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