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 #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_vdev.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 #define RTE_PMD_PCAP_MAX_QUEUES 16
  62
  63 static char errbuf[PCAP_ERRBUF_SIZE];
  64 static unsigned char tx_pcap_data[RTE_ETH_PCAP_SNAPLEN];
  65 static struct timeval start_time;
  66 static uint64_t start_cycles;
  67 static uint64_t hz;
  68
  69 struct queue_stat {
  70         volatile unsigned long pkts;
  71         volatile unsigned long bytes;
  72         volatile unsigned long err_pkts;
  73 };
  74
  75 struct pcap_rx_queue {
  76         pcap_t *pcap;
  77         uint8_t in_port;
  78         struct rte_mempool *mb_pool;
  79         struct queue_stat rx_stat;
  80         char name[PATH_MAX];
  81         char type[ETH_PCAP_ARG_MAXLEN];
  82 };
  83
  84 struct pcap_tx_queue {
  85         pcap_dumper_t *dumper;
  86         pcap_t *pcap;
  87         struct queue_stat tx_stat;
  88         char name[PATH_MAX];
  89         char type[ETH_PCAP_ARG_MAXLEN];
  90 };
  91
  92 struct pmd_internals {
  93         struct pcap_rx_queue rx_queue[RTE_PMD_PCAP_MAX_QUEUES];
  94         struct pcap_tx_queue tx_queue[RTE_PMD_PCAP_MAX_QUEUES];
  95         int if_index;
  96         int single_iface;
  97 };
  98
  99 struct pmd_devargs {
 100         unsigned num_of_queue;
 101         struct devargs_queue {
 102                 pcap_dumper_t *dumper;
 103                 pcap_t *pcap;
 104                 const char *name;
 105                 const char *type;
 106         } queue[RTE_PMD_PCAP_MAX_QUEUES];
 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 = ETH_SPEED_NUM_10G,
 126                 .link_duplex = ETH_LINK_FULL_DUPLEX,
 127                 .link_status = ETH_LINK_DOWN,
 128                 .link_autoneg = ETH_LINK_SPEED_FIXED,
 129 };
 130
 131 static int
 132 eth_pcap_rx_jumbo(struct rte_mempool *mb_pool,
 133                   struct rte_mbuf *mbuf,
 134                   const u_char *data,
 135                   uint16_t data_len)
 136 {
 137         struct rte_mbuf *m = mbuf;
 138
 139         /* Copy the first segment. */
 140         uint16_t len = rte_pktmbuf_tailroom(mbuf);
 141
 142         rte_memcpy(rte_pktmbuf_append(mbuf, len), data, len);
 143         data_len -= len;
 144         data += len;
 145
 146         while (data_len > 0) {
 147                 /* Allocate next mbuf and point to that. */
 148                 m->next = rte_pktmbuf_alloc(mb_pool);
 149
 150                 if (unlikely(!m->next))
 151                         return -1;
 152
 153                 m = m->next;
 154
 155                 /* Headroom is not needed in chained mbufs. */
 156                 rte_pktmbuf_prepend(m, rte_pktmbuf_headroom(m));
 157                 m->pkt_len = 0;
 158                 m->data_len = 0;
 159
 160                 /* Copy next segment. */
 161                 len = RTE_MIN(rte_pktmbuf_tailroom(m), data_len);
 162                 rte_memcpy(rte_pktmbuf_append(m, len), data, len);
 163
 164                 mbuf->nb_segs++;
 165                 data_len -= len;
 166                 data += len;
 167         }
 168
 169         return mbuf->nb_segs;
 170 }
 171
 172 /* Copy data from mbuf chain to a buffer suitable for writing to a PCAP file. */
 173 static void
 174 eth_pcap_gather_data(unsigned char *data, struct rte_mbuf *mbuf)
 175 {
 176         uint16_t data_len = 0;
 177
 178         while (mbuf) {
 179                 rte_memcpy(data + data_len, rte_pktmbuf_mtod(mbuf, void *),
 180                            mbuf->data_len);
 181
 182                 data_len += mbuf->data_len;
 183                 mbuf = mbuf->next;
 184         }
 185 }
 186
 187 static uint16_t
 188 eth_pcap_rx(void *queue,
 189                 struct rte_mbuf **bufs,
 190                 uint16_t nb_pkts)
 191 {
 192         unsigned i;
 193         struct pcap_pkthdr header;
 194         const u_char *packet;
 195         struct rte_mbuf *mbuf;
 196         struct pcap_rx_queue *pcap_q = queue;
 197         uint16_t num_rx = 0;
 198         uint16_t buf_size;
 199         uint32_t rx_bytes = 0;
 200
 201         if (unlikely(pcap_q->pcap == NULL || nb_pkts == 0))
 202                 return 0;
 203
 204         /* Reads the given number of packets from the pcap file one by one
 205          * and copies the packet data into a newly allocated mbuf to return.
