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 <ethdev_driver.h>
11 #include <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);
306 struct rte_udp_hdr *udp_hdr;
308 if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
310 else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
312 /* Don't verify checksum for packets with discontinuous L2 header */
313 if (unlikely(l2_len + sizeof(struct rte_ipv4_hdr) >
314 rte_pktmbuf_data_len(mbuf)))
316 l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
317 if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
318 struct rte_ipv4_hdr *iph = l3_hdr;
320 l3_len = rte_ipv4_hdr_len(iph);
321 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
323 /* check that the total length reported by header is not
324 * greater than the total received size
326 if (l2_len + rte_be_to_cpu_16(iph->total_length) >
327 rte_pktmbuf_data_len(mbuf))
330 cksum = ~rte_raw_cksum(iph, l3_len);
331 mbuf->ol_flags |= cksum ?
332 PKT_RX_IP_CKSUM_BAD :
333 PKT_RX_IP_CKSUM_GOOD;
334 } else if (l3 == RTE_PTYPE_L3_IPV6) {
335 struct rte_ipv6_hdr *iph = l3_hdr;
337 l3_len = sizeof(struct rte_ipv6_hdr);
338 /* check that the total length reported by header is not
339 * greater than the total received size
341 if (l2_len + l3_len + rte_be_to_cpu_16(iph->payload_len) >
342 rte_pktmbuf_data_len(mbuf))
345 /* IPv6 extensions are not supported */
348 if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
351 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
352 /* Don't verify checksum for multi-segment packets. */
353 if (mbuf->nb_segs > 1)
355 if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
356 if (l4 == RTE_PTYPE_L4_UDP) {
357 udp_hdr = (struct rte_udp_hdr *)l4_hdr;
358 if (udp_hdr->dgram_cksum == 0) {
360 * For IPv4, a zero UDP checksum
361 * indicates that the sender did not
362 * generate one [RFC 768].
364 mbuf->ol_flags |= PKT_RX_L4_CKSUM_NONE;
368 cksum = rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
369 } else { /* l3 == RTE_PTYPE_L3_IPV6, checked above */
370 cksum = rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
372 cksum_ok = (cksum == 0) || (cksum == 0xffff);
373 mbuf->ol_flags |= cksum_ok ?
374 PKT_RX_L4_CKSUM_GOOD : PKT_RX_L4_CKSUM_BAD;
379 tap_rx_offload_get_port_capa(void)
382 * No specific port Rx offload capabilities.
388 tap_rx_offload_get_queue_capa(void)
390 return DEV_RX_OFFLOAD_SCATTER |
391 DEV_RX_OFFLOAD_IPV4_CKSUM |
392 DEV_RX_OFFLOAD_UDP_CKSUM |
393 DEV_RX_OFFLOAD_TCP_CKSUM;
397 tap_rxq_pool_free(struct rte_mbuf *pool)
399 struct rte_mbuf *mbuf = pool;
400 uint16_t nb_segs = 1;
409 pool->nb_segs = nb_segs;
410 rte_pktmbuf_free(pool);
413 /* Callback to handle the rx burst of packets to the correct interface and
414 * file descriptor(s) in a multi-queue setup.
417 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
419 struct rx_queue *rxq = queue;
420 struct pmd_process_private *process_private;
422 unsigned long num_rx_bytes = 0;
423 uint32_t trigger = tap_trigger;
425 if (trigger == rxq->trigger_seen)
428 process_private = rte_eth_devices[rxq->in_port].process_private;
429 for (num_rx = 0; num_rx < nb_pkts; ) {
430 struct rte_mbuf *mbuf = rxq->pool;
431 struct rte_mbuf *seg = NULL;
432 struct rte_mbuf *new_tail = NULL;
433 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
436 len = readv(process_private->rxq_fds[rxq->queue_id],
438 1 + (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
439 rxq->nb_rx_desc : 1));
440 if (len < (int)sizeof(struct tun_pi))
443 /* Packet couldn't fit in the provided mbuf */
444 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
445 rxq->stats.ierrors++;
449 len -= sizeof(struct tun_pi);
452 mbuf->port = rxq->in_port;
454 struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
456 if (unlikely(!buf)) {
457 rxq->stats.rx_nombuf++;
458 /* No new buf has been allocated: do nothing */
459 if (!new_tail || !seg)
463 tap_rxq_pool_free(mbuf);
467 seg = seg ? seg->next : mbuf;
468 if (rxq->pool == mbuf)
471 new_tail->next = buf;
473 new_tail->next = seg->next;
475 /* iovecs[0] is reserved for packet info (pi) */
476 (*rxq->iovecs)[mbuf->nb_segs].iov_len =
477 buf->buf_len - data_off;
478 (*rxq->iovecs)[mbuf->nb_segs].iov_base =
479 (char *)buf->buf_addr + data_off;
481 seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
482 seg->data_off = data_off;
484 len -= seg->data_len;
488 /* First segment has headroom, not the others */
492 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
494 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
495 tap_verify_csum(mbuf);
497 /* account for the receive frame */
498 bufs[num_rx++] = mbuf;
499 num_rx_bytes += mbuf->pkt_len;
502 rxq->stats.ipackets += num_rx;
503 rxq->stats.ibytes += num_rx_bytes;
505 if (trigger && num_rx < nb_pkts)
506 rxq->trigger_seen = trigger;
512 tap_tx_offload_get_port_capa(void)
515 * No specific port Tx offload capabilities.
521 tap_tx_offload_get_queue_capa(void)
523 return DEV_TX_OFFLOAD_MULTI_SEGS |
524 DEV_TX_OFFLOAD_IPV4_CKSUM |
525 DEV_TX_OFFLOAD_UDP_CKSUM |
526 DEV_TX_OFFLOAD_TCP_CKSUM |
527 DEV_TX_OFFLOAD_TCP_TSO;
530 /* Finalize l4 checksum calculation */
532 tap_tx_l4_cksum(uint16_t *l4_cksum, uint16_t l4_phdr_cksum,
533 uint32_t l4_raw_cksum)
538 cksum = __rte_raw_cksum_reduce(l4_raw_cksum);
539 cksum += l4_phdr_cksum;
541 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
542 cksum = (~cksum) & 0xffff;
549 /* Accumaulate L4 raw checksums */
551 tap_tx_l4_add_rcksum(char *l4_data, unsigned int l4_len, uint16_t *l4_cksum,
552 uint32_t *l4_raw_cksum)
554 if (l4_cksum == NULL)
557 *l4_raw_cksum = __rte_raw_cksum(l4_data, l4_len, *l4_raw_cksum);
560 /* L3 and L4 pseudo headers checksum offloads */
562 tap_tx_l3_cksum(char *packet, uint64_t ol_flags, unsigned int l2_len,
563 unsigned int l3_len, unsigned int l4_len, uint16_t **l4_cksum,
564 uint16_t *l4_phdr_cksum, uint32_t *l4_raw_cksum)
566 void *l3_hdr = packet + l2_len;
568 if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
569 struct rte_ipv4_hdr *iph = l3_hdr;
572 iph->hdr_checksum = 0;
573 cksum = rte_raw_cksum(iph, l3_len);
574 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
576 if (ol_flags & PKT_TX_L4_MASK) {
579 l4_hdr = packet + l2_len + l3_len;
580 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
581 *l4_cksum = &((struct rte_udp_hdr *)l4_hdr)->dgram_cksum;
582 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
583 *l4_cksum = &((struct rte_tcp_hdr *)l4_hdr)->cksum;
587 if (ol_flags & PKT_TX_IPV4)
588 *l4_phdr_cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
590 *l4_phdr_cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
591 *l4_raw_cksum = __rte_raw_cksum(l4_hdr, l4_len, 0);
596 tap_write_mbufs(struct tx_queue *txq, uint16_t num_mbufs,
597 struct rte_mbuf **pmbufs,
598 uint16_t *num_packets, unsigned long *num_tx_bytes)
602 struct pmd_process_private *process_private;
604 process_private = rte_eth_devices[txq->out_port].process_private;
606 for (i = 0; i < num_mbufs; i++) {
607 struct rte_mbuf *mbuf = pmbufs[i];
608 struct iovec iovecs[mbuf->nb_segs + 2];
609 struct tun_pi pi = { .flags = 0, .proto = 0x00 };
610 struct rte_mbuf *seg = mbuf;
611 char m_copy[mbuf->data_len];
615 int k; /* current index in iovecs for copying segments */
616 uint16_t seg_len; /* length of first segment */
618 uint16_t *l4_cksum; /* l4 checksum (pseudo header + payload) */
619 uint32_t l4_raw_cksum = 0; /* TCP/UDP payload raw checksum */
620 uint16_t l4_phdr_cksum = 0; /* TCP/UDP pseudo header checksum */
621 uint16_t is_cksum = 0; /* in case cksum should be offloaded */
624 if (txq->type == ETH_TUNTAP_TYPE_TUN) {
626 * TUN and TAP are created with IFF_NO_PI disabled.
