xref: /dpdk/drivers/net/pcap/pcap_ethdev.c (revision 0899a87ce7c71576d445cd956036f6f7e555f01c)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2016 Intel Corporation.
3  * Copyright(c) 2014 6WIND S.A.
4  * All rights reserved.
5  */
6 
7 #include <stdlib.h>
8 #include <time.h>
9 
10 #include <pcap.h>
11 
12 #include <rte_cycles.h>
13 #include <ethdev_driver.h>
14 #include <ethdev_vdev.h>
15 #include <rte_kvargs.h>
16 #include <rte_malloc.h>
17 #include <rte_mbuf.h>
18 #include <rte_mbuf_dyn.h>
19 #include <bus_vdev_driver.h>
20 #include <rte_os_shim.h>
21 
22 #include "pcap_osdep.h"
23 
24 #define RTE_ETH_PCAP_SNAPSHOT_LEN 65535
25 #define RTE_ETH_PCAP_SNAPLEN RTE_ETHER_MAX_JUMBO_FRAME_LEN
26 #define RTE_ETH_PCAP_PROMISC 1
27 #define RTE_ETH_PCAP_TIMEOUT -1
28 
29 #define ETH_PCAP_RX_PCAP_ARG  "rx_pcap"
30 #define ETH_PCAP_TX_PCAP_ARG  "tx_pcap"
31 #define ETH_PCAP_RX_IFACE_ARG "rx_iface"
32 #define ETH_PCAP_RX_IFACE_IN_ARG "rx_iface_in"
33 #define ETH_PCAP_TX_IFACE_ARG "tx_iface"
34 #define ETH_PCAP_IFACE_ARG    "iface"
35 #define ETH_PCAP_PHY_MAC_ARG  "phy_mac"
36 #define ETH_PCAP_INFINITE_RX_ARG  "infinite_rx"
37 
38 #define ETH_PCAP_ARG_MAXLEN	64
39 
40 #define RTE_PMD_PCAP_MAX_QUEUES 16
41 
42 static char errbuf[PCAP_ERRBUF_SIZE];
43 static struct timespec start_time;
44 static uint64_t start_cycles;
45 static uint64_t hz;
46 static uint8_t iface_idx;
47 
48 static uint64_t timestamp_rx_dynflag;
49 static int timestamp_dynfield_offset = -1;
50 
51 struct queue_stat {
52 	volatile unsigned long pkts;
53 	volatile unsigned long bytes;
54 	volatile unsigned long err_pkts;
55 	volatile unsigned long rx_nombuf;
56 };
57 
58 struct queue_missed_stat {
59 	/* last value retrieved from pcap */
60 	unsigned int pcap;
61 	/* stores values lost by pcap stop or rollover */
62 	unsigned long mnemonic;
63 	/* value on last reset */
64 	unsigned long reset;
65 };
66 
67 struct pcap_rx_queue {
68 	uint16_t port_id;
69 	uint16_t queue_id;
70 	struct rte_mempool *mb_pool;
71 	struct queue_stat rx_stat;
72 	struct queue_missed_stat missed_stat;
73 	char name[PATH_MAX];
74 	char type[ETH_PCAP_ARG_MAXLEN];
75 
76 	/* Contains pre-generated packets to be looped through */
77 	struct rte_ring *pkts;
78 };
79 
80 struct pcap_tx_queue {
81 	uint16_t port_id;
82 	uint16_t queue_id;
83 	struct queue_stat tx_stat;
84 	char name[PATH_MAX];
85 	char type[ETH_PCAP_ARG_MAXLEN];
86 };
87 
88 struct pmd_internals {
89 	struct pcap_rx_queue rx_queue[RTE_PMD_PCAP_MAX_QUEUES];
90 	struct pcap_tx_queue tx_queue[RTE_PMD_PCAP_MAX_QUEUES];
91 	char devargs[ETH_PCAP_ARG_MAXLEN];
92 	struct rte_ether_addr eth_addr;
93 	int if_index;
94 	int single_iface;
95 	int phy_mac;
96 	unsigned int infinite_rx;
97 };
98 
99 struct pmd_process_private {
100 	pcap_t *rx_pcap[RTE_PMD_PCAP_MAX_QUEUES];
101 	pcap_t *tx_pcap[RTE_PMD_PCAP_MAX_QUEUES];
102 	pcap_dumper_t *tx_dumper[RTE_PMD_PCAP_MAX_QUEUES];
103 };
104 
105 struct pmd_devargs {
106 	unsigned int num_of_queue;
107 	struct devargs_queue {
108 		pcap_dumper_t *dumper;
109 		pcap_t *pcap;
110 		const char *name;
111 		const char *type;
112 	} queue[RTE_PMD_PCAP_MAX_QUEUES];
113 	int phy_mac;
114 };
115 
116 struct pmd_devargs_all {
117 	struct pmd_devargs rx_queues;
118 	struct pmd_devargs tx_queues;
119 	int single_iface;
120 	unsigned int is_tx_pcap;
121 	unsigned int is_tx_iface;
122 	unsigned int is_rx_pcap;
123 	unsigned int is_rx_iface;
124 	unsigned int infinite_rx;
125 };
126 
127 static const char *valid_arguments[] = {
128 	ETH_PCAP_RX_PCAP_ARG,
129 	ETH_PCAP_TX_PCAP_ARG,
130 	ETH_PCAP_RX_IFACE_ARG,
131 	ETH_PCAP_RX_IFACE_IN_ARG,
132 	ETH_PCAP_TX_IFACE_ARG,
133 	ETH_PCAP_IFACE_ARG,
134 	ETH_PCAP_PHY_MAC_ARG,
135 	ETH_PCAP_INFINITE_RX_ARG,
136 	NULL
137 };
138 
139 static struct rte_eth_link pmd_link = {
140 		.link_speed = RTE_ETH_SPEED_NUM_10G,
141 		.link_duplex = RTE_ETH_LINK_FULL_DUPLEX,
142 		.link_status = RTE_ETH_LINK_DOWN,
143 		.link_autoneg = RTE_ETH_LINK_FIXED,
144 };
145 
146 RTE_LOG_REGISTER_DEFAULT(eth_pcap_logtype, NOTICE);
147 
148 static struct queue_missed_stat*
149 queue_missed_stat_update(struct rte_eth_dev *dev, unsigned int qid)
150 {
151 	struct pmd_internals *internals = dev->data->dev_private;
152 	struct queue_missed_stat *missed_stat =
153 			&internals->rx_queue[qid].missed_stat;
154 	const struct pmd_process_private *pp = dev->process_private;
155 	pcap_t *pcap = pp->rx_pcap[qid];
156 	struct pcap_stat stat;
157 
158 	if (!pcap || (pcap_stats(pcap, &stat) != 0))
159 		return missed_stat;
160 
161 	/* rollover check - best effort fixup assuming single rollover */
162 	if (stat.ps_drop < missed_stat->pcap)
163 		missed_stat->mnemonic += UINT_MAX;
164 	missed_stat->pcap = stat.ps_drop;
165 
166 	return missed_stat;
167 }
168 
169 static void
170 queue_missed_stat_on_stop_update(struct rte_eth_dev *dev, unsigned int qid)
171 {
172 	struct queue_missed_stat *missed_stat =
173 			queue_missed_stat_update(dev, qid);
174 
175 	missed_stat->mnemonic += missed_stat->pcap;
176 	missed_stat->pcap = 0;
177 }
178 
179 static void
180 queue_missed_stat_reset(struct rte_eth_dev *dev, unsigned int qid)
181 {
182 	struct queue_missed_stat *missed_stat =
183 			queue_missed_stat_update(dev, qid);
184 
185 	missed_stat->reset = missed_stat->pcap;
186 	missed_stat->mnemonic = 0;
187 }
188 
189 static unsigned long
190 queue_missed_stat_get(struct rte_eth_dev *dev, unsigned int qid)
191 {
192 	const struct queue_missed_stat *missed_stat =
193 			queue_missed_stat_update(dev, qid);
194 
195 	return missed_stat->pcap + missed_stat->mnemonic - missed_stat->reset;
196 }
197 
198 static int
199 eth_pcap_rx_jumbo(struct rte_mempool *mb_pool, struct rte_mbuf *mbuf,
200 		const u_char *data, uint16_t data_len)
201 {
202 	/* Copy the first segment. */
203 	uint16_t len = rte_pktmbuf_tailroom(mbuf);
204 	struct rte_mbuf *m = mbuf;
205 
206 	rte_memcpy(rte_pktmbuf_append(mbuf, len), data, len);
207 	data_len -= len;
