drivers/net/pcap/rte_eth_pcap.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2015 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 ETHER_MAX_JUMBO_FRAME_LEN
  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 unsigned char tx_pcap_data[RTE_ETH_PCAP_SNAPLEN];
  63 static struct timeval start_time;
  64 static uint64_t start_cycles;
  65 static uint64_t hz;
  66
  67 struct pcap_rx_queue {
  68         pcap_t *pcap;
  69         uint8_t in_port;
  70         struct rte_mempool *mb_pool;
  71         volatile unsigned long rx_pkts;
  72         volatile unsigned long rx_bytes;
  73         volatile unsigned long err_pkts;
  74         char name[PATH_MAX];
  75         char type[ETH_PCAP_ARG_MAXLEN];
  76 };
  77
  78 struct pcap_tx_queue {
  79         pcap_dumper_t *dumper;
  80         pcap_t *pcap;
  81         volatile unsigned long tx_pkts;
  82         volatile unsigned long tx_bytes;
  83         volatile unsigned long err_pkts;
  84         char name[PATH_MAX];
  85         char type[ETH_PCAP_ARG_MAXLEN];
  86 };
  87
  88 struct rx_pcaps {
  89         unsigned num_of_rx;
  90         pcap_t *pcaps[RTE_PMD_RING_MAX_RX_RINGS];
  91         const char *names[RTE_PMD_RING_MAX_RX_RINGS];
  92         const char *types[RTE_PMD_RING_MAX_RX_RINGS];
  93 };
  94
  95 struct tx_pcaps {
  96         unsigned num_of_tx;
  97         pcap_dumper_t *dumpers[RTE_PMD_RING_MAX_TX_RINGS];
  98         pcap_t *pcaps[RTE_PMD_RING_MAX_RX_RINGS];
  99         const char *names[RTE_PMD_RING_MAX_RX_RINGS];
 100         const char *types[RTE_PMD_RING_MAX_RX_RINGS];
 101 };
 102
 103 struct pmd_internals {
 104         struct pcap_rx_queue rx_queue[RTE_PMD_RING_MAX_RX_RINGS];
 105         struct pcap_tx_queue tx_queue[RTE_PMD_RING_MAX_TX_RINGS];
 106         int if_index;
 107         int single_iface;
 108 };
 109
 110 const char *valid_arguments[] = {
 111         ETH_PCAP_RX_PCAP_ARG,
 112         ETH_PCAP_TX_PCAP_ARG,
 113         ETH_PCAP_RX_IFACE_ARG,
 114         ETH_PCAP_TX_IFACE_ARG,
 115         ETH_PCAP_IFACE_ARG,
 116         NULL
 117 };
 118
 119 static int open_single_tx_pcap(const char *pcap_filename, pcap_dumper_t **dumper);
 120 static int open_single_rx_pcap(const char *pcap_filename, pcap_t **pcap);
 121 static int open_single_iface(const char *iface, pcap_t **pcap);
 122
 123 static struct ether_addr eth_addr = { .addr_bytes = { 0, 0, 0, 0x1, 0x2, 0x3 } };
 124 static const char *drivername = "Pcap PMD";
 125 static struct rte_eth_link pmd_link = {
 126                 .link_speed = ETH_SPEED_NUM_10G,
 127                 .link_duplex = ETH_LINK_FULL_DUPLEX,
 128                 .link_status = ETH_LINK_DOWN,
 129                 .link_autoneg = ETH_LINK_SPEED_FIXED,
 130 };
 131
 132 static int
 133 eth_pcap_rx_jumbo(struct rte_mempool *mb_pool,
 134                   struct rte_mbuf *mbuf,
 135                   const u_char *data,
 136                   uint16_t data_len)
 137 {
 138         struct rte_mbuf *m = mbuf;
 139
 140         /* Copy the first segment. */
 141         uint16_t len = rte_pktmbuf_tailroom(mbuf);
 142
 143         rte_memcpy(rte_pktmbuf_append(mbuf, len), data, len);
 144         data_len -= len;
 145         data += len;
 146
 147         while (data_len > 0) {
 148                 /* Allocate next mbuf and point to that. */
 149                 m->next = rte_pktmbuf_alloc(mb_pool);
 150
 151                 if (unlikely(!m->next))
 152                         return -1;
 153
 154                 m = m->next;
 155
 156                 /* Headroom is not needed in chained mbufs. */
 157                 rte_pktmbuf_prepend(m, rte_pktmbuf_headroom(m));
 158                 m->pkt_len = 0;
 159                 m->data_len = 0;
 160
 161                 /* Copy next segment. */
 162                 len = RTE_MIN(rte_pktmbuf_tailroom(m), data_len);
 163                 rte_memcpy(rte_pktmbuf_append(m, len), data, len);
 164
 165                 mbuf->nb_segs++;
 166                 data_len -= len;
 167                 data += len;
 168         }
 169
 170         return mbuf->nb_segs;
 171 }
 172
 173 /* Copy data from mbuf chain to a buffer suitable for writing to a PCAP file. */
 174 static void
 175 eth_pcap_gather_data(unsigned char *data, struct rte_mbuf *mbuf)
 176 {
 177         uint16_t data_len = 0;
 178
 179         while (mbuf) {
 180                 rte_memcpy(data + data_len, rte_pktmbuf_mtod(mbuf, void *),
 181                            mbuf->data_len);
 182
 183                 data_len += mbuf->data_len;
 184                 mbuf = mbuf->next;
 185         }
 186 }
 187
 188 static uint16_t
 189 eth_pcap_rx(void *queue,
 190                 struct rte_mbuf **bufs,
 191                 uint16_t nb_pkts)
 192 {
 193         unsigned i;
 194         struct pcap_pkthdr header;
 195         const u_char *packet;
 196         struct rte_mbuf *mbuf;
 197         struct pcap_rx_queue *pcap_q = queue;
 198         uint16_t num_rx = 0;
 199         uint16_t buf_size;
 200         uint32_t rx_bytes = 0;
 201
 202         if (unlikely(pcap_q->pcap == NULL || nb_pkts == 0))
 203                 return 0;
 204
 205         /* Reads the given number of packets from the pcap file one by one
 206          * and copies the packet data into a newly allocated mbuf to return.
