xref: /dpdk/drivers/net/af_packet/rte_eth_af_packet.c (revision ceb1ccd5d50c1a89ba8bdd97cc199e7f07422b98)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright(c) 2014 John W. Linville <linville@tuxdriver.com>
5  *
6  *   Originally based upon librte_pmd_pcap code:
7  *
8  *   Copyright(c) 2010-2015 Intel Corporation. All rights reserved.
9  *   Copyright(c) 2014 6WIND S.A.
10  *   All rights reserved.
11  *
12  *   Redistribution and use in source and binary forms, with or without
13  *   modification, are permitted provided that the following conditions
14  *   are met:
15  *
16  *     * Redistributions of source code must retain the above copyright
17  *       notice, this list of conditions and the following disclaimer.
18  *     * Redistributions in binary form must reproduce the above copyright
19  *       notice, this list of conditions and the following disclaimer in
20  *       the documentation and/or other materials provided with the
21  *       distribution.
22  *     * Neither the name of Intel Corporation nor the names of its
23  *       contributors may be used to endorse or promote products derived
24  *       from this software without specific prior written permission.
25  *
26  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
27  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
28  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
29  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
30  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
31  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
32  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
33  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
34  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
35  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
36  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
37  */
38 
39 #include <rte_mbuf.h>
40 #include <rte_ethdev.h>
41 #include <rte_malloc.h>
42 #include <rte_kvargs.h>
43 #include <rte_dev.h>
44 
45 #include <linux/if_ether.h>
46 #include <linux/if_packet.h>
47 #include <arpa/inet.h>
48 #include <net/if.h>
49 #include <sys/types.h>
50 #include <sys/socket.h>
51 #include <sys/ioctl.h>
52 #include <sys/mman.h>
53 #include <unistd.h>
54 #include <poll.h>
55 
56 #define ETH_AF_PACKET_IFACE_ARG		"iface"
57 #define ETH_AF_PACKET_NUM_Q_ARG		"qpairs"
58 #define ETH_AF_PACKET_BLOCKSIZE_ARG	"blocksz"
59 #define ETH_AF_PACKET_FRAMESIZE_ARG	"framesz"
60 #define ETH_AF_PACKET_FRAMECOUNT_ARG	"framecnt"
61 
62 #define DFLT_BLOCK_SIZE		(1 << 12)
63 #define DFLT_FRAME_SIZE		(1 << 11)
64 #define DFLT_FRAME_COUNT	(1 << 9)
65 
66 #define RTE_PMD_AF_PACKET_MAX_RINGS 16
67 
68 struct pkt_rx_queue {
69 	int sockfd;
70 
71 	struct iovec *rd;
72 	uint8_t *map;
73 	unsigned int framecount;
74 	unsigned int framenum;
75 
76 	struct rte_mempool *mb_pool;
77 	uint8_t in_port;
78 
79 	volatile unsigned long rx_pkts;
80 	volatile unsigned long err_pkts;
81 };
82 
83 struct pkt_tx_queue {
84 	int sockfd;
85 
86 	struct iovec *rd;
87 	uint8_t *map;
88 	unsigned int framecount;
89 	unsigned int framenum;
90 
91 	volatile unsigned long tx_pkts;
92 	volatile unsigned long err_pkts;
93 };
94 
95 struct pmd_internals {
96 	unsigned nb_queues;
97 
98 	int if_index;
99 	struct ether_addr eth_addr;
100 
101 	struct tpacket_req req;
102 
103 	struct pkt_rx_queue rx_queue[RTE_PMD_AF_PACKET_MAX_RINGS];
104 	struct pkt_tx_queue tx_queue[RTE_PMD_AF_PACKET_MAX_RINGS];
105 };
106 
107 static const char *valid_arguments[] = {
108 	ETH_AF_PACKET_IFACE_ARG,
109 	ETH_AF_PACKET_NUM_Q_ARG,
110 	ETH_AF_PACKET_BLOCKSIZE_ARG,
111 	ETH_AF_PACKET_FRAMESIZE_ARG,
112 	ETH_AF_PACKET_FRAMECOUNT_ARG,
113 	NULL
114 };
115 
116 static const char *drivername = "AF_PACKET PMD";
117 
118 static struct rte_eth_link pmd_link = {
119 	.link_speed = 10000,
120 	.link_duplex = ETH_LINK_FULL_DUPLEX,
121 	.link_status = 0
122 };
123 
124 static uint16_t
125 eth_af_packet_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
126 {
127 	unsigned i;
128 	struct tpacket2_hdr *ppd;
129 	struct rte_mbuf *mbuf;
130 	uint8_t *pbuf;
131 	struct pkt_rx_queue *pkt_q = queue;
132 	uint16_t num_rx = 0;
133 	unsigned int framecount, framenum;
134 
135 	if (unlikely(nb_pkts == 0))
136 		return 0;
137 
138 	/*
139 	 * Reads the given number of packets from the AF_PACKET socket one by
140 	 * one and copies the packet data into a newly allocated mbuf.
