]> git.proxmox.com Git - ceph.git/blob - ceph/src/spdk/dpdk/drivers/net/tap/rte_eth_tap.c
projects / ceph.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
update sources to ceph Nautilus 14.2.1
[ceph.git] / ceph / src / spdk / dpdk / drivers / net / tap / rte_eth_tap.c
1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2016-2017 Intel Corporation
3 */
4
5 #include <rte_atomic.h>
6 #include <rte_branch_prediction.h>
7 #include <rte_byteorder.h>
8 #include <rte_common.h>
9 #include <rte_mbuf.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>
15 #include <rte_net.h>
16 #include <rte_debug.h>
17 #include <rte_ip.h>
18 #include <rte_string_fns.h>
19
20 #include <assert.h>
21 #include <sys/types.h>
22 #include <sys/stat.h>
23 #include <sys/socket.h>
24 #include <sys/ioctl.h>
25 #include <sys/utsname.h>
26 #include <sys/mman.h>
27 #include <errno.h>
28 #include <signal.h>
29 #include <stdbool.h>
30 #include <stdint.h>
31 #include <sys/uio.h>
32 #include <unistd.h>
33 #include <arpa/inet.h>
34 #include <net/if.h>
35 #include <linux/if_tun.h>
36 #include <linux/if_ether.h>
37 #include <fcntl.h>
38
39 #include <tap_rss.h>
40 #include <rte_eth_tap.h>
41 #include <tap_flow.h>
42 #include <tap_netlink.h>
43 #include <tap_tcmsgs.h>
44
45 /* Linux based path to the TUN device */
46 #define TUN_TAP_DEV_PATH "/dev/net/tun"
47 #define DEFAULT_TAP_NAME "dtap"
48 #define DEFAULT_TUN_NAME "dtun"
49
50 #define ETH_TAP_IFACE_ARG "iface"
51 #define ETH_TAP_REMOTE_ARG "remote"
52 #define ETH_TAP_MAC_ARG "mac"
53 #define ETH_TAP_MAC_FIXED "fixed"
54
55 #define ETH_TAP_USR_MAC_FMT "xx:xx:xx:xx:xx:xx"
56 #define ETH_TAP_CMP_MAC_FMT "0123456789ABCDEFabcdef"
57 #define ETH_TAP_MAC_ARG_FMT ETH_TAP_MAC_FIXED "|" ETH_TAP_USR_MAC_FMT
58
59 #define TAP_GSO_MBUFS_PER_CORE 128
60 #define TAP_GSO_MBUF_SEG_SIZE 128
61 #define TAP_GSO_MBUF_CACHE_SIZE 4
62 #define TAP_GSO_MBUFS_NUM \
63 (TAP_GSO_MBUFS_PER_CORE * TAP_GSO_MBUF_CACHE_SIZE)
64
65 static struct rte_vdev_driver pmd_tap_drv;
66 static struct rte_vdev_driver pmd_tun_drv;
67
68 static const char *valid_arguments[] = {
69 ETH_TAP_IFACE_ARG,
70 ETH_TAP_REMOTE_ARG,
71 ETH_TAP_MAC_ARG,
72 NULL
73 };
74
75 static unsigned int tap_unit;
76 static unsigned int tun_unit;
77
78 static char tuntap_name[8];
79
80 static volatile uint32_t tap_trigger; /* Rx trigger */
81
82 static struct rte_eth_link pmd_link = {
83 .link_speed = ETH_SPEED_NUM_10G,
84 .link_duplex = ETH_LINK_FULL_DUPLEX,
85 .link_status = ETH_LINK_DOWN,
86 .link_autoneg = ETH_LINK_FIXED,
87 };
88
89 static void
90 tap_trigger_cb(int sig __rte_unused)
91 {
92 /* Valid trigger values are nonzero */
93 tap_trigger = (tap_trigger + 1) | 0x80000000;
94 }
95
96 /* Specifies on what netdevices the ioctl should be applied */
97 enum ioctl_mode {
98 LOCAL_AND_REMOTE,
99 LOCAL_ONLY,
100 REMOTE_ONLY,
101 };
102
103 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
104
105 /**
106 * Tun/Tap allocation routine
107 *
108 * @param[in] pmd
109 * Pointer to private structure.
110 *
111 * @param[in] is_keepalive
112 * Keepalive flag
113 *
114 * @return
115 * -1 on failure, fd on success
116 */
117 static int
118 tun_alloc(struct pmd_internals *pmd, int is_keepalive)
119 {
120 struct ifreq ifr;
121 #ifdef IFF_MULTI_QUEUE
122 unsigned int features;
123 #endif
124 int fd;
125
126 memset(&ifr, 0, sizeof(struct ifreq));
127
128 /*
129 * Do not set IFF_NO_PI as packet information header will be needed
130 * to check if a received packet has been truncated.
131 */
132 ifr.ifr_flags = (pmd->type == ETH_TUNTAP_TYPE_TAP) ?
133 IFF_TAP : IFF_TUN | IFF_POINTOPOINT;
134 snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
135
136 TAP_LOG(DEBUG, "ifr_name '%s'", ifr.ifr_name);
137
138 fd = open(TUN_TAP_DEV_PATH, O_RDWR);
139 if (fd < 0) {
140 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
141 goto error;
142 }
143
144 #ifdef IFF_MULTI_QUEUE
145 /* Grab the TUN features to verify we can work multi-queue */
146 if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
147 TAP_LOG(ERR, "%s unable to get TUN/TAP features",
148 tuntap_name);
149 goto error;
150 }
151 TAP_LOG(DEBUG, "%s Features %08x", tuntap_name, features);
152
153 if (features & IFF_MULTI_QUEUE) {
154 TAP_LOG(DEBUG, " Multi-queue support for %d queues",
155 RTE_PMD_TAP_MAX_QUEUES);
156 ifr.ifr_flags |= IFF_MULTI_QUEUE;
157 } else
158 #endif
159 {
160 ifr.ifr_flags |= IFF_ONE_QUEUE;
161 TAP_LOG(DEBUG, " Single queue only support");
162 }
163
164 /* Set the TUN/TAP configuration and set the name if needed */
165 if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
166 TAP_LOG(WARNING, "Unable to set TUNSETIFF for %s: %s",
167 ifr.ifr_name, strerror(errno));
168 goto error;
169 }
170
171 if (is_keepalive) {
172 /*
173 * Detach the TUN/TAP keep-alive queue
174 * to avoid traffic through it
175 */
176 ifr.ifr_flags = IFF_DETACH_QUEUE;
177 if (ioctl(fd, TUNSETQUEUE, (void *)&ifr) < 0) {
178 TAP_LOG(WARNING,
179 "Unable to detach keep-alive queue for %s: %s",
180 ifr.ifr_name, strerror(errno));
181 goto error;
182 }
183 }
184
185 /* Always set the file descriptor to non-blocking */
186 if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
187 TAP_LOG(WARNING,
188 "Unable to set %s to nonblocking: %s",
189 ifr.ifr_name, strerror(errno));
190 goto error;
191 }
192
193 /* Set up trigger to optimize empty Rx bursts */
194 errno = 0;
195 do {
196 struct sigaction sa;
197 int flags = fcntl(fd, F_GETFL);
198
199 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
200 break;
201 if (sa.sa_handler != tap_trigger_cb) {
202 /*
203 * Make sure SIGIO is not already taken. This is done
204 * as late as possible to leave the application a
205 * chance to set up its own signal handler first.
206 */
207 if (sa.sa_handler != SIG_IGN &&
208 sa.sa_handler != SIG_DFL) {
209 errno = EBUSY;
210 break;
211 }
212 sa = (struct sigaction){
213 .sa_flags = SA_RESTART,
214 .sa_handler = tap_trigger_cb,
215 };
216 if (sigaction(SIGIO, &sa, NULL) == -1)
217 break;
218 }
219 /* Enable SIGIO on file descriptor */
220 fcntl(fd, F_SETFL, flags | O_ASYNC);
221 fcntl(fd, F_SETOWN, getpid());
222 } while (0);
223
224 if (errno) {
225 /* Disable trigger globally in case of error */
226 tap_trigger = 0;
227 TAP_LOG(WARNING, "Rx trigger disabled: %s",
228 strerror(errno));
229 }
230
231 return fd;
232
233 error:
234 if (fd > 0)
235 close(fd);
236 return -1;
237 }
238
239 static void
240 tap_verify_csum(struct rte_mbuf *mbuf)
241 {
242 uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
243 uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
244 uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
245 unsigned int l2_len = sizeof(struct ether_hdr);
246 unsigned int l3_len;
247 uint16_t cksum = 0;
248 void *l3_hdr;
249 void *l4_hdr;
250
251 if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
252 l2_len += 4;
253 else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
254 l2_len += 8;
255 /* Don't verify checksum for packets with discontinuous L2 header */
256 if (unlikely(l2_len + sizeof(struct ipv4_hdr) >
257 rte_pktmbuf_data_len(mbuf)))
258 return;
259 l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
260 if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
261 struct ipv4_hdr *iph = l3_hdr;
262
263 /* ihl contains the number of 4-byte words in the header */
264 l3_len = 4 * (iph->version_ihl & 0xf);
265 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
266 return;
267
268 cksum = ~rte_raw_cksum(iph, l3_len);
269 mbuf->ol_flags |= cksum ?
