drivers/net/pcap/rte_eth_pcap.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2016 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
  37 #include <net/if.h>
  38
  39 #include <pcap.h>
  40
  41 #include <rte_cycles.h>
  42 #include <rte_ethdev.h>
  43 #include <rte_kvargs.h>
  44 #include <rte_malloc.h>
  45 #include <rte_mbuf.h>
  46 #include <rte_vdev.h>
  47
  48 #define RTE_ETH_PCAP_SNAPSHOT_LEN 65535
  49 #define RTE_ETH_PCAP_SNAPLEN ETHER_MAX_JUMBO_FRAME_LEN
  50 #define RTE_ETH_PCAP_PROMISC 1
  51 #define RTE_ETH_PCAP_TIMEOUT -1
  52 #define ETH_PCAP_RX_PCAP_ARG  "rx_pcap"
  53 #define ETH_PCAP_TX_PCAP_ARG  "tx_pcap"
  54 #define ETH_PCAP_RX_IFACE_ARG "rx_iface"
  55 #define ETH_PCAP_TX_IFACE_ARG "tx_iface"
  56 #define ETH_PCAP_IFACE_ARG    "iface"
  57
  58 #define ETH_PCAP_ARG_MAXLEN     64
  59
  60 #define RTE_PMD_PCAP_MAX_QUEUES 16
  61
  62 static char errbuf[PCAP_ERRBUF_SIZE];
  63 static unsigned char tx_pcap_data[RTE_ETH_PCAP_SNAPLEN];
  64 static struct timeval start_time;
  65 static uint64_t start_cycles;
  66 static uint64_t hz;
  67
  68 struct queue_stat {
  69         volatile unsigned long pkts;
  70         volatile unsigned long bytes;
  71         volatile unsigned long err_pkts;
  72 };
  73
  74 struct pcap_rx_queue {
  75         pcap_t *pcap;
  76         uint8_t in_port;
  77         struct rte_mempool *mb_pool;
  78         struct queue_stat rx_stat;
  79         char name[PATH_MAX];
  80         char type[ETH_PCAP_ARG_MAXLEN];
  81 };
  82
  83 struct pcap_tx_queue {
  84         pcap_dumper_t *dumper;
  85         pcap_t *pcap;
  86         struct queue_stat tx_stat;
  87         char name[PATH_MAX];
  88         char type[ETH_PCAP_ARG_MAXLEN];
  89 };
  90
  91 struct pmd_internals {
  92         struct pcap_rx_queue rx_queue[RTE_PMD_PCAP_MAX_QUEUES];
  93         struct pcap_tx_queue tx_queue[RTE_PMD_PCAP_MAX_QUEUES];
  94         int if_index;
  95         int single_iface;
  96 };
  97
  98 struct pmd_devargs {
  99         unsigned num_of_queue;
 100         struct devargs_queue {
 101                 pcap_dumper_t *dumper;
 102                 pcap_t *pcap;
 103                 const char *name;
 104                 const char *type;
 105         } queue[RTE_PMD_PCAP_MAX_QUEUES];
 106 };
 107
 108 static const char *valid_arguments[] = {
 109         ETH_PCAP_RX_PCAP_ARG,
 110         ETH_PCAP_TX_PCAP_ARG,
 111         ETH_PCAP_RX_IFACE_ARG,
 112         ETH_PCAP_TX_IFACE_ARG,
 113         ETH_PCAP_IFACE_ARG,
 114         NULL
 115 };
 116
 117 static int open_single_tx_pcap(const char *pcap_filename, pcap_dumper_t **dumper);
 118 static int open_single_rx_pcap(const char *pcap_filename, pcap_t **pcap);
 119 static int open_single_iface(const char *iface, pcap_t **pcap);
 120
 121 static struct ether_addr eth_addr = { .addr_bytes = { 0, 0, 0, 0x1, 0x2, 0x3 } };
 122 static const char *drivername = "Pcap PMD";
 123 static struct rte_eth_link pmd_link = {
 124                 .link_speed = ETH_SPEED_NUM_10G,
 125                 .link_duplex = ETH_LINK_FULL_DUPLEX,
 126                 .link_status = ETH_LINK_DOWN,
 127                 .link_autoneg = ETH_LINK_SPEED_FIXED,
 128 };
 129
 130 static int
 131 eth_pcap_rx_jumbo(struct rte_mempool *mb_pool,
 132                   struct rte_mbuf *mbuf,
 133                   const u_char *data,
 134                   uint16_t data_len)
 135 {
 136         struct rte_mbuf *m = mbuf;
 137
 138         /* Copy the first segment. */
 139         uint16_t len = rte_pktmbuf_tailroom(mbuf);
 140
 141         rte_memcpy(rte_pktmbuf_append(mbuf, len), data, len);
 142         data_len -= len;
 143         data += len;
 144
 145         while (data_len > 0) {
 146                 /* Allocate next mbuf and point to that. */
 147                 m->next = rte_pktmbuf_alloc(mb_pool);
 148
 149                 if (unlikely(!m->next))
 150                         return -1;
 151
 152                 m = m->next;
 153
 154                 /* Headroom is not needed in chained mbufs. */
