1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2016-2017 Intel Corporation
5 #include <rte_atomic.h>
6 #include <rte_branch_prediction.h>
7 #include <rte_byteorder.h>
8 #include <rte_common.h>
10 #include <rte_ethdev_driver.h>
11 #include <rte_ethdev_vdev.h>
12 #include <rte_malloc.h>
13 #include <rte_bus_vdev.h>
14 #include <rte_kvargs.h>
16 #include <rte_debug.h>
18 #include <rte_string_fns.h>
19 #include <rte_ethdev.h>
20 #include <rte_errno.h>
22 #include <sys/types.h>
24 #include <sys/socket.h>
25 #include <sys/ioctl.h>
26 #include <sys/utsname.h>
34 #include <arpa/inet.h>
36 #include <linux/if_tun.h>
37 #include <linux/if_ether.h>
42 #include <rte_eth_tap.h>
44 #include <tap_netlink.h>
45 #include <tap_tcmsgs.h>
47 /* Linux based path to the TUN device */
48 #define TUN_TAP_DEV_PATH "/dev/net/tun"
49 #define DEFAULT_TAP_NAME "dtap"
50 #define DEFAULT_TUN_NAME "dtun"
52 #define ETH_TAP_IFACE_ARG "iface"
53 #define ETH_TAP_REMOTE_ARG "remote"
54 #define ETH_TAP_MAC_ARG "mac"
55 #define ETH_TAP_MAC_FIXED "fixed"
57 #define ETH_TAP_USR_MAC_FMT "xx:xx:xx:xx:xx:xx"
58 #define ETH_TAP_CMP_MAC_FMT "0123456789ABCDEFabcdef"
59 #define ETH_TAP_MAC_ARG_FMT ETH_TAP_MAC_FIXED "|" ETH_TAP_USR_MAC_FMT
61 #define TAP_GSO_MBUFS_PER_CORE 128
62 #define TAP_GSO_MBUF_SEG_SIZE 128
63 #define TAP_GSO_MBUF_CACHE_SIZE 4
64 #define TAP_GSO_MBUFS_NUM \
65 (TAP_GSO_MBUFS_PER_CORE * TAP_GSO_MBUF_CACHE_SIZE)
67 /* IPC key for queue fds sync */
68 #define TAP_MP_KEY "tap_mp_sync_queues"
70 #define TAP_IOV_DEFAULT_MAX 1024
72 static int tap_devices_count;
74 static const char *valid_arguments[] = {
81 static volatile uint32_t tap_trigger; /* Rx trigger */
83 static struct rte_eth_link pmd_link = {
84 .link_speed = ETH_SPEED_NUM_10G,
85 .link_duplex = ETH_LINK_FULL_DUPLEX,
86 .link_status = ETH_LINK_DOWN,
87 .link_autoneg = ETH_LINK_FIXED,
91 tap_trigger_cb(int sig __rte_unused)
93 /* Valid trigger values are nonzero */
94 tap_trigger = (tap_trigger + 1) | 0x80000000;
97 /* Specifies on what netdevices the ioctl should be applied */
104 /* Message header to synchronize queues via IPC */
106 char port_name[RTE_DEV_NAME_MAX_LEN];
110 * The file descriptors are in the dedicated part
111 * of the Unix message to be translated by the kernel.
115 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
118 * Tun/Tap allocation routine
121 * Pointer to private structure.
123 * @param[in] is_keepalive
127 * -1 on failure, fd on success
130 tun_alloc(struct pmd_internals *pmd, int is_keepalive)
133 #ifdef IFF_MULTI_QUEUE
134 unsigned int features;
138 memset(&ifr, 0, sizeof(struct ifreq));
141 * Do not set IFF_NO_PI as packet information header will be needed
142 * to check if a received packet has been truncated.
144 ifr.ifr_flags = (pmd->type == ETH_TUNTAP_TYPE_TAP) ?
145 IFF_TAP : IFF_TUN | IFF_POINTOPOINT;
146 strlcpy(ifr.ifr_name, pmd->name, IFNAMSIZ);
148 fd = open(TUN_TAP_DEV_PATH, O_RDWR);
150 TAP_LOG(ERR, "Unable to open %s interface", TUN_TAP_DEV_PATH);
154 #ifdef IFF_MULTI_QUEUE
155 /* Grab the TUN features to verify we can work multi-queue */
156 if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
157 TAP_LOG(ERR, "unable to get TUN/TAP features");
160 TAP_LOG(DEBUG, "%s Features %08x", TUN_TAP_DEV_PATH, features);
162 if (features & IFF_MULTI_QUEUE) {
163 TAP_LOG(DEBUG, " Multi-queue support for %d queues",
164 RTE_PMD_TAP_MAX_QUEUES);
165 ifr.ifr_flags |= IFF_MULTI_QUEUE;
169 ifr.ifr_flags |= IFF_ONE_QUEUE;
170 TAP_LOG(DEBUG, " Single queue only support");
173 /* Set the TUN/TAP configuration and set the name if needed */
174 if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
175 TAP_LOG(WARNING, "Unable to set TUNSETIFF for %s: %s",
176 ifr.ifr_name, strerror(errno));
181 * Name passed to kernel might be wildcard like dtun%d
182 * and need to find the resulting device.
184 TAP_LOG(DEBUG, "Device name is '%s'", ifr.ifr_name);
185 strlcpy(pmd->name, ifr.ifr_name, RTE_ETH_NAME_MAX_LEN);
189 * Detach the TUN/TAP keep-alive queue
190 * to avoid traffic through it
192 ifr.ifr_flags = IFF_DETACH_QUEUE;
193 if (ioctl(fd, TUNSETQUEUE, (void *)&ifr) < 0) {
195 "Unable to detach keep-alive queue for %s: %s",
196 ifr.ifr_name, strerror(errno));
201 /* Always set the file descriptor to non-blocking */
202 if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
204 "Unable to set %s to nonblocking: %s",
205 ifr.ifr_name, strerror(errno));
209 /* Set up trigger to optimize empty Rx bursts */
213 int flags = fcntl(fd, F_GETFL);
215 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
217 if (sa.sa_handler != tap_trigger_cb) {
219 * Make sure SIGIO is not already taken. This is done
220 * as late as possible to leave the application a
221 * chance to set up its own signal handler first.
223 if (sa.sa_handler != SIG_IGN &&
224 sa.sa_handler != SIG_DFL) {
228 sa = (struct sigaction){
229 .sa_flags = SA_RESTART,
230 .sa_handler = tap_trigger_cb,
232 if (sigaction(SIGIO, &sa, NULL) == -1)
235 /* Enable SIGIO on file descriptor */
236 fcntl(fd, F_SETFL, flags | O_ASYNC);
237 fcntl(fd, F_SETOWN, getpid());
241 /* Disable trigger globally in case of error */
243 TAP_LOG(WARNING, "Rx trigger disabled: %s",
256 tap_verify_csum(struct rte_mbuf *mbuf)
258 uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
259 uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
260 uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
261 unsigned int l2_len = sizeof(struct rte_ether_hdr);
267 if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
269 else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
271 /* Don't verify checksum for packets with discontinuous L2 header */
272 if (unlikely(l2_len + sizeof(struct rte_ipv4_hdr) >
273 rte_pktmbuf_data_len(mbuf)))
275 l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
276 if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
277 struct rte_ipv4_hdr *iph = l3_hdr;
279 /* ihl contains the number of 4-byte words in the header */
280 l3_len = 4 * (iph->version_ihl & 0xf);
281 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
283 /* check that the total length reported by header is not
284 * greater than the total received size
286 if (l2_len + rte_be_to_cpu_16(iph->total_length) >
287 rte_pktmbuf_data_len(mbuf))
290 cksum = ~rte_raw_cksum(iph, l3_len);
291 mbuf->ol_flags |= cksum ?