270 PKT_RX_IP_CKSUM_BAD :
271 PKT_RX_IP_CKSUM_GOOD;
272 } else if (l3 == RTE_PTYPE_L3_IPV6) {
273 l3_len = sizeof(struct ipv6_hdr);
274 } else {
275 /* IPv6 extensions are not supported */
276 return;
277 }
278 if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
279 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
280 /* Don't verify checksum for multi-segment packets. */
281 if (mbuf->nb_segs > 1)
282 return;
283 if (l3 == RTE_PTYPE_L3_IPV4)
284 cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
285 else if (l3 == RTE_PTYPE_L3_IPV6)
286 cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
287 mbuf->ol_flags |= cksum ?
288 PKT_RX_L4_CKSUM_BAD :
289 PKT_RX_L4_CKSUM_GOOD;
290 }
291 }
292
293 static uint64_t
294 tap_rx_offload_get_port_capa(void)
295 {
296 /*
297 * No specific port Rx offload capabilities.
298 */
299 return 0;
300 }
301
302 static uint64_t
303 tap_rx_offload_get_queue_capa(void)
304 {
305 return DEV_RX_OFFLOAD_SCATTER |
306 DEV_RX_OFFLOAD_IPV4_CKSUM |
307 DEV_RX_OFFLOAD_UDP_CKSUM |
308 DEV_RX_OFFLOAD_TCP_CKSUM |
309 DEV_RX_OFFLOAD_CRC_STRIP;
310 }
311
312 /* Callback to handle the rx burst of packets to the correct interface and
313 * file descriptor(s) in a multi-queue setup.
314 */
315 static uint16_t
316 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
317 {
318 struct rx_queue *rxq = queue;
319 uint16_t num_rx;
320 unsigned long num_rx_bytes = 0;
321 uint32_t trigger = tap_trigger;
322
323 if (trigger == rxq->trigger_seen)
324 return 0;
325 if (trigger)
326 rxq->trigger_seen = trigger;
327 rte_compiler_barrier();
328 for (num_rx = 0; num_rx < nb_pkts; ) {
329 struct rte_mbuf *mbuf = rxq->pool;
330 struct rte_mbuf *seg = NULL;
331 struct rte_mbuf *new_tail = NULL;
332 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
333 int len;
334
335 len = readv(rxq->fd, *rxq->iovecs,
336 1 +
337 (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
338 rxq->nb_rx_desc : 1));
339 if (len < (int)sizeof(struct tun_pi))
340 break;
341
342 /* Packet couldn't fit in the provided mbuf */
343 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
344 rxq->stats.ierrors++;
345 continue;
346 }
347
348 len -= sizeof(struct tun_pi);
349
350 mbuf->pkt_len = len;
351 mbuf->port = rxq->in_port;
352 while (1) {
353 struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
354
355 if (unlikely(!buf)) {
356 rxq->stats.rx_nombuf++;
357 /* No new buf has been allocated: do nothing */
358 if (!new_tail || !seg)
359 goto end;
360
361 seg->next = NULL;
362 rte_pktmbuf_free(mbuf);
363
364 goto end;
365 }
366 seg = seg ? seg->next : mbuf;
367 if (rxq->pool == mbuf)
368 rxq->pool = buf;
369 if (new_tail)
370 new_tail->next = buf;
371 new_tail = buf;
372 new_tail->next = seg->next;
373
374 /* iovecs[0] is reserved for packet info (pi) */
375 (*rxq->iovecs)[mbuf->nb_segs].iov_len =
376 buf->buf_len - data_off;
377 (*rxq->iovecs)[mbuf->nb_segs].iov_base =
378 (char *)buf->buf_addr + data_off;
379
380 seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
381 seg->data_off = data_off;
382
383 len -= seg->data_len;
384 if (len <= 0)
385 break;
386 mbuf->nb_segs++;
387 /* First segment has headroom, not the others */
388 data_off = 0;
389 }
390 seg->next = NULL;
391 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
392 RTE_PTYPE_ALL_MASK);
393 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
394 tap_verify_csum(mbuf);
395
396 /* account for the receive frame */
397 bufs[num_rx++] = mbuf;
398 num_rx_bytes += mbuf->pkt_len;
399 }
400 end:
401 rxq->stats.ipackets += num_rx;
402 rxq->stats.ibytes += num_rx_bytes;
403
404 return num_rx;
405 }
406
407 static uint64_t
408 tap_tx_offload_get_port_capa(void)
409 {
410 /*
411 * No specific port Tx offload capabilities.
412 */
413 return 0;
414 }
415
416 static uint64_t
417 tap_tx_offload_get_queue_capa(void)
418 {
419 return DEV_TX_OFFLOAD_MULTI_SEGS |
420 DEV_TX_OFFLOAD_IPV4_CKSUM |
421 DEV_TX_OFFLOAD_UDP_CKSUM |
422 DEV_TX_OFFLOAD_TCP_CKSUM |
423 DEV_TX_OFFLOAD_TCP_TSO;
424 }
425
426 /* Finalize l4 checksum calculation */
427 static void
428 tap_tx_l4_cksum(uint16_t *l4_cksum, uint16_t l4_phdr_cksum,
429 uint32_t l4_raw_cksum)
430 {
431 if (l4_cksum) {
432 uint32_t cksum;
433
434 cksum = __rte_raw_cksum_reduce(l4_raw_cksum);
435 cksum += l4_phdr_cksum;
436
437 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
438 cksum = (~cksum) & 0xffff;
439 if (cksum == 0)
440 cksum = 0xffff;
441 *l4_cksum = cksum;
442 }
443 }
444
445 /* Accumaulate L4 raw checksums */
446 static void
447 tap_tx_l4_add_rcksum(char *l4_data, unsigned int l4_len, uint16_t *l4_cksum,
448 uint32_t *l4_raw_cksum)
449 {
450 if (l4_cksum == NULL)
451 return;
452
453 *l4_raw_cksum = __rte_raw_cksum(l4_data, l4_len, *l4_raw_cksum);
454 }
455
456 /* L3 and L4 pseudo headers checksum offloads */
457 static void
458 tap_tx_l3_cksum(char *packet, uint64_t ol_flags, unsigned int l2_len,
459 unsigned int l3_len, unsigned int l4_len, uint16_t **l4_cksum,
460 uint16_t *l4_phdr_cksum, uint32_t *l4_raw_cksum)
461 {
462 void *l3_hdr = packet + l2_len;
463
464 if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
465 struct ipv4_hdr *iph = l3_hdr;
466 uint16_t cksum;
467
468 iph->hdr_checksum = 0;
469 cksum = rte_raw_cksum(iph, l3_len);
470 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
471 }
472 if (ol_flags & PKT_TX_L4_MASK) {
473 void *l4_hdr;
474
475 l4_hdr = packet + l2_len + l3_len;
476 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
477 *l4_cksum = &((struct udp_hdr *)l4_hdr)->dgram_cksum;
478 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
479 *l4_cksum = &((struct tcp_hdr *)l4_hdr)->cksum;
480 else
481 return;
482 **l4_cksum = 0;
483 if (ol_flags & PKT_TX_IPV4)
484 *l4_phdr_cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
485 else
486 *l4_phdr_cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
487 *l4_raw_cksum = __rte_raw_cksum(l4_hdr, l4_len, 0);
488 }
489 }
490
491 static inline void
492 tap_write_mbufs(struct tx_queue *txq, uint16_t num_mbufs,
493 struct rte_mbuf **pmbufs,
494 uint16_t *num_packets, unsigned long *num_tx_bytes)
495 {
496 int i;
497 uint16_t l234_hlen;
498
499 for (i = 0; i < num_mbufs; i++) {
500 struct rte_mbuf *mbuf = pmbufs[i];
501 struct iovec iovecs[mbuf->nb_segs + 2];
502 struct tun_pi pi = { .flags = 0, .proto = 0x00 };
503 struct rte_mbuf *seg = mbuf;
504 char m_copy[mbuf->data_len];
505 int proto;
506 int n;
507 int j;
508 int k; /* current index in iovecs for copying segments */
509 uint16_t seg_len; /* length of first segment */
510 uint16_t nb_segs;
511 uint16_t *l4_cksum; /* l4 checksum (pseudo header + payload) */
512 uint32_t l4_raw_cksum = 0; /* TCP/UDP payload raw checksum */
513 uint16_t l4_phdr_cksum = 0; /* TCP/UDP pseudo header checksum */
514 uint16_t is_cksum = 0; /* in case cksum should be offloaded */
515
516 l4_cksum = NULL;
517 if (txq->type == ETH_TUNTAP_TYPE_TUN) {
518 /*
519 * TUN and TAP are created with IFF_NO_PI disabled.