 155                 rte_pktmbuf_prepend(m, rte_pktmbuf_headroom(m));
 156                 m->pkt_len = 0;
 157                 m->data_len = 0;
 158
 159                 /* Copy next segment. */
 160                 len = RTE_MIN(rte_pktmbuf_tailroom(m), data_len);
 161                 rte_memcpy(rte_pktmbuf_append(m, len), data, len);
 162
 163                 mbuf->nb_segs++;
 164                 data_len -= len;
 165                 data += len;
 166         }
 167
 168         return mbuf->nb_segs;
 169 }
 170
 171 /* Copy data from mbuf chain to a buffer suitable for writing to a PCAP file. */
 172 static void
 173 eth_pcap_gather_data(unsigned char *data, struct rte_mbuf *mbuf)
 174 {
 175         uint16_t data_len = 0;
 176
 177         while (mbuf) {
 178                 rte_memcpy(data + data_len, rte_pktmbuf_mtod(mbuf, void *),
 179                            mbuf->data_len);
 180
 181                 data_len += mbuf->data_len;
 182                 mbuf = mbuf->next;
 183         }
 184 }
 185
 186 static uint16_t
 187 eth_pcap_rx(void *queue,
 188                 struct rte_mbuf **bufs,
 189                 uint16_t nb_pkts)
 190 {
 191         unsigned i;
 192         struct pcap_pkthdr header;
 193         const u_char *packet;
 194         struct rte_mbuf *mbuf;
 195         struct pcap_rx_queue *pcap_q = queue;
 196         uint16_t num_rx = 0;
 197         uint16_t buf_size;
 198         uint32_t rx_bytes = 0;
 199
 200         if (unlikely(pcap_q->pcap == NULL || nb_pkts == 0))
 201                 return 0;
 202
 203         /* Reads the given number of packets from the pcap file one by one
 204          * and copies the packet data into a newly allocated mbuf to return.
 205          */
 206         for (i = 0; i < nb_pkts; i++) {
 207                 /* Get the next PCAP packet */
 208                 packet = pcap_next(pcap_q->pcap, &header);
 209                 if (unlikely(packet == NULL))
 210                         break;
 211                 else
 212                         mbuf = rte_pktmbuf_alloc(pcap_q->mb_pool);
 213                 if (unlikely(mbuf == NULL))
 214                         break;
 215
 216                 /* Now get the space available for data in the mbuf */
 217                 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(pcap_q->mb_pool) -
 218                                 RTE_PKTMBUF_HEADROOM);
 219
 220                 if (header.caplen <= buf_size) {
 221                         /* pcap packet will fit in the mbuf, go ahead and copy */
 222                         rte_memcpy(rte_pktmbuf_mtod(mbuf, void *), packet,
 223                                         header.caplen);
 224                         mbuf->data_len = (uint16_t)header.caplen;
 225                 } else {
 226                         /* Try read jumbo frame into multi mbufs. */
 227                         if (unlikely(eth_pcap_rx_jumbo(pcap_q->mb_pool,
 228                                                        mbuf,
 229                                                        packet,
 230                                                        header.caplen) == -1))
 231                                 break;
 232                 }
 233
 234                 mbuf->pkt_len = (uint16_t)header.caplen;
 235                 mbuf->port = pcap_q->in_port;
 236                 bufs[num_rx] = mbuf;
 237                 num_rx++;
 238                 rx_bytes += header.caplen;
 239         }
 240         pcap_q->rx_stat.pkts += num_rx;
 241         pcap_q->rx_stat.bytes += rx_bytes;
 242         return num_rx;
 243 }
 244
 245 static inline void
 246 calculate_timestamp(struct timeval *ts) {
 247         uint64_t cycles;
 248         struct timeval cur_time;
 249
 250         cycles = rte_get_timer_cycles() - start_cycles;
 251         cur_time.tv_sec = cycles / hz;
 252         cur_time.tv_usec = (cycles % hz) * 10e6 / hz;
 253         timeradd(&start_time, &cur_time, ts);
 254 }
 255
 256 /*
 257  * Callback to handle writing packets to a pcap file.
