xref: /dpdk/drivers/net/af_packet/rte_eth_af_packet.c (revision 7917b0d38e92e8b9ec5a870415b791420e10f11a)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2014 John W. Linville <linville@tuxdriver.com>
3  * Originally based upon librte_pmd_pcap code:
4  * Copyright(c) 2010-2015 Intel Corporation.
5  * Copyright(c) 2014 6WIND S.A.
6  * All rights reserved.
7  */
8 
9 #include <rte_common.h>
10 #include <rte_string_fns.h>
11 #include <rte_mbuf.h>
12 #include <ethdev_driver.h>
13 #include <ethdev_vdev.h>
14 #include <rte_malloc.h>
15 #include <rte_kvargs.h>
16 #include <bus_vdev_driver.h>
17 
18 #include <errno.h>
19 #include <linux/if_ether.h>
20 #include <linux/if_packet.h>
21 #include <arpa/inet.h>
22 #include <net/if.h>
23 #include <net/if_arp.h>
24 #include <sys/types.h>
25 #include <sys/socket.h>
26 #include <sys/ioctl.h>
27 #include <stdlib.h>
28 #include <string.h>
29 #include <sys/mman.h>
30 #include <unistd.h>
31 #include <poll.h>
32 
33 #define ETH_AF_PACKET_IFACE_ARG		"iface"
34 #define ETH_AF_PACKET_NUM_Q_ARG		"qpairs"
35 #define ETH_AF_PACKET_BLOCKSIZE_ARG	"blocksz"
36 #define ETH_AF_PACKET_FRAMESIZE_ARG	"framesz"
37 #define ETH_AF_PACKET_FRAMECOUNT_ARG	"framecnt"
38 #define ETH_AF_PACKET_QDISC_BYPASS_ARG	"qdisc_bypass"
39 
40 #define DFLT_FRAME_SIZE		(1 << 11)
41 #define DFLT_FRAME_COUNT	(1 << 9)
42 
43 static uint64_t timestamp_dynflag;
44 static int timestamp_dynfield_offset = -1;
45 
46 struct __rte_cache_aligned pkt_rx_queue {
47 	int sockfd;
48 
49 	struct iovec *rd;
50 	uint8_t *map;
51 	unsigned int framecount;
52 	unsigned int framenum;
53 
54 	struct rte_mempool *mb_pool;
55 	uint16_t in_port;
56 	uint8_t vlan_strip;
57 	uint8_t timestamp_offloading;
58 
59 	volatile unsigned long rx_pkts;
60 	volatile unsigned long rx_bytes;
61 };
62 
63 struct __rte_cache_aligned pkt_tx_queue {
64 	int sockfd;
65 	unsigned int frame_data_size;
66 
67 	struct iovec *rd;
68 	uint8_t *map;
69 	unsigned int framecount;
70 	unsigned int framenum;
71 
72 	volatile unsigned long tx_pkts;
73 	volatile unsigned long err_pkts;
74 	volatile unsigned long tx_bytes;
75 };
76 
77 struct pmd_internals {
78 	unsigned nb_queues;
79 
80 	int if_index;
81 	char *if_name;
82 	struct rte_ether_addr eth_addr;
83 
84 	struct tpacket_req req;
85 
86 	struct pkt_rx_queue *rx_queue;
87 	struct pkt_tx_queue *tx_queue;
88 	uint8_t vlan_strip;
89 	uint8_t timestamp_offloading;
90 };
91 
92 static const char *valid_arguments[] = {
93 	ETH_AF_PACKET_IFACE_ARG,
94 	ETH_AF_PACKET_NUM_Q_ARG,
95 	ETH_AF_PACKET_BLOCKSIZE_ARG,
96 	ETH_AF_PACKET_FRAMESIZE_ARG,
97 	ETH_AF_PACKET_FRAMECOUNT_ARG,
98 	ETH_AF_PACKET_QDISC_BYPASS_ARG,
99 	NULL
100 };
101 
102 static struct rte_eth_link pmd_link = {
103 	.link_speed = RTE_ETH_SPEED_NUM_10G,
104 	.link_duplex = RTE_ETH_LINK_FULL_DUPLEX,
105 	.link_status = RTE_ETH_LINK_DOWN,
106 	.link_autoneg = RTE_ETH_LINK_FIXED,
107 };
108 
109 RTE_LOG_REGISTER_DEFAULT(af_packet_logtype, NOTICE);
110 #define RTE_LOGTYPE_AFPACKET af_packet_logtype
111 
112 #define PMD_LOG(level, ...) \
113 	RTE_LOG_LINE_PREFIX(level, AFPACKET, "%s(): ", __func__, __VA_ARGS__)
114 
115 #define PMD_LOG_ERRNO(level, fmt, ...) \
116 	RTE_LOG_LINE(level, AFPACKET, "%s(): " fmt ":%s", __func__, \
117 		## __VA_ARGS__, strerror(errno))
118 
119 static uint16_t
120 eth_af_packet_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
121 {
122 	unsigned i;
123 	struct tpacket2_hdr *ppd;
124 	struct rte_mbuf *mbuf;
125 	uint8_t *pbuf;
126 	struct pkt_rx_queue *pkt_q = queue;
127 	uint16_t num_rx = 0;
128 	unsigned long num_rx_bytes = 0;
129 	unsigned int framecount, framenum;
130 
131 	if (unlikely(nb_pkts == 0))
132 		return 0;
133 
134 	/*
135 	 * Reads the given number of packets from the AF_PACKET socket one by
136 	 * one and copies the packet data into a newly allocated mbuf.
