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         uint16_t buf_size;
 246         uint32_t rx_bytes = 0;
 247         pcap_t *pcap;
 248
 249         pp = rte_eth_devices[pcap_q->port_id].process_private;
 250         pcap = pp->rx_pcap[pcap_q->queue_id];
 251
 252         if (unlikely(pcap == NULL || nb_pkts == 0))
 253                 return 0;
 254
 255         /* Reads the given number of packets from the pcap file one by one
 256          * and copies the packet data into a newly allocated mbuf to return.
 257          */
 258         for (i = 0; i < nb_pkts; i++) {
 259                 /* Get the next PCAP packet */
 260                 packet = pcap_next(pcap, &header);
 261                 if (unlikely(packet == NULL))
 262                         break;
 263
 264                 mbuf = rte_pktmbuf_alloc(pcap_q->mb_pool);
 265                 if (unlikely(mbuf == NULL))
 266                         break;
 267
 268                 /* Now get the space available for data in the mbuf */
 269                 buf_size = rte_pktmbuf_data_room_size(pcap_q->mb_pool) -
 270                                 RTE_PKTMBUF_HEADROOM;
 271
 272                 if (header.caplen <= buf_size) {
 273                         /* pcap packet will fit in the mbuf, can copy it */
 274                         rte_memcpy(rte_pktmbuf_mtod(mbuf, void *), packet,
 275                                         header.caplen);
 276                         mbuf->data_len = (uint16_t)header.caplen;
 277                 } else {
 278                         /* Try read jumbo frame into multi mbufs. */
 279                         if (unlikely(eth_pcap_rx_jumbo(pcap_q->mb_pool,
 280                                                        mbuf,
 281                                                        packet,
 282                                                        header.caplen) == -1)) {
 283                                 rte_pktmbuf_free(mbuf);
 284                                 break;
 285                         }
 286                 }
 287
 288                 mbuf->pkt_len = (uint16_t)header.caplen;
 289                 mbuf->port = pcap_q->port_id;
 290                 bufs[num_rx] = mbuf;
 291                 num_rx++;
 292                 rx_bytes += header.caplen;
 293         }
 294         pcap_q->rx_stat.pkts += num_rx;
 295         pcap_q->rx_stat.bytes += rx_bytes;
 296
 297         return num_rx;
 298 }
 299
 300 static uint16_t
 301 eth_null_rx(void *queue __rte_unused,
 302                 struct rte_mbuf **bufs __rte_unused,
 303                 uint16_t nb_pkts __rte_unused)
 304 {
 305         return 0;
 306 }
 307
 308 static inline void
 309 calculate_timestamp(struct timeval *ts) {
 310         uint64_t cycles;
 311         struct timeval cur_time;
 312
 313         cycles = rte_get_timer_cycles() - start_cycles;
 314         cur_time.tv_sec = cycles / hz;
 315         cur_time.tv_usec = (cycles % hz) * 1e6 / hz;
 316         timeradd(&start_time, &cur_time, ts);
 317 }
 318
 319 /*
 320  * Callback to handle writing packets to a pcap file.
 321  */
 322 static uint16_t
 323 eth_pcap_tx_dumper(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 324 {
 325         unsigned int i;
 326         struct rte_mbuf *mbuf;
 327         struct pmd_process_private *pp;
 328         struct pcap_tx_queue *dumper_q = queue;
 329         uint16_t num_tx = 0;
 330         uint32_t tx_bytes = 0;
 331         struct pcap_pkthdr header;
 332         pcap_dumper_t *dumper;
 333
 334         pp = rte_eth_devices[dumper_q->port_id].process_private;
 335         dumper = pp->tx_dumper[dumper_q->queue_id];
 336
 337         if (dumper == NULL || nb_pkts == 0)
 338                 return 0;
 339
 340         /* writes the nb_pkts packets to the previously opened pcap file
 341          * dumper */
 342         for (i = 0; i < nb_pkts; i++) {
 343                 mbuf = bufs[i];
 344                 calculate_timestamp(&header.ts);
 345                 header.len = mbuf->pkt_len;
 346                 header.caplen = header.len;
 347
 348                 if (likely(mbuf->nb_segs == 1)) {
 349                         pcap_dump((u_char *)dumper, &header,
 350                                   rte_pktmbuf_mtod(mbuf, void*));
 351                 } else {
 352                         if (mbuf->pkt_len <= RTE_ETHER_MAX_JUMBO_FRAME_LEN) {
 353                                 eth_pcap_gather_data(tx_pcap_data, mbuf);
 354                                 pcap_dump((u_char *)dumper, &header,
 355                                           tx_pcap_data);
 356                         } else {
 357                                 PMD_LOG(ERR,
 358                                         "Dropping PCAP packet. Size (%d) > max jumbo size (%d).",
 359                                         mbuf->pkt_len,
 360                                         RTE_ETHER_MAX_JUMBO_FRAME_LEN);
 361
 362                                 break;
 363                         }
 364                 }
 365
 366                 num_tx++;
 367                 tx_bytes += mbuf->pkt_len;
 368                 rte_pktmbuf_free(mbuf);
 369         }
 370
 371         /*
 372          * Since there's no place to hook a callback when the forwarding
 373          * process stops and to make sure the pcap file is actually written,
 374          * we flush the pcap dumper within each burst.
 375          */
 376         pcap_dump_flush(dumper);
 377         dumper_q->tx_stat.pkts += num_tx;
 378         dumper_q->tx_stat.bytes += tx_bytes;
 379         dumper_q->tx_stat.err_pkts += nb_pkts - num_tx;
 380
 381         return num_tx;
 382 }
 383
 384 /*
 385  * Callback to handle dropping packets in the infinite rx case.
 386  */
 387 static uint16_t
 388 eth_tx_drop(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 389 {
 390         unsigned int i;
 391         uint32_t tx_bytes = 0;
 392         struct pcap_tx_queue *tx_queue = queue;
 393
 394         if (unlikely(nb_pkts == 0))
 395                 return 0;
 396
 397         for (i = 0; i < nb_pkts; i++) {
 398                 tx_bytes += bufs[i]->data_len;
 399                 rte_pktmbuf_free(bufs[i]);
 400         }
 401
 402         tx_queue->tx_stat.pkts += nb_pkts;
 403         tx_queue->tx_stat.bytes += tx_bytes;
 404
 405         return i;
 406 }
 407
 408 /*
 409  * Callback to handle sending packets through a real NIC.