292 PKT_RX_IP_CKSUM_BAD :
293 PKT_RX_IP_CKSUM_GOOD;
294 } else if (l3 == RTE_PTYPE_L3_IPV6) {
295 struct rte_ipv6_hdr *iph = l3_hdr;
297 l3_len = sizeof(struct rte_ipv6_hdr);
298 /* check that the total length reported by header is not
299 * greater than the total received size
301 if (l2_len + l3_len + rte_be_to_cpu_16(iph->payload_len) >
302 rte_pktmbuf_data_len(mbuf))
305 /* IPv6 extensions are not supported */
308 if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
309 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
310 /* Don't verify checksum for multi-segment packets. */
311 if (mbuf->nb_segs > 1)
313 if (l3 == RTE_PTYPE_L3_IPV4)
314 cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
315 else if (l3 == RTE_PTYPE_L3_IPV6)
316 cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
317 mbuf->ol_flags |= cksum ?
318 PKT_RX_L4_CKSUM_BAD :
319 PKT_RX_L4_CKSUM_GOOD;
324 tap_rx_offload_get_port_capa(void)
327 * No specific port Rx offload capabilities.
333 tap_rx_offload_get_queue_capa(void)
335 return DEV_RX_OFFLOAD_SCATTER |
336 DEV_RX_OFFLOAD_IPV4_CKSUM |
337 DEV_RX_OFFLOAD_UDP_CKSUM |
338 DEV_RX_OFFLOAD_TCP_CKSUM;
342 tap_rxq_pool_free(struct rte_mbuf *pool)
344 struct rte_mbuf *mbuf = pool;
345 uint16_t nb_segs = 1;
354 pool->nb_segs = nb_segs;
355 rte_pktmbuf_free(pool);
358 /* Callback to handle the rx burst of packets to the correct interface and
359 * file descriptor(s) in a multi-queue setup.
362 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
364 struct rx_queue *rxq = queue;
365 struct pmd_process_private *process_private;
367 unsigned long num_rx_bytes = 0;
368 uint32_t trigger = tap_trigger;
370 if (trigger == rxq->trigger_seen)
373 process_private = rte_eth_devices[rxq->in_port].process_private;
374 for (num_rx = 0; num_rx < nb_pkts; ) {
375 struct rte_mbuf *mbuf = rxq->pool;
376 struct rte_mbuf *seg = NULL;
377 struct rte_mbuf *new_tail = NULL;
378 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
381 len = readv(process_private->rxq_fds[rxq->queue_id],
383 1 + (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
384 rxq->nb_rx_desc : 1));
385 if (len < (int)sizeof(struct tun_pi))
388 /* Packet couldn't fit in the provided mbuf */
389 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
390 rxq->stats.ierrors++;
394 len -= sizeof(struct tun_pi);
397 mbuf->port = rxq->in_port;
399 struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
401 if (unlikely(!buf)) {
402 rxq->stats.rx_nombuf++;
403 /* No new buf has been allocated: do nothing */
404 if (!new_tail || !seg)
408 tap_rxq_pool_free(mbuf);
412 seg = seg ? seg->next : mbuf;
413 if (rxq->pool == mbuf)
416 new_tail->next = buf;
418 new_tail->next = seg->next;
420 /* iovecs[0] is reserved for packet info (pi) */
421 (*rxq->iovecs)[mbuf->nb_segs].iov_len =
422 buf->buf_len - data_off;
423 (*rxq->iovecs)[mbuf->nb_segs].iov_base =
424 (char *)buf->buf_addr + data_off;
426 seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
427 seg->data_off = data_off;
429 len -= seg->data_len;
433 /* First segment has headroom, not the others */
437 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
439 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
440 tap_verify_csum(mbuf);
442 /* account for the receive frame */
443 bufs[num_rx++] = mbuf;
444 num_rx_bytes += mbuf->pkt_len;
447 rxq->stats.ipackets += num_rx;
448 rxq->stats.ibytes += num_rx_bytes;
450 if (trigger && num_rx < nb_pkts)
451 rxq->trigger_seen = trigger;
457 tap_tx_offload_get_port_capa(void)
460 * No specific port Tx offload capabilities.
466 tap_tx_offload_get_queue_capa(void)
468 return DEV_TX_OFFLOAD_MULTI_SEGS |
469 DEV_TX_OFFLOAD_IPV4_CKSUM |
470 DEV_TX_OFFLOAD_UDP_CKSUM |
471 DEV_TX_OFFLOAD_TCP_CKSUM |
472 DEV_TX_OFFLOAD_TCP_TSO;
475 /* Finalize l4 checksum calculation */
477 tap_tx_l4_cksum(uint16_t *l4_cksum, uint16_t l4_phdr_cksum,
478 uint32_t l4_raw_cksum)
483 cksum = __rte_raw_cksum_reduce(l4_raw_cksum);
484 cksum += l4_phdr_cksum;
486 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
487 cksum = (~cksum) & 0xffff;
494 /* Accumaulate L4 raw checksums */
496 tap_tx_l4_add_rcksum(char *l4_data, unsigned int l4_len, uint16_t *l4_cksum,
497 uint32_t *l4_raw_cksum)
499 if (l4_cksum == NULL)
502 *l4_raw_cksum = __rte_raw_cksum(l4_data, l4_len, *l4_raw_cksum);
505 /* L3 and L4 pseudo headers checksum offloads */
507 tap_tx_l3_cksum(char *packet, uint64_t ol_flags, unsigned int l2_len,
508 unsigned int l3_len, unsigned int l4_len, uint16_t **l4_cksum,
509 uint16_t *l4_phdr_cksum, uint32_t *l4_raw_cksum)
511 void *l3_hdr = packet + l2_len;
513 if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
514 struct rte_ipv4_hdr *iph = l3_hdr;
517 iph->hdr_checksum = 0;
518 cksum = rte_raw_cksum(iph, l3_len);
519 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
521 if (ol_flags & PKT_TX_L4_MASK) {
524 l4_hdr = packet + l2_len + l3_len;
525 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
526 *l4_cksum = &((struct rte_udp_hdr *)l4_hdr)->dgram_cksum;
527 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
528 *l4_cksum = &((struct rte_tcp_hdr *)l4_hdr)->cksum;
532 if (ol_flags & PKT_TX_IPV4)
533 *l4_phdr_cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
535 *l4_phdr_cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
536 *l4_raw_cksum = __rte_raw_cksum(l4_hdr, l4_len, 0);
541 tap_write_mbufs(struct tx_queue *txq, uint16_t num_mbufs,
542 struct rte_mbuf **pmbufs,
543 uint16_t *num_packets, unsigned long *num_tx_bytes)
547 struct pmd_process_private *process_private;
549 process_private = rte_eth_devices[txq->out_port].process_private;
551 for (i = 0; i < num_mbufs; i++) {
552 struct rte_mbuf *mbuf = pmbufs[i];
553 struct iovec iovecs[mbuf->nb_segs + 2];
554 struct tun_pi pi = { .flags = 0, .proto = 0x00 };
555 struct rte_mbuf *seg = mbuf;
556 char m_copy[mbuf->data_len];
560 int k; /* current index in iovecs for copying segments */
561 uint16_t seg_len; /* length of first segment */
563 uint16_t *l4_cksum; /* l4 checksum (pseudo header + payload) */
564 uint32_t l4_raw_cksum = 0; /* TCP/UDP payload raw checksum */
565 uint16_t l4_phdr_cksum = 0; /* TCP/UDP pseudo header checksum */
566 uint16_t is_cksum = 0; /* in case cksum should be offloaded */
569 if (txq->type == ETH_TUNTAP_TYPE_TUN) {
571 * TUN and TAP are created with IFF_NO_PI disabled.
572 * For TUN PMD this mandatory as fields are used by
573 * Kernel tun.c to determine whether its IP or non IP
576 * The logic fetches the first byte of data from mbuf
577 * then compares whether its v4 or v6. If first byte
578 * is 4 or 6, then protocol field is updated.
580 char *buff_data = rte_pktmbuf_mtod(seg, void *);
581 proto = (*buff_data & 0xf0);
582 pi.proto = (proto == 0x40) ?
