net/af_packet: make qdisc bypass configurable
[dpdk.git] / drivers / net / pcap / rte_eth_pcap.c
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
5  *   Copyright(c) 2014 6WIND S.A.
6  *   All rights reserved.
7  *
8  *   Redistribution and use in source and binary forms, with or without
9  *   modification, are permitted provided that the following conditions
10  *   are met:
11  *
12  *     * Redistributions of source code must retain the above copyright
13  *       notice, this list of conditions and the following disclaimer.
14  *     * Redistributions in binary form must reproduce the above copyright
15  *       notice, this list of conditions and the following disclaimer in
16  *       the documentation and/or other materials provided with the
17  *       distribution.
18  *     * Neither the name of Intel Corporation nor the names of its
19  *       contributors may be used to endorse or promote products derived
20  *       from this software without specific prior written permission.
21  *
22  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
25  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
26  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
27  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
28  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
29  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
30  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
31  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
32  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33  */
34
35 #include <time.h>
36
37 #include <net/if.h>
38
39 #include <pcap.h>
40
41 #include <rte_cycles.h>
42 #include <rte_ethdev.h>
43 #include <rte_ethdev_vdev.h>
44 #include <rte_kvargs.h>
45 #include <rte_malloc.h>
46 #include <rte_mbuf.h>
47 #include <rte_vdev.h>
48
49 #define RTE_ETH_PCAP_SNAPSHOT_LEN 65535
50 #define RTE_ETH_PCAP_SNAPLEN ETHER_MAX_JUMBO_FRAME_LEN
51 #define RTE_ETH_PCAP_PROMISC 1
52 #define RTE_ETH_PCAP_TIMEOUT -1
53
54 #define ETH_PCAP_RX_PCAP_ARG  "rx_pcap"
55 #define ETH_PCAP_TX_PCAP_ARG  "tx_pcap"
56 #define ETH_PCAP_RX_IFACE_ARG "rx_iface"
57 #define ETH_PCAP_TX_IFACE_ARG "tx_iface"
58 #define ETH_PCAP_IFACE_ARG    "iface"
59
60 #define ETH_PCAP_ARG_MAXLEN     64
61
62 #define RTE_PMD_PCAP_MAX_QUEUES 16
63
64 static char errbuf[PCAP_ERRBUF_SIZE];
65 static unsigned char tx_pcap_data[RTE_ETH_PCAP_SNAPLEN];
66 static struct timeval start_time;
67 static uint64_t start_cycles;
68 static uint64_t hz;
69
70 struct queue_stat {
71         volatile unsigned long pkts;
72         volatile unsigned long bytes;
73         volatile unsigned long err_pkts;
74 };
75
76 struct pcap_rx_queue {
77         pcap_t *pcap;
78         uint8_t in_port;
79         struct rte_mempool *mb_pool;
80         struct queue_stat rx_stat;
81         char name[PATH_MAX];
82         char type[ETH_PCAP_ARG_MAXLEN];
83 };
84
85 struct pcap_tx_queue {
86         pcap_dumper_t *dumper;
87         pcap_t *pcap;
88         struct queue_stat tx_stat;
89         char name[PATH_MAX];
90         char type[ETH_PCAP_ARG_MAXLEN];
91 };
92
93 struct pmd_internals {
94         struct pcap_rx_queue rx_queue[RTE_PMD_PCAP_MAX_QUEUES];
95         struct pcap_tx_queue tx_queue[RTE_PMD_PCAP_MAX_QUEUES];
96         int if_index;
97         int single_iface;
98 };
99
100 struct pmd_devargs {
101         unsigned int num_of_queue;
102         struct devargs_queue {
103                 pcap_dumper_t *dumper;
104                 pcap_t *pcap;
105                 const char *name;
106                 const char *type;
107         } queue[RTE_PMD_PCAP_MAX_QUEUES];
108 };
109
110 static const char *valid_arguments[] = {
111         ETH_PCAP_RX_PCAP_ARG,
112         ETH_PCAP_TX_PCAP_ARG,
113         ETH_PCAP_RX_IFACE_ARG,
114         ETH_PCAP_TX_IFACE_ARG,
115         ETH_PCAP_IFACE_ARG,
116         NULL
117 };
118
119 static struct ether_addr eth_addr = {
120         .addr_bytes = { 0, 0, 0, 0x1, 0x2, 0x3 }
121 };
122
123 static struct rte_eth_link pmd_link = {
124                 .link_speed = ETH_SPEED_NUM_10G,
125                 .link_duplex = ETH_LINK_FULL_DUPLEX,
126                 .link_status = ETH_LINK_DOWN,
127                 .link_autoneg = ETH_LINK_SPEED_FIXED,
128 };
129
130 static int
131 eth_pcap_rx_jumbo(struct rte_mempool *mb_pool, struct rte_mbuf *mbuf,
132                 const u_char *data, uint16_t data_len)
133 {
134         /* Copy the first segment. */
135         uint16_t len = rte_pktmbuf_tailroom(mbuf);