 207          */
 208         for (i = 0; i < nb_pkts; i++) {
 209                 /* Get the next PCAP packet */
 210                 packet = pcap_next(pcap_q->pcap, &header);
 211                 if (unlikely(packet == NULL))
 212                         break;
 213                 else
 214                         mbuf = rte_pktmbuf_alloc(pcap_q->mb_pool);
 215                 if (unlikely(mbuf == NULL))
 216                         break;
 217
 218                 /* Now get the space available for data in the mbuf */
 219                 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(pcap_q->mb_pool) -
 220                                 RTE_PKTMBUF_HEADROOM);
 221
 222                 if (header.caplen <= buf_size) {
 223                         /* pcap packet will fit in the mbuf, go ahead and copy */
 224                         rte_memcpy(rte_pktmbuf_mtod(mbuf, void *), packet,
 225                                         header.caplen);
 226                         mbuf->data_len = (uint16_t)header.caplen;
 227                 } else {
 228                         /* Try read jumbo frame into multi mbufs. */
 229                         if (unlikely(eth_pcap_rx_jumbo(pcap_q->mb_pool,
 230                                                        mbuf,
 231                                                        packet,
 232                                                        header.caplen) == -1))
 233                                 break;
 234                 }
 235
 236                 mbuf->pkt_len = (uint16_t)header.caplen;
 237                 mbuf->port = pcap_q->in_port;
 238                 bufs[num_rx] = mbuf;
 239                 num_rx++;
 240                 rx_bytes += header.caplen;
 241         }
 242         pcap_q->rx_pkts += num_rx;
 243         pcap_q->rx_bytes += rx_bytes;
 244         return num_rx;
 245 }
 246
 247 static inline void
 248 calculate_timestamp(struct timeval *ts) {
 249         uint64_t cycles;
 250         struct timeval cur_time;
 251
 252         cycles = rte_get_timer_cycles() - start_cycles;
 253         cur_time.tv_sec = cycles / hz;
 254         cur_time.tv_usec = (cycles % hz) * 10e6 / hz;
 255         timeradd(&start_time, &cur_time, ts);
 256 }
 257
 258 /*
 259  * Callback to handle writing packets to a pcap file.
 260  */
 261 static uint16_t
 262 eth_pcap_tx_dumper(void *queue,
 263                 struct rte_mbuf **bufs,
 264                 uint16_t nb_pkts)
 265 {
 266         unsigned i;
 267         struct rte_mbuf *mbuf;
 268         struct pcap_tx_queue *dumper_q = queue;
 269         uint16_t num_tx = 0;
 270         uint32_t tx_bytes = 0;
 271         struct pcap_pkthdr header;
 272
 273         if (dumper_q->dumper == NULL || nb_pkts == 0)
 274                 return 0;
 275
 276         /* writes the nb_pkts packets to the previously opened pcap file dumper */
 277         for (i = 0; i < nb_pkts; i++) {
 278                 mbuf = bufs[i];
 279                 calculate_timestamp(&header.ts);
 280                 header.len = mbuf->pkt_len;
 281                 header.caplen = header.len;
 282
 283                 if (likely(mbuf->nb_segs == 1)) {
 284                         pcap_dump((u_char *)dumper_q->dumper, &header,
 285                                   rte_pktmbuf_mtod(mbuf, void*));
 286                 } else {
 287                         if (mbuf->pkt_len <= ETHER_MAX_JUMBO_FRAME_LEN) {
 288                                 eth_pcap_gather_data(tx_pcap_data, mbuf);
 289                                 pcap_dump((u_char *)dumper_q->dumper, &header,
 290                                           tx_pcap_data);
 291                         } else {
 292                                 RTE_LOG(ERR, PMD,
 293                                         "Dropping PCAP packet. "
 294                                         "Size (%d) > max jumbo size (%d).\n",
 295                                         mbuf->pkt_len,
 296                                         ETHER_MAX_JUMBO_FRAME_LEN);
 297
 298                                 rte_pktmbuf_free(mbuf);
 299                                 break;
 300                         }
 301                 }
 302
 303                 rte_pktmbuf_free(mbuf);
 304                 num_tx++;
 305                 tx_bytes += mbuf->pkt_len;
 306         }
 307
 308         /*
 309          * Since there's no place to hook a callback when the forwarding
 310          * process stops and to make sure the pcap file is actually written,
 311          * we flush the pcap dumper within each burst.