627 * For TUN PMD this mandatory as fields are used by
628 * Kernel tun.c to determine whether its IP or non IP
631 * The logic fetches the first byte of data from mbuf
632 * then compares whether its v4 or v6. If first byte
633 * is 4 or 6, then protocol field is updated.
635 char *buff_data = rte_pktmbuf_mtod(seg, void *);
636 proto = (*buff_data & 0xf0);
637 pi.proto = (proto == 0x40) ?
638 rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4) :
640 rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6) :
645 iovecs[k].iov_base = π
646 iovecs[k].iov_len = sizeof(pi);
649 nb_segs = mbuf->nb_segs;
651 ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
652 (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
653 (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
656 /* Support only packets with at least layer 4
657 * header included in the first segment
659 seg_len = rte_pktmbuf_data_len(mbuf);
660 l234_hlen = mbuf->l2_len + mbuf->l3_len + mbuf->l4_len;
661 if (seg_len < l234_hlen)
664 /* To change checksums, work on a * copy of l2, l3
665 * headers + l4 pseudo header
667 rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
669 tap_tx_l3_cksum(m_copy, mbuf->ol_flags,
670 mbuf->l2_len, mbuf->l3_len, mbuf->l4_len,
671 &l4_cksum, &l4_phdr_cksum,
673 iovecs[k].iov_base = m_copy;
674 iovecs[k].iov_len = l234_hlen;
677 /* Update next iovecs[] beyond l2, l3, l4 headers */
678 if (seg_len > l234_hlen) {
679 iovecs[k].iov_len = seg_len - l234_hlen;
681 rte_pktmbuf_mtod(seg, char *) +
683 tap_tx_l4_add_rcksum(iovecs[k].iov_base,
684 iovecs[k].iov_len, l4_cksum,
692 for (j = k; j <= nb_segs; j++) {
693 iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
694 iovecs[j].iov_base = rte_pktmbuf_mtod(seg, void *);
696 tap_tx_l4_add_rcksum(iovecs[j].iov_base,
697 iovecs[j].iov_len, l4_cksum,
703 tap_tx_l4_cksum(l4_cksum, l4_phdr_cksum, l4_raw_cksum);
705 /* copy the tx frame data */
706 n = writev(process_private->txq_fds[txq->queue_id], iovecs, j);
711 (*num_tx_bytes) += rte_pktmbuf_pkt_len(mbuf);
716 /* Callback to handle sending packets from the tap interface
719 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
721 struct tx_queue *txq = queue;
723 uint16_t num_packets = 0;
724 unsigned long num_tx_bytes = 0;
728 if (unlikely(nb_pkts == 0))
731 struct rte_mbuf *gso_mbufs[MAX_GSO_MBUFS];
732 max_size = *txq->mtu + (RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN + 4);
733 for (i = 0; i < nb_pkts; i++) {
734 struct rte_mbuf *mbuf_in = bufs[num_tx];
735 struct rte_mbuf **mbuf;
736 uint16_t num_mbufs = 0;
737 uint16_t tso_segsz = 0;
743 tso = mbuf_in->ol_flags & PKT_TX_TCP_SEG;
745 struct rte_gso_ctx *gso_ctx = &txq->gso_ctx;
747 /* TCP segmentation implies TCP checksum offload */
748 mbuf_in->ol_flags |= PKT_TX_TCP_CKSUM;
750 /* gso size is calculated without RTE_ETHER_CRC_LEN */
751 hdrs_len = mbuf_in->l2_len + mbuf_in->l3_len +
753 tso_segsz = mbuf_in->tso_segsz + hdrs_len;
754 if (unlikely(tso_segsz == hdrs_len) ||
755 tso_segsz > *txq->mtu) {
759 gso_ctx->gso_size = tso_segsz;
760 /* 'mbuf_in' packet to segment */
761 num_tso_mbufs = rte_gso_segment(mbuf_in,
762 gso_ctx, /* gso control block */
763 (struct rte_mbuf **)&gso_mbufs, /* out mbufs */
764 RTE_DIM(gso_mbufs)); /* max tso mbufs */
766 /* ret contains the number of new created mbufs */
767 if (num_tso_mbufs < 0)
770 if (num_tso_mbufs >= 1) {
772 num_mbufs = num_tso_mbufs;
774 /* 0 means it can be transmitted directly
781 /* stats.errs will be incremented */
782 if (rte_pktmbuf_pkt_len(mbuf_in) > max_size)
785 /* ret 0 indicates no new mbufs were created */
791 ret = tap_write_mbufs(txq, num_mbufs, mbuf,
792 &num_packets, &num_tx_bytes);
796 if (num_tso_mbufs > 0)
797 rte_pktmbuf_free_bulk(mbuf, num_tso_mbufs);
801 /* free original mbuf */
802 rte_pktmbuf_free(mbuf_in);
804 if (num_tso_mbufs > 0)
805 rte_pktmbuf_free_bulk(mbuf, num_tso_mbufs);
808 txq->stats.opackets += num_packets;
809 txq->stats.errs += nb_pkts - num_tx;
810 txq->stats.obytes += num_tx_bytes;
816 tap_ioctl_req2str(unsigned long request)
820 return "SIOCSIFFLAGS";
822 return "SIOCGIFFLAGS";
824 return "SIOCGIFHWADDR";
826 return "SIOCSIFHWADDR";
834 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
835 struct ifreq *ifr, int set, enum ioctl_mode mode)
837 short req_flags = ifr->ifr_flags;
838 int remote = pmd->remote_if_index &&
839 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
841 if (!pmd->remote_if_index && mode == REMOTE_ONLY)
844 * If there is a remote netdevice, apply ioctl on it, then apply it on
849 strlcpy(ifr->ifr_name, pmd->remote_iface, IFNAMSIZ);
850 else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
851 strlcpy(ifr->ifr_name, pmd->name, IFNAMSIZ);
854 /* fetch current flags to leave other flags untouched */
855 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
858 ifr->ifr_flags |= req_flags;
860 ifr->ifr_flags &= ~req_flags;
868 TAP_LOG(WARNING, "%s: ioctl() called with wrong arg",
872 if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
874 if (remote-- && mode == LOCAL_AND_REMOTE)
879 TAP_LOG(DEBUG, "%s(%s) failed: %s(%d)", ifr->ifr_name,
880 tap_ioctl_req2str(request), strerror(errno), errno);
885 tap_link_set_down(struct rte_eth_dev *dev)
887 struct pmd_internals *pmd = dev->data->dev_private;
888 struct ifreq ifr = { .ifr_flags = IFF_UP };
890 dev->data->dev_link.link_status = ETH_LINK_DOWN;
891 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
895 tap_link_set_up(struct rte_eth_dev *dev)
897 struct pmd_internals *pmd = dev->data->dev_private;
898 struct ifreq ifr = { .ifr_flags = IFF_UP };
900 dev->data->dev_link.link_status = ETH_LINK_UP;
901 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
905 tap_dev_start(struct rte_eth_dev *dev)
909 err = tap_intr_handle_set(dev, 1);
913 err = tap_link_set_up(dev);
917 for (i = 0; i < dev->data->nb_tx_queues; i++)
918 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
919 for (i = 0; i < dev->data->nb_rx_queues; i++)
920 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
925 /* This function gets called when the current port gets stopped.