208 	data += len;
209 
210 	while (data_len > 0) {
211 		/* Allocate next mbuf and point to that. */
212 		m->next = rte_pktmbuf_alloc(mb_pool);
213 
214 		if (unlikely(!m->next))
215 			return -1;
216 
217 		m = m->next;
218 
219 		/* Headroom is not needed in chained mbufs. */
220 		rte_pktmbuf_prepend(m, rte_pktmbuf_headroom(m));
221 		m->pkt_len = 0;
222 		m->data_len = 0;
223 
224 		/* Copy next segment. */
225 		len = RTE_MIN(rte_pktmbuf_tailroom(m), data_len);
226 		rte_memcpy(rte_pktmbuf_append(m, len), data, len);
227 
228 		mbuf->nb_segs++;
229 		data_len -= len;
230 		data += len;
231 	}
232 
233 	return mbuf->nb_segs;
234 }
235 
236 static uint16_t
237 eth_pcap_rx_infinite(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
238 {
239 	int i;
240 	struct pcap_rx_queue *pcap_q = queue;
241 	uint32_t rx_bytes = 0;
242 
243 	if (unlikely(nb_pkts == 0))
244 		return 0;
245 
246 	if (rte_pktmbuf_alloc_bulk(pcap_q->mb_pool, bufs, nb_pkts) != 0)
247 		return 0;
248 
249 	for (i = 0; i < nb_pkts; i++) {
250 		struct rte_mbuf *pcap_buf;
251 		int err = rte_ring_dequeue(pcap_q->pkts, (void **)&pcap_buf);
252 		if (err)
253 			return i;
254 
255 		rte_memcpy(rte_pktmbuf_mtod(bufs[i], void *),
256 				rte_pktmbuf_mtod(pcap_buf, void *),
257 				pcap_buf->data_len);
258 		bufs[i]->data_len = pcap_buf->data_len;
259 		bufs[i]->pkt_len = pcap_buf->pkt_len;
260 		bufs[i]->port = pcap_q->port_id;
261 		rx_bytes += pcap_buf->data_len;
262 
263 		/* Enqueue packet back on ring to allow infinite rx. */
264 		rte_ring_enqueue(pcap_q->pkts, pcap_buf);
265 	}
266 
267 	pcap_q->rx_stat.pkts += i;
268 	pcap_q->rx_stat.bytes += rx_bytes;
269 
270 	return i;
271 }
272 
273 static uint16_t
274 eth_pcap_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
275 {
276 	unsigned int i;
277 	struct pcap_pkthdr header;
278 	struct pmd_process_private *pp;
279 	const u_char *packet;
280 	struct rte_mbuf *mbuf;
281 	struct pcap_rx_queue *pcap_q = queue;
282 	uint16_t num_rx = 0;
283 	uint32_t rx_bytes = 0;
284 	pcap_t *pcap;
285 
286 	pp = rte_eth_devices[pcap_q->port_id].process_private;
287 	pcap = pp->rx_pcap[pcap_q->queue_id];
288 
289 	if (unlikely(pcap == NULL || nb_pkts == 0))
290 		return 0;
291 
292 	/* Reads the given number of packets from the pcap file one by one
293 	 * and copies the packet data into a newly allocated mbuf to return.
294 	 */
295 	for (i = 0; i < nb_pkts; i++) {
296 		/* Get the next PCAP packet */
297 		packet = pcap_next(pcap, &header);
298 		if (unlikely(packet == NULL))
299 			break;
300 
301 		mbuf = rte_pktmbuf_alloc(pcap_q->mb_pool);
302 		if (unlikely(mbuf == NULL)) {
303 			pcap_q->rx_stat.rx_nombuf++;
304 			break;
305 		}
306 
307 		if (header.caplen <= rte_pktmbuf_tailroom(mbuf)) {
308 			/* pcap packet will fit in the mbuf, can copy it */
309 			rte_memcpy(rte_pktmbuf_mtod(mbuf, void *), packet,
310 					header.caplen);
311 			mbuf->data_len = (uint16_t)header.caplen;
312 		} else {
313 			/* Try read jumbo frame into multi mbufs. */
314 			if (unlikely(eth_pcap_rx_jumbo(pcap_q->mb_pool,
315 						       mbuf,
316 						       packet,
317 						       header.caplen) == -1)) {
318 				pcap_q->rx_stat.err_pkts++;
319 				rte_pktmbuf_free(mbuf);
320 				break;
321 			}
322 		}
323 
324 		mbuf->pkt_len = (uint16_t)header.caplen;
325 		*RTE_MBUF_DYNFIELD(mbuf, timestamp_dynfield_offset,
326 			rte_mbuf_timestamp_t *) =
327 				(uint64_t)header.ts.tv_sec * 1000000 +
328 				header.ts.tv_usec;
329 		mbuf->ol_flags |= timestamp_rx_dynflag;
330 		mbuf->port = pcap_q->port_id;
331 		bufs[num_rx] = mbuf;
332 		num_rx++;
333 		rx_bytes += header.caplen;
334 	}
335 	pcap_q->rx_stat.pkts += num_rx;
336 	pcap_q->rx_stat.bytes += rx_bytes;
337 
338 	return num_rx;
339 }
340 
341 static uint16_t
342 eth_null_rx(void *queue __rte_unused,
343 		struct rte_mbuf **bufs __rte_unused,
344 		uint16_t nb_pkts __rte_unused)
345 {
346 	return 0;
347 }
348 
349 #define NSEC_PER_SEC	1000000000L
350 
351 /*
352  * This function stores nanoseconds in `tv_usec` field of `struct timeval`,
353  * because `ts` goes directly to nanosecond-precision dump.
354  */
355 static inline void
356 calculate_timestamp(struct timeval *ts) {
357 	uint64_t cycles;
358 	struct timespec cur_time;
359 
360 	cycles = rte_get_timer_cycles() - start_cycles;
361 	cur_time.tv_sec = cycles / hz;
362 	cur_time.tv_nsec = (cycles % hz) * NSEC_PER_SEC / hz;
363 
364 	ts->tv_sec = start_time.tv_sec + cur_time.tv_sec;
365 	ts->tv_usec = start_time.tv_nsec + cur_time.tv_nsec;
366 	if (ts->tv_usec >= NSEC_PER_SEC) {
367 		ts->tv_usec -= NSEC_PER_SEC;
368 		ts->tv_sec += 1;
369 	}
370 }
371 
372 /*
373  * Callback to handle writing packets to a pcap file.
374  */
375 static uint16_t
376 eth_pcap_tx_dumper(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
377 {
378 	unsigned int i;
379 	struct rte_mbuf *mbuf;
380 	struct pmd_process_private *pp;
381 	struct pcap_tx_queue *dumper_q = queue;
382 	uint16_t num_tx = 0;
383 	uint32_t tx_bytes = 0;
384 	struct pcap_pkthdr header;
385 	pcap_dumper_t *dumper;
386 	unsigned char temp_data[RTE_ETH_PCAP_SNAPLEN];
387 	size_t len, caplen;
388 
389 	pp = rte_eth_devices[dumper_q->port_id].process_private;
390 	dumper = pp->tx_dumper[dumper_q->queue_id];
391 
392 	if (dumper == NULL || nb_pkts == 0)
393 		return 0;
394 
395 	/* writes the nb_pkts packets to the previously opened pcap file
396 	 * dumper */
397 	for (i = 0; i < nb_pkts; i++) {
398 		mbuf = bufs[i];
399 		len = caplen = rte_pktmbuf_pkt_len(mbuf);
400 		if (unlikely(!rte_pktmbuf_is_contiguous(mbuf) &&
401 				len > sizeof(temp_data))) {
402 			caplen = sizeof(temp_data);
403 		}
404 
405 		calculate_timestamp(&header.ts);
406 		header.len = len;
407 		header.caplen = caplen;
408 		/* rte_pktmbuf_read() returns a pointer to the data directly
409 		 * in the mbuf (when the mbuf is contiguous) or, otherwise,
410 		 * a pointer to temp_data after copying into it.
