lib/librte_pmd_pcap/rte_eth_pcap.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
   5  *   Copyright(c) 2014 6WIND S.A.
   6  *   All rights reserved.
   7  *
   8  *   Redistribution and use in source and binary forms, with or without
   9  *   modification, are permitted provided that the following conditions
  10  *   are met:
  11  *
  12  *     * Redistributions of source code must retain the above copyright
  13  *       notice, this list of conditions and the following disclaimer.
  14  *     * Redistributions in binary form must reproduce the above copyright
  15  *       notice, this list of conditions and the following disclaimer in
  16  *       the documentation and/or other materials provided with the
  17  *       distribution.
  18  *     * Neither the name of Intel Corporation nor the names of its
  19  *       contributors may be used to endorse or promote products derived
  20  *       from this software without specific prior written permission.
  21  *
  22  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  23  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  24  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  25  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  26  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  27  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  28  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  29  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  30  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  31  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  32  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  33  */
  34
  35 #include <time.h>
  36 #include <rte_mbuf.h>
  37 #include <rte_ethdev.h>
  38 #include <rte_malloc.h>
  39 #include <rte_memcpy.h>
  40 #include <rte_string_fns.h>
  41 #include <rte_cycles.h>
  42 #include <rte_kvargs.h>
  43 #include <rte_dev.h>
  44
  45 #include <net/if.h>
  46
  47 #include <pcap.h>
  48
  49 #define RTE_ETH_PCAP_SNAPSHOT_LEN 65535
  50 #define RTE_ETH_PCAP_SNAPLEN 4096
  51 #define RTE_ETH_PCAP_PROMISC 1
  52 #define RTE_ETH_PCAP_TIMEOUT -1
  53 #define ETH_PCAP_RX_PCAP_ARG  "rx_pcap"
  54 #define ETH_PCAP_TX_PCAP_ARG  "tx_pcap"
  55 #define ETH_PCAP_RX_IFACE_ARG "rx_iface"
  56 #define ETH_PCAP_TX_IFACE_ARG "tx_iface"
  57 #define ETH_PCAP_IFACE_ARG    "iface"
  58
  59 #define ETH_PCAP_ARG_MAXLEN     64
  60
  61 static char errbuf[PCAP_ERRBUF_SIZE];
  62 static struct timeval start_time;
  63 static uint64_t start_cycles;
  64 static uint64_t hz;
  65
  66 struct pcap_rx_queue {
  67         pcap_t *pcap;
  68         uint8_t in_port;
  69         struct rte_mempool *mb_pool;
  70         volatile unsigned long rx_pkts;
  71         volatile unsigned long err_pkts;
  72         char name[PATH_MAX];
  73         char type[ETH_PCAP_ARG_MAXLEN];
  74 };
  75
  76 struct pcap_tx_queue {
  77         pcap_dumper_t *dumper;
  78         pcap_t *pcap;
  79         volatile unsigned long tx_pkts;
  80         volatile unsigned long err_pkts;
  81         char name[PATH_MAX];
  82         char type[ETH_PCAP_ARG_MAXLEN];
  83 };
  84
  85 struct rx_pcaps {
  86         unsigned num_of_rx;
  87         pcap_t *pcaps[RTE_PMD_RING_MAX_RX_RINGS];
  88         const char *names[RTE_PMD_RING_MAX_RX_RINGS];
  89         const char *types[RTE_PMD_RING_MAX_RX_RINGS];
  90 };
  91
  92 struct tx_pcaps {
  93         unsigned num_of_tx;
  94         pcap_dumper_t *dumpers[RTE_PMD_RING_MAX_TX_RINGS];
  95         pcap_t *pcaps[RTE_PMD_RING_MAX_RX_RINGS];
  96         const char *names[RTE_PMD_RING_MAX_RX_RINGS];
  97         const char *types[RTE_PMD_RING_MAX_RX_RINGS];
  98 };
  99
 100 struct pmd_internals {
 101         struct pcap_rx_queue rx_queue[RTE_PMD_RING_MAX_RX_RINGS];
 102         struct pcap_tx_queue tx_queue[RTE_PMD_RING_MAX_TX_RINGS];
 103         unsigned nb_rx_queues;
 104         unsigned nb_tx_queues;
 105         int if_index;
 106         int single_iface;
 107 };
 108
 109 const char *valid_arguments[] = {
 110         ETH_PCAP_RX_PCAP_ARG,
 111         ETH_PCAP_TX_PCAP_ARG,
 112         ETH_PCAP_RX_IFACE_ARG,
 113         ETH_PCAP_TX_IFACE_ARG,
 114         ETH_PCAP_IFACE_ARG,
 115         NULL
 116 };
 117
 118 static int open_single_tx_pcap(const char *pcap_filename, pcap_dumper_t **dumper);
 119 static int open_single_rx_pcap(const char *pcap_filename, pcap_t **pcap);
 120 static int open_single_iface(const char *iface, pcap_t **pcap);
 121
 122 static struct ether_addr eth_addr = { .addr_bytes = { 0, 0, 0, 0x1, 0x2, 0x3 } };
 123 static const char *drivername = "Pcap PMD";
 124 static struct rte_eth_link pmd_link = {
 125                 .link_speed = 10000,
 126                 .link_duplex = ETH_LINK_FULL_DUPLEX,
 127                 .link_status = 0
 128 };
 129
 130
 131 static uint16_t
 132 eth_pcap_rx(void *queue,
 133                 struct rte_mbuf **bufs,
 134                 uint16_t nb_pkts)
 135 {
 136         unsigned i;
 137         struct pcap_pkthdr header;
 138         const u_char *packet;
 139         struct rte_mbuf *mbuf;
 140         struct pcap_rx_queue *pcap_q = queue;
 141         uint16_t num_rx = 0;
 142         uint16_t buf_size;
 143
 144         if (unlikely(pcap_q->pcap == NULL || nb_pkts == 0))
 145                 return 0;
 146
 147         /* Reads the given number of packets from the pcap file one by one
 148          * and copies the packet data into a newly allocated mbuf to return.