 410  */
 411 static uint16_t
 412 eth_pcap_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 413 {
 414         unsigned int i;
 415         int ret;
 416         struct rte_mbuf *mbuf;
 417         struct pmd_process_private *pp;
 418         struct pcap_tx_queue *tx_queue = queue;
 419         uint16_t num_tx = 0;
 420         uint32_t tx_bytes = 0;
 421         pcap_t *pcap;
 422
 423         pp = rte_eth_devices[tx_queue->port_id].process_private;
 424         pcap = pp->tx_pcap[tx_queue->queue_id];
 425
 426         if (unlikely(nb_pkts == 0 || pcap == NULL))
 427                 return 0;
 428
 429         for (i = 0; i < nb_pkts; i++) {
 430                 mbuf = bufs[i];
 431
 432                 if (likely(mbuf->nb_segs == 1)) {
 433                         ret = pcap_sendpacket(pcap,
 434                                         rte_pktmbuf_mtod(mbuf, u_char *),
 435                                         mbuf->pkt_len);
 436                 } else {
 437                         if (mbuf->pkt_len <= RTE_ETHER_MAX_JUMBO_FRAME_LEN) {
 438                                 eth_pcap_gather_data(tx_pcap_data, mbuf);
 439                                 ret = pcap_sendpacket(pcap,
 440                                                 tx_pcap_data, mbuf->pkt_len);
 441                         } else {
 442                                 PMD_LOG(ERR,
 443                                         "Dropping PCAP packet. Size (%d) > max jumbo size (%d).",
 444                                         mbuf->pkt_len,
 445                                         RTE_ETHER_MAX_JUMBO_FRAME_LEN);
 446
 447                                 break;
 448                         }
 449                 }
 450
 451                 if (unlikely(ret != 0))
 452                         break;
 453                 num_tx++;
 454                 tx_bytes += mbuf->pkt_len;
 455                 rte_pktmbuf_free(mbuf);
 456         }
 457
 458         tx_queue->tx_stat.pkts += num_tx;
 459         tx_queue->tx_stat.bytes += tx_bytes;
 460         tx_queue->tx_stat.err_pkts += nb_pkts - num_tx;
 461
 462         return num_tx;
 463 }
 464
 465 /*
 466  * pcap_open_live wrapper function
 467  */
 468 static inline int
 469 open_iface_live(const char *iface, pcap_t **pcap) {
 470         *pcap = pcap_open_live(iface, RTE_ETH_PCAP_SNAPLEN,
 471                         RTE_ETH_PCAP_PROMISC, RTE_ETH_PCAP_TIMEOUT, errbuf);
 472
 473         if (*pcap == NULL) {
 474                 PMD_LOG(ERR, "Couldn't open %s: %s", iface, errbuf);
 475                 return -1;
 476         }
 477
 478         return 0;
 479 }
 480
 481 static int
 482 open_single_iface(const char *iface, pcap_t **pcap)
 483 {
 484         if (open_iface_live(iface, pcap) < 0) {
 485                 PMD_LOG(ERR, "Couldn't open interface %s", iface);
 486                 return -1;
 487         }
 488
 489         return 0;
 490 }
 491
 492 static int
 493 open_single_tx_pcap(const char *pcap_filename, pcap_dumper_t **dumper)
 494 {
 495         pcap_t *tx_pcap;
 496
 497         /*
 498          * We need to create a dummy empty pcap_t to use it
 499          * with pcap_dump_open(). We create big enough an Ethernet
 500          * pcap holder.
 501          */
 502         tx_pcap = pcap_open_dead(DLT_EN10MB, RTE_ETH_PCAP_SNAPSHOT_LEN);
 503         if (tx_pcap == NULL) {
 504                 PMD_LOG(ERR, "Couldn't create dead pcap");
 505                 return -1;
 506         }
 507
 508         /* The dumper is created using the previous pcap_t reference */
 509         *dumper = pcap_dump_open(tx_pcap, pcap_filename);
 510         if (*dumper == NULL) {
 511                 pcap_close(tx_pcap);
 512                 PMD_LOG(ERR, "Couldn't open %s for writing.",
 513                         pcap_filename);
 514                 return -1;
 515         }
 516
 517         pcap_close(tx_pcap);
 518         return 0;
 519 }
 520
 521 static int
 522 open_single_rx_pcap(const char *pcap_filename, pcap_t **pcap)
 523 {
 524         *pcap = pcap_open_offline(pcap_filename, errbuf);
 525         if (*pcap == NULL) {
 526                 PMD_LOG(ERR, "Couldn't open %s: %s", pcap_filename,
 527                         errbuf);
 528                 return -1;
 529         }
 530
 531         return 0;
 532 }
 533
 534 static uint64_t
 535 count_packets_in_pcap(pcap_t **pcap, struct pcap_rx_queue *pcap_q)
 536 {
 537         const u_char *packet;
 538         struct pcap_pkthdr header;
 539         uint64_t pcap_pkt_count = 0;
 540
 541         while ((packet = pcap_next(*pcap, &header)))
 542                 pcap_pkt_count++;
 543
 544         /* The pcap is reopened so it can be used as normal later. */
 545         pcap_close(*pcap);
 546         *pcap = NULL;
 547         open_single_rx_pcap(pcap_q->name, pcap);
 548
 549         return pcap_pkt_count;
 550 }
 551
 552 static int
 553 eth_dev_start(struct rte_eth_dev *dev)
 554 {
 555         unsigned int i;
 556         struct pmd_internals *internals = dev->data->dev_private;
 557         struct pmd_process_private *pp = dev->process_private;
 558         struct pcap_tx_queue *tx;
 559         struct pcap_rx_queue *rx;
 560
 561         /* Special iface case. Single pcap is open and shared between tx/rx. */
 562         if (internals->single_iface) {
 563                 tx = &internals->tx_queue[0];
 564                 rx = &internals->rx_queue[0];
 565
 566                 if (!pp->tx_pcap[0] &&
 567                         strcmp(tx->type, ETH_PCAP_IFACE_ARG) == 0) {
 568                         if (open_single_iface(tx->name, &pp->tx_pcap[0]) < 0)
 569                                 return -1;
 570                         pp->rx_pcap[0] = pp->tx_pcap[0];
 571                 }
 572
 573                 goto status_up;
 574         }
 575
 576         /* If not open already, open tx pcaps/dumpers */
 577         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 578                 tx = &internals->tx_queue[i];
 579
 580                 if (!pp->tx_dumper[i] &&
 581                                 strcmp(tx->type, ETH_PCAP_TX_PCAP_ARG) == 0) {
 582                         if (open_single_tx_pcap(tx->name,
 583                                 &pp->tx_dumper[i]) < 0)
 584                                 return -1;
 585                 } else if (!pp->tx_pcap[i] &&
 586                                 strcmp(tx->type, ETH_PCAP_TX_IFACE_ARG) == 0) {
 587                         if (open_single_iface(tx->name, &pp->tx_pcap[i]) < 0)
 588                                 return -1;
 589                 }
 590         }
 591
 592         /* If not open already, open rx pcaps */
 593         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 594                 rx = &internals->rx_queue[i];
 595
 596                 if (pp->rx_pcap[i] != NULL)
 597                         continue;
 598
 599                 if (strcmp(rx->type, ETH_PCAP_RX_PCAP_ARG) == 0) {
 600                         if (open_single_rx_pcap(rx->name, &pp->rx_pcap[i]) < 0)
 601                                 return -1;
 602                 } else if (strcmp(rx->type, ETH_PCAP_RX_IFACE_ARG) == 0) {
 603                         if (open_single_iface(rx->name, &pp->rx_pcap[i]) < 0)
 604                                 return -1;
 605                 }
 606         }
 607
 608 status_up:
 609         for (i = 0; i < dev->data->nb_rx_queues; i++)
 610                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 611
 612         for (i = 0; i < dev->data->nb_tx_queues; i++)
 613                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 614
 615         dev->data->dev_link.link_status = ETH_LINK_UP;
 616
 617         return 0;
 618 }
 619
 620 /*
 621  * This function gets called when the current port gets stopped.