411 		 */
412 		pcap_dump((u_char *)dumper, &header,
413 			rte_pktmbuf_read(mbuf, 0, caplen, temp_data));
414 
415 		num_tx++;
416 		tx_bytes += caplen;
417 		rte_pktmbuf_free(mbuf);
418 	}
419 
420 	/*
421 	 * Since there's no place to hook a callback when the forwarding
422 	 * process stops and to make sure the pcap file is actually written,
423 	 * we flush the pcap dumper within each burst.
424 	 */
425 	pcap_dump_flush(dumper);
426 	dumper_q->tx_stat.pkts += num_tx;
427 	dumper_q->tx_stat.bytes += tx_bytes;
428 	dumper_q->tx_stat.err_pkts += nb_pkts - num_tx;
429 
430 	return nb_pkts;
431 }
432 
433 /*
434  * Callback to handle dropping packets in the infinite rx case.
435  */
436 static uint16_t
437 eth_tx_drop(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
438 {
439 	unsigned int i;
440 	uint32_t tx_bytes = 0;
441 	struct pcap_tx_queue *tx_queue = queue;
442 
443 	if (unlikely(nb_pkts == 0))
444 		return 0;
445 
446 	for (i = 0; i < nb_pkts; i++) {
447 		tx_bytes += bufs[i]->pkt_len;
448 		rte_pktmbuf_free(bufs[i]);
449 	}
450 
451 	tx_queue->tx_stat.pkts += nb_pkts;
452 	tx_queue->tx_stat.bytes += tx_bytes;
453 
454 	return i;
455 }
456 
457 /*
458  * Callback to handle sending packets through a real NIC.
459  */
460 static uint16_t
461 eth_pcap_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
462 {
463 	unsigned int i;
464 	int ret;
465 	struct rte_mbuf *mbuf;
466 	struct pmd_process_private *pp;
467 	struct pcap_tx_queue *tx_queue = queue;
468 	uint16_t num_tx = 0;
469 	uint32_t tx_bytes = 0;
470 	pcap_t *pcap;
471 	unsigned char temp_data[RTE_ETH_PCAP_SNAPLEN];
472 	size_t len;
473 
474 	pp = rte_eth_devices[tx_queue->port_id].process_private;
475 	pcap = pp->tx_pcap[tx_queue->queue_id];
476 
477 	if (unlikely(nb_pkts == 0 || pcap == NULL))
478 		return 0;
479 
480 	for (i = 0; i < nb_pkts; i++) {
481 		mbuf = bufs[i];
482 		len = rte_pktmbuf_pkt_len(mbuf);
483 		if (unlikely(!rte_pktmbuf_is_contiguous(mbuf) &&
484 				len > sizeof(temp_data))) {
485 			PMD_LOG(ERR,
486 				"Dropping multi segment PCAP packet. Size (%zd) > max size (%zd).",
487 				len, sizeof(temp_data));
488 			rte_pktmbuf_free(mbuf);
489 			continue;
490 		}
491 
492 		/* rte_pktmbuf_read() returns a pointer to the data directly
493 		 * in the mbuf (when the mbuf is contiguous) or, otherwise,
494 		 * a pointer to temp_data after copying into it.
495 		 */
496 		ret = pcap_sendpacket(pcap,
497 			rte_pktmbuf_read(mbuf, 0, len, temp_data), len);
498 		if (unlikely(ret != 0))
499 			break;
500 		num_tx++;
501 		tx_bytes += len;
502 		rte_pktmbuf_free(mbuf);
503 	}
504 
505 	tx_queue->tx_stat.pkts += num_tx;
506 	tx_queue->tx_stat.bytes += tx_bytes;
507 	tx_queue->tx_stat.err_pkts += i - num_tx;
508 
509 	return i;
510 }
511 
512 /*
513  * pcap_open_live wrapper function
514  */
515 static inline int
516 open_iface_live(const char *iface, pcap_t **pcap) {
517 	*pcap = pcap_open_live(iface, RTE_ETH_PCAP_SNAPLEN,
518 			RTE_ETH_PCAP_PROMISC, RTE_ETH_PCAP_TIMEOUT, errbuf);
519 
520 	if (*pcap == NULL) {
521 		PMD_LOG(ERR, "Couldn't open %s: %s", iface, errbuf);
522 		return -1;
523 	}
524 
525 	return 0;
526 }
527 
528 static int
529 open_single_iface(const char *iface, pcap_t **pcap)
530 {
531 	if (open_iface_live(iface, pcap) < 0) {
532 		PMD_LOG(ERR, "Couldn't open interface %s", iface);
533 		return -1;
534 	}
535 
536 	return 0;
537 }
538 
539 static int
540 open_single_tx_pcap(const char *pcap_filename, pcap_dumper_t **dumper)
541 {
542 	pcap_t *tx_pcap;
543 
544 	/*
545 	 * We need to create a dummy empty pcap_t to use it
546 	 * with pcap_dump_open(). We create big enough an Ethernet
547 	 * pcap holder.
548 	 */
549 	tx_pcap = pcap_open_dead_with_tstamp_precision(DLT_EN10MB,
550 			RTE_ETH_PCAP_SNAPSHOT_LEN, PCAP_TSTAMP_PRECISION_NANO);
551 	if (tx_pcap == NULL) {
552 		PMD_LOG(ERR, "Couldn't create dead pcap");
553 		return -1;
554 	}
555 
556 	/* The dumper is created using the previous pcap_t reference */
557 	*dumper = pcap_dump_open(tx_pcap, pcap_filename);
558 	if (*dumper == NULL) {
559 		pcap_close(tx_pcap);
560 		PMD_LOG(ERR, "Couldn't open %s for writing.",
561 			pcap_filename);
562 		return -1;
563 	}
564 
565 	pcap_close(tx_pcap);
566 	return 0;
567 }
568 
569 static int
570 open_single_rx_pcap(const char *pcap_filename, pcap_t **pcap)
571 {
572 	*pcap = pcap_open_offline(pcap_filename, errbuf);
573 	if (*pcap == NULL) {
574 		PMD_LOG(ERR, "Couldn't open %s: %s", pcap_filename,
575 			errbuf);
576 		return -1;
577 	}
578 
579 	return 0;
580 }
581 
582 static uint64_t
583 count_packets_in_pcap(pcap_t **pcap, struct pcap_rx_queue *pcap_q)
584 {
585 	const u_char *packet;
586 	struct pcap_pkthdr header;
587 	uint64_t pcap_pkt_count = 0;
588 
589 	while ((packet = pcap_next(*pcap, &header)))
590 		pcap_pkt_count++;
591 
592 	/* The pcap is reopened so it can be used as normal later. */
593 	pcap_close(*pcap);
594 	*pcap = NULL;
595 	open_single_rx_pcap(pcap_q->name, pcap);
596 
597 	return pcap_pkt_count;
598 }
599 
600 static int
601 eth_dev_start(struct rte_eth_dev *dev)
602 {
603 	unsigned int i;
604 	struct pmd_internals *internals = dev->data->dev_private;
605 	struct pmd_process_private *pp = dev->process_private;
606 	struct pcap_tx_queue *tx;
607 	struct pcap_rx_queue *rx;
608 
609 	/* Special iface case. Single pcap is open and shared between tx/rx. */
610 	if (internals->single_iface) {
611 		tx = &internals->tx_queue[0];
612 		rx = &internals->rx_queue[0];
613 
614 		if (!pp->tx_pcap[0] &&
615 			strcmp(tx->type, ETH_PCAP_IFACE_ARG) == 0) {
616 			if (open_single_iface(tx->name, &pp->tx_pcap[0]) < 0)
617 				return -1;
618 			pp->rx_pcap[0] = pp->tx_pcap[0];
619 		}
620 
621 		goto status_up;
622 	}
623 
624 	/* If not open already, open tx pcaps/dumpers */
625 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
626 		tx = &internals->tx_queue[i];
627 
628 		if (!pp->tx_dumper[i] &&
629 				strcmp(tx->type, ETH_PCAP_TX_PCAP_ARG) == 0) {
630 			if (open_single_tx_pcap(tx->name,
631 				&pp->tx_dumper[i]) < 0)
632 				return -1;
633 		} else if (!pp->tx_pcap[i] &&
634 				strcmp(tx->type, ETH_PCAP_TX_IFACE_ARG) == 0) {
635 			if (open_single_iface(tx->name, &pp->tx_pcap[i]) < 0)
636 				return -1;
637 		}
638 	}
639 
640 	/* If not open already, open rx pcaps */
641 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
642 		rx = &internals->rx_queue[i];
643 
644 		if (pp->rx_pcap[i] != NULL)
645 			continue;
646 
647 		if (strcmp(rx->type, ETH_PCAP_RX_PCAP_ARG) == 0) {
648 			if (open_single_rx_pcap(rx->name, &pp->rx_pcap[i]) < 0)
649 				return -1;
650 		} else if (strcmp(rx->type, ETH_PCAP_RX_IFACE_ARG) == 0) {
651 			if (open_single_iface(rx->name, &pp->rx_pcap[i]) < 0)
652 				return -1;
653 		}
654 	}
655 
656 status_up:
657 	for (i = 0; i < dev->data->nb_rx_queues; i++)
658 		dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
659 
660 	for (i = 0; i < dev->data->nb_tx_queues; i++)
661 		dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
662 
663 	dev->data->dev_link.link_status = RTE_ETH_LINK_UP;
664 
665 	return 0;
666 }
667 
668 /*
669  * This function gets called when the current port gets stopped.