928 tap_dev_stop(struct rte_eth_dev *dev)
932 for (i = 0; i < dev->data->nb_tx_queues; i++)
933 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
934 for (i = 0; i < dev->data->nb_rx_queues; i++)
935 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
937 tap_intr_handle_set(dev, 0);
938 tap_link_set_down(dev);
944 tap_dev_configure(struct rte_eth_dev *dev)
946 struct pmd_internals *pmd = dev->data->dev_private;
948 if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
950 "%s: number of rx queues %d exceeds max num of queues %d",
952 dev->data->nb_rx_queues,
953 RTE_PMD_TAP_MAX_QUEUES);
956 if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
958 "%s: number of tx queues %d exceeds max num of queues %d",
960 dev->data->nb_tx_queues,
961 RTE_PMD_TAP_MAX_QUEUES);
965 TAP_LOG(INFO, "%s: %s: TX configured queues number: %u",
966 dev->device->name, pmd->name, dev->data->nb_tx_queues);
968 TAP_LOG(INFO, "%s: %s: RX configured queues number: %u",
969 dev->device->name, pmd->name, dev->data->nb_rx_queues);
975 tap_dev_speed_capa(void)
977 uint32_t speed = pmd_link.link_speed;
980 if (speed >= ETH_SPEED_NUM_10M)
981 capa |= ETH_LINK_SPEED_10M;
982 if (speed >= ETH_SPEED_NUM_100M)
983 capa |= ETH_LINK_SPEED_100M;
984 if (speed >= ETH_SPEED_NUM_1G)
985 capa |= ETH_LINK_SPEED_1G;
986 if (speed >= ETH_SPEED_NUM_5G)
987 capa |= ETH_LINK_SPEED_2_5G;
988 if (speed >= ETH_SPEED_NUM_5G)
989 capa |= ETH_LINK_SPEED_5G;
990 if (speed >= ETH_SPEED_NUM_10G)
991 capa |= ETH_LINK_SPEED_10G;
992 if (speed >= ETH_SPEED_NUM_20G)
993 capa |= ETH_LINK_SPEED_20G;
994 if (speed >= ETH_SPEED_NUM_25G)
995 capa |= ETH_LINK_SPEED_25G;
996 if (speed >= ETH_SPEED_NUM_40G)
997 capa |= ETH_LINK_SPEED_40G;
998 if (speed >= ETH_SPEED_NUM_50G)
999 capa |= ETH_LINK_SPEED_50G;
1000 if (speed >= ETH_SPEED_NUM_56G)
1001 capa |= ETH_LINK_SPEED_56G;
1002 if (speed >= ETH_SPEED_NUM_100G)
1003 capa |= ETH_LINK_SPEED_100G;
1009 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1011 struct pmd_internals *internals = dev->data->dev_private;
1013 dev_info->if_index = internals->if_index;
1014 dev_info->max_mac_addrs = 1;
1015 dev_info->max_rx_pktlen = (uint32_t)RTE_ETHER_MAX_VLAN_FRAME_LEN;
1016 dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
1017 dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
1018 dev_info->min_rx_bufsize = 0;
1019 dev_info->speed_capa = tap_dev_speed_capa();
1020 dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
1021 dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
1022 dev_info->rx_queue_offload_capa;
1023 dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
1024 dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
1025 dev_info->tx_queue_offload_capa;
1026 dev_info->hash_key_size = TAP_RSS_HASH_KEY_SIZE;
1028 * limitation: TAP supports all of IP, UDP and TCP hash
1029 * functions together and not in partial combinations
1031 dev_info->flow_type_rss_offloads = ~TAP_RSS_HF_MASK;
1037 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
1039 unsigned int i, imax;
1040 unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
1041 unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
1042 unsigned long rx_nombuf = 0, ierrors = 0;
1043 const struct pmd_internals *pmd = dev->data->dev_private;
1045 /* rx queue statistics */
1046 imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
1047 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
1048 for (i = 0; i < imax; i++) {
1049 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
1050 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
1051 rx_total += tap_stats->q_ipackets[i];
1052 rx_bytes_total += tap_stats->q_ibytes[i];
1053 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
1054 ierrors += pmd->rxq[i].stats.ierrors;
1057 /* tx queue statistics */
1058 imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
1059 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
1061 for (i = 0; i < imax; i++) {
1062 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
1063 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
1064 tx_total += tap_stats->q_opackets[i];
1065 tx_err_total += pmd->txq[i].stats.errs;
1066 tx_bytes_total += tap_stats->q_obytes[i];
1069 tap_stats->ipackets = rx_total;
1070 tap_stats->ibytes = rx_bytes_total;
1071 tap_stats->ierrors = ierrors;
1072 tap_stats->rx_nombuf = rx_nombuf;
1073 tap_stats->opackets = tx_total;
1074 tap_stats->oerrors = tx_err_total;
1075 tap_stats->obytes = tx_bytes_total;
1080 tap_stats_reset(struct rte_eth_dev *dev)
1083 struct pmd_internals *pmd = dev->data->dev_private;
1085 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1086 pmd->rxq[i].stats.ipackets = 0;
1087 pmd->rxq[i].stats.ibytes = 0;
1088 pmd->rxq[i].stats.ierrors = 0;
1089 pmd->rxq[i].stats.rx_nombuf = 0;
1091 pmd->txq[i].stats.opackets = 0;
1092 pmd->txq[i].stats.errs = 0;
1093 pmd->txq[i].stats.obytes = 0;
1100 tap_dev_close(struct rte_eth_dev *dev)
1103 struct pmd_internals *internals = dev->data->dev_private;
1104 struct pmd_process_private *process_private = dev->process_private;
1105 struct rx_queue *rxq;
1107 if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
1108 rte_free(dev->process_private);
1112 tap_link_set_down(dev);
1113 if (internals->nlsk_fd != -1) {
1114 tap_flow_flush(dev, NULL);
1115 tap_flow_implicit_flush(internals, NULL);
1116 tap_nl_final(internals->nlsk_fd);
1117 internals->nlsk_fd = -1;
1120 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1121 if (process_private->rxq_fds[i] != -1) {
1122 rxq = &internals->rxq[i];
1123 close(process_private->rxq_fds[i]);
1124 process_private->rxq_fds[i] = -1;
1125 tap_rxq_pool_free(rxq->pool);
1126 rte_free(rxq->iovecs);
1130 if (process_private->txq_fds[i] != -1) {
1131 close(process_private->txq_fds[i]);
1132 process_private->txq_fds[i] = -1;
1136 if (internals->remote_if_index) {
1137 /* Restore initial remote state */
1138 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