137 	 */
138 	framecount = pkt_q->framecount;
139 	framenum = pkt_q->framenum;
140 	for (i = 0; i < nb_pkts; i++) {
141 		/* point at the next incoming frame */
142 		ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
143 		if ((ppd->tp_status & TP_STATUS_USER) == 0)
144 			break;
145 
146 		/* allocate the next mbuf */
147 		mbuf = rte_pktmbuf_alloc(pkt_q->mb_pool);
148 		if (unlikely(mbuf == NULL))
149 			break;
150 
151 		/* packet will fit in the mbuf, go ahead and receive it */
152 		rte_pktmbuf_pkt_len(mbuf) = rte_pktmbuf_data_len(mbuf) = ppd->tp_snaplen;
153 		pbuf = (uint8_t *) ppd + ppd->tp_mac;
154 		memcpy(rte_pktmbuf_mtod(mbuf, void *), pbuf, rte_pktmbuf_data_len(mbuf));
155 
156 		/* check for vlan info */
157 		if (ppd->tp_status & TP_STATUS_VLAN_VALID) {
158 			mbuf->vlan_tci = ppd->tp_vlan_tci;
159 			mbuf->ol_flags |= (RTE_MBUF_F_RX_VLAN | RTE_MBUF_F_RX_VLAN_STRIPPED);
160 
161 			if (!pkt_q->vlan_strip && rte_vlan_insert(&mbuf))
162 				PMD_LOG(ERR, "Failed to reinsert VLAN tag");
163 		}
164 
165 		/* add kernel provided timestamp when offloading is enabled */
166 		if (pkt_q->timestamp_offloading) {
167 			/* since TPACKET_V2 timestamps are provided in nanoseconds resolution */
168 			*RTE_MBUF_DYNFIELD(mbuf, timestamp_dynfield_offset,
169 				rte_mbuf_timestamp_t *) =
170 					(uint64_t)ppd->tp_sec * 1000000000 + ppd->tp_nsec;
171 
172 			mbuf->ol_flags |= timestamp_dynflag;
173 		}
174 
175 		/* release incoming frame and advance ring buffer */
176 		ppd->tp_status = TP_STATUS_KERNEL;
177 		if (++framenum >= framecount)
178 			framenum = 0;
179 		mbuf->port = pkt_q->in_port;
180 
181 		/* account for the receive frame */
182 		bufs[i] = mbuf;
183 		num_rx++;
184 		num_rx_bytes += mbuf->pkt_len;
185 	}
186 	pkt_q->framenum = framenum;
187 	pkt_q->rx_pkts += num_rx;
188 	pkt_q->rx_bytes += num_rx_bytes;
189 	return num_rx;
190 }
191 
192 /*
193  * Check if there is an available frame in the ring
194  */
195 static inline bool
196 tx_ring_status_available(uint32_t tp_status)
197 {
198 	/*
199 	 * We eliminate the timestamp status from the packet status.
200 	 * This should only matter if timestamping is enabled on the socket,
201 	 * but there is a bug in the kernel which is fixed in newer releases.
202 	 *
203 	 * See the following kernel commit for reference:
204 	 *     commit 171c3b151118a2fe0fc1e2a9d1b5a1570cfe82d2
205 	 *     net: packetmmap: fix only tx timestamp on request
206 	 */
207 	tp_status &= ~(TP_STATUS_TS_SOFTWARE | TP_STATUS_TS_RAW_HARDWARE);
208 
209 	return tp_status == TP_STATUS_AVAILABLE;
210 }
211 
212 /*
213  * Callback to handle sending packets through a real NIC.
214  */
215 static uint16_t
216 eth_af_packet_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
217 {
218 	struct tpacket2_hdr *ppd;
219 	struct rte_mbuf *mbuf;
220 	uint8_t *pbuf;
221 	unsigned int framecount, framenum;
222 	struct pollfd pfd;
223 	struct pkt_tx_queue *pkt_q = queue;
224 	uint16_t num_tx = 0;
225 	unsigned long num_tx_bytes = 0;
226 	int i;
227 
228 	if (unlikely(nb_pkts == 0))
229 		return 0;
230 
231 	memset(&pfd, 0, sizeof(pfd));
232 	pfd.fd = pkt_q->sockfd;
233 	pfd.events = POLLOUT;
234 	pfd.revents = 0;
235 
236 	framecount = pkt_q->framecount;
237 	framenum = pkt_q->framenum;
238 	ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
239 	for (i = 0; i < nb_pkts; i++) {
240 		mbuf = *bufs++;
241 
242 		/* drop oversized packets */
243 		if (mbuf->pkt_len > pkt_q->frame_data_size) {
244 			rte_pktmbuf_free(mbuf);
245 			continue;
246 		}
247 
248 		/* insert vlan info if necessary */
249 		if (mbuf->ol_flags & RTE_MBUF_F_TX_VLAN) {
250 			if (rte_vlan_insert(&mbuf)) {
251 				rte_pktmbuf_free(mbuf);
252 				continue;
253 			}
254 		}
255 
256 		/* point at the next incoming frame */
257 		if (!tx_ring_status_available(ppd->tp_status)) {
258 			if (poll(&pfd, 1, -1) < 0)
259 				break;
260 
261 			/* poll() can return POLLERR if the interface is down */
262 			if (pfd.revents & POLLERR)
263 				break;
264 		}
265 
266 		/*
267 		 * poll() will almost always return POLLOUT, even if there
268 		 * are no extra buffers available
269 		 *
270 		 * This happens, because packet_poll() calls datagram_poll()
271 		 * which checks the space left in the socket buffer and,
272 		 * in the case of packet_mmap, the default socket buffer length
273 		 * doesn't match the requested size for the tx_ring.
274 		 * As such, there is almost always space left in socket buffer,
275 		 * which doesn't seem to be correlated to the requested size
276 		 * for the tx_ring in packet_mmap.
277 		 *
278 		 * This results in poll() returning POLLOUT.