 622  * Is the only place for us to close all the tx streams dumpers.
 623  * If not called the dumpers will be flushed within each tx burst.
 624  */
 625 static void
 626 eth_dev_stop(struct rte_eth_dev *dev)
 627 {
 628         unsigned int i;
 629         struct pmd_internals *internals = dev->data->dev_private;
 630         struct pmd_process_private *pp = dev->process_private;
 631
 632         /* Special iface case. Single pcap is open and shared between tx/rx. */
 633         if (internals->single_iface) {
 634                 pcap_close(pp->tx_pcap[0]);
 635                 pp->tx_pcap[0] = NULL;
 636                 pp->rx_pcap[0] = NULL;
 637                 goto status_down;
 638         }
 639
 640         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 641                 if (pp->tx_dumper[i] != NULL) {
 642                         pcap_dump_close(pp->tx_dumper[i]);
 643                         pp->tx_dumper[i] = NULL;
 644                 }
 645
 646                 if (pp->tx_pcap[i] != NULL) {
 647                         pcap_close(pp->tx_pcap[i]);
 648                         pp->tx_pcap[i] = NULL;
 649                 }
 650         }
 651
 652         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 653                 if (pp->rx_pcap[i] != NULL) {
 654                         pcap_close(pp->rx_pcap[i]);
 655                         pp->rx_pcap[i] = NULL;
 656                 }
 657         }
 658
 659 status_down:
 660         for (i = 0; i < dev->data->nb_rx_queues; i++)
 661                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 662
 663         for (i = 0; i < dev->data->nb_tx_queues; i++)
 664                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 665
 666         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 667 }
 668
 669 static int
 670 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 671 {
 672         return 0;
 673 }
 674
 675 static void
 676 eth_dev_info(struct rte_eth_dev *dev,
 677                 struct rte_eth_dev_info *dev_info)
 678 {
 679         struct pmd_internals *internals = dev->data->dev_private;
 680
 681         dev_info->if_index = internals->if_index;
 682         dev_info->max_mac_addrs = 1;
 683         dev_info->max_rx_pktlen = (uint32_t) -1;
 684         dev_info->max_rx_queues = dev->data->nb_rx_queues;
 685         dev_info->max_tx_queues = dev->data->nb_tx_queues;
 686         dev_info->min_rx_bufsize = 0;
 687 }
 688
 689 static int
 690 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 691 {
 692         unsigned int i;
 693         unsigned long rx_packets_total = 0, rx_bytes_total = 0;
 694         unsigned long tx_packets_total = 0, tx_bytes_total = 0;
 695         unsigned long tx_packets_err_total = 0;
 696         const struct pmd_internals *internal = dev->data->dev_private;
 697
 698         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 699                         i < dev->data->nb_rx_queues; i++) {
 700                 stats->q_ipackets[i] = internal->rx_queue[i].rx_stat.pkts;
 701                 stats->q_ibytes[i] = internal->rx_queue[i].rx_stat.bytes;
 702                 rx_packets_total += stats->q_ipackets[i];
 703                 rx_bytes_total += stats->q_ibytes[i];
 704         }
 705
 706         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 707                         i < dev->data->nb_tx_queues; i++) {
 708                 stats->q_opackets[i] = internal->tx_queue[i].tx_stat.pkts;
 709                 stats->q_obytes[i] = internal->tx_queue[i].tx_stat.bytes;
 710                 tx_packets_total += stats->q_opackets[i];
 711                 tx_bytes_total += stats->q_obytes[i];
 712                 tx_packets_err_total += internal->tx_queue[i].tx_stat.err_pkts;
 713         }
 714
 715         stats->ipackets = rx_packets_total;
 716         stats->ibytes = rx_bytes_total;
 717         stats->opackets = tx_packets_total;
 718         stats->obytes = tx_bytes_total;
 719         stats->oerrors = tx_packets_err_total;
 720
 721         return 0;
 722 }
 723
 724 static void
 725 eth_stats_reset(struct rte_eth_dev *dev)
 726 {
 727         unsigned int i;
 728         struct pmd_internals *internal = dev->data->dev_private;
 729
 730         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 731                 internal->rx_queue[i].rx_stat.pkts = 0;
 732                 internal->rx_queue[i].rx_stat.bytes = 0;
 733         }
 734
 735         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 736                 internal->tx_queue[i].tx_stat.pkts = 0;
 737                 internal->tx_queue[i].tx_stat.bytes = 0;
 738                 internal->tx_queue[i].tx_stat.err_pkts = 0;
 739         }
 740 }
 741
 742 static void
 743 eth_dev_close(struct rte_eth_dev *dev)
 744 {
 745         unsigned int i;
 746         struct pmd_internals *internals = dev->data->dev_private;
 747
 748         /* Device wide flag, but cleanup must be performed per queue. */
 749         if (internals->infinite_rx) {
 750                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
 751                         struct pcap_rx_queue *pcap_q = &internals->rx_queue[i];
 752                         struct rte_mbuf *pcap_buf;
 753
 754                         while (!rte_ring_dequeue(pcap_q->pkts,
 755                                         (void **)&pcap_buf))
 756                                 rte_pktmbuf_free(pcap_buf);
 757
 758                         rte_ring_free(pcap_q->pkts);
 759                 }
 760         }
 761
 762 }
 763
 764 static void
 765 eth_queue_release(void *q __rte_unused)
 766 {
 767 }
 768
 769 static int
 770 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 771                 int wait_to_complete __rte_unused)
 772 {
 773         return 0;
 774 }
 775
 776 static int
 777 eth_rx_queue_setup(struct rte_eth_dev *dev,
 778                 uint16_t rx_queue_id,
 779                 uint16_t nb_rx_desc __rte_unused,
 780                 unsigned int socket_id __rte_unused,
 781                 const struct rte_eth_rxconf *rx_conf __rte_unused,
 782                 struct rte_mempool *mb_pool)
 783 {
 784         struct pmd_internals *internals = dev->data->dev_private;
 785         struct pcap_rx_queue *pcap_q = &internals->rx_queue[rx_queue_id];