136         struct rte_mbuf *m = mbuf;
137
138         rte_memcpy(rte_pktmbuf_append(mbuf, len), data, len);
139         data_len -= len;
140         data += len;
141
142         while (data_len > 0) {
143                 /* Allocate next mbuf and point to that. */
144                 m->next = rte_pktmbuf_alloc(mb_pool);
145
146                 if (unlikely(!m->next))
147                         return -1;
148
149                 m = m->next;
150
151                 /* Headroom is not needed in chained mbufs. */
152                 rte_pktmbuf_prepend(m, rte_pktmbuf_headroom(m));
153                 m->pkt_len = 0;
154                 m->data_len = 0;
155
156                 /* Copy next segment. */
157                 len = RTE_MIN(rte_pktmbuf_tailroom(m), data_len);
158                 rte_memcpy(rte_pktmbuf_append(m, len), data, len);
159
160                 mbuf->nb_segs++;
161                 data_len -= len;
162                 data += len;
163         }
164
165         return mbuf->nb_segs;
166 }
167
168 /* Copy data from mbuf chain to a buffer suitable for writing to a PCAP file. */
169 static void
170 eth_pcap_gather_data(unsigned char *data, struct rte_mbuf *mbuf)
171 {
172         uint16_t data_len = 0;
173
174         while (mbuf) {
175                 rte_memcpy(data + data_len, rte_pktmbuf_mtod(mbuf, void *),
176                         mbuf->data_len);
177
178                 data_len += mbuf->data_len;
179                 mbuf = mbuf->next;
180         }
181 }
182
183 static uint16_t
184 eth_pcap_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
185 {
186         unsigned int i;
187         struct pcap_pkthdr header;
188         const u_char *packet;
189         struct rte_mbuf *mbuf;
190         struct pcap_rx_queue *pcap_q = queue;
191         uint16_t num_rx = 0;
192         uint16_t buf_size;
193         uint32_t rx_bytes = 0;
194
195         if (unlikely(pcap_q->pcap == NULL || nb_pkts == 0))
196                 return 0;
197
198         /* Reads the given number of packets from the pcap file one by one
199          * and copies the packet data into a newly allocated mbuf to return.
200          */
201         for (i = 0; i < nb_pkts; i++) {
202                 /* Get the next PCAP packet */
203                 packet = pcap_next(pcap_q->pcap, &header);
204                 if (unlikely(packet == NULL))
205                         break;
206
207                 mbuf = rte_pktmbuf_alloc(pcap_q->mb_pool);
208                 if (unlikely(mbuf == NULL))
209                         break;
210
211                 /* Now get the space available for data in the mbuf */
212                 buf_size = rte_pktmbuf_data_room_size(pcap_q->mb_pool) -
213                                 RTE_PKTMBUF_HEADROOM;
214
215                 if (header.caplen <= buf_size) {
216                         /* pcap packet will fit in the mbuf, can copy it */
217                         rte_memcpy(rte_pktmbuf_mtod(mbuf, void *), packet,
218                                         header.caplen);
219                         mbuf->data_len = (uint16_t)header.caplen;
220                 } else {
221                         /* Try read jumbo frame into multi mbufs. */
222                         if (unlikely(eth_pcap_rx_jumbo(pcap_q->mb_pool,
223                                                        mbuf,
224                                                        packet,
225                                                        header.caplen) == -1)) {
226                                 rte_pktmbuf_free(mbuf);
227                                 break;
228                         }
229                 }
230
231                 mbuf->pkt_len = (uint16_t)header.caplen;
232                 mbuf->port = pcap_q->in_port;
233                 bufs[num_rx] = mbuf;
234                 num_rx++;
235                 rx_bytes += header.caplen;
236         }
237         pcap_q->rx_stat.pkts += num_rx;
238         pcap_q->rx_stat.bytes += rx_bytes;
239
240         return num_rx;
241 }
242
243 static inline void
244 calculate_timestamp(struct timeval *ts) {
245         uint64_t cycles;
246         struct timeval cur_time;
247
248         cycles = rte_get_timer_cycles() - start_cycles;
249         cur_time.tv_sec = cycles / hz;
250         cur_time.tv_usec = (cycles % hz) * 1e6 / hz;
251         timeradd(&start_time, &cur_time, ts);
252 }
253
254 /*
255  * Callback to handle writing packets to a pcap file.