279 		 */
280 		if (!tx_ring_status_available(ppd->tp_status))
281 			break;
282 
283 		/* copy the tx frame data */
284 		pbuf = (uint8_t *) ppd + TPACKET2_HDRLEN -
285 			sizeof(struct sockaddr_ll);
286 
287 		struct rte_mbuf *tmp_mbuf = mbuf;
288 		while (tmp_mbuf) {
289 			uint16_t data_len = rte_pktmbuf_data_len(tmp_mbuf);
290 			memcpy(pbuf, rte_pktmbuf_mtod(tmp_mbuf, void*), data_len);
291 			pbuf += data_len;
292 			tmp_mbuf = tmp_mbuf->next;
293 		}
294 
295 		ppd->tp_len = mbuf->pkt_len;
296 		ppd->tp_snaplen = mbuf->pkt_len;
297 
298 		/* release incoming frame and advance ring buffer */
299 		ppd->tp_status = TP_STATUS_SEND_REQUEST;
300 		if (++framenum >= framecount)
301 			framenum = 0;
302 		ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
303 
304 		num_tx++;
305 		num_tx_bytes += mbuf->pkt_len;
306 		rte_pktmbuf_free(mbuf);
307 	}
308 
309 	/* kick-off transmits */
310 	if (sendto(pkt_q->sockfd, NULL, 0, MSG_DONTWAIT, NULL, 0) == -1 &&
311 			errno != ENOBUFS && errno != EAGAIN) {
312 		/*
313 		 * In case of a ENOBUFS/EAGAIN error all of the enqueued
314 		 * packets will be considered successful even though only some
315 		 * are sent.
316 		 */
317 
318 		num_tx = 0;
319 		num_tx_bytes = 0;
320 	}
321 
322 	pkt_q->framenum = framenum;
323 	pkt_q->tx_pkts += num_tx;
324 	pkt_q->err_pkts += i - num_tx;
325 	pkt_q->tx_bytes += num_tx_bytes;
326 	return i;
327 }
328 
329 static int
330 eth_dev_start(struct rte_eth_dev *dev)
331 {
332 	struct pmd_internals *internals = dev->data->dev_private;
333 	uint16_t i;
334 
335 	if (internals->timestamp_offloading) {
336 		/* Register mbuf field and flag for Rx timestamp */
337 		int rc = rte_mbuf_dyn_rx_timestamp_register(&timestamp_dynfield_offset,
338 				&timestamp_dynflag);
339 		if (rc) {
340 			PMD_LOG(ERR, "Cannot register mbuf field/flag for timestamp");
341 			return rc;
342 		}
343 	}
344 
345 	dev->data->dev_link.link_status = RTE_ETH_LINK_UP;
346 	for (i = 0; i < internals->nb_queues; i++) {
347 		dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
348 		dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
349 	}
350 	return 0;
351 }
352 
353 /*
354  * This function gets called when the current port gets stopped.
355  */
356 static int
357 eth_dev_stop(struct rte_eth_dev *dev)
358 {
359 	unsigned i;
360 	int sockfd;
361 	struct pmd_internals *internals = dev->data->dev_private;
362 
363 	for (i = 0; i < internals->nb_queues; i++) {
364 		sockfd = internals->rx_queue[i].sockfd;
365 		if (sockfd != -1)
366 			close(sockfd);
367 
368 		/* Prevent use after free in case tx fd == rx fd */
369 		if (sockfd != internals->tx_queue[i].sockfd) {
370 			sockfd = internals->tx_queue[i].sockfd;
371 			if (sockfd != -1)
372 				close(sockfd);
373 		}
374 
375 		internals->rx_queue[i].sockfd = -1;
376 		internals->tx_queue[i].sockfd = -1;
377 		dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
378 		dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
379 	}
380 
381 	dev->data->dev_link.link_status = RTE_ETH_LINK_DOWN;
382 	return 0;
383 }
384 
385 static int
386 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
387 {
388 	struct rte_eth_conf *dev_conf = &dev->data->dev_conf;
389 	const struct rte_eth_rxmode *rxmode = &dev_conf->rxmode;
390 	struct pmd_internals *internals = dev->data->dev_private;
391 
392 	internals->vlan_strip = !!(rxmode->offloads & RTE_ETH_RX_OFFLOAD_VLAN_STRIP);
393 	internals->timestamp_offloading = !!(rxmode->offloads & RTE_ETH_RX_OFFLOAD_TIMESTAMP);
394 	return 0;
395 }
396 
397 static int
398 eth_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
399 {
400 	struct pmd_internals *internals = dev->data->dev_private;
401 
402 	dev_info->if_index = internals->if_index;
403 	dev_info->max_mac_addrs = 1;
404 	dev_info->max_rx_pktlen = RTE_ETHER_MAX_LEN;
405 	dev_info->max_rx_queues = (uint16_t)internals->nb_queues;
406 	dev_info->max_tx_queues = (uint16_t)internals->nb_queues;
407 	dev_info->min_rx_bufsize = 0;
408 	dev_info->tx_offload_capa = RTE_ETH_TX_OFFLOAD_MULTI_SEGS |
409 		RTE_ETH_TX_OFFLOAD_VLAN_INSERT;
410 	dev_info->rx_offload_capa = RTE_ETH_RX_OFFLOAD_VLAN_STRIP |
411 		RTE_ETH_RX_OFFLOAD_TIMESTAMP;
412 
413 	return 0;
414 }
415 
416 static int
417 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *igb_stats)
418 {
419 	unsigned i, imax;
420 	unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
421 	unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
422 	const struct pmd_internals *internal = dev->data->dev_private;
423 
424 	imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