1139 &internals->remote_initial_flags);
1142 rte_mempool_free(internals->gso_ctx_mp);
1143 internals->gso_ctx_mp = NULL;
1145 if (internals->ka_fd != -1) {
1146 close(internals->ka_fd);
1147 internals->ka_fd = -1;
1150 /* mac_addrs must not be freed alone because part of dev_private */
1151 dev->data->mac_addrs = NULL;
1153 internals = dev->data->dev_private;
1154 TAP_LOG(DEBUG, "Closing %s Ethernet device on numa %u",
1155 tuntap_types[internals->type], rte_socket_id());
1157 if (internals->ioctl_sock != -1) {
1158 close(internals->ioctl_sock);
1159 internals->ioctl_sock = -1;
1161 rte_free(dev->process_private);
1162 if (tap_devices_count == 1)
1163 rte_mp_action_unregister(TAP_MP_KEY);
1164 tap_devices_count--;
1166 * Since TUN device has no more opened file descriptors
1167 * it will be removed from kernel
1174 tap_rx_queue_release(void *queue)
1176 struct rx_queue *rxq = queue;
1177 struct pmd_process_private *process_private;
1181 process_private = rte_eth_devices[rxq->in_port].process_private;
1182 if (process_private->rxq_fds[rxq->queue_id] != -1) {
1183 close(process_private->rxq_fds[rxq->queue_id]);
1184 process_private->rxq_fds[rxq->queue_id] = -1;
1185 tap_rxq_pool_free(rxq->pool);
1186 rte_free(rxq->iovecs);
1193 tap_tx_queue_release(void *queue)
1195 struct tx_queue *txq = queue;
1196 struct pmd_process_private *process_private;
1200 process_private = rte_eth_devices[txq->out_port].process_private;
1202 if (process_private->txq_fds[txq->queue_id] != -1) {
1203 close(process_private->txq_fds[txq->queue_id]);
1204 process_private->txq_fds[txq->queue_id] = -1;
1209 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
1211 struct rte_eth_link *dev_link = &dev->data->dev_link;
1212 struct pmd_internals *pmd = dev->data->dev_private;
1213 struct ifreq ifr = { .ifr_flags = 0 };
1215 if (pmd->remote_if_index) {
1216 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
1217 if (!(ifr.ifr_flags & IFF_UP) ||
1218 !(ifr.ifr_flags & IFF_RUNNING)) {
1219 dev_link->link_status = ETH_LINK_DOWN;
1223 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
1224 dev_link->link_status =
1225 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
1232 tap_promisc_enable(struct rte_eth_dev *dev)
1234 struct pmd_internals *pmd = dev->data->dev_private;
1235 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1238 ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1242 if (pmd->remote_if_index && !pmd->flow_isolate) {
1243 dev->data->promiscuous = 1;
1244 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
1246 /* Rollback promisc flag */
1247 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1249 * rte_eth_dev_promiscuous_enable() rollback
1250 * dev->data->promiscuous in the case of failure.
1260 tap_promisc_disable(struct rte_eth_dev *dev)
1262 struct pmd_internals *pmd = dev->data->dev_private;
1263 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1266 ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1270 if (pmd->remote_if_index && !pmd->flow_isolate) {
1271 dev->data->promiscuous = 0;
1272 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
1274 /* Rollback promisc flag */
1275 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1277 * rte_eth_dev_promiscuous_disable() rollback
1278 * dev->data->promiscuous in the case of failure.
1288 tap_allmulti_enable(struct rte_eth_dev *dev)
1290 struct pmd_internals *pmd = dev->data->dev_private;
1291 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1294 ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1298 if (pmd->remote_if_index && !pmd->flow_isolate) {
1299 dev->data->all_multicast = 1;
1300 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
1302 /* Rollback allmulti flag */
1303 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1305 * rte_eth_dev_allmulticast_enable() rollback
1306 * dev->data->all_multicast in the case of failure.
1316 tap_allmulti_disable(struct rte_eth_dev *dev)
1318 struct pmd_internals *pmd = dev->data->dev_private;
1319 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1322 ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1326 if (pmd->remote_if_index && !pmd->flow_isolate) {
1327 dev->data->all_multicast = 0;
1328 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
1330 /* Rollback allmulti flag */
1331 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1333 * rte_eth_dev_allmulticast_disable() rollback
1334 * dev->data->all_multicast in the case of failure.
1344 tap_mac_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1346 struct pmd_internals *pmd = dev->data->dev_private;
1347 enum ioctl_mode mode = LOCAL_ONLY;
1351 if (pmd->type == ETH_TUNTAP_TYPE_TUN) {
1352 TAP_LOG(ERR, "%s: can't MAC address for TUN",
1357 if (rte_is_zero_ether_addr(mac_addr)) {
1358 TAP_LOG(ERR, "%s: can't set an empty MAC address",
1362 /* Check the actual current MAC address on the tap netdevice */
1363 ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY);
1366 if (rte_is_same_ether_addr(
1367 (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1370 /* Check the current MAC address on the remote */
1371 ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY);
1374 if (!rte_is_same_ether_addr(
1375 (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1377 mode = LOCAL_AND_REMOTE;
1378 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1379 rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, RTE_ETHER_ADDR_LEN);
1380 ret = tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode);
1383 rte_memcpy(&pmd->eth_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1384 if (pmd->remote_if_index && !pmd->flow_isolate) {
1385 /* Replace MAC redirection rule after a MAC change */
1386 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC);
1389 "%s: Couldn't delete MAC redirection rule",
1393 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC);
1396 "%s: Couldn't add MAC redirection rule",
1406 tap_gso_ctx_setup(struct rte_gso_ctx *gso_ctx, struct rte_eth_dev *dev)
