drivers/net/pcap/pcap_ethdev.c

   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 <time.h>
   8
   9 #include <pcap.h>
  10
  11 #include <rte_cycles.h>
  12 #include <ethdev_driver.h>
  13 #include <ethdev_vdev.h>
  14 #include <rte_kvargs.h>
  15 #include <rte_malloc.h>
  16 #include <rte_mbuf.h>
  17 #include <rte_mbuf_dyn.h>
  18 #include <rte_bus_vdev.h>
  19
  20 #include "pcap_osdep.h"
  21
  22 #define RTE_ETH_PCAP_SNAPSHOT_LEN 65535
  23 #define RTE_ETH_PCAP_SNAPLEN RTE_ETHER_MAX_JUMBO_FRAME_LEN
  24 #define RTE_ETH_PCAP_PROMISC 1
  25 #define RTE_ETH_PCAP_TIMEOUT -1
  26
  27 #define ETH_PCAP_RX_PCAP_ARG  "rx_pcap"
  28 #define ETH_PCAP_TX_PCAP_ARG  "tx_pcap"
  29 #define ETH_PCAP_RX_IFACE_ARG "rx_iface"
  30 #define ETH_PCAP_RX_IFACE_IN_ARG "rx_iface_in"
  31 #define ETH_PCAP_TX_IFACE_ARG "tx_iface"
  32 #define ETH_PCAP_IFACE_ARG    "iface"
  33 #define ETH_PCAP_PHY_MAC_ARG  "phy_mac"
  34 #define ETH_PCAP_INFINITE_RX_ARG  "infinite_rx"
  35
  36 #define ETH_PCAP_ARG_MAXLEN     64
  37
  38 #define RTE_PMD_PCAP_MAX_QUEUES 16
  39
  40 static char errbuf[PCAP_ERRBUF_SIZE];
  41 static struct timespec start_time;
  42 static uint64_t start_cycles;
  43 static uint64_t hz;
  44 static uint8_t iface_idx;
  45
  46 static uint64_t timestamp_rx_dynflag;
  47 static int timestamp_dynfield_offset = -1;
  48
  49 struct queue_stat {
  50         volatile unsigned long pkts;
  51         volatile unsigned long bytes;
  52         volatile unsigned long err_pkts;
  53 };
  54
  55 struct queue_missed_stat {
  56         /* last value retrieved from pcap */
  57         unsigned int pcap;
  58         /* stores values lost by pcap stop or rollover */
  59         unsigned long mnemonic;
  60         /* value on last reset */
  61         unsigned long reset;
  62 };
  63
  64 struct pcap_rx_queue {
  65         uint16_t port_id;
  66         uint16_t queue_id;
  67         struct rte_mempool *mb_pool;
  68         struct queue_stat rx_stat;
  69         struct queue_missed_stat missed_stat;
  70         char name[PATH_MAX];
  71         char type[ETH_PCAP_ARG_MAXLEN];
  72
  73         /* Contains pre-generated packets to be looped through */
  74         struct rte_ring *pkts;
  75 };
  76
  77 struct pcap_tx_queue {
  78         uint16_t port_id;
  79         uint16_t queue_id;
  80         struct queue_stat tx_stat;
  81         char name[PATH_MAX];
  82         char type[ETH_PCAP_ARG_MAXLEN];
  83 };
  84
  85 struct pmd_internals {
  86         struct pcap_rx_queue rx_queue[RTE_PMD_PCAP_MAX_QUEUES];
  87         struct pcap_tx_queue tx_queue[RTE_PMD_PCAP_MAX_QUEUES];
  88         char devargs[ETH_PCAP_ARG_MAXLEN];
  89         struct rte_ether_addr eth_addr;
  90         int if_index;
  91         int single_iface;
  92         int phy_mac;
  93         unsigned int infinite_rx;
  94 };
  95
  96 struct pmd_process_private {
  97         pcap_t *rx_pcap[RTE_PMD_PCAP_MAX_QUEUES];
  98         pcap_t *tx_pcap[RTE_PMD_PCAP_MAX_QUEUES];
  99         pcap_dumper_t *tx_dumper[RTE_PMD_PCAP_MAX_QUEUES];
 100 };
 101
 102 struct pmd_devargs {
 103         unsigned int num_of_queue;
 104         struct devargs_queue {
 105                 pcap_dumper_t *dumper;
 106                 pcap_t *pcap;
 107                 const char *name;
 108                 const char *type;
 109         } queue[RTE_PMD_PCAP_MAX_QUEUES];
 110         int phy_mac;
 111 };
 112
 113 struct pmd_devargs_all {
 114         struct pmd_devargs rx_queues;
 115         struct pmd_devargs tx_queues;
 116         int single_iface;
 117         unsigned int is_tx_pcap;
 118         unsigned int is_tx_iface;
 119         unsigned int is_rx_pcap;
 120         unsigned int is_rx_iface;
 121         unsigned int infinite_rx;
 122 };
 123
 124 static const char *valid_arguments[] = {
 125         ETH_PCAP_RX_PCAP_ARG,
 126         ETH_PCAP_TX_PCAP_ARG,
 127         ETH_PCAP_RX_IFACE_ARG,
 128         ETH_PCAP_RX_IFACE_IN_ARG,
 129         ETH_PCAP_TX_IFACE_ARG,
 130         ETH_PCAP_IFACE_ARG,
 131         ETH_PCAP_PHY_MAC_ARG,
 132         ETH_PCAP_INFINITE_RX_ARG,
 133         NULL
 134 };
 135
 136 static struct rte_eth_link pmd_link = {
 137                 .link_speed = ETH_SPEED_NUM_10G,
 138                 .link_duplex = ETH_LINK_FULL_DUPLEX,
 139                 .link_status = ETH_LINK_DOWN,
 140                 .link_autoneg = ETH_LINK_FIXED,
 141 };
 142
 143 RTE_LOG_REGISTER(eth_pcap_logtype, pmd.net.pcap, NOTICE);
 144
 145 #define PMD_LOG(level, fmt, args...) \
 146         rte_log(RTE_LOG_ ## level, eth_pcap_logtype, \
 147                 "%s(): " fmt "\n", __func__, ##args)
 148
 149 static struct queue_missed_stat*
 150 queue_missed_stat_update(struct rte_eth_dev *dev, unsigned int qid)
 151 {
 152         struct pmd_internals *internals = dev->data->dev_private;
 153         struct queue_missed_stat *missed_stat =
 154                         &internals->rx_queue[qid].missed_stat;
 155         const struct pmd_process_private *pp = dev->process_private;
 156         pcap_t *pcap = pp->rx_pcap[qid];
 157         struct pcap_stat stat;
 158
 159         if (!pcap || (pcap_stats(pcap, &stat) != 0))
 160                 return missed_stat;
 161
 162         /* rollover check - best effort fixup assuming single rollover */
 163         if (stat.ps_drop < missed_stat->pcap)
 164                 missed_stat->mnemonic += UINT_MAX;
 165         missed_stat->pcap = stat.ps_drop;
 166
 167         return missed_stat;
 168 }
 169
 170 static void
 171 queue_missed_stat_on_stop_update(struct rte_eth_dev *dev, unsigned int qid)
 172 {
 173         struct queue_missed_stat *missed_stat =
 174                         queue_missed_stat_update(dev, qid);
 175
 176         missed_stat->mnemonic += missed_stat->pcap;
 177         missed_stat->pcap = 0;
 178 }
 179
 180 static void
 181 queue_missed_stat_reset(struct rte_eth_dev *dev, unsigned int qid)
 182 {
 183         struct queue_missed_stat *missed_stat =
 184                         queue_missed_stat_update(dev, qid);
 185
 186         missed_stat->reset = missed_stat->pcap;
 187         missed_stat->mnemonic = 0;
 188 }
 189
 190 static unsigned long
 191 queue_missed_stat_get(struct rte_eth_dev *dev, unsigned int qid)
 192 {
 193         const struct queue_missed_stat *missed_stat =
 194                         queue_missed_stat_update(dev, qid);
 195
 196         return missed_stat->pcap + missed_stat->mnemonic - missed_stat->reset;
 197 }
 198
 199 static int
 200 eth_pcap_rx_jumbo(struct rte_mempool *mb_pool, struct rte_mbuf *mbuf,