256  */
257 static uint16_t
258 eth_pcap_tx_dumper(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
259 {
260         unsigned int i;
261         struct rte_mbuf *mbuf;
262         struct pcap_tx_queue *dumper_q = queue;
263         uint16_t num_tx = 0;
264         uint32_t tx_bytes = 0;
265         struct pcap_pkthdr header;
266
267         if (dumper_q->dumper == NULL || nb_pkts == 0)
268                 return 0;
269
270         /* writes the nb_pkts packets to the previously opened pcap file
271          * dumper */
272         for (i = 0; i < nb_pkts; i++) {
273                 mbuf = bufs[i];
274                 calculate_timestamp(&header.ts);
275                 header.len = mbuf->pkt_len;
276                 header.caplen = header.len;
277
278                 if (likely(mbuf->nb_segs == 1)) {
279                         pcap_dump((u_char *)dumper_q->dumper, &header,
280                                   rte_pktmbuf_mtod(mbuf, void*));
281                 } else {
282                         if (mbuf->pkt_len <= ETHER_MAX_JUMBO_FRAME_LEN) {
283                                 eth_pcap_gather_data(tx_pcap_data, mbuf);
284                                 pcap_dump((u_char *)dumper_q->dumper, &header,
285                                           tx_pcap_data);
286                         } else {
287                                 RTE_LOG(ERR, PMD,
288                                         "Dropping PCAP packet. Size (%d) > max jumbo size (%d).\n",
289                                         mbuf->pkt_len,
290                                         ETHER_MAX_JUMBO_FRAME_LEN);
291
292                                 rte_pktmbuf_free(mbuf);
293                                 break;
294                         }
295                 }
296
297                 num_tx++;
298                 tx_bytes += mbuf->pkt_len;
299                 rte_pktmbuf_free(mbuf);
300         }
301
302         /*
303          * Since there's no place to hook a callback when the forwarding
304          * process stops and to make sure the pcap file is actually written,
305          * we flush the pcap dumper within each burst.
306          */
307         pcap_dump_flush(dumper_q->dumper);
308         dumper_q->tx_stat.pkts += num_tx;
309         dumper_q->tx_stat.bytes += tx_bytes;
310         dumper_q->tx_stat.err_pkts += nb_pkts - num_tx;
311
312         return num_tx;
313 }
314
315 /*
316  * Callback to handle sending packets through a real NIC.
317  */
318 static uint16_t
319 eth_pcap_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
320 {
321         unsigned int i;
322         int ret;
323         struct rte_mbuf *mbuf;
324         struct pcap_tx_queue *tx_queue = queue;
325         uint16_t num_tx = 0;
326         uint32_t tx_bytes = 0;
327
328         if (unlikely(nb_pkts == 0 || tx_queue->pcap == NULL))
329                 return 0;
330
331         for (i = 0; i < nb_pkts; i++) {
332                 mbuf = bufs[i];
333
334                 if (likely(mbuf->nb_segs == 1)) {
335                         ret = pcap_sendpacket(tx_queue->pcap,
336                                         rte_pktmbuf_mtod(mbuf, u_char *),
337                                         mbuf->pkt_len);
338                 } else {
339                         if (mbuf->pkt_len <= ETHER_MAX_JUMBO_FRAME_LEN) {
340                                 eth_pcap_gather_data(tx_pcap_data, mbuf);
341                                 ret = pcap_sendpacket(tx_queue->pcap,
342                                                 tx_pcap_data, mbuf->pkt_len);
343                         } else {
344                                 RTE_LOG(ERR, PMD,
345                                         "Dropping PCAP packet. Size (%d) > max jumbo size (%d).\n",
346                                         mbuf->pkt_len,
347                                         ETHER_MAX_JUMBO_FRAME_LEN);
348
349                                 rte_pktmbuf_free(mbuf);
350                                 break;
351                         }
352                 }
353
354                 if (unlikely(ret != 0))
355                         break;
356                 num_tx++;
357                 tx_bytes += mbuf->pkt_len;
358                 rte_pktmbuf_free(mbuf);
359         }
360
361         tx_queue->tx_stat.pkts += num_tx;
362         tx_queue->tx_stat.bytes += tx_bytes;
363         tx_queue->tx_stat.err_pkts += nb_pkts - num_tx;
364
365         return num_tx;
366 }
367
368 /*
369  * pcap_open_live wrapper function
370  */
371 static inline int
372 open_iface_live(const char *iface, pcap_t **pcap) {
373         *pcap = pcap_open_live(iface, RTE_ETH_PCAP_SNAPLEN,
374                         RTE_ETH_PCAP_PROMISC, RTE_ETH_PCAP_TIMEOUT, errbuf);
375
376         if (*pcap == NULL) {
377                 RTE_LOG(ERR, PMD, "Couldn't open %s: %s\n", iface, errbuf);
378                 return -1;
379         }
380
381         return 0;
382 }
383
384 static int
385 open_single_iface(const char *iface, pcap_t **pcap)
386 {
387         if (open_iface_live(iface, pcap) < 0) {
388                 RTE_LOG(ERR, PMD, "Couldn't open interface %s\n", iface);
389                 return -1;
390         }
391
392         return 0;
393 }
394
395 static int
396 open_single_tx_pcap(const char *pcap_filename, pcap_dumper_t **dumper)
397 {
398         pcap_t *tx_pcap;
399
400         /*
401          * We need to create a dummy empty pcap_t to use it
402          * with pcap_dump_open(). We create big enough an Ethernet
403          * pcap holder.
