drivers/net/pcap/rte_eth_pcap.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 <net/if.h>
  10 #include <sys/socket.h>
  11 #include <sys/ioctl.h>
  12 #include <unistd.h>
  13
  14 #if defined(RTE_EXEC_ENV_FREEBSD)
  15 #include <sys/sysctl.h>
  16 #include <net/if_dl.h>
  17 #endif
  18
  19 #include <pcap.h>
  20
  21 #include <rte_cycles.h>
  22 #include <rte_ethdev_driver.h>
  23 #include <rte_ethdev_vdev.h>
  24 #include <rte_kvargs.h>
  25 #include <rte_malloc.h>
  26 #include <rte_mbuf.h>
  27 #include <rte_bus_vdev.h>
  28 #include <rte_string_fns.h>
  29
  30 #define RTE_ETH_PCAP_SNAPSHOT_LEN 65535
  31 #define RTE_ETH_PCAP_SNAPLEN RTE_ETHER_MAX_JUMBO_FRAME_LEN
  32 #define RTE_ETH_PCAP_PROMISC 1
  33 #define RTE_ETH_PCAP_TIMEOUT -1
  34
  35 #define ETH_PCAP_RX_PCAP_ARG  "rx_pcap"
  36 #define ETH_PCAP_TX_PCAP_ARG  "tx_pcap"
  37 #define ETH_PCAP_RX_IFACE_ARG "rx_iface"
  38 #define ETH_PCAP_RX_IFACE_IN_ARG "rx_iface_in"
  39 #define ETH_PCAP_TX_IFACE_ARG "tx_iface"
  40 #define ETH_PCAP_IFACE_ARG    "iface"
  41 #define ETH_PCAP_PHY_MAC_ARG  "phy_mac"
  42 #define ETH_PCAP_INFINITE_RX_ARG  "infinite_rx"
  43
  44 #define ETH_PCAP_ARG_MAXLEN     64
  45
  46 #define RTE_PMD_PCAP_MAX_QUEUES 16
  47
  48 static char errbuf[PCAP_ERRBUF_SIZE];
  49 static unsigned char tx_pcap_data[RTE_ETH_PCAP_SNAPLEN];
  50 static struct timeval start_time;
  51 static uint64_t start_cycles;
  52 static uint64_t hz;
  53 static uint8_t iface_idx;
  54
  55 struct queue_stat {
  56         volatile unsigned long pkts;
  57         volatile unsigned long bytes;
  58         volatile unsigned long err_pkts;
  59 };
  60
  61 struct pcap_rx_queue {
  62         uint16_t port_id;
  63         uint16_t queue_id;
  64         struct rte_mempool *mb_pool;
  65         struct queue_stat rx_stat;
  66         char name[PATH_MAX];
  67         char type[ETH_PCAP_ARG_MAXLEN];
  68
  69         /* Contains pre-generated packets to be looped through */
  70         struct rte_ring *pkts;
  71 };
  72
  73 struct pcap_tx_queue {
  74         uint16_t port_id;
  75         uint16_t queue_id;
  76         struct queue_stat tx_stat;
  77         char name[PATH_MAX];
  78         char type[ETH_PCAP_ARG_MAXLEN];
  79 };
  80
  81 struct pmd_internals {
  82         struct pcap_rx_queue rx_queue[RTE_PMD_PCAP_MAX_QUEUES];
  83         struct pcap_tx_queue tx_queue[RTE_PMD_PCAP_MAX_QUEUES];
  84         char devargs[ETH_PCAP_ARG_MAXLEN];
  85         struct rte_ether_addr eth_addr;
  86         int if_index;
  87         int single_iface;
  88         int phy_mac;
  89         unsigned int infinite_rx;
  90 };
  91
  92 struct pmd_process_private {
  93         pcap_t *rx_pcap[RTE_PMD_PCAP_MAX_QUEUES];
  94         pcap_t *tx_pcap[RTE_PMD_PCAP_MAX_QUEUES];
  95         pcap_dumper_t *tx_dumper[RTE_PMD_PCAP_MAX_QUEUES];
  96 };
  97
  98 struct pmd_devargs {
  99         unsigned int num_of_queue;
 100         struct devargs_queue {
 101                 pcap_dumper_t *dumper;
 102                 pcap_t *pcap;
 103                 const char *name;
 104                 const char *type;
 105         } queue[RTE_PMD_PCAP_MAX_QUEUES];
 106         int phy_mac;
 107 };
 108
 109 struct pmd_devargs_all {
 110         struct pmd_devargs rx_queues;
 111         struct pmd_devargs tx_queues;
 112         int single_iface;
 113         unsigned int is_tx_pcap;
 114         unsigned int is_tx_iface;
 115         unsigned int is_rx_pcap;
 116         unsigned int is_rx_iface;
 117         unsigned int infinite_rx;
 118 };
 119
 120 static const char *valid_arguments[] = {
 121         ETH_PCAP_RX_PCAP_ARG,
 122         ETH_PCAP_TX_PCAP_ARG,
 123         ETH_PCAP_RX_IFACE_ARG,
 124         ETH_PCAP_RX_IFACE_IN_ARG,
 125         ETH_PCAP_TX_IFACE_ARG,
 126         ETH_PCAP_IFACE_ARG,
 127         ETH_PCAP_PHY_MAC_ARG,
 128         ETH_PCAP_INFINITE_RX_ARG,
 129         NULL
 130 };
 131
 132 static struct rte_eth_link pmd_link = {
 133                 .link_speed = ETH_SPEED_NUM_10G,
 134                 .link_duplex = ETH_LINK_FULL_DUPLEX,
 135                 .link_status = ETH_LINK_DOWN,
 136                 .link_autoneg = ETH_LINK_FIXED,
 137 };
 138
 139 static int eth_pcap_logtype;
 140
 141 #define PMD_LOG(level, fmt, args...) \
 142         rte_log(RTE_LOG_ ## level, eth_pcap_logtype, \
 143                 "%s(): " fmt "\n", __func__, ##args)
 144
 145 static int
 146 eth_pcap_rx_jumbo(struct rte_mempool *mb_pool, struct rte_mbuf *mbuf,
 147                 const u_char *data, uint16_t data_len)
 148 {
 149         /* Copy the first segment. */
 150         uint16_t len = rte_pktmbuf_tailroom(mbuf);
 151         struct rte_mbuf *m = mbuf;
 152
 153         rte_memcpy(rte_pktmbuf_append(mbuf, len), data, len);
 154         data_len -= len;
 155         data += len;
 156
 157         while (data_len > 0) {
 158                 /* Allocate next mbuf and point to that. */
 159                 m->next = rte_pktmbuf_alloc(mb_pool);
 160
 161                 if (unlikely(!m->next))
 162                         return -1;
 163
 164                 m = m->next;
 165
 166                 /* Headroom is not needed in chained mbufs. */
 167                 rte_pktmbuf_prepend(m, rte_pktmbuf_headroom(m));
 168                 m->pkt_len = 0;
 169                 m->data_len = 0;
 170
 171                 /* Copy next segment. */
 172                 len = RTE_MIN(rte_pktmbuf_tailroom(m), data_len);
 173                 rte_memcpy(rte_pktmbuf_append(m, len), data, len);
 174
 175                 mbuf->nb_segs++;
 176                 data_len -= len;
 177                 data += len;
 178         }
 179
 180         return mbuf->nb_segs;
 181 }
 182
 183 /* Copy data from mbuf chain to a buffer suitable for writing to a PCAP file. */
 184 static void
 185 eth_pcap_gather_data(unsigned char *data, struct rte_mbuf *mbuf)
 186 {
 187         uint16_t data_len = 0;
 188
 189         while (mbuf) {
 190                 rte_memcpy(data + data_len, rte_pktmbuf_mtod(mbuf, void *),
 191                         mbuf->data_len);
 192
 193                 data_len += mbuf->data_len;
 194                 mbuf = mbuf->next;
 195         }
 196 }
 197
 198 static uint16_t
 199 eth_pcap_rx_infinite(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 200 {
 201         int i;
 202         struct pcap_rx_queue *pcap_q = queue;
 203         uint32_t rx_bytes = 0;
 204
 205         if (unlikely(nb_pkts == 0))
 206                 return 0;
 207
 208         if (rte_pktmbuf_alloc_bulk(pcap_q->mb_pool, bufs, nb_pkts) != 0)
 209                 return 0;
 210
 211         for (i = 0; i < nb_pkts; i++) {
 212                 struct rte_mbuf *pcap_buf;