 201                 const u_char *data, uint16_t data_len)
 202 {
 203         /* Copy the first segment. */
 204         uint16_t len = rte_pktmbuf_tailroom(mbuf);
 205         struct rte_mbuf *m = mbuf;
 206
 207         rte_memcpy(rte_pktmbuf_append(mbuf, len), data, len);
 208         data_len -= len;
 209         data += len;
 210
 211         while (data_len > 0) {
 212                 /* Allocate next mbuf and point to that. */
 213                 m->next = rte_pktmbuf_alloc(mb_pool);
 214
 215                 if (unlikely(!m->next))
 216                         return -1;
 217
 218                 m = m->next;
 219
 220                 /* Headroom is not needed in chained mbufs. */
 221                 rte_pktmbuf_prepend(m, rte_pktmbuf_headroom(m));
 222                 m->pkt_len = 0;
 223                 m->data_len = 0;
 224
 225                 /* Copy next segment. */
 226                 len = RTE_MIN(rte_pktmbuf_tailroom(m), data_len);
 227                 rte_memcpy(rte_pktmbuf_append(m, len), data, len);
 228
 229                 mbuf->nb_segs++;
 230                 data_len -= len;
 231                 data += len;
 232         }
 233
 234         return mbuf->nb_segs;
 235 }
 236
 237 static uint16_t
 238 eth_pcap_rx_infinite(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 239 {
 240         int i;
 241         struct pcap_rx_queue *pcap_q = queue;
 242         uint32_t rx_bytes = 0;
 243
 244         if (unlikely(nb_pkts == 0))
 245                 return 0;
 246
 247         if (rte_pktmbuf_alloc_bulk(pcap_q->mb_pool, bufs, nb_pkts) != 0)
 248                 return 0;
 249
 250         for (i = 0; i < nb_pkts; i++) {
 251                 struct rte_mbuf *pcap_buf;
 252                 int err = rte_ring_dequeue(pcap_q->pkts, (void **)&pcap_buf);
 253                 if (err)
 254                         return i;
 255
 256                 rte_memcpy(rte_pktmbuf_mtod(bufs[i], void *),
 257                                 rte_pktmbuf_mtod(pcap_buf, void *),
 258                                 pcap_buf->data_len);
 259                 bufs[i]->data_len = pcap_buf->data_len;
 260                 bufs[i]->pkt_len = pcap_buf->pkt_len;
 261                 bufs[i]->port = pcap_q->port_id;
 262                 rx_bytes += pcap_buf->data_len;
 263
 264                 /* Enqueue packet back on ring to allow infinite rx. */
 265                 rte_ring_enqueue(pcap_q->pkts, pcap_buf);
 266         }
 267
 268         pcap_q->rx_stat.pkts += i;
 269         pcap_q->rx_stat.bytes += rx_bytes;
 270
 271         return i;
 272 }
 273
 274 static uint16_t
 275 eth_pcap_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 276 {
 277         unsigned int i;
 278         struct pcap_pkthdr header;
 279         struct pmd_process_private *pp;
 280         const u_char *packet;
 281         struct rte_mbuf *mbuf;
 282         struct pcap_rx_queue *pcap_q = queue;
 283         uint16_t num_rx = 0;
 284         uint32_t rx_bytes = 0;
 285         pcap_t *pcap;
 286
 287         pp = rte_eth_devices[pcap_q->port_id].process_private;
 288         pcap = pp->rx_pcap[pcap_q->queue_id];
 289
 290         if (unlikely(pcap == NULL || nb_pkts == 0))
 291                 return 0;
 292
 293         /* Reads the given number of packets from the pcap file one by one
 294          * and copies the packet data into a newly allocated mbuf to return.
 295          */
 296         for (i = 0; i < nb_pkts; i++) {
 297                 /* Get the next PCAP packet */
 298                 packet = pcap_next(pcap, &header);
 299                 if (unlikely(packet == NULL))
 300                         break;
 301
 302                 mbuf = rte_pktmbuf_alloc(pcap_q->mb_pool);
 303                 if (unlikely(mbuf == NULL))
 304                         break;
 305
 306                 if (header.caplen <= rte_pktmbuf_tailroom(mbuf)) {
 307                         /* pcap packet will fit in the mbuf, can copy it */
 308                         rte_memcpy(rte_pktmbuf_mtod(mbuf, void *), packet,
 309                                         header.caplen);
 310                         mbuf->data_len = (uint16_t)header.caplen;
 311                 } else {
 312                         /* Try read jumbo frame into multi mbufs. */
 313                         if (unlikely(eth_pcap_rx_jumbo(pcap_q->mb_pool,
 314                                                        mbuf,
 315                                                        packet,
 316                                                        header.caplen) == -1)) {
 317                                 rte_pktmbuf_free(mbuf);
 318                                 break;
 319                         }
 320                 }
 321
 322                 mbuf->pkt_len = (uint16_t)header.caplen;
 323                 *RTE_MBUF_DYNFIELD(mbuf, timestamp_dynfield_offset,
 324                         rte_mbuf_timestamp_t *) =
 325                                 (uint64_t)header.ts.tv_sec * 1000000 +
 326                                 header.ts.tv_usec;
 327                 mbuf->ol_flags |= timestamp_rx_dynflag;
 328                 mbuf->port = pcap_q->port_id;
 329                 bufs[num_rx] = mbuf;
 330                 num_rx++;
 331                 rx_bytes += header.caplen;
 332         }
 333         pcap_q->rx_stat.pkts += num_rx;
 334         pcap_q->rx_stat.bytes += rx_bytes;
 335
 336         return num_rx;
 337 }
 338
 339 static uint16_t
 340 eth_null_rx(void *queue __rte_unused,
 341                 struct rte_mbuf **bufs __rte_unused,
 342                 uint16_t nb_pkts __rte_unused)
 343 {
 344         return 0;
 345 }
 346
 347 #define NSEC_PER_SEC    1000000000L
 348
 349 /*
 350  * This function stores nanoseconds in `tv_usec` field of `struct timeval`,
 351  * because `ts` goes directly to nanosecond-precision dump.
 352  */
 353 static inline void
 354 calculate_timestamp(struct timeval *ts) {
 355         uint64_t cycles;
 356         struct timespec cur_time;
 357
 358         cycles = rte_get_timer_cycles() - start_cycles;
 359         cur_time.tv_sec = cycles / hz;
 360         cur_time.tv_nsec = (cycles % hz) * NSEC_PER_SEC / hz;
 361
 362         ts->tv_sec = start_time.tv_sec + cur_time.tv_sec;
 363         ts->tv_usec = start_time.tv_nsec + cur_time.tv_nsec;
 364         if (ts->tv_usec >= NSEC_PER_SEC) {
 365                 ts->tv_usec -= NSEC_PER_SEC;
 366                 ts->tv_sec += 1;
 367         }
 368 }
 369
 370 /*
 371  * Callback to handle writing packets to a pcap file.