 258  */
 259 static uint16_t
 260 eth_pcap_tx_dumper(void *queue,
 261                 struct rte_mbuf **bufs,
 262                 uint16_t nb_pkts)
 263 {
 264         unsigned i;
 265         struct rte_mbuf *mbuf;
 266         struct pcap_tx_queue *dumper_q = queue;
 267         uint16_t num_tx = 0;
 268         uint32_t tx_bytes = 0;
 269         struct pcap_pkthdr header;
 270
 271         if (dumper_q->dumper == NULL || nb_pkts == 0)
 272                 return 0;
 273
 274         /* writes the nb_pkts packets to the previously opened pcap file dumper */
 275         for (i = 0; i < nb_pkts; i++) {
 276                 mbuf = bufs[i];
 277                 calculate_timestamp(&header.ts);
 278                 header.len = mbuf->pkt_len;
 279                 header.caplen = header.len;
 280
 281                 if (likely(mbuf->nb_segs == 1)) {
 282                         pcap_dump((u_char *)dumper_q->dumper, &header,
 283                                   rte_pktmbuf_mtod(mbuf, void*));
 284                 } else {
 285                         if (mbuf->pkt_len <= ETHER_MAX_JUMBO_FRAME_LEN) {
 286                                 eth_pcap_gather_data(tx_pcap_data, mbuf);
 287                                 pcap_dump((u_char *)dumper_q->dumper, &header,
 288                                           tx_pcap_data);
 289                         } else {
 290                                 RTE_LOG(ERR, PMD,
 291                                         "Dropping PCAP packet. "
 292                                         "Size (%d) > max jumbo size (%d).\n",
 293                                         mbuf->pkt_len,
 294                                         ETHER_MAX_JUMBO_FRAME_LEN);
 295
 296                                 rte_pktmbuf_free(mbuf);
 297                                 break;
 298                         }
 299                 }
 300
 301                 rte_pktmbuf_free(mbuf);
 302                 num_tx++;
 303                 tx_bytes += mbuf->pkt_len;
 304         }
 305
 306         /*
 307          * Since there's no place to hook a callback when the forwarding
 308          * process stops and to make sure the pcap file is actually written,
 309          * we flush the pcap dumper within each burst.
 310          */
 311         pcap_dump_flush(dumper_q->dumper);
 312         dumper_q->tx_stat.pkts += num_tx;
 313         dumper_q->tx_stat.bytes += tx_bytes;
 314         dumper_q->tx_stat.err_pkts += nb_pkts - num_tx;
 315         return num_tx;
 316 }
 317
 318 /*
 319  * Callback to handle sending packets through a real NIC.
 320  */
 321 static uint16_t
 322 eth_pcap_tx(void *queue,
 323                 struct rte_mbuf **bufs,
 324                 uint16_t nb_pkts)
 325 {
 326         unsigned i;
 327         int ret;
 328         struct rte_mbuf *mbuf;
 329         struct pcap_tx_queue *tx_queue = queue;
 330         uint16_t num_tx = 0;
 331         uint32_t tx_bytes = 0;
 332
 333         if (unlikely(nb_pkts == 0 || tx_queue->pcap == NULL))
 334                 return 0;
 335
 336         for (i = 0; i < nb_pkts; i++) {
 337                 mbuf = bufs[i];
 338
 339                 if (likely(mbuf->nb_segs == 1)) {
 340                         ret = pcap_sendpacket(tx_queue->pcap,
 341                                               rte_pktmbuf_mtod(mbuf, u_char *),
 342                                               mbuf->pkt_len);
 343                 } else {
 344                         if (mbuf->pkt_len <= ETHER_MAX_JUMBO_FRAME_LEN) {
 345                                 eth_pcap_gather_data(tx_pcap_data, mbuf);
 346                                 ret = pcap_sendpacket(tx_queue->pcap,
 347                                                       tx_pcap_data,
 348                                                       mbuf->pkt_len);
 349                         } else {
 350                                 RTE_LOG(ERR, PMD,
 351                                         "Dropping PCAP packet. "
 352                                         "Size (%d) > max jumbo size (%d).\n",
 353                                         mbuf->pkt_len,
 354                                         ETHER_MAX_JUMBO_FRAME_LEN);
 355
 356                                 rte_pktmbuf_free(mbuf);
 357                                 break;
 358                         }
 359                 }
 360
 361                 if (unlikely(ret != 0))
 362                         break;
 363                 num_tx++;
 364                 tx_bytes += mbuf->pkt_len;
 365                 rte_pktmbuf_free(mbuf);
 366         }
 367
 368         tx_queue->tx_stat.pkts += num_tx;
 369         tx_queue->tx_stat.bytes += tx_bytes;
 370         tx_queue->tx_stat.err_pkts += nb_pkts - num_tx;
 371         return num_tx;
 372 }
 373
 374 static int
 375 eth_dev_start(struct rte_eth_dev *dev)
 376 {
 377         unsigned i;
 378         struct pmd_internals *internals = dev->data->dev_private;
 379         struct pcap_tx_queue *tx;
 380         struct pcap_rx_queue *rx;
 381
 382         /* Special iface case. Single pcap is open and shared between tx/rx. */
 383         if (internals->single_iface) {
 384                 tx = &internals->tx_queue[0];
 385                 rx = &internals->rx_queue[0];
 386
 387                 if (!tx->pcap && strcmp(tx->type, ETH_PCAP_IFACE_ARG) == 0) {
 388                         if (open_single_iface(tx->name, &tx->pcap) < 0)
 389                                 return -1;
 390                         rx->pcap = tx->pcap;
 391                 }
 392                 goto status_up;
 393         }
 394
 395         /* If not open already, open tx pcaps/dumpers */
 396         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 397                 tx = &internals->tx_queue[i];
 398
 399                 if (!tx->dumper && strcmp(tx->type, ETH_PCAP_TX_PCAP_ARG) == 0) {
 400                         if (open_single_tx_pcap(tx->name, &tx->dumper) < 0)
 401                                 return -1;
 402                 }
 403
 404                 else if (!tx->pcap && strcmp(tx->type, ETH_PCAP_TX_IFACE_ARG) == 0) {
 405                         if (open_single_iface(tx->name, &tx->pcap) < 0)
 406                                 return -1;
 407                 }
 408         }
 409
 410         /* If not open already, open rx pcaps */
 411         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 412                 rx = &internals->rx_queue[i];
 413
 414                 if (rx->pcap != NULL)
 415                         continue;
 416
 417                 if (strcmp(rx->type, ETH_PCAP_RX_PCAP_ARG) == 0) {
 418                         if (open_single_rx_pcap(rx->name, &rx->pcap) < 0)
 419                                 return -1;
 420                 }
 421
 422                 else if (strcmp(rx->type, ETH_PCAP_RX_IFACE_ARG) == 0) {
 423                         if (open_single_iface(rx->name, &rx->pcap) < 0)
 424                                 return -1;
 425                 }
 426         }
 427
 428 status_up:
 429
 430         dev->data->dev_link.link_status = ETH_LINK_UP;
 431         return 0;
 432 }
 433
 434 /*
 435  * This function gets called when the current port gets stopped.