141 	 */
142 	framecount = pkt_q->framecount;
143 	framenum = pkt_q->framenum;
144 	for (i = 0; i < nb_pkts; i++) {
145 		/* point at the next incoming frame */
146 		ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
147 		if ((ppd->tp_status & TP_STATUS_USER) == 0)
148 			break;
149 
150 		/* allocate the next mbuf */
151 		mbuf = rte_pktmbuf_alloc(pkt_q->mb_pool);
152 		if (unlikely(mbuf == NULL))
153 			break;
154 
155 		/* packet will fit in the mbuf, go ahead and receive it */
156 		rte_pktmbuf_pkt_len(mbuf) = rte_pktmbuf_data_len(mbuf) = ppd->tp_snaplen;
157 		pbuf = (uint8_t *) ppd + ppd->tp_mac;
158 		memcpy(rte_pktmbuf_mtod(mbuf, void *), pbuf, rte_pktmbuf_data_len(mbuf));
159 
160 		/* release incoming frame and advance ring buffer */
161 		ppd->tp_status = TP_STATUS_KERNEL;
162 		if (++framenum >= framecount)
163 			framenum = 0;
164 		mbuf->port = pkt_q->in_port;
165 
166 		/* account for the receive frame */
167 		bufs[i] = mbuf;
168 		num_rx++;
169 	}
170 	pkt_q->framenum = framenum;
171 	pkt_q->rx_pkts += num_rx;
172 	return num_rx;
173 }
174 
175 /*
176  * Callback to handle sending packets through a real NIC.
177  */
178 static uint16_t
179 eth_af_packet_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
180 {
181 	struct tpacket2_hdr *ppd;
182 	struct rte_mbuf *mbuf;
183 	uint8_t *pbuf;
184 	unsigned int framecount, framenum;
185 	struct pollfd pfd;
186 	struct pkt_tx_queue *pkt_q = queue;
187 	uint16_t num_tx = 0;
188 	int i;
189 
190 	if (unlikely(nb_pkts == 0))
191 		return 0;
192 
193 	memset(&pfd, 0, sizeof(pfd));
194 	pfd.fd = pkt_q->sockfd;
195 	pfd.events = POLLOUT;
196 	pfd.revents = 0;
197 
198 	framecount = pkt_q->framecount;
199 	framenum = pkt_q->framenum;
200 	ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
201 	for (i = 0; i < nb_pkts; i++) {
202 		/* point at the next incoming frame */
203 		if ((ppd->tp_status != TP_STATUS_AVAILABLE) &&
204 		    (poll(&pfd, 1, -1) < 0))
205 				continue;
206 
207 		/* copy the tx frame data */
208 		mbuf = bufs[num_tx];
209 		pbuf = (uint8_t *) ppd + TPACKET2_HDRLEN -
210 			sizeof(struct sockaddr_ll);
211 		memcpy(pbuf, rte_pktmbuf_mtod(mbuf, void*), rte_pktmbuf_data_len(mbuf));
212 		ppd->tp_len = ppd->tp_snaplen = rte_pktmbuf_data_len(mbuf);
213 
214 		/* release incoming frame and advance ring buffer */
215 		ppd->tp_status = TP_STATUS_SEND_REQUEST;
216 		if (++framenum >= framecount)
217 			framenum = 0;
218 		ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
219 
220 		num_tx++;
221 		rte_pktmbuf_free(mbuf);
222 	}
223 
224 	/* kick-off transmits */
225 	if (sendto(pkt_q->sockfd, NULL, 0, MSG_DONTWAIT, NULL, 0) == -1)
226 		return 0; /* error sending -- no packets transmitted */
227 
228 	pkt_q->framenum = framenum;
229 	pkt_q->tx_pkts += num_tx;
230 	pkt_q->err_pkts += nb_pkts - num_tx;
231 	return num_tx;
232 }
233 
234 static int
235 eth_dev_start(struct rte_eth_dev *dev)
236 {
237 	dev->data->dev_link.link_status = 1;
238 	return 0;
239 }
240 
241 /*
242  * This function gets called when the current port gets stopped.