404          */
405         tx_pcap = pcap_open_dead(DLT_EN10MB, RTE_ETH_PCAP_SNAPSHOT_LEN);
406         if (tx_pcap == NULL) {
407                 RTE_LOG(ERR, PMD, "Couldn't create dead pcap\n");
408                 return -1;
409         }
410
411         /* The dumper is created using the previous pcap_t reference */
412         *dumper = pcap_dump_open(tx_pcap, pcap_filename);
413         if (*dumper == NULL) {
414                 RTE_LOG(ERR, PMD, "Couldn't open %s for writing.\n",
415                         pcap_filename);
416                 return -1;
417         }
418
419         return 0;
420 }
421
422 static int
423 open_single_rx_pcap(const char *pcap_filename, pcap_t **pcap)
424 {
425         *pcap = pcap_open_offline(pcap_filename, errbuf);
426         if (*pcap == NULL) {
427                 RTE_LOG(ERR, PMD, "Couldn't open %s: %s\n", pcap_filename,
428                         errbuf);
429                 return -1;
430         }
431
432         return 0;
433 }
434
435 static int
436 eth_dev_start(struct rte_eth_dev *dev)
437 {
438         unsigned int i;
439         struct pmd_internals *internals = dev->data->dev_private;
440         struct pcap_tx_queue *tx;
441         struct pcap_rx_queue *rx;
442
443         /* Special iface case. Single pcap is open and shared between tx/rx. */
444         if (internals->single_iface) {
445                 tx = &internals->tx_queue[0];
446                 rx = &internals->rx_queue[0];
447
448                 if (!tx->pcap && strcmp(tx->type, ETH_PCAP_IFACE_ARG) == 0) {
449                         if (open_single_iface(tx->name, &tx->pcap) < 0)
450                                 return -1;
451                         rx->pcap = tx->pcap;
452                 }
453                 goto status_up;
454         }
455
456         /* If not open already, open tx pcaps/dumpers */
457         for (i = 0; i < dev->data->nb_tx_queues; i++) {
458                 tx = &internals->tx_queue[i];
459
460                 if (!tx->dumper &&
461                                 strcmp(tx->type, ETH_PCAP_TX_PCAP_ARG) == 0) {
462                         if (open_single_tx_pcap(tx->name, &tx->dumper) < 0)
463                                 return -1;
464                 } else if (!tx->pcap &&
465                                 strcmp(tx->type, ETH_PCAP_TX_IFACE_ARG) == 0) {
466                         if (open_single_iface(tx->name, &tx->pcap) < 0)
467                                 return -1;
468                 }
469         }
470
471         /* If not open already, open rx pcaps */
472         for (i = 0; i < dev->data->nb_rx_queues; i++) {
473                 rx = &internals->rx_queue[i];
474
475                 if (rx->pcap != NULL)
476                         continue;
477
478                 if (strcmp(rx->type, ETH_PCAP_RX_PCAP_ARG) == 0) {
479                         if (open_single_rx_pcap(rx->name, &rx->pcap) < 0)
480                                 return -1;
481                 } else if (strcmp(rx->type, ETH_PCAP_RX_IFACE_ARG) == 0) {
482                         if (open_single_iface(rx->name, &rx->pcap) < 0)
483                                 return -1;
484                 }
485         }
486
487 status_up:
488         dev->data->dev_link.link_status = ETH_LINK_UP;
489
490         return 0;
491 }
492
493 /*
494  * This function gets called when the current port gets stopped.
495  * Is the only place for us to close all the tx streams dumpers.
496  * If not called the dumpers will be flushed within each tx burst.