 213                 int err = rte_ring_dequeue(pcap_q->pkts, (void **)&pcap_buf);
 214                 if (err)
 215                         return i;
 216
 217                 rte_memcpy(rte_pktmbuf_mtod(bufs[i], void *),
 218                                 rte_pktmbuf_mtod(pcap_buf, void *),
 219                                 pcap_buf->data_len);
 220                 bufs[i]->data_len = pcap_buf->data_len;
 221                 bufs[i]->pkt_len = pcap_buf->pkt_len;
 222                 bufs[i]->port = pcap_q->port_id;
 223                 rx_bytes += pcap_buf->data_len;
 224
 225                 /* Enqueue packet back on ring to allow infinite rx. */
 226                 rte_ring_enqueue(pcap_q->pkts, pcap_buf);
 227         }
 228
 229         pcap_q->rx_stat.pkts += i;
 230         pcap_q->rx_stat.bytes += rx_bytes;
 231
 232         return i;
 233 }
 234
 235 static uint16_t
 236 eth_pcap_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 237 {
 238         unsigned int i;
 239         struct pcap_pkthdr header;
 240         struct pmd_process_private *pp;
 241         const u_char *packet;
 242         struct rte_mbuf *mbuf;
 243         struct pcap_rx_queue *pcap_q = queue;
 244         uint16_t num_rx = 0;
 245         uint32_t rx_bytes = 0;
 246         pcap_t *pcap;
 247
 248         pp = rte_eth_devices[pcap_q->port_id].process_private;
 249         pcap = pp->rx_pcap[pcap_q->queue_id];
 250
 251         if (unlikely(pcap == NULL || nb_pkts == 0))
 252                 return 0;
 253
 254         /* Reads the given number of packets from the pcap file one by one
 255          * and copies the packet data into a newly allocated mbuf to return.
 256          */
 257         for (i = 0; i < nb_pkts; i++) {
 258                 /* Get the next PCAP packet */
 259                 packet = pcap_next(pcap, &header);
 260                 if (unlikely(packet == NULL))
 261                         break;
 262
 263                 mbuf = rte_pktmbuf_alloc(pcap_q->mb_pool);
 264                 if (unlikely(mbuf == NULL))
 265                         break;
 266
 267                 if (header.caplen <= rte_pktmbuf_tailroom(mbuf)) {
 268                         /* pcap packet will fit in the mbuf, can copy it */
 269                         rte_memcpy(rte_pktmbuf_mtod(mbuf, void *), packet,
 270                                         header.caplen);
 271                         mbuf->data_len = (uint16_t)header.caplen;
 272                 } else {
 273                         /* Try read jumbo frame into multi mbufs. */
 274                         if (unlikely(eth_pcap_rx_jumbo(pcap_q->mb_pool,
 275                                                        mbuf,
 276                                                        packet,
 277                                                        header.caplen) == -1)) {
 278                                 rte_pktmbuf_free(mbuf);
 279                                 break;
 280                         }
 281                 }
 282
 283                 mbuf->pkt_len = (uint16_t)header.caplen;
 284                 mbuf->port = pcap_q->port_id;
 285                 bufs[num_rx] = mbuf;
 286                 num_rx++;
 287                 rx_bytes += header.caplen;
 288         }
 289         pcap_q->rx_stat.pkts += num_rx;
 290         pcap_q->rx_stat.bytes += rx_bytes;
 291
 292         return num_rx;
 293 }
 294
 295 static uint16_t
 296 eth_null_rx(void *queue __rte_unused,
 297                 struct rte_mbuf **bufs __rte_unused,
 298                 uint16_t nb_pkts __rte_unused)
 299 {
 300         return 0;
 301 }
 302
 303 static inline void
 304 calculate_timestamp(struct timeval *ts) {
 305         uint64_t cycles;
 306         struct timeval cur_time;
 307
 308         cycles = rte_get_timer_cycles() - start_cycles;
 309         cur_time.tv_sec = cycles / hz;
 310         cur_time.tv_usec = (cycles % hz) * 1e6 / hz;
 311         timeradd(&start_time, &cur_time, ts);
 312 }
 313
 314 /*
 315  * Callback to handle writing packets to a pcap file.
 316  */
 317 static uint16_t
 318 eth_pcap_tx_dumper(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 319 {
 320         unsigned int i;
 321         struct rte_mbuf *mbuf;
 322         struct pmd_process_private *pp;
 323         struct pcap_tx_queue *dumper_q = queue;
 324         uint16_t num_tx = 0;
 325         uint32_t tx_bytes = 0;
 326         struct pcap_pkthdr header;
 327         pcap_dumper_t *dumper;
 328
 329         pp = rte_eth_devices[dumper_q->port_id].process_private;
 330         dumper = pp->tx_dumper[dumper_q->queue_id];
 331
 332         if (dumper == NULL || nb_pkts == 0)
 333                 return 0;
 334
 335         /* writes the nb_pkts packets to the previously opened pcap file
 336          * dumper */
 337         for (i = 0; i < nb_pkts; i++) {
 338                 mbuf = bufs[i];
 339                 calculate_timestamp(&header.ts);
 340                 header.len = mbuf->pkt_len;
 341                 header.caplen = header.len;
 342
 343                 if (likely(mbuf->nb_segs == 1)) {
 344                         pcap_dump((u_char *)dumper, &header,
 345                                   rte_pktmbuf_mtod(mbuf, void*));
 346                 } else {
 347                         if (mbuf->pkt_len <= RTE_ETHER_MAX_JUMBO_FRAME_LEN) {
 348                                 eth_pcap_gather_data(tx_pcap_data, mbuf);
 349                                 pcap_dump((u_char *)dumper, &header,
 350                                           tx_pcap_data);
 351                         } else {
 352                                 PMD_LOG(ERR,
 353                                         "Dropping PCAP packet. Size (%d) > max jumbo size (%d).",
 354                                         mbuf->pkt_len,
 355                                         RTE_ETHER_MAX_JUMBO_FRAME_LEN);
 356
 357                                 rte_pktmbuf_free(mbuf);
 358                                 continue;
 359                         }
 360                 }
 361
 362                 num_tx++;
 363                 tx_bytes += mbuf->pkt_len;
 364                 rte_pktmbuf_free(mbuf);
 365         }
 366
 367         /*
 368          * Since there's no place to hook a callback when the forwarding
 369          * process stops and to make sure the pcap file is actually written,
 370          * we flush the pcap dumper within each burst.
 371          */
 372         pcap_dump_flush(dumper);
 373         dumper_q->tx_stat.pkts += num_tx;
 374         dumper_q->tx_stat.bytes += tx_bytes;
 375         dumper_q->tx_stat.err_pkts += nb_pkts - num_tx;
 376
 377         return nb_pkts;
 378 }
 379
 380 /*
 381  * Callback to handle dropping packets in the infinite rx case.