 436  * Is the only place for us to close all the tx streams dumpers.
 437  * If not called the dumpers will be flushed within each tx burst.
 438  */
 439 static void
 440 eth_dev_stop(struct rte_eth_dev *dev)
 441 {
 442         unsigned i;
 443         struct pmd_internals *internals = dev->data->dev_private;
 444         struct pcap_tx_queue *tx;
 445         struct pcap_rx_queue *rx;
 446
 447         /* Special iface case. Single pcap is open and shared between tx/rx. */
 448         if (internals->single_iface) {
 449                 tx = &internals->tx_queue[0];
 450                 rx = &internals->rx_queue[0];
 451                 pcap_close(tx->pcap);
 452                 tx->pcap = NULL;
 453                 rx->pcap = NULL;
 454                 goto status_down;
 455         }
 456
 457         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 458                 tx = &internals->tx_queue[i];
 459
 460                 if (tx->dumper != NULL) {
 461                         pcap_dump_close(tx->dumper);
 462                         tx->dumper = NULL;
 463                 }
 464
 465                 if (tx->pcap != NULL) {
 466                         pcap_close(tx->pcap);
 467                         tx->pcap = NULL;
 468                 }
 469         }
 470
 471         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 472                 rx = &internals->rx_queue[i];
 473
 474                 if (rx->pcap != NULL) {
 475                         pcap_close(rx->pcap);
 476                         rx->pcap = NULL;
 477                 }
 478         }
 479
 480 status_down:
 481         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 482 }
 483
 484 static int
 485 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 486 {
 487         return 0;
 488 }
 489
 490 static void
 491 eth_dev_info(struct rte_eth_dev *dev,
 492                 struct rte_eth_dev_info *dev_info)
 493 {
 494         struct pmd_internals *internals = dev->data->dev_private;
 495         dev_info->driver_name = drivername;
 496         dev_info->if_index = internals->if_index;
 497         dev_info->max_mac_addrs = 1;
 498         dev_info->max_rx_pktlen = (uint32_t) -1;
 499         dev_info->max_rx_queues = dev->data->nb_rx_queues;
 500         dev_info->max_tx_queues = dev->data->nb_tx_queues;
 501         dev_info->min_rx_bufsize = 0;
 502         dev_info->pci_dev = NULL;
 503 }
 504
 505 static void
 506 eth_stats_get(struct rte_eth_dev *dev,
 507                 struct rte_eth_stats *stats)
 508 {
 509         unsigned i;
 510         unsigned long rx_packets_total = 0, rx_bytes_total = 0;
 511         unsigned long tx_packets_total = 0, tx_bytes_total = 0;
 512         unsigned long tx_packets_err_total = 0;
 513         const struct pmd_internals *internal = dev->data->dev_private;
 514
 515         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 516                         i < dev->data->nb_rx_queues; i++) {
 517                 stats->q_ipackets[i] = internal->rx_queue[i].rx_stat.pkts;
 518                 stats->q_ibytes[i] = internal->rx_queue[i].rx_stat.bytes;
 519                 rx_packets_total += stats->q_ipackets[i];
 520                 rx_bytes_total += stats->q_ibytes[i];
 521         }
 522
 523         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 524                         i < dev->data->nb_tx_queues; i++) {
 525                 stats->q_opackets[i] = internal->tx_queue[i].tx_stat.pkts;
 526                 stats->q_obytes[i] = internal->tx_queue[i].tx_stat.bytes;
 527                 stats->q_errors[i] = internal->tx_queue[i].tx_stat.err_pkts;
 528                 tx_packets_total += stats->q_opackets[i];
 529                 tx_bytes_total += stats->q_obytes[i];
 530                 tx_packets_err_total += stats->q_errors[i];
 531         }
 532
 533         stats->ipackets = rx_packets_total;
 534         stats->ibytes = rx_bytes_total;
 535         stats->opackets = tx_packets_total;
 536         stats->obytes = tx_bytes_total;
 537         stats->oerrors = tx_packets_err_total;
 538 }
 539
 540 static void
 541 eth_stats_reset(struct rte_eth_dev *dev)
 542 {
 543         unsigned i;
 544         struct pmd_internals *internal = dev->data->dev_private;
 545         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 546                 internal->rx_queue[i].rx_stat.pkts = 0;
 547                 internal->rx_queue[i].rx_stat.bytes = 0;
 548         }
