2028 pmd->name, pmd->remote_iface);
2031 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
2032 RTE_ETHER_ADDR_LEN);
2033 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
2034 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
2035 TAP_LOG(ERR, "%s: failed to get %s MAC address.",
2036 pmd->name, remote_iface);
2041 * Flush usually returns negative value because it tries to
2042 * delete every QDISC (and on a running device, one QDISC at
2043 * least is needed). Ignore negative return value.
2045 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
2046 if (qdisc_create_ingress(pmd->nlsk_fd,
2047 pmd->remote_if_index) < 0) {
2048 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
2052 LIST_INIT(&pmd->implicit_flows);
2053 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
2054 tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
2055 tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
2056 tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
2058 "%s: failed to create implicit rules.",
2064 rte_eth_dev_probing_finish(dev);
2068 TAP_LOG(ERR, " Disabling rte flow support: %s(%d)",
2069 strerror(errno), errno);
2070 if (strlen(remote_iface)) {
2071 TAP_LOG(ERR, "Remote feature requires flow support.");
2074 rte_eth_dev_probing_finish(dev);
2078 TAP_LOG(ERR, " Can't set up remote feature: %s(%d)",
2079 strerror(errno), errno);
2080 tap_flow_implicit_flush(pmd, NULL);
2083 if (pmd->nlsk_fd != -1)
2084 close(pmd->nlsk_fd);
2085 if (pmd->ka_fd != -1)
2087 if (pmd->ioctl_sock != -1)
2088 close(pmd->ioctl_sock);
2089 /* mac_addrs must not be freed alone because part of dev_private */
2090 dev->data->mac_addrs = NULL;
2091 rte_eth_dev_release_port(dev);
2094 TAP_LOG(ERR, "%s Unable to initialize %s",
2095 tuntap_name, rte_vdev_device_name(vdev));
2100 /* make sure name is a possible Linux network device name */
2102 is_valid_iface(const char *name)
2107 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
2111 if (*name == '/' || *name == ':' || isspace(*name))
2119 set_interface_name(const char *key __rte_unused,
2123 char *name = (char *)extra_args;
2126 if (!is_valid_iface(value)) {
2127 TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
2131 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
2133 /* use tap%d which causes kernel to choose next available */
2134 strlcpy(name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2140 set_remote_iface(const char *key __rte_unused,
2144 char *name = (char *)extra_args;
2147 if (!is_valid_iface(value)) {
2148 TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
2152 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
2158 static int parse_user_mac(struct rte_ether_addr *user_mac,
2161 unsigned int index = 0;
2162 char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
2164 if (user_mac == NULL || value == NULL)
2167 strlcpy(mac_temp, value, sizeof(mac_temp));
2168 mac_byte = strtok(mac_temp, ":");
2170 while ((mac_byte != NULL) &&
2171 (strlen(mac_byte) <= 2) &&
2172 (strlen(mac_byte) == strspn(mac_byte,
2173 ETH_TAP_CMP_MAC_FMT))) {
2174 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
2175 mac_byte = strtok(NULL, ":");
2182 set_mac_type(const char *key __rte_unused,
2186 struct rte_ether_addr *user_mac = extra_args;
2191 if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
2192 static int iface_idx;
2194 /* fixed mac = 00:64:74:61:70:<iface_idx> */
2195 memcpy((char *)user_mac->addr_bytes, "\0dtap",
2196 RTE_ETHER_ADDR_LEN);
2197 user_mac->addr_bytes[RTE_ETHER_ADDR_LEN - 1] =
2202 if (parse_user_mac(user_mac, value) != 6)
2205 TAP_LOG(DEBUG, "TAP user MAC param (%s)", value);
2209 TAP_LOG(ERR, "TAP user MAC (%s) is not in format (%s|%s)",
2210 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
2215 * Open a TUN interface device. TUN PMD
2216 * 1) sets tap_type as false
2217 * 2) intakes iface as argument.