 382  */
 383 static uint16_t
 384 eth_tx_drop(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 385 {
 386         unsigned int i;
 387         uint32_t tx_bytes = 0;
 388         struct pcap_tx_queue *tx_queue = queue;
 389
 390         if (unlikely(nb_pkts == 0))
 391                 return 0;
 392
 393         for (i = 0; i < nb_pkts; i++) {
 394                 tx_bytes += bufs[i]->data_len;
 395                 rte_pktmbuf_free(bufs[i]);
 396         }
 397
 398         tx_queue->tx_stat.pkts += nb_pkts;
 399         tx_queue->tx_stat.bytes += tx_bytes;
 400
 401         return i;
 402 }
 403
 404 /*
 405  * Callback to handle sending packets through a real NIC.
 406  */
 407 static uint16_t
 408 eth_pcap_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 409 {
 410         unsigned int i;
 411         int ret;
 412         struct rte_mbuf *mbuf;
 413         struct pmd_process_private *pp;
 414         struct pcap_tx_queue *tx_queue = queue;
 415         uint16_t num_tx = 0;
 416         uint32_t tx_bytes = 0;
 417         pcap_t *pcap;
 418
 419         pp = rte_eth_devices[tx_queue->port_id].process_private;
 420         pcap = pp->tx_pcap[tx_queue->queue_id];
 421
 422         if (unlikely(nb_pkts == 0 || pcap == NULL))
 423                 return 0;
 424
 425         for (i = 0; i < nb_pkts; i++) {
 426                 mbuf = bufs[i];
 427
 428                 if (likely(mbuf->nb_segs == 1)) {
 429                         ret = pcap_sendpacket(pcap,
 430                                         rte_pktmbuf_mtod(mbuf, u_char *),
 431                                         mbuf->pkt_len);
 432                 } else {
 433                         if (mbuf->pkt_len <= RTE_ETHER_MAX_JUMBO_FRAME_LEN) {
 434                                 eth_pcap_gather_data(tx_pcap_data, mbuf);
 435                                 ret = pcap_sendpacket(pcap,
 436                                                 tx_pcap_data, mbuf->pkt_len);
 437                         } else {
 438                                 PMD_LOG(ERR,
 439                                         "Dropping PCAP packet. Size (%d) > max jumbo size (%d).",
 440                                         mbuf->pkt_len,
 441                                         RTE_ETHER_MAX_JUMBO_FRAME_LEN);
 442
 443                                 rte_pktmbuf_free(mbuf);
 444                                 continue;
 445                         }
 446                 }
 447
 448                 if (unlikely(ret != 0))
 449                         break;
 450                 num_tx++;
 451                 tx_bytes += mbuf->pkt_len;
 452                 rte_pktmbuf_free(mbuf);
 453         }
 454
 455         tx_queue->tx_stat.pkts += num_tx;
 456         tx_queue->tx_stat.bytes += tx_bytes;
 457         tx_queue->tx_stat.err_pkts += i - num_tx;
 458
 459         return i;
 460 }
 461
 462 /*
 463  * pcap_open_live wrapper function
 464  */
 465 static inline int
 466 open_iface_live(const char *iface, pcap_t **pcap) {
 467         *pcap = pcap_open_live(iface, RTE_ETH_PCAP_SNAPLEN,
 468                         RTE_ETH_PCAP_PROMISC, RTE_ETH_PCAP_TIMEOUT, errbuf);
 469
 470         if (*pcap == NULL) {
 471                 PMD_LOG(ERR, "Couldn't open %s: %s", iface, errbuf);
 472                 return -1;
 473         }
 474
 475         return 0;
 476 }
 477
 478 static int
 479 open_single_iface(const char *iface, pcap_t **pcap)
 480 {
 481         if (open_iface_live(iface, pcap) < 0) {
 482                 PMD_LOG(ERR, "Couldn't open interface %s", iface);
 483                 return -1;
 484         }
 485
 486         return 0;
 487 }
 488
 489 static int
 490 open_single_tx_pcap(const char *pcap_filename, pcap_dumper_t **dumper)
 491 {
 492         pcap_t *tx_pcap;
 493
 494         /*
 495          * We need to create a dummy empty pcap_t to use it
 496          * with pcap_dump_open(). We create big enough an Ethernet
 497          * pcap holder.
 498          */
 499         tx_pcap = pcap_open_dead(DLT_EN10MB, RTE_ETH_PCAP_SNAPSHOT_LEN);
 500         if (tx_pcap == NULL) {
 501                 PMD_LOG(ERR, "Couldn't create dead pcap");
 502                 return -1;
 503         }
 504
 505         /* The dumper is created using the previous pcap_t reference */
 506         *dumper = pcap_dump_open(tx_pcap, pcap_filename);
 507         if (*dumper == NULL) {
 508                 pcap_close(tx_pcap);
 509                 PMD_LOG(ERR, "Couldn't open %s for writing.",
 510                         pcap_filename);
 511                 return -1;
 512         }
 513
 514         pcap_close(tx_pcap);
 515         return 0;
 516 }
 517
 518 static int
 519 open_single_rx_pcap(const char *pcap_filename, pcap_t **pcap)
 520 {
 521         *pcap = pcap_open_offline(pcap_filename, errbuf);
 522         if (*pcap == NULL) {
 523                 PMD_LOG(ERR, "Couldn't open %s: %s", pcap_filename,
 524                         errbuf);
 525                 return -1;
 526         }
 527
 528         return 0;
 529 }
 530
 531 static uint64_t
 532 count_packets_in_pcap(pcap_t **pcap, struct pcap_rx_queue *pcap_q)
 533 {
 534         const u_char *packet;
 535         struct pcap_pkthdr header;
 536         uint64_t pcap_pkt_count = 0;
 537
 538         while ((packet = pcap_next(*pcap, &header)))
 539                 pcap_pkt_count++;
 540
 541         /* The pcap is reopened so it can be used as normal later. */
 542         pcap_close(*pcap);
 543         *pcap = NULL;
 544         open_single_rx_pcap(pcap_q->name, pcap);
 545
 546         return pcap_pkt_count;
 547 }
 548
 549 static int
 550 eth_dev_start(struct rte_eth_dev *dev)
 551 {
 552         unsigned int i;
 553         struct pmd_internals *internals = dev->data->dev_private;
 554         struct pmd_process_private *pp = dev->process_private;
 555         struct pcap_tx_queue *tx;
 556         struct pcap_rx_queue *rx;
 557
 558         /* Special iface case. Single pcap is open and shared between tx/rx. */
 559         if (internals->single_iface) {
 560                 tx = &internals->tx_queue[0];
 561                 rx = &internals->rx_queue[0];
 562
 563                 if (!pp->tx_pcap[0] &&
 564                         strcmp(tx->type, ETH_PCAP_IFACE_ARG) == 0) {
 565                         if (open_single_iface(tx->name, &pp->tx_pcap[0]) < 0)
 566                                 return -1;
 567                         pp->rx_pcap[0] = pp->tx_pcap[0];
 568                 }
 569
 570                 goto status_up;
 571         }
 572
 573         /* If not open already, open tx pcaps/dumpers */
 574         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 575                 tx = &internals->tx_queue[i];
 576
 577                 if (!pp->tx_dumper[i] &&
 578                                 strcmp(tx->type, ETH_PCAP_TX_PCAP_ARG) == 0) {
 579                         if (open_single_tx_pcap(tx->name,
 580                                 &pp->tx_dumper[i]) < 0)
 581                                 return -1;
 582                 } else if (!pp->tx_pcap[i] &&
 583                                 strcmp(tx->type, ETH_PCAP_TX_IFACE_ARG) == 0) {
 584                         if (open_single_iface(tx->name, &pp->tx_pcap[i]) < 0)
 585                                 return -1;
 586                 }
 587         }
 588
 589         /* If not open already, open rx pcaps */
 590         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 591                 rx = &internals->rx_queue[i];
 592
 593                 if (pp->rx_pcap[i] != NULL)
 594                         continue;
 595
 596                 if (strcmp(rx->type, ETH_PCAP_RX_PCAP_ARG) == 0) {
 597                         if (open_single_rx_pcap(rx->name, &pp->rx_pcap[i]) < 0)
 598                                 return -1;
 599                 } else if (strcmp(rx->type, ETH_PCAP_RX_IFACE_ARG) == 0) {
 600                         if (open_single_iface(rx->name, &pp->rx_pcap[i]) < 0)
 601                                 return -1;
 602                 }
 603         }
 604
 605 status_up:
 606         for (i = 0; i < dev->data->nb_rx_queues; i++)
 607                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 608
 609         for (i = 0; i < dev->data->nb_tx_queues; i++)
 610                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 611
 612         dev->data->dev_link.link_status = ETH_LINK_UP;
 613
 614         return 0;
 615 }
 616
 617 /*
 618  * This function gets called when the current port gets stopped.