583 rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4) :
585 rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6) :
590 iovecs[k].iov_base = π
591 iovecs[k].iov_len = sizeof(pi);
594 nb_segs = mbuf->nb_segs;
596 ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
597 (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
598 (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
601 /* Support only packets with at least layer 4
602 * header included in the first segment
604 seg_len = rte_pktmbuf_data_len(mbuf);
605 l234_hlen = mbuf->l2_len + mbuf->l3_len + mbuf->l4_len;
606 if (seg_len < l234_hlen)
609 /* To change checksums, work on a * copy of l2, l3
610 * headers + l4 pseudo header
612 rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
614 tap_tx_l3_cksum(m_copy, mbuf->ol_flags,
615 mbuf->l2_len, mbuf->l3_len, mbuf->l4_len,
616 &l4_cksum, &l4_phdr_cksum,
618 iovecs[k].iov_base = m_copy;
619 iovecs[k].iov_len = l234_hlen;
622 /* Update next iovecs[] beyond l2, l3, l4 headers */
623 if (seg_len > l234_hlen) {
624 iovecs[k].iov_len = seg_len - l234_hlen;
626 rte_pktmbuf_mtod(seg, char *) +
628 tap_tx_l4_add_rcksum(iovecs[k].iov_base,
629 iovecs[k].iov_len, l4_cksum,
637 for (j = k; j <= nb_segs; j++) {
638 iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
639 iovecs[j].iov_base = rte_pktmbuf_mtod(seg, void *);
641 tap_tx_l4_add_rcksum(iovecs[j].iov_base,
642 iovecs[j].iov_len, l4_cksum,
648 tap_tx_l4_cksum(l4_cksum, l4_phdr_cksum, l4_raw_cksum);
650 /* copy the tx frame data */
651 n = writev(process_private->txq_fds[txq->queue_id], iovecs, j);
656 (*num_tx_bytes) += rte_pktmbuf_pkt_len(mbuf);
661 /* Callback to handle sending packets from the tap interface
664 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
666 struct tx_queue *txq = queue;
668 uint16_t num_packets = 0;
669 unsigned long num_tx_bytes = 0;
673 if (unlikely(nb_pkts == 0))
676 struct rte_mbuf *gso_mbufs[MAX_GSO_MBUFS];
677 max_size = *txq->mtu + (RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN + 4);
678 for (i = 0; i < nb_pkts; i++) {
679 struct rte_mbuf *mbuf_in = bufs[num_tx];
680 struct rte_mbuf **mbuf;
681 uint16_t num_mbufs = 0;
682 uint16_t tso_segsz = 0;
688 tso = mbuf_in->ol_flags & PKT_TX_TCP_SEG;
690 struct rte_gso_ctx *gso_ctx = &txq->gso_ctx;
692 /* TCP segmentation implies TCP checksum offload */
693 mbuf_in->ol_flags |= PKT_TX_TCP_CKSUM;
695 /* gso size is calculated without RTE_ETHER_CRC_LEN */
696 hdrs_len = mbuf_in->l2_len + mbuf_in->l3_len +
698 tso_segsz = mbuf_in->tso_segsz + hdrs_len;
699 if (unlikely(tso_segsz == hdrs_len) ||
700 tso_segsz > *txq->mtu) {
704 gso_ctx->gso_size = tso_segsz;
705 /* 'mbuf_in' packet to segment */
706 num_tso_mbufs = rte_gso_segment(mbuf_in,
707 gso_ctx, /* gso control block */
708 (struct rte_mbuf **)&gso_mbufs, /* out mbufs */
709 RTE_DIM(gso_mbufs)); /* max tso mbufs */
711 /* ret contains the number of new created mbufs */
712 if (num_tso_mbufs < 0)
716 num_mbufs = num_tso_mbufs;
718 /* stats.errs will be incremented */
719 if (rte_pktmbuf_pkt_len(mbuf_in) > max_size)
722 /* ret 0 indicates no new mbufs were created */
728 ret = tap_write_mbufs(txq, num_mbufs, mbuf,
729 &num_packets, &num_tx_bytes);
733 if (num_tso_mbufs > 0)
734 rte_pktmbuf_free_bulk(mbuf, num_tso_mbufs);
738 /* free original mbuf */
739 rte_pktmbuf_free(mbuf_in);
741 if (num_tso_mbufs > 0)
742 rte_pktmbuf_free_bulk(mbuf, num_tso_mbufs);
745 txq->stats.opackets += num_packets;
746 txq->stats.errs += nb_pkts - num_tx;
747 txq->stats.obytes += num_tx_bytes;
753 tap_ioctl_req2str(unsigned long request)
757 return "SIOCSIFFLAGS";
759 return "SIOCGIFFLAGS";
761 return "SIOCGIFHWADDR";
763 return "SIOCSIFHWADDR";
771 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
772 struct ifreq *ifr, int set, enum ioctl_mode mode)
774 short req_flags = ifr->ifr_flags;
775 int remote = pmd->remote_if_index &&
776 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
778 if (!pmd->remote_if_index && mode == REMOTE_ONLY)
781 * If there is a remote netdevice, apply ioctl on it, then apply it on
786 strlcpy(ifr->ifr_name, pmd->remote_iface, IFNAMSIZ);
787 else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
788 strlcpy(ifr->ifr_name, pmd->name, IFNAMSIZ);
791 /* fetch current flags to leave other flags untouched */
792 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
795 ifr->ifr_flags |= req_flags;
797 ifr->ifr_flags &= ~req_flags;
805 TAP_LOG(WARNING, "%s: ioctl() called with wrong arg",
809 if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
811 if (remote-- && mode == LOCAL_AND_REMOTE)
816 TAP_LOG(DEBUG, "%s(%s) failed: %s(%d)", ifr->ifr_name,
817 tap_ioctl_req2str(request), strerror(errno), errno);
822 tap_link_set_down(struct rte_eth_dev *dev)
824 struct pmd_internals *pmd = dev->data->dev_private;
825 struct ifreq ifr = { .ifr_flags = IFF_UP };
827 dev->data->dev_link.link_status = ETH_LINK_DOWN;
828 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
832 tap_link_set_up(struct rte_eth_dev *dev)
834 struct pmd_internals *pmd = dev->data->dev_private;
835 struct ifreq ifr = { .ifr_flags = IFF_UP };
837 dev->data->dev_link.link_status = ETH_LINK_UP;
838 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
842 tap_dev_start(struct rte_eth_dev *dev)
846 err = tap_intr_handle_set(dev, 1);
850 err = tap_link_set_up(dev);
854 for (i = 0; i < dev->data->nb_tx_queues; i++)
855 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
856 for (i = 0; i < dev->data->nb_rx_queues; i++)
857 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
862 /* This function gets called when the current port gets stopped.