520 * For TUN PMD this mandatory as fields are used by
521 * Kernel tun.c to determine whether its IP or non IP
522 * packets.
523 *
524 * The logic fetches the first byte of data from mbuf
525 * then compares whether its v4 or v6. If first byte
526 * is 4 or 6, then protocol field is updated.
527 */
528 char *buff_data = rte_pktmbuf_mtod(seg, void *);
529 proto = (*buff_data & 0xf0);
530 pi.proto = (proto == 0x40) ?
531 rte_cpu_to_be_16(ETHER_TYPE_IPv4) :
532 ((proto == 0x60) ?
533 rte_cpu_to_be_16(ETHER_TYPE_IPv6) :
534 0x00);
535 }
536
537 k = 0;
538 iovecs[k].iov_base = &pi;
539 iovecs[k].iov_len = sizeof(pi);
540 k++;
541
542 nb_segs = mbuf->nb_segs;
543 if (txq->csum &&
544 ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
545 (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
546 (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
547 is_cksum = 1;
548
549 /* Support only packets with at least layer 4
550 * header included in the first segment
551 */
552 seg_len = rte_pktmbuf_data_len(mbuf);
553 l234_hlen = mbuf->l2_len + mbuf->l3_len + mbuf->l4_len;
554 if (seg_len < l234_hlen)
555 break;
556
557 /* To change checksums, work on a * copy of l2, l3
558 * headers + l4 pseudo header
559 */
560 rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
561 l234_hlen);
562 tap_tx_l3_cksum(m_copy, mbuf->ol_flags,
563 mbuf->l2_len, mbuf->l3_len, mbuf->l4_len,
564 &l4_cksum, &l4_phdr_cksum,
565 &l4_raw_cksum);
566 iovecs[k].iov_base = m_copy;
567 iovecs[k].iov_len = l234_hlen;
568 k++;
569
570 /* Update next iovecs[] beyond l2, l3, l4 headers */
571 if (seg_len > l234_hlen) {
572 iovecs[k].iov_len = seg_len - l234_hlen;
573 iovecs[k].iov_base =
574 rte_pktmbuf_mtod(seg, char *) +
575 l234_hlen;
576 tap_tx_l4_add_rcksum(iovecs[k].iov_base,
577 iovecs[k].iov_len, l4_cksum,
578 &l4_raw_cksum);
579 k++;
580 nb_segs++;
581 }
582 seg = seg->next;
583 }
584
585 for (j = k; j <= nb_segs; j++) {
586 iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
587 iovecs[j].iov_base = rte_pktmbuf_mtod(seg, void *);
588 if (is_cksum)
589 tap_tx_l4_add_rcksum(iovecs[j].iov_base,
590 iovecs[j].iov_len, l4_cksum,
591 &l4_raw_cksum);
592 seg = seg->next;
593 }
594
595 if (is_cksum)
596 tap_tx_l4_cksum(l4_cksum, l4_phdr_cksum, l4_raw_cksum);
597
598 /* copy the tx frame data */
599 n = writev(txq->fd, iovecs, j);
600 if (n <= 0)
601 break;
602 (*num_packets)++;
603 (*num_tx_bytes) += rte_pktmbuf_pkt_len(mbuf);
604 }
605 }
606
607 /* Callback to handle sending packets from the tap interface
608 */
609 static uint16_t
610 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
611 {
612 struct tx_queue *txq = queue;
613 uint16_t num_tx = 0;
614 uint16_t num_packets = 0;
615 unsigned long num_tx_bytes = 0;
616 uint32_t max_size;
617 int i;
618
619 if (unlikely(nb_pkts == 0))
620 return 0;
621
622 struct rte_mbuf *gso_mbufs[MAX_GSO_MBUFS];
623 max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
624 for (i = 0; i < nb_pkts; i++) {
625 struct rte_mbuf *mbuf_in = bufs[num_tx];
626 struct rte_mbuf **mbuf;
627 uint16_t num_mbufs = 0;
628 uint16_t tso_segsz = 0;
629 int ret;
630 uint16_t hdrs_len;
631 int j;
632 uint64_t tso;
633
634 tso = mbuf_in->ol_flags & PKT_TX_TCP_SEG;
635 if (tso) {
636 struct rte_gso_ctx *gso_ctx = &txq->gso_ctx;
637
638 assert(gso_ctx != NULL);
639
640 /* TCP segmentation implies TCP checksum offload */
641 mbuf_in->ol_flags |= PKT_TX_TCP_CKSUM;
642
643 /* gso size is calculated without ETHER_CRC_LEN */
644 hdrs_len = mbuf_in->l2_len + mbuf_in->l3_len +
645 mbuf_in->l4_len;
646 tso_segsz = mbuf_in->tso_segsz + hdrs_len;
647 if (unlikely(tso_segsz == hdrs_len) ||
648 tso_segsz > *txq->mtu) {
649 txq->stats.errs++;
650 break;
651 }
652 gso_ctx->gso_size = tso_segsz;
653 ret = rte_gso_segment(mbuf_in, /* packet to segment */
654 gso_ctx, /* gso control block */
655 (struct rte_mbuf **)&gso_mbufs, /* out mbufs */
656 RTE_DIM(gso_mbufs)); /* max tso mbufs */
657
658 /* ret contains the number of new created mbufs */
659 if (ret < 0)
660 break;
661
662 mbuf = gso_mbufs;
663 num_mbufs = ret;
664 } else {
665 /* stats.errs will be incremented */
666 if (rte_pktmbuf_pkt_len(mbuf_in) > max_size)
667 break;
668
669 /* ret 0 indicates no new mbufs were created */
670 ret = 0;
671 mbuf = &mbuf_in;
672 num_mbufs = 1;
673 }
674
675 tap_write_mbufs(txq, num_mbufs, mbuf,
676 &num_packets, &num_tx_bytes);
677 num_tx++;
678 /* free original mbuf */
679 rte_pktmbuf_free(mbuf_in);
680 /* free tso mbufs */
681 for (j = 0; j < ret; j++)
682 rte_pktmbuf_free(mbuf[j]);
683 }
684
685 txq->stats.opackets += num_packets;
686 txq->stats.errs += nb_pkts - num_tx;
687 txq->stats.obytes += num_tx_bytes;
688
689 return num_tx;
690 }
691
692 static const char *
693 tap_ioctl_req2str(unsigned long request)
694 {
695 switch (request) {
696 case SIOCSIFFLAGS:
697 return "SIOCSIFFLAGS";
698 case SIOCGIFFLAGS:
699 return "SIOCGIFFLAGS";
700 case SIOCGIFHWADDR:
701 return "SIOCGIFHWADDR";
702 case SIOCSIFHWADDR:
703 return "SIOCSIFHWADDR";
704 case SIOCSIFMTU:
705 return "SIOCSIFMTU";
706 }
707 return "UNKNOWN";
708 }
709
710 static int
711 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
712 struct ifreq *ifr, int set, enum ioctl_mode mode)
713 {
714 short req_flags = ifr->ifr_flags;
715 int remote = pmd->remote_if_index &&
716 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
717
718 if (!pmd->remote_if_index && mode == REMOTE_ONLY)
719 return 0;
720 /*
721 * If there is a remote netdevice, apply ioctl on it, then apply it on
722 * the tap netdevice.
723 */
724 apply:
725 if (remote)
726 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
727 else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
728 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
729 switch (request) {
730 case SIOCSIFFLAGS:
731 /* fetch current flags to leave other flags untouched */
732 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
733 goto error;
734 if (set)
735 ifr->ifr_flags |= req_flags;
736 else
737 ifr->ifr_flags &= ~req_flags;
738 break;
739 case SIOCGIFFLAGS:
740 case SIOCGIFHWADDR:
741 case SIOCSIFHWADDR:
742 case SIOCSIFMTU:
743 break;
744 default:
745 RTE_LOG(WARNING, PMD, "%s: ioctl() called with wrong arg\n",
746 pmd->name);
747 return -EINVAL;
748 }
749 if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
750 goto error;
751 if (remote-- && mode == LOCAL_AND_REMOTE)
752 goto apply;
753 return 0;
754
755 error:
756 TAP_LOG(DEBUG, "%s(%s) failed: %s(%d)", ifr->ifr_name,
757 tap_ioctl_req2str(request), strerror(errno), errno);
758 return -errno;
759 }
760
761 static int
762 tap_link_set_down(struct rte_eth_dev *dev)
763 {
764 struct pmd_internals *pmd = dev->data->dev_private;
765 struct ifreq ifr = { .ifr_flags = IFF_UP };
766
767 dev->data->dev_link.link_status = ETH_LINK_DOWN;
768 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
769 }
770
771 static int
772 tap_link_set_up(struct rte_eth_dev *dev)
773 {
774 struct pmd_internals *pmd = dev->data->dev_private;
775 struct ifreq ifr = { .ifr_flags = IFF_UP };
776
777 dev->data->dev_link.link_status = ETH_LINK_UP;
778 return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
779 }
780
781 static int
782 tap_dev_start(struct rte_eth_dev *dev)
783 {
784 int err, i;
785
786 err = tap_intr_handle_set(dev, 1);
787 if (err)
788 return err;
789
790 err = tap_link_set_up(dev);
791 if (err)
792 return err;
793
794 for (i = 0; i < dev->data->nb_tx_queues; i++)
795 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
796 for (i = 0; i < dev->data->nb_rx_queues; i++)
797 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
798
799 return err;
800 }
801
802 /* This function gets called when the current port gets stopped.