 372  */
 373 static uint16_t
 374 eth_pcap_tx_dumper(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 375 {
 376         unsigned int i;
 377         struct rte_mbuf *mbuf;
 378         struct pmd_process_private *pp;
 379         struct pcap_tx_queue *dumper_q = queue;
 380         uint16_t num_tx = 0;
 381         uint32_t tx_bytes = 0;
 382         struct pcap_pkthdr header;
 383         pcap_dumper_t *dumper;
 384         unsigned char temp_data[RTE_ETH_PCAP_SNAPLEN];
 385         size_t len, caplen;
 386
 387         pp = rte_eth_devices[dumper_q->port_id].process_private;
 388         dumper = pp->tx_dumper[dumper_q->queue_id];
 389
 390         if (dumper == NULL || nb_pkts == 0)
 391                 return 0;
 392
 393         /* writes the nb_pkts packets to the previously opened pcap file
 394          * dumper */
 395         for (i = 0; i < nb_pkts; i++) {
 396                 mbuf = bufs[i];
 397                 len = caplen = rte_pktmbuf_pkt_len(mbuf);
 398                 if (unlikely(!rte_pktmbuf_is_contiguous(mbuf) &&
 399                                 len > sizeof(temp_data))) {
 400                         caplen = sizeof(temp_data);
 401                 }
 402
 403                 calculate_timestamp(&header.ts);
 404                 header.len = len;
 405                 header.caplen = caplen;
 406                 /* rte_pktmbuf_read() returns a pointer to the data directly
 407                  * in the mbuf (when the mbuf is contiguous) or, otherwise,
 408                  * a pointer to temp_data after copying into it.
 409                  */
 410                 pcap_dump((u_char *)dumper, &header,
 411                         rte_pktmbuf_read(mbuf, 0, caplen, temp_data));
 412
 413                 num_tx++;
 414                 tx_bytes += caplen;
 415                 rte_pktmbuf_free(mbuf);
 416         }
 417
 418         /*
 419          * Since there's no place to hook a callback when the forwarding
 420          * process stops and to make sure the pcap file is actually written,
 421          * we flush the pcap dumper within each burst.
 422          */
 423         pcap_dump_flush(dumper);
 424         dumper_q->tx_stat.pkts += num_tx;
 425         dumper_q->tx_stat.bytes += tx_bytes;
 426         dumper_q->tx_stat.err_pkts += nb_pkts - num_tx;
 427
 428         return nb_pkts;
 429 }
 430
 431 /*
 432  * Callback to handle dropping packets in the infinite rx case.
 433  */
 434 static uint16_t
 435 eth_tx_drop(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 436 {
 437         unsigned int i;
 438         uint32_t tx_bytes = 0;
 439         struct pcap_tx_queue *tx_queue = queue;
 440
 441         if (unlikely(nb_pkts == 0))
 442                 return 0;
 443
 444         for (i = 0; i < nb_pkts; i++) {
 445                 tx_bytes += bufs[i]->pkt_len;
 446                 rte_pktmbuf_free(bufs[i]);
 447         }
 448
 449         tx_queue->tx_stat.pkts += nb_pkts;
 450         tx_queue->tx_stat.bytes += tx_bytes;
 451
 452         return i;
 453 }
 454
 455 /*
 456  * Callback to handle sending packets through a real NIC.
 457  */
 458 static uint16_t
 459 eth_pcap_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 460 {
 461         unsigned int i;
 462         int ret;
 463         struct rte_mbuf *mbuf;
 464         struct pmd_process_private *pp;
 465         struct pcap_tx_queue *tx_queue = queue;
 466         uint16_t num_tx = 0;
 467         uint32_t tx_bytes = 0;
 468         pcap_t *pcap;
 469         unsigned char temp_data[RTE_ETH_PCAP_SNAPLEN];
 470         size_t len;
 471
 472         pp = rte_eth_devices[tx_queue->port_id].process_private;
 473         pcap = pp->tx_pcap[tx_queue->queue_id];
 474
 475         if (unlikely(nb_pkts == 0 || pcap == NULL))
 476                 return 0;
 477
 478         for (i = 0; i < nb_pkts; i++) {
 479                 mbuf = bufs[i];
 480                 len = rte_pktmbuf_pkt_len(mbuf);
 481                 if (unlikely(!rte_pktmbuf_is_contiguous(mbuf) &&
 482                                 len > sizeof(temp_data))) {
 483                         PMD_LOG(ERR,
 484                                 "Dropping multi segment PCAP packet. Size (%zd) > max size (%zd).",
 485                                 len, sizeof(temp_data));
 486                         rte_pktmbuf_free(mbuf);
 487                         continue;
 488                 }
 489
 490                 /* rte_pktmbuf_read() returns a pointer to the data directly
 491                  * in the mbuf (when the mbuf is contiguous) or, otherwise,
 492                  * a pointer to temp_data after copying into it.
 493                  */
 494                 ret = pcap_sendpacket(pcap,
 495                         rte_pktmbuf_read(mbuf, 0, len, temp_data), len);
 496                 if (unlikely(ret != 0))
 497                         break;
 498                 num_tx++;
 499                 tx_bytes += len;
 500                 rte_pktmbuf_free(mbuf);
 501         }
 502
 503         tx_queue->tx_stat.pkts += num_tx;
 504         tx_queue->tx_stat.bytes += tx_bytes;
 505         tx_queue->tx_stat.err_pkts += i - num_tx;
 506
 507         return i;
 508 }
 509
 510 /*
 511  * pcap_open_live wrapper function
 512  */
 513 static inline int
 514 open_iface_live(const char *iface, pcap_t **pcap) {
 515         *pcap = pcap_open_live(iface, RTE_ETH_PCAP_SNAPLEN,
 516                         RTE_ETH_PCAP_PROMISC, RTE_ETH_PCAP_TIMEOUT, errbuf);
 517
 518         if (*pcap == NULL) {
 519                 PMD_LOG(ERR, "Couldn't open %s: %s", iface, errbuf);
 520                 return -1;
 521         }
 522
 523         return 0;
 524 }
 525
 526 static int
 527 open_single_iface(const char *iface, pcap_t **pcap)
 528 {
 529         if (open_iface_live(iface, pcap) < 0) {
 530                 PMD_LOG(ERR, "Couldn't open interface %s", iface);
 531                 return -1;
 532         }
 533
 534         return 0;
 535 }
 536
 537 static int
 538 open_single_tx_pcap(const char *pcap_filename, pcap_dumper_t **dumper)
 539 {
 540         pcap_t *tx_pcap;
 541
 542         /*
 543          * We need to create a dummy empty pcap_t to use it
 544          * with pcap_dump_open(). We create big enough an Ethernet
 545          * pcap holder.