 206          */
 207         for (i = 0; i < nb_pkts; i++) {
 208                 /* Get the next PCAP packet */
 209                 packet = pcap_next(pcap_q->pcap, &header);
 210                 if (unlikely(packet == NULL))
 211                         break;
 212                 else
 213                         mbuf = rte_pktmbuf_alloc(pcap_q->mb_pool);
 214                 if (unlikely(mbuf == NULL))
 215                         break;
 216
 217                 /* Now get the space available for data in the mbuf */
 218                 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(pcap_q->mb_pool) -
 219                                 RTE_PKTMBUF_HEADROOM);
 220
 221                 if (header.caplen <= buf_size) {
 222                         /* pcap packet will fit in the mbuf, go ahead and copy */
 223                         rte_memcpy(rte_pktmbuf_mtod(mbuf, void *), packet,
 224                                         header.caplen);
 225                         mbuf->data_len = (uint16_t)header.caplen;
 226                 } else {
 227                         /* Try read jumbo frame into multi mbufs. */
 228                         if (unlikely(eth_pcap_rx_jumbo(pcap_q->mb_pool,
 229                                                        mbuf,
 230                                                        packet,
 231                                                        header.caplen) == -1))
 232                                 break;
 233                 }
 234
 235                 mbuf->pkt_len = (uint16_t)header.caplen;
 236                 mbuf->port = pcap_q->in_port;
 237                 bufs[num_rx] = mbuf;
 238                 num_rx++;
 239                 rx_bytes += header.caplen;
 240         }
 241         pcap_q->rx_stat.pkts += num_rx;
 242         pcap_q->rx_stat.bytes += rx_bytes;
 243         return num_rx;
 244 }
 245
 246 static inline void
 247 calculate_timestamp(struct timeval *ts) {
 248         uint64_t cycles;
 249         struct timeval cur_time;
 250
 251         cycles = rte_get_timer_cycles() - start_cycles;
 252         cur_time.tv_sec = cycles / hz;
 253         cur_time.tv_usec = (cycles % hz) * 10e6 / hz;
 254         timeradd(&start_time, &cur_time, ts);
 255 }
 256
 257 /*
 258  * Callback to handle writing packets to a pcap file.
 259  */
 260 static uint16_t
 261 eth_pcap_tx_dumper(void *queue,
 262                 struct rte_mbuf **bufs,
 263                 uint16_t nb_pkts)
 264 {
 265         unsigned i;
 266         struct rte_mbuf *mbuf;
 267         struct pcap_tx_queue *dumper_q = queue;
 268         uint16_t num_tx = 0;
 269         uint32_t tx_bytes = 0;
 270         struct pcap_pkthdr header;
 271
 272         if (dumper_q->dumper == NULL || nb_pkts == 0)
 273                 return 0;
 274
 275         /* writes the nb_pkts packets to the previously opened pcap file dumper */
 276         for (i = 0; i < nb_pkts; i++) {
 277                 mbuf = bufs[i];
 278                 calculate_timestamp(&header.ts);
 279                 header.len = mbuf->pkt_len;
 280                 header.caplen = header.len;
 281
 282                 if (likely(mbuf->nb_segs == 1)) {
 283                         pcap_dump((u_char *)dumper_q->dumper, &header,
 284                                   rte_pktmbuf_mtod(mbuf, void*));
 285                 } else {
 286                         if (mbuf->pkt_len <= ETHER_MAX_JUMBO_FRAME_LEN) {
 287                                 eth_pcap_gather_data(tx_pcap_data, mbuf);
 288                                 pcap_dump((u_char *)dumper_q->dumper, &header,
 289                                           tx_pcap_data);
 290                         } else {
 291                                 RTE_LOG(ERR, PMD,
 292                                         "Dropping PCAP packet. "
 293                                         "Size (%d) > max jumbo size (%d).\n",
 294                                         mbuf->pkt_len,
 295                                         ETHER_MAX_JUMBO_FRAME_LEN);
 296
 297                                 rte_pktmbuf_free(mbuf);
 298                                 break;
 299                         }
 300                 }
 301
 302                 rte_pktmbuf_free(mbuf);
 303                 num_tx++;
 304                 tx_bytes += mbuf->pkt_len;
 305         }
 306
 307         /*
 308          * Since there's no place to hook a callback when the forwarding
 309          * process stops and to make sure the pcap file is actually written,
 310          * we flush the pcap dumper within each burst.