865 tap_dev_stop(struct rte_eth_dev *dev)
869 for (i = 0; i < dev->data->nb_tx_queues; i++)
870 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
871 for (i = 0; i < dev->data->nb_rx_queues; i++)
872 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
874 tap_intr_handle_set(dev, 0);
875 tap_link_set_down(dev);
879 tap_dev_configure(struct rte_eth_dev *dev)
881 struct pmd_internals *pmd = dev->data->dev_private;
883 if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
885 "%s: number of rx queues %d exceeds max num of queues %d",
887 dev->data->nb_rx_queues,
888 RTE_PMD_TAP_MAX_QUEUES);
891 if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
893 "%s: number of tx queues %d exceeds max num of queues %d",
895 dev->data->nb_tx_queues,
896 RTE_PMD_TAP_MAX_QUEUES);
900 TAP_LOG(INFO, "%s: %s: TX configured queues number: %u",
901 dev->device->name, pmd->name, dev->data->nb_tx_queues);
903 TAP_LOG(INFO, "%s: %s: RX configured queues number: %u",
904 dev->device->name, pmd->name, dev->data->nb_rx_queues);
910 tap_dev_speed_capa(void)
912 uint32_t speed = pmd_link.link_speed;
915 if (speed >= ETH_SPEED_NUM_10M)
916 capa |= ETH_LINK_SPEED_10M;
917 if (speed >= ETH_SPEED_NUM_100M)
918 capa |= ETH_LINK_SPEED_100M;
919 if (speed >= ETH_SPEED_NUM_1G)
920 capa |= ETH_LINK_SPEED_1G;
921 if (speed >= ETH_SPEED_NUM_5G)
922 capa |= ETH_LINK_SPEED_2_5G;
923 if (speed >= ETH_SPEED_NUM_5G)
924 capa |= ETH_LINK_SPEED_5G;
925 if (speed >= ETH_SPEED_NUM_10G)
926 capa |= ETH_LINK_SPEED_10G;
927 if (speed >= ETH_SPEED_NUM_20G)
928 capa |= ETH_LINK_SPEED_20G;
929 if (speed >= ETH_SPEED_NUM_25G)
930 capa |= ETH_LINK_SPEED_25G;
931 if (speed >= ETH_SPEED_NUM_40G)
932 capa |= ETH_LINK_SPEED_40G;
933 if (speed >= ETH_SPEED_NUM_50G)
934 capa |= ETH_LINK_SPEED_50G;
935 if (speed >= ETH_SPEED_NUM_56G)
936 capa |= ETH_LINK_SPEED_56G;
937 if (speed >= ETH_SPEED_NUM_100G)
938 capa |= ETH_LINK_SPEED_100G;
944 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
946 struct pmd_internals *internals = dev->data->dev_private;
948 dev_info->if_index = internals->if_index;
949 dev_info->max_mac_addrs = 1;
950 dev_info->max_rx_pktlen = (uint32_t)RTE_ETHER_MAX_VLAN_FRAME_LEN;
951 dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
952 dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
953 dev_info->min_rx_bufsize = 0;
954 dev_info->speed_capa = tap_dev_speed_capa();
955 dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
956 dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
957 dev_info->rx_queue_offload_capa;
958 dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
959 dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
960 dev_info->tx_queue_offload_capa;
961 dev_info->hash_key_size = TAP_RSS_HASH_KEY_SIZE;
963 * limitation: TAP supports all of IP, UDP and TCP hash
964 * functions together and not in partial combinations
966 dev_info->flow_type_rss_offloads = ~TAP_RSS_HF_MASK;
972 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
974 unsigned int i, imax;
975 unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
976 unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
977 unsigned long rx_nombuf = 0, ierrors = 0;
978 const struct pmd_internals *pmd = dev->data->dev_private;
980 /* rx queue statistics */
981 imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
982 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
983 for (i = 0; i < imax; i++) {
984 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
985 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
986 rx_total += tap_stats->q_ipackets[i];
987 rx_bytes_total += tap_stats->q_ibytes[i];
988 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
989 ierrors += pmd->rxq[i].stats.ierrors;
992 /* tx queue statistics */
993 imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
994 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
996 for (i = 0; i < imax; i++) {
997 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
998 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
999 tx_total += tap_stats->q_opackets[i];
1000 tx_err_total += pmd->txq[i].stats.errs;
1001 tx_bytes_total += tap_stats->q_obytes[i];
1004 tap_stats->ipackets = rx_total;
1005 tap_stats->ibytes = rx_bytes_total;
1006 tap_stats->ierrors = ierrors;
1007 tap_stats->rx_nombuf = rx_nombuf;
1008 tap_stats->opackets = tx_total;
1009 tap_stats->oerrors = tx_err_total;
1010 tap_stats->obytes = tx_bytes_total;
1015 tap_stats_reset(struct rte_eth_dev *dev)
1018 struct pmd_internals *pmd = dev->data->dev_private;
1020 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1021 pmd->rxq[i].stats.ipackets = 0;
1022 pmd->rxq[i].stats.ibytes = 0;
1023 pmd->rxq[i].stats.ierrors = 0;
1024 pmd->rxq[i].stats.rx_nombuf = 0;
1026 pmd->txq[i].stats.opackets = 0;
1027 pmd->txq[i].stats.errs = 0;
1028 pmd->txq[i].stats.obytes = 0;
1035 tap_dev_close(struct rte_eth_dev *dev)
1038 struct pmd_internals *internals = dev->data->dev_private;
1039 struct pmd_process_private *process_private = dev->process_private;
1040 struct rx_queue *rxq;
1042 tap_link_set_down(dev);
1043 if (internals->nlsk_fd != -1) {
1044 tap_flow_flush(dev, NULL);
1045 tap_flow_implicit_flush(internals, NULL);
1046 tap_nl_final(internals->nlsk_fd);
1047 internals->nlsk_fd = -1;
1050 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1051 if (process_private->rxq_fds[i] != -1) {
1052 rxq = &internals->rxq[i];
1053 close(process_private->rxq_fds[i]);
1054 process_private->rxq_fds[i] = -1;
1055 tap_rxq_pool_free(rxq->pool);
1056 rte_free(rxq->iovecs);
1060 if (process_private->txq_fds[i] != -1) {
1061 close(process_private->txq_fds[i]);
1062 process_private->txq_fds[i] = -1;
1066 if (internals->remote_if_index) {
1067 /* Restore initial remote state */
1068 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
1069 &internals->remote_initial_flags);
1072 if (internals->ka_fd != -1) {
1073 close(internals->ka_fd);
1074 internals->ka_fd = -1;
1077 * Since TUN device has no more opened file descriptors
1078 * it will be removed from kernel
1083 tap_rx_queue_release(void *queue)
1085 struct rx_queue *rxq = queue;
1086 struct pmd_process_private *process_private;
1090 process_private = rte_eth_devices[rxq->in_port].process_private;
1091 if (process_private->rxq_fds[rxq->queue_id] > 0) {
1092 close(process_private->rxq_fds[rxq->queue_id]);
1093 process_private->rxq_fds[rxq->queue_id] = -1;
1094 tap_rxq_pool_free(rxq->pool);
1095 rte_free(rxq->iovecs);
1102 tap_tx_queue_release(void *queue)
1104 struct tx_queue *txq = queue;
1105 struct pmd_process_private *process_private;
1109 process_private = rte_eth_devices[txq->out_port].process_private;
1111 if (process_private->txq_fds[txq->queue_id] > 0) {
1112 close(process_private->txq_fds[txq->queue_id]);
1113 process_private->txq_fds[txq->queue_id] = -1;
1118 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
1120 struct rte_eth_link *dev_link = &dev->data->dev_link;
1121 struct pmd_internals *pmd = dev->data->dev_private;
1122 struct ifreq ifr = { .ifr_flags = 0 };
1124 if (pmd->remote_if_index) {
1125 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
1126 if (!(ifr.ifr_flags & IFF_UP) ||
1127 !(ifr.ifr_flags & IFF_RUNNING)) {
1128 dev_link->link_status = ETH_LINK_DOWN;
1132 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
1133 dev_link->link_status =
1134 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
1141 tap_promisc_enable(struct rte_eth_dev *dev)
1143 struct pmd_internals *pmd = dev->data->dev_private;
1144 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1147 ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1151 if (pmd->remote_if_index && !pmd->flow_isolate) {
1152 dev->data->promiscuous = 1;
1153 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
1155 /* Rollback promisc flag */
1156 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1158 * rte_eth_dev_promiscuous_enable() rollback
1159 * dev->data->promiscuous in the case of failure.
1169 tap_promisc_disable(struct rte_eth_dev *dev)
1171 struct pmd_internals *pmd = dev->data->dev_private;
1172 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1175 ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1179 if (pmd->remote_if_index && !pmd->flow_isolate) {
1180 dev->data->promiscuous = 0;
1181 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
1183 /* Rollback promisc flag */
1184 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1186 * rte_eth_dev_promiscuous_disable() rollback
1187 * dev->data->promiscuous in the case of failure.
1197 tap_allmulti_enable(struct rte_eth_dev *dev)
1199 struct pmd_internals *pmd = dev->data->dev_private;
1200 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1203 ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1207 if (pmd->remote_if_index && !pmd->flow_isolate) {
1208 dev->data->all_multicast = 1;
1209 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
1211 /* Rollback allmulti flag */
1212 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1214 * rte_eth_dev_allmulticast_enable() rollback
1215 * dev->data->all_multicast in the case of failure.
1225 tap_allmulti_disable(struct rte_eth_dev *dev)
1227 struct pmd_internals *pmd = dev->data->dev_private;
1228 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1231 ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1235 if (pmd->remote_if_index && !pmd->flow_isolate) {
1236 dev->data->all_multicast = 0;
1237 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
1239 /* Rollback allmulti flag */
1240 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1242 * rte_eth_dev_allmulticast_disable() rollback
1243 * dev->data->all_multicast in the case of failure.