 619  * Is the only place for us to close all the tx streams dumpers.
 620  * If not called the dumpers will be flushed within each tx burst.
 621  */
 622 static void
 623 eth_dev_stop(struct rte_eth_dev *dev)
 624 {
 625         unsigned int i;
 626         struct pmd_internals *internals = dev->data->dev_private;
 627         struct pmd_process_private *pp = dev->process_private;
 628
 629         /* Special iface case. Single pcap is open and shared between tx/rx. */
 630         if (internals->single_iface) {
 631                 pcap_close(pp->tx_pcap[0]);
 632                 pp->tx_pcap[0] = NULL;
 633                 pp->rx_pcap[0] = NULL;
 634                 goto status_down;
 635         }
 636
 637         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 638                 if (pp->tx_dumper[i] != NULL) {
 639                         pcap_dump_close(pp->tx_dumper[i]);
 640                         pp->tx_dumper[i] = NULL;
 641                 }
 642
 643                 if (pp->tx_pcap[i] != NULL) {
 644                         pcap_close(pp->tx_pcap[i]);
 645                         pp->tx_pcap[i] = NULL;
 646                 }
 647         }
 648
 649         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 650                 if (pp->rx_pcap[i] != NULL) {
 651                         pcap_close(pp->rx_pcap[i]);
 652                         pp->rx_pcap[i] = NULL;
 653                 }
 654         }
 655
 656 status_down:
 657         for (i = 0; i < dev->data->nb_rx_queues; i++)
 658                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 659
 660         for (i = 0; i < dev->data->nb_tx_queues; i++)
 661                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 662
 663         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 664 }
 665
 666 static int
 667 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 668 {
 669         return 0;
 670 }
 671
 672 static void
 673 eth_dev_info(struct rte_eth_dev *dev,
 674                 struct rte_eth_dev_info *dev_info)
 675 {
 676         struct pmd_internals *internals = dev->data->dev_private;
 677
 678         dev_info->if_index = internals->if_index;
 679         dev_info->max_mac_addrs = 1;
 680         dev_info->max_rx_pktlen = (uint32_t) -1;
 681         dev_info->max_rx_queues = dev->data->nb_rx_queues;
 682         dev_info->max_tx_queues = dev->data->nb_tx_queues;
 683         dev_info->min_rx_bufsize = 0;
 684 }
 685
 686 static int
 687 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 688 {
 689         unsigned int i;
 690         unsigned long rx_packets_total = 0, rx_bytes_total = 0;
 691         unsigned long tx_packets_total = 0, tx_bytes_total = 0;
 692         unsigned long tx_packets_err_total = 0;
 693         const struct pmd_internals *internal = dev->data->dev_private;
 694
 695         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 696                         i < dev->data->nb_rx_queues; i++) {
 697                 stats->q_ipackets[i] = internal->rx_queue[i].rx_stat.pkts;
 698                 stats->q_ibytes[i] = internal->rx_queue[i].rx_stat.bytes;
 699                 rx_packets_total += stats->q_ipackets[i];
 700                 rx_bytes_total += stats->q_ibytes[i];
 701         }
 702
 703         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 704                         i < dev->data->nb_tx_queues; i++) {
 705                 stats->q_opackets[i] = internal->tx_queue[i].tx_stat.pkts;
 706                 stats->q_obytes[i] = internal->tx_queue[i].tx_stat.bytes;
 707                 tx_packets_total += stats->q_opackets[i];
 708                 tx_bytes_total += stats->q_obytes[i];
 709                 tx_packets_err_total += internal->tx_queue[i].tx_stat.err_pkts;
 710         }
 711
 712         stats->ipackets = rx_packets_total;
 713         stats->ibytes = rx_bytes_total;
 714         stats->opackets = tx_packets_total;
 715         stats->obytes = tx_bytes_total;
 716         stats->oerrors = tx_packets_err_total;
 717
 718         return 0;
 719 }
 720
 721 static void
 722 eth_stats_reset(struct rte_eth_dev *dev)
 723 {
 724         unsigned int i;
 725         struct pmd_internals *internal = dev->data->dev_private;
 726
 727         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 728                 internal->rx_queue[i].rx_stat.pkts = 0;
 729                 internal->rx_queue[i].rx_stat.bytes = 0;
 730         }
 731
 732         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 733                 internal->tx_queue[i].tx_stat.pkts = 0;
 734                 internal->tx_queue[i].tx_stat.bytes = 0;
 735                 internal->tx_queue[i].tx_stat.err_pkts = 0;
 736         }
 737 }
 738
 739 static void
 740 eth_dev_close(struct rte_eth_dev *dev)
 741 {
 742         unsigned int i;
 743         struct pmd_internals *internals = dev->data->dev_private;
 744
 745         /* Device wide flag, but cleanup must be performed per queue. */
 746         if (internals->infinite_rx) {
 747                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
 748                         struct pcap_rx_queue *pcap_q = &internals->rx_queue[i];
 749                         struct rte_mbuf *pcap_buf;
 750
 751                         while (!rte_ring_dequeue(pcap_q->pkts,
 752                                         (void **)&pcap_buf))
 753                                 rte_pktmbuf_free(pcap_buf);
 754
 755                         rte_ring_free(pcap_q->pkts);
 756                 }
 757         }
 758
 759 }
 760
 761 static void
 762 eth_queue_release(void *q __rte_unused)
 763 {
 764 }
 765
 766 static int
 767 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 768                 int wait_to_complete __rte_unused)
 769 {
 770         return 0;
 771 }
 772
 773 static int
 774 eth_rx_queue_setup(struct rte_eth_dev *dev,
 775                 uint16_t rx_queue_id,
 776                 uint16_t nb_rx_desc __rte_unused,
 777                 unsigned int socket_id __rte_unused,
 778                 const struct rte_eth_rxconf *rx_conf __rte_unused,
 779                 struct rte_mempool *mb_pool)
 780 {
 781         struct pmd_internals *internals = dev->data->dev_private;
 782         struct pcap_rx_queue *pcap_q = &internals->rx_queue[rx_queue_id];
 783
 784         pcap_q->mb_pool = mb_pool;
 785         pcap_q->port_id = dev->data->port_id;
 786         pcap_q->queue_id = rx_queue_id;
 787         dev->data->rx_queues[rx_queue_id] = pcap_q;
 788
 789         if (internals->infinite_rx) {
 790                 struct pmd_process_private *pp;
 791                 char ring_name[NAME_MAX];
 792                 static uint32_t ring_number;
 793                 uint64_t pcap_pkt_count = 0;
 794                 struct rte_mbuf *bufs[1];
 795                 pcap_t **pcap;
 796
 797                 pp = rte_eth_devices[pcap_q->port_id].process_private;
 798                 pcap = &pp->rx_pcap[pcap_q->queue_id];
 799
 800                 if (unlikely(*pcap == NULL))
 801                         return -ENOENT;
 802
 803                 pcap_pkt_count = count_packets_in_pcap(pcap, pcap_q);
 804
 805                 snprintf(ring_name, sizeof(ring_name), "PCAP_RING%" PRIu16,
 806                                 ring_number);
 807
 808                 pcap_q->pkts = rte_ring_create(ring_name,
 809                                 rte_align64pow2(pcap_pkt_count + 1), 0,
 810                                 RING_F_SP_ENQ | RING_F_SC_DEQ);
 811                 ring_number++;
 812                 if (!pcap_q->pkts)
 813                         return -ENOENT;
 814
 815                 /* Fill ring with packets from PCAP file one by one. */
 816                 while (eth_pcap_rx(pcap_q, bufs, 1)) {
 817                         /* Check for multiseg mbufs. */
 818                         if (bufs[0]->nb_segs != 1) {
 819                                 rte_pktmbuf_free(*bufs);
 820
 821                                 while (!rte_ring_dequeue(pcap_q->pkts,
 822                                                 (void **)bufs))
 823                                         rte_pktmbuf_free(*bufs);
 824
 825                                 rte_ring_free(pcap_q->pkts);
 826                                 PMD_LOG(ERR, "Multiseg mbufs are not supported in infinite_rx "
 827                                                 "mode.");
 828                                 return -EINVAL;
 829                         }
 830
 831                         rte_ring_enqueue_bulk(pcap_q->pkts,
 832                                         (void * const *)bufs, 1, NULL);
 833                 }
 834                 /*
 835                  * Reset the stats for this queue since eth_pcap_rx calls above
 836                  * didn't result in the application receiving packets.