803 */
804 static void
805 tap_dev_stop(struct rte_eth_dev *dev)
806 {
807 int i;
808
809 for (i = 0; i < dev->data->nb_tx_queues; i++)
810 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
811 for (i = 0; i < dev->data->nb_rx_queues; i++)
812 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
813
814 tap_intr_handle_set(dev, 0);
815 tap_link_set_down(dev);
816 }
817
818 static int
819 tap_dev_configure(struct rte_eth_dev *dev)
820 {
821 if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
822 TAP_LOG(ERR,
823 "%s: number of rx queues %d exceeds max num of queues %d",
824 dev->device->name,
825 dev->data->nb_rx_queues,
826 RTE_PMD_TAP_MAX_QUEUES);
827 return -1;
828 }
829 if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
830 TAP_LOG(ERR,
831 "%s: number of tx queues %d exceeds max num of queues %d",
832 dev->device->name,
833 dev->data->nb_tx_queues,
834 RTE_PMD_TAP_MAX_QUEUES);
835 return -1;
836 }
837
838 TAP_LOG(INFO, "%s: %p: TX configured queues number: %u",
839 dev->device->name, (void *)dev, dev->data->nb_tx_queues);
840
841 TAP_LOG(INFO, "%s: %p: RX configured queues number: %u",
842 dev->device->name, (void *)dev, dev->data->nb_rx_queues);
843
844 return 0;
845 }
846
847 static uint32_t
848 tap_dev_speed_capa(void)
849 {
850 uint32_t speed = pmd_link.link_speed;
851 uint32_t capa = 0;
852
853 if (speed >= ETH_SPEED_NUM_10M)
854 capa |= ETH_LINK_SPEED_10M;
855 if (speed >= ETH_SPEED_NUM_100M)
856 capa |= ETH_LINK_SPEED_100M;
857 if (speed >= ETH_SPEED_NUM_1G)
858 capa |= ETH_LINK_SPEED_1G;
859 if (speed >= ETH_SPEED_NUM_5G)
860 capa |= ETH_LINK_SPEED_2_5G;
861 if (speed >= ETH_SPEED_NUM_5G)
862 capa |= ETH_LINK_SPEED_5G;
863 if (speed >= ETH_SPEED_NUM_10G)
864 capa |= ETH_LINK_SPEED_10G;
865 if (speed >= ETH_SPEED_NUM_20G)
866 capa |= ETH_LINK_SPEED_20G;
867 if (speed >= ETH_SPEED_NUM_25G)
868 capa |= ETH_LINK_SPEED_25G;
869 if (speed >= ETH_SPEED_NUM_40G)
870 capa |= ETH_LINK_SPEED_40G;
871 if (speed >= ETH_SPEED_NUM_50G)
872 capa |= ETH_LINK_SPEED_50G;
873 if (speed >= ETH_SPEED_NUM_56G)
874 capa |= ETH_LINK_SPEED_56G;
875 if (speed >= ETH_SPEED_NUM_100G)
876 capa |= ETH_LINK_SPEED_100G;
877
878 return capa;
879 }
880
881 static void
882 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
883 {
884 struct pmd_internals *internals = dev->data->dev_private;
885
886 dev_info->if_index = internals->if_index;
887 dev_info->max_mac_addrs = 1;
888 dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
889 dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
890 dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
891 dev_info->min_rx_bufsize = 0;
892 dev_info->speed_capa = tap_dev_speed_capa();
893 dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
894 dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
895 dev_info->rx_queue_offload_capa;
896 dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
897 dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
898 dev_info->tx_queue_offload_capa;
899 dev_info->hash_key_size = TAP_RSS_HASH_KEY_SIZE;
900 /*
901 * limitation: TAP supports all of IP, UDP and TCP hash
902 * functions together and not in partial combinations
903 */
904 dev_info->flow_type_rss_offloads = ~TAP_RSS_HF_MASK;
905 }
906
907 static int
908 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
909 {
910 unsigned int i, imax;
911 unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
912 unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
913 unsigned long rx_nombuf = 0, ierrors = 0;
914 const struct pmd_internals *pmd = dev->data->dev_private;
915
916 /* rx queue statistics */
917 imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
918 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
919 for (i = 0; i < imax; i++) {
920 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
921 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
922 rx_total += tap_stats->q_ipackets[i];
923 rx_bytes_total += tap_stats->q_ibytes[i];
924 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
925 ierrors += pmd->rxq[i].stats.ierrors;
926 }
927
928 /* tx queue statistics */
929 imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
930 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
931
932 for (i = 0; i < imax; i++) {
933 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
934 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
935 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
936 tx_total += tap_stats->q_opackets[i];
937 tx_err_total += tap_stats->q_errors[i];
938 tx_bytes_total += tap_stats->q_obytes[i];
939 }
940
941 tap_stats->ipackets = rx_total;
942 tap_stats->ibytes = rx_bytes_total;
943 tap_stats->ierrors = ierrors;
944 tap_stats->rx_nombuf = rx_nombuf;
945 tap_stats->opackets = tx_total;
946 tap_stats->oerrors = tx_err_total;
947 tap_stats->obytes = tx_bytes_total;
948 return 0;
949 }
950
951 static void
952 tap_stats_reset(struct rte_eth_dev *dev)
953 {
954 int i;
955 struct pmd_internals *pmd = dev->data->dev_private;
956
957 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
958 pmd->rxq[i].stats.ipackets = 0;
959 pmd->rxq[i].stats.ibytes = 0;
960 pmd->rxq[i].stats.ierrors = 0;
961 pmd->rxq[i].stats.rx_nombuf = 0;
962
963 pmd->txq[i].stats.opackets = 0;
964 pmd->txq[i].stats.errs = 0;
965 pmd->txq[i].stats.obytes = 0;
966 }
967 }
968
969 static void
970 tap_dev_close(struct rte_eth_dev *dev)
971 {
972 int i;
973 struct pmd_internals *internals = dev->data->dev_private;
974
975 tap_link_set_down(dev);
976 tap_flow_flush(dev, NULL);
977 tap_flow_implicit_flush(internals, NULL);
978
979 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
980 if (internals->rxq[i].fd != -1) {
981 close(internals->rxq[i].fd);
982 internals->rxq[i].fd = -1;
983 }
984 if (internals->txq[i].fd != -1) {
985 close(internals->txq[i].fd);
986 internals->txq[i].fd = -1;
987 }
988 }
989
990 if (internals->remote_if_index) {
991 /* Restore initial remote state */
992 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
993 &internals->remote_initial_flags);
994 }
995
996 if (internals->ka_fd != -1) {
997 close(internals->ka_fd);
998 internals->ka_fd = -1;
999 }
1000 /*
1001 * Since TUN device has no more opened file descriptors
1002 * it will be removed from kernel
1003 */
1004 }
1005
1006 static void
1007 tap_rx_queue_release(void *queue)
1008 {
1009 struct rx_queue *rxq = queue;
1010
1011 if (rxq && (rxq->fd > 0)) {
1012 close(rxq->fd);
1013 rxq->fd = -1;
1014 rte_pktmbuf_free(rxq->pool);
1015 rte_free(rxq->iovecs);
1016 rxq->pool = NULL;
1017 rxq->iovecs = NULL;
1018 }
1019 }
1020
1021 static void
1022 tap_tx_queue_release(void *queue)
1023 {
1024 struct tx_queue *txq = queue;
1025
1026 if (txq && (txq->fd > 0)) {
1027 close(txq->fd);
1028 txq->fd = -1;
1029 }
1030 }
1031
1032 static int
1033 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
1034 {
1035 struct rte_eth_link *dev_link = &dev->data->dev_link;
1036 struct pmd_internals *pmd = dev->data->dev_private;
1037 struct ifreq ifr = { .ifr_flags = 0 };
1038
1039 if (pmd->remote_if_index) {
1040 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