425 	        internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
426 	for (i = 0; i < imax; i++) {
427 		igb_stats->q_ipackets[i] = internal->rx_queue[i].rx_pkts;
428 		igb_stats->q_ibytes[i] = internal->rx_queue[i].rx_bytes;
429 		rx_total += igb_stats->q_ipackets[i];
430 		rx_bytes_total += igb_stats->q_ibytes[i];
431 	}
432 
433 	imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
434 	        internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
435 	for (i = 0; i < imax; i++) {
436 		igb_stats->q_opackets[i] = internal->tx_queue[i].tx_pkts;
437 		igb_stats->q_obytes[i] = internal->tx_queue[i].tx_bytes;
438 		tx_total += igb_stats->q_opackets[i];
439 		tx_err_total += internal->tx_queue[i].err_pkts;
440 		tx_bytes_total += igb_stats->q_obytes[i];
441 	}
442 
443 	igb_stats->ipackets = rx_total;
444 	igb_stats->ibytes = rx_bytes_total;
445 	igb_stats->opackets = tx_total;
446 	igb_stats->oerrors = tx_err_total;
447 	igb_stats->obytes = tx_bytes_total;
448 	return 0;
449 }
450 
451 static int
452 eth_stats_reset(struct rte_eth_dev *dev)
453 {
454 	unsigned i;
455 	struct pmd_internals *internal = dev->data->dev_private;
456 
457 	for (i = 0; i < internal->nb_queues; i++) {
458 		internal->rx_queue[i].rx_pkts = 0;
459 		internal->rx_queue[i].rx_bytes = 0;
460 	}
461 
462 	for (i = 0; i < internal->nb_queues; i++) {
463 		internal->tx_queue[i].tx_pkts = 0;
464 		internal->tx_queue[i].err_pkts = 0;
465 		internal->tx_queue[i].tx_bytes = 0;
466 	}
467 
468 	return 0;
469 }
470 
471 static int
472 eth_dev_close(struct rte_eth_dev *dev)
473 {
474 	struct pmd_internals *internals;
475 	struct tpacket_req *req;
476 	unsigned int q;
477 
478 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
479 		return 0;
480 
481 	PMD_LOG(INFO, "Closing AF_PACKET ethdev on NUMA socket %u",
482 		rte_socket_id());
483 
484 	internals = dev->data->dev_private;
485 	req = &internals->req;
486 	for (q = 0; q < internals->nb_queues; q++) {
487 		munmap(internals->rx_queue[q].map,
488 			2 * req->tp_block_size * req->tp_block_nr);
489 		rte_free(internals->rx_queue[q].rd);
490 		rte_free(internals->tx_queue[q].rd);
491 	}
492 	free(internals->if_name);
493 	rte_free(internals->rx_queue);
494 	rte_free(internals->tx_queue);
495 
496 	/* mac_addrs must not be freed alone because part of dev_private */
497 	dev->data->mac_addrs = NULL;
498 	return 0;
499 }
500 
501 static int
502 eth_link_update(struct rte_eth_dev *dev __rte_unused,
503                 int wait_to_complete __rte_unused)
504 {
505 	return 0;
506 }
507 
508 static int
509 eth_rx_queue_setup(struct rte_eth_dev *dev,
510                    uint16_t rx_queue_id,
511                    uint16_t nb_rx_desc __rte_unused,
512                    unsigned int socket_id __rte_unused,
513                    const struct rte_eth_rxconf *rx_conf __rte_unused,
514                    struct rte_mempool *mb_pool)
515 {
516 	struct pmd_internals *internals = dev->data->dev_private;
517 	struct pkt_rx_queue *pkt_q = &internals->rx_queue[rx_queue_id];
518 	unsigned int buf_size, data_size;
519 
520 	pkt_q->mb_pool = mb_pool;
521 
522 	/* Now get the space available for data in the mbuf */
523 	buf_size = rte_pktmbuf_data_room_size(pkt_q->mb_pool) -
524 		RTE_PKTMBUF_HEADROOM;
525 	data_size = internals->req.tp_frame_size;
526 	data_size -= TPACKET2_HDRLEN - sizeof(struct sockaddr_ll);
527 
528 	if (data_size > buf_size) {
529 		PMD_LOG(ERR,
530 			"%s: %d bytes will not fit in mbuf (%d bytes)",
531 			dev->device->name, data_size, buf_size);
532 		return -ENOMEM;
533 	}
534 
535 	dev->data->rx_queues[rx_queue_id] = pkt_q;
536 	pkt_q->in_port = dev->data->port_id;
537 	pkt_q->vlan_strip = internals->vlan_strip;
538 	pkt_q->timestamp_offloading = internals->timestamp_offloading;
539 
540 	return 0;
541 }
542 
543 static int
544 eth_tx_queue_setup(struct rte_eth_dev *dev,
545                    uint16_t tx_queue_id,
546                    uint16_t nb_tx_desc __rte_unused,
547                    unsigned int socket_id __rte_unused,
548                    const struct rte_eth_txconf *tx_conf __rte_unused)
549 {
550 
551 	struct pmd_internals *internals = dev->data->dev_private;
552 
553 	dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
554 	return 0;
555 }
556 
557 static int
558 eth_dev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
559 {
560 	struct pmd_internals *internals = dev->data->dev_private;
561 	struct ifreq ifr = { .ifr_mtu = mtu };
562 	int ret;
563 	int s;
564 	unsigned int data_size = internals->req.tp_frame_size -
565 				 TPACKET2_HDRLEN;