1253 tap_mac_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1255 struct pmd_internals *pmd = dev->data->dev_private;
1256 enum ioctl_mode mode = LOCAL_ONLY;
1260 if (pmd->type == ETH_TUNTAP_TYPE_TUN) {
1261 TAP_LOG(ERR, "%s: can't MAC address for TUN",
1266 if (rte_is_zero_ether_addr(mac_addr)) {
1267 TAP_LOG(ERR, "%s: can't set an empty MAC address",
1271 /* Check the actual current MAC address on the tap netdevice */
1272 ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY);
1275 if (rte_is_same_ether_addr(
1276 (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1279 /* Check the current MAC address on the remote */
1280 ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY);
1283 if (!rte_is_same_ether_addr(
1284 (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1286 mode = LOCAL_AND_REMOTE;
1287 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1288 rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, RTE_ETHER_ADDR_LEN);
1289 ret = tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode);
1292 rte_memcpy(&pmd->eth_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1293 if (pmd->remote_if_index && !pmd->flow_isolate) {
1294 /* Replace MAC redirection rule after a MAC change */
1295 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC);
1298 "%s: Couldn't delete MAC redirection rule",
1302 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC);
1305 "%s: Couldn't add MAC redirection rule",
1315 tap_gso_ctx_setup(struct rte_gso_ctx *gso_ctx, struct rte_eth_dev *dev)
1321 * Create private mbuf pool with TAP_GSO_MBUF_SEG_SIZE bytes
1322 * size per mbuf use this pool for both direct and indirect mbufs
1325 struct rte_mempool *mp; /* Mempool for GSO packets */
1327 /* initialize GSO context */
1328 gso_types = DEV_TX_OFFLOAD_TCP_TSO;
1329 snprintf(pool_name, sizeof(pool_name), "mp_%s", dev->device->name);
1330 mp = rte_mempool_lookup((const char *)pool_name);
1332 mp = rte_pktmbuf_pool_create(pool_name, TAP_GSO_MBUFS_NUM,
1333 TAP_GSO_MBUF_CACHE_SIZE, 0,
1334 RTE_PKTMBUF_HEADROOM + TAP_GSO_MBUF_SEG_SIZE,
1337 struct pmd_internals *pmd = dev->data->dev_private;
1340 "%s: failed to create mbuf pool for device %s\n",
1341 pmd->name, dev->device->name);
1346 gso_ctx->direct_pool = mp;
1347 gso_ctx->indirect_pool = mp;
1348 gso_ctx->gso_types = gso_types;
1349 gso_ctx->gso_size = 0; /* gso_size is set in tx_burst() per packet */
1356 tap_setup_queue(struct rte_eth_dev *dev,
1357 struct pmd_internals *internals,
1365 struct pmd_internals *pmd = dev->data->dev_private;
1366 struct pmd_process_private *process_private = dev->process_private;
1367 struct rx_queue *rx = &internals->rxq[qid];
1368 struct tx_queue *tx = &internals->txq[qid];
1369 struct rte_gso_ctx *gso_ctx;
1372 fd = &process_private->rxq_fds[qid];
1373 other_fd = &process_private->txq_fds[qid];
1377 fd = &process_private->txq_fds[qid];
1378 other_fd = &process_private->rxq_fds[qid];
1380 gso_ctx = &tx->gso_ctx;
1383 /* fd for this queue already exists */
1384 TAP_LOG(DEBUG, "%s: fd %d for %s queue qid %d exists",
1385 pmd->name, *fd, dir, qid);
1387 } else if (*other_fd != -1) {
1388 /* Only other_fd exists. dup it */
1389 *fd = dup(*other_fd);
1392 TAP_LOG(ERR, "%s: dup() failed.", pmd->name);
1395 TAP_LOG(DEBUG, "%s: dup fd %d for %s queue qid %d (%d)",
1396 pmd->name, *other_fd, dir, qid, *fd);
1398 /* Both RX and TX fds do not exist (equal -1). Create fd */
1399 *fd = tun_alloc(pmd, 0);
1401 *fd = -1; /* restore original value */
1402 TAP_LOG(ERR, "%s: tun_alloc() failed.", pmd->name);
1405 TAP_LOG(DEBUG, "%s: add %s queue for qid %d fd %d",
1406 pmd->name, dir, qid, *fd);
1409 tx->mtu = &dev->data->mtu;
1410 rx->rxmode = &dev->data->dev_conf.rxmode;
1412 ret = tap_gso_ctx_setup(gso_ctx, dev);
1417 tx->type = pmd->type;
1423 tap_rx_queue_setup(struct rte_eth_dev *dev,
1424 uint16_t rx_queue_id,
1425 uint16_t nb_rx_desc,
1426 unsigned int socket_id,
1427 const struct rte_eth_rxconf *rx_conf __rte_unused,
1428 struct rte_mempool *mp)
1430 struct pmd_internals *internals = dev->data->dev_private;
1431 struct pmd_process_private *process_private = dev->process_private;
1432 struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1433 struct rte_mbuf **tmp = &rxq->pool;
1434 long iov_max = sysconf(_SC_IOV_MAX);
1438 "_SC_IOV_MAX is not defined. Using %d as default",
1439 TAP_IOV_DEFAULT_MAX);
1440 iov_max = TAP_IOV_DEFAULT_MAX;
1442 uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1443 struct iovec (*iovecs)[nb_desc + 1];
1444 int data_off = RTE_PKTMBUF_HEADROOM;
1449 if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1451 "nb_rx_queues %d too small or mempool NULL",
1452 dev->data->nb_rx_queues);
1457 rxq->trigger_seen = 1; /* force initial burst */
1458 rxq->in_port = dev->data->port_id;
1459 rxq->queue_id = rx_queue_id;
1460 rxq->nb_rx_desc = nb_desc;
1461 iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1465 "%s: Couldn't allocate %d RX descriptors",
1466 dev->device->name, nb_desc);
1469 rxq->iovecs = iovecs;
1471 dev->data->rx_queues[rx_queue_id] = rxq;
1472 fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1478 (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1479 (*rxq->iovecs)[0].iov_base = &rxq->pi;
1481 for (i = 1; i <= nb_desc; i++) {
1482 *tmp = rte_pktmbuf_alloc(rxq->mp);
1485 "%s: couldn't allocate memory for queue %d",
1486 dev->device->name, rx_queue_id);
1490 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1491 (*rxq->iovecs)[i].iov_base =
1492 (char *)(*tmp)->buf_addr + data_off;
1494 tmp = &(*tmp)->next;
1497 TAP_LOG(DEBUG, " RX TUNTAP device name %s, qid %d on fd %d",
1498 internals->name, rx_queue_id,
1499 process_private->rxq_fds[rx_queue_id]);
1504 tap_rxq_pool_free(rxq->pool);
1506 rte_free(rxq->iovecs);
1512 tap_tx_queue_setup(struct rte_eth_dev *dev,
1513 uint16_t tx_queue_id,
1514 uint16_t nb_tx_desc __rte_unused,
1515 unsigned int socket_id __rte_unused,
1516 const struct rte_eth_txconf *tx_conf)
1518 struct pmd_internals *internals = dev->data->dev_private;
1519 struct pmd_process_private *process_private = dev->process_private;
1520 struct tx_queue *txq;
1524 if (tx_queue_id >= dev->data->nb_tx_queues)
1526 dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1527 txq = dev->data->tx_queues[tx_queue_id];
1528 txq->out_port = dev->data->port_id;
1529 txq->queue_id = tx_queue_id;
1531 offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1532 txq->csum = !!(offloads &
1533 (DEV_TX_OFFLOAD_IPV4_CKSUM |
1534 DEV_TX_OFFLOAD_UDP_CKSUM |
1535 DEV_TX_OFFLOAD_TCP_CKSUM));
1537 ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1541 " TX TUNTAP device name %s, qid %d on fd %d csum %s",
1542 internals->name, tx_queue_id,
1543 process_private->txq_fds[tx_queue_id],
1544 txq->csum ? "on" : "off");
1550 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1552 struct pmd_internals *pmd = dev->data->dev_private;
1553 struct ifreq ifr = { .ifr_mtu = mtu };
1556 err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1558 dev->data->mtu = mtu;
1564 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1565 struct rte_ether_addr *mc_addr_set __rte_unused,
1566 uint32_t nb_mc_addr __rte_unused)
1569 * Nothing to do actually: the tap has no filtering whatsoever, every
1570 * packet is received.