 312          */
 313         pcap_dump_flush(dumper_q->dumper);
 314         dumper_q->tx_pkts += num_tx;
 315         dumper_q->tx_bytes += tx_bytes;
 316         dumper_q->err_pkts += nb_pkts - num_tx;
 317         return num_tx;
 318 }
 319
 320 /*
 321  * Callback to handle sending packets through a real NIC.
 322  */
 323 static uint16_t
 324 eth_pcap_tx(void *queue,
 325                 struct rte_mbuf **bufs,
 326                 uint16_t nb_pkts)
 327 {
 328         unsigned i;
 329         int ret;
 330         struct rte_mbuf *mbuf;
 331         struct pcap_tx_queue *tx_queue = queue;
 332         uint16_t num_tx = 0;
 333         uint32_t tx_bytes = 0;
 334
 335         if (unlikely(nb_pkts == 0 || tx_queue->pcap == NULL))
 336                 return 0;
 337
 338         for (i = 0; i < nb_pkts; i++) {
 339                 mbuf = bufs[i];
 340
 341                 if (likely(mbuf->nb_segs == 1)) {
 342                         ret = pcap_sendpacket(tx_queue->pcap,
 343                                               rte_pktmbuf_mtod(mbuf, u_char *),
 344                                               mbuf->pkt_len);
 345                 } else {
 346                         if (mbuf->pkt_len <= ETHER_MAX_JUMBO_FRAME_LEN) {
 347                                 eth_pcap_gather_data(tx_pcap_data, mbuf);
 348                                 ret = pcap_sendpacket(tx_queue->pcap,
 349                                                       tx_pcap_data,
 350                                                       mbuf->pkt_len);
 351                         } else {
 352                                 RTE_LOG(ERR, PMD,
 353                                         "Dropping PCAP packet. "
 354                                         "Size (%d) > max jumbo size (%d).\n",
 355                                         mbuf->pkt_len,
 356                                         ETHER_MAX_JUMBO_FRAME_LEN);
 357
 358                                 rte_pktmbuf_free(mbuf);
 359                                 break;
 360                         }
 361                 }
 362
 363                 if (unlikely(ret != 0))
 364                         break;
 365                 num_tx++;
 366                 tx_bytes += mbuf->pkt_len;
 367                 rte_pktmbuf_free(mbuf);
 368         }
 369
 370         tx_queue->tx_pkts += num_tx;
 371         tx_queue->tx_bytes += tx_bytes;
 372         tx_queue->err_pkts += nb_pkts - num_tx;
 373         return num_tx;
 374 }
 375
 376 static int
 377 eth_dev_start(struct rte_eth_dev *dev)
 378 {
 379         unsigned i;
 380         struct pmd_internals *internals = dev->data->dev_private;
 381         struct pcap_tx_queue *tx;
 382         struct pcap_rx_queue *rx;
 383
 384         /* Special iface case. Single pcap is open and shared between tx/rx. */
 385         if (internals->single_iface) {
 386                 tx = &internals->tx_queue[0];
 387                 rx = &internals->rx_queue[0];
 388
 389                 if (!tx->pcap && strcmp(tx->type, ETH_PCAP_IFACE_ARG) == 0) {
 390                         if (open_single_iface(tx->name, &tx->pcap) < 0)
 391                                 return -1;
 392                         rx->pcap = tx->pcap;
 393                 }
 394                 goto status_up;
 395         }
 396
 397         /* If not open already, open tx pcaps/dumpers */
 398         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 399                 tx = &internals->tx_queue[i];
 400
 401                 if (!tx->dumper && strcmp(tx->type, ETH_PCAP_TX_PCAP_ARG) == 0) {
 402                         if (open_single_tx_pcap(tx->name, &tx->dumper) < 0)
 403                                 return -1;
 404                 }
 405
 406                 else if (!tx->pcap && strcmp(tx->type, ETH_PCAP_TX_IFACE_ARG) == 0) {
 407                         if (open_single_iface(tx->name, &tx->pcap) < 0)
 408                                 return -1;
 409                 }
 410         }
 411
 412         /* If not open already, open rx pcaps */
 413         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 414                 rx = &internals->rx_queue[i];
 415
 416                 if (rx->pcap != NULL)
 417                         continue;
 418
 419                 if (strcmp(rx->type, ETH_PCAP_RX_PCAP_ARG) == 0) {
 420                         if (open_single_rx_pcap(rx->name, &rx->pcap) < 0)
 421                                 return -1;
 422                 }
 423
 424                 else if (strcmp(rx->type, ETH_PCAP_RX_IFACE_ARG) == 0) {
 425                         if (open_single_iface(rx->name, &rx->pcap) < 0)
 426                                 return -1;
 427                 }
 428         }
 429
 430 status_up:
 431
 432         dev->data->dev_link.link_status = ETH_LINK_UP;
 433         return 0;
 434 }
 435
 436 /*
 437  * This function gets called when the current port gets stopped.