 837                  */
 838                 pcap_q->rx_stat.pkts = 0;
 839                 pcap_q->rx_stat.bytes = 0;
 840         }
 841
 842         return 0;
 843 }
 844
 845 static int
 846 eth_tx_queue_setup(struct rte_eth_dev *dev,
 847                 uint16_t tx_queue_id,
 848                 uint16_t nb_tx_desc __rte_unused,
 849                 unsigned int socket_id __rte_unused,
 850                 const struct rte_eth_txconf *tx_conf __rte_unused)
 851 {
 852         struct pmd_internals *internals = dev->data->dev_private;
 853         struct pcap_tx_queue *pcap_q = &internals->tx_queue[tx_queue_id];
 854
 855         pcap_q->port_id = dev->data->port_id;
 856         pcap_q->queue_id = tx_queue_id;
 857         dev->data->tx_queues[tx_queue_id] = pcap_q;
 858
 859         return 0;
 860 }
 861
 862 static int
 863 eth_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 864 {
 865         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
 866
 867         return 0;
 868 }
 869
 870 static int
 871 eth_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
 872 {
 873         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
 874
 875         return 0;
 876 }
 877
 878 static int
 879 eth_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 880 {
 881         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
 882
 883         return 0;
 884 }
 885
 886 static int
 887 eth_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
 888 {
 889         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
 890
 891         return 0;
 892 }
 893
 894 static const struct eth_dev_ops ops = {
 895         .dev_start = eth_dev_start,
 896         .dev_stop = eth_dev_stop,
 897         .dev_close = eth_dev_close,
 898         .dev_configure = eth_dev_configure,
 899         .dev_infos_get = eth_dev_info,
 900         .rx_queue_setup = eth_rx_queue_setup,
 901         .tx_queue_setup = eth_tx_queue_setup,
 902         .rx_queue_start = eth_rx_queue_start,
 903         .tx_queue_start = eth_tx_queue_start,
 904         .rx_queue_stop = eth_rx_queue_stop,
 905         .tx_queue_stop = eth_tx_queue_stop,
 906         .rx_queue_release = eth_queue_release,
 907         .tx_queue_release = eth_queue_release,
 908         .link_update = eth_link_update,
 909         .stats_get = eth_stats_get,
 910         .stats_reset = eth_stats_reset,
 911 };
 912
 913 static int
 914 add_queue(struct pmd_devargs *pmd, const char *name, const char *type,
 915                 pcap_t *pcap, pcap_dumper_t *dumper)
 916 {
 917         if (pmd->num_of_queue >= RTE_PMD_PCAP_MAX_QUEUES)
 918                 return -1;
 919         if (pcap)
 920                 pmd->queue[pmd->num_of_queue].pcap = pcap;
 921         if (dumper)
 922                 pmd->queue[pmd->num_of_queue].dumper = dumper;
 923         pmd->queue[pmd->num_of_queue].name = name;
 924         pmd->queue[pmd->num_of_queue].type = type;
 925         pmd->num_of_queue++;
 926         return 0;
 927 }
 928
 929 /*
 930  * Function handler that opens the pcap file for reading a stores a
 931  * reference of it for use it later on.
 932  */
 933 static int
 934 open_rx_pcap(const char *key, const char *value, void *extra_args)
 935 {
 936         const char *pcap_filename = value;
 937         struct pmd_devargs *rx = extra_args;
 938         pcap_t *pcap = NULL;
 939
 940         if (open_single_rx_pcap(pcap_filename, &pcap) < 0)
 941                 return -1;
 942
 943         if (add_queue(rx, pcap_filename, key, pcap, NULL) < 0) {
 944                 pcap_close(pcap);
 945                 return -1;
 946         }
 947
 948         return 0;
 949 }
 950
 951 /*
 952  * Opens a pcap file for writing and stores a reference to it
 953  * for use it later on.
 954  */
 955 static int
 956 open_tx_pcap(const char *key, const char *value, void *extra_args)
 957 {
 958         const char *pcap_filename = value;
 959         struct pmd_devargs *dumpers = extra_args;
 960         pcap_dumper_t *dumper;
 961
 962         if (open_single_tx_pcap(pcap_filename, &dumper) < 0)
 963                 return -1;
 964
 965         if (add_queue(dumpers, pcap_filename, key, NULL, dumper) < 0) {
 966                 pcap_dump_close(dumper);
 967                 return -1;
 968         }
 969
 970         return 0;
 971 }
 972
 973 /*
 974  * Opens an interface for reading and writing
 975  */
 976 static inline int
 977 open_rx_tx_iface(const char *key, const char *value, void *extra_args)
 978 {
 979         const char *iface = value;
 980         struct pmd_devargs *tx = extra_args;
 981         pcap_t *pcap = NULL;
 982
 983         if (open_single_iface(iface, &pcap) < 0)
 984                 return -1;
 985
 986         tx->queue[0].pcap = pcap;
 987         tx->queue[0].name = iface;
 988         tx->queue[0].type = key;
 989
 990         return 0;
 991 }
 992
 993 static inline int
 994 set_iface_direction(const char *iface, pcap_t *pcap,
 995                 pcap_direction_t direction)
 996 {
 997         const char *direction_str = (direction == PCAP_D_IN) ? "IN" : "OUT";
 998         if (pcap_setdirection(pcap, direction) < 0) {
 999                 PMD_LOG(ERR, "Setting %s pcap direction %s failed - %s\n",
1000                                 iface, direction_str, pcap_geterr(pcap));
1001                 return -1;
1002         }
1003         PMD_LOG(INFO, "Setting %s pcap direction %s\n",
1004                         iface, direction_str);
1005         return 0;
1006 }
1007
1008 static inline int
1009 open_iface(const char *key, const char *value, void *extra_args)
1010 {
1011         const char *iface = value;
1012         struct pmd_devargs *pmd = extra_args;
1013         pcap_t *pcap = NULL;
1014
1015         if (open_single_iface(iface, &pcap) < 0)
1016                 return -1;
1017         if (add_queue(pmd, iface, key, pcap, NULL) < 0) {
1018                 pcap_close(pcap);
1019                 return -1;
1020         }
1021
1022         return 0;
1023 }
1024
1025 /*
1026  * Opens a NIC for reading packets from it
1027  */
1028 static inline int
1029 open_rx_iface(const char *key, const char *value, void *extra_args)
1030 {
1031         int ret = open_iface(key, value, extra_args);
1032         if (ret < 0)
1033                 return ret;
1034         if (strcmp(key, ETH_PCAP_RX_IFACE_IN_ARG) == 0) {
1035                 struct pmd_devargs *pmd = extra_args;
1036                 unsigned int qid = pmd->num_of_queue - 1;
1037
1038                 set_iface_direction(pmd->queue[qid].name,
1039                                 pmd->queue[qid].pcap,
1040                                 PCAP_D_IN);
1041         }
1042
1043         return 0;
1044 }
1045
1046 static inline int
1047 rx_iface_args_process(const char *key, const char *value, void *extra_args)
1048 {
1049         if (strcmp(key, ETH_PCAP_RX_IFACE_ARG) == 0 ||
1050                         strcmp(key, ETH_PCAP_RX_IFACE_IN_ARG) == 0)
1051                 return open_rx_iface(key, value, extra_args);