 546          */
 547         tx_pcap = pcap_open_dead_with_tstamp_precision(DLT_EN10MB,
 548                         RTE_ETH_PCAP_SNAPSHOT_LEN, PCAP_TSTAMP_PRECISION_NANO);
 549         if (tx_pcap == NULL) {
 550                 PMD_LOG(ERR, "Couldn't create dead pcap");
 551                 return -1;
 552         }
 553
 554         /* The dumper is created using the previous pcap_t reference */
 555         *dumper = pcap_dump_open(tx_pcap, pcap_filename);
 556         if (*dumper == NULL) {
 557                 pcap_close(tx_pcap);
 558                 PMD_LOG(ERR, "Couldn't open %s for writing.",
 559                         pcap_filename);
 560                 return -1;
 561         }
 562
 563         pcap_close(tx_pcap);
 564         return 0;
 565 }
 566
 567 static int
 568 open_single_rx_pcap(const char *pcap_filename, pcap_t **pcap)
 569 {
 570         *pcap = pcap_open_offline(pcap_filename, errbuf);
 571         if (*pcap == NULL) {
 572                 PMD_LOG(ERR, "Couldn't open %s: %s", pcap_filename,
 573                         errbuf);
 574                 return -1;
 575         }
 576
 577         return 0;
 578 }
 579
 580 static uint64_t
 581 count_packets_in_pcap(pcap_t **pcap, struct pcap_rx_queue *pcap_q)
 582 {
 583         const u_char *packet;
 584         struct pcap_pkthdr header;
 585         uint64_t pcap_pkt_count = 0;
 586
 587         while ((packet = pcap_next(*pcap, &header)))
 588                 pcap_pkt_count++;
 589
 590         /* The pcap is reopened so it can be used as normal later. */
 591         pcap_close(*pcap);
 592         *pcap = NULL;
 593         open_single_rx_pcap(pcap_q->name, pcap);
 594
 595         return pcap_pkt_count;
 596 }
 597
 598 static int
 599 eth_dev_start(struct rte_eth_dev *dev)
 600 {
 601         unsigned int i;
 602         struct pmd_internals *internals = dev->data->dev_private;
 603         struct pmd_process_private *pp = dev->process_private;
 604         struct pcap_tx_queue *tx;
 605         struct pcap_rx_queue *rx;
 606
 607         /* Special iface case. Single pcap is open and shared between tx/rx. */
 608         if (internals->single_iface) {
 609                 tx = &internals->tx_queue[0];
 610                 rx = &internals->rx_queue[0];
 611
 612                 if (!pp->tx_pcap[0] &&
 613                         strcmp(tx->type, ETH_PCAP_IFACE_ARG) == 0) {
 614                         if (open_single_iface(tx->name, &pp->tx_pcap[0]) < 0)
 615                                 return -1;
 616                         pp->rx_pcap[0] = pp->tx_pcap[0];
 617                 }
 618
 619                 goto status_up;
 620         }
 621
 622         /* If not open already, open tx pcaps/dumpers */
 623         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 624                 tx = &internals->tx_queue[i];
 625
 626                 if (!pp->tx_dumper[i] &&
 627                                 strcmp(tx->type, ETH_PCAP_TX_PCAP_ARG) == 0) {
 628                         if (open_single_tx_pcap(tx->name,
 629                                 &pp->tx_dumper[i]) < 0)
 630                                 return -1;
 631                 } else if (!pp->tx_pcap[i] &&
 632                                 strcmp(tx->type, ETH_PCAP_TX_IFACE_ARG) == 0) {
 633                         if (open_single_iface(tx->name, &pp->tx_pcap[i]) < 0)
 634                                 return -1;
 635                 }
 636         }
 637
 638         /* If not open already, open rx pcaps */
 639         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 640                 rx = &internals->rx_queue[i];
 641
 642                 if (pp->rx_pcap[i] != NULL)
 643                         continue;
 644
 645                 if (strcmp(rx->type, ETH_PCAP_RX_PCAP_ARG) == 0) {
 646                         if (open_single_rx_pcap(rx->name, &pp->rx_pcap[i]) < 0)
 647                                 return -1;
 648                 } else if (strcmp(rx->type, ETH_PCAP_RX_IFACE_ARG) == 0) {
 649                         if (open_single_iface(rx->name, &pp->rx_pcap[i]) < 0)
 650                                 return -1;
 651                 }
 652         }
 653
 654 status_up:
 655         for (i = 0; i < dev->data->nb_rx_queues; i++)
 656                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 657
 658         for (i = 0; i < dev->data->nb_tx_queues; i++)
 659                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 660
 661         dev->data->dev_link.link_status = ETH_LINK_UP;
 662
 663         return 0;
 664 }
 665
 666 /*
 667  * This function gets called when the current port gets stopped.
 668  * Is the only place for us to close all the tx streams dumpers.
 669  * If not called the dumpers will be flushed within each tx burst.
 670  */
 671 static int
 672 eth_dev_stop(struct rte_eth_dev *dev)
 673 {
 674         unsigned int i;
 675         struct pmd_internals *internals = dev->data->dev_private;
 676         struct pmd_process_private *pp = dev->process_private;
 677
 678         /* Special iface case. Single pcap is open and shared between tx/rx. */
 679         if (internals->single_iface) {
 680                 queue_missed_stat_on_stop_update(dev, 0);
 681                 if (pp->tx_pcap[0] != NULL) {
 682                         pcap_close(pp->tx_pcap[0]);
 683                         pp->tx_pcap[0] = NULL;
 684                         pp->rx_pcap[0] = NULL;
 685                 }
 686                 goto status_down;
 687         }
 688
 689         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 690                 if (pp->tx_dumper[i] != NULL) {
 691                         pcap_dump_close(pp->tx_dumper[i]);
 692                         pp->tx_dumper[i] = NULL;
 693                 }
 694
 695                 if (pp->tx_pcap[i] != NULL) {
 696                         pcap_close(pp->tx_pcap[i]);
 697                         pp->tx_pcap[i] = NULL;
 698                 }
 699         }
 700
 701         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 702                 if (pp->rx_pcap[i] != NULL) {
 703                         queue_missed_stat_on_stop_update(dev, i);
 704                         pcap_close(pp->rx_pcap[i]);
 705                         pp->rx_pcap[i] = NULL;
 706                 }
 707         }
 708
 709 status_down:
 710         for (i = 0; i < dev->data->nb_rx_queues; i++)
 711                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 712
 713         for (i = 0; i < dev->data->nb_tx_queues; i++)
 714                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 715
 716         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 717
 718         return 0;
 719 }
 720
 721 static int
 722 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 723 {
 724         return 0;
 725 }
 726
 727 static int
 728 eth_dev_info(struct rte_eth_dev *dev,
 729                 struct rte_eth_dev_info *dev_info)
 730 {
 731         struct pmd_internals *internals = dev->data->dev_private;
 732
 733         dev_info->if_index = internals->if_index;
 734         dev_info->max_mac_addrs = 1;
 735         dev_info->max_rx_pktlen = (uint32_t) -1;
 736         dev_info->max_rx_queues = dev->data->nb_rx_queues;
 737         dev_info->max_tx_queues = dev->data->nb_tx_queues;
 738         dev_info->min_rx_bufsize = 0;
 739
 740         return 0;
 741 }
 742
 743 static int
 744 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 745 {
 746         unsigned int i;
 747         unsigned long rx_packets_total = 0, rx_bytes_total = 0;
 748         unsigned long rx_missed_total = 0;
 749         unsigned long tx_packets_total = 0, tx_bytes_total = 0;
 750         unsigned long tx_packets_err_total = 0;
 751         const struct pmd_internals *internal = dev->data->dev_private;
 752
 753         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 754                         i < dev->data->nb_rx_queues; i++) {
 755                 stats->q_ipackets[i] = internal->rx_queue[i].rx_stat.pkts;
 756                 stats->q_ibytes[i] = internal->rx_queue[i].rx_stat.bytes;
 757                 rx_packets_total += stats->q_ipackets[i];
 758                 rx_bytes_total += stats->q_ibytes[i];
 759                 rx_missed_total += queue_missed_stat_get(dev, i);
 760         }
 761
 762         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 763                         i < dev->data->nb_tx_queues; i++) {
 764                 stats->q_opackets[i] = internal->tx_queue[i].tx_stat.pkts;
 765                 stats->q_obytes[i] = internal->tx_queue[i].tx_stat.bytes;
 766                 tx_packets_total += stats->q_opackets[i];
 767                 tx_bytes_total += stats->q_obytes[i];
 768                 tx_packets_err_total += internal->tx_queue[i].tx_stat.err_pkts;
 769         }
 770
 771         stats->ipackets = rx_packets_total;
 772         stats->ibytes = rx_bytes_total;
 773         stats->imissed = rx_missed_total;
 774         stats->opackets = tx_packets_total;
 775         stats->obytes = tx_bytes_total;
 776         stats->oerrors = tx_packets_err_total;