 549         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 550                 internal->tx_queue[i].tx_stat.pkts = 0;
 551                 internal->tx_queue[i].tx_stat.bytes = 0;
 552                 internal->tx_queue[i].tx_stat.err_pkts = 0;
 553         }
 554 }
 555
 556 static void
 557 eth_dev_close(struct rte_eth_dev *dev __rte_unused)
 558 {
 559 }
 560
 561 static void
 562 eth_queue_release(void *q __rte_unused)
 563 {
 564 }
 565
 566 static int
 567 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 568                 int wait_to_complete __rte_unused)
 569 {
 570         return 0;
 571 }
 572
 573 static int
 574 eth_rx_queue_setup(struct rte_eth_dev *dev,
 575                 uint16_t rx_queue_id,
 576                 uint16_t nb_rx_desc __rte_unused,
 577                 unsigned int socket_id __rte_unused,
 578                 const struct rte_eth_rxconf *rx_conf __rte_unused,
 579                 struct rte_mempool *mb_pool)
 580 {
 581         struct pmd_internals *internals = dev->data->dev_private;
 582         struct pcap_rx_queue *pcap_q = &internals->rx_queue[rx_queue_id];
 583         pcap_q->mb_pool = mb_pool;
 584         dev->data->rx_queues[rx_queue_id] = pcap_q;
 585         pcap_q->in_port = dev->data->port_id;
 586         return 0;
 587 }
 588
 589 static int
 590 eth_tx_queue_setup(struct rte_eth_dev *dev,
 591                 uint16_t tx_queue_id,
 592                 uint16_t nb_tx_desc __rte_unused,
 593                 unsigned int socket_id __rte_unused,
 594                 const struct rte_eth_txconf *tx_conf __rte_unused)
 595 {
 596
 597         struct pmd_internals *internals = dev->data->dev_private;
 598         dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
 599         return 0;
 600 }
 601
 602 static const struct eth_dev_ops ops = {
 603         .dev_start = eth_dev_start,
 604         .dev_stop =     eth_dev_stop,
 605         .dev_close = eth_dev_close,
 606         .dev_configure = eth_dev_configure,
 607         .dev_infos_get = eth_dev_info,
 608         .rx_queue_setup = eth_rx_queue_setup,
 609         .tx_queue_setup = eth_tx_queue_setup,
 610         .rx_queue_release = eth_queue_release,
 611         .tx_queue_release = eth_queue_release,
 612         .link_update = eth_link_update,
 613         .stats_get = eth_stats_get,
 614         .stats_reset = eth_stats_reset,
 615 };
 616
 617 /*
 618  * Function handler that opens the pcap file for reading a stores a
 619  * reference of it for use it later on.
 620  */
 621 static int
 622 open_rx_pcap(const char *key, const char *value, void *extra_args)
 623 {
 624         unsigned i;
 625         const char *pcap_filename = value;
 626         struct pmd_devargs *rx = extra_args;
 627         pcap_t *pcap = NULL;
 628
 629         for (i = 0; i < rx->num_of_queue; i++) {
 630                 if (open_single_rx_pcap(pcap_filename, &pcap) < 0)
 631                         return -1;
 632
 633                 rx->queue[i].pcap = pcap;
 634                 rx->queue[i].name = pcap_filename;
 635                 rx->queue[i].type = key;
 636         }
 637
 638         return 0;
 639 }
 640
 641 static int
 642 open_single_rx_pcap(const char *pcap_filename, pcap_t **pcap)
 643 {
 644         if ((*pcap = pcap_open_offline(pcap_filename, errbuf)) == NULL) {
 645                 RTE_LOG(ERR, PMD, "Couldn't open %s: %s\n", pcap_filename, errbuf);
 646                 return -1;
 647         }
 648         return 0;
 649 }
 650
 651 /*
 652  * Opens a pcap file for writing and stores a reference to it
 653  * for use it later on.
 654  */
 655 static int
 656 open_tx_pcap(const char *key, const char *value, void *extra_args)
 657 {
 658         unsigned i;
 659         const char *pcap_filename = value;
 660         struct pmd_devargs *dumpers = extra_args;
 661         pcap_dumper_t *dumper;
 662
 663         for (i = 0; i < dumpers->num_of_queue; i++) {
 664                 if (open_single_tx_pcap(pcap_filename, &dumper) < 0)
 665                         return -1;
 666
 667                 dumpers->queue[i].dumper = dumper;
 668                 dumpers->queue[i].name = pcap_filename;
 669                 dumpers->queue[i].type = key;
 670         }
 671
 672         return 0;
 673 }
 674
 675 static int
 676 open_single_tx_pcap(const char *pcap_filename, pcap_dumper_t **dumper)
 677 {
 678         pcap_t *tx_pcap;
 679         /*
 680          * We need to create a dummy empty pcap_t to use it
 681          * with pcap_dump_open(). We create big enough an Ethernet
 682          * pcap holder.