 149          */
 150         for (i = 0; i < nb_pkts; i++) {
 151                 /* Get the next PCAP packet */
 152                 packet = pcap_next(pcap_q->pcap, &header);
 153                 if (unlikely(packet == NULL))
 154                         break;
 155                 else
 156                         mbuf = rte_pktmbuf_alloc(pcap_q->mb_pool);
 157                 if (unlikely(mbuf == NULL))
 158                         break;
 159
 160                 /* Now get the space available for data in the mbuf */
 161                 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(pcap_q->mb_pool) -
 162                                 RTE_PKTMBUF_HEADROOM);
 163
 164                 if (header.len <= buf_size) {
 165                         /* pcap packet will fit in the mbuf, go ahead and copy */
 166                         rte_memcpy(rte_pktmbuf_mtod(mbuf, void *), packet,
 167                                         header.len);
 168                         mbuf->data_len = (uint16_t)header.len;
 169                         mbuf->pkt_len = mbuf->data_len;
 170                         mbuf->port = pcap_q->in_port;
 171                         bufs[num_rx] = mbuf;
 172                         num_rx++;
 173                 } else {
 174                         /* pcap packet will not fit in the mbuf, so drop packet */
 175                         RTE_LOG(ERR, PMD,
 176                                         "PCAP packet %d bytes will not fit in mbuf (%d bytes)\n",
 177                                         header.len, buf_size);
 178                         rte_pktmbuf_free(mbuf);
 179                 }
 180         }
 181         pcap_q->rx_pkts += num_rx;
 182         return num_rx;
 183 }
 184
 185 static inline void
 186 calculate_timestamp(struct timeval *ts) {
 187         uint64_t cycles;
 188         struct timeval cur_time;
 189
 190         cycles = rte_get_timer_cycles() - start_cycles;
 191         cur_time.tv_sec = cycles / hz;
 192         cur_time.tv_usec = (cycles % hz) * 10e6 / hz;
 193         timeradd(&start_time, &cur_time, ts);
 194 }
 195
 196 /*
 197  * Callback to handle writing packets to a pcap file.
 198  */
 199 static uint16_t
 200 eth_pcap_tx_dumper(void *queue,
 201                 struct rte_mbuf **bufs,
 202                 uint16_t nb_pkts)
 203 {
 204         unsigned i;
 205         struct rte_mbuf *mbuf;
 206         struct pcap_tx_queue *dumper_q = queue;
 207         uint16_t num_tx = 0;
 208         struct pcap_pkthdr header;
 209
 210         if (dumper_q->dumper == NULL || nb_pkts == 0)
 211                 return 0;
 212
 213         /* writes the nb_pkts packets to the previously opened pcap file dumper */
 214         for (i = 0; i < nb_pkts; i++) {
 215                 mbuf = bufs[i];
 216                 calculate_timestamp(&header.ts);
 217                 header.len = mbuf->data_len;
 218                 header.caplen = header.len;
 219                 pcap_dump((u_char *)dumper_q->dumper, &header,
 220                                 rte_pktmbuf_mtod(mbuf, void*));
 221                 rte_pktmbuf_free(mbuf);
 222                 num_tx++;
 223         }
 224
 225         /*
 226          * Since there's no place to hook a callback when the forwarding
 227          * process stops and to make sure the pcap file is actually written,
 228          * we flush the pcap dumper within each burst.
 229          */
 230         pcap_dump_flush(dumper_q->dumper);
 231         dumper_q->tx_pkts += num_tx;
 232         dumper_q->err_pkts += nb_pkts - num_tx;
 233         return num_tx;
 234 }
 235
 236 /*
 237  * Callback to handle sending packets through a real NIC.