2218 * 3) as interface is virtual set speed to 10G
2221 rte_pmd_tun_probe(struct rte_vdev_device *dev)
2223 const char *name, *params;
2225 struct rte_kvargs *kvlist = NULL;
2226 char tun_name[RTE_ETH_NAME_MAX_LEN];
2227 char remote_iface[RTE_ETH_NAME_MAX_LEN];
2228 struct rte_eth_dev *eth_dev;
2230 name = rte_vdev_device_name(dev);
2231 params = rte_vdev_device_args(dev);
2232 memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2234 if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
2235 strlen(params) == 0) {
2236 eth_dev = rte_eth_dev_attach_secondary(name);
2238 TAP_LOG(ERR, "Failed to probe %s", name);
2241 eth_dev->dev_ops = &ops;
2242 eth_dev->device = &dev->device;
2243 rte_eth_dev_probing_finish(eth_dev);
2247 /* use tun%d which causes kernel to choose next available */
2248 strlcpy(tun_name, DEFAULT_TUN_NAME "%d", 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,
2266 pmd_link.link_speed = ETH_SPEED_NUM_10G;
2268 TAP_LOG(DEBUG, "Initializing pmd_tun for %s", name);
2270 ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0,
2271 ETH_TUNTAP_TYPE_TUN);
2275 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2278 rte_kvargs_free(kvlist);
2283 /* Request queue file descriptors from secondary to primary. */
2285 tap_mp_attach_queues(const char *port_name, struct rte_eth_dev *dev)
2288 struct timespec timeout = {.tv_sec = 1, .tv_nsec = 0};
2289 struct rte_mp_msg request, *reply;
2290 struct rte_mp_reply replies;
2291 struct ipc_queues *request_param = (struct ipc_queues *)request.param;
2292 struct ipc_queues *reply_param;
2293 struct pmd_process_private *process_private = dev->process_private;
2294 int queue, fd_iterator;
2296 /* Prepare the request */
2297 memset(&request, 0, sizeof(request));
2298 strlcpy(request.name, TAP_MP_KEY, sizeof(request.name));
2299 strlcpy(request_param->port_name, port_name,
2300 sizeof(request_param->port_name));
2301 request.len_param = sizeof(*request_param);
2302 /* Send request and receive reply */
2303 ret = rte_mp_request_sync(&request, &replies, &timeout);
2304 if (ret < 0 || replies.nb_received != 1) {
2305 TAP_LOG(ERR, "Failed to request queues from primary: %d",
2309 reply = &replies.msgs[0];
2310 reply_param = (struct ipc_queues *)reply->param;
2311 TAP_LOG(DEBUG, "Received IPC reply for %s", reply_param->port_name);
2313 /* Attach the queues from received file descriptors */
2314 if (reply_param->rxq_count + reply_param->txq_count != reply->num_fds) {
2315 TAP_LOG(ERR, "Unexpected number of fds received");
2319 dev->data->nb_rx_queues = reply_param->rxq_count;
2320 dev->data->nb_tx_queues = reply_param->txq_count;
2322 for (queue = 0; queue < reply_param->rxq_count; queue++)
2323 process_private->rxq_fds[queue] = reply->fds[fd_iterator++];
2324 for (queue = 0; queue < reply_param->txq_count; queue++)
2325 process_private->txq_fds[queue] = reply->fds[fd_iterator++];
2330 /* Send the queue file descriptors from the primary process to secondary. */
2332 tap_mp_sync_queues(const struct rte_mp_msg *request, const void *peer)
2334 struct rte_eth_dev *dev;
2335 struct pmd_process_private *process_private;
2336 struct rte_mp_msg reply;
2337 const struct ipc_queues *request_param =
2338 (const struct ipc_queues *)request->param;
2339 struct ipc_queues *reply_param =
2340 (struct ipc_queues *)reply.param;
2345 /* Get requested port */
2346 TAP_LOG(DEBUG, "Received IPC request for %s", request_param->port_name);
2347 ret = rte_eth_dev_get_port_by_name(request_param->port_name, &port_id);
2349 TAP_LOG(ERR, "Failed to get port id for %s",
2350 request_param->port_name);
2353 dev = &rte_eth_devices[port_id];
2354 process_private = dev->process_private;
2356 /* Fill file descriptors for all queues */
2358 reply_param->rxq_count = 0;
2359 if (dev->data->nb_rx_queues + dev->data->nb_tx_queues >
2361 TAP_LOG(ERR, "Number of rx/tx queues exceeds max number of fds");
2365 for (queue = 0; queue < dev->data->nb_rx_queues; queue++) {
2366 reply.fds[reply.num_fds++] = process_private->rxq_fds[queue];
2367 reply_param->rxq_count++;
2369 RTE_ASSERT(reply_param->rxq_count == dev->data->nb_rx_queues);
2371 reply_param->txq_count = 0;
2372 for (queue = 0; queue < dev->data->nb_tx_queues; queue++) {
2373 reply.fds[reply.num_fds++] = process_private->txq_fds[queue];
2374 reply_param->txq_count++;
2376 RTE_ASSERT(reply_param->txq_count == dev->data->nb_tx_queues);
2379 strlcpy(reply.name, request->name, sizeof(reply.name));
2380 strlcpy(reply_param->port_name, request_param->port_name,
2381 sizeof(reply_param->port_name));
2382 reply.len_param = sizeof(*reply_param);
2383 if (rte_mp_reply(&reply, peer) < 0) {
2384 TAP_LOG(ERR, "Failed to reply an IPC request to sync queues");
2390 /* Open a TAP interface device.