 438  * Is the only place for us to close all the tx streams dumpers.
 439  * If not called the dumpers will be flushed within each tx burst.
 440  */
 441 static void
 442 eth_dev_stop(struct rte_eth_dev *dev)
 443 {
 444         unsigned i;
 445         struct pmd_internals *internals = dev->data->dev_private;
 446         struct pcap_tx_queue *tx;
 447         struct pcap_rx_queue *rx;
 448
 449         /* Special iface case. Single pcap is open and shared between tx/rx. */
 450         if (internals->single_iface) {
 451                 tx = &internals->tx_queue[0];
 452                 rx = &internals->rx_queue[0];
 453                 pcap_close(tx->pcap);
 454                 tx->pcap = NULL;
 455                 rx->pcap = NULL;
 456                 goto status_down;
 457         }
 458
 459         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 460                 tx = &internals->tx_queue[i];
 461
 462                 if (tx->dumper != NULL) {
 463                         pcap_dump_close(tx->dumper);
 464                         tx->dumper = NULL;
 465                 }
 466
 467                 if (tx->pcap != NULL) {
 468                         pcap_close(tx->pcap);
 469                         tx->pcap = NULL;
 470                 }
 471         }
 472
 473         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 474                 rx = &internals->rx_queue[i];
 475
 476                 if (rx->pcap != NULL) {
 477                         pcap_close(rx->pcap);
 478                         rx->pcap = NULL;
 479                 }
 480         }
 481
 482 status_down:
 483         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 484 }
 485
 486 static int
 487 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 488 {
 489         return 0;
 490 }
 491
 492 static void
 493 eth_dev_info(struct rte_eth_dev *dev,
 494                 struct rte_eth_dev_info *dev_info)
 495 {
 496         struct pmd_internals *internals = dev->data->dev_private;
 497         dev_info->driver_name = drivername;
 498         dev_info->if_index = internals->if_index;
 499         dev_info->max_mac_addrs = 1;
 500         dev_info->max_rx_pktlen = (uint32_t) -1;
 501         dev_info->max_rx_queues = dev->data->nb_rx_queues;
 502         dev_info->max_tx_queues = dev->data->nb_tx_queues;
 503         dev_info->min_rx_bufsize = 0;
 504         dev_info->pci_dev = NULL;
 505 }
 506
 507 static void
 508 eth_stats_get(struct rte_eth_dev *dev,
 509                 struct rte_eth_stats *igb_stats)
 510 {
 511         unsigned i;
 512         unsigned long rx_packets_total = 0, rx_bytes_total = 0;
 513         unsigned long tx_packets_total = 0, tx_bytes_total = 0;
 514         unsigned long tx_packets_err_total = 0;
 515         const struct pmd_internals *internal = dev->data->dev_private;
 516
 517         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 518                         i < dev->data->nb_rx_queues; i++) {
 519                 igb_stats->q_ipackets[i] = internal->rx_queue[i].rx_pkts;
 520                 igb_stats->q_ibytes[i] = internal->rx_queue[i].rx_bytes;
 521                 rx_packets_total += igb_stats->q_ipackets[i];
 522                 rx_bytes_total += igb_stats->q_ibytes[i];
 523         }
 524
 525         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 526                         i < dev->data->nb_tx_queues; i++) {
 527                 igb_stats->q_opackets[i] = internal->tx_queue[i].tx_pkts;
 528                 igb_stats->q_obytes[i] = internal->tx_queue[i].tx_bytes;
 529                 igb_stats->q_errors[i] = internal->tx_queue[i].err_pkts;
 530                 tx_packets_total += igb_stats->q_opackets[i];
 531                 tx_bytes_total += igb_stats->q_obytes[i];
 532                 tx_packets_err_total += igb_stats->q_errors[i];
 533         }
 534
 535         igb_stats->ipackets = rx_packets_total;
 536         igb_stats->ibytes = rx_bytes_total;
 537         igb_stats->opackets = tx_packets_total;
 538         igb_stats->obytes = tx_bytes_total;
 539         igb_stats->oerrors = tx_packets_err_total;
 540 }
 541
 542 static void
 543 eth_stats_reset(struct rte_eth_dev *dev)
 544 {
 545         unsigned i;
 546         struct pmd_internals *internal = dev->data->dev_private;
 547         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 548                 internal->rx_queue[i].rx_pkts = 0;
 549                 internal->rx_queue[i].rx_bytes = 0;
 550         }