 238  */
 239 static uint16_t
 240 eth_pcap_tx(void *queue,
 241                 struct rte_mbuf **bufs,
 242                 uint16_t nb_pkts)
 243 {
 244         unsigned i;
 245         int ret;
 246         struct rte_mbuf *mbuf;
 247         struct pcap_tx_queue *tx_queue = queue;
 248         uint16_t num_tx = 0;
 249
 250         if (unlikely(nb_pkts == 0 || tx_queue->pcap == NULL))
 251                 return 0;
 252
 253         for (i = 0; i < nb_pkts; i++) {
 254                 mbuf = bufs[i];
 255                 ret = pcap_sendpacket(tx_queue->pcap,
 256                                 rte_pktmbuf_mtod(mbuf, u_char *),
 257                                 mbuf->data_len);
 258                 if (unlikely(ret != 0))
 259                         break;
 260                 num_tx++;
 261                 rte_pktmbuf_free(mbuf);
 262         }
 263
 264         tx_queue->tx_pkts += num_tx;
 265         tx_queue->err_pkts += nb_pkts - num_tx;
 266         return num_tx;
 267 }
 268
 269 static int
 270 eth_dev_start(struct rte_eth_dev *dev)
 271 {
 272         unsigned i;
 273         struct pmd_internals *internals = dev->data->dev_private;
 274         struct pcap_tx_queue *tx;
 275         struct pcap_rx_queue *rx;
 276
 277         /* Special iface case. Single pcap is open and shared between tx/rx. */
 278         if (internals->single_iface) {
 279                 tx = &internals->tx_queue[0];
 280                 rx = &internals->rx_queue[0];
 281
 282                 if (!tx->pcap && strcmp(tx->type, ETH_PCAP_IFACE_ARG) == 0) {
 283                         if (open_single_iface(tx->name, &tx->pcap) < 0)
 284                                 return -1;
 285                         rx->pcap = tx->pcap;
 286                 }
 287                 goto status_up;
 288         }
 289
 290         /* If not open already, open tx pcaps/dumpers */
 291         for (i = 0; i < internals->nb_tx_queues; i++) {
 292                 tx = &internals->tx_queue[i];
 293
 294                 if (!tx->dumper && strcmp(tx->type, ETH_PCAP_TX_PCAP_ARG) == 0) {
 295                         if (open_single_tx_pcap(tx->name, &tx->dumper) < 0)
 296                                 return -1;
 297                 }
 298
 299                 else if (!tx->pcap && strcmp(tx->type, ETH_PCAP_TX_IFACE_ARG) == 0) {
 300                         if (open_single_iface(tx->name, &tx->pcap) < 0)
 301                                 return -1;
 302                 }
 303         }
 304
 305         /* If not open already, open rx pcaps */
 306         for (i = 0; i < internals->nb_rx_queues; i++) {
 307                 rx = &internals->rx_queue[i];
 308
 309                 if (rx->pcap != NULL)
 310                         continue;
 311
 312                 if (strcmp(rx->type, ETH_PCAP_RX_PCAP_ARG) == 0) {
 313                         if (open_single_rx_pcap(rx->name, &rx->pcap) < 0)
 314                                 return -1;
 315                 }
 316
 317                 else if (strcmp(rx->type, ETH_PCAP_RX_IFACE_ARG) == 0) {
 318                         if (open_single_iface(rx->name, &rx->pcap) < 0)
 319                                 return -1;
 320                 }
 321         }
 322
 323 status_up:
 324
 325         dev->data->dev_link.link_status = 1;
 326         return 0;
 327 }
 328
 329 /*
 330  * This function gets called when the current port gets stopped.
 331  * Is the only place for us to close all the tx streams dumpers.
 332  * If not called the dumpers will be flushed within each tx burst.