243  */
244 static void
245 eth_dev_stop(struct rte_eth_dev *dev)
246 {
247 	unsigned i;
248 	int sockfd;
249 	struct pmd_internals *internals = dev->data->dev_private;
250 
251 	for (i = 0; i < internals->nb_queues; i++) {
252 		sockfd = internals->rx_queue[i].sockfd;
253 		if (sockfd != -1)
254 			close(sockfd);
255 		sockfd = internals->tx_queue[i].sockfd;
256 		if (sockfd != -1)
257 			close(sockfd);
258 	}
259 
260 	dev->data->dev_link.link_status = 0;
261 }
262 
263 static int
264 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
265 {
266 	return 0;
267 }
268 
269 static void
270 eth_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
271 {
272 	struct pmd_internals *internals = dev->data->dev_private;
273 
274 	dev_info->driver_name = drivername;
275 	dev_info->if_index = internals->if_index;
276 	dev_info->max_mac_addrs = 1;
277 	dev_info->max_rx_pktlen = (uint32_t)ETH_FRAME_LEN;
278 	dev_info->max_rx_queues = (uint16_t)internals->nb_queues;
279 	dev_info->max_tx_queues = (uint16_t)internals->nb_queues;
280 	dev_info->min_rx_bufsize = 0;
281 	dev_info->pci_dev = NULL;
282 }
283 
284 static void
285 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *igb_stats)
286 {
287 	unsigned i, imax;
288 	unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
289 	const struct pmd_internals *internal = dev->data->dev_private;
290 
291 	imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
292 	        internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
293 	for (i = 0; i < imax; i++) {
294 		igb_stats->q_ipackets[i] = internal->rx_queue[i].rx_pkts;
295 		rx_total += igb_stats->q_ipackets[i];
296 	}
297 
298 	imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
299 	        internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
300 	for (i = 0; i < imax; i++) {
301 		igb_stats->q_opackets[i] = internal->tx_queue[i].tx_pkts;
302 		igb_stats->q_errors[i] = internal->tx_queue[i].err_pkts;
303 		tx_total += igb_stats->q_opackets[i];
304 		tx_err_total += igb_stats->q_errors[i];
305 	}
306 
307 	igb_stats->ipackets = rx_total;
308 	igb_stats->opackets = tx_total;
309 	igb_stats->oerrors = tx_err_total;
310 }
311 
312 static void
313 eth_stats_reset(struct rte_eth_dev *dev)
314 {
315 	unsigned i;
316 	struct pmd_internals *internal = dev->data->dev_private;
317 
318 	for (i = 0; i < internal->nb_queues; i++)
319 		internal->rx_queue[i].rx_pkts = 0;
320 
321 	for (i = 0; i < internal->nb_queues; i++) {
322 		internal->tx_queue[i].tx_pkts = 0;
323 		internal->tx_queue[i].err_pkts = 0;
324 	}
325 }
326 
327 static void
328 eth_dev_close(struct rte_eth_dev *dev __rte_unused)
329 {
330 }
331 
332 static void
333 eth_queue_release(void *q __rte_unused)
334 {
335 }
336 
337 static int
338 eth_link_update(struct rte_eth_dev *dev __rte_unused,
339                 int wait_to_complete __rte_unused)
340 {
341 	return 0;
342 }
343 
344 static int
345 eth_rx_queue_setup(struct rte_eth_dev *dev,
346                    uint16_t rx_queue_id,
347                    uint16_t nb_rx_desc __rte_unused,
348                    unsigned int socket_id __rte_unused,
349                    const struct rte_eth_rxconf *rx_conf __rte_unused,
350                    struct rte_mempool *mb_pool)
351 {