 786
 787         pcap_q->mb_pool = mb_pool;
 788         pcap_q->port_id = dev->data->port_id;
 789         pcap_q->queue_id = rx_queue_id;
 790         dev->data->rx_queues[rx_queue_id] = pcap_q;
 791
 792         if (internals->infinite_rx) {
 793                 struct pmd_process_private *pp;
 794                 char ring_name[NAME_MAX];
 795                 static uint32_t ring_number;
 796                 uint64_t pcap_pkt_count = 0;
 797                 struct rte_mbuf *bufs[1];
 798                 pcap_t **pcap;
 799
 800                 pp = rte_eth_devices[pcap_q->port_id].process_private;
 801                 pcap = &pp->rx_pcap[pcap_q->queue_id];
 802
 803                 if (unlikely(*pcap == NULL))
 804                         return -ENOENT;
 805
 806                 pcap_pkt_count = count_packets_in_pcap(pcap, pcap_q);
 807
 808                 snprintf(ring_name, sizeof(ring_name), "PCAP_RING%" PRIu16,
 809                                 ring_number);
 810
 811                 pcap_q->pkts = rte_ring_create(ring_name,
 812                                 rte_align64pow2(pcap_pkt_count + 1), 0,
 813                                 RING_F_SP_ENQ | RING_F_SC_DEQ);
 814                 ring_number++;
 815                 if (!pcap_q->pkts)
 816                         return -ENOENT;
 817
 818                 /* Fill ring with packets from PCAP file one by one. */
 819                 while (eth_pcap_rx(pcap_q, bufs, 1)) {
 820                         /* Check for multiseg mbufs. */
 821                         if (bufs[0]->nb_segs != 1) {
 822                                 rte_pktmbuf_free(*bufs);
 823
 824                                 while (!rte_ring_dequeue(pcap_q->pkts,
 825                                                 (void **)bufs))
 826                                         rte_pktmbuf_free(*bufs);
 827
 828                                 rte_ring_free(pcap_q->pkts);
 829                                 PMD_LOG(ERR, "Multiseg mbufs are not supported in infinite_rx "
 830                                                 "mode.");
 831                                 return -EINVAL;
 832                         }
 833
 834                         rte_ring_enqueue_bulk(pcap_q->pkts,
 835                                         (void * const *)bufs, 1, NULL);
 836                 }
 837                 /*
 838                  * Reset the stats for this queue since eth_pcap_rx calls above
 839                  * didn't result in the application receiving packets.
 840                  */
 841                 pcap_q->rx_stat.pkts = 0;
 842                 pcap_q->rx_stat.bytes = 0;
 843         }
 844
 845         return 0;
 846 }
 847
 848 static int
 849 eth_tx_queue_setup(struct rte_eth_dev *dev,
 850                 uint16_t tx_queue_id,
 851                 uint16_t nb_tx_desc __rte_unused,
 852                 unsigned int socket_id __rte_unused,
 853                 const struct rte_eth_txconf *tx_conf __rte_unused)
 854 {
 855         struct pmd_internals *internals = dev->data->dev_private;
 856         struct pcap_tx_queue *pcap_q = &internals->tx_queue[tx_queue_id];
 857
 858         pcap_q->port_id = dev->data->port_id;
 859         pcap_q->queue_id = tx_queue_id;
 860         dev->data->tx_queues[tx_queue_id] = pcap_q;
 861
 862         return 0;
 863 }
 864
 865 static int
 866 eth_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 867 {
 868         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
 869
 870         return 0;
 871 }
 872
 873 static int
 874 eth_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
 875 {
 876         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
 877
 878         return 0;
 879 }
 880
 881 static int
 882 eth_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 883 {
 884         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
 885
 886         return 0;
 887 }
 888
 889 static int
 890 eth_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
 891 {
 892         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
 893
 894         return 0;
 895 }
 896
 897 static const struct eth_dev_ops ops = {
 898         .dev_start = eth_dev_start,
 899         .dev_stop = eth_dev_stop,
 900         .dev_close = eth_dev_close,
 901         .dev_configure = eth_dev_configure,
 902         .dev_infos_get = eth_dev_info,
 903         .rx_queue_setup = eth_rx_queue_setup,
 904         .tx_queue_setup = eth_tx_queue_setup,
 905         .rx_queue_start = eth_rx_queue_start,
 906         .tx_queue_start = eth_tx_queue_start,
 907         .rx_queue_stop = eth_rx_queue_stop,
 908         .tx_queue_stop = eth_tx_queue_stop,
 909         .rx_queue_release = eth_queue_release,
 910         .tx_queue_release = eth_queue_release,
 911         .link_update = eth_link_update,
 912         .stats_get = eth_stats_get,
 913         .stats_reset = eth_stats_reset,
 914 };
 915
 916 static int
 917 add_queue(struct pmd_devargs *pmd, const char *name, const char *type,
 918                 pcap_t *pcap, pcap_dumper_t *dumper)
 919 {
 920         if (pmd->num_of_queue >= RTE_PMD_PCAP_MAX_QUEUES)
 921                 return -1;
 922         if (pcap)
 923                 pmd->queue[pmd->num_of_queue].pcap = pcap;
 924         if (dumper)
 925                 pmd->queue[pmd->num_of_queue].dumper = dumper;
 926         pmd->queue[pmd->num_of_queue].name = name;
 927         pmd->queue[pmd->num_of_queue].type = type;
 928         pmd->num_of_queue++;
 929         return 0;
 930 }
 931
 932 /*
 933  * Function handler that opens the pcap file for reading a stores a
 934  * reference of it for use it later on.
 935  */
 936 static int
 937 open_rx_pcap(const char *key, const char *value, void *extra_args)
 938 {
 939         const char *pcap_filename = value;
 940         struct pmd_devargs *rx = extra_args;
 941         pcap_t *pcap = NULL;
 942
 943         if (open_single_rx_pcap(pcap_filename, &pcap) < 0)
 944                 return -1;
 945
 946         if (add_queue(rx, pcap_filename, key, pcap, NULL) < 0) {
 947                 pcap_close(pcap);
 948                 return -1;
 949         }
 950
 951         return 0;
 952 }
 953
 954 /*
 955  * Opens a pcap file for writing and stores a reference to it
 956  * for use it later on.