566 
567 	if (mtu > data_size)
568 		return -EINVAL;
569 
570 	s = socket(PF_INET, SOCK_DGRAM, 0);
571 	if (s < 0)
572 		return -EINVAL;
573 
574 	strlcpy(ifr.ifr_name, internals->if_name, IFNAMSIZ);
575 	ret = ioctl(s, SIOCSIFMTU, &ifr);
576 	close(s);
577 
578 	if (ret < 0)
579 		return -EINVAL;
580 
581 	return 0;
582 }
583 
584 static int
585 eth_dev_macaddr_set(struct rte_eth_dev *dev, struct rte_ether_addr *addr)
586 {
587 	struct pmd_internals *internals = dev->data->dev_private;
588 	struct ifreq ifr = { };
589 	int sockfd = internals->rx_queue[0].sockfd;
590 	int ret;
591 
592 	if (sockfd == -1) {
593 		PMD_LOG(ERR, "receive socket not found");
594 		return -EINVAL;
595 	}
596 
597 	strlcpy(ifr.ifr_name, internals->if_name, IFNAMSIZ);
598 	ifr.ifr_hwaddr.sa_family = ARPHRD_ETHER;
599 	memcpy(ifr.ifr_hwaddr.sa_data, addr, sizeof(*addr));
600 	ret = ioctl(sockfd, SIOCSIFHWADDR, &ifr);
601 
602 	if (ret < 0) {
603 		PMD_LOG_ERRNO(ERR, "ioctl(SIOCSIFHWADDR) failed");
604 		return -EINVAL;
605 	}
606 
607 	return 0;
608 }
609 
610 static int
611 eth_dev_change_flags(char *if_name, uint32_t flags, uint32_t mask)
612 {
613 	struct ifreq ifr;
614 	int ret = 0;
615 	int s;
616 
617 	s = socket(PF_INET, SOCK_DGRAM, 0);
618 	if (s < 0)
619 		return -errno;
620 
621 	strlcpy(ifr.ifr_name, if_name, IFNAMSIZ);
622 	if (ioctl(s, SIOCGIFFLAGS, &ifr) < 0) {
623 		ret = -errno;
624 		goto out;
625 	}
626 	ifr.ifr_flags &= mask;
627 	ifr.ifr_flags |= flags;
628 	if (ioctl(s, SIOCSIFFLAGS, &ifr) < 0) {
629 		ret = -errno;
630 		goto out;
631 	}
632 out:
633 	close(s);
634 	return ret;
635 }
636 
637 static int
638 eth_dev_promiscuous_enable(struct rte_eth_dev *dev)
639 {
640 	struct pmd_internals *internals = dev->data->dev_private;
641 
642 	return eth_dev_change_flags(internals->if_name, IFF_PROMISC, ~0);
643 }
644 
645 static int
646 eth_dev_promiscuous_disable(struct rte_eth_dev *dev)
647 {
648 	struct pmd_internals *internals = dev->data->dev_private;
649 
650 	return eth_dev_change_flags(internals->if_name, 0, ~IFF_PROMISC);
651 }
652 
653 static const struct eth_dev_ops ops = {
654 	.dev_start = eth_dev_start,
655 	.dev_stop = eth_dev_stop,
656 	.dev_close = eth_dev_close,
657 	.dev_configure = eth_dev_configure,
658 	.dev_infos_get = eth_dev_info,
659 	.mac_addr_set = eth_dev_macaddr_set,
660 	.mtu_set = eth_dev_mtu_set,
661 	.promiscuous_enable = eth_dev_promiscuous_enable,
662 	.promiscuous_disable = eth_dev_promiscuous_disable,
663 	.rx_queue_setup = eth_rx_queue_setup,
664 	.tx_queue_setup = eth_tx_queue_setup,
665 	.link_update = eth_link_update,
666 	.stats_get = eth_stats_get,
667 	.stats_reset = eth_stats_reset,
668 };
669 
670 /*
671  * Opens an AF_PACKET socket
672  */
673 static int
674 open_packet_iface(const char *key __rte_unused,
675                   const char *value __rte_unused,
676                   void *extra_args)
677 {
678 	int *sockfd = extra_args;
679 
680 	/* Open an AF_PACKET socket... */
681 	*sockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
682 	if (*sockfd == -1) {
683 		PMD_LOG(ERR, "Could not open AF_PACKET socket");
684 		return -1;
685 	}
686 
687 	return 0;
688 }
689 
690 static int
691 rte_pmd_init_internals(struct rte_vdev_device *dev,
692                        const int sockfd,
693                        const unsigned nb_queues,
694                        unsigned int blocksize,
695                        unsigned int blockcnt,
696                        unsigned int framesize,
697                        unsigned int framecnt,
698 		       unsigned int qdisc_bypass,
699                        struct pmd_internals **internals,
700                        struct rte_eth_dev **eth_dev,
701                        struct rte_kvargs *kvlist)
702 {
703 	const char *name = rte_vdev_device_name(dev);
704 	const unsigned int numa_node = dev->device.numa_node;
705 	struct rte_eth_dev_data *data = NULL;
706 	struct rte_kvargs_pair *pair = NULL;
707 	struct ifreq ifr;
708 	size_t ifnamelen;
709 	unsigned k_idx;
710 	struct sockaddr_ll sockaddr;
711 	struct tpacket_req *req;
712 	struct pkt_rx_queue *rx_queue;
713 	struct pkt_tx_queue *tx_queue;
714 	int rc, tpver, discard;
715 	int qsockfd = -1;
716 	unsigned int i, q, rdsize;
717 #if defined(PACKET_FANOUT)
718 	int fanout_arg;
719 #endif
720 
721 	for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
722 		pair = &kvlist->pairs[k_idx];
723 		if (strstr(pair->key, ETH_AF_PACKET_IFACE_ARG) != NULL)
724 			break;
725 	}
726 	if (pair == NULL) {
727 		PMD_LOG(ERR,
728 			"%s: no interface specified for AF_PACKET ethdev",