352 	struct pmd_internals *internals = dev->data->dev_private;
353 	struct pkt_rx_queue *pkt_q = &internals->rx_queue[rx_queue_id];
354 	uint16_t buf_size;
355 
356 	pkt_q->mb_pool = mb_pool;
357 
358 	/* Now get the space available for data in the mbuf */
359 	buf_size = (uint16_t)(rte_pktmbuf_data_room_size(pkt_q->mb_pool) -
360 		RTE_PKTMBUF_HEADROOM);
361 
362 	if (ETH_FRAME_LEN > buf_size) {
363 		RTE_LOG(ERR, PMD,
364 			"%s: %d bytes will not fit in mbuf (%d bytes)\n",
365 			dev->data->name, ETH_FRAME_LEN, buf_size);
366 		return -ENOMEM;
367 	}
368 
369 	dev->data->rx_queues[rx_queue_id] = pkt_q;
370 	pkt_q->in_port = dev->data->port_id;
371 
372 	return 0;
373 }
374 
375 static int
376 eth_tx_queue_setup(struct rte_eth_dev *dev,
377                    uint16_t tx_queue_id,
378                    uint16_t nb_tx_desc __rte_unused,
379                    unsigned int socket_id __rte_unused,
380                    const struct rte_eth_txconf *tx_conf __rte_unused)
381 {
382 
383 	struct pmd_internals *internals = dev->data->dev_private;
384 
385 	dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
386 	return 0;
387 }
388 
389 static const struct eth_dev_ops ops = {
390 	.dev_start = eth_dev_start,
391 	.dev_stop = eth_dev_stop,
392 	.dev_close = eth_dev_close,
393 	.dev_configure = eth_dev_configure,
394 	.dev_infos_get = eth_dev_info,
395 	.rx_queue_setup = eth_rx_queue_setup,
396 	.tx_queue_setup = eth_tx_queue_setup,
397 	.rx_queue_release = eth_queue_release,
398 	.tx_queue_release = eth_queue_release,
399 	.link_update = eth_link_update,
400 	.stats_get = eth_stats_get,
401 	.stats_reset = eth_stats_reset,
402 };
403 
404 /*
405  * Opens an AF_PACKET socket
406  */
407 static int
408 open_packet_iface(const char *key __rte_unused,
409                   const char *value __rte_unused,
410                   void *extra_args)
411 {
412 	int *sockfd = extra_args;
413 
414 	/* Open an AF_PACKET socket... */
415 	*sockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
416 	if (*sockfd == -1) {
417 		RTE_LOG(ERR, PMD, "Could not open AF_PACKET socket\n");
418 		return -1;
419 	}
420 
421 	return 0;
422 }
423 
424 static int
425 rte_pmd_init_internals(const char *name,
426                        const int sockfd,
427                        const unsigned nb_queues,
428                        unsigned int blocksize,
429                        unsigned int blockcnt,
430                        unsigned int framesize,
431                        unsigned int framecnt,
432                        const unsigned numa_node,
433                        struct pmd_internals **internals,
434                        struct rte_eth_dev **eth_dev,
435                        struct rte_kvargs *kvlist)
436 {
437 	struct rte_eth_dev_data *data = NULL;
438 	struct rte_kvargs_pair *pair = NULL;
439 	struct ifreq ifr;
440 	size_t ifnamelen;
441 	unsigned k_idx;
442 	struct sockaddr_ll sockaddr;
443 	struct tpacket_req *req;
444 	struct pkt_rx_queue *rx_queue;
445 	struct pkt_tx_queue *tx_queue;
446 	int rc, tpver, discard;
447 	int qsockfd = -1;
448 	unsigned int i, q, rdsize;
449 	int fanout_arg __rte_unused, bypass __rte_unused;
450 
451 	for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