 333  */
 334 static void
 335 eth_dev_stop(struct rte_eth_dev *dev)
 336 {
 337         unsigned i;
 338         struct pmd_internals *internals = dev->data->dev_private;
 339         struct pcap_tx_queue *tx;
 340         struct pcap_rx_queue *rx;
 341
 342         /* Special iface case. Single pcap is open and shared between tx/rx. */
 343         if (internals->single_iface) {
 344                 tx = &internals->tx_queue[0];
 345                 rx = &internals->rx_queue[0];
 346                 pcap_close(tx->pcap);
 347                 tx->pcap = NULL;
 348                 rx->pcap = NULL;
 349                 goto status_down;
 350         }
 351
 352         for (i = 0; i < internals->nb_tx_queues; i++) {
 353                 tx = &internals->tx_queue[i];
 354
 355                 if (tx->dumper != NULL) {
 356                         pcap_dump_close(tx->dumper);
 357                         tx->dumper = NULL;
 358                 }
 359
 360                 if (tx->pcap != NULL) {
 361                         pcap_close(tx->pcap);
 362                         tx->pcap = NULL;
 363                 }
 364         }
 365
 366         for (i = 0; i < internals->nb_rx_queues; i++) {
 367                 rx = &internals->rx_queue[i];
 368
 369                 if (rx->pcap != NULL) {
 370                         pcap_close(rx->pcap);
 371                         rx->pcap = NULL;
 372                 }
 373         }
 374
 375 status_down:
 376         dev->data->dev_link.link_status = 0;
 377 }
 378
 379 static int
 380 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 381 {
 382         return 0;
 383 }
 384
 385 static void
 386 eth_dev_info(struct rte_eth_dev *dev,
 387                 struct rte_eth_dev_info *dev_info)
 388 {
 389         struct pmd_internals *internals = dev->data->dev_private;
 390         dev_info->driver_name = drivername;
 391         dev_info->if_index = internals->if_index;
 392         dev_info->max_mac_addrs = 1;
 393         dev_info->max_rx_pktlen = (uint32_t) -1;
 394         dev_info->max_rx_queues = (uint16_t)internals->nb_rx_queues;
 395         dev_info->max_tx_queues = (uint16_t)internals->nb_tx_queues;
 396         dev_info->min_rx_bufsize = 0;
 397         dev_info->pci_dev = NULL;
 398 }
 399
 400 static void
 401 eth_stats_get(struct rte_eth_dev *dev,
 402                 struct rte_eth_stats *igb_stats)
 403 {
 404         unsigned i;
 405         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 406         const struct pmd_internals *internal = dev->data->dev_private;
 407
 408         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS && i < internal->nb_rx_queues;
 409                         i++) {
 410                 igb_stats->q_ipackets[i] = internal->rx_queue[i].rx_pkts;
 411                 rx_total += igb_stats->q_ipackets[i];
 412         }
 413
 414         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS && i < internal->nb_tx_queues;
 415                         i++) {
 416                 igb_stats->q_opackets[i] = internal->tx_queue[i].tx_pkts;
 417                 igb_stats->q_errors[i] = internal->tx_queue[i].err_pkts;
 418                 tx_total += igb_stats->q_opackets[i];
 419                 tx_err_total += igb_stats->q_errors[i];
 420         }
 421
 422         igb_stats->ipackets = rx_total;
 423         igb_stats->opackets = tx_total;
 424         igb_stats->oerrors = tx_err_total;
 425 }
 426
 427 static void
 428 eth_stats_reset(struct rte_eth_dev *dev)
 429 {
 430         unsigned i;
 431         struct pmd_internals *internal = dev->data->dev_private;
 432         for (i = 0; i < internal->nb_rx_queues; i++)
 433                 internal->rx_queue[i].rx_pkts = 0;
 434         for (i = 0; i < internal->nb_tx_queues; i++) {
 435                 internal->tx_queue[i].tx_pkts = 0;
 436                 internal->tx_queue[i].err_pkts = 0;
 437         }
 438 }
 439
 440 static void
 441 eth_dev_close(struct rte_eth_dev *dev __rte_unused)
 442 {
 443 }
 444
 445 static void
 446 eth_queue_release(void *q __rte_unused)
 447 {
 448 }
 449
 450 static int
 451 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 452                 int wait_to_complete __rte_unused)
 453 {
 454         return 0;
 455 }
 456
 457 static int
 458 eth_rx_queue_setup(struct rte_eth_dev *dev,
 459                 uint16_t rx_queue_id,
 460                 uint16_t nb_rx_desc __rte_unused,
 461                 unsigned int socket_id __rte_unused,
 462                 const struct rte_eth_rxconf *rx_conf __rte_unused,
 463                 struct rte_mempool *mb_pool)
 464 {
 465         struct pmd_internals *internals = dev->data->dev_private;
 466         struct pcap_rx_queue *pcap_q = &internals->rx_queue[rx_queue_id];
 467         pcap_q->mb_pool = mb_pool;
 468         dev->data->rx_queues[rx_queue_id] = pcap_q;
 469         pcap_q->in_port = dev->data->port_id;
 470         return 0;
 471 }
 472
 473 static int
 474 eth_tx_queue_setup(struct rte_eth_dev *dev,
 475                 uint16_t tx_queue_id,
 476                 uint16_t nb_tx_desc __rte_unused,
 477                 unsigned int socket_id __rte_unused,
 478                 const struct rte_eth_txconf *tx_conf __rte_unused)
 479 {
 480
 481         struct pmd_internals *internals = dev->data->dev_private;
 482         dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
 483         return 0;
 484 }
 485
 486 static const struct eth_dev_ops ops = {
 487         .dev_start = eth_dev_start,
 488         .dev_stop =     eth_dev_stop,
 489         .dev_close = eth_dev_close,
 490         .dev_configure = eth_dev_configure,
 491         .dev_infos_get = eth_dev_info,
 492         .rx_queue_setup = eth_rx_queue_setup,
 493         .tx_queue_setup = eth_tx_queue_setup,
 494         .rx_queue_release = eth_queue_release,
 495         .tx_queue_release = eth_queue_release,
 496         .link_update = eth_link_update,
 497         .stats_get = eth_stats_get,
 498         .stats_reset = eth_stats_reset,
 499 };
 500
 501 static struct eth_driver rte_pcap_pmd = {
 502         .pci_drv = {
 503                 .name = "rte_pcap_pmd",
 504                 .drv_flags = RTE_PCI_DRV_DETACHABLE,
 505         },
 506 };
 507
 508 /*
 509  * Function handler that opens the pcap file for reading a stores a
 510  * reference of it for use it later on.