1041 if (!(ifr.ifr_flags & IFF_UP) ||
1042 !(ifr.ifr_flags & IFF_RUNNING)) {
1043 dev_link->link_status = ETH_LINK_DOWN;
1044 return 0;
1045 }
1046 }
1047 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
1048 dev_link->link_status =
1049 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
1050 ETH_LINK_UP :
1051 ETH_LINK_DOWN);
1052 return 0;
1053 }
1054
1055 static void
1056 tap_promisc_enable(struct rte_eth_dev *dev)
1057 {
1058 struct pmd_internals *pmd = dev->data->dev_private;
1059 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1060
1061 dev->data->promiscuous = 1;
1062 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1063 if (pmd->remote_if_index && !pmd->flow_isolate)
1064 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
1065 }
1066
1067 static void
1068 tap_promisc_disable(struct rte_eth_dev *dev)
1069 {
1070 struct pmd_internals *pmd = dev->data->dev_private;
1071 struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1072
1073 dev->data->promiscuous = 0;
1074 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1075 if (pmd->remote_if_index && !pmd->flow_isolate)
1076 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
1077 }
1078
1079 static void
1080 tap_allmulti_enable(struct rte_eth_dev *dev)
1081 {
1082 struct pmd_internals *pmd = dev->data->dev_private;
1083 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1084
1085 dev->data->all_multicast = 1;
1086 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1087 if (pmd->remote_if_index && !pmd->flow_isolate)
1088 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
1089 }
1090
1091 static void
1092 tap_allmulti_disable(struct rte_eth_dev *dev)
1093 {
1094 struct pmd_internals *pmd = dev->data->dev_private;
1095 struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1096
1097 dev->data->all_multicast = 0;
1098 tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1099 if (pmd->remote_if_index && !pmd->flow_isolate)
1100 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
1101 }
1102
1103 static int
1104 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1105 {
1106 struct pmd_internals *pmd = dev->data->dev_private;
1107 enum ioctl_mode mode = LOCAL_ONLY;
1108 struct ifreq ifr;
1109 int ret;
1110
1111 if (pmd->type == ETH_TUNTAP_TYPE_TUN) {
1112 TAP_LOG(ERR, "%s: can't MAC address for TUN",
1113 dev->device->name);
1114 return -ENOTSUP;
1115 }
1116
1117 if (is_zero_ether_addr(mac_addr)) {
1118 TAP_LOG(ERR, "%s: can't set an empty MAC address",
1119 dev->device->name);
1120 return -EINVAL;
1121 }
1122 /* Check the actual current MAC address on the tap netdevice */
1123 ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY);
1124 if (ret < 0)
1125 return ret;
1126 if (is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
1127 mac_addr))
1128 return 0;
1129 /* Check the current MAC address on the remote */
1130 ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY);
1131 if (ret < 0)
1132 return ret;
1133 if (!is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
1134 mac_addr))
1135 mode = LOCAL_AND_REMOTE;
1136 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1137 rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
1138 ret = tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode);
1139 if (ret < 0)
1140 return ret;
1141 rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
1142 if (pmd->remote_if_index && !pmd->flow_isolate) {
1143 /* Replace MAC redirection rule after a MAC change */
1144 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC);
1145 if (ret < 0) {
1146 TAP_LOG(ERR,
1147 "%s: Couldn't delete MAC redirection rule",
1148 dev->device->name);
1149 return ret;
1150 }
1151 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC);
1152 if (ret < 0) {
1153 TAP_LOG(ERR,
1154 "%s: Couldn't add MAC redirection rule",
1155 dev->device->name);
1156 return ret;
1157 }
1158 }
1159
1160 return 0;
1161 }
1162
1163 static int
1164 tap_gso_ctx_setup(struct rte_gso_ctx *gso_ctx, struct rte_eth_dev *dev)
1165 {
1166 uint32_t gso_types;
1167 char pool_name[64];
1168
1169 /*
1170 * Create private mbuf pool with TAP_GSO_MBUF_SEG_SIZE bytes
1171 * size per mbuf use this pool for both direct and indirect mbufs
1172 */
1173
1174 struct rte_mempool *mp; /* Mempool for GSO packets */
1175
1176 /* initialize GSO context */
1177 gso_types = DEV_TX_OFFLOAD_TCP_TSO;
1178 snprintf(pool_name, sizeof(pool_name), "mp_%s", dev->device->name);
1179 mp = rte_mempool_lookup((const char *)pool_name);
1180 if (!mp) {
1181 mp = rte_pktmbuf_pool_create(pool_name, TAP_GSO_MBUFS_NUM,
1182 TAP_GSO_MBUF_CACHE_SIZE, 0,
1183 RTE_PKTMBUF_HEADROOM + TAP_GSO_MBUF_SEG_SIZE,
1184 SOCKET_ID_ANY);
1185 if (!mp) {
1186 struct pmd_internals *pmd = dev->data->dev_private;
1187 RTE_LOG(DEBUG, PMD, "%s: failed to create mbuf pool for device %s\n",
1188 pmd->name, dev->device->name);
1189 return -1;
1190 }
1191 }
1192
1193 gso_ctx->direct_pool = mp;
1194 gso_ctx->indirect_pool = mp;
1195 gso_ctx->gso_types = gso_types;
1196 gso_ctx->gso_size = 0; /* gso_size is set in tx_burst() per packet */
1197 gso_ctx->flag = 0;
1198
1199 return 0;
1200 }
1201
1202 static int
1203 tap_setup_queue(struct rte_eth_dev *dev,
1204 struct pmd_internals *internals,
1205 uint16_t qid,
1206 int is_rx)
1207 {
1208 int ret;
1209 int *fd;
1210 int *other_fd;
1211 const char *dir;
1212 struct pmd_internals *pmd = dev->data->dev_private;
1213 struct rx_queue *rx = &internals->rxq[qid];
1214 struct tx_queue *tx = &internals->txq[qid];
1215 struct rte_gso_ctx *gso_ctx;
1216
1217 if (is_rx) {
1218 fd = &rx->fd;
1219 other_fd = &tx->fd;
1220 dir = "rx";
1221 gso_ctx = NULL;
1222 } else {
1223 fd = &tx->fd;
1224 other_fd = &rx->fd;
1225 dir = "tx";
1226 gso_ctx = &tx->gso_ctx;
1227 }
1228 if (*fd != -1) {
1229 /* fd for this queue already exists */
1230 TAP_LOG(DEBUG, "%s: fd %d for %s queue qid %d exists",
1231 pmd->name, *fd, dir, qid);
1232 gso_ctx = NULL;
1233 } else if (*other_fd != -1) {
1234 /* Only other_fd exists. dup it */
1235 *fd = dup(*other_fd);
1236 if (*fd < 0) {
1237 *fd = -1;
1238 TAP_LOG(ERR, "%s: dup() failed.", pmd->name);
1239 return -1;
1240 }
1241 TAP_LOG(DEBUG, "%s: dup fd %d for %s queue qid %d (%d)",
1242 pmd->name, *other_fd, dir, qid, *fd);
1243 } else {
1244 /* Both RX and TX fds do not exist (equal -1). Create fd */
1245 *fd = tun_alloc(pmd, 0);
1246 if (*fd < 0) {
1247 *fd = -1; /* restore original value */
1248 TAP_LOG(ERR, "%s: tun_alloc() failed.", pmd->name);
1249 return -1;
1250 }
1251 TAP_LOG(DEBUG, "%s: add %s queue for qid %d fd %d",
1252 pmd->name, dir, qid, *fd);
1253 }
1254
1255 tx->mtu = &dev->data->mtu;
1256 rx->rxmode = &dev->data->dev_conf.rxmode;
1257 if (gso_ctx) {
1258 ret = tap_gso_ctx_setup(gso_ctx, dev);
1259 if (ret)
1260 return -1;
1261 }
1262
1263 tx->type = pmd->type;
1264
1265 return *fd;
1266 }
1267
1268 static int
1269 tap_rx_queue_setup(struct rte_eth_dev *dev,
1270 uint16_t rx_queue_id,
1271 uint16_t nb_rx_desc,
1272 unsigned int socket_id,
1273 const struct rte_eth_rxconf *rx_conf __rte_unused,