 957  */
 958 static int
 959 open_tx_pcap(const char *key, const char *value, void *extra_args)
 960 {
 961         const char *pcap_filename = value;
 962         struct pmd_devargs *dumpers = extra_args;
 963         pcap_dumper_t *dumper;
 964
 965         if (open_single_tx_pcap(pcap_filename, &dumper) < 0)
 966                 return -1;
 967
 968         if (add_queue(dumpers, pcap_filename, key, NULL, dumper) < 0) {
 969                 pcap_dump_close(dumper);
 970                 return -1;
 971         }
 972
 973         return 0;
 974 }
 975
 976 /*
 977  * Opens an interface for reading and writing
 978  */
 979 static inline int
 980 open_rx_tx_iface(const char *key, const char *value, void *extra_args)
 981 {
 982         const char *iface = value;
 983         struct pmd_devargs *tx = extra_args;
 984         pcap_t *pcap = NULL;
 985
 986         if (open_single_iface(iface, &pcap) < 0)
 987                 return -1;
 988
 989         tx->queue[0].pcap = pcap;
 990         tx->queue[0].name = iface;
 991         tx->queue[0].type = key;
 992
 993         return 0;
 994 }
 995
 996 static inline int
 997 set_iface_direction(const char *iface, pcap_t *pcap,
 998                 pcap_direction_t direction)
 999 {
1000         const char *direction_str = (direction == PCAP_D_IN) ? "IN" : "OUT";
1001         if (pcap_setdirection(pcap, direction) < 0) {
1002                 PMD_LOG(ERR, "Setting %s pcap direction %s failed - %s\n",
1003                                 iface, direction_str, pcap_geterr(pcap));
1004                 return -1;
1005         }
1006         PMD_LOG(INFO, "Setting %s pcap direction %s\n",
1007                         iface, direction_str);
1008         return 0;
1009 }
1010
1011 static inline int
1012 open_iface(const char *key, const char *value, void *extra_args)
1013 {
1014         const char *iface = value;
1015         struct pmd_devargs *pmd = extra_args;
1016         pcap_t *pcap = NULL;
1017
1018         if (open_single_iface(iface, &pcap) < 0)
1019                 return -1;
1020         if (add_queue(pmd, iface, key, pcap, NULL) < 0) {
1021                 pcap_close(pcap);
1022                 return -1;
1023         }
1024
1025         return 0;
1026 }
1027
1028 /*
1029  * Opens a NIC for reading packets from it
1030  */
1031 static inline int
1032 open_rx_iface(const char *key, const char *value, void *extra_args)
1033 {
1034         int ret = open_iface(key, value, extra_args);
1035         if (ret < 0)
1036                 return ret;
1037         if (strcmp(key, ETH_PCAP_RX_IFACE_IN_ARG) == 0) {
1038                 struct pmd_devargs *pmd = extra_args;
1039                 unsigned int qid = pmd->num_of_queue - 1;
1040
1041                 set_iface_direction(pmd->queue[qid].name,
1042                                 pmd->queue[qid].pcap,
1043                                 PCAP_D_IN);
1044         }
1045
1046         return 0;
1047 }
1048
1049 static inline int
1050 rx_iface_args_process(const char *key, const char *value, void *extra_args)
1051 {
1052         if (strcmp(key, ETH_PCAP_RX_IFACE_ARG) == 0 ||
1053                         strcmp(key, ETH_PCAP_RX_IFACE_IN_ARG) == 0)
1054                 return open_rx_iface(key, value, extra_args);
1055
1056         return 0;
1057 }
1058
1059 /*
1060  * Opens a NIC for writing packets to it
1061  */
1062 static int
1063 open_tx_iface(const char *key, const char *value, void *extra_args)
1064 {
1065         return open_iface(key, value, extra_args);
1066 }
1067
1068 static int
1069 select_phy_mac(const char *key __rte_unused, const char *value,
1070                 void *extra_args)
1071 {
1072         if (extra_args) {
1073                 const int phy_mac = atoi(value);
1074                 int *enable_phy_mac = extra_args;
1075
1076                 if (phy_mac)
1077                         *enable_phy_mac = 1;
1078         }
1079         return 0;
1080 }
1081
1082 static int
1083 get_infinite_rx_arg(const char *key __rte_unused,
1084                 const char *value, void *extra_args)
1085 {
1086         if (extra_args) {
1087                 const int infinite_rx = atoi(value);
1088                 int *enable_infinite_rx = extra_args;
1089
1090                 if (infinite_rx > 0)
1091                         *enable_infinite_rx = 1;
1092         }
1093         return 0;
1094 }
1095
1096 static int
1097 pmd_init_internals(struct rte_vdev_device *vdev,
1098                 const unsigned int nb_rx_queues,
1099                 const unsigned int nb_tx_queues,
1100                 struct pmd_internals **internals,
1101                 struct rte_eth_dev **eth_dev)
1102 {
1103         struct rte_eth_dev_data *data;
1104         struct pmd_process_private *pp;
1105         unsigned int numa_node = vdev->device.numa_node;
1106
1107         PMD_LOG(INFO, "Creating pcap-backed ethdev on numa socket %d",
1108                 numa_node);
1109
1110         pp = (struct pmd_process_private *)
1111                 rte_zmalloc(NULL, sizeof(struct pmd_process_private),
1112                                 RTE_CACHE_LINE_SIZE);
1113
1114         if (pp == NULL) {
1115                 PMD_LOG(ERR,
1116                         "Failed to allocate memory for process private");
1117                 return -1;
1118         }
1119
1120         /* reserve an ethdev entry */
1121         *eth_dev = rte_eth_vdev_allocate(vdev, sizeof(**internals));
1122         if (!(*eth_dev)) {
1123                 rte_free(pp);
1124                 return -1;
1125         }
1126         (*eth_dev)->process_private = pp;
1127         /* now put it all together
1128          * - store queue data in internals,
1129          * - store numa_node info in eth_dev
1130          * - point eth_dev_data to internals
1131          * - and point eth_dev structure to new eth_dev_data structure
1132          */
1133         *internals = (*eth_dev)->data->dev_private;
1134         /*
1135          * Interface MAC = 02:70:63:61:70:<iface_idx>
1136          * derived from: 'locally administered':'p':'c':'a':'p':'iface_idx'
1137          * where the middle 4 characters are converted to hex.