 511  */
 512 static int
 513 open_rx_pcap(const char *key, const char *value, void *extra_args)
 514 {
 515         unsigned i;
 516         const char *pcap_filename = value;
 517         struct rx_pcaps *pcaps = extra_args;
 518         pcap_t *pcap = NULL;
 519
 520         for (i = 0; i < pcaps->num_of_rx; i++) {
 521                 if (open_single_rx_pcap(pcap_filename, &pcap) < 0)
 522                         return -1;
 523
 524                 pcaps->pcaps[i] = pcap;
 525                 pcaps->names[i] = pcap_filename;
 526                 pcaps->types[i] = key;
 527         }
 528
 529         return 0;
 530 }
 531
 532 static int
 533 open_single_rx_pcap(const char *pcap_filename, pcap_t **pcap)
 534 {
 535         if ((*pcap = pcap_open_offline(pcap_filename, errbuf)) == NULL) {
 536                 RTE_LOG(ERR, PMD, "Couldn't open %s: %s\n", pcap_filename, errbuf);
 537                 return -1;
 538         }
 539         return 0;
 540 }
 541
 542 /*
 543  * Opens a pcap file for writing and stores a reference to it
 544  * for use it later on.
 545  */
 546 static int
 547 open_tx_pcap(const char *key, const char *value, void *extra_args)
 548 {
 549         unsigned i;
 550         const char *pcap_filename = value;
 551         struct tx_pcaps *dumpers = extra_args;
 552         pcap_dumper_t *dumper;
 553
 554         for (i = 0; i < dumpers->num_of_tx; i++) {
 555                 if (open_single_tx_pcap(pcap_filename, &dumper) < 0)
 556                         return -1;
 557
 558                 dumpers->dumpers[i] = dumper;
 559                 dumpers->names[i] = pcap_filename;
 560                 dumpers->types[i] = key;
 561         }
 562
 563         return 0;
 564 }
 565
 566 static int
 567 open_single_tx_pcap(const char *pcap_filename, pcap_dumper_t **dumper)
 568 {
 569         pcap_t *tx_pcap;
 570         /*
 571          * We need to create a dummy empty pcap_t to use it
 572          * with pcap_dump_open(). We create big enough an Ethernet
 573          * pcap holder.
 574          */
 575
 576         if ((tx_pcap = pcap_open_dead(DLT_EN10MB, RTE_ETH_PCAP_SNAPSHOT_LEN))
 577                         == NULL) {
 578                 RTE_LOG(ERR, PMD, "Couldn't create dead pcap\n");
 579                 return -1;
 580         }
 581
 582         /* The dumper is created using the previous pcap_t reference */
 583         if ((*dumper = pcap_dump_open(tx_pcap, pcap_filename)) == NULL) {
 584                 RTE_LOG(ERR, PMD, "Couldn't open %s for writing.\n", pcap_filename);
 585                 return -1;
 586         }
 587
 588         return 0;
 589 }
 590
 591 /*
 592  * pcap_open_live wrapper function
 593  */
 594 static inline int
 595 open_iface_live(const char *iface, pcap_t **pcap) {
 596         *pcap = pcap_open_live(iface, RTE_ETH_PCAP_SNAPLEN,
 597                         RTE_ETH_PCAP_PROMISC, RTE_ETH_PCAP_TIMEOUT, errbuf);
 598
 599         if (*pcap == NULL) {
 600                 RTE_LOG(ERR, PMD, "Couldn't open %s: %s\n", iface, errbuf);
 601                 return -1;
 602         }
 603         return 0;
 604 }
 605
 606 /*
 607  * Opens an interface for reading and writing
 608  */
 609 static inline int
 610 open_rx_tx_iface(const char *key, const char *value, void *extra_args)
 611 {
 612         const char *iface = value;
 613         struct rx_pcaps *pcaps = extra_args;
 614         pcap_t *pcap = NULL;
 615
 616         if (open_single_iface(iface, &pcap) < 0)
 617                 return -1;
 618
 619         pcaps->pcaps[0] = pcap;
 620         pcaps->names[0] = iface;
 621         pcaps->types[0] = key;
 622
 623         return 0;
 624 }
 625
 626 /*
 627  * Opens a NIC for reading packets from it
 628  */
 629 static inline int
 630 open_rx_iface(const char *key, const char *value, void *extra_args)
 631 {
 632         unsigned i;
 633         const char *iface = value;
 634         struct rx_pcaps *pcaps = extra_args;
 635         pcap_t *pcap = NULL;
 636
 637         for (i = 0; i < pcaps->num_of_rx; i++) {
 638                 if (open_single_iface(iface, &pcap) < 0)
 639                         return -1;
 640                 pcaps->pcaps[i] = pcap;
 641                 pcaps->names[i] = iface;
 642                 pcaps->types[i] = key;
 643         }
 644
 645         return 0;
 646 }
 647
 648 /*
 649  * Opens a NIC for writing packets to it
 650  */
 651 static int