1052
1053         return 0;
1054 }
1055
1056 /*
1057  * Opens a NIC for writing packets to it
1058  */
1059 static int
1060 open_tx_iface(const char *key, const char *value, void *extra_args)
1061 {
1062         return open_iface(key, value, extra_args);
1063 }
1064
1065 static int
1066 select_phy_mac(const char *key __rte_unused, const char *value,
1067                 void *extra_args)
1068 {
1069         if (extra_args) {
1070                 const int phy_mac = atoi(value);
1071                 int *enable_phy_mac = extra_args;
1072
1073                 if (phy_mac)
1074                         *enable_phy_mac = 1;
1075         }
1076         return 0;
1077 }
1078
1079 static int
1080 get_infinite_rx_arg(const char *key __rte_unused,
1081                 const char *value, void *extra_args)
1082 {
1083         if (extra_args) {
1084                 const int infinite_rx = atoi(value);
1085                 int *enable_infinite_rx = extra_args;
1086
1087                 if (infinite_rx > 0)
1088                         *enable_infinite_rx = 1;
1089         }
1090         return 0;
1091 }
1092
1093 static int
1094 pmd_init_internals(struct rte_vdev_device *vdev,
1095                 const unsigned int nb_rx_queues,
1096                 const unsigned int nb_tx_queues,
1097                 struct pmd_internals **internals,
1098                 struct rte_eth_dev **eth_dev)
1099 {
1100         struct rte_eth_dev_data *data;
1101         struct pmd_process_private *pp;
1102         unsigned int numa_node = vdev->device.numa_node;
1103
1104         PMD_LOG(INFO, "Creating pcap-backed ethdev on numa socket %d",
1105                 numa_node);
1106
1107         pp = (struct pmd_process_private *)
1108                 rte_zmalloc(NULL, sizeof(struct pmd_process_private),
1109                                 RTE_CACHE_LINE_SIZE);
1110
1111         if (pp == NULL) {
1112                 PMD_LOG(ERR,
1113                         "Failed to allocate memory for process private");
1114                 return -1;
1115         }
1116
1117         /* reserve an ethdev entry */
1118         *eth_dev = rte_eth_vdev_allocate(vdev, sizeof(**internals));
1119         if (!(*eth_dev)) {
1120                 rte_free(pp);
1121                 return -1;
1122         }
1123         (*eth_dev)->process_private = pp;
1124         /* now put it all together
1125          * - store queue data in internals,
1126          * - store numa_node info in eth_dev
1127          * - point eth_dev_data to internals
1128          * - and point eth_dev structure to new eth_dev_data structure
1129          */
1130         *internals = (*eth_dev)->data->dev_private;
1131         /*
1132          * Interface MAC = 02:70:63:61:70:<iface_idx>
1133          * derived from: 'locally administered':'p':'c':'a':'p':'iface_idx'
1134          * where the middle 4 characters are converted to hex.
1135          */
1136         (*internals)->eth_addr = (struct rte_ether_addr) {
1137                 .addr_bytes = { 0x02, 0x70, 0x63, 0x61, 0x70, iface_idx++ }
1138         };
1139         (*internals)->phy_mac = 0;
1140         data = (*eth_dev)->data;
1141         data->nb_rx_queues = (uint16_t)nb_rx_queues;
1142         data->nb_tx_queues = (uint16_t)nb_tx_queues;
1143         data->dev_link = pmd_link;
1144         data->mac_addrs = &(*internals)->eth_addr;
1145
1146         /*
1147          * NOTE: we'll replace the data element, of originally allocated
1148          * eth_dev so the rings are local per-process
1149          */
1150         (*eth_dev)->dev_ops = &ops;
1151
1152         strlcpy((*internals)->devargs, rte_vdev_device_args(vdev),
1153                         ETH_PCAP_ARG_MAXLEN);
1154
1155         return 0;
1156 }
1157
1158 static int
1159 eth_pcap_update_mac(const char *if_name, struct rte_eth_dev *eth_dev,
1160                 const unsigned int numa_node)
1161 {
1162 #if defined(RTE_EXEC_ENV_LINUX)
1163         void *mac_addrs;
1164         struct ifreq ifr;
1165         int if_fd = socket(AF_INET, SOCK_DGRAM, 0);
1166
1167         if (if_fd == -1)
1168                 return -1;
1169
1170         rte_strscpy(ifr.ifr_name, if_name, sizeof(ifr.ifr_name));
1171         if (ioctl(if_fd, SIOCGIFHWADDR, &ifr)) {
1172                 close(if_fd);
1173                 return -1;
1174         }
1175
1176         mac_addrs = rte_zmalloc_socket(NULL, RTE_ETHER_ADDR_LEN, 0, numa_node);
1177         if (!mac_addrs) {
1178                 close(if_fd);
1179                 return -1;
1180         }
1181
1182         PMD_LOG(INFO, "Setting phy MAC for %s", if_name);
1183         eth_dev->data->mac_addrs = mac_addrs;
1184         rte_memcpy(eth_dev->data->mac_addrs[0].addr_bytes,
1185                         ifr.ifr_hwaddr.sa_data, RTE_ETHER_ADDR_LEN);
1186
1187         close(if_fd);
1188
1189         return 0;
1190
1191 #elif defined(RTE_EXEC_ENV_FREEBSD)
1192         void *mac_addrs;
1193         struct if_msghdr *ifm;
1194         struct sockaddr_dl *sdl;
1195         int mib[6];
1196         size_t len = 0;
1197         char *buf;
1198
1199         mib[0] = CTL_NET;
1200         mib[1] = AF_ROUTE;
1201         mib[2] = 0;
1202         mib[3] = AF_LINK;
1203         mib[4] = NET_RT_IFLIST;
1204         mib[5] = if_nametoindex(if_name);
1205
1206         if (sysctl(mib, 6, NULL, &len, NULL, 0) < 0)
1207                 return -1;
1208
1209         if (len == 0)
1210                 return -1;
1211
1212         buf = rte_malloc(NULL, len, 0);
1213         if (!buf)
1214                 return -1;
1215
1216         if (sysctl(mib, 6, buf, &len, NULL, 0) < 0) {
1217                 rte_free(buf);
1218                 return -1;
1219         }
1220         ifm = (struct if_msghdr *)buf;
1221         sdl = (struct sockaddr_dl *)(ifm + 1);
1222
1223         mac_addrs = rte_zmalloc_socket(NULL, RTE_ETHER_ADDR_LEN, 0, numa_node);
1224         if (!mac_addrs) {
1225                 rte_free(buf);
1226                 return -1;
1227         }
1228
1229         PMD_LOG(INFO, "Setting phy MAC for %s", if_name);
1230         eth_dev->data->mac_addrs = mac_addrs;
1231         rte_memcpy(eth_dev->data->mac_addrs[0].addr_bytes,
1232                         LLADDR(sdl), RTE_ETHER_ADDR_LEN);
1233
1234         rte_free(buf);
1235
1236         return 0;
1237 #else
1238         return -1;
1239 #endif
1240 }
1241
1242 static int
1243 eth_from_pcaps_common(struct rte_vdev_device *vdev,
1244                 struct pmd_devargs_all *devargs_all,
1245                 struct pmd_internals **internals, struct rte_eth_dev **eth_dev)
1246 {
1247         struct pmd_process_private *pp;
1248         struct pmd_devargs *rx_queues = &devargs_all->rx_queues;
1249         struct pmd_devargs *tx_queues = &devargs_all->tx_queues;
1250         const unsigned int nb_rx_queues = rx_queues->num_of_queue;
1251         const unsigned int nb_tx_queues = tx_queues->num_of_queue;
1252         unsigned int i;
1253
1254         /* do some parameter checking */