 777
 778         return 0;
 779 }
 780
 781 static int
 782 eth_stats_reset(struct rte_eth_dev *dev)
 783 {
 784         unsigned int i;
 785         struct pmd_internals *internal = dev->data->dev_private;
 786
 787         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 788                 internal->rx_queue[i].rx_stat.pkts = 0;
 789                 internal->rx_queue[i].rx_stat.bytes = 0;
 790                 queue_missed_stat_reset(dev, i);
 791         }
 792
 793         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 794                 internal->tx_queue[i].tx_stat.pkts = 0;
 795                 internal->tx_queue[i].tx_stat.bytes = 0;
 796                 internal->tx_queue[i].tx_stat.err_pkts = 0;
 797         }
 798
 799         return 0;
 800 }
 801
 802 static inline void
 803 infinite_rx_ring_free(struct rte_ring *pkts)
 804 {
 805         struct rte_mbuf *bufs;
 806
 807         while (!rte_ring_dequeue(pkts, (void **)&bufs))
 808                 rte_pktmbuf_free(bufs);
 809
 810         rte_ring_free(pkts);
 811 }
 812
 813 static int
 814 eth_dev_close(struct rte_eth_dev *dev)
 815 {
 816         unsigned int i;
 817         struct pmd_internals *internals = dev->data->dev_private;
 818
 819         PMD_LOG(INFO, "Closing pcap ethdev on NUMA socket %d",
 820                         rte_socket_id());
 821
 822         eth_dev_stop(dev);
 823
 824         rte_free(dev->process_private);
 825
 826         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 827                 return 0;
 828
 829         /* Device wide flag, but cleanup must be performed per queue. */
 830         if (internals->infinite_rx) {
 831                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
 832                         struct pcap_rx_queue *pcap_q = &internals->rx_queue[i];
 833
 834                         /*
 835                          * 'pcap_q->pkts' can be NULL if 'eth_dev_close()'
 836                          * called before 'eth_rx_queue_setup()' has been called
 837                          */
 838                         if (pcap_q->pkts == NULL)
 839                                 continue;
 840
 841                         infinite_rx_ring_free(pcap_q->pkts);
 842                 }
 843         }
 844
 845         if (internals->phy_mac == 0)
 846                 /* not dynamically allocated, must not be freed */
 847                 dev->data->mac_addrs = NULL;
 848
 849         return 0;
 850 }
 851
 852 static void
 853 eth_queue_release(void *q __rte_unused)
 854 {
 855 }
 856
 857 static int
 858 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 859                 int wait_to_complete __rte_unused)
 860 {
 861         return 0;
 862 }
 863
 864 static int
 865 eth_rx_queue_setup(struct rte_eth_dev *dev,
 866                 uint16_t rx_queue_id,
 867                 uint16_t nb_rx_desc __rte_unused,
 868                 unsigned int socket_id __rte_unused,
 869                 const struct rte_eth_rxconf *rx_conf __rte_unused,
 870                 struct rte_mempool *mb_pool)
 871 {
 872         struct pmd_internals *internals = dev->data->dev_private;
 873         struct pcap_rx_queue *pcap_q = &internals->rx_queue[rx_queue_id];
 874
 875         pcap_q->mb_pool = mb_pool;
 876         pcap_q->port_id = dev->data->port_id;
 877         pcap_q->queue_id = rx_queue_id;
 878         dev->data->rx_queues[rx_queue_id] = pcap_q;
 879
 880         if (internals->infinite_rx) {
 881                 struct pmd_process_private *pp;
 882                 char ring_name[RTE_RING_NAMESIZE];
 883                 static uint32_t ring_number;
 884                 uint64_t pcap_pkt_count = 0;
 885                 struct rte_mbuf *bufs[1];
 886                 pcap_t **pcap;
 887
 888                 pp = rte_eth_devices[pcap_q->port_id].process_private;
 889                 pcap = &pp->rx_pcap[pcap_q->queue_id];
 890
 891                 if (unlikely(*pcap == NULL))
 892                         return -ENOENT;
 893
 894                 pcap_pkt_count = count_packets_in_pcap(pcap, pcap_q);
 895
 896                 snprintf(ring_name, sizeof(ring_name), "PCAP_RING%" PRIu32,
 897                                 ring_number);
 898
 899                 pcap_q->pkts = rte_ring_create(ring_name,
 900                                 rte_align64pow2(pcap_pkt_count + 1), 0,
 901                                 RING_F_SP_ENQ | RING_F_SC_DEQ);
 902                 ring_number++;
 903                 if (!pcap_q->pkts)
 904                         return -ENOENT;
 905
 906                 /* Fill ring with packets from PCAP file one by one. */
 907                 while (eth_pcap_rx(pcap_q, bufs, 1)) {
 908                         /* Check for multiseg mbufs. */
 909                         if (bufs[0]->nb_segs != 1) {
 910                                 infinite_rx_ring_free(pcap_q->pkts);
 911                                 PMD_LOG(ERR,
 912                                         "Multiseg mbufs are not supported in infinite_rx mode.");
 913                                 return -EINVAL;
 914                         }
 915
 916                         rte_ring_enqueue_bulk(pcap_q->pkts,
 917                                         (void * const *)bufs, 1, NULL);
 918                 }
 919
 920                 if (rte_ring_count(pcap_q->pkts) < pcap_pkt_count) {
 921                         infinite_rx_ring_free(pcap_q->pkts);
 922                         PMD_LOG(ERR,
 923                                 "Not enough mbufs to accommodate packets in pcap file. "
 924                                 "At least %" PRIu64 " mbufs per queue is required.",
 925                                 pcap_pkt_count);
 926                         return -EINVAL;
 927                 }
 928
 929                 /*
 930                  * Reset the stats for this queue since eth_pcap_rx calls above
 931                  * didn't result in the application receiving packets.
 932                  */
 933                 pcap_q->rx_stat.pkts = 0;
 934                 pcap_q->rx_stat.bytes = 0;
 935         }
 936
 937         return 0;
 938 }
 939
 940 static int
 941 eth_tx_queue_setup(struct rte_eth_dev *dev,
 942                 uint16_t tx_queue_id,
 943                 uint16_t nb_tx_desc __rte_unused,
 944                 unsigned int socket_id __rte_unused,
 945                 const struct rte_eth_txconf *tx_conf __rte_unused)
 946 {
 947         struct pmd_internals *internals = dev->data->dev_private;
 948         struct pcap_tx_queue *pcap_q = &internals->tx_queue[tx_queue_id];
 949
 950         pcap_q->port_id = dev->data->port_id;
 951         pcap_q->queue_id = tx_queue_id;
 952         dev->data->tx_queues[tx_queue_id] = pcap_q;
 953
 954         return 0;
 955 }
 956
 957 static int
 958 eth_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 959 {
 960         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
 961
 962         return 0;
 963 }
 964
 965 static int
 966 eth_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
 967 {
 968         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
 969
 970         return 0;
 971 }
 972
 973 static int
 974 eth_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 975 {
 976         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
 977
 978         return 0;
 979 }
 980
 981 static int
 982 eth_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
 983 {
 984         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
 985
 986         return 0;
 987 }
 988
 989 static const struct eth_dev_ops ops = {
 990         .dev_start = eth_dev_start,
 991         .dev_stop = eth_dev_stop,
 992         .dev_close = eth_dev_close,
 993         .dev_configure = eth_dev_configure,
 994         .dev_infos_get = eth_dev_info,
 995         .rx_queue_setup = eth_rx_queue_setup,
 996         .tx_queue_setup = eth_tx_queue_setup,
 997         .rx_queue_start = eth_rx_queue_start,
 998         .tx_queue_start = eth_tx_queue_start,
 999         .rx_queue_stop = eth_rx_queue_stop,
1000         .tx_queue_stop = eth_tx_queue_stop,
1001         .rx_queue_release = eth_queue_release,
1002         .tx_queue_release = eth_queue_release,
1003         .link_update = eth_link_update,
1004         .stats_get = eth_stats_get,
1005         .stats_reset = eth_stats_reset,
1006 };
1007
1008 static int
1009 add_queue(struct pmd_devargs *pmd, const char *name, const char *type,
1010                 pcap_t *pcap, pcap_dumper_t *dumper)
1011 {
1012         if (pmd->num_of_queue >= RTE_PMD_PCAP_MAX_QUEUES)
1013                 return -1;
1014         if (pcap)
1015                 pmd->queue[pmd->num_of_queue].pcap = pcap;
1016         if (dumper)
1017                 pmd->queue[pmd->num_of_queue].dumper = dumper;
1018         pmd->queue[pmd->num_of_queue].name = name;
1019         pmd->queue[pmd->num_of_queue].type = type;
1020         pmd->num_of_queue++;
1021         return 0;
1022 }
1023
1024 /*
1025  * Function handler that opens the pcap file for reading a stores a
1026  * reference of it for use it later on.