452 		pair = &kvlist->pairs[k_idx];
453 		if (strstr(pair->key, ETH_AF_PACKET_IFACE_ARG) != NULL)
454 			break;
455 	}
456 	if (pair == NULL) {
457 		RTE_LOG(ERR, PMD,
458 			"%s: no interface specified for AF_PACKET ethdev\n",
459 		        name);
460 		goto error_early;
461 	}
462 
463 	RTE_LOG(INFO, PMD,
464 		"%s: creating AF_PACKET-backed ethdev on numa socket %u\n",
465 		name, numa_node);
466 
467 	/*
468 	 * now do all data allocation - for eth_dev structure, dummy pci driver
469 	 * and internal (private) data
470 	 */
471 	data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
472 	if (data == NULL)
473 		goto error_early;
474 
475 	*internals = rte_zmalloc_socket(name, sizeof(**internals),
476 	                                0, numa_node);
477 	if (*internals == NULL)
478 		goto error_early;
479 
480 	for (q = 0; q < nb_queues; q++) {
481 		(*internals)->rx_queue[q].map = MAP_FAILED;
482 		(*internals)->tx_queue[q].map = MAP_FAILED;
483 	}
484 
485 	req = &((*internals)->req);
486 
487 	req->tp_block_size = blocksize;
488 	req->tp_block_nr = blockcnt;
489 	req->tp_frame_size = framesize;
490 	req->tp_frame_nr = framecnt;
491 
492 	ifnamelen = strlen(pair->value);
493 	if (ifnamelen < sizeof(ifr.ifr_name)) {
494 		memcpy(ifr.ifr_name, pair->value, ifnamelen);
495 		ifr.ifr_name[ifnamelen] = '\0';
496 	} else {
497 		RTE_LOG(ERR, PMD,
498 			"%s: I/F name too long (%s)\n",
499 			name, pair->value);
500 		goto error_early;
501 	}
502 	if (ioctl(sockfd, SIOCGIFINDEX, &ifr) == -1) {
503 		RTE_LOG(ERR, PMD,
504 			"%s: ioctl failed (SIOCGIFINDEX)\n",
505 		        name);
506 		goto error_early;
507 	}
508 	(*internals)->if_index = ifr.ifr_ifindex;
509 
510 	if (ioctl(sockfd, SIOCGIFHWADDR, &ifr) == -1) {
511 		RTE_LOG(ERR, PMD,
512 			"%s: ioctl failed (SIOCGIFHWADDR)\n",
513 		        name);
514 		goto error_early;
515 	}
516 	memcpy(&(*internals)->eth_addr, ifr.ifr_hwaddr.sa_data, ETH_ALEN);
517 
518 	memset(&sockaddr, 0, sizeof(sockaddr));
519 	sockaddr.sll_family = AF_PACKET;
520 	sockaddr.sll_protocol = htons(ETH_P_ALL);
521 	sockaddr.sll_ifindex = (*internals)->if_index;
522 
523 #if defined(PACKET_FANOUT)
524 	fanout_arg = (getpid() ^ (*internals)->if_index) & 0xffff;
525 	fanout_arg |= (PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_DEFRAG) << 16;
526 #if defined(PACKET_FANOUT_FLAG_ROLLOVER)
527 	fanout_arg |= PACKET_FANOUT_FLAG_ROLLOVER << 16;
528 #endif
529 #endif
530 
531 	for (q = 0; q < nb_queues; q++) {
532 		/* Open an AF_PACKET socket for this queue... */
533 		qsockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
534 		if (qsockfd == -1) {
535 			RTE_LOG(ERR, PMD,
536 			        "%s: could not open AF_PACKET socket\n",
537 			        name);
538 			return -1;
539 		}
540 
541 		tpver = TPACKET_V2;
542 		rc = setsockopt(qsockfd, SOL_PACKET, PACKET_VERSION,
543 				&tpver, sizeof(tpver));
544 		if (rc == -1) {
545 			RTE_LOG(ERR, PMD,
546 				"%s: could not set PACKET_VERSION on AF_PACKET "
547 				"socket for %s\n", name, pair->value);
548 			goto error;
549 		}
550 
551 		discard = 1;
552 		rc = setsockopt(qsockfd, SOL_PACKET, PACKET_LOSS,
553 				&discard, sizeof(discard));
554 		if (rc == -1) {
555 			RTE_LOG(ERR, PMD,
556 				"%s: could not set PACKET_LOSS on "