1138          */
1139         (*internals)->eth_addr = (struct rte_ether_addr) {
1140                 .addr_bytes = { 0x02, 0x70, 0x63, 0x61, 0x70, iface_idx++ }
1141         };
1142         (*internals)->phy_mac = 0;
1143         data = (*eth_dev)->data;
1144         data->nb_rx_queues = (uint16_t)nb_rx_queues;
1145         data->nb_tx_queues = (uint16_t)nb_tx_queues;
1146         data->dev_link = pmd_link;
1147         data->mac_addrs = &(*internals)->eth_addr;
1148
1149         /*
1150          * NOTE: we'll replace the data element, of originally allocated
1151          * eth_dev so the rings are local per-process
1152          */
1153         (*eth_dev)->dev_ops = &ops;
1154
1155         strlcpy((*internals)->devargs, rte_vdev_device_args(vdev),
1156                         ETH_PCAP_ARG_MAXLEN);
1157
1158         return 0;
1159 }
1160
1161 static int
1162 eth_pcap_update_mac(const char *if_name, struct rte_eth_dev *eth_dev,
1163                 const unsigned int numa_node)
1164 {
1165 #if defined(RTE_EXEC_ENV_LINUX)
1166         void *mac_addrs;
1167         struct ifreq ifr;
1168         int if_fd = socket(AF_INET, SOCK_DGRAM, 0);
1169
1170         if (if_fd == -1)
1171                 return -1;
1172
1173         rte_strscpy(ifr.ifr_name, if_name, sizeof(ifr.ifr_name));
1174         if (ioctl(if_fd, SIOCGIFHWADDR, &ifr)) {
1175                 close(if_fd);
1176                 return -1;
1177         }
1178
1179         mac_addrs = rte_zmalloc_socket(NULL, RTE_ETHER_ADDR_LEN, 0, numa_node);
1180         if (!mac_addrs) {
1181                 close(if_fd);
1182                 return -1;
1183         }
1184
1185         PMD_LOG(INFO, "Setting phy MAC for %s", if_name);
1186         eth_dev->data->mac_addrs = mac_addrs;
1187         rte_memcpy(eth_dev->data->mac_addrs[0].addr_bytes,
1188                         ifr.ifr_hwaddr.sa_data, RTE_ETHER_ADDR_LEN);
1189
1190         close(if_fd);
1191
1192         return 0;
1193
1194 #elif defined(RTE_EXEC_ENV_FREEBSD)
1195         void *mac_addrs;
1196         struct if_msghdr *ifm;
1197         struct sockaddr_dl *sdl;
1198         int mib[6];
1199         size_t len = 0;
1200         char *buf;
1201
1202         mib[0] = CTL_NET;
1203         mib[1] = AF_ROUTE;
1204         mib[2] = 0;
1205         mib[3] = AF_LINK;
1206         mib[4] = NET_RT_IFLIST;
1207         mib[5] = if_nametoindex(if_name);
1208
1209         if (sysctl(mib, 6, NULL, &len, NULL, 0) < 0)
1210                 return -1;
1211
1212         if (len == 0)
1213                 return -1;
1214
1215         buf = rte_malloc(NULL, len, 0);
1216         if (!buf)
1217                 return -1;
1218
1219         if (sysctl(mib, 6, buf, &len, NULL, 0) < 0) {
1220                 rte_free(buf);
1221                 return -1;
1222         }
1223         ifm = (struct if_msghdr *)buf;
1224         sdl = (struct sockaddr_dl *)(ifm + 1);
1225
1226         mac_addrs = rte_zmalloc_socket(NULL, RTE_ETHER_ADDR_LEN, 0, numa_node);
1227         if (!mac_addrs) {
1228                 rte_free(buf);
1229                 return -1;
1230         }
1231
1232         PMD_LOG(INFO, "Setting phy MAC for %s", if_name);
1233         eth_dev->data->mac_addrs = mac_addrs;
1234         rte_memcpy(eth_dev->data->mac_addrs[0].addr_bytes,
1235                         LLADDR(sdl), RTE_ETHER_ADDR_LEN);
1236
1237         rte_free(buf);
1238
1239         return 0;
1240 #else
1241         return -1;
1242 #endif
1243 }
1244
1245 static int
1246 eth_from_pcaps_common(struct rte_vdev_device *vdev,
1247                 struct pmd_devargs_all *devargs_all,
1248                 struct pmd_internals **internals, struct rte_eth_dev **eth_dev)
1249 {
1250         struct pmd_process_private *pp;
1251         struct pmd_devargs *rx_queues = &devargs_all->rx_queues;
1252         struct pmd_devargs *tx_queues = &devargs_all->tx_queues;
1253         const unsigned int nb_rx_queues = rx_queues->num_of_queue;
1254         const unsigned int nb_tx_queues = tx_queues->num_of_queue;
1255         unsigned int i;
1256
1257         /* do some parameter checking */
1258         if (rx_queues == NULL && nb_rx_queues > 0)
1259                 return -1;
1260         if (tx_queues == NULL && nb_tx_queues > 0)
1261                 return -1;
1262
1263         if (pmd_init_internals(vdev, nb_rx_queues, nb_tx_queues, internals,
1264                         eth_dev) < 0)
1265                 return -1;
1266
1267         pp = (*eth_dev)->process_private;
1268         for (i = 0; i < nb_rx_queues; i++) {
1269                 struct pcap_rx_queue *rx = &(*internals)->rx_queue[i];
1270                 struct devargs_queue *queue = &rx_queues->queue[i];
1271
1272                 pp->rx_pcap[i] = queue->pcap;
1273                 strlcpy(rx->name, queue->name, sizeof(rx->name));
1274                 strlcpy(rx->type, queue->type, sizeof(rx->type));
1275         }
1276
1277         for (i = 0; i < nb_tx_queues; i++) {
1278                 struct pcap_tx_queue *tx = &(*internals)->tx_queue[i];
1279                 struct devargs_queue *queue = &tx_queues->queue[i];
1280
1281                 pp->tx_dumper[i] = queue->dumper;
1282                 pp->tx_pcap[i] = queue->pcap;
1283                 strlcpy(tx->name, queue->name, sizeof(tx->name));
1284                 strlcpy(tx->type, queue->type, sizeof(tx->type));
1285         }
1286
1287         return 0;
1288 }
1289
1290 static int
1291 eth_from_pcaps(struct rte_vdev_device *vdev,
1292                 struct pmd_devargs_all *devargs_all)
1293 {