 683          */
 684
 685         if ((tx_pcap = pcap_open_dead(DLT_EN10MB, RTE_ETH_PCAP_SNAPSHOT_LEN))
 686                         == NULL) {
 687                 RTE_LOG(ERR, PMD, "Couldn't create dead pcap\n");
 688                 return -1;
 689         }
 690
 691         /* The dumper is created using the previous pcap_t reference */
 692         if ((*dumper = pcap_dump_open(tx_pcap, pcap_filename)) == NULL) {
 693                 RTE_LOG(ERR, PMD, "Couldn't open %s for writing.\n", pcap_filename);
 694                 return -1;
 695         }
 696
 697         return 0;
 698 }
 699
 700 /*
 701  * pcap_open_live wrapper function
 702  */
 703 static inline int
 704 open_iface_live(const char *iface, pcap_t **pcap) {
 705         *pcap = pcap_open_live(iface, RTE_ETH_PCAP_SNAPLEN,
 706                         RTE_ETH_PCAP_PROMISC, RTE_ETH_PCAP_TIMEOUT, errbuf);
 707
 708         if (*pcap == NULL) {
 709                 RTE_LOG(ERR, PMD, "Couldn't open %s: %s\n", iface, errbuf);
 710                 return -1;
 711         }
 712         return 0;
 713 }
 714
 715 /*
 716  * Opens an interface for reading and writing
 717  */
 718 static inline int
 719 open_rx_tx_iface(const char *key, const char *value, void *extra_args)
 720 {
 721         const char *iface = value;
 722         struct pmd_devargs *tx = extra_args;
 723         pcap_t *pcap = NULL;
 724
 725         if (open_single_iface(iface, &pcap) < 0)
 726                 return -1;
 727
 728         tx->queue[0].pcap = pcap;
 729         tx->queue[0].name = iface;
 730         tx->queue[0].type = key;
 731
 732         return 0;
 733 }
 734
 735 /*
 736  * Opens a NIC for reading packets from it
 737  */
 738 static inline int
 739 open_rx_iface(const char *key, const char *value, void *extra_args)
 740 {
 741         unsigned i;
 742         const char *iface = value;
 743         struct pmd_devargs *rx = extra_args;
 744         pcap_t *pcap = NULL;
 745
 746         for (i = 0; i < rx->num_of_queue; i++) {
 747                 if (open_single_iface(iface, &pcap) < 0)
 748                         return -1;
 749                 rx->queue[i].pcap = pcap;
 750                 rx->queue[i].name = iface;
 751                 rx->queue[i].type = key;
 752         }
 753
 754         return 0;
 755 }
 756
 757 /*
 758  * Opens a NIC for writing packets to it
 759  */
 760 static int
 761 open_tx_iface(const char *key, const char *value, void *extra_args)
 762 {
 763         unsigned i;
 764         const char *iface = value;
 765         struct pmd_devargs *tx = extra_args;
 766         pcap_t *pcap;
 767
 768         for (i = 0; i < tx->num_of_queue; i++) {
 769                 if (open_single_iface(iface, &pcap) < 0)
 770                         return -1;
 771                 tx->queue[i].pcap = pcap;
 772                 tx->queue[i].name = iface;
 773                 tx->queue[i].type = key;
 774         }
 775
 776         return 0;
 777 }
 778
 779 static int
 780 open_single_iface(const char *iface, pcap_t **pcap)
 781 {
 782         if (open_iface_live(iface, pcap) < 0) {
 783                 RTE_LOG(ERR, PMD, "Couldn't open interface %s\n", iface);
 784                 return -1;
 785         }
 786
 787         return 0;
 788 }
 789
 790 static int
 791 rte_pmd_init_internals(const char *name, const unsigned nb_rx_queues,
 792                 const unsigned nb_tx_queues, struct pmd_internals **internals,
 793                 struct rte_eth_dev **eth_dev)
 794 {
 795         struct rte_eth_dev_data *data = NULL;
 796         unsigned int numa_node = rte_socket_id();
 797
 798         RTE_LOG(INFO, PMD, "Creating pcap-backed ethdev on numa socket %u\n",
 799                 numa_node);
 800
 801         /* now do all data allocation - for eth_dev structure
 802          * and internal (private) data
 803          */
 804         data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
 805         if (data == NULL)
 806                 goto error;
 807
 808         *internals = rte_zmalloc_socket(name, sizeof(**internals), 0,
 809                         numa_node);
 810         if (*internals == NULL)
 811                 goto error;
 812
 813         /* reserve an ethdev entry */
 814         *eth_dev = rte_eth_dev_allocate(name);
 815         if (*eth_dev == NULL)
 816                 goto error;
 817
 818         /* check length of device name */
 819         if ((strlen((*eth_dev)->data->name) + 1) > sizeof(data->name))
 820                 goto error;
 821
 822         /* now put it all together
 823          * - store queue data in internals,
 824          * - store numa_node info in eth_dev
 825          * - point eth_dev_data to internals
 826          * - and point eth_dev structure to new eth_dev_data structure