497  */
498 static void
499 eth_dev_stop(struct rte_eth_dev *dev)
500 {
501         unsigned int i;
502         struct pmd_internals *internals = dev->data->dev_private;
503         struct pcap_tx_queue *tx;
504         struct pcap_rx_queue *rx;
505
506         /* Special iface case. Single pcap is open and shared between tx/rx. */
507         if (internals->single_iface) {
508                 tx = &internals->tx_queue[0];
509                 rx = &internals->rx_queue[0];
510                 pcap_close(tx->pcap);
511                 tx->pcap = NULL;
512                 rx->pcap = NULL;
513                 goto status_down;
514         }
515
516         for (i = 0; i < dev->data->nb_tx_queues; i++) {
517                 tx = &internals->tx_queue[i];
518
519                 if (tx->dumper != NULL) {
520                         pcap_dump_close(tx->dumper);
521                         tx->dumper = NULL;
522                 }
523
524                 if (tx->pcap != NULL) {
525                         pcap_close(tx->pcap);
526                         tx->pcap = NULL;
527                 }
528         }
529
530         for (i = 0; i < dev->data->nb_rx_queues; i++) {
531                 rx = &internals->rx_queue[i];
532
533                 if (rx->pcap != NULL) {
534                         pcap_close(rx->pcap);
535                         rx->pcap = NULL;
536                 }
537         }
538
539 status_down:
540         dev->data->dev_link.link_status = ETH_LINK_DOWN;
541 }
542
543 static int
544 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
545 {
546         return 0;
547 }
548
549 static void
550 eth_dev_info(struct rte_eth_dev *dev,
551                 struct rte_eth_dev_info *dev_info)
552 {
553         struct pmd_internals *internals = dev->data->dev_private;
554
555         dev_info->if_index = internals->if_index;
556         dev_info->max_mac_addrs = 1;
557         dev_info->max_rx_pktlen = (uint32_t) -1;
558         dev_info->max_rx_queues = dev->data->nb_rx_queues;
559         dev_info->max_tx_queues = dev->data->nb_tx_queues;
560         dev_info->min_rx_bufsize = 0;
561 }
562
563 static void
564 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
565 {
566         unsigned int i;
567         unsigned long rx_packets_total = 0, rx_bytes_total = 0;
568         unsigned long tx_packets_total = 0, tx_bytes_total = 0;
569         unsigned long tx_packets_err_total = 0;
570         const struct pmd_internals *internal = dev->data->dev_private;
571
572         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
573                         i < dev->data->nb_rx_queues; i++) {
574                 stats->q_ipackets[i] = internal->rx_queue[i].rx_stat.pkts;
575                 stats->q_ibytes[i] = internal->rx_queue[i].rx_stat.bytes;
576                 rx_packets_total += stats->q_ipackets[i];
577                 rx_bytes_total += stats->q_ibytes[i];
578         }
579
580         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
581                         i < dev->data->nb_tx_queues; i++) {
582                 stats->q_opackets[i] = internal->tx_queue[i].tx_stat.pkts;
583                 stats->q_obytes[i] = internal->tx_queue[i].tx_stat.bytes;
584                 stats->q_errors[i] = internal->tx_queue[i].tx_stat.err_pkts;
585                 tx_packets_total += stats->q_opackets[i];
586                 tx_bytes_total += stats->q_obytes[i];
587                 tx_packets_err_total += stats->q_errors[i];
588         }
589
590         stats->ipackets = rx_packets_total;
591         stats->ibytes = rx_bytes_total;
592         stats->opackets = tx_packets_total;
593         stats->obytes = tx_bytes_total;
594         stats->oerrors = tx_packets_err_total;
595 }
596
597 static void
598 eth_stats_reset(struct rte_eth_dev *dev)
599 {
600         unsigned int i;
601         struct pmd_internals *internal = dev->data->dev_private;
602
603         for (i = 0; i < dev->data->nb_rx_queues; i++) {
604                 internal->rx_queue[i].rx_stat.pkts = 0;
605                 internal->rx_queue[i].rx_stat.bytes = 0;
606         }
607
608         for (i = 0; i < dev->data->nb_tx_queues; i++) {
609                 internal->tx_queue[i].tx_stat.pkts = 0;
610                 internal->tx_queue[i].tx_stat.bytes = 0;
611                 internal->tx_queue[i].tx_stat.err_pkts = 0;
612         }
613 }
614
615 static void
616 eth_dev_close(struct rte_eth_dev *dev __rte_unused)
617 {
618 }
619
620 static void
621 eth_queue_release(void *q __rte_unused)
622 {
623 }
624
625 static int
626 eth_link_update(struct rte_eth_dev *dev __rte_unused,
627                 int wait_to_complete __rte_unused)
628 {
629         return 0;
630 }
631
632 static int
633 eth_rx_queue_setup(struct rte_eth_dev *dev,
634                 uint16_t rx_queue_id,
635                 uint16_t nb_rx_desc __rte_unused,
636                 unsigned int socket_id __rte_unused,
637                 const struct rte_eth_rxconf *rx_conf __rte_unused,
638                 struct rte_mempool *mb_pool)
639 {
640         struct pmd_internals *internals = dev->data->dev_private;
641         struct pcap_rx_queue *pcap_q = &internals->rx_queue[rx_queue_id];
642
643         pcap_q->mb_pool = mb_pool;
644         dev->data->rx_queues[rx_queue_id] = pcap_q;
645         pcap_q->in_port = dev->data->port_id;
646
647         return 0;
648 }
649
650 static int
651 eth_tx_queue_setup(struct rte_eth_dev *dev,
652                 uint16_t tx_queue_id,
653                 uint16_t nb_tx_desc __rte_unused,
654                 unsigned int socket_id __rte_unused,
655                 const struct rte_eth_txconf *tx_conf __rte_unused)
656 {
657         struct pmd_internals *internals = dev->data->dev_private;
658
659         dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
660
661         return 0;
662 }
663
664 static const struct eth_dev_ops ops = {
665         .dev_start = eth_dev_start,
666         .dev_stop = eth_dev_stop,
667         .dev_close = eth_dev_close,
668         .dev_configure = eth_dev_configure,
669         .dev_infos_get = eth_dev_info,
670         .rx_queue_setup = eth_rx_queue_setup,
671         .tx_queue_setup = eth_tx_queue_setup,
672         .rx_queue_release = eth_queue_release,
673         .tx_queue_release = eth_queue_release,
674         .link_update = eth_link_update,
675         .stats_get = eth_stats_get,
676         .stats_reset = eth_stats_reset,
677 };
678
679 /*
680  * Function handler that opens the pcap file for reading a stores a
681  * reference of it for use it later on.