1294         struct pmd_internals *internals = NULL;
1295         struct rte_eth_dev *eth_dev = NULL;
1296         struct pmd_devargs *rx_queues = &devargs_all->rx_queues;
1297         int single_iface = devargs_all->single_iface;
1298         unsigned int infinite_rx = devargs_all->infinite_rx;
1299         int ret;
1300
1301         ret = eth_from_pcaps_common(vdev, devargs_all, &internals, &eth_dev);
1302
1303         if (ret < 0)
1304                 return ret;
1305
1306         /* store weather we are using a single interface for rx/tx or not */
1307         internals->single_iface = single_iface;
1308
1309         if (single_iface) {
1310                 internals->if_index = if_nametoindex(rx_queues->queue[0].name);
1311
1312                 /* phy_mac arg is applied only only if "iface" devarg is provided */
1313                 if (rx_queues->phy_mac) {
1314                         int ret = eth_pcap_update_mac(rx_queues->queue[0].name,
1315                                         eth_dev, vdev->device.numa_node);
1316                         if (ret == 0)
1317                                 internals->phy_mac = 1;
1318                 }
1319         }
1320
1321         internals->infinite_rx = infinite_rx;
1322         /* Assign rx ops. */
1323         if (infinite_rx)
1324                 eth_dev->rx_pkt_burst = eth_pcap_rx_infinite;
1325         else if (devargs_all->is_rx_pcap || devargs_all->is_rx_iface ||
1326                         single_iface)
1327                 eth_dev->rx_pkt_burst = eth_pcap_rx;
1328         else
1329                 eth_dev->rx_pkt_burst = eth_null_rx;
1330
1331         /* Assign tx ops. */
1332         if (devargs_all->is_tx_pcap)
1333                 eth_dev->tx_pkt_burst = eth_pcap_tx_dumper;
1334         else if (devargs_all->is_tx_iface || single_iface)
1335                 eth_dev->tx_pkt_burst = eth_pcap_tx;
1336         else
1337                 eth_dev->tx_pkt_burst = eth_tx_drop;
1338
1339         rte_eth_dev_probing_finish(eth_dev);
1340         return 0;
1341 }
1342
1343 static int
1344 pmd_pcap_probe(struct rte_vdev_device *dev)
1345 {
1346         const char *name;
1347         struct rte_kvargs *kvlist;
1348         struct pmd_devargs pcaps = {0};
1349         struct pmd_devargs dumpers = {0};
1350         struct rte_eth_dev *eth_dev =  NULL;
1351         struct pmd_internals *internal;
1352         int ret = 0;
1353
1354         struct pmd_devargs_all devargs_all = {
1355                 .single_iface = 0,
1356                 .is_tx_pcap = 0,
1357                 .is_tx_iface = 0,
1358                 .infinite_rx = 0,
1359         };
1360
1361         name = rte_vdev_device_name(dev);
1362         PMD_LOG(INFO, "Initializing pmd_pcap for %s", name);
1363
1364         gettimeofday(&start_time, NULL);
1365         start_cycles = rte_get_timer_cycles();
1366         hz = rte_get_timer_hz();
1367
1368         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1369                 eth_dev = rte_eth_dev_attach_secondary(name);
1370                 if (!eth_dev) {
1371                         PMD_LOG(ERR, "Failed to probe %s", name);
1372                         return -1;
1373                 }
1374
1375                 internal = eth_dev->data->dev_private;
1376
1377                 kvlist = rte_kvargs_parse(internal->devargs, valid_arguments);
1378                 if (kvlist == NULL)
1379                         return -1;
1380         } else {
1381                 kvlist = rte_kvargs_parse(rte_vdev_device_args(dev),
1382                                 valid_arguments);
1383                 if (kvlist == NULL)
1384                         return -1;
1385         }
1386
1387         /*
1388          * If iface argument is passed we open the NICs and use them for
1389          * reading / writing
1390          */
1391         if (rte_kvargs_count(kvlist, ETH_PCAP_IFACE_ARG) == 1) {
1392
1393                 ret = rte_kvargs_process(kvlist, ETH_PCAP_IFACE_ARG,
1394                                 &open_rx_tx_iface, &pcaps);
1395                 if (ret < 0)
1396                         goto free_kvlist;
1397
1398                 dumpers.queue[0] = pcaps.queue[0];
1399
1400                 ret = rte_kvargs_process(kvlist, ETH_PCAP_PHY_MAC_ARG,
1401                                 &select_phy_mac, &pcaps.phy_mac);
1402                 if (ret < 0)
1403                         goto free_kvlist;
1404
1405                 dumpers.phy_mac = pcaps.phy_mac;
1406
1407                 devargs_all.single_iface = 1;
1408                 pcaps.num_of_queue = 1;
1409                 dumpers.num_of_queue = 1;
1410
1411                 goto create_eth;
1412         }
1413
1414         /*
1415          * We check whether we want to open a RX stream from a real NIC, a
1416          * pcap file or open a dummy RX stream
1417          */
1418         devargs_all.is_rx_pcap =
1419                 rte_kvargs_count(kvlist, ETH_PCAP_RX_PCAP_ARG) ? 1 : 0;
1420         devargs_all.is_rx_iface =
1421                 rte_kvargs_count(kvlist, ETH_PCAP_RX_IFACE_ARG) ? 1 : 0;
1422         pcaps.num_of_queue = 0;
1423
1424         devargs_all.is_tx_pcap =
1425                 rte_kvargs_count(kvlist, ETH_PCAP_TX_PCAP_ARG) ? 1 : 0;
1426         devargs_all.is_tx_iface =
1427                 rte_kvargs_count(kvlist, ETH_PCAP_TX_IFACE_ARG) ? 1 : 0;
1428         dumpers.num_of_queue = 0;
1429
1430         if (devargs_all.is_rx_pcap) {
1431                 /*
1432                  * We check whether we want to infinitely rx the pcap file.