 827          */
 828         data->dev_private = *internals;
 829         data->port_id = (*eth_dev)->data->port_id;
 830         snprintf(data->name, sizeof(data->name), "%s", (*eth_dev)->data->name);
 831         data->nb_rx_queues = (uint16_t)nb_rx_queues;
 832         data->nb_tx_queues = (uint16_t)nb_tx_queues;
 833         data->dev_link = pmd_link;
 834         data->mac_addrs = &eth_addr;
 835         strncpy(data->name,
 836                 (*eth_dev)->data->name, strlen((*eth_dev)->data->name));
 837
 838         /*
 839          * NOTE: we'll replace the data element, of originally allocated
 840          * eth_dev so the rings are local per-process
 841          */
 842         (*eth_dev)->data = data;
 843         (*eth_dev)->dev_ops = &ops;
 844         (*eth_dev)->driver = NULL;
 845         data->dev_flags = RTE_ETH_DEV_DETACHABLE;
 846         data->kdrv = RTE_KDRV_NONE;
 847         data->drv_name = drivername;
 848         data->numa_node = numa_node;
 849
 850         return 0;
 851
 852 error:
 853         rte_free(data);
 854         rte_free(*internals);
 855
 856         return -1;
 857 }
 858
 859 static int
 860 rte_eth_from_pcaps_common(const char *name, struct pmd_devargs *rx_queues,
 861                 const unsigned nb_rx_queues, struct pmd_devargs *tx_queues,
 862                 const unsigned nb_tx_queues, struct rte_kvargs *kvlist,
 863                 struct pmd_internals **internals, struct rte_eth_dev **eth_dev)
 864 {
 865         struct rte_kvargs_pair *pair = NULL;
 866         unsigned k_idx;
 867         unsigned i;
 868
 869         /* do some parameter checking */
 870         if (rx_queues == NULL && nb_rx_queues > 0)
 871                 return -1;
 872         if (tx_queues == NULL && nb_tx_queues > 0)
 873                 return -1;
 874
 875         if (rte_pmd_init_internals(name, nb_rx_queues, nb_tx_queues, internals,
 876                         eth_dev) < 0)
 877                 return -1;
 878
 879         for (i = 0; i < nb_rx_queues; i++) {
 880                 (*internals)->rx_queue[i].pcap = rx_queues->queue[i].pcap;
 881                 snprintf((*internals)->rx_queue[i].name,
 882                         sizeof((*internals)->rx_queue[i].name), "%s",
 883                         rx_queues->queue[i].name);
 884                 snprintf((*internals)->rx_queue[i].type,
 885                         sizeof((*internals)->rx_queue[i].type), "%s",
 886                         rx_queues->queue[i].type);
 887         }
 888         for (i = 0; i < nb_tx_queues; i++) {
 889                 (*internals)->tx_queue[i].dumper = tx_queues->queue[i].dumper;
 890                 snprintf((*internals)->tx_queue[i].name,
 891                         sizeof((*internals)->tx_queue[i].name), "%s",
 892                         tx_queues->queue[i].name);
 893                 snprintf((*internals)->tx_queue[i].type,
 894                         sizeof((*internals)->tx_queue[i].type), "%s",
 895                         tx_queues->queue[i].type);
 896         }
 897
 898         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 899                 pair = &kvlist->pairs[k_idx];
 900                 if (strstr(pair->key, ETH_PCAP_IFACE_ARG) != NULL)
 901                         break;
 902         }
 903
 904         if (pair == NULL)
 905                 (*internals)->if_index = 0;
 906         else
 907                 (*internals)->if_index = if_nametoindex(pair->value);
 908
 909         return 0;
 910 }
 911
 912 static int
 913 rte_eth_from_pcaps(const char *name, struct pmd_devargs *rx_queues,
 914                 const unsigned nb_rx_queues, struct pmd_devargs *tx_queues,
 915                 const unsigned nb_tx_queues, struct rte_kvargs *kvlist,
 916                 int single_iface, unsigned int using_dumpers)
 917 {
 918         struct pmd_internals *internals = NULL;
 919         struct rte_eth_dev *eth_dev = NULL;
 920         int ret;
 921
 922         ret = rte_eth_from_pcaps_common(name, rx_queues, nb_rx_queues,
 923                 tx_queues, nb_tx_queues, kvlist, &internals, &eth_dev);
 924
 925         if (ret < 0)
 926                 return ret;
 927
 928         /* store wether we are using a single interface for rx/tx or not */
 929         internals->single_iface = single_iface;
 930
 931         eth_dev->rx_pkt_burst = eth_pcap_rx;
 932
 933         if (using_dumpers)
 934                 eth_dev->tx_pkt_burst = eth_pcap_tx_dumper;
 935         else
 936                 eth_dev->tx_pkt_burst = eth_pcap_tx;
 937
 938         return 0;
 939 }
 940
 941
 942 static int
 943 rte_pmd_pcap_devinit(const char *name, const char *params)
 944 {
 945         unsigned int is_rx_pcap = 0, is_tx_pcap = 0;
 946         struct rte_kvargs *kvlist;