1274 struct rte_mempool *mp)
1275 {
1276 struct pmd_internals *internals = dev->data->dev_private;
1277 struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1278 struct rte_mbuf **tmp = &rxq->pool;
1279 long iov_max = sysconf(_SC_IOV_MAX);
1280 uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1281 struct iovec (*iovecs)[nb_desc + 1];
1282 int data_off = RTE_PKTMBUF_HEADROOM;
1283 int ret = 0;
1284 int fd;
1285 int i;
1286
1287 if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1288 TAP_LOG(WARNING,
1289 "nb_rx_queues %d too small or mempool NULL",
1290 dev->data->nb_rx_queues);
1291 return -1;
1292 }
1293
1294 rxq->mp = mp;
1295 rxq->trigger_seen = 1; /* force initial burst */
1296 rxq->in_port = dev->data->port_id;
1297 rxq->nb_rx_desc = nb_desc;
1298 iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1299 socket_id);
1300 if (!iovecs) {
1301 TAP_LOG(WARNING,
1302 "%s: Couldn't allocate %d RX descriptors",
1303 dev->device->name, nb_desc);
1304 return -ENOMEM;
1305 }
1306 rxq->iovecs = iovecs;
1307
1308 dev->data->rx_queues[rx_queue_id] = rxq;
1309 fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1310 if (fd == -1) {
1311 ret = fd;
1312 goto error;
1313 }
1314
1315 (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1316 (*rxq->iovecs)[0].iov_base = &rxq->pi;
1317
1318 for (i = 1; i <= nb_desc; i++) {
1319 *tmp = rte_pktmbuf_alloc(rxq->mp);
1320 if (!*tmp) {
1321 TAP_LOG(WARNING,
1322 "%s: couldn't allocate memory for queue %d",
1323 dev->device->name, rx_queue_id);
1324 ret = -ENOMEM;
1325 goto error;
1326 }
1327 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1328 (*rxq->iovecs)[i].iov_base =
1329 (char *)(*tmp)->buf_addr + data_off;
1330 data_off = 0;
1331 tmp = &(*tmp)->next;
1332 }
1333
1334 TAP_LOG(DEBUG, " RX TUNTAP device name %s, qid %d on fd %d",
1335 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
1336
1337 return 0;
1338
1339 error:
1340 rte_pktmbuf_free(rxq->pool);
1341 rxq->pool = NULL;
1342 rte_free(rxq->iovecs);
1343 rxq->iovecs = NULL;
1344 return ret;
1345 }
1346
1347 static int
1348 tap_tx_queue_setup(struct rte_eth_dev *dev,
1349 uint16_t tx_queue_id,
1350 uint16_t nb_tx_desc __rte_unused,
1351 unsigned int socket_id __rte_unused,
1352 const struct rte_eth_txconf *tx_conf)
1353 {
1354 struct pmd_internals *internals = dev->data->dev_private;
1355 struct tx_queue *txq;
1356 int ret;
1357 uint64_t offloads;
1358
1359 if (tx_queue_id >= dev->data->nb_tx_queues)
1360 return -1;
1361 dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1362 txq = dev->data->tx_queues[tx_queue_id];
1363
1364 offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1365 txq->csum = !!(offloads &
1366 (DEV_TX_OFFLOAD_IPV4_CKSUM |
1367 DEV_TX_OFFLOAD_UDP_CKSUM |
1368 DEV_TX_OFFLOAD_TCP_CKSUM));
1369
1370 ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1371 if (ret == -1)
1372 return -1;
1373 TAP_LOG(DEBUG,
1374 " TX TUNTAP device name %s, qid %d on fd %d csum %s",
1375 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd,
1376 txq->csum ? "on" : "off");
1377
1378 return 0;
1379 }
1380
1381 static int
1382 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1383 {
1384 struct pmd_internals *pmd = dev->data->dev_private;
1385 struct ifreq ifr = { .ifr_mtu = mtu };
1386 int err = 0;
1387
1388 err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1389 if (!err)
1390 dev->data->mtu = mtu;
1391
1392 return err;
1393 }
1394
1395 static int
1396 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1397 struct ether_addr *mc_addr_set __rte_unused,
1398 uint32_t nb_mc_addr __rte_unused)
1399 {
1400 /*
1401 * Nothing to do actually: the tap has no filtering whatsoever, every
1402 * packet is received.
1403 */
1404 return 0;
1405 }
1406
1407 static int
1408 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1409 {
1410 struct rte_eth_dev *dev = arg;
1411 struct pmd_internals *pmd = dev->data->dev_private;
1412 struct ifinfomsg *info = NLMSG_DATA(nh);
1413
1414 if (nh->nlmsg_type != RTM_NEWLINK ||
1415 (info->ifi_index != pmd->if_index &&
1416 info->ifi_index != pmd->remote_if_index))
1417 return 0;
1418 return tap_link_update(dev, 0);
1419 }
1420
1421 static void
1422 tap_dev_intr_handler(void *cb_arg)
1423 {
1424 struct rte_eth_dev *dev = cb_arg;
1425 struct pmd_internals *pmd = dev->data->dev_private;
1426
1427 tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1428 }
1429
1430 static int
1431 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1432 {
1433 struct pmd_internals *pmd = dev->data->dev_private;
1434
1435 /* In any case, disable interrupt if the conf is no longer there. */
1436 if (!dev->data->dev_conf.intr_conf.lsc) {
1437 if (pmd->intr_handle.fd != -1) {
1438 tap_nl_final(pmd->intr_handle.fd);
1439 rte_intr_callback_unregister(&pmd->intr_handle,
1440 tap_dev_intr_handler, dev);
1441 }
1442 return 0;
1443 }
1444 if (set) {
1445 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1446 if (unlikely(pmd->intr_handle.fd == -1))
1447 return -EBADF;
1448 return rte_intr_callback_register(
1449 &pmd->intr_handle, tap_dev_intr_handler, dev);
1450 }
1451 tap_nl_final(pmd->intr_handle.fd);
1452 return rte_intr_callback_unregister(&pmd->intr_handle,
1453 tap_dev_intr_handler, dev);
1454 }
1455
1456 static int
1457 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1458 {
1459 int err;
1460
1461 err = tap_lsc_intr_handle_set(dev, set);
1462 if (err)
1463 return err;
1464 err = tap_rx_intr_vec_set(dev, set);
1465 if (err && set)
1466 tap_lsc_intr_handle_set(dev, 0);
1467 return err;
1468 }
1469
1470 static const uint32_t*
1471 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1472 {
1473 static const uint32_t ptypes[] = {
1474 RTE_PTYPE_INNER_L2_ETHER,
1475 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1476 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1477 RTE_PTYPE_INNER_L3_IPV4,
1478 RTE_PTYPE_INNER_L3_IPV4_EXT,
1479 RTE_PTYPE_INNER_L3_IPV6,
1480 RTE_PTYPE_INNER_L3_IPV6_EXT,
1481 RTE_PTYPE_INNER_L4_FRAG,
1482 RTE_PTYPE_INNER_L4_UDP,
1483 RTE_PTYPE_INNER_L4_TCP,
1484 RTE_PTYPE_INNER_L4_SCTP,
1485 RTE_PTYPE_L2_ETHER,
1486 RTE_PTYPE_L2_ETHER_VLAN,
1487 RTE_PTYPE_L2_ETHER_QINQ,
1488 RTE_PTYPE_L3_IPV4,
1489 RTE_PTYPE_L3_IPV4_EXT,
1490 RTE_PTYPE_L3_IPV6_EXT,
1491 RTE_PTYPE_L3_IPV6,
1492 RTE_PTYPE_L4_FRAG,
1493 RTE_PTYPE_L4_UDP,
1494 RTE_PTYPE_L4_TCP,
1495 RTE_PTYPE_L4_SCTP,
1496 };
1497
1498 return ptypes;
1499 }
1500
1501 static int
1502 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1503 struct rte_eth_fc_conf *fc_conf)
1504 {
1505 fc_conf->mode = RTE_FC_NONE;
1506 return 0;
1507 }
1508
1509 static int
1510 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1511 struct rte_eth_fc_conf *fc_conf)
1512 {
1513 if (fc_conf->mode != RTE_FC_NONE)
1514 return -ENOTSUP;
1515 return 0;
1516 }
1517
1518 /**
1519 * DPDK callback to update the RSS hash configuration.
1520 *
1521 * @param dev
1522 * Pointer to Ethernet device structure.
1523 * @param[in] rss_conf
1524 * RSS configuration data.
1525 *
1526 * @return
1527 * 0 on success, a negative errno value otherwise and rte_errno is set.