 551         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 552                 internal->tx_queue[i].tx_pkts = 0;
 553                 internal->tx_queue[i].tx_bytes = 0;
 554                 internal->tx_queue[i].err_pkts = 0;
 555         }
 556 }
 557
 558 static void
 559 eth_dev_close(struct rte_eth_dev *dev __rte_unused)
 560 {
 561 }
 562
 563 static void
 564 eth_queue_release(void *q __rte_unused)
 565 {
 566 }
 567
 568 static int
 569 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 570                 int wait_to_complete __rte_unused)
 571 {
 572         return 0;
 573 }
 574
 575 static int
 576 eth_rx_queue_setup(struct rte_eth_dev *dev,
 577                 uint16_t rx_queue_id,
 578                 uint16_t nb_rx_desc __rte_unused,
 579                 unsigned int socket_id __rte_unused,
 580                 const struct rte_eth_rxconf *rx_conf __rte_unused,
 581                 struct rte_mempool *mb_pool)
 582 {
 583         struct pmd_internals *internals = dev->data->dev_private;
 584         struct pcap_rx_queue *pcap_q = &internals->rx_queue[rx_queue_id];
 585         pcap_q->mb_pool = mb_pool;
 586         dev->data->rx_queues[rx_queue_id] = pcap_q;
 587         pcap_q->in_port = dev->data->port_id;
 588         return 0;
 589 }
 590
 591 static int
 592 eth_tx_queue_setup(struct rte_eth_dev *dev,
 593                 uint16_t tx_queue_id,
 594                 uint16_t nb_tx_desc __rte_unused,
 595                 unsigned int socket_id __rte_unused,
 596                 const struct rte_eth_txconf *tx_conf __rte_unused)
 597 {
 598
 599         struct pmd_internals *internals = dev->data->dev_private;
 600         dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
 601         return 0;
 602 }
 603
 604 static const struct eth_dev_ops ops = {
 605         .dev_start = eth_dev_start,
 606         .dev_stop =     eth_dev_stop,
 607         .dev_close = eth_dev_close,
 608         .dev_configure = eth_dev_configure,
 609         .dev_infos_get = eth_dev_info,
 610         .rx_queue_setup = eth_rx_queue_setup,
 611         .tx_queue_setup = eth_tx_queue_setup,
 612         .rx_queue_release = eth_queue_release,
 613         .tx_queue_release = eth_queue_release,
 614         .link_update = eth_link_update,
 615         .stats_get = eth_stats_get,
 616         .stats_reset = eth_stats_reset,
 617 };
 618
 619 /*
 620  * Function handler that opens the pcap file for reading a stores a
 621  * reference of it for use it later on.
 622  */
 623 static int
 624 open_rx_pcap(const char *key, const char *value, void *extra_args)
 625 {
 626         unsigned i;
 627         const char *pcap_filename = value;
 628         struct rx_pcaps *pcaps = extra_args;
 629         pcap_t *pcap = NULL;
 630
 631         for (i = 0; i < pcaps->num_of_rx; i++) {
 632                 if (open_single_rx_pcap(pcap_filename, &pcap) < 0)
 633                         return -1;
 634
 635                 pcaps->pcaps[i] = pcap;
 636                 pcaps->names[i] = pcap_filename;
 637                 pcaps->types[i] = key;
 638         }
 639
 640         return 0;
 641 }
 642
 643 static int
 644 open_single_rx_pcap(const char *pcap_filename, pcap_t **pcap)
 645 {
 646         if ((*pcap = pcap_open_offline(pcap_filename, errbuf)) == NULL) {
 647                 RTE_LOG(ERR, PMD, "Couldn't open %s: %s\n", pcap_filename, errbuf);
 648                 return -1;
 649         }
 650         return 0;
 651 }
 652
 653 /*
 654  * Opens a pcap file for writing and stores a reference to it
 655  * for use it later on.
 656  */
 657 static int
 658 open_tx_pcap(const char *key, const char *value, void *extra_args)
 659 {
 660         unsigned i;
 661         const char *pcap_filename = value;
 662         struct tx_pcaps *dumpers = extra_args;
 663         pcap_dumper_t *dumper;
 664
 665         for (i = 0; i < dumpers->num_of_tx; i++) {
 666                 if (open_single_tx_pcap(pcap_filename, &dumper) < 0)
 667                         return -1;
 668
 669                 dumpers->dumpers[i] = dumper;
 670                 dumpers->names[i] = pcap_filename;
 671                 dumpers->types[i] = key;
 672         }
 673
 674         return 0;
 675 }
 676
 677 static int
 678 open_single_tx_pcap(const char *pcap_filename, pcap_dumper_t **dumper)
 679 {
 680         pcap_t *tx_pcap;
 681         /*
 682          * We need to create a dummy empty pcap_t to use it
 683          * with pcap_dump_open(). We create big enough an Ethernet
 684          * pcap holder.