682  */
683 static int
684 open_rx_pcap(const char *key, const char *value, void *extra_args)
685 {
686         unsigned int i;
687         const char *pcap_filename = value;
688         struct pmd_devargs *rx = extra_args;
689         pcap_t *pcap = NULL;
690
691         for (i = 0; i < rx->num_of_queue; i++) {
692                 if (open_single_rx_pcap(pcap_filename, &pcap) < 0)
693                         return -1;
694
695                 rx->queue[i].pcap = pcap;
696                 rx->queue[i].name = pcap_filename;
697                 rx->queue[i].type = key;
698         }
699
700         return 0;
701 }
702
703 /*
704  * Opens a pcap file for writing and stores a reference to it
705  * for use it later on.
706  */
707 static int
708 open_tx_pcap(const char *key, const char *value, void *extra_args)
709 {
710         unsigned int i;
711         const char *pcap_filename = value;
712         struct pmd_devargs *dumpers = extra_args;
713         pcap_dumper_t *dumper;
714
715         for (i = 0; i < dumpers->num_of_queue; i++) {
716                 if (open_single_tx_pcap(pcap_filename, &dumper) < 0)
717                         return -1;
718
719                 dumpers->queue[i].dumper = dumper;
720                 dumpers->queue[i].name = pcap_filename;
721                 dumpers->queue[i].type = key;
722         }
723
724         return 0;
725 }
726
727 /*
728  * Opens an interface for reading and writing
729  */
730 static inline int
731 open_rx_tx_iface(const char *key, const char *value, void *extra_args)
732 {
733         const char *iface = value;
734         struct pmd_devargs *tx = extra_args;
735         pcap_t *pcap = NULL;
736
737         if (open_single_iface(iface, &pcap) < 0)
738                 return -1;
739
740         tx->queue[0].pcap = pcap;
741         tx->queue[0].name = iface;
742         tx->queue[0].type = key;
743
744         return 0;
745 }
746
747 /*
748  * Opens a NIC for reading packets from it
749  */
750 static inline int
751 open_rx_iface(const char *key, const char *value, void *extra_args)
752 {
753         unsigned int i;
754         const char *iface = value;
755         struct pmd_devargs *rx = extra_args;
756         pcap_t *pcap = NULL;
757
758         for (i = 0; i < rx->num_of_queue; i++) {
759                 if (open_single_iface(iface, &pcap) < 0)
760                         return -1;
761                 rx->queue[i].pcap = pcap;
762                 rx->queue[i].name = iface;
763                 rx->queue[i].type = key;
764         }
765
766         return 0;
767 }
768
769 /*
770  * Opens a NIC for writing packets to it
771  */
772 static int
773 open_tx_iface(const char *key, const char *value, void *extra_args)
774 {
775         unsigned int i;
776         const char *iface = value;
777         struct pmd_devargs *tx = extra_args;
778         pcap_t *pcap;
779
780         for (i = 0; i < tx->num_of_queue; i++) {
781                 if (open_single_iface(iface, &pcap) < 0)
782                         return -1;
783                 tx->queue[i].pcap = pcap;
784                 tx->queue[i].name = iface;
785                 tx->queue[i].type = key;
786         }
787
788         return 0;
789 }
790
791 static struct rte_vdev_driver pmd_pcap_drv;
792
793 static int
794 pmd_init_internals(struct rte_vdev_device *vdev,
795                 const unsigned int nb_rx_queues,
796                 const unsigned int nb_tx_queues,
797                 struct pmd_internals **internals,
798                 struct rte_eth_dev **eth_dev)
799 {
800         struct rte_eth_dev_data *data = NULL;
801         unsigned int numa_node = vdev->device.numa_node;
802         const char *name;
803
804         name = rte_vdev_device_name(vdev);
805         RTE_LOG(INFO, PMD, "Creating pcap-backed ethdev on numa socket %u\n",
806                 numa_node);
807
808         /* now do all data allocation - for eth_dev structure
809          * and internal (private) data
810          */
811         data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
812         if (data == NULL)
813                 return -1;
814
815         /* reserve an ethdev entry */
816         *eth_dev = rte_eth_vdev_allocate(vdev, sizeof(**internals));
817         if (*eth_dev == NULL) {
818                 rte_free(data);
819                 return -1;
820         }
821
822         /* now put it all together
823          * - store queue data in internals,
824          * - store numa_node info in eth_dev