1528 */
1529 static int
1530 tap_rss_hash_update(struct rte_eth_dev *dev,
1531 struct rte_eth_rss_conf *rss_conf)
1532 {
1533 if (rss_conf->rss_hf & TAP_RSS_HF_MASK) {
1534 rte_errno = EINVAL;
1535 return -rte_errno;
1536 }
1537 if (rss_conf->rss_key && rss_conf->rss_key_len) {
1538 /*
1539 * Currently TAP RSS key is hard coded
1540 * and cannot be updated
1541 */
1542 TAP_LOG(ERR,
1543 "port %u RSS key cannot be updated",
1544 dev->data->port_id);
1545 rte_errno = EINVAL;
1546 return -rte_errno;
1547 }
1548 return 0;
1549 }
1550
1551 static int
1552 tap_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1553 {
1554 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1555
1556 return 0;
1557 }
1558
1559 static int
1560 tap_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1561 {
1562 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1563
1564 return 0;
1565 }
1566
1567 static int
1568 tap_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1569 {
1570 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1571
1572 return 0;
1573 }
1574
1575 static int
1576 tap_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1577 {
1578 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1579
1580 return 0;
1581 }
1582 static const struct eth_dev_ops ops = {
1583 .dev_start = tap_dev_start,
1584 .dev_stop = tap_dev_stop,
1585 .dev_close = tap_dev_close,
1586 .dev_configure = tap_dev_configure,
1587 .dev_infos_get = tap_dev_info,
1588 .rx_queue_setup = tap_rx_queue_setup,
1589 .tx_queue_setup = tap_tx_queue_setup,
1590 .rx_queue_start = tap_rx_queue_start,
1591 .tx_queue_start = tap_tx_queue_start,
1592 .rx_queue_stop = tap_rx_queue_stop,
1593 .tx_queue_stop = tap_tx_queue_stop,
1594 .rx_queue_release = tap_rx_queue_release,
1595 .tx_queue_release = tap_tx_queue_release,
1596 .flow_ctrl_get = tap_flow_ctrl_get,
1597 .flow_ctrl_set = tap_flow_ctrl_set,
1598 .link_update = tap_link_update,
1599 .dev_set_link_up = tap_link_set_up,
1600 .dev_set_link_down = tap_link_set_down,
1601 .promiscuous_enable = tap_promisc_enable,
1602 .promiscuous_disable = tap_promisc_disable,
1603 .allmulticast_enable = tap_allmulti_enable,
1604 .allmulticast_disable = tap_allmulti_disable,
1605 .mac_addr_set = tap_mac_set,
1606 .mtu_set = tap_mtu_set,
1607 .set_mc_addr_list = tap_set_mc_addr_list,
1608 .stats_get = tap_stats_get,
1609 .stats_reset = tap_stats_reset,
1610 .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1611 .rss_hash_update = tap_rss_hash_update,
1612 .filter_ctrl = tap_dev_filter_ctrl,
1613 };
1614
1615 static int
1616 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
1617 char *remote_iface, struct ether_addr *mac_addr,
1618 enum rte_tuntap_type type)
1619 {
1620 int numa_node = rte_socket_id();
1621 struct rte_eth_dev *dev;
1622 struct pmd_internals *pmd;
1623 struct rte_eth_dev_data *data;
1624 struct ifreq ifr;
1625 int i;
1626
1627 TAP_LOG(DEBUG, "%s device on numa %u",
1628 tuntap_name, rte_socket_id());
1629
1630 dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1631 if (!dev) {
1632 TAP_LOG(ERR, "%s Unable to allocate device struct",
1633 tuntap_name);
1634 goto error_exit_nodev;
1635 }
1636
1637 pmd = dev->data->dev_private;
1638 pmd->dev = dev;
1639 snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1640 pmd->type = type;
1641
1642 pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1643 if (pmd->ioctl_sock == -1) {
1644 TAP_LOG(ERR,
1645 "%s Unable to get a socket for management: %s",
1646 tuntap_name, strerror(errno));
1647 goto error_exit;
1648 }
1649
1650 /* Setup some default values */
1651 data = dev->data;
1652 data->dev_private = pmd;
1653 data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1654 data->numa_node = numa_node;
1655
1656 data->dev_link = pmd_link;
1657 data->mac_addrs = &pmd->eth_addr;
1658 /* Set the number of RX and TX queues */
1659 data->nb_rx_queues = 0;
1660 data->nb_tx_queues = 0;
1661
1662 dev->dev_ops = &ops;
1663 dev->rx_pkt_burst = pmd_rx_burst;
1664 dev->tx_pkt_burst = pmd_tx_burst;
1665
1666 pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1667 pmd->intr_handle.fd = -1;
1668 dev->intr_handle = &pmd->intr_handle;
1669
1670 /* Presetup the fds to -1 as being not valid */
1671 pmd->ka_fd = -1;
1672 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1673 pmd->rxq[i].fd = -1;
1674 pmd->txq[i].fd = -1;
1675 }
1676
1677 if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1678 if (is_zero_ether_addr(mac_addr))
1679 eth_random_addr((uint8_t *)&pmd->eth_addr);
1680 else
1681 rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1682 }
1683
1684 /*
1685 * Allocate a TUN device keep-alive file descriptor that will only be
1686 * closed when the TUN device itself is closed or removed.
1687 * This keep-alive file descriptor will guarantee that the TUN device
1688 * exists even when all of its queues are closed
1689 */
1690 pmd->ka_fd = tun_alloc(pmd, 1);
1691 if (pmd->ka_fd == -1) {
1692 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
1693 goto error_exit;
1694 }
1695
1696 ifr.ifr_mtu = dev->data->mtu;
1697 if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1698 goto error_exit;
1699
1700 if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1701 memset(&ifr, 0, sizeof(struct ifreq));
1702 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1703 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
1704 ETHER_ADDR_LEN);
1705 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1706 goto error_exit;
1707 }
1708
1709 /*
1710 * Set up everything related to rte_flow:
1711 * - netlink socket
1712 * - tap / remote if_index
1713 * - mandatory QDISCs
1714 * - rte_flow actual/implicit lists
1715 * - implicit rules
1716 */
1717 pmd->nlsk_fd = tap_nl_init(0);
1718 if (pmd->nlsk_fd == -1) {
1719 TAP_LOG(WARNING, "%s: failed to create netlink socket.",
1720 pmd->name);
1721 goto disable_rte_flow;
1722 }
1723 pmd->if_index = if_nametoindex(pmd->name);
1724 if (!pmd->if_index) {
1725 TAP_LOG(ERR, "%s: failed to get if_index.", pmd->name);
1726 goto disable_rte_flow;
1727 }
1728 if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1729 TAP_LOG(ERR, "%s: failed to create multiq qdisc.",
1730 pmd->name);
1731 goto disable_rte_flow;
1732 }
1733 if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1734 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1735 pmd->name);
1736 goto disable_rte_flow;
1737 }
1738 LIST_INIT(&pmd->flows);
1739
1740 if (strlen(remote_iface)) {
1741 pmd->remote_if_index = if_nametoindex(remote_iface);
1742 if (!pmd->remote_if_index) {
1743 TAP_LOG(ERR, "%s: failed to get %s if_index.",
1744 pmd->name, remote_iface);
1745 goto error_remote;
1746 }
1747 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1748 "%s", remote_iface);
1749
1750 /* Save state of remote device */
1751 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1752
1753 /* Replicate remote MAC address */
1754 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1755 TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1756 pmd->name, pmd->remote_iface);
1757 goto error_remote;
1758 }
1759 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1760 ETHER_ADDR_LEN);
1761 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1762 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1763 TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1764 pmd->name, remote_iface);
1765 goto error_remote;
1766 }
1767
1768 /*
1769 * Flush usually returns negative value because it tries to
1770 * delete every QDISC (and on a running device, one QDISC at
1771 * least is needed). Ignore negative return value.
1772 */
1773 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1774 if (qdisc_create_ingress(pmd->nlsk_fd,
1775 pmd->remote_if_index) < 0) {
1776 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1777 pmd->remote_iface);
1778 goto error_remote;
1779 }
1780 LIST_INIT(&pmd->implicit_flows);
1781 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1782 tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1783 tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1784 tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1785 TAP_LOG(ERR,
1786 "%s: failed to create implicit rules.",
1787 pmd->name);
1788 goto error_remote;
1789 }
1790 }
1791
1792 rte_eth_dev_probing_finish(dev);
1793 return 0;
1794
1795 disable_rte_flow:
1796 TAP_LOG(ERR, " Disabling rte flow support: %s(%d)",
1797 strerror(errno), errno);
1798 if (strlen(remote_iface)) {
1799 TAP_LOG(ERR, "Remote feature requires flow support.");
1800 goto error_exit;
1801 }
1802 return 0;
1803
1804 error_remote:
1805 TAP_LOG(ERR, " Can't set up remote feature: %s(%d)",
1806 strerror(errno), errno);
1807 tap_flow_implicit_flush(pmd, NULL);
1808
1809 error_exit:
1810 if (pmd->ioctl_sock > 0)
1811 close(pmd->ioctl_sock);
1812 rte_eth_dev_release_port(dev);
1813
1814 error_exit_nodev:
1815 TAP_LOG(ERR, "%s Unable to initialize %s",
1816 tuntap_name, rte_vdev_device_name(vdev));
1817
1818 return -EINVAL;
1819 }
1820
1821 static int
1822 set_interface_name(const char *key __rte_unused,
1823 const char *value,
1824 void *extra_args)
1825 {
1826 char *name = (char *)extra_args;
1827
1828 if (value)
1829 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN - 1);
1830 else
1831 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1832 DEFAULT_TAP_NAME, (tap_unit - 1));
1833
1834 return 0;
1835 }
1836
1837 static int
1838 set_remote_iface(const char *key __rte_unused,
1839 const char *value,
1840 void *extra_args)
1841 {
1842 char *name = (char *)extra_args;
1843
1844 if (value)
1845 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
1846
1847 return 0;
1848 }
1849
1850 static int parse_user_mac(struct ether_addr *user_mac,
1851 const char *value)
1852 {
1853 unsigned int index = 0;
1854 char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
1855
1856 if (user_mac == NULL || value == NULL)
1857 return 0;
1858
1859 strlcpy(mac_temp, value, sizeof(mac_temp));
1860 mac_byte = strtok(mac_temp, ":");
1861
1862 while ((mac_byte != NULL) &&
1863 (strlen(mac_byte) <= 2) &&
1864 (strlen(mac_byte) == strspn(mac_byte,
1865 ETH_TAP_CMP_MAC_FMT))) {
1866 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
1867 mac_byte = strtok(NULL, ":");
1868 }
1869
1870 return index;
1871 }
1872
1873 static int
1874 set_mac_type(const char *key __rte_unused,
1875 const char *value,
1876 void *extra_args)
1877 {
1878 struct ether_addr *user_mac = extra_args;
1879
1880 if (!value)
1881 return 0;
1882
1883 if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
1884 static int iface_idx;
1885
1886 /* fixed mac = 00:64:74:61:70:<iface_idx> */
1887 memcpy((char *)user_mac->addr_bytes, "\0dtap", ETHER_ADDR_LEN);
1888 user_mac->addr_bytes[ETHER_ADDR_LEN - 1] = iface_idx++ + '0';
1889 goto success;
1890 }
1891
1892 if (parse_user_mac(user_mac, value) != 6)
1893 goto error;
1894 success:
1895 TAP_LOG(DEBUG, "TAP user MAC param (%s)", value);
1896 return 0;
1897
1898 error:
1899 TAP_LOG(ERR, "TAP user MAC (%s) is not in format (%s|%s)",
1900 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
1901 return -1;
1902 }
1903
1904 /*
1905 * Open a TUN interface device. TUN PMD
1906 * 1) sets tap_type as false
1907 * 2) intakes iface as argument.