1410 struct pmd_internals *pmd = dev->data->dev_private;
1413 /* initialize GSO context */
1414 gso_types = DEV_TX_OFFLOAD_TCP_TSO;
1415 if (!pmd->gso_ctx_mp) {
1417 * Create private mbuf pool with TAP_GSO_MBUF_SEG_SIZE
1418 * bytes size per mbuf use this pool for both direct and
1421 ret = snprintf(pool_name, sizeof(pool_name), "mp_%s",
1423 if (ret < 0 || ret >= (int)sizeof(pool_name)) {
1425 "%s: failed to create mbuf pool name for device %s,"
1426 "device name too long or output error, ret: %d\n",
1427 pmd->name, dev->device->name, ret);
1428 return -ENAMETOOLONG;
1430 pmd->gso_ctx_mp = rte_pktmbuf_pool_create(pool_name,
1431 TAP_GSO_MBUFS_NUM, TAP_GSO_MBUF_CACHE_SIZE, 0,
1432 RTE_PKTMBUF_HEADROOM + TAP_GSO_MBUF_SEG_SIZE,
1434 if (!pmd->gso_ctx_mp) {
1436 "%s: failed to create mbuf pool for device %s\n",
1437 pmd->name, dev->device->name);
1442 gso_ctx->direct_pool = pmd->gso_ctx_mp;
1443 gso_ctx->indirect_pool = pmd->gso_ctx_mp;
1444 gso_ctx->gso_types = gso_types;
1445 gso_ctx->gso_size = 0; /* gso_size is set in tx_burst() per packet */
1452 tap_setup_queue(struct rte_eth_dev *dev,
1453 struct pmd_internals *internals,
1461 struct pmd_internals *pmd = dev->data->dev_private;
1462 struct pmd_process_private *process_private = dev->process_private;
1463 struct rx_queue *rx = &internals->rxq[qid];
1464 struct tx_queue *tx = &internals->txq[qid];
1465 struct rte_gso_ctx *gso_ctx;
1468 fd = &process_private->rxq_fds[qid];
1469 other_fd = &process_private->txq_fds[qid];
1473 fd = &process_private->txq_fds[qid];
1474 other_fd = &process_private->rxq_fds[qid];
1476 gso_ctx = &tx->gso_ctx;
1479 /* fd for this queue already exists */
1480 TAP_LOG(DEBUG, "%s: fd %d for %s queue qid %d exists",
1481 pmd->name, *fd, dir, qid);
1483 } else if (*other_fd != -1) {
1484 /* Only other_fd exists. dup it */
1485 *fd = dup(*other_fd);
1488 TAP_LOG(ERR, "%s: dup() failed.", pmd->name);
1491 TAP_LOG(DEBUG, "%s: dup fd %d for %s queue qid %d (%d)",
1492 pmd->name, *other_fd, dir, qid, *fd);
1494 /* Both RX and TX fds do not exist (equal -1). Create fd */
1495 *fd = tun_alloc(pmd, 0);
1497 *fd = -1; /* restore original value */
1498 TAP_LOG(ERR, "%s: tun_alloc() failed.", pmd->name);
1501 TAP_LOG(DEBUG, "%s: add %s queue for qid %d fd %d",
1502 pmd->name, dir, qid, *fd);
1505 tx->mtu = &dev->data->mtu;
1506 rx->rxmode = &dev->data->dev_conf.rxmode;
1508 ret = tap_gso_ctx_setup(gso_ctx, dev);
1513 tx->type = pmd->type;
1519 tap_rx_queue_setup(struct rte_eth_dev *dev,
1520 uint16_t rx_queue_id,
1521 uint16_t nb_rx_desc,
1522 unsigned int socket_id,
1523 const struct rte_eth_rxconf *rx_conf __rte_unused,
1524 struct rte_mempool *mp)
1526 struct pmd_internals *internals = dev->data->dev_private;
1527 struct pmd_process_private *process_private = dev->process_private;
1528 struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1529 struct rte_mbuf **tmp = &rxq->pool;
1530 long iov_max = sysconf(_SC_IOV_MAX);
1534 "_SC_IOV_MAX is not defined. Using %d as default",
1535 TAP_IOV_DEFAULT_MAX);
1536 iov_max = TAP_IOV_DEFAULT_MAX;
1538 uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1539 struct iovec (*iovecs)[nb_desc + 1];
1540 int data_off = RTE_PKTMBUF_HEADROOM;
1545 if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1547 "nb_rx_queues %d too small or mempool NULL",
1548 dev->data->nb_rx_queues);
1553 rxq->trigger_seen = 1; /* force initial burst */
1554 rxq->in_port = dev->data->port_id;
1555 rxq->queue_id = rx_queue_id;
1556 rxq->nb_rx_desc = nb_desc;
1557 iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1561 "%s: Couldn't allocate %d RX descriptors",
1562 dev->device->name, nb_desc);
1565 rxq->iovecs = iovecs;
1567 dev->data->rx_queues[rx_queue_id] = rxq;
1568 fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1574 (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1575 (*rxq->iovecs)[0].iov_base = &rxq->pi;
1577 for (i = 1; i <= nb_desc; i++) {
1578 *tmp = rte_pktmbuf_alloc(rxq->mp);
1581 "%s: couldn't allocate memory for queue %d",
1582 dev->device->name, rx_queue_id);
1586 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1587 (*rxq->iovecs)[i].iov_base =
1588 (char *)(*tmp)->buf_addr + data_off;
1590 tmp = &(*tmp)->next;
1593 TAP_LOG(DEBUG, " RX TUNTAP device name %s, qid %d on fd %d",
1594 internals->name, rx_queue_id,
1595 process_private->rxq_fds[rx_queue_id]);
1600 tap_rxq_pool_free(rxq->pool);
1602 rte_free(rxq->iovecs);
1608 tap_tx_queue_setup(struct rte_eth_dev *dev,
1609 uint16_t tx_queue_id,
1610 uint16_t nb_tx_desc __rte_unused,
1611 unsigned int socket_id __rte_unused,
1612 const struct rte_eth_txconf *tx_conf)
1614 struct pmd_internals *internals = dev->data->dev_private;
1615 struct pmd_process_private *process_private = dev->process_private;
1616 struct tx_queue *txq;
1620 if (tx_queue_id >= dev->data->nb_tx_queues)
1622 dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1623 txq = dev->data->tx_queues[tx_queue_id];
1624 txq->out_port = dev->data->port_id;
1625 txq->queue_id = tx_queue_id;
1627 offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1628 txq->csum = !!(offloads &
1629 (DEV_TX_OFFLOAD_IPV4_CKSUM |
1630 DEV_TX_OFFLOAD_UDP_CKSUM |
1631 DEV_TX_OFFLOAD_TCP_CKSUM));
1633 ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1637 " TX TUNTAP device name %s, qid %d on fd %d csum %s",
1638 internals->name, tx_queue_id,
1639 process_private->txq_fds[tx_queue_id],
1640 txq->csum ? "on" : "off");
1646 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1648 struct pmd_internals *pmd = dev->data->dev_private;
1649 struct ifreq ifr = { .ifr_mtu = mtu };
1652 err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1654 dev->data->mtu = mtu;
1660 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1661 struct rte_ether_addr *mc_addr_set __rte_unused,
1662 uint32_t nb_mc_addr __rte_unused)
1665 * Nothing to do actually: the tap has no filtering whatsoever, every
1666 * packet is received.