1027  */
1028 static int
1029 open_rx_pcap(const char *key, const char *value, void *extra_args)
1030 {
1031         const char *pcap_filename = value;
1032         struct pmd_devargs *rx = extra_args;
1033         pcap_t *pcap = NULL;
1034
1035         if (open_single_rx_pcap(pcap_filename, &pcap) < 0)
1036                 return -1;
1037
1038         if (add_queue(rx, pcap_filename, key, pcap, NULL) < 0) {
1039                 pcap_close(pcap);
1040                 return -1;
1041         }
1042
1043         return 0;
1044 }
1045
1046 /*
1047  * Opens a pcap file for writing and stores a reference to it
1048  * for use it later on.
1049  */
1050 static int
1051 open_tx_pcap(const char *key, const char *value, void *extra_args)
1052 {
1053         const char *pcap_filename = value;
1054         struct pmd_devargs *dumpers = extra_args;
1055         pcap_dumper_t *dumper;
1056
1057         if (open_single_tx_pcap(pcap_filename, &dumper) < 0)
1058                 return -1;
1059
1060         if (add_queue(dumpers, pcap_filename, key, NULL, dumper) < 0) {
1061                 pcap_dump_close(dumper);
1062                 return -1;
1063         }
1064
1065         return 0;
1066 }
1067
1068 /*
1069  * Opens an interface for reading and writing
1070  */
1071 static inline int
1072 open_rx_tx_iface(const char *key, const char *value, void *extra_args)
1073 {
1074         const char *iface = value;
1075         struct pmd_devargs *tx = extra_args;
1076         pcap_t *pcap = NULL;
1077
1078         if (open_single_iface(iface, &pcap) < 0)
1079                 return -1;
1080
1081         tx->queue[0].pcap = pcap;
1082         tx->queue[0].name = iface;
1083         tx->queue[0].type = key;
1084
1085         return 0;
1086 }
1087
1088 static inline int
1089 set_iface_direction(const char *iface, pcap_t *pcap,
1090                 pcap_direction_t direction)
1091 {
1092         const char *direction_str = (direction == PCAP_D_IN) ? "IN" : "OUT";
1093         if (pcap_setdirection(pcap, direction) < 0) {
1094                 PMD_LOG(ERR, "Setting %s pcap direction %s failed - %s\n",
1095                                 iface, direction_str, pcap_geterr(pcap));
1096                 return -1;
1097         }
1098         PMD_LOG(INFO, "Setting %s pcap direction %s\n",
1099                         iface, direction_str);
1100         return 0;
1101 }
1102
1103 static inline int
1104 open_iface(const char *key, const char *value, void *extra_args)
1105 {
1106         const char *iface = value;
1107         struct pmd_devargs *pmd = extra_args;
1108         pcap_t *pcap = NULL;
1109
1110         if (open_single_iface(iface, &pcap) < 0)
1111                 return -1;
1112         if (add_queue(pmd, iface, key, pcap, NULL) < 0) {
1113                 pcap_close(pcap);
1114                 return -1;
1115         }
1116
1117         return 0;
1118 }
1119
1120 /*
1121  * Opens a NIC for reading packets from it
1122  */
1123 static inline int
1124 open_rx_iface(const char *key, const char *value, void *extra_args)
1125 {
1126         int ret = open_iface(key, value, extra_args);
1127         if (ret < 0)
1128                 return ret;
1129         if (strcmp(key, ETH_PCAP_RX_IFACE_IN_ARG) == 0) {
1130                 struct pmd_devargs *pmd = extra_args;
1131                 unsigned int qid = pmd->num_of_queue - 1;
1132
1133                 set_iface_direction(pmd->queue[qid].name,
1134                                 pmd->queue[qid].pcap,
1135                                 PCAP_D_IN);
1136         }
1137
1138         return 0;
1139 }
1140
1141 static inline int
1142 rx_iface_args_process(const char *key, const char *value, void *extra_args)
1143 {
1144         if (strcmp(key, ETH_PCAP_RX_IFACE_ARG) == 0 ||
1145                         strcmp(key, ETH_PCAP_RX_IFACE_IN_ARG) == 0)
1146                 return open_rx_iface(key, value, extra_args);
1147
1148         return 0;
1149 }
1150
1151 /*
1152  * Opens a NIC for writing packets to it
1153  */
1154 static int
1155 open_tx_iface(const char *key, const char *value, void *extra_args)
1156 {
1157         return open_iface(key, value, extra_args);
1158 }
1159
1160 static int
1161 select_phy_mac(const char *key __rte_unused, const char *value,
1162                 void *extra_args)
1163 {
1164         if (extra_args) {
1165                 const int phy_mac = atoi(value);
1166                 int *enable_phy_mac = extra_args;
1167
1168                 if (phy_mac)
1169                         *enable_phy_mac = 1;
1170         }
1171         return 0;
1172 }
1173
1174 static int
1175 get_infinite_rx_arg(const char *key __rte_unused,
1176                 const char *value, void *extra_args)
1177 {
1178         if (extra_args) {
1179                 const int infinite_rx = atoi(value);
1180                 int *enable_infinite_rx = extra_args;
1181
1182                 if (infinite_rx > 0)
1183                         *enable_infinite_rx = 1;
1184         }
1185         return 0;
1186 }
1187
1188 static int
1189 pmd_init_internals(struct rte_vdev_device *vdev,
1190                 const unsigned int nb_rx_queues,
1191                 const unsigned int nb_tx_queues,
1192                 struct pmd_internals **internals,
1193                 struct rte_eth_dev **eth_dev)
1194 {
1195         struct rte_eth_dev_data *data;
1196         struct pmd_process_private *pp;
1197         unsigned int numa_node = vdev->device.numa_node;
1198
1199         PMD_LOG(INFO, "Creating pcap-backed ethdev on numa socket %d",
1200                 numa_node);
1201
1202         pp = (struct pmd_process_private *)
1203                 rte_zmalloc(NULL, sizeof(struct pmd_process_private),
1204                                 RTE_CACHE_LINE_SIZE);
1205
1206         if (pp == NULL) {
1207                 PMD_LOG(ERR,
1208                         "Failed to allocate memory for process private");
1209                 return -1;
1210         }
1211
1212         /* reserve an ethdev entry */
1213         *eth_dev = rte_eth_vdev_allocate(vdev, sizeof(**internals));
1214         if (!(*eth_dev)) {
1215                 rte_free(pp);
1216                 return -1;
1217         }
1218         (*eth_dev)->process_private = pp;
1219         /* now put it all together
1220          * - store queue data in internals,
1221          * - store numa_node info in eth_dev
1222          * - point eth_dev_data to internals
1223          * - and point eth_dev structure to new eth_dev_data structure
1224          */
1225         *internals = (*eth_dev)->data->dev_private;
1226         /*
1227          * Interface MAC = 02:70:63:61:70:<iface_idx>
1228          * derived from: 'locally administered':'p':'c':'a':'p':'iface_idx'
1229          * where the middle 4 characters are converted to hex.