670  * Is the only place for us to close all the tx streams dumpers.
671  * If not called the dumpers will be flushed within each tx burst.
672  */
673 static int
674 eth_dev_stop(struct rte_eth_dev *dev)
675 {
676 	unsigned int i;
677 	struct pmd_internals *internals = dev->data->dev_private;
678 	struct pmd_process_private *pp = dev->process_private;
679 
680 	/* Special iface case. Single pcap is open and shared between tx/rx. */
681 	if (internals->single_iface) {
682 		queue_missed_stat_on_stop_update(dev, 0);
683 		if (pp->tx_pcap[0] != NULL) {
684 			pcap_close(pp->tx_pcap[0]);
685 			pp->tx_pcap[0] = NULL;
686 			pp->rx_pcap[0] = NULL;
687 		}
688 		goto status_down;
689 	}
690 
691 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
692 		if (pp->tx_dumper[i] != NULL) {
693 			pcap_dump_close(pp->tx_dumper[i]);
694 			pp->tx_dumper[i] = NULL;
695 		}
696 
697 		if (pp->tx_pcap[i] != NULL) {
698 			pcap_close(pp->tx_pcap[i]);
699 			pp->tx_pcap[i] = NULL;
700 		}
701 	}
702 
703 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
704 		if (pp->rx_pcap[i] != NULL) {
705 			queue_missed_stat_on_stop_update(dev, i);
706 			pcap_close(pp->rx_pcap[i]);
707 			pp->rx_pcap[i] = NULL;
708 		}
709 	}
710 
711 status_down:
712 	for (i = 0; i < dev->data->nb_rx_queues; i++)
713 		dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
714 
715 	for (i = 0; i < dev->data->nb_tx_queues; i++)
716 		dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
717 
718 	dev->data->dev_link.link_status = RTE_ETH_LINK_DOWN;
719 
720 	return 0;
721 }
722 
723 static int
724 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
725 {
726 	return 0;
727 }
728 
729 static int
730 eth_dev_info(struct rte_eth_dev *dev,
731 		struct rte_eth_dev_info *dev_info)
732 {
733 	struct pmd_internals *internals = dev->data->dev_private;
734 
735 	dev_info->if_index = internals->if_index;
736 	dev_info->max_mac_addrs = 1;
737 	dev_info->max_rx_pktlen = (uint32_t) -1;
738 	dev_info->max_rx_queues = dev->data->nb_rx_queues;
739 	dev_info->max_tx_queues = dev->data->nb_tx_queues;
740 	dev_info->min_rx_bufsize = 0;
741 
742 	return 0;
743 }
744 
745 static int
746 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
747 {
748 	unsigned int i;
749 	unsigned long rx_packets_total = 0, rx_bytes_total = 0;
750 	unsigned long rx_missed_total = 0;
751 	unsigned long rx_nombuf_total = 0, rx_err_total = 0;
752 	unsigned long tx_packets_total = 0, tx_bytes_total = 0;
753 	unsigned long tx_packets_err_total = 0;
754 	const struct pmd_internals *internal = dev->data->dev_private;
755 
756 	for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
757 			i < dev->data->nb_rx_queues; i++) {
758 		stats->q_ipackets[i] = internal->rx_queue[i].rx_stat.pkts;
759 		stats->q_ibytes[i] = internal->rx_queue[i].rx_stat.bytes;
760 		rx_nombuf_total += internal->rx_queue[i].rx_stat.rx_nombuf;
761 		rx_err_total += internal->rx_queue[i].rx_stat.err_pkts;
762 		rx_packets_total += stats->q_ipackets[i];
763 		rx_bytes_total += stats->q_ibytes[i];
764 		rx_missed_total += queue_missed_stat_get(dev, i);
765 	}
766 
767 	for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
768 			i < dev->data->nb_tx_queues; i++) {
769 		stats->q_opackets[i] = internal->tx_queue[i].tx_stat.pkts;
770 		stats->q_obytes[i] = internal->tx_queue[i].tx_stat.bytes;
771 		tx_packets_total += stats->q_opackets[i];
772 		tx_bytes_total += stats->q_obytes[i];
773 		tx_packets_err_total += internal->tx_queue[i].tx_stat.err_pkts;
774 	}
775 
776 	stats->ipackets = rx_packets_total;
777 	stats->ibytes = rx_bytes_total;
778 	stats->imissed = rx_missed_total;
779 	stats->ierrors = rx_err_total;
780 	stats->rx_nombuf = rx_nombuf_total;
781 	stats->opackets = tx_packets_total;
782 	stats->obytes = tx_bytes_total;
783 	stats->oerrors = tx_packets_err_total;
784 
785 	return 0;
786 }
787 
788 static int
789 eth_stats_reset(struct rte_eth_dev *dev)
790 {
791 	unsigned int i;
792 	struct pmd_internals *internal = dev->data->dev_private;
793 
794 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
795 		internal->rx_queue[i].rx_stat.pkts = 0;
796 		internal->rx_queue[i].rx_stat.bytes = 0;
797 		internal->rx_queue[i].rx_stat.err_pkts = 0;
798 		internal->rx_queue[i].rx_stat.rx_nombuf = 0;
799 		queue_missed_stat_reset(dev, i);
800 	}
801 
802 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
803 		internal->tx_queue[i].tx_stat.pkts = 0;
804 		internal->tx_queue[i].tx_stat.bytes = 0;
805 		internal->tx_queue[i].tx_stat.err_pkts = 0;
806 	}
807 
808 	return 0;
809 }
810 
811 static inline void
812 infinite_rx_ring_free(struct rte_ring *pkts)
813 {
814 	struct rte_mbuf *bufs;
815 
816 	while (!rte_ring_dequeue(pkts, (void **)&bufs))
817 		rte_pktmbuf_free(bufs);
818 
819 	rte_ring_free(pkts);
820 }
821 
822 static int
823 eth_dev_close(struct rte_eth_dev *dev)
824 {
825 	unsigned int i;
826 	struct pmd_internals *internals = dev->data->dev_private;
827 
828 	PMD_LOG(INFO, "Closing pcap ethdev on NUMA socket %d",
829 			rte_socket_id());
830 
831 	eth_dev_stop(dev);
832 
833 	rte_free(dev->process_private);
834 
835 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
836 		return 0;
837 
838 	/* Device wide flag, but cleanup must be performed per queue. */
839 	if (internals->infinite_rx) {
840 		for (i = 0; i < dev->data->nb_rx_queues; i++) {
841 			struct pcap_rx_queue *pcap_q = &internals->rx_queue[i];
842 
843 			/*
844 			 * 'pcap_q->pkts' can be NULL if 'eth_dev_close()'
845 			 * called before 'eth_rx_queue_setup()' has been called
846 			 */
847 			if (pcap_q->pkts == NULL)
848 				continue;
849 
850 			infinite_rx_ring_free(pcap_q->pkts);
851 		}
852 	}
853 