1672 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1674 struct rte_eth_dev *dev = arg;
1675 struct pmd_internals *pmd = dev->data->dev_private;
1676 struct ifinfomsg *info = NLMSG_DATA(nh);
1678 if (nh->nlmsg_type != RTM_NEWLINK ||
1679 (info->ifi_index != pmd->if_index &&
1680 info->ifi_index != pmd->remote_if_index))
1682 return tap_link_update(dev, 0);
1686 tap_dev_intr_handler(void *cb_arg)
1688 struct rte_eth_dev *dev = cb_arg;
1689 struct pmd_internals *pmd = dev->data->dev_private;
1691 tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1695 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1697 struct pmd_internals *pmd = dev->data->dev_private;
1700 /* In any case, disable interrupt if the conf is no longer there. */
1701 if (!dev->data->dev_conf.intr_conf.lsc) {
1702 if (pmd->intr_handle.fd != -1) {
1708 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1709 if (unlikely(pmd->intr_handle.fd == -1))
1711 return rte_intr_callback_register(
1712 &pmd->intr_handle, tap_dev_intr_handler, dev);
1717 ret = rte_intr_callback_unregister(&pmd->intr_handle,
1718 tap_dev_intr_handler, dev);
1721 } else if (ret == -EAGAIN) {
1724 TAP_LOG(ERR, "intr callback unregister failed: %d",
1730 tap_nl_final(pmd->intr_handle.fd);
1731 pmd->intr_handle.fd = -1;
1737 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1741 err = tap_lsc_intr_handle_set(dev, set);
1744 tap_rx_intr_vec_set(dev, 0);
1747 err = tap_rx_intr_vec_set(dev, set);
1749 tap_lsc_intr_handle_set(dev, 0);
1753 static const uint32_t*
1754 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1756 static const uint32_t ptypes[] = {
1757 RTE_PTYPE_INNER_L2_ETHER,
1758 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1759 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1760 RTE_PTYPE_INNER_L3_IPV4,
1761 RTE_PTYPE_INNER_L3_IPV4_EXT,
1762 RTE_PTYPE_INNER_L3_IPV6,
1763 RTE_PTYPE_INNER_L3_IPV6_EXT,
1764 RTE_PTYPE_INNER_L4_FRAG,
1765 RTE_PTYPE_INNER_L4_UDP,
1766 RTE_PTYPE_INNER_L4_TCP,
1767 RTE_PTYPE_INNER_L4_SCTP,
1769 RTE_PTYPE_L2_ETHER_VLAN,
1770 RTE_PTYPE_L2_ETHER_QINQ,
1772 RTE_PTYPE_L3_IPV4_EXT,
1773 RTE_PTYPE_L3_IPV6_EXT,
1785 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1786 struct rte_eth_fc_conf *fc_conf)
1788 fc_conf->mode = RTE_FC_NONE;
1793 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1794 struct rte_eth_fc_conf *fc_conf)
1796 if (fc_conf->mode != RTE_FC_NONE)
1802 * DPDK callback to update the RSS hash configuration.
1805 * Pointer to Ethernet device structure.
1806 * @param[in] rss_conf
1807 * RSS configuration data.
1810 * 0 on success, a negative errno value otherwise and rte_errno is set.
1813 tap_rss_hash_update(struct rte_eth_dev *dev,
1814 struct rte_eth_rss_conf *rss_conf)
1816 if (rss_conf->rss_hf & TAP_RSS_HF_MASK) {
1820 if (rss_conf->rss_key && rss_conf->rss_key_len) {
1822 * Currently TAP RSS key is hard coded
1823 * and cannot be updated
1826 "port %u RSS key cannot be updated",
1827 dev->data->port_id);
1835 tap_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1837 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1843 tap_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1845 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1851 tap_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1853 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1859 tap_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1861 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1865 static const struct eth_dev_ops ops = {
1866 .dev_start = tap_dev_start,
1867 .dev_stop = tap_dev_stop,
1868 .dev_close = tap_dev_close,
1869 .dev_configure = tap_dev_configure,
1870 .dev_infos_get = tap_dev_info,
1871 .rx_queue_setup = tap_rx_queue_setup,
1872 .tx_queue_setup = tap_tx_queue_setup,
1873 .rx_queue_start = tap_rx_queue_start,
1874 .tx_queue_start = tap_tx_queue_start,
1875 .rx_queue_stop = tap_rx_queue_stop,
1876 .tx_queue_stop = tap_tx_queue_stop,
1877 .rx_queue_release = tap_rx_queue_release,
1878 .tx_queue_release = tap_tx_queue_release,
1879 .flow_ctrl_get = tap_flow_ctrl_get,
1880 .flow_ctrl_set = tap_flow_ctrl_set,
1881 .link_update = tap_link_update,
1882 .dev_set_link_up = tap_link_set_up,
1883 .dev_set_link_down = tap_link_set_down,
1884 .promiscuous_enable = tap_promisc_enable,
1885 .promiscuous_disable = tap_promisc_disable,
1886 .allmulticast_enable = tap_allmulti_enable,
1887 .allmulticast_disable = tap_allmulti_disable,
1888 .mac_addr_set = tap_mac_set,
1889 .mtu_set = tap_mtu_set,
1890 .set_mc_addr_list = tap_set_mc_addr_list,
1891 .stats_get = tap_stats_get,
1892 .stats_reset = tap_stats_reset,
1893 .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1894 .rss_hash_update = tap_rss_hash_update,
1895 .flow_ops_get = tap_dev_flow_ops_get,
1899 eth_dev_tap_create(struct rte_vdev_device *vdev, const char *tap_name,
1900 char *remote_iface, struct rte_ether_addr *mac_addr,
1901 enum rte_tuntap_type type)
1903 int numa_node = rte_socket_id();
1904 struct rte_eth_dev *dev;
1905 struct pmd_internals *pmd;
1906 struct pmd_process_private *process_private;
1907 const char *tuntap_name = tuntap_types[type];
1908 struct rte_eth_dev_data *data;
1912 TAP_LOG(DEBUG, "%s device on numa %u", tuntap_name, rte_socket_id());
1914 dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1916 TAP_LOG(ERR, "%s Unable to allocate device struct",
1918 goto error_exit_nodev;
1921 process_private = (struct pmd_process_private *)
1922 rte_zmalloc_socket(tap_name, sizeof(struct pmd_process_private),
1923 RTE_CACHE_LINE_SIZE, dev->device->numa_node);
1925 if (process_private == NULL) {
1926 TAP_LOG(ERR, "Failed to alloc memory for process private");
1929 pmd = dev->data->dev_private;
1930 dev->process_private = process_private;
1932 strlcpy(pmd->name, tap_name, sizeof(pmd->name));
1936 pmd->gso_ctx_mp = NULL;
1938 pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1939 if (pmd->ioctl_sock == -1) {
1941 "%s Unable to get a socket for management: %s",
1942 tuntap_name, strerror(errno));
1946 /* Setup some default values */
1948 data->dev_private = pmd;
1949 data->dev_flags = RTE_ETH_DEV_INTR_LSC |
1950 RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
1951 data->numa_node = numa_node;
1953 data->dev_link = pmd_link;
1954 data->mac_addrs = &pmd->eth_addr;
1955 /* Set the number of RX and TX queues */
1956 data->nb_rx_queues = 0;
1957 data->nb_tx_queues = 0;
1959 dev->dev_ops = &ops;
1960 dev->rx_pkt_burst = pmd_rx_burst;
1961 dev->tx_pkt_burst = pmd_tx_burst;
1963 pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1964 pmd->intr_handle.fd = -1;
1965 dev->intr_handle = &pmd->intr_handle;
1967 /* Presetup the fds to -1 as being not valid */
1968 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1969 process_private->rxq_fds[i] = -1;
1970 process_private->txq_fds[i] = -1;
1973 if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1974 if (rte_is_zero_ether_addr(mac_addr))
1975 rte_eth_random_addr((uint8_t *)&pmd->eth_addr);
1977 rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1981 * Allocate a TUN device keep-alive file descriptor that will only be
1982 * closed when the TUN device itself is closed or removed.