825          * - point eth_dev_data to internals
826          * - and point eth_dev structure to new eth_dev_data structure
827          */
828         *internals = (*eth_dev)->data->dev_private;
829         rte_memcpy(data, (*eth_dev)->data, sizeof(*data));
830         data->nb_rx_queues = (uint16_t)nb_rx_queues;
831         data->nb_tx_queues = (uint16_t)nb_tx_queues;
832         data->dev_link = pmd_link;
833         data->mac_addrs = &eth_addr;
834
835         /*
836          * NOTE: we'll replace the data element, of originally allocated
837          * eth_dev so the rings are local per-process
838          */
839         (*eth_dev)->data = data;
840         (*eth_dev)->dev_ops = &ops;
841         data->dev_flags = RTE_ETH_DEV_DETACHABLE;
842
843         return 0;
844 }
845
846 static int
847 eth_from_pcaps_common(struct rte_vdev_device *vdev,
848                 struct pmd_devargs *rx_queues, const unsigned int nb_rx_queues,
849                 struct pmd_devargs *tx_queues, const unsigned int nb_tx_queues,
850                 struct rte_kvargs *kvlist, struct pmd_internals **internals,
851                 struct rte_eth_dev **eth_dev)
852 {
853         struct rte_kvargs_pair *pair = NULL;
854         unsigned int k_idx;
855         unsigned int i;
856
857         /* do some parameter checking */
858         if (rx_queues == NULL && nb_rx_queues > 0)
859                 return -1;
860         if (tx_queues == NULL && nb_tx_queues > 0)
861                 return -1;
862
863         if (pmd_init_internals(vdev, nb_rx_queues, nb_tx_queues, internals,
864                         eth_dev) < 0)
865                 return -1;
866
867         for (i = 0; i < nb_rx_queues; i++) {
868                 struct pcap_rx_queue *rx = &(*internals)->rx_queue[i];
869                 struct devargs_queue *queue = &rx_queues->queue[i];
870
871                 rx->pcap = queue->pcap;
872                 snprintf(rx->name, sizeof(rx->name), "%s", queue->name);
873                 snprintf(rx->type, sizeof(rx->type), "%s", queue->type);
874         }
875
876         for (i = 0; i < nb_tx_queues; i++) {
877                 struct pcap_tx_queue *tx = &(*internals)->tx_queue[i];
878                 struct devargs_queue *queue = &tx_queues->queue[i];
879
880                 tx->dumper = queue->dumper;
881                 tx->pcap = queue->pcap;
882                 snprintf(tx->name, sizeof(tx->name), "%s", queue->name);
883                 snprintf(tx->type, sizeof(tx->type), "%s", queue->type);
884         }
885
886         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
887                 pair = &kvlist->pairs[k_idx];
888                 if (strstr(pair->key, ETH_PCAP_IFACE_ARG) != NULL)
889                         break;
890         }
891
892         if (pair == NULL)
893                 (*internals)->if_index = 0;
894         else
895                 (*internals)->if_index = if_nametoindex(pair->value);
896
897         return 0;
898 }
899
900 static int
901 eth_from_pcaps(struct rte_vdev_device *vdev,
902                 struct pmd_devargs *rx_queues, const unsigned int nb_rx_queues,
903                 struct pmd_devargs *tx_queues, const unsigned int nb_tx_queues,
904                 struct rte_kvargs *kvlist, int single_iface,
905                 unsigned int using_dumpers)
906 {
907         struct pmd_internals *internals = NULL;
908         struct rte_eth_dev *eth_dev = NULL;
909         int ret;
910
911         ret = eth_from_pcaps_common(vdev, rx_queues, nb_rx_queues,
912                 tx_queues, nb_tx_queues, kvlist, &internals, &eth_dev);
913
914         if (ret < 0)
915                 return ret;
916
917         /* store weather we are using a single interface for rx/tx or not */
918         internals->single_iface = single_iface;
919
920         eth_dev->rx_pkt_burst = eth_pcap_rx;
921
922         if (using_dumpers)
923                 eth_dev->tx_pkt_burst = eth_pcap_tx_dumper;
924         else
925                 eth_dev->tx_pkt_burst = eth_pcap_tx;
926
927         return 0;
928 }
929
930 static int
931 pmd_pcap_probe(struct rte_vdev_device *dev)
932 {
933         const char *name;
934         unsigned int is_rx_pcap = 0, is_tx_pcap = 0;
935         struct rte_kvargs *kvlist;
936         struct pmd_devargs pcaps = {0};
937         struct pmd_devargs dumpers = {0};
938         int single_iface = 0;
939         int ret;
940
941         name = rte_vdev_device_name(dev);