 652 open_tx_iface(const char *key, const char *value, void *extra_args)
 653 {
 654         unsigned i;
 655         const char *iface = value;
 656         struct tx_pcaps *pcaps = extra_args;
 657         pcap_t *pcap;
 658
 659         for (i = 0; i < pcaps->num_of_tx; i++) {
 660                 if (open_single_iface(iface, &pcap) < 0)
 661                         return -1;
 662                 pcaps->pcaps[i] = pcap;
 663                 pcaps->names[i] = iface;
 664                 pcaps->types[i] = key;
 665         }
 666
 667         return 0;
 668 }
 669
 670 static int
 671 open_single_iface(const char *iface, pcap_t **pcap)
 672 {
 673         if (open_iface_live(iface, pcap) < 0) {
 674                 RTE_LOG(ERR, PMD, "Couldn't open interface %s\n", iface);
 675                 return -1;
 676         }
 677
 678         return 0;
 679 }
 680
 681 static int
 682 rte_pmd_init_internals(const char *name, const unsigned nb_rx_queues,
 683                 const unsigned nb_tx_queues,
 684                 const unsigned numa_node,
 685                 struct pmd_internals **internals,
 686                 struct rte_eth_dev **eth_dev,
 687                 struct rte_kvargs *kvlist)
 688 {
 689         struct rte_eth_dev_data *data = NULL;
 690         struct rte_pci_device *pci_dev = NULL;
 691         unsigned k_idx;
 692         struct rte_kvargs_pair *pair = NULL;
 693
 694         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 695                 pair = &kvlist->pairs[k_idx];
 696                 if (strstr(pair->key, ETH_PCAP_IFACE_ARG) != NULL)
 697                         break;
 698         }
 699
 700         RTE_LOG(INFO, PMD,
 701                         "Creating pcap-backed ethdev on numa socket %u\n", numa_node);
 702
 703         /* now do all data allocation - for eth_dev structure, dummy pci driver
 704          * and internal (private) data
 705          */
 706         data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
 707         if (data == NULL)
 708                 goto error;
 709
 710         pci_dev = rte_zmalloc_socket(name, sizeof(*pci_dev), 0, numa_node);
 711         if (pci_dev == NULL)
 712                 goto error;
 713
 714         *internals = rte_zmalloc_socket(name, sizeof(**internals), 0, numa_node);
 715         if (*internals == NULL)
 716                 goto error;
 717
 718         /* reserve an ethdev entry */
 719         *eth_dev = rte_eth_dev_allocate(name, RTE_ETH_DEV_VIRTUAL);
 720         if (*eth_dev == NULL)
 721                 goto error;
 722
 723         /* check length of device name */
 724         if ((strlen((*eth_dev)->data->name) + 1) > sizeof(data->name))
 725                 goto error;
 726
 727         /* now put it all together
 728          * - store queue data in internals,
 729          * - store numa_node info in pci_driver
 730          * - point eth_dev_data to internals and pci_driver
 731          * - and point eth_dev structure to new eth_dev_data structure
 732          */
 733         /* NOTE: we'll replace the data element, of originally allocated eth_dev
 734          * so the rings are local per-process */
 735
 736         (*internals)->nb_rx_queues = nb_rx_queues;
 737         (*internals)->nb_tx_queues = nb_tx_queues;
 738
 739         if (pair == NULL)
 740                 (*internals)->if_index = 0;
 741         else
 742                 (*internals)->if_index = if_nametoindex(pair->value);
 743
 744         pci_dev->numa_node = numa_node;
 745
 746         data->dev_private = *internals;
 747         data->port_id = (*eth_dev)->data->port_id;
 748         snprintf(data->name, sizeof(data->name), "%s", (*eth_dev)->data->name);
 749         data->nb_rx_queues = (uint16_t)nb_rx_queues;
 750         data->nb_tx_queues = (uint16_t)nb_tx_queues;
 751         data->dev_link = pmd_link;
 752         data->mac_addrs = &eth_addr;
 753         strncpy(data->name,
 754                 (*eth_dev)->data->name, strlen((*eth_dev)->data->name));
 755
 756         (*eth_dev)->data = data;
 757         (*eth_dev)->dev_ops = &ops;
 758         (*eth_dev)->pci_dev = pci_dev;
 759         (*eth_dev)->driver = &rte_pcap_pmd;
 760
 761         return 0;
 762
 763 error:
 764         rte_free(data);
 765         rte_free(pci_dev);
 766         rte_free(*internals);
 767
 768         return -1;
 769 }
 770
 771 static int
 772 rte_eth_from_pcaps_n_dumpers(const char *name,