1983 * This keep-alive file descriptor will guarantee that the TUN device
1984 * exists even when all of its queues are closed
1986 pmd->ka_fd = tun_alloc(pmd, 1);
1987 if (pmd->ka_fd == -1) {
1988 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
1991 TAP_LOG(DEBUG, "allocated %s", pmd->name);
1993 ifr.ifr_mtu = dev->data->mtu;
1994 if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1997 if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1998 memset(&ifr, 0, sizeof(struct ifreq));
1999 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
2000 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
2001 RTE_ETHER_ADDR_LEN);
2002 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
2007 * Set up everything related to rte_flow:
2009 * - tap / remote if_index
2010 * - mandatory QDISCs
2011 * - rte_flow actual/implicit lists
2014 pmd->nlsk_fd = tap_nl_init(0);
2015 if (pmd->nlsk_fd == -1) {
2016 TAP_LOG(WARNING, "%s: failed to create netlink socket.",
2018 goto disable_rte_flow;
2020 pmd->if_index = if_nametoindex(pmd->name);
2021 if (!pmd->if_index) {
2022 TAP_LOG(ERR, "%s: failed to get if_index.", pmd->name);
2023 goto disable_rte_flow;
2025 if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
2026 TAP_LOG(ERR, "%s: failed to create multiq qdisc.",
2028 goto disable_rte_flow;
2030 if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
2031 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
2033 goto disable_rte_flow;
2035 LIST_INIT(&pmd->flows);
2037 if (strlen(remote_iface)) {
2038 pmd->remote_if_index = if_nametoindex(remote_iface);
2039 if (!pmd->remote_if_index) {
2040 TAP_LOG(ERR, "%s: failed to get %s if_index.",
2041 pmd->name, remote_iface);
2044 strlcpy(pmd->remote_iface, remote_iface, RTE_ETH_NAME_MAX_LEN);
2046 /* Save state of remote device */
2047 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
2049 /* Replicate remote MAC address */
2050 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
2051 TAP_LOG(ERR, "%s: failed to get %s MAC address.",
2052 pmd->name, pmd->remote_iface);
2055 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
2056 RTE_ETHER_ADDR_LEN);
2057 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
2058 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
2059 TAP_LOG(ERR, "%s: failed to get %s MAC address.",
2060 pmd->name, remote_iface);
2065 * Flush usually returns negative value because it tries to
2066 * delete every QDISC (and on a running device, one QDISC at
2067 * least is needed). Ignore negative return value.
2069 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
2070 if (qdisc_create_ingress(pmd->nlsk_fd,
2071 pmd->remote_if_index) < 0) {
2072 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
2076 LIST_INIT(&pmd->implicit_flows);
2077 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
2078 tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
2079 tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
2080 tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
2082 "%s: failed to create implicit rules.",
2088 rte_eth_dev_probing_finish(dev);
2092 TAP_LOG(ERR, " Disabling rte flow support: %s(%d)",
2093 strerror(errno), errno);
2094 if (strlen(remote_iface)) {
2095 TAP_LOG(ERR, "Remote feature requires flow support.");
2098 rte_eth_dev_probing_finish(dev);
2102 TAP_LOG(ERR, " Can't set up remote feature: %s(%d)",
2103 strerror(errno), errno);
2104 tap_flow_implicit_flush(pmd, NULL);
2107 if (pmd->nlsk_fd != -1)
2108 close(pmd->nlsk_fd);
2109 if (pmd->ka_fd != -1)
2111 if (pmd->ioctl_sock != -1)
2112 close(pmd->ioctl_sock);
2113 /* mac_addrs must not be freed alone because part of dev_private */
2114 dev->data->mac_addrs = NULL;
2115 rte_eth_dev_release_port(dev);
2118 TAP_LOG(ERR, "%s Unable to initialize %s",
2119 tuntap_name, rte_vdev_device_name(vdev));
2124 /* make sure name is a possible Linux network device name */
2126 is_valid_iface(const char *name)
2131 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
2135 if (*name == '/' || *name == ':' || isspace(*name))
2143 set_interface_name(const char *key __rte_unused,
2147 char *name = (char *)extra_args;
2150 if (!is_valid_iface(value)) {
2151 TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
2155 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
2157 /* use tap%d which causes kernel to choose next available */
2158 strlcpy(name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2164 set_remote_iface(const char *key __rte_unused,
2168 char *name = (char *)extra_args;
2171 if (!is_valid_iface(value)) {
2172 TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
2176 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
2182 static int parse_user_mac(struct rte_ether_addr *user_mac,
2185 unsigned int index = 0;
2186 char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
2188 if (user_mac == NULL || value == NULL)
2191 strlcpy(mac_temp, value, sizeof(mac_temp));
2192 mac_byte = strtok(mac_temp, ":");
2194 while ((mac_byte != NULL) &&
2195 (strlen(mac_byte) <= 2) &&
2196 (strlen(mac_byte) == strspn(mac_byte,
2197 ETH_TAP_CMP_MAC_FMT))) {
2198 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
2199 mac_byte = strtok(NULL, ":");
2206 set_mac_type(const char *key __rte_unused,
2210 struct rte_ether_addr *user_mac = extra_args;
2215 if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
2216 static int iface_idx;
2218 /* fixed mac = 00:64:74:61:70:<iface_idx> */
2219 memcpy((char *)user_mac->addr_bytes, "\0dtap",
2220 RTE_ETHER_ADDR_LEN);
2221 user_mac->addr_bytes[RTE_ETHER_ADDR_LEN - 1] =
2226 if (parse_user_mac(user_mac, value) != 6)
2229 TAP_LOG(DEBUG, "TAP user MAC param (%s)", value);
2233 TAP_LOG(ERR, "TAP user MAC (%s) is not in format (%s|%s)",
2234 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
2239 * Open a TUN interface device. TUN PMD
2240 * 1) sets tap_type as false
2241 * 2) intakes iface as argument.
2242 * 3) as interface is virtual set speed to 10G
2245 rte_pmd_tun_probe(struct rte_vdev_device *dev)
2247 const char *name, *params;
2249 struct rte_kvargs *kvlist = NULL;
2250 char tun_name[RTE_ETH_NAME_MAX_LEN];
2251 char remote_iface[RTE_ETH_NAME_MAX_LEN];
2252 struct rte_eth_dev *eth_dev;
2254 name = rte_vdev_device_name(dev);
2255 params = rte_vdev_device_args(dev);
2256 memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2258 if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
2259 strlen(params) == 0) {
2260 eth_dev = rte_eth_dev_attach_secondary(name);
2262 TAP_LOG(ERR, "Failed to probe %s", name);
2265 eth_dev->dev_ops = &ops;
2266 eth_dev->device = &dev->device;
2267 rte_eth_dev_probing_finish(eth_dev);
2271 /* use tun%d which causes kernel to choose next available */
2272 strlcpy(tun_name, DEFAULT_TUN_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2274 if (params && (params[0] != '\0')) {
2275 TAP_LOG(DEBUG, "parameters (%s)", params);
2277 kvlist = rte_kvargs_parse(params, valid_arguments);
2279 if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2280 ret = rte_kvargs_process(kvlist,
2282 &set_interface_name,
2290 pmd_link.link_speed = ETH_SPEED_NUM_10G;
2292 TAP_LOG(DEBUG, "Initializing pmd_tun for %s", name);
2294 ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0,
2295 ETH_TUNTAP_TYPE_TUN);
2299 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2302 rte_kvargs_free(kvlist);