1576 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1578 struct rte_eth_dev *dev = arg;
1579 struct pmd_internals *pmd = dev->data->dev_private;
1580 struct ifinfomsg *info = NLMSG_DATA(nh);
1582 if (nh->nlmsg_type != RTM_NEWLINK ||
1583 (info->ifi_index != pmd->if_index &&
1584 info->ifi_index != pmd->remote_if_index))
1586 return tap_link_update(dev, 0);
1590 tap_dev_intr_handler(void *cb_arg)
1592 struct rte_eth_dev *dev = cb_arg;
1593 struct pmd_internals *pmd = dev->data->dev_private;
1595 tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1599 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1601 struct pmd_internals *pmd = dev->data->dev_private;
1603 /* In any case, disable interrupt if the conf is no longer there. */
1604 if (!dev->data->dev_conf.intr_conf.lsc) {
1605 if (pmd->intr_handle.fd != -1) {
1606 tap_nl_final(pmd->intr_handle.fd);
1607 rte_intr_callback_unregister(&pmd->intr_handle,
1608 tap_dev_intr_handler, dev);
1613 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1614 if (unlikely(pmd->intr_handle.fd == -1))
1616 return rte_intr_callback_register(
1617 &pmd->intr_handle, tap_dev_intr_handler, dev);
1619 tap_nl_final(pmd->intr_handle.fd);
1620 return rte_intr_callback_unregister(&pmd->intr_handle,
1621 tap_dev_intr_handler, dev);
1625 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1629 err = tap_lsc_intr_handle_set(dev, set);
1632 tap_rx_intr_vec_set(dev, 0);
1635 err = tap_rx_intr_vec_set(dev, set);
1637 tap_lsc_intr_handle_set(dev, 0);
1641 static const uint32_t*
1642 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1644 static const uint32_t ptypes[] = {
1645 RTE_PTYPE_INNER_L2_ETHER,
1646 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1647 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1648 RTE_PTYPE_INNER_L3_IPV4,
1649 RTE_PTYPE_INNER_L3_IPV4_EXT,
1650 RTE_PTYPE_INNER_L3_IPV6,
1651 RTE_PTYPE_INNER_L3_IPV6_EXT,
1652 RTE_PTYPE_INNER_L4_FRAG,
1653 RTE_PTYPE_INNER_L4_UDP,
1654 RTE_PTYPE_INNER_L4_TCP,
1655 RTE_PTYPE_INNER_L4_SCTP,
1657 RTE_PTYPE_L2_ETHER_VLAN,
1658 RTE_PTYPE_L2_ETHER_QINQ,
1660 RTE_PTYPE_L3_IPV4_EXT,
1661 RTE_PTYPE_L3_IPV6_EXT,
1673 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1674 struct rte_eth_fc_conf *fc_conf)
1676 fc_conf->mode = RTE_FC_NONE;
1681 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1682 struct rte_eth_fc_conf *fc_conf)
1684 if (fc_conf->mode != RTE_FC_NONE)
1690 * DPDK callback to update the RSS hash configuration.
1693 * Pointer to Ethernet device structure.
1694 * @param[in] rss_conf
1695 * RSS configuration data.
1698 * 0 on success, a negative errno value otherwise and rte_errno is set.
1701 tap_rss_hash_update(struct rte_eth_dev *dev,
1702 struct rte_eth_rss_conf *rss_conf)
1704 if (rss_conf->rss_hf & TAP_RSS_HF_MASK) {
1708 if (rss_conf->rss_key && rss_conf->rss_key_len) {
1710 * Currently TAP RSS key is hard coded
1711 * and cannot be updated
1714 "port %u RSS key cannot be updated",
1715 dev->data->port_id);
1723 tap_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1725 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1731 tap_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1733 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1739 tap_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1741 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1747 tap_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1749 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1753 static const struct eth_dev_ops ops = {
1754 .dev_start = tap_dev_start,
1755 .dev_stop = tap_dev_stop,
1756 .dev_close = tap_dev_close,
1757 .dev_configure = tap_dev_configure,
1758 .dev_infos_get = tap_dev_info,
1759 .rx_queue_setup = tap_rx_queue_setup,
1760 .tx_queue_setup = tap_tx_queue_setup,
1761 .rx_queue_start = tap_rx_queue_start,
1762 .tx_queue_start = tap_tx_queue_start,
1763 .rx_queue_stop = tap_rx_queue_stop,
1764 .tx_queue_stop = tap_tx_queue_stop,
1765 .rx_queue_release = tap_rx_queue_release,
1766 .tx_queue_release = tap_tx_queue_release,
1767 .flow_ctrl_get = tap_flow_ctrl_get,
1768 .flow_ctrl_set = tap_flow_ctrl_set,
1769 .link_update = tap_link_update,
1770 .dev_set_link_up = tap_link_set_up,
1771 .dev_set_link_down = tap_link_set_down,
1772 .promiscuous_enable = tap_promisc_enable,
1773 .promiscuous_disable = tap_promisc_disable,
1774 .allmulticast_enable = tap_allmulti_enable,
1775 .allmulticast_disable = tap_allmulti_disable,
1776 .mac_addr_set = tap_mac_set,
1777 .mtu_set = tap_mtu_set,
1778 .set_mc_addr_list = tap_set_mc_addr_list,
1779 .stats_get = tap_stats_get,
1780 .stats_reset = tap_stats_reset,
1781 .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1782 .rss_hash_update = tap_rss_hash_update,
1783 .filter_ctrl = tap_dev_filter_ctrl,
1786 static const char *tuntap_types[ETH_TUNTAP_TYPE_MAX] = {
1787 "UNKNOWN", "TUN", "TAP"
1791 eth_dev_tap_create(struct rte_vdev_device *vdev, const char *tap_name,
1792 char *remote_iface, struct rte_ether_addr *mac_addr,
1793 enum rte_tuntap_type type)
1795 int numa_node = rte_socket_id();
1796 struct rte_eth_dev *dev;
1797 struct pmd_internals *pmd;
1798 struct pmd_process_private *process_private;
1799 const char *tuntap_name = tuntap_types[type];
1800 struct rte_eth_dev_data *data;
1804 TAP_LOG(DEBUG, "%s device on numa %u", tuntap_name, rte_socket_id());
1806 dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1808 TAP_LOG(ERR, "%s Unable to allocate device struct",
1810 goto error_exit_nodev;
1813 process_private = (struct pmd_process_private *)
1814 rte_zmalloc_socket(tap_name, sizeof(struct pmd_process_private),
1815 RTE_CACHE_LINE_SIZE, dev->device->numa_node);
1817 if (process_private == NULL) {
1818 TAP_LOG(ERR, "Failed to alloc memory for process private");
1821 pmd = dev->data->dev_private;
1822 dev->process_private = process_private;
1824 strlcpy(pmd->name, tap_name, sizeof(pmd->name));
1829 pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1830 if (pmd->ioctl_sock == -1) {
1832 "%s Unable to get a socket for management: %s",
1833 tuntap_name, strerror(errno));
1837 /* Setup some default values */
1839 data->dev_private = pmd;
1840 data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1841 data->numa_node = numa_node;
1843 data->dev_link = pmd_link;
1844 data->mac_addrs = &pmd->eth_addr;
1845 /* Set the number of RX and TX queues */
1846 data->nb_rx_queues = 0;
1847 data->nb_tx_queues = 0;
1849 dev->dev_ops = &ops;
1850 dev->rx_pkt_burst = pmd_rx_burst;
1851 dev->tx_pkt_burst = pmd_tx_burst;
1853 pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1854 pmd->intr_handle.fd = -1;
1855 dev->intr_handle = &pmd->intr_handle;
1857 /* Presetup the fds to -1 as being not valid */
1858 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1859 process_private->rxq_fds[i] = -1;
1860 process_private->txq_fds[i] = -1;
1863 if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1864 if (rte_is_zero_ether_addr(mac_addr))
1865 rte_eth_random_addr((uint8_t *)&pmd->eth_addr);
1867 rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1871 * Allocate a TUN device keep-alive file descriptor that will only be
1872 * closed when the TUN device itself is closed or removed.