 773                 struct rx_pcaps *rx_queues,
 774                 const unsigned nb_rx_queues,
 775                 struct tx_pcaps *tx_queues,
 776                 const unsigned nb_tx_queues,
 777                 const unsigned numa_node,
 778                 struct rte_kvargs *kvlist)
 779 {
 780         struct pmd_internals *internals = NULL;
 781         struct rte_eth_dev *eth_dev = NULL;
 782         unsigned i;
 783
 784         /* do some parameter checking */
 785         if (rx_queues == NULL && nb_rx_queues > 0)
 786                 return -1;
 787         if (tx_queues == NULL && nb_tx_queues > 0)
 788                 return -1;
 789
 790         if (rte_pmd_init_internals(name, nb_rx_queues, nb_tx_queues, numa_node,
 791                         &internals, &eth_dev, kvlist) < 0)
 792                 return -1;
 793
 794         for (i = 0; i < nb_rx_queues; i++) {
 795                 internals->rx_queue[i].pcap = rx_queues->pcaps[i];
 796                 snprintf(internals->rx_queue[i].name,
 797                         sizeof(internals->rx_queue[i].name), "%s",
 798                         rx_queues->names[i]);
 799                 snprintf(internals->rx_queue[i].type,
 800                         sizeof(internals->rx_queue[i].type), "%s",
 801                         rx_queues->types[i]);
 802         }
 803         for (i = 0; i < nb_tx_queues; i++) {
 804                 internals->tx_queue[i].dumper = tx_queues->dumpers[i];
 805                 snprintf(internals->tx_queue[i].name,
 806                         sizeof(internals->tx_queue[i].name), "%s",
 807                         tx_queues->names[i]);
 808                 snprintf(internals->tx_queue[i].type,
 809                         sizeof(internals->tx_queue[i].type), "%s",
 810                         tx_queues->types[i]);
 811         }
 812
 813         /* using multiple pcaps/interfaces */
 814         internals->single_iface = 0;
 815
 816         eth_dev->rx_pkt_burst = eth_pcap_rx;
 817         eth_dev->tx_pkt_burst = eth_pcap_tx_dumper;
 818
 819         return 0;
 820 }
 821
 822 static int
 823 rte_eth_from_pcaps(const char *name,
 824                 struct rx_pcaps *rx_queues,
 825                 const unsigned nb_rx_queues,
 826                 struct tx_pcaps *tx_queues,
 827                 const unsigned nb_tx_queues,
 828                 const unsigned numa_node,
 829                 struct rte_kvargs *kvlist,
 830                 int single_iface)
 831 {
 832         struct pmd_internals *internals = NULL;
 833         struct rte_eth_dev *eth_dev = NULL;
 834         unsigned i;
 835
 836         /* do some parameter checking */
 837         if (rx_queues == NULL && nb_rx_queues > 0)
 838                 return -1;
 839         if (tx_queues == NULL && nb_tx_queues > 0)
 840                 return -1;
 841
 842         if (rte_pmd_init_internals(name, nb_rx_queues, nb_tx_queues, numa_node,
 843                         &internals, &eth_dev, kvlist) < 0)
 844                 return -1;
 845
 846         for (i = 0; i < nb_rx_queues; i++) {
 847                 internals->rx_queue[i].pcap = rx_queues->pcaps[i];
 848                 snprintf(internals->rx_queue[i].name,
 849                         sizeof(internals->rx_queue[i].name), "%s",
 850                         rx_queues->names[i]);
 851                 snprintf(internals->rx_queue[i].type,
 852                         sizeof(internals->rx_queue[i].type), "%s",
 853                         rx_queues->types[i]);
 854         }
 855         for (i = 0; i < nb_tx_queues; i++) {
 856                 internals->tx_queue[i].dumper = tx_queues->dumpers[i];
 857                 snprintf(internals->tx_queue[i].name,
 858                         sizeof(internals->tx_queue[i].name), "%s",
 859                         tx_queues->names[i]);
 860                 snprintf(internals->tx_queue[i].type,
 861                         sizeof(internals->tx_queue[i].type), "%s",
 862                         tx_queues->types[i]);
 863         }
 864
 865         /* store wether we are using a single interface for rx/tx or not */
 866         internals->single_iface = single_iface;
 867
 868         eth_dev->rx_pkt_burst = eth_pcap_rx;
 869         eth_dev->tx_pkt_burst = eth_pcap_tx;
 870
 871         return 0;
 872 }
 873
 874
 875 static int
 876 rte_pmd_pcap_devinit(const char *name, const char *params)
 877 {