942         RTE_LOG(INFO, PMD, "Initializing pmd_pcap for %s\n", name);
943
944         gettimeofday(&start_time, NULL);
945         start_cycles = rte_get_timer_cycles();
946         hz = rte_get_timer_hz();
947
948         kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_arguments);
949         if (kvlist == NULL)
950                 return -1;
951
952         /*
953          * If iface argument is passed we open the NICs and use them for
954          * reading / writing
955          */
956         if (rte_kvargs_count(kvlist, ETH_PCAP_IFACE_ARG) == 1) {
957
958                 ret = rte_kvargs_process(kvlist, ETH_PCAP_IFACE_ARG,
959                                 &open_rx_tx_iface, &pcaps);
960
961                 if (ret < 0)
962                         goto free_kvlist;
963
964                 dumpers.queue[0] = pcaps.queue[0];
965
966                 single_iface = 1;
967                 pcaps.num_of_queue = 1;
968                 dumpers.num_of_queue = 1;
969
970                 goto create_eth;
971         }
972
973         /*
974          * We check whether we want to open a RX stream from a real NIC or a
975          * pcap file
976          */
977         pcaps.num_of_queue = rte_kvargs_count(kvlist, ETH_PCAP_RX_PCAP_ARG);
978         if (pcaps.num_of_queue)
979                 is_rx_pcap = 1;
980         else
981                 pcaps.num_of_queue = rte_kvargs_count(kvlist,
982                                 ETH_PCAP_RX_IFACE_ARG);
983
984         if (pcaps.num_of_queue > RTE_PMD_PCAP_MAX_QUEUES)
985                 pcaps.num_of_queue = RTE_PMD_PCAP_MAX_QUEUES;
986
987         if (is_rx_pcap)
988                 ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_PCAP_ARG,
989                                 &open_rx_pcap, &pcaps);
990         else
991                 ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_IFACE_ARG,
992                                 &open_rx_iface, &pcaps);
993
994         if (ret < 0)
995                 goto free_kvlist;
996
997         /*
998          * We check whether we want to open a TX stream to a real NIC or a
999          * pcap file
1000          */
1001         dumpers.num_of_queue = rte_kvargs_count(kvlist, ETH_PCAP_TX_PCAP_ARG);
1002         if (dumpers.num_of_queue)
1003                 is_tx_pcap = 1;
1004         else
1005                 dumpers.num_of_queue = rte_kvargs_count(kvlist,
1006                                 ETH_PCAP_TX_IFACE_ARG);
1007
1008         if (dumpers.num_of_queue > RTE_PMD_PCAP_MAX_QUEUES)
1009                 dumpers.num_of_queue = RTE_PMD_PCAP_MAX_QUEUES;
1010
1011         if (is_tx_pcap)
1012                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_PCAP_ARG,
1013                                 &open_tx_pcap, &dumpers);
1014         else
1015                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_IFACE_ARG,
1016                                 &open_tx_iface, &dumpers);
1017
1018         if (ret < 0)
1019                 goto free_kvlist;
1020
1021 create_eth:
1022         ret = eth_from_pcaps(dev, &pcaps, pcaps.num_of_queue, &dumpers,
1023                 dumpers.num_of_queue, kvlist, single_iface, is_tx_pcap);
1024
1025 free_kvlist:
1026         rte_kvargs_free(kvlist);
1027
1028         return ret;
1029 }
1030
1031 static int
1032 pmd_pcap_remove(struct rte_vdev_device *dev)
1033 {
1034         struct rte_eth_dev *eth_dev = NULL;
1035
1036         RTE_LOG(INFO, PMD, "Closing pcap ethdev on numa socket %u\n",
1037                         rte_socket_id());
1038
1039         if (!dev)
1040                 return -1;
1041
1042         /* reserve an ethdev entry */
1043         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1044         if (eth_dev == NULL)
1045                 return -1;
1046
1047         rte_free(eth_dev->data->dev_private);
1048         rte_free(eth_dev->data);
1049
1050         rte_eth_dev_release_port(eth_dev);
1051
1052         return 0;
1053 }
1054
1055 static struct rte_vdev_driver pmd_pcap_drv = {
1056         .probe = pmd_pcap_probe,
1057         .remove = pmd_pcap_remove,
1058 };
1059
1060 RTE_PMD_REGISTER_VDEV(net_pcap, pmd_pcap_drv);
1061 RTE_PMD_REGISTER_ALIAS(net_pcap, eth_pcap);
1062 RTE_PMD_REGISTER_PARAM_STRING(net_pcap,
1063         ETH_PCAP_RX_PCAP_ARG "=<string> "
1064         ETH_PCAP_TX_PCAP_ARG "=<string> "
1065         ETH_PCAP_RX_IFACE_ARG "=<ifc> "
1066         ETH_PCAP_TX_IFACE_ARG "=<ifc> "
1067         ETH_PCAP_IFACE_ARG "=<ifc>");