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