 311          */
 312         pcap_dump_flush(dumper_q->dumper);
 313         dumper_q->tx_stat.pkts += num_tx;
 314         dumper_q->tx_stat.bytes += tx_bytes;
 315         dumper_q->tx_stat.err_pkts += nb_pkts - num_tx;
 316         return num_tx;
 317 }
 318
 319 /*
 320  * Callback to handle sending packets through a real NIC.
 321  */
 322 static uint16_t
 323 eth_pcap_tx(void *queue,
 324                 struct rte_mbuf **bufs,
 325                 uint16_t nb_pkts)
 326 {
 327         unsigned i;
 328         int ret;
 329         struct rte_mbuf *mbuf;
 330         struct pcap_tx_queue *tx_queue = queue;
 331         uint16_t num_tx = 0;
 332         uint32_t tx_bytes = 0;
 333
 334         if (unlikely(nb_pkts == 0 || tx_queue->pcap == NULL))
 335                 return 0;
 336
 337         for (i = 0; i < nb_pkts; i++) {
 338                 mbuf = bufs[i];
 339
 340                 if (likely(mbuf->nb_segs == 1)) {
 341                         ret = pcap_sendpacket(tx_queue->pcap,
 342                                               rte_pktmbuf_mtod(mbuf, u_char *),
 343                                               mbuf->pkt_len);
 344                 } else {
 345                         if (mbuf->pkt_len <= ETHER_MAX_JUMBO_FRAME_LEN) {
 346                                 eth_pcap_gather_data(tx_pcap_data, mbuf);
 347                                 ret = pcap_sendpacket(tx_queue->pcap,
 348                                                       tx_pcap_data,
 349                                                       mbuf->pkt_len);
 350                         } else {
 351                                 RTE_LOG(ERR, PMD,
 352                                         "Dropping PCAP packet. "
 353                                         "Size (%d) > max jumbo size (%d).\n",
 354                                         mbuf->pkt_len,
 355                                         ETHER_MAX_JUMBO_FRAME_LEN);
 356
 357                                 rte_pktmbuf_free(mbuf);
 358                                 break;
 359                         }
 360                 }
 361
 362                 if (unlikely(ret != 0))
 363                         break;
 364                 num_tx++;
 365                 tx_bytes += mbuf->pkt_len;
 366                 rte_pktmbuf_free(mbuf);
 367         }
 368
 369         tx_queue->tx_stat.pkts += num_tx;
 370         tx_queue->tx_stat.bytes += tx_bytes;
 371         tx_queue->tx_stat.err_pkts += nb_pkts - num_tx;
 372         return num_tx;
 373 }
 374
 375 static int
 376 eth_dev_start(struct rte_eth_dev *dev)
 377 {
 378         unsigned i;
 379         struct pmd_internals *internals = dev->data->dev_private;
 380         struct pcap_tx_queue *tx;
 381         struct pcap_rx_queue *rx;
 382
 383         /* Special iface case. Single pcap is open and shared between tx/rx. */
 384         if (internals->single_iface) {
 385                 tx = &internals->tx_queue[0];
 386                 rx = &internals->rx_queue[0];
 387
 388                 if (!tx->pcap && strcmp(tx->type, ETH_PCAP_IFACE_ARG) == 0) {
 389                         if (open_single_iface(tx->name, &tx->pcap) < 0)
 390                                 return -1;
 391                         rx->pcap = tx->pcap;
 392                 }
 393                 goto status_up;
 394         }
 395
 396         /* If not open already, open tx pcaps/dumpers */
 397         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 398                 tx = &internals->tx_queue[i];
 399
 400                 if (!tx->dumper && strcmp(tx->type, ETH_PCAP_TX_PCAP_ARG) == 0) {
 401                         if (open_single_tx_pcap(tx->name, &tx->dumper) < 0)
 402                                 return -1;
 403                 }
 404
 405                 else if (!tx->pcap && strcmp(tx->type, ETH_PCAP_TX_IFACE_ARG) == 0) {
 406                         if (open_single_iface(tx->name, &tx->pcap) < 0)
 407                                 return -1;
 408                 }
 409         }
 410
 411         /* If not open already, open rx pcaps */
 412         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 413                 rx = &internals->rx_queue[i];
 414
 415                 if (rx->pcap != NULL)
 416                         continue;
 417
 418                 if (strcmp(rx->type, ETH_PCAP_RX_PCAP_ARG) == 0) {
 419                         if (open_single_rx_pcap(rx->name, &rx->pcap) < 0)
 420                                 return -1;
 421                 }
 422
 423                 else if (strcmp(rx->type, ETH_PCAP_RX_IFACE_ARG) == 0) {
 424                         if (open_single_iface(rx->name, &rx->pcap) < 0)
 425                                 return -1;
 426                 }
 427         }
 428
 429 status_up:
 430
 431         dev->data->dev_link.link_status = ETH_LINK_UP;
 432         return 0;
 433 }
 434
 435 /*
 436  * This function gets called when the current port gets stopped.