 685          */
 686
 687         if ((tx_pcap = pcap_open_dead(DLT_EN10MB, RTE_ETH_PCAP_SNAPSHOT_LEN))
 688                         == NULL) {
 689                 RTE_LOG(ERR, PMD, "Couldn't create dead pcap\n");
 690                 return -1;
 691         }
 692
 693         /* The dumper is created using the previous pcap_t reference */
 694         if ((*dumper = pcap_dump_open(tx_pcap, pcap_filename)) == NULL) {
 695                 RTE_LOG(ERR, PMD, "Couldn't open %s for writing.\n", pcap_filename);
 696                 return -1;
 697         }
 698
 699         return 0;
 700 }
 701
 702 /*
 703  * pcap_open_live wrapper function
 704  */
 705 static inline int
 706 open_iface_live(const char *iface, pcap_t **pcap) {
 707         *pcap = pcap_open_live(iface, RTE_ETH_PCAP_SNAPLEN,
 708                         RTE_ETH_PCAP_PROMISC, RTE_ETH_PCAP_TIMEOUT, errbuf);
 709
 710         if (*pcap == NULL) {
 711                 RTE_LOG(ERR, PMD, "Couldn't open %s: %s\n", iface, errbuf);
 712                 return -1;
 713         }
 714         return 0;
 715 }
 716
 717 /*
 718  * Opens an interface for reading and writing
 719  */
 720 static inline int
 721 open_rx_tx_iface(const char *key, const char *value, void *extra_args)
 722 {
 723         const char *iface = value;
 724         struct rx_pcaps *pcaps = extra_args;
 725         pcap_t *pcap = NULL;
 726
 727         if (open_single_iface(iface, &pcap) < 0)
 728                 return -1;
 729
 730         pcaps->pcaps[0] = pcap;
 731         pcaps->names[0] = iface;
 732         pcaps->types[0] = key;
 733
 734         return 0;
 735 }
 736
 737 /*
 738  * Opens a NIC for reading packets from it
 739  */
 740 static inline int
 741 open_rx_iface(const char *key, const char *value, void *extra_args)
 742 {
 743         unsigned i;
 744         const char *iface = value;
 745         struct rx_pcaps *pcaps = extra_args;
 746         pcap_t *pcap = NULL;
 747
 748         for (i = 0; i < pcaps->num_of_rx; i++) {
 749                 if (open_single_iface(iface, &pcap) < 0)
 750                         return -1;
 751                 pcaps->pcaps[i] = pcap;
 752                 pcaps->names[i] = iface;
 753                 pcaps->types[i] = key;
 754         }
 755
 756         return 0;
 757 }
 758
 759 /*
 760  * Opens a NIC for writing packets to it
 761  */
 762 static int
 763 open_tx_iface(const char *key, const char *value, void *extra_args)
 764 {
 765         unsigned i;
 766         const char *iface = value;
 767         struct tx_pcaps *pcaps = extra_args;
 768         pcap_t *pcap;
 769
 770         for (i = 0; i < pcaps->num_of_tx; i++) {
 771                 if (open_single_iface(iface, &pcap) < 0)
 772                         return -1;
 773                 pcaps->pcaps[i] = pcap;
 774                 pcaps->names[i] = iface;
 775                 pcaps->types[i] = key;
 776         }
 777
 778         return 0;
 779 }
 780
 781 static int
 782 open_single_iface(const char *iface, pcap_t **pcap)
 783 {
 784         if (open_iface_live(iface, pcap) < 0) {
 785                 RTE_LOG(ERR, PMD, "Couldn't open interface %s\n", iface);
 786                 return -1;
 787         }
 788
 789         return 0;
 790 }
 791
 792 static int
 793 rte_pmd_init_internals(const char *name, const unsigned nb_rx_queues,
 794                 const unsigned nb_tx_queues,
 795                 const unsigned numa_node,
 796                 struct pmd_internals **internals,
 797                 struct rte_eth_dev **eth_dev,
 798                 struct rte_kvargs *kvlist)
 799 {
 800         struct rte_eth_dev_data *data = NULL;
 801         unsigned k_idx;
 802         struct rte_kvargs_pair *pair = NULL;
 803
 804         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 805                 pair = &kvlist->pairs[k_idx];
 806                 if (strstr(pair->key, ETH_PCAP_IFACE_ARG) != NULL)
 807                         break;
 808         }
 809
 810         RTE_LOG(INFO, PMD,
 811                         "Creating pcap-backed ethdev on numa socket %u\n", numa_node);
 812
 813         /* now do all data allocation - for eth_dev structure
 814          * and internal (private) data
 815          */
 816         data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
 817         if (data == NULL)
 818                 goto error;
 819
 820         *internals = rte_zmalloc_socket(name, sizeof(**internals), 0, numa_node);
 821         if (*internals == NULL)
 822                 goto error;
 823
 824         /* reserve an ethdev entry */
 825         *eth_dev = rte_eth_dev_allocate(name, RTE_ETH_DEV_VIRTUAL);
 826         if (*eth_dev == NULL)
 827                 goto error;
 828
 829         /* check length of device name */
 830         if ((strlen((*eth_dev)->data->name) + 1) > sizeof(data->name))
 831                 goto error;
 832
 833         /* now put it all together
 834          * - store queue data in internals,
 835          * - store numa_node info in eth_dev