1230          */
1231         (*internals)->eth_addr = (struct rte_ether_addr) {
1232                 .addr_bytes = { 0x02, 0x70, 0x63, 0x61, 0x70, iface_idx++ }
1233         };
1234         (*internals)->phy_mac = 0;
1235         data = (*eth_dev)->data;
1236         data->nb_rx_queues = (uint16_t)nb_rx_queues;
1237         data->nb_tx_queues = (uint16_t)nb_tx_queues;
1238         data->dev_link = pmd_link;
1239         data->mac_addrs = &(*internals)->eth_addr;
1240         data->promiscuous = 1;
1241         data->all_multicast = 1;
1242         data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
1243
1244         /*
1245          * NOTE: we'll replace the data element, of originally allocated
1246          * eth_dev so the rings are local per-process
1247          */
1248         (*eth_dev)->dev_ops = &ops;
1249
1250         strlcpy((*internals)->devargs, rte_vdev_device_args(vdev),
1251                         ETH_PCAP_ARG_MAXLEN);
1252
1253         return 0;
1254 }
1255
1256 static int
1257 eth_pcap_update_mac(const char *if_name, struct rte_eth_dev *eth_dev,
1258                 const unsigned int numa_node)
1259 {
1260         void *mac_addrs;
1261         struct rte_ether_addr mac;
1262
1263         if (osdep_iface_mac_get(if_name, &mac) < 0)
1264                 return -1;
1265
1266         mac_addrs = rte_zmalloc_socket(NULL, RTE_ETHER_ADDR_LEN, 0, numa_node);
1267         if (mac_addrs == NULL)
1268                 return -1;
1269
1270         PMD_LOG(INFO, "Setting phy MAC for %s", if_name);
1271         rte_memcpy(mac_addrs, mac.addr_bytes, RTE_ETHER_ADDR_LEN);
1272         eth_dev->data->mac_addrs = mac_addrs;
1273         return 0;
1274 }
1275
1276 static int
1277 eth_from_pcaps_common(struct rte_vdev_device *vdev,
1278                 struct pmd_devargs_all *devargs_all,
1279                 struct pmd_internals **internals, struct rte_eth_dev **eth_dev)
1280 {
1281         struct pmd_process_private *pp;
1282         struct pmd_devargs *rx_queues = &devargs_all->rx_queues;
1283         struct pmd_devargs *tx_queues = &devargs_all->tx_queues;
1284         const unsigned int nb_rx_queues = rx_queues->num_of_queue;
1285         const unsigned int nb_tx_queues = tx_queues->num_of_queue;
1286         unsigned int i;
1287
1288         if (pmd_init_internals(vdev, nb_rx_queues, nb_tx_queues, internals,
1289                         eth_dev) < 0)
1290                 return -1;
1291
1292         pp = (*eth_dev)->process_private;
1293         for (i = 0; i < nb_rx_queues; i++) {
1294                 struct pcap_rx_queue *rx = &(*internals)->rx_queue[i];
1295                 struct devargs_queue *queue = &rx_queues->queue[i];
1296
1297                 pp->rx_pcap[i] = queue->pcap;
1298                 strlcpy(rx->name, queue->name, sizeof(rx->name));
1299                 strlcpy(rx->type, queue->type, sizeof(rx->type));
1300         }
1301
1302         for (i = 0; i < nb_tx_queues; i++) {
1303                 struct pcap_tx_queue *tx = &(*internals)->tx_queue[i];
1304                 struct devargs_queue *queue = &tx_queues->queue[i];
1305
1306                 pp->tx_dumper[i] = queue->dumper;
1307                 pp->tx_pcap[i] = queue->pcap;
1308                 strlcpy(tx->name, queue->name, sizeof(tx->name));
1309                 strlcpy(tx->type, queue->type, sizeof(tx->type));
1310         }
1311
1312         return 0;
1313 }
1314
1315 static int
1316 eth_from_pcaps(struct rte_vdev_device *vdev,
1317                 struct pmd_devargs_all *devargs_all)
1318 {
1319         struct pmd_internals *internals = NULL;
1320         struct rte_eth_dev *eth_dev = NULL;
1321         struct pmd_devargs *rx_queues = &devargs_all->rx_queues;
1322         int single_iface = devargs_all->single_iface;
1323         unsigned int infinite_rx = devargs_all->infinite_rx;
1324         int ret;
1325
1326         ret = eth_from_pcaps_common(vdev, devargs_all, &internals, &eth_dev);
1327
1328         if (ret < 0)
1329                 return ret;
1330
1331         /* store weather we are using a single interface for rx/tx or not */
1332         internals->single_iface = single_iface;
1333
1334         if (single_iface) {
1335                 internals->if_index =
1336                         osdep_iface_index_get(rx_queues->queue[0].name);
1337
1338                 /* phy_mac arg is applied only only if "iface" devarg is provided */
1339                 if (rx_queues->phy_mac) {
1340                         if (eth_pcap_update_mac(rx_queues->queue[0].name,
1341                                         eth_dev, vdev->device.numa_node) == 0)
1342                                 internals->phy_mac = 1;
1343                 }
1344         }
1345
1346         internals->infinite_rx = infinite_rx;
1347         /* Assign rx ops. */
1348         if (infinite_rx)
1349                 eth_dev->rx_pkt_burst = eth_pcap_rx_infinite;
1350         else if (devargs_all->is_rx_pcap || devargs_all->is_rx_iface ||
1351                         single_iface)
1352                 eth_dev->rx_pkt_burst = eth_pcap_rx;
1353         else
1354                 eth_dev->rx_pkt_burst = eth_null_rx;
1355
1356         /* Assign tx ops. */
1357         if (devargs_all->is_tx_pcap)
1358                 eth_dev->tx_pkt_burst = eth_pcap_tx_dumper;
1359         else if (devargs_all->is_tx_iface || single_iface)
1360                 eth_dev->tx_pkt_burst = eth_pcap_tx;
1361         else
1362                 eth_dev->tx_pkt_burst = eth_tx_drop;
1363
1364         rte_eth_dev_probing_finish(eth_dev);
1365         return 0;
1366 }
1367
1368 static int
1369 pmd_pcap_probe(struct rte_vdev_device *dev)
1370 {
1371         const char *name;
1372         struct rte_kvargs *kvlist;
1373         struct pmd_devargs pcaps = {0};
1374         struct pmd_devargs dumpers = {0};
1375         struct rte_eth_dev *eth_dev =  NULL;
1376         struct pmd_internals *internal;
1377         int ret = 0;
1378
1379         struct pmd_devargs_all devargs_all = {
1380                 .single_iface = 0,
1381                 .is_tx_pcap = 0,
1382                 .is_tx_iface = 0,
1383                 .infinite_rx = 0,
1384         };
1385
1386         name = rte_vdev_device_name(dev);
1387         PMD_LOG(INFO, "Initializing pmd_pcap for %s", name);
1388
1389         timespec_get(&start_time, TIME_UTC);
1390         start_cycles = rte_get_timer_cycles();
1391         hz = rte_get_timer_hz();
1392
1393         ret = rte_mbuf_dyn_rx_timestamp_register(&timestamp_dynfield_offset,
1394                         &timestamp_rx_dynflag);
1395         if (ret != 0) {
1396                 PMD_LOG(ERR, "Failed to register Rx timestamp field/flag");
1397                 return -1;
1398         }
1399
1400         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1401                 eth_dev = rte_eth_dev_attach_secondary(name);
1402                 if (!eth_dev) {
1403                         PMD_LOG(ERR, "Failed to probe %s", name);
1404                         return -1;
1405                 }
1406
1407                 internal = eth_dev->data->dev_private;
1408
1409                 kvlist = rte_kvargs_parse(internal->devargs, valid_arguments);
1410                 if (kvlist == NULL)
1411                         return -1;
1412         } else {
1413                 kvlist = rte_kvargs_parse(rte_vdev_device_args(dev),
1414                                 valid_arguments);
1415                 if (kvlist == NULL)
1416                         return -1;
1417         }
1418
1419         /*
1420          * If iface argument is passed we open the NICs and use them for
1421          * reading / writing
1422          */
1423         if (rte_kvargs_count(kvlist, ETH_PCAP_IFACE_ARG) == 1) {
1424
1425                 ret = rte_kvargs_process(kvlist, ETH_PCAP_IFACE_ARG,
1426                                 &open_rx_tx_iface, &pcaps);
1427                 if (ret < 0)
1428                         goto free_kvlist;
1429
1430                 dumpers.queue[0] = pcaps.queue[0];
1431
1432                 ret = rte_kvargs_process(kvlist, ETH_PCAP_PHY_MAC_ARG,
1433                                 &select_phy_mac, &pcaps.phy_mac);
1434                 if (ret < 0)
1435                         goto free_kvlist;
1436
1437                 dumpers.phy_mac = pcaps.phy_mac;
1438
1439                 devargs_all.single_iface = 1;
1440                 pcaps.num_of_queue = 1;
1441                 dumpers.num_of_queue = 1;
1442
1443                 goto create_eth;
1444         }
1445
1446         /*
1447          * We check whether we want to open a RX stream from a real NIC, a
1448          * pcap file or open a dummy RX stream
1449          */
1450         devargs_all.is_rx_pcap =
1451                 rte_kvargs_count(kvlist, ETH_PCAP_RX_PCAP_ARG) ? 1 : 0;
1452         devargs_all.is_rx_iface =
1453                 (rte_kvargs_count(kvlist, ETH_PCAP_RX_IFACE_ARG) +
1454                  rte_kvargs_count(kvlist, ETH_PCAP_RX_IFACE_IN_ARG)) ? 1 : 0;
1455         pcaps.num_of_queue = 0;
1456
1457         devargs_all.is_tx_pcap =
1458                 rte_kvargs_count(kvlist, ETH_PCAP_TX_PCAP_ARG) ? 1 : 0;
1459         devargs_all.is_tx_iface =
1460                 rte_kvargs_count(kvlist, ETH_PCAP_TX_IFACE_ARG) ? 1 : 0;
1461         dumpers.num_of_queue = 0;
1462
1463         if (devargs_all.is_rx_pcap) {
1464                 /*
1465                  * We check whether we want to infinitely rx the pcap file.