2393 rte_pmd_tap_probe(struct rte_vdev_device *dev)
2395 const char *name, *params;
2397 struct rte_kvargs *kvlist = NULL;
2399 char tap_name[RTE_ETH_NAME_MAX_LEN];
2400 char remote_iface[RTE_ETH_NAME_MAX_LEN];
2401 struct rte_ether_addr user_mac = { .addr_bytes = {0} };
2402 struct rte_eth_dev *eth_dev;
2403 int tap_devices_count_increased = 0;
2405 name = rte_vdev_device_name(dev);
2406 params = rte_vdev_device_args(dev);
2408 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
2409 eth_dev = rte_eth_dev_attach_secondary(name);
2411 TAP_LOG(ERR, "Failed to probe %s", name);
2414 eth_dev->dev_ops = &ops;
2415 eth_dev->device = &dev->device;
2416 eth_dev->rx_pkt_burst = pmd_rx_burst;
2417 eth_dev->tx_pkt_burst = pmd_tx_burst;
2418 if (!rte_eal_primary_proc_alive(NULL)) {
2419 TAP_LOG(ERR, "Primary process is missing");
2422 eth_dev->process_private = (struct pmd_process_private *)
2423 rte_zmalloc_socket(name,
2424 sizeof(struct pmd_process_private),
2425 RTE_CACHE_LINE_SIZE,
2426 eth_dev->device->numa_node);
2427 if (eth_dev->process_private == NULL) {
2429 "Failed to alloc memory for process private");
2433 ret = tap_mp_attach_queues(name, eth_dev);
2436 rte_eth_dev_probing_finish(eth_dev);
2440 speed = ETH_SPEED_NUM_10G;
2442 /* use tap%d which causes kernel to choose next available */
2443 strlcpy(tap_name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2444 memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2446 if (params && (params[0] != '\0')) {
2447 TAP_LOG(DEBUG, "parameters (%s)", params);
2449 kvlist = rte_kvargs_parse(params, valid_arguments);
2451 if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2452 ret = rte_kvargs_process(kvlist,
2454 &set_interface_name,
2460 if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
2461 ret = rte_kvargs_process(kvlist,
2469 if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
2470 ret = rte_kvargs_process(kvlist,
2479 pmd_link.link_speed = speed;
2481 TAP_LOG(DEBUG, "Initializing pmd_tap for %s", name);
2483 /* Register IPC feed callback */
2484 if (!tap_devices_count) {
2485 ret = rte_mp_action_register(TAP_MP_KEY, tap_mp_sync_queues);
2486 if (ret < 0 && rte_errno != ENOTSUP) {
2487 TAP_LOG(ERR, "tap: Failed to register IPC callback: %s",
2488 strerror(rte_errno));
2492 tap_devices_count++;
2493 tap_devices_count_increased = 1;
2494 ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac,
2495 ETH_TUNTAP_TYPE_TAP);
2499 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2501 if (tap_devices_count_increased == 1) {
2502 if (tap_devices_count == 1)
2503 rte_mp_action_unregister(TAP_MP_KEY);
2504 tap_devices_count--;
2507 rte_kvargs_free(kvlist);
2512 /* detach a TUNTAP device.
2515 rte_pmd_tap_remove(struct rte_vdev_device *dev)
2517 struct rte_eth_dev *eth_dev = NULL;
2519 /* find the ethdev entry */
2520 eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
2524 tap_dev_close(eth_dev);
2525 rte_eth_dev_release_port(eth_dev);
2530 static struct rte_vdev_driver pmd_tun_drv = {
2531 .probe = rte_pmd_tun_probe,
2532 .remove = rte_pmd_tap_remove,
2535 static struct rte_vdev_driver pmd_tap_drv = {
2536 .probe = rte_pmd_tap_probe,
2537 .remove = rte_pmd_tap_remove,
2540 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
2541 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
2542 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
2543 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
2544 ETH_TAP_IFACE_ARG "=<string> ");
2545 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
2546 ETH_TAP_IFACE_ARG "=<string> "
2547 ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
2548 ETH_TAP_REMOTE_ARG "=<string>");
2549 RTE_LOG_REGISTER(tap_logtype, pmd.net.tap, NOTICE);