2307 /* Request queue file descriptors from secondary to primary. */
2309 tap_mp_attach_queues(const char *port_name, struct rte_eth_dev *dev)
2312 struct timespec timeout = {.tv_sec = 1, .tv_nsec = 0};
2313 struct rte_mp_msg request, *reply;
2314 struct rte_mp_reply replies;
2315 struct ipc_queues *request_param = (struct ipc_queues *)request.param;
2316 struct ipc_queues *reply_param;
2317 struct pmd_process_private *process_private = dev->process_private;
2318 int queue, fd_iterator;
2320 /* Prepare the request */
2321 memset(&request, 0, sizeof(request));
2322 strlcpy(request.name, TAP_MP_KEY, sizeof(request.name));
2323 strlcpy(request_param->port_name, port_name,
2324 sizeof(request_param->port_name));
2325 request.len_param = sizeof(*request_param);
2326 /* Send request and receive reply */
2327 ret = rte_mp_request_sync(&request, &replies, &timeout);
2328 if (ret < 0 || replies.nb_received != 1) {
2329 TAP_LOG(ERR, "Failed to request queues from primary: %d",
2333 reply = &replies.msgs[0];
2334 reply_param = (struct ipc_queues *)reply->param;
2335 TAP_LOG(DEBUG, "Received IPC reply for %s", reply_param->port_name);
2337 /* Attach the queues from received file descriptors */
2338 if (reply_param->rxq_count + reply_param->txq_count != reply->num_fds) {
2339 TAP_LOG(ERR, "Unexpected number of fds received");
2343 dev->data->nb_rx_queues = reply_param->rxq_count;
2344 dev->data->nb_tx_queues = reply_param->txq_count;
2346 for (queue = 0; queue < reply_param->rxq_count; queue++)
2347 process_private->rxq_fds[queue] = reply->fds[fd_iterator++];
2348 for (queue = 0; queue < reply_param->txq_count; queue++)
2349 process_private->txq_fds[queue] = reply->fds[fd_iterator++];
2354 /* Send the queue file descriptors from the primary process to secondary. */
2356 tap_mp_sync_queues(const struct rte_mp_msg *request, const void *peer)
2358 struct rte_eth_dev *dev;
2359 struct pmd_process_private *process_private;
2360 struct rte_mp_msg reply;
2361 const struct ipc_queues *request_param =
2362 (const struct ipc_queues *)request->param;
2363 struct ipc_queues *reply_param =
2364 (struct ipc_queues *)reply.param;
2369 /* Get requested port */
2370 TAP_LOG(DEBUG, "Received IPC request for %s", request_param->port_name);
2371 ret = rte_eth_dev_get_port_by_name(request_param->port_name, &port_id);
2373 TAP_LOG(ERR, "Failed to get port id for %s",
2374 request_param->port_name);
2377 dev = &rte_eth_devices[port_id];
2378 process_private = dev->process_private;
2380 /* Fill file descriptors for all queues */
2382 reply_param->rxq_count = 0;
2383 if (dev->data->nb_rx_queues + dev->data->nb_tx_queues >
2385 TAP_LOG(ERR, "Number of rx/tx queues exceeds max number of fds");
2389 for (queue = 0; queue < dev->data->nb_rx_queues; queue++) {
2390 reply.fds[reply.num_fds++] = process_private->rxq_fds[queue];
2391 reply_param->rxq_count++;
2393 RTE_ASSERT(reply_param->rxq_count == dev->data->nb_rx_queues);
2395 reply_param->txq_count = 0;
2396 for (queue = 0; queue < dev->data->nb_tx_queues; queue++) {
2397 reply.fds[reply.num_fds++] = process_private->txq_fds[queue];
2398 reply_param->txq_count++;
2400 RTE_ASSERT(reply_param->txq_count == dev->data->nb_tx_queues);
2403 strlcpy(reply.name, request->name, sizeof(reply.name));
2404 strlcpy(reply_param->port_name, request_param->port_name,
2405 sizeof(reply_param->port_name));
2406 reply.len_param = sizeof(*reply_param);
2407 if (rte_mp_reply(&reply, peer) < 0) {
2408 TAP_LOG(ERR, "Failed to reply an IPC request to sync queues");
2414 /* Open a TAP interface device.
2417 rte_pmd_tap_probe(struct rte_vdev_device *dev)
2419 const char *name, *params;
2421 struct rte_kvargs *kvlist = NULL;
2423 char tap_name[RTE_ETH_NAME_MAX_LEN];
2424 char remote_iface[RTE_ETH_NAME_MAX_LEN];
2425 struct rte_ether_addr user_mac = { .addr_bytes = {0} };
2426 struct rte_eth_dev *eth_dev;
2427 int tap_devices_count_increased = 0;
2429 name = rte_vdev_device_name(dev);
2430 params = rte_vdev_device_args(dev);
2432 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
2433 eth_dev = rte_eth_dev_attach_secondary(name);
2435 TAP_LOG(ERR, "Failed to probe %s", name);
2438 eth_dev->dev_ops = &ops;
2439 eth_dev->device = &dev->device;
2440 eth_dev->rx_pkt_burst = pmd_rx_burst;
2441 eth_dev->tx_pkt_burst = pmd_tx_burst;
2442 if (!rte_eal_primary_proc_alive(NULL)) {
2443 TAP_LOG(ERR, "Primary process is missing");
2446 eth_dev->process_private = (struct pmd_process_private *)
2447 rte_zmalloc_socket(name,
2448 sizeof(struct pmd_process_private),
2449 RTE_CACHE_LINE_SIZE,
2450 eth_dev->device->numa_node);
2451 if (eth_dev->process_private == NULL) {
2453 "Failed to alloc memory for process private");
2457 ret = tap_mp_attach_queues(name, eth_dev);
2460 rte_eth_dev_probing_finish(eth_dev);
2464 speed = ETH_SPEED_NUM_10G;
2466 /* use tap%d which causes kernel to choose next available */
2467 strlcpy(tap_name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2468 memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2470 if (params && (params[0] != '\0')) {
2471 TAP_LOG(DEBUG, "parameters (%s)", params);
2473 kvlist = rte_kvargs_parse(params, valid_arguments);
2475 if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2476 ret = rte_kvargs_process(kvlist,
2478 &set_interface_name,
2484 if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
2485 ret = rte_kvargs_process(kvlist,
2493 if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
2494 ret = rte_kvargs_process(kvlist,
2503 pmd_link.link_speed = speed;
2505 TAP_LOG(DEBUG, "Initializing pmd_tap for %s", name);
2507 /* Register IPC feed callback */
2508 if (!tap_devices_count) {
2509 ret = rte_mp_action_register(TAP_MP_KEY, tap_mp_sync_queues);
2510 if (ret < 0 && rte_errno != ENOTSUP) {
2511 TAP_LOG(ERR, "tap: Failed to register IPC callback: %s",
2512 strerror(rte_errno));
2516 tap_devices_count++;
2517 tap_devices_count_increased = 1;
2518 ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac,
2519 ETH_TUNTAP_TYPE_TAP);
2523 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2525 if (tap_devices_count_increased == 1) {
2526 if (tap_devices_count == 1)
2527 rte_mp_action_unregister(TAP_MP_KEY);
2528 tap_devices_count--;
2531 rte_kvargs_free(kvlist);
2536 /* detach a TUNTAP device.
2539 rte_pmd_tap_remove(struct rte_vdev_device *dev)
2541 struct rte_eth_dev *eth_dev = NULL;
2543 /* find the ethdev entry */
2544 eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
2548 tap_dev_close(eth_dev);
2549 rte_eth_dev_release_port(eth_dev);
2554 static struct rte_vdev_driver pmd_tun_drv = {
2555 .probe = rte_pmd_tun_probe,
2556 .remove = rte_pmd_tap_remove,
2559 static struct rte_vdev_driver pmd_tap_drv = {
2560 .probe = rte_pmd_tap_probe,
2561 .remove = rte_pmd_tap_remove,
2564 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
2565 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
2566 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
2567 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
2568 ETH_TAP_IFACE_ARG "=<string> ");
2569 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
2570 ETH_TAP_IFACE_ARG "=<string> "
2571 ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
2572 ETH_TAP_REMOTE_ARG "=<string>");
2573 RTE_LOG_REGISTER(tap_logtype, pmd.net.tap, NOTICE);