1433                  */
1434                 unsigned int infinite_rx_arg_cnt = rte_kvargs_count(kvlist,
1435                                 ETH_PCAP_INFINITE_RX_ARG);
1436
1437                 if (infinite_rx_arg_cnt == 1) {
1438                         ret = rte_kvargs_process(kvlist,
1439                                         ETH_PCAP_INFINITE_RX_ARG,
1440                                         &get_infinite_rx_arg,
1441                                         &devargs_all.infinite_rx);
1442                         if (ret < 0)
1443                                 goto free_kvlist;
1444                         PMD_LOG(INFO, "infinite_rx has been %s for %s",
1445                                         devargs_all.infinite_rx ? "enabled" : "disabled",
1446                                         name);
1447
1448                 } else if (infinite_rx_arg_cnt > 1) {
1449                         PMD_LOG(WARNING, "infinite_rx has not been enabled since the "
1450                                         "argument has been provided more than once "
1451                                         "for %s", name);
1452                 }
1453
1454                 ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_PCAP_ARG,
1455                                 &open_rx_pcap, &pcaps);
1456         } else if (devargs_all.is_rx_iface) {
1457                 ret = rte_kvargs_process(kvlist, NULL,
1458                                 &rx_iface_args_process, &pcaps);
1459         } else if (devargs_all.is_tx_iface || devargs_all.is_tx_pcap) {
1460                 unsigned int i;
1461
1462                 /* Count number of tx queue args passed before dummy rx queue
1463                  * creation so a dummy rx queue can be created for each tx queue
1464                  */
1465                 unsigned int num_tx_queues =
1466                         (rte_kvargs_count(kvlist, ETH_PCAP_TX_PCAP_ARG) +
1467                         rte_kvargs_count(kvlist, ETH_PCAP_TX_IFACE_ARG));
1468
1469                 PMD_LOG(INFO, "Creating null rx queue since no rx queues were provided.");
1470
1471                 /* Creating a dummy rx queue for each tx queue passed */
1472                 for (i = 0; i < num_tx_queues; i++)
1473                         ret = add_queue(&pcaps, "dummy_rx", "rx_null", NULL,
1474                                         NULL);
1475         } else {
1476                 PMD_LOG(ERR, "Error - No rx or tx queues provided");
1477                 ret = -ENOENT;
1478         }
1479         if (ret < 0)
1480                 goto free_kvlist;
1481
1482         /*
1483          * We check whether we want to open a TX stream to a real NIC,
1484          * a pcap file, or drop packets on tx
1485          */
1486         if (devargs_all.is_tx_pcap) {
1487                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_PCAP_ARG,
1488                                 &open_tx_pcap, &dumpers);
1489         } else if (devargs_all.is_tx_iface) {
1490                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_IFACE_ARG,
1491                                 &open_tx_iface, &dumpers);
1492         } else {
1493                 unsigned int i;
1494
1495                 PMD_LOG(INFO, "Dropping packets on tx since no tx queues were provided.");
1496
1497                 /* Add 1 dummy queue per rxq which counts and drops packets. */
1498                 for (i = 0; i < pcaps.num_of_queue; i++)
1499                         ret = add_queue(&dumpers, "dummy_tx", "tx_drop", NULL,
1500                                         NULL);
1501         }
1502
1503         if (ret < 0)
1504                 goto free_kvlist;
1505
1506 create_eth:
1507         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1508                 struct pmd_process_private *pp;
1509                 unsigned int i;
1510
1511                 internal = eth_dev->data->dev_private;
1512                         pp = (struct pmd_process_private *)
1513                                 rte_zmalloc(NULL,
1514                                         sizeof(struct pmd_process_private),
1515                                         RTE_CACHE_LINE_SIZE);
1516
1517                 if (pp == NULL) {
1518                         PMD_LOG(ERR,
1519                                 "Failed to allocate memory for process private");
1520                         ret = -1;
1521                         goto free_kvlist;
1522                 }
1523
1524                 eth_dev->dev_ops = &ops;
1525                 eth_dev->device = &dev->device;
1526
1527                 /* setup process private */
1528                 for (i = 0; i < pcaps.num_of_queue; i++)
1529                         pp->rx_pcap[i] = pcaps.queue[i].pcap;
1530
1531                 for (i = 0; i < dumpers.num_of_queue; i++) {
1532                         pp->tx_dumper[i] = dumpers.queue[i].dumper;
1533                         pp->tx_pcap[i] = dumpers.queue[i].pcap;
1534                 }
1535
1536                 eth_dev->process_private = pp;
1537                 eth_dev->rx_pkt_burst = eth_pcap_rx;
1538                 if (devargs_all.is_tx_pcap)
1539                         eth_dev->tx_pkt_burst = eth_pcap_tx_dumper;
1540                 else
1541                         eth_dev->tx_pkt_burst = eth_pcap_tx;
1542
1543                 rte_eth_dev_probing_finish(eth_dev);
1544                 goto free_kvlist;
1545         }
1546
1547         devargs_all.rx_queues = pcaps;
1548         devargs_all.tx_queues = dumpers;
1549
1550         ret = eth_from_pcaps(dev, &devargs_all);
1551
1552 free_kvlist:
1553         rte_kvargs_free(kvlist);
1554
1555         return ret;
1556 }
1557
1558 static int
1559 pmd_pcap_remove(struct rte_vdev_device *dev)
1560 {
1561         struct pmd_internals *internals = NULL;
1562         struct rte_eth_dev *eth_dev = NULL;
1563
1564         PMD_LOG(INFO, "Closing pcap ethdev on numa socket %d",
1565                         rte_socket_id());
1566
1567         if (!dev)
1568                 return -1;
1569
1570         /* reserve an ethdev entry */
1571         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1572         if (eth_dev == NULL)
1573                 return -1;
1574
1575         if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
1576                 internals = eth_dev->data->dev_private;
1577                 if (internals != NULL && internals->phy_mac == 0)
1578                         /* not dynamically allocated, must not be freed */
1579                         eth_dev->data->mac_addrs = NULL;
1580         }
1581
1582         eth_dev_close(eth_dev);
1583
1584         rte_free(eth_dev->process_private);
1585         rte_eth_dev_release_port(eth_dev);
1586
1587         return 0;
1588 }
1589
1590 static struct rte_vdev_driver pmd_pcap_drv = {
1591         .probe = pmd_pcap_probe,
1592         .remove = pmd_pcap_remove,
1593 };
1594
1595 RTE_PMD_REGISTER_VDEV(net_pcap, pmd_pcap_drv);
1596 RTE_PMD_REGISTER_ALIAS(net_pcap, eth_pcap);
1597 RTE_PMD_REGISTER_PARAM_STRING(net_pcap,
1598         ETH_PCAP_RX_PCAP_ARG "=<string> "
1599         ETH_PCAP_TX_PCAP_ARG "=<string> "
1600         ETH_PCAP_RX_IFACE_ARG "=<ifc> "
1601         ETH_PCAP_RX_IFACE_IN_ARG "=<ifc> "
1602         ETH_PCAP_TX_IFACE_ARG "=<ifc> "
1603         ETH_PCAP_IFACE_ARG "=<ifc> "
1604         ETH_PCAP_PHY_MAC_ARG "=<int>"
1605         ETH_PCAP_INFINITE_RX_ARG "=<0|1>");
1606
1607 RTE_INIT(eth_pcap_init_log)
1608 {
1609         eth_pcap_logtype = rte_log_register("pmd.net.pcap");
1610         if (eth_pcap_logtype >= 0)
1611                 rte_log_set_level(eth_pcap_logtype, RTE_LOG_NOTICE);
1612 }