854 	if (internals->phy_mac == 0)
855 		/* not dynamically allocated, must not be freed */
856 		dev->data->mac_addrs = NULL;
857 
858 	return 0;
859 }
860 
861 static int
862 eth_link_update(struct rte_eth_dev *dev __rte_unused,
863 		int wait_to_complete __rte_unused)
864 {
865 	return 0;
866 }
867 
868 static int
869 eth_rx_queue_setup(struct rte_eth_dev *dev,
870 		uint16_t rx_queue_id,
871 		uint16_t nb_rx_desc __rte_unused,
872 		unsigned int socket_id __rte_unused,
873 		const struct rte_eth_rxconf *rx_conf __rte_unused,
874 		struct rte_mempool *mb_pool)
875 {
876 	struct pmd_internals *internals = dev->data->dev_private;
877 	struct pcap_rx_queue *pcap_q = &internals->rx_queue[rx_queue_id];
878 
879 	pcap_q->mb_pool = mb_pool;
880 	pcap_q->port_id = dev->data->port_id;
881 	pcap_q->queue_id = rx_queue_id;
882 	dev->data->rx_queues[rx_queue_id] = pcap_q;
883 
884 	if (internals->infinite_rx) {
885 		struct pmd_process_private *pp;
886 		char ring_name[RTE_RING_NAMESIZE];
887 		static uint32_t ring_number;
888 		uint64_t pcap_pkt_count = 0;
889 		struct rte_mbuf *bufs[1];
890 		pcap_t **pcap;
891 
892 		pp = rte_eth_devices[pcap_q->port_id].process_private;
893 		pcap = &pp->rx_pcap[pcap_q->queue_id];
894 
895 		if (unlikely(*pcap == NULL))
896 			return -ENOENT;
897 
898 		pcap_pkt_count = count_packets_in_pcap(pcap, pcap_q);
899 
900 		snprintf(ring_name, sizeof(ring_name), "PCAP_RING%" PRIu32,
901 				ring_number);
902 
903 		pcap_q->pkts = rte_ring_create(ring_name,
904 				rte_align64pow2(pcap_pkt_count + 1), 0,
905 				RING_F_SP_ENQ | RING_F_SC_DEQ);
906 		ring_number++;
907 		if (!pcap_q->pkts)
908 			return -ENOENT;
909 
910 		/* Fill ring with packets from PCAP file one by one. */
911 		while (eth_pcap_rx(pcap_q, bufs, 1)) {
912 			/* Check for multiseg mbufs. */
913 			if (bufs[0]->nb_segs != 1) {
914 				infinite_rx_ring_free(pcap_q->pkts);
915 				PMD_LOG(ERR,
916 					"Multiseg mbufs are not supported in infinite_rx mode.");
917 				return -EINVAL;
918 			}
919 
920 			rte_ring_enqueue_bulk(pcap_q->pkts,
921 					(void * const *)bufs, 1, NULL);
922 		}
923 
924 		if (rte_ring_count(pcap_q->pkts) < pcap_pkt_count) {
925 			infinite_rx_ring_free(pcap_q->pkts);
926 			PMD_LOG(ERR,
927 				"Not enough mbufs to accommodate packets in pcap file. "
928 				"At least %" PRIu64 " mbufs per queue is required.",
929 				pcap_pkt_count);
930 			return -EINVAL;
931 		}
932 
933 		/*
934 		 * Reset the stats for this queue since eth_pcap_rx calls above
935 		 * didn't result in the application receiving packets.
936 		 */
937 		pcap_q->rx_stat.pkts = 0;
938 		pcap_q->rx_stat.bytes = 0;
939 	}
940 
941 	return 0;
942 }
943 
944 static int
945 eth_tx_queue_setup(struct rte_eth_dev *dev,
946 		uint16_t tx_queue_id,
947 		uint16_t nb_tx_desc __rte_unused,
948 		unsigned int socket_id __rte_unused,
949 		const struct rte_eth_txconf *tx_conf __rte_unused)
950 {
951 	struct pmd_internals *internals = dev->data->dev_private;
952 	struct pcap_tx_queue *pcap_q = &internals->tx_queue[tx_queue_id];
953 
954 	pcap_q->port_id = dev->data->port_id;
955 	pcap_q->queue_id = tx_queue_id;
956 	dev->data->tx_queues[tx_queue_id] = pcap_q;
957 
958 	return 0;
959 }
960 
961 static int
962 eth_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
963 {
964 	dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
965 
966 	return 0;
967 }
968 
969 static int
970 eth_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
971 {
972 	dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
973 
974 	return 0;
975 }
976 
977 static int
978 eth_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
979 {
980 	dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
981 
982 	return 0;
983 }
984 
985 static int
986 eth_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
987 {
988 	dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
989 
990 	return 0;
991 }
992 
993 static const struct eth_dev_ops ops = {
994 	.dev_start = eth_dev_start,
995 	.dev_stop = eth_dev_stop,
996 	.dev_close = eth_dev_close,
997 	.dev_configure = eth_dev_configure,
998 	.dev_infos_get = eth_dev_info,
999 	.rx_queue_setup = eth_rx_queue_setup,
1000 	.tx_queue_setup = eth_tx_queue_setup,
1001 	.rx_queue_start = eth_rx_queue_start,
1002 	.tx_queue_start = eth_tx_queue_start,
1003 	.rx_queue_stop = eth_rx_queue_stop,
1004 	.tx_queue_stop = eth_tx_queue_stop,
1005 	.link_update = eth_link_update,
1006 	.stats_get = eth_stats_get,
1007 	.stats_reset = eth_stats_reset,
1008 };
1009 
1010 static int
1011 add_queue(struct pmd_devargs *pmd, const char *name, const char *type,
1012 		pcap_t *pcap, pcap_dumper_t *dumper)
1013 {
1014 	if (pmd->num_of_queue >= RTE_PMD_PCAP_MAX_QUEUES)
1015 		return -1;
1016 	if (pcap)
1017 		pmd->queue[pmd->num_of_queue].pcap = pcap;
1018 	if (dumper)
1019 		pmd->queue[pmd->num_of_queue].dumper = dumper;
1020 	pmd->queue[pmd->num_of_queue].name = name;
1021 	pmd->queue[pmd->num_of_queue].type = type;
1022 	pmd->num_of_queue++;
1023 	return 0;
1024 }
1025 
1026 /*
1027  * Function handler that opens the pcap file for reading a stores a
1028  * reference of it for use it later on.
1029  */
1030 static int
1031 open_rx_pcap(const char *key, const char *value, void *extra_args)
1032 {
1033 	const char *pcap_filename = value;
1034 	struct pmd_devargs *rx = extra_args;
1035 	pcap_t *pcap = NULL;
1036 
1037 	if (open_single_rx_pcap(pcap_filename, &pcap) < 0)
1038 		return -1;
1039 
1040 	if (add_queue(rx, pcap_filename, key, pcap, NULL) < 0) {
1041 		pcap_close(pcap);
1042 		return -1;
1043 	}
1044 
1045 	return 0;
1046 }
1047 
1048 /*
1049  * Opens a pcap file for writing and stores a reference to it
1050  * for use it later on.