 947         struct pmd_devargs pcaps = {0};
 948         struct pmd_devargs dumpers = {0};
 949         int single_iface = 0;
 950         int ret;
 951
 952         RTE_LOG(INFO, PMD, "Initializing pmd_pcap for %s\n", name);
 953
 954         gettimeofday(&start_time, NULL);
 955         start_cycles = rte_get_timer_cycles();
 956         hz = rte_get_timer_hz();
 957
 958         kvlist = rte_kvargs_parse(params, valid_arguments);
 959         if (kvlist == NULL)
 960                 return -1;
 961
 962         /*
 963          * If iface argument is passed we open the NICs and use them for
 964          * reading / writing
 965          */
 966         if (rte_kvargs_count(kvlist, ETH_PCAP_IFACE_ARG) == 1) {
 967
 968                 ret = rte_kvargs_process(kvlist, ETH_PCAP_IFACE_ARG,
 969                                 &open_rx_tx_iface, &pcaps);
 970
 971                 if (ret < 0)
 972                         goto free_kvlist;
 973                 dumpers.queue[0].pcap = pcaps.queue[0].pcap;
 974                 dumpers.queue[0].name = pcaps.queue[0].name;
 975                 dumpers.queue[0].type = pcaps.queue[0].type;
 976
 977                 single_iface = 1;
 978
 979                 ret = rte_eth_from_pcaps(name, &pcaps, 1, &dumpers, 1,
 980                         kvlist, single_iface, is_tx_pcap);
 981
 982                 goto free_kvlist;
 983         }
 984
 985         /*
 986          * We check whether we want to open a RX stream from a real NIC or a
 987          * pcap file
 988          */
 989         pcaps.num_of_queue = rte_kvargs_count(kvlist, ETH_PCAP_RX_PCAP_ARG);
 990         if (pcaps.num_of_queue)
 991                 is_rx_pcap = 1;
 992         else
 993                 pcaps.num_of_queue = rte_kvargs_count(kvlist,
 994                                 ETH_PCAP_RX_IFACE_ARG);
 995
 996         if (pcaps.num_of_queue > RTE_PMD_PCAP_MAX_QUEUES)
 997                 pcaps.num_of_queue = RTE_PMD_PCAP_MAX_QUEUES;
 998
 999         if (is_rx_pcap)
1000                 ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_PCAP_ARG,
1001                                 &open_rx_pcap, &pcaps);
1002         else
1003                 ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_IFACE_ARG,
1004                                 &open_rx_iface, &pcaps);
1005
1006         if (ret < 0)
1007                 goto free_kvlist;
1008
1009         /*
1010          * We check whether we want to open a TX stream to a real NIC or a
1011          * pcap file
1012          */
1013         dumpers.num_of_queue = rte_kvargs_count(kvlist, ETH_PCAP_TX_PCAP_ARG);
1014         if (dumpers.num_of_queue)
1015                 is_tx_pcap = 1;
1016         else
1017                 dumpers.num_of_queue = rte_kvargs_count(kvlist,
1018                                 ETH_PCAP_TX_IFACE_ARG);
1019
1020         if (dumpers.num_of_queue > RTE_PMD_PCAP_MAX_QUEUES)
1021                 dumpers.num_of_queue = RTE_PMD_PCAP_MAX_QUEUES;
1022
1023         if (is_tx_pcap)
1024                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_PCAP_ARG,
1025                                 &open_tx_pcap, &dumpers);
1026         else
1027                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_IFACE_ARG,
1028                                 &open_tx_iface, &dumpers);
1029
1030         if (ret < 0)
1031                 goto free_kvlist;
1032
1033         ret = rte_eth_from_pcaps(name, &pcaps, pcaps.num_of_queue, &dumpers,
1034                 dumpers.num_of_queue, kvlist, single_iface, is_tx_pcap);
1035
1036 free_kvlist:
1037         rte_kvargs_free(kvlist);
1038         return ret;
1039 }
1040
1041 static int
1042 rte_pmd_pcap_devuninit(const char *name)
1043 {
1044         struct rte_eth_dev *eth_dev = NULL;
1045
1046         RTE_LOG(INFO, PMD, "Closing pcap ethdev on numa socket %u\n",
1047                         rte_socket_id());
1048
1049         if (name == NULL)
1050                 return -1;
1051
1052         /* reserve an ethdev entry */
1053         eth_dev = rte_eth_dev_allocated(name);
1054         if (eth_dev == NULL)
1055                 return -1;
1056
1057         rte_free(eth_dev->data->dev_private);
1058         rte_free(eth_dev->data);
1059
1060         rte_eth_dev_release_port(eth_dev);
1061
1062         return 0;
1063 }
1064
1065 static struct rte_vdev_driver pmd_pcap_drv = {
1066         .init = rte_pmd_pcap_devinit,
1067         .uninit = rte_pmd_pcap_devuninit,
1068 };
1069
1070 DRIVER_REGISTER_VDEV(net_pcap, pmd_pcap_drv);
1071 DRIVER_REGISTER_PARAM_STRING(net_pcap,
1072         ETH_PCAP_RX_PCAP_ARG "=<string> "
1073         ETH_PCAP_TX_PCAP_ARG "=<string> "
1074         ETH_PCAP_RX_IFACE_ARG "=<ifc> "
1075         ETH_PCAP_TX_IFACE_ARG "=<ifc> "
1076         ETH_PCAP_IFACE_ARG "=<ifc>");