1873 * This keep-alive file descriptor will guarantee that the TUN device
1874 * exists even when all of its queues are closed
1876 pmd->ka_fd = tun_alloc(pmd, 1);
1877 if (pmd->ka_fd == -1) {
1878 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
1881 TAP_LOG(DEBUG, "allocated %s", pmd->name);
1883 ifr.ifr_mtu = dev->data->mtu;
1884 if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1887 if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1888 memset(&ifr, 0, sizeof(struct ifreq));
1889 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1890 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
1891 RTE_ETHER_ADDR_LEN);
1892 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1897 * Set up everything related to rte_flow:
1899 * - tap / remote if_index
1900 * - mandatory QDISCs
1901 * - rte_flow actual/implicit lists
1904 pmd->nlsk_fd = tap_nl_init(0);
1905 if (pmd->nlsk_fd == -1) {
1906 TAP_LOG(WARNING, "%s: failed to create netlink socket.",
1908 goto disable_rte_flow;
1910 pmd->if_index = if_nametoindex(pmd->name);
1911 if (!pmd->if_index) {
1912 TAP_LOG(ERR, "%s: failed to get if_index.", pmd->name);
1913 goto disable_rte_flow;
1915 if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1916 TAP_LOG(ERR, "%s: failed to create multiq qdisc.",
1918 goto disable_rte_flow;
1920 if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1921 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1923 goto disable_rte_flow;
1925 LIST_INIT(&pmd->flows);
1927 if (strlen(remote_iface)) {
1928 pmd->remote_if_index = if_nametoindex(remote_iface);
1929 if (!pmd->remote_if_index) {
1930 TAP_LOG(ERR, "%s: failed to get %s if_index.",
1931 pmd->name, remote_iface);
1934 strlcpy(pmd->remote_iface, remote_iface, RTE_ETH_NAME_MAX_LEN);
1936 /* Save state of remote device */
1937 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1939 /* Replicate remote MAC address */
1940 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1941 TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1942 pmd->name, pmd->remote_iface);
1945 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1946 RTE_ETHER_ADDR_LEN);
1947 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1948 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1949 TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1950 pmd->name, remote_iface);
1955 * Flush usually returns negative value because it tries to
1956 * delete every QDISC (and on a running device, one QDISC at
1957 * least is needed). Ignore negative return value.
1959 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1960 if (qdisc_create_ingress(pmd->nlsk_fd,
1961 pmd->remote_if_index) < 0) {
1962 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1966 LIST_INIT(&pmd->implicit_flows);
1967 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1968 tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1969 tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1970 tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1972 "%s: failed to create implicit rules.",
1978 rte_eth_dev_probing_finish(dev);
1982 TAP_LOG(ERR, " Disabling rte flow support: %s(%d)",
1983 strerror(errno), errno);
1984 if (strlen(remote_iface)) {
1985 TAP_LOG(ERR, "Remote feature requires flow support.");
1988 rte_eth_dev_probing_finish(dev);
1992 TAP_LOG(ERR, " Can't set up remote feature: %s(%d)",
1993 strerror(errno), errno);
1994 tap_flow_implicit_flush(pmd, NULL);
1997 if (pmd->nlsk_fd != -1)
1998 close(pmd->nlsk_fd);
1999 if (pmd->ka_fd != -1)
2001 if (pmd->ioctl_sock != -1)
2002 close(pmd->ioctl_sock);
2003 /* mac_addrs must not be freed alone because part of dev_private */
2004 dev->data->mac_addrs = NULL;
2005 rte_eth_dev_release_port(dev);
2008 TAP_LOG(ERR, "%s Unable to initialize %s",
2009 tuntap_name, rte_vdev_device_name(vdev));
2014 /* make sure name is a possible Linux network device name */
2016 is_valid_iface(const char *name)
2021 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
2025 if (*name == '/' || *name == ':' || isspace(*name))
2033 set_interface_name(const char *key __rte_unused,
2037 char *name = (char *)extra_args;
2040 if (!is_valid_iface(value)) {
2041 TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
2045 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
2047 /* use tap%d which causes kernel to choose next available */
2048 strlcpy(name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2054 set_remote_iface(const char *key __rte_unused,
2058 char *name = (char *)extra_args;
2061 if (!is_valid_iface(value)) {
2062 TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
2066 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
2072 static int parse_user_mac(struct rte_ether_addr *user_mac,
2075 unsigned int index = 0;
2076 char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
2078 if (user_mac == NULL || value == NULL)
2081 strlcpy(mac_temp, value, sizeof(mac_temp));
2082 mac_byte = strtok(mac_temp, ":");
2084 while ((mac_byte != NULL) &&
2085 (strlen(mac_byte) <= 2) &&
2086 (strlen(mac_byte) == strspn(mac_byte,
2087 ETH_TAP_CMP_MAC_FMT))) {
2088 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
2089 mac_byte = strtok(NULL, ":");
2096 set_mac_type(const char *key __rte_unused,
2100 struct rte_ether_addr *user_mac = extra_args;
2105 if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
2106 static int iface_idx;
2108 /* fixed mac = 00:64:74:61:70:<iface_idx> */
2109 memcpy((char *)user_mac->addr_bytes, "\0dtap",
2110 RTE_ETHER_ADDR_LEN);
2111 user_mac->addr_bytes[RTE_ETHER_ADDR_LEN - 1] =
2116 if (parse_user_mac(user_mac, value) != 6)
2119 TAP_LOG(DEBUG, "TAP user MAC param (%s)", value);
2123 TAP_LOG(ERR, "TAP user MAC (%s) is not in format (%s|%s)",
2124 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
2129 * Open a TUN interface device. TUN PMD
2130 * 1) sets tap_type as false
2131 * 2) intakes iface as argument.
2132 * 3) as interface is virtual set speed to 10G
2135 rte_pmd_tun_probe(struct rte_vdev_device *dev)
2137 const char *name, *params;
2139 struct rte_kvargs *kvlist = NULL;
2140 char tun_name[RTE_ETH_NAME_MAX_LEN];
2141 char remote_iface[RTE_ETH_NAME_MAX_LEN];
2142 struct rte_eth_dev *eth_dev;
2144 name = rte_vdev_device_name(dev);
2145 params = rte_vdev_device_args(dev);
2146 memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2148 if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
2149 strlen(params) == 0) {
2150 eth_dev = rte_eth_dev_attach_secondary(name);
2152 TAP_LOG(ERR, "Failed to probe %s", name);
2155 eth_dev->dev_ops = &ops;
2156 eth_dev->device = &dev->device;
2157 rte_eth_dev_probing_finish(eth_dev);
2161 /* use tun%d which causes kernel to choose next available */
2162 strlcpy(tun_name, DEFAULT_TUN_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2164 if (params && (params[0] != '\0')) {
2165 TAP_LOG(DEBUG, "parameters (%s)", params);
2167 kvlist = rte_kvargs_parse(params, valid_arguments);
2169 if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2170 ret = rte_kvargs_process(kvlist,
2172 &set_interface_name,
2180 pmd_link.link_speed = ETH_SPEED_NUM_10G;
2182 TAP_LOG(DEBUG, "Initializing pmd_tun for %s", name);
2184 ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0,
2185 ETH_TUNTAP_TYPE_TUN);
2189 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2192 rte_kvargs_free(kvlist);
2197 /* Request queue file descriptors from secondary to primary. */
2199 tap_mp_attach_queues(const char *port_name, struct rte_eth_dev *dev)
2202 struct timespec timeout = {.tv_sec = 1, .tv_nsec = 0};
2203 struct rte_mp_msg request, *reply;
2204 struct rte_mp_reply replies;
2205 struct ipc_queues *request_param = (struct ipc_queues *)request.param;