 836          * - point eth_dev_data to internals
 837          * - and point eth_dev structure to new eth_dev_data structure
 838          */
 839         /* NOTE: we'll replace the data element, of originally allocated eth_dev
 840          * so the rings are local per-process */
 841
 842         if (pair == NULL)
 843                 (*internals)->if_index = 0;
 844         else
 845                 (*internals)->if_index = if_nametoindex(pair->value);
 846
 847         data->dev_private = *internals;
 848         data->port_id = (*eth_dev)->data->port_id;
 849         snprintf(data->name, sizeof(data->name), "%s", (*eth_dev)->data->name);
 850         data->nb_rx_queues = (uint16_t)nb_rx_queues;
 851         data->nb_tx_queues = (uint16_t)nb_tx_queues;
 852         data->dev_link = pmd_link;
 853         data->mac_addrs = &eth_addr;
 854         strncpy(data->name,
 855                 (*eth_dev)->data->name, strlen((*eth_dev)->data->name));
 856
 857         (*eth_dev)->data = data;
 858         (*eth_dev)->dev_ops = &ops;
 859         (*eth_dev)->driver = NULL;
 860         data->dev_flags = RTE_ETH_DEV_DETACHABLE;
 861         data->kdrv = RTE_KDRV_NONE;
 862         data->drv_name = drivername;
 863         data->numa_node = numa_node;
 864
 865         return 0;
 866
 867 error:
 868         rte_free(data);
 869         rte_free(*internals);
 870
 871         return -1;
 872 }
 873
 874 static int
 875 rte_eth_from_pcaps_common(const char *name, struct rx_pcaps *rx_queues,
 876                 const unsigned nb_rx_queues, struct tx_pcaps *tx_queues,
 877                 const unsigned nb_tx_queues, const unsigned numa_node,
 878                 struct rte_kvargs *kvlist, struct pmd_internals **internals,
 879                 struct rte_eth_dev **eth_dev)
 880 {
 881         unsigned i;
 882
 883         /* do some parameter checking */
 884         if (rx_queues == NULL && nb_rx_queues > 0)
 885                 return -1;
 886         if (tx_queues == NULL && nb_tx_queues > 0)
 887                 return -1;
 888
 889         if (rte_pmd_init_internals(name, nb_rx_queues, nb_tx_queues, numa_node,
 890                         internals, eth_dev, kvlist) < 0)
 891                 return -1;
 892
 893         for (i = 0; i < nb_rx_queues; i++) {
 894                 (*internals)->rx_queue[i].pcap = rx_queues->pcaps[i];
 895                 snprintf((*internals)->rx_queue[i].name,
 896                         sizeof((*internals)->rx_queue[i].name), "%s",
 897                         rx_queues->names[i]);
 898                 snprintf((*internals)->rx_queue[i].type,
 899                         sizeof((*internals)->rx_queue[i].type), "%s",
 900                         rx_queues->types[i]);
 901         }
 902         for (i = 0; i < nb_tx_queues; i++) {
 903                 (*internals)->tx_queue[i].dumper = tx_queues->dumpers[i];
 904                 snprintf((*internals)->tx_queue[i].name,
 905                         sizeof((*internals)->tx_queue[i].name), "%s",
 906                         tx_queues->names[i]);
 907                 snprintf((*internals)->tx_queue[i].type,
 908                         sizeof((*internals)->tx_queue[i].type), "%s",
 909                         tx_queues->types[i]);
 910         }
 911
 912         return 0;
 913 }
 914
 915 static int
 916 rte_eth_from_pcaps_n_dumpers(const char *name,
 917                 struct rx_pcaps *rx_queues,
 918                 const unsigned nb_rx_queues,
 919                 struct tx_pcaps *tx_queues,
 920                 const unsigned nb_tx_queues,
 921                 const unsigned numa_node,
 922                 struct rte_kvargs *kvlist)
 923 {
 924         struct pmd_internals *internals = NULL;
 925         struct rte_eth_dev *eth_dev = NULL;
 926         int ret;
 927
 928         ret = rte_eth_from_pcaps_common(name, rx_queues, nb_rx_queues,
 929                         tx_queues, nb_tx_queues, numa_node, kvlist,
 930                         &internals, &eth_dev);
 931
 932         if (ret < 0)
 933                 return ret;
 934
 935         /* using multiple pcaps/interfaces */
 936         internals->single_iface = 0;
 937
 938         eth_dev->rx_pkt_burst = eth_pcap_rx;
 939         eth_dev->tx_pkt_burst = eth_pcap_tx_dumper;
 940
 941         return 0;
 942 }
 943
 944 static int
 945 rte_eth_from_pcaps(const char *name,
 946                 struct rx_pcaps *rx_queues,
 947                 const unsigned nb_rx_queues,
 948                 struct tx_pcaps *tx_queues,
 949                 const unsigned nb_tx_queues,
 950                 const unsigned numa_node,
 951                 struct rte_kvargs *kvlist,
 952                 int single_iface)
 953 {
 954         struct pmd_internals *internals = NULL;
 955         struct rte_eth_dev *eth_dev = NULL;
 956         int ret;
 957
 958         ret = rte_eth_from_pcaps_common(name, rx_queues, nb_rx_queues,
 959                         tx_queues, nb_tx_queues, numa_node, kvlist,
 960                         &internals, &eth_dev);