729 		        name);
730 		return -1;
731 	}
732 
733 	PMD_LOG(INFO,
734 		"%s: creating AF_PACKET-backed ethdev on numa socket %u",
735 		name, numa_node);
736 
737 	*internals = rte_zmalloc_socket(name, sizeof(**internals),
738 	                                0, numa_node);
739 	if (*internals == NULL)
740 		return -1;
741 
742 
743 	(*internals)->rx_queue = rte_calloc_socket("af_packet_rx",
744 						nb_queues,
745 						sizeof(struct pkt_rx_queue),
746 						0, numa_node);
747 	(*internals)->tx_queue = rte_calloc_socket("af_packet_tx",
748 						nb_queues,
749 						sizeof(struct pkt_tx_queue),
750 						0, numa_node);
751 	if (!(*internals)->rx_queue || !(*internals)->tx_queue) {
752 		goto free_internals;
753 	}
754 
755 	for (q = 0; q < nb_queues; q++) {
756 		(*internals)->rx_queue[q].map = MAP_FAILED;
757 		(*internals)->tx_queue[q].map = MAP_FAILED;
758 		(*internals)->rx_queue[q].sockfd = -1;
759 		(*internals)->tx_queue[q].sockfd = -1;
760 	}
761 
762 	req = &((*internals)->req);
763 
764 	req->tp_block_size = blocksize;
765 	req->tp_block_nr = blockcnt;
766 	req->tp_frame_size = framesize;
767 	req->tp_frame_nr = framecnt;
768 
769 	ifnamelen = strlen(pair->value);
770 	if (ifnamelen < sizeof(ifr.ifr_name)) {
771 		memcpy(ifr.ifr_name, pair->value, ifnamelen);
772 		ifr.ifr_name[ifnamelen] = '\0';
773 	} else {
774 		PMD_LOG(ERR,
775 			"%s: I/F name too long (%s)",
776 			name, pair->value);
777 		goto free_internals;
778 	}
779 	if (ioctl(sockfd, SIOCGIFINDEX, &ifr) == -1) {
780 		PMD_LOG_ERRNO(ERR, "%s: ioctl failed (SIOCGIFINDEX)", name);
781 		goto free_internals;
782 	}
783 	(*internals)->if_name = strdup(pair->value);
784 	if ((*internals)->if_name == NULL)
785 		goto free_internals;
786 	(*internals)->if_index = ifr.ifr_ifindex;
787 
788 	if (ioctl(sockfd, SIOCGIFHWADDR, &ifr) == -1) {
789 		PMD_LOG_ERRNO(ERR, "%s: ioctl failed (SIOCGIFHWADDR)", name);
790 		goto free_internals;
791 	}
792 	memcpy(&(*internals)->eth_addr, ifr.ifr_hwaddr.sa_data, ETH_ALEN);
793 
794 	memset(&sockaddr, 0, sizeof(sockaddr));
795 	sockaddr.sll_family = AF_PACKET;
796 	sockaddr.sll_protocol = htons(ETH_P_ALL);
797 	sockaddr.sll_ifindex = (*internals)->if_index;
798 
799 #if defined(PACKET_FANOUT)
800 	fanout_arg = (getpid() ^ (*internals)->if_index) & 0xffff;
801 	fanout_arg |= (PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_DEFRAG) << 16;
802 #if defined(PACKET_FANOUT_FLAG_ROLLOVER)
803 	fanout_arg |= PACKET_FANOUT_FLAG_ROLLOVER << 16;
804 #endif
805 #endif
806 
807 	for (q = 0; q < nb_queues; q++) {
808 		/* Open an AF_PACKET socket for this queue... */
809 		qsockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
810 		if (qsockfd == -1) {
811 			PMD_LOG_ERRNO(ERR,
812 				"%s: could not open AF_PACKET socket",
813 				name);
814 			goto error;
815 		}
816 
817 		tpver = TPACKET_V2;
818 		rc = setsockopt(qsockfd, SOL_PACKET, PACKET_VERSION,
819 				&tpver, sizeof(tpver));
820 		if (rc == -1) {
821 			PMD_LOG_ERRNO(ERR,
822 				"%s: could not set PACKET_VERSION on AF_PACKET socket for %s",
823 				name, pair->value);
824 			goto error;
825 		}
826 
827 		discard = 1;
828 		rc = setsockopt(qsockfd, SOL_PACKET, PACKET_LOSS,
829 				&discard, sizeof(discard));
830 		if (rc == -1) {
831 			PMD_LOG_ERRNO(ERR,
832 				"%s: could not set PACKET_LOSS on AF_PACKET socket for %s",
833 				name, pair->value);
834 			goto error;
835 		}
836 
837 		if (qdisc_bypass) {
838 #if defined(PACKET_QDISC_BYPASS)
839 			rc = setsockopt(qsockfd, SOL_PACKET, PACKET_QDISC_BYPASS,
840 					&qdisc_bypass, sizeof(qdisc_bypass));
841 			if (rc == -1) {
842 				PMD_LOG_ERRNO(ERR,
843 					"%s: could not set PACKET_QDISC_BYPASS on AF_PACKET socket for %s",
844 					name, pair->value);
845 				goto error;
846 			}
847 #endif
848 		}
849 
850 		rc = setsockopt(qsockfd, SOL_PACKET, PACKET_RX_RING, req, sizeof(*req));
851 		if (rc == -1) {
852 			PMD_LOG_ERRNO(ERR,
853 				"%s: could not set PACKET_RX_RING on AF_PACKET socket for %s",
854 				name, pair->value);
855 			goto error;
856 		}
857 
858 		rc = setsockopt(qsockfd, SOL_PACKET, PACKET_TX_RING, req, sizeof(*req));
859 		if (rc == -1) {
860 			PMD_LOG_ERRNO(ERR,
861 				"%s: could not set PACKET_TX_RING on AF_PACKET "
862 				"socket for %s", name, pair->value);
863 			goto error;
864 		}
865 
866 		rx_queue = &((*internals)->rx_queue[q]);
867 		rx_queue->framecount = req->tp_frame_nr;
868 
869 		rx_queue->map = mmap(NULL, 2 * req->tp_block_size * req->tp_block_nr,
870 				    PROT_READ | PROT_WRITE, MAP_SHARED | MAP_LOCKED,
871 				    qsockfd, 0);