2206 struct ipc_queues *reply_param;
2207 struct pmd_process_private *process_private = dev->process_private;
2208 int queue, fd_iterator;
2210 /* Prepare the request */
2211 memset(&request, 0, sizeof(request));
2212 strlcpy(request.name, TAP_MP_KEY, sizeof(request.name));
2213 strlcpy(request_param->port_name, port_name,
2214 sizeof(request_param->port_name));
2215 request.len_param = sizeof(*request_param);
2216 /* Send request and receive reply */
2217 ret = rte_mp_request_sync(&request, &replies, &timeout);
2218 if (ret < 0 || replies.nb_received != 1) {
2219 TAP_LOG(ERR, "Failed to request queues from primary: %d",
2223 reply = &replies.msgs[0];
2224 reply_param = (struct ipc_queues *)reply->param;
2225 TAP_LOG(DEBUG, "Received IPC reply for %s", reply_param->port_name);
2227 /* Attach the queues from received file descriptors */
2228 if (reply_param->rxq_count + reply_param->txq_count != reply->num_fds) {
2229 TAP_LOG(ERR, "Unexpected number of fds received");
2233 dev->data->nb_rx_queues = reply_param->rxq_count;
2234 dev->data->nb_tx_queues = reply_param->txq_count;
2236 for (queue = 0; queue < reply_param->rxq_count; queue++)
2237 process_private->rxq_fds[queue] = reply->fds[fd_iterator++];
2238 for (queue = 0; queue < reply_param->txq_count; queue++)
2239 process_private->txq_fds[queue] = reply->fds[fd_iterator++];
2244 /* Send the queue file descriptors from the primary process to secondary. */
2246 tap_mp_sync_queues(const struct rte_mp_msg *request, const void *peer)
2248 struct rte_eth_dev *dev;
2249 struct pmd_process_private *process_private;
2250 struct rte_mp_msg reply;
2251 const struct ipc_queues *request_param =
2252 (const struct ipc_queues *)request->param;
2253 struct ipc_queues *reply_param =
2254 (struct ipc_queues *)reply.param;
2259 /* Get requested port */
2260 TAP_LOG(DEBUG, "Received IPC request for %s", request_param->port_name);
2261 ret = rte_eth_dev_get_port_by_name(request_param->port_name, &port_id);
2263 TAP_LOG(ERR, "Failed to get port id for %s",
2264 request_param->port_name);
2267 dev = &rte_eth_devices[port_id];
2268 process_private = dev->process_private;
2270 /* Fill file descriptors for all queues */
2272 reply_param->rxq_count = 0;
2273 if (dev->data->nb_rx_queues + dev->data->nb_tx_queues >
2275 TAP_LOG(ERR, "Number of rx/tx queues exceeds max number of fds");
2279 for (queue = 0; queue < dev->data->nb_rx_queues; queue++) {
2280 reply.fds[reply.num_fds++] = process_private->rxq_fds[queue];
2281 reply_param->rxq_count++;
2283 RTE_ASSERT(reply_param->rxq_count == dev->data->nb_rx_queues);
2285 reply_param->txq_count = 0;
2286 for (queue = 0; queue < dev->data->nb_tx_queues; queue++) {
2287 reply.fds[reply.num_fds++] = process_private->txq_fds[queue];
2288 reply_param->txq_count++;
2290 RTE_ASSERT(reply_param->txq_count == dev->data->nb_tx_queues);
2293 strlcpy(reply.name, request->name, sizeof(reply.name));
2294 strlcpy(reply_param->port_name, request_param->port_name,
2295 sizeof(reply_param->port_name));
2296 reply.len_param = sizeof(*reply_param);
2297 if (rte_mp_reply(&reply, peer) < 0) {
2298 TAP_LOG(ERR, "Failed to reply an IPC request to sync queues");
2304 /* Open a TAP interface device.
2307 rte_pmd_tap_probe(struct rte_vdev_device *dev)
2309 const char *name, *params;
2311 struct rte_kvargs *kvlist = NULL;
2313 char tap_name[RTE_ETH_NAME_MAX_LEN];
2314 char remote_iface[RTE_ETH_NAME_MAX_LEN];
2315 struct rte_ether_addr user_mac = { .addr_bytes = {0} };
2316 struct rte_eth_dev *eth_dev;
2317 int tap_devices_count_increased = 0;
2319 name = rte_vdev_device_name(dev);
2320 params = rte_vdev_device_args(dev);
2322 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
2323 eth_dev = rte_eth_dev_attach_secondary(name);
2325 TAP_LOG(ERR, "Failed to probe %s", name);
2328 eth_dev->dev_ops = &ops;
2329 eth_dev->device = &dev->device;
2330 eth_dev->rx_pkt_burst = pmd_rx_burst;
2331 eth_dev->tx_pkt_burst = pmd_tx_burst;
2332 if (!rte_eal_primary_proc_alive(NULL)) {
2333 TAP_LOG(ERR, "Primary process is missing");
2336 eth_dev->process_private = (struct pmd_process_private *)
2337 rte_zmalloc_socket(name,
2338 sizeof(struct pmd_process_private),
2339 RTE_CACHE_LINE_SIZE,
2340 eth_dev->device->numa_node);
2341 if (eth_dev->process_private == NULL) {
2343 "Failed to alloc memory for process private");
2347 ret = tap_mp_attach_queues(name, eth_dev);
2350 rte_eth_dev_probing_finish(eth_dev);
2354 speed = ETH_SPEED_NUM_10G;
2356 /* use tap%d which causes kernel to choose next available */
2357 strlcpy(tap_name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2358 memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2360 if (params && (params[0] != '\0')) {
2361 TAP_LOG(DEBUG, "parameters (%s)", params);
2363 kvlist = rte_kvargs_parse(params, valid_arguments);
2365 if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2366 ret = rte_kvargs_process(kvlist,
2368 &set_interface_name,
2374 if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
2375 ret = rte_kvargs_process(kvlist,
2383 if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
2384 ret = rte_kvargs_process(kvlist,
2393 pmd_link.link_speed = speed;
2395 TAP_LOG(DEBUG, "Initializing pmd_tap for %s", name);
2397 /* Register IPC feed callback */
2398 if (!tap_devices_count) {
2399 ret = rte_mp_action_register(TAP_MP_KEY, tap_mp_sync_queues);
2400 if (ret < 0 && rte_errno != ENOTSUP) {
2401 TAP_LOG(ERR, "tap: Failed to register IPC callback: %s",
2402 strerror(rte_errno));
2406 tap_devices_count++;
2407 tap_devices_count_increased = 1;
2408 ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac,
2409 ETH_TUNTAP_TYPE_TAP);
2413 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2415 if (tap_devices_count_increased == 1) {
2416 if (tap_devices_count == 1)
2417 rte_mp_action_unregister(TAP_MP_KEY);
2418 tap_devices_count--;
2421 rte_kvargs_free(kvlist);
2426 /* detach a TUNTAP device.
2429 rte_pmd_tap_remove(struct rte_vdev_device *dev)
2431 struct rte_eth_dev *eth_dev = NULL;
2432 struct pmd_internals *internals;
2434 /* find the ethdev entry */
2435 eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
2439 /* mac_addrs must not be freed alone because part of dev_private */
2440 eth_dev->data->mac_addrs = NULL;
2442 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2443 return rte_eth_dev_release_port(eth_dev);
2445 tap_dev_close(eth_dev);
2447 internals = eth_dev->data->dev_private;
2448 TAP_LOG(DEBUG, "Closing %s Ethernet device on numa %u",
2449 tuntap_types[internals->type], rte_socket_id());
2451 close(internals->ioctl_sock);
2452 rte_free(eth_dev->process_private);
2453 if (tap_devices_count == 1)
2454 rte_mp_action_unregister(TAP_MP_KEY);
2455 tap_devices_count--;
2456 rte_eth_dev_release_port(eth_dev);
2461 static struct rte_vdev_driver pmd_tun_drv = {
2462 .probe = rte_pmd_tun_probe,
2463 .remove = rte_pmd_tap_remove,
2466 static struct rte_vdev_driver pmd_tap_drv = {
2467 .probe = rte_pmd_tap_probe,
2468 .remove = rte_pmd_tap_remove,
2471 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
2472 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
2473 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
2474 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
2475 ETH_TAP_IFACE_ARG "=<string> ");
2476 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
2477 ETH_TAP_IFACE_ARG "=<string> "
2478 ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
2479 ETH_TAP_REMOTE_ARG "=<string>");
2482 RTE_INIT(tap_init_log)
2484 tap_logtype = rte_log_register("pmd.net.tap");
2485 if (tap_logtype >= 0)
2486 rte_log_set_level(tap_logtype, RTE_LOG_NOTICE);