1255         if (rx_queues == NULL && nb_rx_queues > 0)
1256                 return -1;
1257         if (tx_queues == NULL && nb_tx_queues > 0)
1258                 return -1;
1259
1260         if (pmd_init_internals(vdev, nb_rx_queues, nb_tx_queues, internals,
1261                         eth_dev) < 0)
1262                 return -1;
1263
1264         pp = (*eth_dev)->process_private;
1265         for (i = 0; i < nb_rx_queues; i++) {
1266                 struct pcap_rx_queue *rx = &(*internals)->rx_queue[i];
1267                 struct devargs_queue *queue = &rx_queues->queue[i];
1268
1269                 pp->rx_pcap[i] = queue->pcap;
1270                 strlcpy(rx->name, queue->name, sizeof(rx->name));
1271                 strlcpy(rx->type, queue->type, sizeof(rx->type));
1272         }
1273
1274         for (i = 0; i < nb_tx_queues; i++) {
1275                 struct pcap_tx_queue *tx = &(*internals)->tx_queue[i];
1276                 struct devargs_queue *queue = &tx_queues->queue[i];
1277
1278                 pp->tx_dumper[i] = queue->dumper;
1279                 pp->tx_pcap[i] = queue->pcap;
1280                 strlcpy(tx->name, queue->name, sizeof(tx->name));
1281                 strlcpy(tx->type, queue->type, sizeof(tx->type));
1282         }
1283
1284         return 0;
1285 }
1286
1287 static int
1288 eth_from_pcaps(struct rte_vdev_device *vdev,
1289                 struct pmd_devargs_all *devargs_all)
1290 {
1291         struct pmd_internals *internals = NULL;
1292         struct rte_eth_dev *eth_dev = NULL;
1293         struct pmd_devargs *rx_queues = &devargs_all->rx_queues;
1294         int single_iface = devargs_all->single_iface;
1295         unsigned int infinite_rx = devargs_all->infinite_rx;
1296         int ret;
1297
1298         ret = eth_from_pcaps_common(vdev, devargs_all, &internals, &eth_dev);
1299
1300         if (ret < 0)
1301                 return ret;
1302
1303         /* store weather we are using a single interface for rx/tx or not */
1304         internals->single_iface = single_iface;
1305
1306         if (single_iface) {
1307                 internals->if_index = if_nametoindex(rx_queues->queue[0].name);
1308
1309                 /* phy_mac arg is applied only only if "iface" devarg is provided */
1310                 if (rx_queues->phy_mac) {
1311                         int ret = eth_pcap_update_mac(rx_queues->queue[0].name,
1312                                         eth_dev, vdev->device.numa_node);
1313                         if (ret == 0)
1314                                 internals->phy_mac = 1;
1315                 }
1316         }
1317
1318         internals->infinite_rx = infinite_rx;
1319         /* Assign rx ops. */
1320         if (infinite_rx)
1321                 eth_dev->rx_pkt_burst = eth_pcap_rx_infinite;
1322         else if (devargs_all->is_rx_pcap || devargs_all->is_rx_iface ||
1323                         single_iface)
1324                 eth_dev->rx_pkt_burst = eth_pcap_rx;
1325         else
1326                 eth_dev->rx_pkt_burst = eth_null_rx;
1327
1328         /* Assign tx ops. */
1329         if (devargs_all->is_tx_pcap)
1330                 eth_dev->tx_pkt_burst = eth_pcap_tx_dumper;
1331         else if (devargs_all->is_tx_iface || single_iface)
1332                 eth_dev->tx_pkt_burst = eth_pcap_tx;
1333         else
1334                 eth_dev->tx_pkt_burst = eth_tx_drop;
1335
1336         rte_eth_dev_probing_finish(eth_dev);
1337         return 0;
1338 }
1339
1340 static int
1341 pmd_pcap_probe(struct rte_vdev_device *dev)
1342 {
1343         const char *name;
1344         struct rte_kvargs *kvlist;
1345         struct pmd_devargs pcaps = {0};
1346         struct pmd_devargs dumpers = {0};
1347         struct rte_eth_dev *eth_dev =  NULL;
1348         struct pmd_internals *internal;
1349         int ret = 0;
1350
1351         struct pmd_devargs_all devargs_all = {
1352                 .single_iface = 0,
1353                 .is_tx_pcap = 0,
1354                 .is_tx_iface = 0,
1355                 .infinite_rx = 0,
1356         };
1357
1358         name = rte_vdev_device_name(dev);
1359         PMD_LOG(INFO, "Initializing pmd_pcap for %s", name);
1360
1361         gettimeofday(&start_time, NULL);
1362         start_cycles = rte_get_timer_cycles();
1363         hz = rte_get_timer_hz();
1364
1365         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1366                 eth_dev = rte_eth_dev_attach_secondary(name);
1367                 if (!eth_dev) {
1368                         PMD_LOG(ERR, "Failed to probe %s", name);
1369                         return -1;
1370                 }
1371
1372                 internal = eth_dev->data->dev_private;
1373
1374                 kvlist = rte_kvargs_parse(internal->devargs, valid_arguments);
1375                 if (kvlist == NULL)
1376                         return -1;
1377         } else {
1378                 kvlist = rte_kvargs_parse(rte_vdev_device_args(dev),
1379                                 valid_arguments);
1380                 if (kvlist == NULL)
1381                         return -1;
1382         }
1383
1384         /*
1385          * If iface argument is passed we open the NICs and use them for
1386          * reading / writing
1387          */
1388         if (rte_kvargs_count(kvlist, ETH_PCAP_IFACE_ARG) == 1) {
1389
1390                 ret = rte_kvargs_process(kvlist, ETH_PCAP_IFACE_ARG,
1391                                 &open_rx_tx_iface, &pcaps);
1392                 if (ret < 0)
1393                         goto free_kvlist;
1394
1395                 dumpers.queue[0] = pcaps.queue[0];
1396
1397                 ret = rte_kvargs_process(kvlist, ETH_PCAP_PHY_MAC_ARG,
1398                                 &select_phy_mac, &pcaps.phy_mac);
1399                 if (ret < 0)
1400                         goto free_kvlist;
1401
1402                 dumpers.phy_mac = pcaps.phy_mac;
1403
1404                 devargs_all.single_iface = 1;
1405                 pcaps.num_of_queue = 1;
1406                 dumpers.num_of_queue = 1;
1407
1408                 goto create_eth;
1409         }
1410
1411         /*
1412          * We check whether we want to open a RX stream from a real NIC, a
1413          * pcap file or open a dummy RX stream
1414          */
1415         devargs_all.is_rx_pcap =
1416                 rte_kvargs_count(kvlist, ETH_PCAP_RX_PCAP_ARG) ? 1 : 0;
1417         devargs_all.is_rx_iface =
1418                 rte_kvargs_count(kvlist, ETH_PCAP_RX_IFACE_ARG) ? 1 : 0;
1419         pcaps.num_of_queue = 0;
1420
1421         devargs_all.is_tx_pcap =
1422                 rte_kvargs_count(kvlist, ETH_PCAP_TX_PCAP_ARG) ? 1 : 0;
1423         devargs_all.is_tx_iface =
1424                 rte_kvargs_count(kvlist, ETH_PCAP_TX_IFACE_ARG) ? 1 : 0;
1425         dumpers.num_of_queue = 0;
1426
1427         if (devargs_all.is_rx_pcap) {
1428                 /*
1429                  * We check whether we want to infinitely rx the pcap file.