872 		if (rx_queue->map == MAP_FAILED) {
873 			PMD_LOG_ERRNO(ERR,
874 				"%s: call to mmap failed on AF_PACKET socket for %s",
875 				name, pair->value);
876 			goto error;
877 		}
878 
879 		/* rdsize is same for both Tx and Rx */
880 		rdsize = req->tp_frame_nr * sizeof(*(rx_queue->rd));
881 
882 		rx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
883 		if (rx_queue->rd == NULL)
884 			goto error;
885 		for (i = 0; i < req->tp_frame_nr; ++i) {
886 			rx_queue->rd[i].iov_base = rx_queue->map + (i * framesize);
887 			rx_queue->rd[i].iov_len = req->tp_frame_size;
888 		}
889 		rx_queue->sockfd = qsockfd;
890 
891 		tx_queue = &((*internals)->tx_queue[q]);
892 		tx_queue->framecount = req->tp_frame_nr;
893 		tx_queue->frame_data_size = req->tp_frame_size;
894 		tx_queue->frame_data_size -= TPACKET2_HDRLEN -
895 			sizeof(struct sockaddr_ll);
896 
897 		tx_queue->map = rx_queue->map + req->tp_block_size * req->tp_block_nr;
898 
899 		tx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
900 		if (tx_queue->rd == NULL)
901 			goto error;
902 		for (i = 0; i < req->tp_frame_nr; ++i) {
903 			tx_queue->rd[i].iov_base = tx_queue->map + (i * framesize);
904 			tx_queue->rd[i].iov_len = req->tp_frame_size;
905 		}
906 		tx_queue->sockfd = qsockfd;
907 
908 		rc = bind(qsockfd, (const struct sockaddr*)&sockaddr, sizeof(sockaddr));
909 		if (rc == -1) {
910 			PMD_LOG_ERRNO(ERR,
911 				"%s: could not bind AF_PACKET socket to %s",
912 				name, pair->value);
913 			goto error;
914 		}
915 
916 #if defined(PACKET_FANOUT)
917 		rc = setsockopt(qsockfd, SOL_PACKET, PACKET_FANOUT,
918 				&fanout_arg, sizeof(fanout_arg));
919 		if (rc == -1) {
920 			PMD_LOG_ERRNO(ERR,
921 				"%s: could not set PACKET_FANOUT on AF_PACKET socket for %s",
922 				name, pair->value);
923 			goto error;
924 		}
925 #endif
926 	}
927 
928 	/* reserve an ethdev entry */
929 	*eth_dev = rte_eth_vdev_allocate(dev, 0);
930 	if (*eth_dev == NULL)
931 		goto error;
932 
933 	/*
934 	 * now put it all together
935 	 * - store queue data in internals,
936 	 * - store numa_node in eth_dev
937 	 * - point eth_dev_data to internals
938 	 * - and point eth_dev structure to new eth_dev_data structure
939 	 */
940 
941 	(*internals)->nb_queues = nb_queues;
942 
943 	data = (*eth_dev)->data;
944 	data->dev_private = *internals;
945 	data->nb_rx_queues = (uint16_t)nb_queues;
946 	data->nb_tx_queues = (uint16_t)nb_queues;
947 	data->dev_link = pmd_link;
948 	data->mac_addrs = &(*internals)->eth_addr;
949 	data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
950 
951 	(*eth_dev)->dev_ops = &ops;
952 
953 	return 0;
954 
955 error:
956 	if (qsockfd != -1)
957 		close(qsockfd);
958 	for (q = 0; q < nb_queues; q++) {
959 		if ((*internals)->rx_queue[q].map != MAP_FAILED)
960 			munmap((*internals)->rx_queue[q].map,
961 			       2 * req->tp_block_size * req->tp_block_nr);
962 
963 		rte_free((*internals)->rx_queue[q].rd);
964 		rte_free((*internals)->tx_queue[q].rd);
965 		if (((*internals)->rx_queue[q].sockfd >= 0) &&
966 			((*internals)->rx_queue[q].sockfd != qsockfd))
967 			close((*internals)->rx_queue[q].sockfd);
968 	}
969 free_internals:
970 	rte_free((*internals)->rx_queue);
971 	rte_free((*internals)->tx_queue);
972 	free((*internals)->if_name);
973 	rte_free(*internals);
974 	return -1;
975 }
976 
977 static int
978 rte_eth_from_packet(struct rte_vdev_device *dev,
979                     int const *sockfd,
980                     struct rte_kvargs *kvlist)
981 {
982 	const char *name = rte_vdev_device_name(dev);
983 	struct pmd_internals *internals = NULL;
984 	struct rte_eth_dev *eth_dev = NULL;
985 	struct rte_kvargs_pair *pair = NULL;
986 	unsigned k_idx;
987 	unsigned int blockcount;
988 	unsigned int blocksize;
989 	unsigned int framesize = DFLT_FRAME_SIZE;
990 	unsigned int framecount = DFLT_FRAME_COUNT;
991 	unsigned int qpairs = 1;
992 	unsigned int qdisc_bypass = 1;
993 
994 	/* do some parameter checking */
995 	if (*sockfd < 0)
996 		return -1;
997 
998 	blocksize = getpagesize();
999 
1000 	/*
1001 	 * Walk arguments for configurable settings
1002 	 */
1003 	for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
1004 		pair = &kvlist->pairs[k_idx];
1005 		if (strstr(pair->key, ETH_AF_PACKET_NUM_Q_ARG) != NULL) {
1006 			qpairs = atoi(pair->value);
1007 			if (qpairs < 1) {
1008 				PMD_LOG(ERR,
1009 					"%s: invalid qpairs value",
1010 				        name);
1011 				return -1;
1012 			}
1013 			continue;
1014 		}
1015 		if (strstr(pair->key, ETH_AF_PACKET_BLOCKSIZE_ARG) != NULL) {
1016 			blocksize = atoi(pair->value);