 878         unsigned numa_node, using_dumpers = 0;
 879         int ret;
 880         struct rte_kvargs *kvlist;
 881         struct rx_pcaps pcaps;
 882         struct tx_pcaps dumpers;
 883
 884         RTE_LOG(INFO, PMD, "Initializing pmd_pcap for %s\n", name);
 885
 886         numa_node = rte_socket_id();
 887
 888         gettimeofday(&start_time, NULL);
 889         start_cycles = rte_get_timer_cycles();
 890         hz = rte_get_timer_hz();
 891
 892         kvlist = rte_kvargs_parse(params, valid_arguments);
 893         if (kvlist == NULL)
 894                 return -1;
 895
 896         /*
 897          * If iface argument is passed we open the NICs and use them for
 898          * reading / writing
 899          */
 900         if (rte_kvargs_count(kvlist, ETH_PCAP_IFACE_ARG) == 1) {
 901
 902                 ret = rte_kvargs_process(kvlist, ETH_PCAP_IFACE_ARG,
 903                                 &open_rx_tx_iface, &pcaps);
 904                 if (ret < 0)
 905                         goto free_kvlist;
 906                 dumpers.pcaps[0] = pcaps.pcaps[0];
 907                 dumpers.names[0] = pcaps.names[0];
 908                 dumpers.types[0] = pcaps.types[0];
 909                 ret = rte_eth_from_pcaps(name, &pcaps, 1, &dumpers, 1,
 910                                 numa_node, kvlist, 1);
 911                 goto free_kvlist;
 912         }
 913
 914         /*
 915          * We check whether we want to open a RX stream from a real NIC or a
 916          * pcap file
 917          */
 918         if ((pcaps.num_of_rx = rte_kvargs_count(kvlist, ETH_PCAP_RX_PCAP_ARG))) {
 919                 ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_PCAP_ARG,
 920                                 &open_rx_pcap, &pcaps);
 921         } else {
 922                 pcaps.num_of_rx = rte_kvargs_count(kvlist,
 923                                 ETH_PCAP_RX_IFACE_ARG);
 924                 ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_IFACE_ARG,
 925                                 &open_rx_iface, &pcaps);
 926         }
 927
 928         if (ret < 0)
 929                 goto free_kvlist;
 930
 931         /*
 932          * We check whether we want to open a TX stream to a real NIC or a
 933          * pcap file
 934          */
 935         if ((dumpers.num_of_tx = rte_kvargs_count(kvlist,
 936                         ETH_PCAP_TX_PCAP_ARG))) {
 937                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_PCAP_ARG,
 938                                 &open_tx_pcap, &dumpers);
 939                 using_dumpers = 1;
 940         } else {
 941                 dumpers.num_of_tx = rte_kvargs_count(kvlist,
 942                                 ETH_PCAP_TX_IFACE_ARG);
 943                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_IFACE_ARG,
 944                                 &open_tx_iface, &dumpers);
 945         }
 946
 947         if (ret < 0)
 948                 goto free_kvlist;
 949
 950         if (using_dumpers)
 951                 ret = rte_eth_from_pcaps_n_dumpers(name, &pcaps, pcaps.num_of_rx,
 952                                 &dumpers, dumpers.num_of_tx, numa_node, kvlist);
 953         else
 954                 ret = rte_eth_from_pcaps(name, &pcaps, pcaps.num_of_rx, &dumpers,
 955                         dumpers.num_of_tx, numa_node, kvlist, 0);
 956
 957 free_kvlist:
 958         rte_kvargs_free(kvlist);
 959         return ret;
 960 }
 961
 962 static int
 963 rte_pmd_pcap_devuninit(const char *name)
 964 {
 965         struct rte_eth_dev *eth_dev = NULL;
 966
 967         RTE_LOG(INFO, PMD, "Closing pcap ethdev on numa socket %u\n",
 968                         rte_socket_id());
 969
 970         if (name == NULL)
 971                 return -1;
 972
 973         /* reserve an ethdev entry */
 974         eth_dev = rte_eth_dev_allocated(name);
 975         if (eth_dev == NULL)
 976                 return -1;
 977
 978         rte_free(eth_dev->data->dev_private);
 979         rte_free(eth_dev->data);
 980         rte_free(eth_dev->pci_dev);
 981
 982         rte_eth_dev_release_port(eth_dev);
 983
 984         return 0;
 985 }
 986
 987 static struct rte_driver pmd_pcap_drv = {
 988         .name = "eth_pcap",
 989         .type = PMD_VDEV,
 990         .init = rte_pmd_pcap_devinit,
 991         .uninit = rte_pmd_pcap_devuninit,
 992 };
 993
 994 PMD_REGISTER_DRIVER(pmd_pcap_drv);