557 			        "AF_PACKET socket for %s\n", name, pair->value);
558 			goto error;
559 		}
560 
561 #if defined(PACKET_QDISC_BYPASS)
562 		bypass = 1;
563 		rc = setsockopt(qsockfd, SOL_PACKET, PACKET_QDISC_BYPASS,
564 				&bypass, sizeof(bypass));
565 		if (rc == -1) {
566 			RTE_LOG(ERR, PMD,
567 				"%s: could not set PACKET_QDISC_BYPASS "
568 			        "on AF_PACKET socket for %s\n", name,
569 			        pair->value);
570 			goto error;
571 		}
572 #endif
573 
574 		rc = setsockopt(qsockfd, SOL_PACKET, PACKET_RX_RING, req, sizeof(*req));
575 		if (rc == -1) {
576 			RTE_LOG(ERR, PMD,
577 				"%s: could not set PACKET_RX_RING on AF_PACKET "
578 				"socket for %s\n", name, pair->value);
579 			goto error;
580 		}
581 
582 		rc = setsockopt(qsockfd, SOL_PACKET, PACKET_TX_RING, req, sizeof(*req));
583 		if (rc == -1) {
584 			RTE_LOG(ERR, PMD,
585 				"%s: could not set PACKET_TX_RING on AF_PACKET "
586 				"socket for %s\n", name, pair->value);
587 			goto error;
588 		}
589 
590 		rx_queue = &((*internals)->rx_queue[q]);
591 		rx_queue->framecount = req->tp_frame_nr;
592 
593 		rx_queue->map = mmap(NULL, 2 * req->tp_block_size * req->tp_block_nr,
594 				    PROT_READ | PROT_WRITE, MAP_SHARED | MAP_LOCKED,
595 				    qsockfd, 0);
596 		if (rx_queue->map == MAP_FAILED) {
597 			RTE_LOG(ERR, PMD,
598 				"%s: call to mmap failed on AF_PACKET socket for %s\n",
599 				name, pair->value);
600 			goto error;
601 		}
602 
603 		/* rdsize is same for both Tx and Rx */
604 		rdsize = req->tp_frame_nr * sizeof(*(rx_queue->rd));
605 
606 		rx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
607 		if (rx_queue->rd == NULL)
608 			goto error;
609 		for (i = 0; i < req->tp_frame_nr; ++i) {
610 			rx_queue->rd[i].iov_base = rx_queue->map + (i * framesize);
611 			rx_queue->rd[i].iov_len = req->tp_frame_size;
612 		}
613 		rx_queue->sockfd = qsockfd;
614 
615 		tx_queue = &((*internals)->tx_queue[q]);
616 		tx_queue->framecount = req->tp_frame_nr;
617 
618 		tx_queue->map = rx_queue->map + req->tp_block_size * req->tp_block_nr;
619 
620 		tx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
621 		if (tx_queue->rd == NULL)
622 			goto error;
623 		for (i = 0; i < req->tp_frame_nr; ++i) {
624 			tx_queue->rd[i].iov_base = tx_queue->map + (i * framesize);
625 			tx_queue->rd[i].iov_len = req->tp_frame_size;
626 		}
627 		tx_queue->sockfd = qsockfd;
628 
629 		rc = bind(qsockfd, (const struct sockaddr*)&sockaddr, sizeof(sockaddr));
630 		if (rc == -1) {
631 			RTE_LOG(ERR, PMD,
632 				"%s: could not bind AF_PACKET socket to %s\n",
633 			        name, pair->value);
634 			goto error;
635 		}
636 
637 #if defined(PACKET_FANOUT)
638 		rc = setsockopt(qsockfd, SOL_PACKET, PACKET_FANOUT,
639 				&fanout_arg, sizeof(fanout_arg));
640 		if (rc == -1) {
641 			RTE_LOG(ERR, PMD,
642 				"%s: could not set PACKET_FANOUT on AF_PACKET socket "
643 				"for %s\n", name, pair->value);
644 			goto error;
645 		}
646 #endif
647 	}
648 
649 	/* reserve an ethdev entry */
650 	*eth_dev = rte_eth_dev_allocate(name, RTE_ETH_DEV_VIRTUAL);
651 	if (*eth_dev == NULL)
652 		goto error;
653 
654 	/*
655 	 * now put it all together
656 	 * - store queue data in internals,
657 	 * - store numa_node in eth_dev