 961
 962         if (ret < 0)
 963                 return ret;
 964
 965         /* store wether we are using a single interface for rx/tx or not */
 966         internals->single_iface = single_iface;
 967
 968         eth_dev->rx_pkt_burst = eth_pcap_rx;
 969         eth_dev->tx_pkt_burst = eth_pcap_tx;
 970
 971         return 0;
 972 }
 973
 974
 975 static int
 976 rte_pmd_pcap_devinit(const char *name, const char *params)
 977 {
 978         unsigned numa_node, using_dumpers = 0;
 979         int ret;
 980         struct rte_kvargs *kvlist;
 981         struct rx_pcaps pcaps = {0};
 982         struct tx_pcaps dumpers = {0};
 983
 984         RTE_LOG(INFO, PMD, "Initializing pmd_pcap for %s\n", name);
 985
 986         numa_node = rte_socket_id();
 987
 988         gettimeofday(&start_time, NULL);
 989         start_cycles = rte_get_timer_cycles();
 990         hz = rte_get_timer_hz();
 991
 992         kvlist = rte_kvargs_parse(params, valid_arguments);
 993         if (kvlist == NULL)
 994                 return -1;
 995
 996         /*
 997          * If iface argument is passed we open the NICs and use them for
 998          * reading / writing
 999          */
1000         if (rte_kvargs_count(kvlist, ETH_PCAP_IFACE_ARG) == 1) {
1001
1002                 ret = rte_kvargs_process(kvlist, ETH_PCAP_IFACE_ARG,
1003                                 &open_rx_tx_iface, &pcaps);
1004                 if (ret < 0)
1005                         goto free_kvlist;
1006                 dumpers.pcaps[0] = pcaps.pcaps[0];
1007                 dumpers.names[0] = pcaps.names[0];
1008                 dumpers.types[0] = pcaps.types[0];
1009                 ret = rte_eth_from_pcaps(name, &pcaps, 1, &dumpers, 1,
1010                                 numa_node, kvlist, 1);
1011                 goto free_kvlist;
1012         }
1013
1014         /*
1015          * We check whether we want to open a RX stream from a real NIC or a
1016          * pcap file
1017          */
1018         if ((pcaps.num_of_rx = rte_kvargs_count(kvlist, ETH_PCAP_RX_PCAP_ARG))) {
1019                 ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_PCAP_ARG,
1020                                 &open_rx_pcap, &pcaps);
1021         } else {
1022                 pcaps.num_of_rx = rte_kvargs_count(kvlist,
1023                                 ETH_PCAP_RX_IFACE_ARG);
1024                 ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_IFACE_ARG,
1025                                 &open_rx_iface, &pcaps);
1026         }
1027
1028         if (ret < 0)
1029                 goto free_kvlist;
1030
1031         /*
1032          * We check whether we want to open a TX stream to a real NIC or a
1033          * pcap file
1034          */
1035         if ((dumpers.num_of_tx = rte_kvargs_count(kvlist,
1036                         ETH_PCAP_TX_PCAP_ARG))) {
1037                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_PCAP_ARG,
1038                                 &open_tx_pcap, &dumpers);
1039                 using_dumpers = 1;
1040         } else {
1041                 dumpers.num_of_tx = rte_kvargs_count(kvlist,
1042                                 ETH_PCAP_TX_IFACE_ARG);
1043                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_IFACE_ARG,
1044                                 &open_tx_iface, &dumpers);
1045         }
1046
1047         if (ret < 0)
1048                 goto free_kvlist;
1049
1050         if (using_dumpers)
1051                 ret = rte_eth_from_pcaps_n_dumpers(name, &pcaps, pcaps.num_of_rx,
1052                                 &dumpers, dumpers.num_of_tx, numa_node, kvlist);
1053         else
1054                 ret = rte_eth_from_pcaps(name, &pcaps, pcaps.num_of_rx, &dumpers,
1055                         dumpers.num_of_tx, numa_node, kvlist, 0);
1056
1057 free_kvlist:
1058         rte_kvargs_free(kvlist);
1059         return ret;
1060 }
1061
1062 static int
1063 rte_pmd_pcap_devuninit(const char *name)
1064 {
1065         struct rte_eth_dev *eth_dev = NULL;
1066
1067         RTE_LOG(INFO, PMD, "Closing pcap ethdev on numa socket %u\n",
1068                         rte_socket_id());
1069
1070         if (name == NULL)
1071                 return -1;
1072
1073         /* reserve an ethdev entry */
1074         eth_dev = rte_eth_dev_allocated(name);
1075         if (eth_dev == NULL)
1076                 return -1;
1077
1078         rte_free(eth_dev->data->dev_private);
1079         rte_free(eth_dev->data);
1080
1081         rte_eth_dev_release_port(eth_dev);
1082
1083         return 0;
1084 }
1085
1086 static struct rte_driver pmd_pcap_drv = {
1087         .type = PMD_VDEV,
1088         .init = rte_pmd_pcap_devinit,
1089         .uninit = rte_pmd_pcap_devuninit,
1090 };
1091
1092 PMD_REGISTER_DRIVER(pmd_pcap_drv, eth_pcap);
1093 DRIVER_REGISTER_PARAM_STRING(eth_pcap,
1094         "rx_pcap=<string> "
1095         "tx_pcap=<string> "
1096         "rx_iface=<ifc> "
1097         "tx_iface=<ifc> "
1098         "iface=<ifc>");