1051  */
1052 static int
1053 open_tx_pcap(const char *key, const char *value, void *extra_args)
1054 {
1055 	const char *pcap_filename = value;
1056 	struct pmd_devargs *dumpers = extra_args;
1057 	pcap_dumper_t *dumper;
1058 
1059 	if (open_single_tx_pcap(pcap_filename, &dumper) < 0)
1060 		return -1;
1061 
1062 	if (add_queue(dumpers, pcap_filename, key, NULL, dumper) < 0) {
1063 		pcap_dump_close(dumper);
1064 		return -1;
1065 	}
1066 
1067 	return 0;
1068 }
1069 
1070 /*
1071  * Opens an interface for reading and writing
1072  */
1073 static inline int
1074 open_rx_tx_iface(const char *key, const char *value, void *extra_args)
1075 {
1076 	const char *iface = value;
1077 	struct pmd_devargs *tx = extra_args;
1078 	pcap_t *pcap = NULL;
1079 
1080 	if (open_single_iface(iface, &pcap) < 0)
1081 		return -1;
1082 
1083 	tx->queue[0].pcap = pcap;
1084 	tx->queue[0].name = iface;
1085 	tx->queue[0].type = key;
1086 
1087 	return 0;
1088 }
1089 
1090 static inline int
1091 set_iface_direction(const char *iface, pcap_t *pcap,
1092 		pcap_direction_t direction)
1093 {
1094 	const char *direction_str = (direction == PCAP_D_IN) ? "IN" : "OUT";
1095 	if (pcap_setdirection(pcap, direction) < 0) {
1096 		PMD_LOG(ERR, "Setting %s pcap direction %s failed - %s\n",
1097 				iface, direction_str, pcap_geterr(pcap));
1098 		return -1;
1099 	}
1100 	PMD_LOG(INFO, "Setting %s pcap direction %s\n",
1101 			iface, direction_str);
1102 	return 0;
1103 }
1104 
1105 static inline int
1106 open_iface(const char *key, const char *value, void *extra_args)
1107 {
1108 	const char *iface = value;
1109 	struct pmd_devargs *pmd = extra_args;
1110 	pcap_t *pcap = NULL;
1111 
1112 	if (open_single_iface(iface, &pcap) < 0)
1113 		return -1;
1114 	if (add_queue(pmd, iface, key, pcap, NULL) < 0) {
1115 		pcap_close(pcap);
1116 		return -1;
1117 	}
1118 
1119 	return 0;
1120 }
1121 
1122 /*
1123  * Opens a NIC for reading packets from it
1124  */
1125 static inline int
1126 open_rx_iface(const char *key, const char *value, void *extra_args)
1127 {
1128 	int ret = open_iface(key, value, extra_args);
1129 	if (ret < 0)
1130 		return ret;
1131 	if (strcmp(key, ETH_PCAP_RX_IFACE_IN_ARG) == 0) {
1132 		struct pmd_devargs *pmd = extra_args;
1133 		unsigned int qid = pmd->num_of_queue - 1;
1134 
1135 		set_iface_direction(pmd->queue[qid].name,
1136 				pmd->queue[qid].pcap,
1137 				PCAP_D_IN);
1138 	}
1139 
1140 	return 0;
1141 }
1142 
1143 static inline int
1144 rx_iface_args_process(const char *key, const char *value, void *extra_args)
1145 {
1146 	if (strcmp(key, ETH_PCAP_RX_IFACE_ARG) == 0 ||
1147 			strcmp(key, ETH_PCAP_RX_IFACE_IN_ARG) == 0)
1148 		return open_rx_iface(key, value, extra_args);
1149 
1150 	return 0;
1151 }
1152 
1153 /*
1154  * Opens a NIC for writing packets to it
1155  */
1156 static int
1157 open_tx_iface(const char *key, const char *value, void *extra_args)
1158 {
1159 	return open_iface(key, value, extra_args);
1160 }
1161 
1162 static int
1163 select_phy_mac(const char *key __rte_unused, const char *value,
1164 		void *extra_args)
1165 {
1166 	if (extra_args) {
1167 		const int phy_mac = atoi(value);
1168 		int *enable_phy_mac = extra_args;
1169 
1170 		if (phy_mac)
1171 			*enable_phy_mac = 1;
1172 	}
1173 	return 0;
1174 }
1175 
1176 static int
1177 get_infinite_rx_arg(const char *key __rte_unused,
1178 		const char *value, void *extra_args)
1179 {
1180 	if (extra_args) {
1181 		const int infinite_rx = atoi(value);
1182 		int *enable_infinite_rx = extra_args;
1183 
1184 		if (infinite_rx > 0)
1185 			*enable_infinite_rx = 1;
1186 	}
1187 	return 0;
1188 }
1189 
1190 static int
1191 pmd_init_internals(struct rte_vdev_device *vdev,
1192 		const unsigned int nb_rx_queues,
1193 		const unsigned int nb_tx_queues,
1194 		struct pmd_internals **internals,
1195 		struct rte_eth_dev **eth_dev)
1196 {
1197 	struct rte_eth_dev_data *data;
1198 	struct pmd_process_private *pp;
1199 	unsigned int numa_node = vdev->device.numa_node;
1200 
1201 	PMD_LOG(INFO, "Creating pcap-backed ethdev on numa socket %d",
1202 		numa_node);
1203 
1204 	pp = (struct pmd_process_private *)
1205 		rte_zmalloc(NULL, sizeof(struct pmd_process_private),
1206 				RTE_CACHE_LINE_SIZE);
1207 
1208 	if (pp == NULL) {
1209 		PMD_LOG(ERR,
1210 			"Failed to allocate memory for process private");
1211 		return -1;
1212 	}
1213 
1214 	/* reserve an ethdev entry */
1215 	*eth_dev = rte_eth_vdev_allocate(vdev, sizeof(**internals));
1216 	if (!(*eth_dev)) {
1217 		rte_free(pp);
1218 		return -1;
1219 	}
1220 	(*eth_dev)->process_private = pp;
1221 	/* now put it all together
1222 	 * - store queue data in internals,
1223 	 * - store numa_node info in eth_dev
1224 	 * - point eth_dev_data to internals
1225 	 * - and point eth_dev structure to new eth_dev_data structure
1226 	 */
1227 	*internals = (*eth_dev)->data->dev_private;
1228 	/*
1229 	 * Interface MAC = 02:70:63:61:70:<iface_idx>
1230 	 * derived from: 'locally administered':'p':'c':'a':'p':'iface_idx'
1231 	 * where the middle 4 characters are converted to hex.