1017 			if (!blocksize) {
1018 				PMD_LOG(ERR,
1019 					"%s: invalid blocksize value",
1020 				        name);
1021 				return -1;
1022 			}
1023 			continue;
1024 		}
1025 		if (strstr(pair->key, ETH_AF_PACKET_FRAMESIZE_ARG) != NULL) {
1026 			framesize = atoi(pair->value);
1027 			if (!framesize) {
1028 				PMD_LOG(ERR,
1029 					"%s: invalid framesize value",
1030 				        name);
1031 				return -1;
1032 			}
1033 			continue;
1034 		}
1035 		if (strstr(pair->key, ETH_AF_PACKET_FRAMECOUNT_ARG) != NULL) {
1036 			framecount = atoi(pair->value);
1037 			if (!framecount) {
1038 				PMD_LOG(ERR,
1039 					"%s: invalid framecount value",
1040 				        name);
1041 				return -1;
1042 			}
1043 			continue;
1044 		}
1045 		if (strstr(pair->key, ETH_AF_PACKET_QDISC_BYPASS_ARG) != NULL) {
1046 			qdisc_bypass = atoi(pair->value);
1047 			if (qdisc_bypass > 1) {
1048 				PMD_LOG(ERR,
1049 					"%s: invalid bypass value",
1050 					name);
1051 				return -1;
1052 			}
1053 			continue;
1054 		}
1055 	}
1056 
1057 	if (framesize > blocksize) {
1058 		PMD_LOG(ERR,
1059 			"%s: AF_PACKET MMAP frame size exceeds block size!",
1060 		        name);
1061 		return -1;
1062 	}
1063 
1064 	blockcount = framecount / (blocksize / framesize);
1065 	if (!blockcount) {
1066 		PMD_LOG(ERR,
1067 			"%s: invalid AF_PACKET MMAP parameters", name);
1068 		return -1;
1069 	}
1070 
1071 	PMD_LOG(INFO, "%s: AF_PACKET MMAP parameters:", name);
1072 	PMD_LOG(INFO, "%s:\tblock size %d", name, blocksize);
1073 	PMD_LOG(INFO, "%s:\tblock count %d", name, blockcount);
1074 	PMD_LOG(INFO, "%s:\tframe size %d", name, framesize);
1075 	PMD_LOG(INFO, "%s:\tframe count %d", name, framecount);
1076 
1077 	if (rte_pmd_init_internals(dev, *sockfd, qpairs,
1078 				   blocksize, blockcount,
1079 				   framesize, framecount,
1080 				   qdisc_bypass,
1081 				   &internals, &eth_dev,
1082 				   kvlist) < 0)
1083 		return -1;
1084 
1085 	eth_dev->rx_pkt_burst = eth_af_packet_rx;
1086 	eth_dev->tx_pkt_burst = eth_af_packet_tx;
1087 
1088 	rte_eth_dev_probing_finish(eth_dev);
1089 	return 0;
1090 }
1091 
1092 static int
1093 rte_pmd_af_packet_probe(struct rte_vdev_device *dev)
1094 {
1095 	int ret = 0;
1096 	struct rte_kvargs *kvlist;
1097 	int sockfd = -1;
1098 	struct rte_eth_dev *eth_dev;
1099 	const char *name = rte_vdev_device_name(dev);
1100 
1101 	PMD_LOG(INFO, "Initializing pmd_af_packet for %s", name);
1102 
1103 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1104 		eth_dev = rte_eth_dev_attach_secondary(name);
1105 		if (!eth_dev) {
1106 			PMD_LOG(ERR, "Failed to probe %s", name);
1107 			return -1;
1108 		}
1109 		/* TODO: request info from primary to set up Rx and Tx */
1110 		eth_dev->dev_ops = &ops;
1111 		eth_dev->device = &dev->device;
1112 		rte_eth_dev_probing_finish(eth_dev);
1113 		return 0;
1114 	}
1115 
1116 	kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_arguments);
1117 	if (kvlist == NULL) {
1118 		ret = -1;
1119 		goto exit;
1120 	}
1121 
1122 	/*
1123 	 * If iface argument is passed we open the NICs and use them for
1124 	 * reading / writing
1125 	 */
1126 	if (rte_kvargs_count(kvlist, ETH_AF_PACKET_IFACE_ARG) == 1) {
1127 
1128 		ret = rte_kvargs_process(kvlist, ETH_AF_PACKET_IFACE_ARG,
1129 		                         &open_packet_iface, &sockfd);
1130 		if (ret < 0)
1131 			goto exit;
1132 	}
1133 
1134 	if (dev->device.numa_node == SOCKET_ID_ANY)
1135 		dev->device.numa_node = rte_socket_id();
1136 
1137 	ret = rte_eth_from_packet(dev, &sockfd, kvlist);
1138 	close(sockfd); /* no longer needed */
1139 
1140 exit:
1141 	rte_kvargs_free(kvlist);
1142 	return ret;
1143 }
1144 
1145 static int
1146 rte_pmd_af_packet_remove(struct rte_vdev_device *dev)
1147 {
1148 	struct rte_eth_dev *eth_dev;
1149 
1150 	if (dev == NULL)
1151 		return -1;
1152 
1153 	/* find the ethdev entry */
1154 	eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1155 	if (eth_dev == NULL)
1156 		return 0; /* port already released */
1157 
1158 	eth_dev_close(eth_dev);
1159 	rte_eth_dev_release_port(eth_dev);
1160 
1161 	return 0;
1162 }
1163 
1164 static struct rte_vdev_driver pmd_af_packet_drv = {
1165 	.probe = rte_pmd_af_packet_probe,
1166 	.remove = rte_pmd_af_packet_remove,
1167 };
1168 
1169 RTE_PMD_REGISTER_VDEV(net_af_packet, pmd_af_packet_drv);
1170 RTE_PMD_REGISTER_ALIAS(net_af_packet, eth_af_packet);
1171 RTE_PMD_REGISTER_PARAM_STRING(net_af_packet,
1172 	"iface=<string> "
1173 	"qpairs=<int> "
1174 	"blocksz=<int> "
1175 	"framesz=<int> "
1176 	"framecnt=<int> "
1177 	"qdisc_bypass=<0|1>");
1178