1908 * 3) as interface is virtual set speed to 10G
1909 */
1910 static int
1911 rte_pmd_tun_probe(struct rte_vdev_device *dev)
1912 {
1913 const char *name, *params;
1914 int ret;
1915 struct rte_kvargs *kvlist = NULL;
1916 char tun_name[RTE_ETH_NAME_MAX_LEN];
1917 char remote_iface[RTE_ETH_NAME_MAX_LEN];
1918 struct rte_eth_dev *eth_dev;
1919
1920 strcpy(tuntap_name, "TUN");
1921
1922 name = rte_vdev_device_name(dev);
1923 params = rte_vdev_device_args(dev);
1924 memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1925
1926 if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
1927 strlen(params) == 0) {
1928 eth_dev = rte_eth_dev_attach_secondary(name);
1929 if (!eth_dev) {
1930 TAP_LOG(ERR, "Failed to probe %s", name);
1931 return -1;
1932 }
1933 eth_dev->dev_ops = &ops;
1934 eth_dev->device = &dev->device;
1935 rte_eth_dev_probing_finish(eth_dev);
1936 return 0;
1937 }
1938
1939 snprintf(tun_name, sizeof(tun_name), "%s%u",
1940 DEFAULT_TUN_NAME, tun_unit++);
1941
1942 if (params && (params[0] != '\0')) {
1943 TAP_LOG(DEBUG, "parameters (%s)", params);
1944
1945 kvlist = rte_kvargs_parse(params, valid_arguments);
1946 if (kvlist) {
1947 if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1948 ret = rte_kvargs_process(kvlist,
1949 ETH_TAP_IFACE_ARG,
1950 &set_interface_name,
1951 tun_name);
1952
1953 if (ret == -1)
1954 goto leave;
1955 }
1956 }
1957 }
1958 pmd_link.link_speed = ETH_SPEED_NUM_10G;
1959
1960 TAP_LOG(NOTICE, "Initializing pmd_tun for %s as %s",
1961 name, tun_name);
1962
1963 ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0,
1964 ETH_TUNTAP_TYPE_TUN);
1965
1966 leave:
1967 if (ret == -1) {
1968 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
1969 name, tun_name);
1970 tun_unit--; /* Restore the unit number */
1971 }
1972 rte_kvargs_free(kvlist);
1973
1974 return ret;
1975 }
1976
1977 /* Open a TAP interface device.
1978 */
1979 static int
1980 rte_pmd_tap_probe(struct rte_vdev_device *dev)
1981 {
1982 const char *name, *params;
1983 int ret;
1984 struct rte_kvargs *kvlist = NULL;
1985 int speed;
1986 char tap_name[RTE_ETH_NAME_MAX_LEN];
1987 char remote_iface[RTE_ETH_NAME_MAX_LEN];
1988 struct ether_addr user_mac = { .addr_bytes = {0} };
1989 struct rte_eth_dev *eth_dev;
1990
1991 strcpy(tuntap_name, "TAP");
1992
1993 name = rte_vdev_device_name(dev);
1994 params = rte_vdev_device_args(dev);
1995
1996 if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
1997 strlen(params) == 0) {
1998 eth_dev = rte_eth_dev_attach_secondary(name);
1999 if (!eth_dev) {
2000 TAP_LOG(ERR, "Failed to probe %s", name);
2001 return -1;
2002 }
2003 /* TODO: request info from primary to set up Rx and Tx */
2004 eth_dev->dev_ops = &ops;
2005 eth_dev->device = &dev->device;
2006 rte_eth_dev_probing_finish(eth_dev);
2007 return 0;
2008 }
2009
2010 speed = ETH_SPEED_NUM_10G;
2011 snprintf(tap_name, sizeof(tap_name), "%s%u",
2012 DEFAULT_TAP_NAME, tap_unit++);
2013 memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2014
2015 if (params && (params[0] != '\0')) {
2016 TAP_LOG(DEBUG, "parameters (%s)", params);
2017
2018 kvlist = rte_kvargs_parse(params, valid_arguments);
2019 if (kvlist) {
2020 if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2021 ret = rte_kvargs_process(kvlist,
2022 ETH_TAP_IFACE_ARG,
2023 &set_interface_name,
2024 tap_name);
2025 if (ret == -1)
2026 goto leave;
2027 }
2028
2029 if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
2030 ret = rte_kvargs_process(kvlist,
2031 ETH_TAP_REMOTE_ARG,
2032 &set_remote_iface,
2033 remote_iface);
2034 if (ret == -1)
2035 goto leave;
2036 }
2037
2038 if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
2039 ret = rte_kvargs_process(kvlist,
2040 ETH_TAP_MAC_ARG,
2041 &set_mac_type,
2042 &user_mac);
2043 if (ret == -1)
2044 goto leave;
2045 }
2046 }
2047 }
2048 pmd_link.link_speed = speed;
2049
2050 TAP_LOG(NOTICE, "Initializing pmd_tap for %s as %s",
2051 name, tap_name);
2052
2053 ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac,
2054 ETH_TUNTAP_TYPE_TAP);
2055
2056 leave:
2057 if (ret == -1) {
2058 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2059 name, tap_name);
2060 tap_unit--; /* Restore the unit number */
2061 }
2062 rte_kvargs_free(kvlist);
2063
2064 return ret;
2065 }
2066
2067 /* detach a TUNTAP device.
2068 */
2069 static int
2070 rte_pmd_tap_remove(struct rte_vdev_device *dev)
2071 {
2072 struct rte_eth_dev *eth_dev = NULL;
2073 struct pmd_internals *internals;
2074 int i;
2075
2076 /* find the ethdev entry */
2077 eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
2078 if (!eth_dev)
2079 return 0;
2080
2081 internals = eth_dev->data->dev_private;
2082
2083 TAP_LOG(DEBUG, "Closing %s Ethernet device on numa %u",
2084 (internals->type == ETH_TUNTAP_TYPE_TAP) ? "TAP" : "TUN",
2085 rte_socket_id());
2086
2087 if (internals->nlsk_fd) {
2088 tap_flow_flush(eth_dev, NULL);
2089 tap_flow_implicit_flush(internals, NULL);
2090 tap_nl_final(internals->nlsk_fd);
2091 }
2092 for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
2093 if (internals->rxq[i].fd != -1) {
2094 close(internals->rxq[i].fd);
2095 internals->rxq[i].fd = -1;
2096 }
2097 if (internals->txq[i].fd != -1) {
2098 close(internals->txq[i].fd);
2099 internals->txq[i].fd = -1;
2100 }
2101 }
2102
2103 close(internals->ioctl_sock);
2104 rte_free(eth_dev->data->dev_private);
2105 rte_eth_dev_release_port(eth_dev);
2106
2107 if (internals->ka_fd != -1) {
2108 close(internals->ka_fd);
2109 internals->ka_fd = -1;
2110 }
2111 return 0;
2112 }
2113
2114 static struct rte_vdev_driver pmd_tun_drv = {
2115 .probe = rte_pmd_tun_probe,
2116 .remove = rte_pmd_tap_remove,
2117 };
2118
2119 static struct rte_vdev_driver pmd_tap_drv = {
2120 .probe = rte_pmd_tap_probe,
2121 .remove = rte_pmd_tap_remove,
2122 };
2123
2124 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
2125 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
2126 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
2127 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
2128 ETH_TAP_IFACE_ARG "=<string> ");
2129 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
2130 ETH_TAP_IFACE_ARG "=<string> "
2131 ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
2132 ETH_TAP_REMOTE_ARG "=<string>");
2133 int tap_logtype;
2134
2135 RTE_INIT(tap_init_log)
2136 {
2137 tap_logtype = rte_log_register("pmd.net.tap");
2138 if (tap_logtype >= 0)
2139 rte_log_set_level(tap_logtype, RTE_LOG_NOTICE);
2140 }
Ceph