1232 	 */
1233 	(*internals)->eth_addr = (struct rte_ether_addr) {
1234 		.addr_bytes = { 0x02, 0x70, 0x63, 0x61, 0x70, iface_idx++ }
1235 	};
1236 	(*internals)->phy_mac = 0;
1237 	data = (*eth_dev)->data;
1238 	data->nb_rx_queues = (uint16_t)nb_rx_queues;
1239 	data->nb_tx_queues = (uint16_t)nb_tx_queues;
1240 	data->dev_link = pmd_link;
1241 	data->mac_addrs = &(*internals)->eth_addr;
1242 	data->promiscuous = 1;
1243 	data->all_multicast = 1;
1244 	data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
1245 
1246 	/*
1247 	 * NOTE: we'll replace the data element, of originally allocated
1248 	 * eth_dev so the rings are local per-process
1249 	 */
1250 	(*eth_dev)->dev_ops = &ops;
1251 
1252 	strlcpy((*internals)->devargs, rte_vdev_device_args(vdev),
1253 			ETH_PCAP_ARG_MAXLEN);
1254 
1255 	return 0;
1256 }
1257 
1258 static int
1259 eth_pcap_update_mac(const char *if_name, struct rte_eth_dev *eth_dev,
1260 		const unsigned int numa_node)
1261 {
1262 	void *mac_addrs;
1263 	struct rte_ether_addr mac;
1264 
1265 	if (osdep_iface_mac_get(if_name, &mac) < 0)
1266 		return -1;
1267 
1268 	mac_addrs = rte_zmalloc_socket(NULL, RTE_ETHER_ADDR_LEN, 0, numa_node);
1269 	if (mac_addrs == NULL)
1270 		return -1;
1271 
1272 	PMD_LOG(INFO, "Setting phy MAC for %s", if_name);
1273 	rte_memcpy(mac_addrs, mac.addr_bytes, RTE_ETHER_ADDR_LEN);
1274 	eth_dev->data->mac_addrs = mac_addrs;
1275 	return 0;
1276 }
1277 
1278 static int
1279 eth_from_pcaps_common(struct rte_vdev_device *vdev,
1280 		struct pmd_devargs_all *devargs_all,
1281 		struct pmd_internals **internals, struct rte_eth_dev **eth_dev)
1282 {
1283 	struct pmd_process_private *pp;
1284 	struct pmd_devargs *rx_queues = &devargs_all->rx_queues;
1285 	struct pmd_devargs *tx_queues = &devargs_all->tx_queues;
1286 	const unsigned int nb_rx_queues = rx_queues->num_of_queue;
1287 	const unsigned int nb_tx_queues = tx_queues->num_of_queue;
1288 	unsigned int i;
1289 
1290 	if (pmd_init_internals(vdev, nb_rx_queues, nb_tx_queues, internals,
1291 			eth_dev) < 0)
1292 		return -1;
1293 
1294 	pp = (*eth_dev)->process_private;
1295 	for (i = 0; i < nb_rx_queues; i++) {
1296 		struct pcap_rx_queue *rx = &(*internals)->rx_queue[i];
1297 		struct devargs_queue *queue = &rx_queues->queue[i];
1298 
1299 		pp->rx_pcap[i] = queue->pcap;
1300 		strlcpy(rx->name, queue->name, sizeof(rx->name));
1301 		strlcpy(rx->type, queue->type, sizeof(rx->type));
1302 	}
1303 
1304 	for (i = 0; i < nb_tx_queues; i++) {
1305 		struct pcap_tx_queue *tx = &(*internals)->tx_queue[i];
1306 		struct devargs_queue *queue = &tx_queues->queue[i];
1307 
1308 		pp->tx_dumper[i] = queue->dumper;
1309 		pp->tx_pcap[i] = queue->pcap;
1310 		strlcpy(tx->name, queue->name, sizeof(tx->name));
1311 		strlcpy(tx->type, queue->type, sizeof(tx->type));
1312 	}
1313 
1314 	return 0;
1315 }
1316 
1317 static int
1318 eth_from_pcaps(struct rte_vdev_device *vdev,
1319 		struct pmd_devargs_all *devargs_all)
1320 {
1321 	struct pmd_internals *internals = NULL;
1322 	struct rte_eth_dev *eth_dev = NULL;
1323 	struct pmd_devargs *rx_queues = &devargs_all->rx_queues;
1324 	int single_iface = devargs_all->single_iface;
1325 	unsigned int infinite_rx = devargs_all->infinite_rx;
1326 	int ret;
1327 
1328 	ret = eth_from_pcaps_common(vdev, devargs_all, &internals, &eth_dev);
1329 
1330 	if (ret < 0)
1331 		return ret;
1332 
1333 	/* store weather we are using a single interface for rx/tx or not */
1334 	internals->single_iface = single_iface;
1335 
1336 	if (single_iface) {
1337 		internals->if_index =
1338 			osdep_iface_index_get(rx_queues->queue[0].name);
1339 
1340 		/* phy_mac arg is applied only only if "iface" devarg is provided */
1341 		if (rx_queues->phy_mac) {
1342 			if (eth_pcap_update_mac(rx_queues->queue[0].name,
1343 					eth_dev, vdev->device.numa_node) == 0)
1344 				internals->phy_mac = 1;
1345 		}
1346 	}
1347 
1348 	internals->infinite_rx = infinite_rx;
1349 	/* Assign rx ops. */
1350 	if (infinite_rx)
1351 		eth_dev->rx_pkt_burst = eth_pcap_rx_infinite;
1352 	else if (devargs_all->is_rx_pcap || devargs_all->is_rx_iface ||
1353 			single_iface)
1354 		eth_dev->rx_pkt_burst = eth_pcap_rx;
1355 	else
1356 		eth_dev->rx_pkt_burst = eth_null_rx;
1357 
1358 	/* Assign tx ops. */
1359 	if (devargs_all->is_tx_pcap)
1360 		eth_dev->tx_pkt_burst = eth_pcap_tx_dumper;
1361 	else if (devargs_all->is_tx_iface || single_iface)
1362 		eth_dev->tx_pkt_burst = eth_pcap_tx;
1363 	else
1364 		eth_dev->tx_pkt_burst = eth_tx_drop;
1365 
1366 	rte_eth_dev_probing_finish(eth_dev);
1367 	return 0;
1368 }
1369 
1370 static void
1371 eth_release_pcaps(struct pmd_devargs *pcaps,
1372 		struct pmd_devargs *dumpers,
1373 		int single_iface)
1374 {
1375 	unsigned int i;
1376 
1377 	if (single_iface) {
1378 		if (pcaps->queue[0].pcap)
1379 			pcap_close(pcaps->queue[0].pcap);
1380 		return;
1381 	}
1382 
1383 	for (i = 0; i < dumpers->num_of_queue; i++) {
1384 		if (dumpers->queue[i].dumper)
1385 			pcap_dump_close(dumpers->queue[i].dumper);
1386 
1387 		if (dumpers->queue[i].pcap)
1388 			pcap_close(dumpers->queue[i].pcap);
1389 	}
1390 
1391 	for (i = 0; i < pcaps->num_of_queue; i++) {
1392 		if (pcaps->queue[i].pcap)
1393 			pcap_close(pcaps->queue[i].pcap);
1394 	}
1395 }
1396 
1397 static int
1398 pmd_pcap_probe(struct rte_vdev_device *dev)
1399 {
1400 	const char *name;
1401 	struct rte_kvargs *kvlist;
1402 	struct pmd_devargs pcaps = {0};
1403 	struct pmd_devargs dumpers = {0};
1404 	struct rte_eth_dev *eth_dev =  NULL;
1405 	struct pmd_internals *internal;
1406 	int ret = 0;
1407 
1408 	struct pmd_devargs_all devargs_all = {
1409 		.single_iface = 0,
1410 		.is_tx_pcap = 0,
1411 		.is_tx_iface = 0,
1412 		.infinite_rx = 0,
1413 	};
1414 
1415 	name = rte_vdev_device_name(dev);
1416 	PMD_LOG(INFO, "Initializing pmd_pcap for %s", name);
1417 
1418 	timespec_get(&start_time, TIME_UTC);
1419 	start_cycles = rte_get_timer_cycles();
1420 	hz = rte_get_timer_hz();
1421 
1422 	ret = rte_mbuf_dyn_rx_timestamp_register(&timestamp_dynfield_offset,
1423 			&timestamp_rx_dynflag);
1424 	if (ret != 0) {
1425 		PMD_LOG(ERR, "Failed to register Rx timestamp field/flag");
1426 		return -1;
1427 	}
1428 
1429 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1430 		eth_dev = rte_eth_dev_attach_secondary(name);
1431 		if (!eth_dev) {
1432 			PMD_LOG(ERR, "Failed to probe %s", name);
1433 			return -1;
1434 		}
1435 
1436 		internal = eth_dev->data->dev_private;
1437 
1438 		kvlist = rte_kvargs_parse(internal->devargs, valid_arguments);
1439 		if (kvlist == NULL)
1440 			return -1;
1441 	} else {
1442 		kvlist = rte_kvargs_parse(rte_vdev_device_args(dev),
1443 				valid_arguments);
1444 		if (kvlist == NULL)
1445 			return -1;
1446 	}
1447 
1448 	/*
1449 	 * If iface argument is passed we open the NICs and use them for
1450 	 * reading / writing
1451 	 */
1452 	if (rte_kvargs_count(kvlist, ETH_PCAP_IFACE_ARG) == 1) {
1453 
1454 		ret = rte_kvargs_process(kvlist, ETH_PCAP_IFACE_ARG,
1455 				&open_rx_tx_iface, &pcaps);
1456 		if (ret < 0)
1457 			goto free_kvlist;
1458 
1459 		dumpers.queue[0] = pcaps.queue[0];
1460 
1461 		ret = rte_kvargs_process(kvlist, ETH_PCAP_PHY_MAC_ARG,
1462 				&select_phy_mac, &pcaps.phy_mac);
1463 		if (ret < 0)
1464 			goto free_kvlist;
1465 
1466 		dumpers.phy_mac = pcaps.phy_mac;
1467 
1468 		devargs_all.single_iface = 1;
1469 		pcaps.num_of_queue = 1;
1470 		dumpers.num_of_queue = 1;
1471 
1472 		goto create_eth;
1473 	}
1474 
1475 	/*
1476 	 * We check whether we want to open a RX stream from a real NIC, a
1477 	 * pcap file or open a dummy RX stream
1478 	 */
1479 	devargs_all.is_rx_pcap =
1480 		rte_kvargs_count(kvlist, ETH_PCAP_RX_PCAP_ARG) ? 1 : 0;
1481 	devargs_all.is_rx_iface =
1482 		(rte_kvargs_count(kvlist, ETH_PCAP_RX_IFACE_ARG) +
1483 		 rte_kvargs_count(kvlist, ETH_PCAP_RX_IFACE_IN_ARG)) ? 1 : 0;
1484 	pcaps.num_of_queue = 0;
1485 
1486 	devargs_all.is_tx_pcap =
1487 		rte_kvargs_count(kvlist, ETH_PCAP_TX_PCAP_ARG) ? 1 : 0;
1488 	devargs_all.is_tx_iface =
1489 		rte_kvargs_count(kvlist, ETH_PCAP_TX_IFACE_ARG) ? 1 : 0;
1490 	dumpers.num_of_queue = 0;
1491 
1492 	if (devargs_all.is_rx_pcap) {
1493 		/*
1494 		 * We check whether we want to infinitely rx the pcap file.