1466                  */
1467                 unsigned int infinite_rx_arg_cnt = rte_kvargs_count(kvlist,
1468                                 ETH_PCAP_INFINITE_RX_ARG);
1469
1470                 if (infinite_rx_arg_cnt == 1) {
1471                         ret = rte_kvargs_process(kvlist,
1472                                         ETH_PCAP_INFINITE_RX_ARG,
1473                                         &get_infinite_rx_arg,
1474                                         &devargs_all.infinite_rx);
1475                         if (ret < 0)
1476                                 goto free_kvlist;
1477                         PMD_LOG(INFO, "infinite_rx has been %s for %s",
1478                                         devargs_all.infinite_rx ? "enabled" : "disabled",
1479                                         name);
1480
1481                 } else if (infinite_rx_arg_cnt > 1) {
1482                         PMD_LOG(WARNING, "infinite_rx has not been enabled since the "
1483                                         "argument has been provided more than once "
1484                                         "for %s", name);
1485                 }
1486
1487                 ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_PCAP_ARG,
1488                                 &open_rx_pcap, &pcaps);
1489         } else if (devargs_all.is_rx_iface) {
1490                 ret = rte_kvargs_process(kvlist, NULL,
1491                                 &rx_iface_args_process, &pcaps);
1492         } else if (devargs_all.is_tx_iface || devargs_all.is_tx_pcap) {
1493                 unsigned int i;
1494
1495                 /* Count number of tx queue args passed before dummy rx queue
1496                  * creation so a dummy rx queue can be created for each tx queue
1497                  */
1498                 unsigned int num_tx_queues =
1499                         (rte_kvargs_count(kvlist, ETH_PCAP_TX_PCAP_ARG) +
1500                         rte_kvargs_count(kvlist, ETH_PCAP_TX_IFACE_ARG));
1501
1502                 PMD_LOG(INFO, "Creating null rx queue since no rx queues were provided.");
1503
1504                 /* Creating a dummy rx queue for each tx queue passed */
1505                 for (i = 0; i < num_tx_queues; i++)
1506                         ret = add_queue(&pcaps, "dummy_rx", "rx_null", NULL,
1507                                         NULL);
1508         } else {
1509                 PMD_LOG(ERR, "Error - No rx or tx queues provided");
1510                 ret = -ENOENT;
1511         }
1512         if (ret < 0)
1513                 goto free_kvlist;
1514
1515         /*
1516          * We check whether we want to open a TX stream to a real NIC,
1517          * a pcap file, or drop packets on tx
1518          */
1519         if (devargs_all.is_tx_pcap) {
1520                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_PCAP_ARG,
1521                                 &open_tx_pcap, &dumpers);
1522         } else if (devargs_all.is_tx_iface) {
1523                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_IFACE_ARG,
1524                                 &open_tx_iface, &dumpers);
1525         } else {
1526                 unsigned int i;
1527
1528                 PMD_LOG(INFO, "Dropping packets on tx since no tx queues were provided.");
1529
1530                 /* Add 1 dummy queue per rxq which counts and drops packets. */
1531                 for (i = 0; i < pcaps.num_of_queue; i++)
1532                         ret = add_queue(&dumpers, "dummy_tx", "tx_drop", NULL,
1533                                         NULL);
1534         }
1535
1536         if (ret < 0)
1537                 goto free_kvlist;
1538
1539 create_eth:
1540         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1541                 struct pmd_process_private *pp;
1542                 unsigned int i;
1543
1544                 internal = eth_dev->data->dev_private;
1545                         pp = (struct pmd_process_private *)
1546                                 rte_zmalloc(NULL,
1547                                         sizeof(struct pmd_process_private),
1548                                         RTE_CACHE_LINE_SIZE);
1549
1550                 if (pp == NULL) {
1551                         PMD_LOG(ERR,
1552                                 "Failed to allocate memory for process private");
1553                         ret = -1;
1554                         goto free_kvlist;
1555                 }
1556
1557                 eth_dev->dev_ops = &ops;
1558                 eth_dev->device = &dev->device;
1559
1560                 /* setup process private */
1561                 for (i = 0; i < pcaps.num_of_queue; i++)
1562                         pp->rx_pcap[i] = pcaps.queue[i].pcap;
1563
1564                 for (i = 0; i < dumpers.num_of_queue; i++) {
1565                         pp->tx_dumper[i] = dumpers.queue[i].dumper;
1566                         pp->tx_pcap[i] = dumpers.queue[i].pcap;
1567                 }
1568
1569                 eth_dev->process_private = pp;
1570                 eth_dev->rx_pkt_burst = eth_pcap_rx;
1571                 if (devargs_all.is_tx_pcap)
1572                         eth_dev->tx_pkt_burst = eth_pcap_tx_dumper;
1573                 else
1574                         eth_dev->tx_pkt_burst = eth_pcap_tx;
1575
1576                 rte_eth_dev_probing_finish(eth_dev);
1577                 goto free_kvlist;
1578         }
1579
1580         devargs_all.rx_queues = pcaps;
1581         devargs_all.tx_queues = dumpers;
1582
1583         ret = eth_from_pcaps(dev, &devargs_all);
1584
1585 free_kvlist:
1586         rte_kvargs_free(kvlist);
1587
1588         return ret;
1589 }
1590
1591 static int
1592 pmd_pcap_remove(struct rte_vdev_device *dev)
1593 {
1594         struct rte_eth_dev *eth_dev = NULL;
1595
1596         if (!dev)
1597                 return -1;
1598
1599         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1600         if (eth_dev == NULL)
1601                 return 0; /* port already released */
1602
1603         eth_dev_close(eth_dev);
1604         rte_eth_dev_release_port(eth_dev);
1605
1606         return 0;
1607 }
1608
1609 static struct rte_vdev_driver pmd_pcap_drv = {
1610         .probe = pmd_pcap_probe,
1611         .remove = pmd_pcap_remove,
1612 };
1613
1614 RTE_PMD_REGISTER_VDEV(net_pcap, pmd_pcap_drv);
1615 RTE_PMD_REGISTER_ALIAS(net_pcap, eth_pcap);
1616 RTE_PMD_REGISTER_PARAM_STRING(net_pcap,
1617         ETH_PCAP_RX_PCAP_ARG "=<string> "
1618         ETH_PCAP_TX_PCAP_ARG "=<string> "
1619         ETH_PCAP_RX_IFACE_ARG "=<ifc> "
1620         ETH_PCAP_RX_IFACE_IN_ARG "=<ifc> "
1621         ETH_PCAP_TX_IFACE_ARG "=<ifc> "
1622         ETH_PCAP_IFACE_ARG "=<ifc> "
1623         ETH_PCAP_PHY_MAC_ARG "=<int>"
1624         ETH_PCAP_INFINITE_RX_ARG "=<0|1>");