1430                  */
1431                 unsigned int infinite_rx_arg_cnt = rte_kvargs_count(kvlist,
1432                                 ETH_PCAP_INFINITE_RX_ARG);
1433
1434                 if (infinite_rx_arg_cnt == 1) {
1435                         ret = rte_kvargs_process(kvlist,
1436                                         ETH_PCAP_INFINITE_RX_ARG,
1437                                         &get_infinite_rx_arg,
1438                                         &devargs_all.infinite_rx);
1439                         if (ret < 0)
1440                                 goto free_kvlist;
1441                         PMD_LOG(INFO, "infinite_rx has been %s for %s",
1442                                         devargs_all.infinite_rx ? "enabled" : "disabled",
1443                                         name);
1444
1445                 } else if (infinite_rx_arg_cnt > 1) {
1446                         PMD_LOG(WARNING, "infinite_rx has not been enabled since the "
1447                                         "argument has been provided more than once "
1448                                         "for %s", name);
1449                 }
1450
1451                 ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_PCAP_ARG,
1452                                 &open_rx_pcap, &pcaps);
1453         } else if (devargs_all.is_rx_iface) {
1454                 ret = rte_kvargs_process(kvlist, NULL,
1455                                 &rx_iface_args_process, &pcaps);
1456         } else if (devargs_all.is_tx_iface || devargs_all.is_tx_pcap) {
1457                 unsigned int i;
1458
1459                 /* Count number of tx queue args passed before dummy rx queue
1460                  * creation so a dummy rx queue can be created for each tx queue
1461                  */
1462                 unsigned int num_tx_queues =
1463                         (rte_kvargs_count(kvlist, ETH_PCAP_TX_PCAP_ARG) +
1464                         rte_kvargs_count(kvlist, ETH_PCAP_TX_IFACE_ARG));
1465
1466                 PMD_LOG(INFO, "Creating null rx queue since no rx queues were provided.");
1467
1468                 /* Creating a dummy rx queue for each tx queue passed */
1469                 for (i = 0; i < num_tx_queues; i++)
1470                         ret = add_queue(&pcaps, "dummy_rx", "rx_null", NULL,
1471                                         NULL);
1472         } else {
1473                 PMD_LOG(ERR, "Error - No rx or tx queues provided");
1474                 ret = -ENOENT;
1475         }
1476         if (ret < 0)
1477                 goto free_kvlist;
1478
1479         /*
1480          * We check whether we want to open a TX stream to a real NIC,
1481          * a pcap file, or drop packets on tx
1482          */
1483         if (devargs_all.is_tx_pcap) {
1484                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_PCAP_ARG,
1485                                 &open_tx_pcap, &dumpers);
1486         } else if (devargs_all.is_tx_iface) {
1487                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_IFACE_ARG,
1488                                 &open_tx_iface, &dumpers);
1489         } else {
1490                 unsigned int i;
1491
1492                 PMD_LOG(INFO, "Dropping packets on tx since no tx queues were provided.");
1493
1494                 /* Add 1 dummy queue per rxq which counts and drops packets. */
1495                 for (i = 0; i < pcaps.num_of_queue; i++)
1496                         ret = add_queue(&dumpers, "dummy_tx", "tx_drop", NULL,
1497                                         NULL);
1498         }
1499
1500         if (ret < 0)
1501                 goto free_kvlist;
1502
1503 create_eth:
1504         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1505                 struct pmd_process_private *pp;
1506                 unsigned int i;
1507
1508                 internal = eth_dev->data->dev_private;
1509                         pp = (struct pmd_process_private *)
1510                                 rte_zmalloc(NULL,
1511                                         sizeof(struct pmd_process_private),
1512                                         RTE_CACHE_LINE_SIZE);
1513
1514                 if (pp == NULL) {
1515                         PMD_LOG(ERR,
1516                                 "Failed to allocate memory for process private");
1517                         ret = -1;
1518                         goto free_kvlist;
1519                 }
1520
1521                 eth_dev->dev_ops = &ops;
1522                 eth_dev->device = &dev->device;
1523
1524                 /* setup process private */
1525                 for (i = 0; i < pcaps.num_of_queue; i++)
1526                         pp->rx_pcap[i] = pcaps.queue[i].pcap;
1527
1528                 for (i = 0; i < dumpers.num_of_queue; i++) {
1529                         pp->tx_dumper[i] = dumpers.queue[i].dumper;
1530                         pp->tx_pcap[i] = dumpers.queue[i].pcap;
1531                 }
1532
1533                 eth_dev->process_private = pp;
1534                 eth_dev->rx_pkt_burst = eth_pcap_rx;
1535                 if (devargs_all.is_tx_pcap)
1536                         eth_dev->tx_pkt_burst = eth_pcap_tx_dumper;
1537                 else
1538                         eth_dev->tx_pkt_burst = eth_pcap_tx;
1539
1540                 rte_eth_dev_probing_finish(eth_dev);
1541                 goto free_kvlist;
1542         }
1543
1544         devargs_all.rx_queues = pcaps;
1545         devargs_all.tx_queues = dumpers;
1546
1547         ret = eth_from_pcaps(dev, &devargs_all);
1548
1549 free_kvlist:
1550         rte_kvargs_free(kvlist);
1551
1552         return ret;
1553 }
1554
1555 static int
1556 pmd_pcap_remove(struct rte_vdev_device *dev)
1557 {
1558         struct pmd_internals *internals = NULL;
1559         struct rte_eth_dev *eth_dev = NULL;
1560
1561         PMD_LOG(INFO, "Closing pcap ethdev on numa socket %d",
1562                         rte_socket_id());
1563
1564         if (!dev)
1565                 return -1;
1566
1567         /* reserve an ethdev entry */
1568         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1569         if (eth_dev == NULL)
1570                 return -1;
1571
1572         if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
1573                 internals = eth_dev->data->dev_private;
1574                 if (internals != NULL && internals->phy_mac == 0)
1575                         /* not dynamically allocated, must not be freed */
1576                         eth_dev->data->mac_addrs = NULL;
1577         }
1578
1579         eth_dev_close(eth_dev);
1580
1581         rte_free(eth_dev->process_private);
1582         rte_eth_dev_release_port(eth_dev);
1583
1584         return 0;
1585 }
1586
1587 static struct rte_vdev_driver pmd_pcap_drv = {
1588         .probe = pmd_pcap_probe,
1589         .remove = pmd_pcap_remove,
1590 };
1591
1592 RTE_PMD_REGISTER_VDEV(net_pcap, pmd_pcap_drv);
1593 RTE_PMD_REGISTER_ALIAS(net_pcap, eth_pcap);
1594 RTE_PMD_REGISTER_PARAM_STRING(net_pcap,
1595         ETH_PCAP_RX_PCAP_ARG "=<string> "
1596         ETH_PCAP_TX_PCAP_ARG "=<string> "
1597         ETH_PCAP_RX_IFACE_ARG "=<ifc> "
1598         ETH_PCAP_RX_IFACE_IN_ARG "=<ifc> "
1599         ETH_PCAP_TX_IFACE_ARG "=<ifc> "
1600         ETH_PCAP_IFACE_ARG "=<ifc> "
1601         ETH_PCAP_PHY_MAC_ARG "=<int>"
1602         ETH_PCAP_INFINITE_RX_ARG "=<0|1>");
1603
1604 RTE_INIT(eth_pcap_init_log)
1605 {
1606         eth_pcap_logtype = rte_log_register("pmd.net.pcap");
1607         if (eth_pcap_logtype >= 0)
1608                 rte_log_set_level(eth_pcap_logtype, RTE_LOG_NOTICE);
1609 }