1495 		 */
1496 		unsigned int infinite_rx_arg_cnt = rte_kvargs_count(kvlist,
1497 				ETH_PCAP_INFINITE_RX_ARG);
1498 
1499 		if (infinite_rx_arg_cnt == 1) {
1500 			ret = rte_kvargs_process(kvlist,
1501 					ETH_PCAP_INFINITE_RX_ARG,
1502 					&get_infinite_rx_arg,
1503 					&devargs_all.infinite_rx);
1504 			if (ret < 0)
1505 				goto free_kvlist;
1506 			PMD_LOG(INFO, "infinite_rx has been %s for %s",
1507 					devargs_all.infinite_rx ? "enabled" : "disabled",
1508 					name);
1509 
1510 		} else if (infinite_rx_arg_cnt > 1) {
1511 			PMD_LOG(WARNING, "infinite_rx has not been enabled since the "
1512 					"argument has been provided more than once "
1513 					"for %s", name);
1514 		}
1515 
1516 		ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_PCAP_ARG,
1517 				&open_rx_pcap, &pcaps);
1518 	} else if (devargs_all.is_rx_iface) {
1519 		ret = rte_kvargs_process(kvlist, NULL,
1520 				&rx_iface_args_process, &pcaps);
1521 	} else if (devargs_all.is_tx_iface || devargs_all.is_tx_pcap) {
1522 		unsigned int i;
1523 
1524 		/* Count number of tx queue args passed before dummy rx queue
1525 		 * creation so a dummy rx queue can be created for each tx queue
1526 		 */
1527 		unsigned int num_tx_queues =
1528 			(rte_kvargs_count(kvlist, ETH_PCAP_TX_PCAP_ARG) +
1529 			rte_kvargs_count(kvlist, ETH_PCAP_TX_IFACE_ARG));
1530 
1531 		PMD_LOG(INFO, "Creating null rx queue since no rx queues were provided.");
1532 
1533 		/* Creating a dummy rx queue for each tx queue passed */
1534 		for (i = 0; i < num_tx_queues; i++)
1535 			ret = add_queue(&pcaps, "dummy_rx", "rx_null", NULL,
1536 					NULL);
1537 	} else {
1538 		PMD_LOG(ERR, "Error - No rx or tx queues provided");
1539 		ret = -ENOENT;
1540 	}
1541 	if (ret < 0)
1542 		goto free_kvlist;
1543 
1544 	/*
1545 	 * We check whether we want to open a TX stream to a real NIC,
1546 	 * a pcap file, or drop packets on tx
1547 	 */
1548 	if (devargs_all.is_tx_pcap) {
1549 		ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_PCAP_ARG,
1550 				&open_tx_pcap, &dumpers);
1551 	} else if (devargs_all.is_tx_iface) {
1552 		ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_IFACE_ARG,
1553 				&open_tx_iface, &dumpers);
1554 	} else {
1555 		unsigned int i;
1556 
1557 		PMD_LOG(INFO, "Dropping packets on tx since no tx queues were provided.");
1558 
1559 		/* Add 1 dummy queue per rxq which counts and drops packets. */
1560 		for (i = 0; i < pcaps.num_of_queue; i++)
1561 			ret = add_queue(&dumpers, "dummy_tx", "tx_drop", NULL,
1562 					NULL);
1563 	}
1564 
1565 	if (ret < 0)
1566 		goto free_kvlist;
1567 
1568 create_eth:
1569 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1570 		struct pmd_process_private *pp;
1571 		unsigned int i;
1572 
1573 		internal = eth_dev->data->dev_private;
1574 			pp = (struct pmd_process_private *)
1575 				rte_zmalloc(NULL,
1576 					sizeof(struct pmd_process_private),
1577 					RTE_CACHE_LINE_SIZE);
1578 
1579 		if (pp == NULL) {
1580 			PMD_LOG(ERR,
1581 				"Failed to allocate memory for process private");
1582 			ret = -1;
1583 			goto free_kvlist;
1584 		}
1585 
1586 		eth_dev->dev_ops = &ops;
1587 		eth_dev->device = &dev->device;
1588 
1589 		/* setup process private */
1590 		for (i = 0; i < pcaps.num_of_queue; i++)
1591 			pp->rx_pcap[i] = pcaps.queue[i].pcap;
1592 
1593 		for (i = 0; i < dumpers.num_of_queue; i++) {
1594 			pp->tx_dumper[i] = dumpers.queue[i].dumper;
1595 			pp->tx_pcap[i] = dumpers.queue[i].pcap;
1596 		}
1597 
1598 		eth_dev->process_private = pp;
1599 		eth_dev->rx_pkt_burst = eth_pcap_rx;
1600 		if (devargs_all.is_tx_pcap)
1601 			eth_dev->tx_pkt_burst = eth_pcap_tx_dumper;
1602 		else
1603 			eth_dev->tx_pkt_burst = eth_pcap_tx;
1604 
1605 		rte_eth_dev_probing_finish(eth_dev);
1606 		goto free_kvlist;
1607 	}
1608 
1609 	devargs_all.rx_queues = pcaps;
1610 	devargs_all.tx_queues = dumpers;
1611 
1612 	ret = eth_from_pcaps(dev, &devargs_all);
1613 
1614 free_kvlist:
1615 	rte_kvargs_free(kvlist);
1616 
1617 	if (ret < 0)
1618 		eth_release_pcaps(&pcaps, &dumpers, devargs_all.single_iface);
1619 
1620 	return ret;
1621 }
1622 
1623 static int
1624 pmd_pcap_remove(struct rte_vdev_device *dev)
1625 {
1626 	struct rte_eth_dev *eth_dev = NULL;
1627 
1628 	if (!dev)
1629 		return -1;
1630 
1631 	eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1632 	if (eth_dev == NULL)
1633 		return 0; /* port already released */
1634 
1635 	eth_dev_close(eth_dev);
1636 	rte_eth_dev_release_port(eth_dev);
1637 
1638 	return 0;
1639 }
1640 
1641 static struct rte_vdev_driver pmd_pcap_drv = {
1642 	.probe = pmd_pcap_probe,
1643 	.remove = pmd_pcap_remove,
1644 };
1645 
1646 RTE_PMD_REGISTER_VDEV(net_pcap, pmd_pcap_drv);
1647 RTE_PMD_REGISTER_ALIAS(net_pcap, eth_pcap);
1648 RTE_PMD_REGISTER_PARAM_STRING(net_pcap,
1649 	ETH_PCAP_RX_PCAP_ARG "=<string> "
1650 	ETH_PCAP_TX_PCAP_ARG "=<string> "
1651 	ETH_PCAP_RX_IFACE_ARG "=<ifc> "
1652 	ETH_PCAP_RX_IFACE_IN_ARG "=<ifc> "
1653 	ETH_PCAP_TX_IFACE_ARG "=<ifc> "
1654 	ETH_PCAP_IFACE_ARG "=<ifc> "
1655 	ETH_PCAP_PHY_MAC_ARG "=<int>"
1656 	ETH_PCAP_INFINITE_RX_ARG "=<0|1>");
1657