658 	 * - point eth_dev_data to internals
659 	 * - and point eth_dev structure to new eth_dev_data structure
660 	 */
661 
662 	(*internals)->nb_queues = nb_queues;
663 
664 	data->dev_private = *internals;
665 	data->port_id = (*eth_dev)->data->port_id;
666 	data->nb_rx_queues = (uint16_t)nb_queues;
667 	data->nb_tx_queues = (uint16_t)nb_queues;
668 	data->dev_link = pmd_link;
669 	data->mac_addrs = &(*internals)->eth_addr;
670 	strncpy(data->name,
671 		(*eth_dev)->data->name, strlen((*eth_dev)->data->name));
672 
673 	(*eth_dev)->data = data;
674 	(*eth_dev)->dev_ops = &ops;
675 	(*eth_dev)->driver = NULL;
676 	(*eth_dev)->data->dev_flags = RTE_ETH_DEV_DETACHABLE;
677 	(*eth_dev)->data->drv_name = drivername;
678 	(*eth_dev)->data->kdrv = RTE_KDRV_NONE;
679 	(*eth_dev)->data->numa_node = numa_node;
680 
681 	return 0;
682 
683 error:
684 	if (qsockfd != -1)
685 		close(qsockfd);
686 	for (q = 0; q < nb_queues; q++) {
687 		munmap((*internals)->rx_queue[q].map,
688 		       2 * req->tp_block_size * req->tp_block_nr);
689 
690 		rte_free((*internals)->rx_queue[q].rd);
691 		rte_free((*internals)->tx_queue[q].rd);
692 		if (((*internals)->rx_queue[q].sockfd != 0) &&
693 			((*internals)->rx_queue[q].sockfd != qsockfd))
694 			close((*internals)->rx_queue[q].sockfd);
695 	}
696 	rte_free(*internals);
697 error_early:
698 	rte_free(data);
699 	return -1;
700 }
701 
702 static int
703 rte_eth_from_packet(const char *name,
704                     int const *sockfd,
705                     const unsigned numa_node,
706                     struct rte_kvargs *kvlist)
707 {
708 	struct pmd_internals *internals = NULL;
709 	struct rte_eth_dev *eth_dev = NULL;
710 	struct rte_kvargs_pair *pair = NULL;
711 	unsigned k_idx;
712 	unsigned int blockcount;
713 	unsigned int blocksize = DFLT_BLOCK_SIZE;
714 	unsigned int framesize = DFLT_FRAME_SIZE;
715 	unsigned int framecount = DFLT_FRAME_COUNT;
716 	unsigned int qpairs = 1;
717 
718 	/* do some parameter checking */
719 	if (*sockfd < 0)
720 		return -1;
721 
722 	/*
723 	 * Walk arguments for configurable settings
724 	 */
725 	for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
726 		pair = &kvlist->pairs[k_idx];
727 		if (strstr(pair->key, ETH_AF_PACKET_NUM_Q_ARG) != NULL) {
728 			qpairs = atoi(pair->value);
729 			if (qpairs < 1 ||
730 			    qpairs > RTE_PMD_AF_PACKET_MAX_RINGS) {
731 				RTE_LOG(ERR, PMD,
732 					"%s: invalid qpairs value\n",
733 				        name);
734 				return -1;
735 			}
736 			continue;
737 		}
738 		if (strstr(pair->key, ETH_AF_PACKET_BLOCKSIZE_ARG) != NULL) {
739 			blocksize = atoi(pair->value);
740 			if (!blocksize) {
741 				RTE_LOG(ERR, PMD,
742 					"%s: invalid blocksize value\n",
743 				        name);
744 				return -1;
745 			}
746 			continue;
747 		}
748 		if (strstr(pair->key, ETH_AF_PACKET_FRAMESIZE_ARG) != NULL) {
749 			framesize = atoi(pair->value);
750 			if (!framesize) {
751 				RTE_LOG(ERR, PMD,
752 					"%s: invalid framesize value\n",
753 				        name);
754 				return -1;
755 			}
756 			continue;
757 		}
758 		if (strstr(pair->key, ETH_AF_PACKET_FRAMECOUNT_ARG) != NULL) {
759 			framecount = atoi(pair->value);
760 			if (!framecount) {
761 				RTE_LOG(ERR, PMD,
762 					"%s: invalid framecount value\n",
763 				        name);
764 				return -1;
765 			}
766 			continue;
767 		}
768 	}
769 
770 	if (framesize > blocksize) {
771 		RTE_LOG(ERR, PMD,
772 			"%s: AF_PACKET MMAP frame size exceeds block size!\n",
773 		        name);
774 		return -1;
775 	}
776 
777 	blockcount = framecount / (blocksize / framesize);
778 	if (!blockcount) {
779 		RTE_LOG(ERR, PMD,
780 			"%s: invalid AF_PACKET MMAP parameters\n", name);
781 		return -1;
782 	}
783 
784 	RTE_LOG(INFO, PMD, "%s: AF_PACKET MMAP parameters:\n", name);
785 	RTE_LOG(INFO, PMD, "%s:\tblock size %d\n", name, blocksize);
786 	RTE_LOG(INFO, PMD, "%s:\tblock count %d\n", name, blockcount);
787 	RTE_LOG(INFO, PMD, "%s:\tframe size %d\n", name, framesize);
788 	RTE_LOG(INFO, PMD, "%s:\tframe count %d\n", name, framecount);
789 
790 	if (rte_pmd_init_internals(name, *sockfd, qpairs,
791 	                           blocksize, blockcount,
792 	                           framesize, framecount,
793 	                           numa_node, &internals, &eth_dev,
794 	                           kvlist) < 0)
795 		return -1;
796 
797 	eth_dev->rx_pkt_burst = eth_af_packet_rx;
798 	eth_dev->tx_pkt_burst = eth_af_packet_tx;
799 
800 	return 0;
801 }
802 
803 static int
804 rte_pmd_af_packet_devinit(const char *name, const char *params)
805 {
806 	unsigned numa_node;
807 	int ret = 0;
808 	struct rte_kvargs *kvlist;
809 	int sockfd = -1;
810 
811 	RTE_LOG(INFO, PMD, "Initializing pmd_af_packet for %s\n", name);
812 
813 	numa_node = rte_socket_id();
814 
815 	kvlist = rte_kvargs_parse(params, valid_arguments);
816 	if (kvlist == NULL) {
817 		ret = -1;
818 		goto exit;
819 	}
820 
821 	/*
822 	 * If iface argument is passed we open the NICs and use them for
823 	 * reading / writing
824 	 */
825 	if (rte_kvargs_count(kvlist, ETH_AF_PACKET_IFACE_ARG) == 1) {
826 
827 		ret = rte_kvargs_process(kvlist, ETH_AF_PACKET_IFACE_ARG,
828 		                         &open_packet_iface, &sockfd);
829 		if (ret < 0)
830 			goto exit;
831 	}
832 
833 	ret = rte_eth_from_packet(name, &sockfd, numa_node, kvlist);
834 	close(sockfd); /* no longer needed */
835 
836 exit:
837 	rte_kvargs_free(kvlist);
838 	return ret;
839 }
840 
841 static int
842 rte_pmd_af_packet_devuninit(const char *name)
843 {
844 	struct rte_eth_dev *eth_dev = NULL;
845 	struct pmd_internals *internals;
846 	unsigned q;
847 
848 	RTE_LOG(INFO, PMD, "Closing AF_PACKET ethdev on numa socket %u\n",
849 			rte_socket_id());
850 
851 	if (name == NULL)
852 		return -1;
853 
854 	/* find the ethdev entry */
855 	eth_dev = rte_eth_dev_allocated(name);
856 	if (eth_dev == NULL)
857 		return -1;
858 
859 	internals = eth_dev->data->dev_private;
860 	for (q = 0; q < internals->nb_queues; q++) {
861 		rte_free(internals->rx_queue[q].rd);
862 		rte_free(internals->tx_queue[q].rd);
863 	}
864 
865 	rte_free(eth_dev->data->dev_private);
866 	rte_free(eth_dev->data);
867 
868 	rte_eth_dev_release_port(eth_dev);
869 
870 	return 0;
871 }
872 
873 static struct rte_driver pmd_af_packet_drv = {
874 	.name = "eth_af_packet",
875 	.type = PMD_VDEV,
876 	.init = rte_pmd_af_packet_devinit,
877 	.uninit = rte_pmd_af_packet_devuninit,
878 };
879 
880 PMD_REGISTER_DRIVER(pmd_af_packet_drv);
881