 437  * Is the only place for us to close all the tx streams dumpers.
 438  * If not called the dumpers will be flushed within each tx burst.
 439  */
 440 static void
 441 eth_dev_stop(struct rte_eth_dev *dev)
 442 {
 443         unsigned i;
 444         struct pmd_internals *internals = dev->data->dev_private;
 445         struct pcap_tx_queue *tx;
 446         struct pcap_rx_queue *rx;
 447
 448         /* Special iface case. Single pcap is open and shared between tx/rx. */
 449         if (internals->single_iface) {
 450                 tx = &internals->tx_queue[0];
 451                 rx = &internals->rx_queue[0];
 452                 pcap_close(tx->pcap);
 453                 tx->pcap = NULL;
 454                 rx->pcap = NULL;
 455                 goto status_down;
 456         }
 457
 458         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 459                 tx = &internals->tx_queue[i];
 460
 461                 if (tx->dumper != NULL) {
 462                         pcap_dump_close(tx->dumper);
 463                         tx->dumper = NULL;
 464                 }
 465
 466                 if (tx->pcap != NULL) {
 467                         pcap_close(tx->pcap);
 468                         tx->pcap = NULL;
 469                 }
 470         }
 471
 472         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 473                 rx = &internals->rx_queue[i];
 474
 475                 if (rx->pcap != NULL) {
 476                         pcap_close(rx->pcap);
 477                         rx->pcap = NULL;
 478                 }
 479         }
 480
 481 status_down:
 482         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 483 }
 484
 485 static int
 486 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 487 {
 488         return 0;
 489 }
 490
 491 static void
 492 eth_dev_info(struct rte_eth_dev *dev,
 493                 struct rte_eth_dev_info *dev_info)
 494 {
 495         struct pmd_internals *internals = dev->data->dev_private;
 496         dev_info->driver_name = drivername;
 497         dev_info->if_index = internals->if_index;
 498         dev_info->max_mac_addrs = 1;
 499         dev_info->max_rx_pktlen = (uint32_t) -1;
 500         dev_info->max_rx_queues = dev->data->nb_rx_queues;
 501         dev_info->max_tx_queues = dev->data->nb_tx_queues;
 502         dev_info->min_rx_bufsize = 0;
 503         dev_info->pci_dev = NULL;
 504 }
 505
 506 static void
 507 eth_stats_get(struct rte_eth_dev *dev,
 508                 struct rte_eth_stats *stats)
 509 {
 510         unsigned i;
 511         unsigned long rx_packets_total = 0, rx_bytes_total = 0;
 512         unsigned long tx_packets_total = 0, tx_bytes_total = 0;
 513         unsigned long tx_packets_err_total = 0;
 514         const struct pmd_internals *internal = dev->data->dev_private;
 515
 516         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 517                         i < dev->data->nb_rx_queues; i++) {
 518                 stats->q_ipackets[i] = internal->rx_queue[i].rx_stat.pkts;
 519                 stats->q_ibytes[i] = internal->rx_queue[i].rx_stat.bytes;
 520                 rx_packets_total += stats->q_ipackets[i];
 521                 rx_bytes_total += stats->q_ibytes[i];
 522         }
 523
 524         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 525                         i < dev->data->nb_tx_queues; i++) {
 526                 stats->q_opackets[i] = internal->tx_queue[i].tx_stat.pkts;
 527                 stats->q_obytes[i] = internal->tx_queue[i].tx_stat.bytes;
 528                 stats->q_errors[i] = internal->tx_queue[i].tx_stat.err_pkts;
 529                 tx_packets_total += stats->q_opackets[i];
 530                 tx_bytes_total += stats->q_obytes[i];
 531                 tx_packets_err_total += stats->q_errors[i];
 532         }
 533
 534         stats->ipackets = rx_packets_total;
 535         stats->ibytes = rx_bytes_total;
 536         stats->opackets = tx_packets_total;
 537         stats->obytes = tx_bytes_total;
 538         stats->oerrors = tx_packets_err_total;
 539 }
 540
 541 static void
 542 eth_stats_reset(struct rte_eth_dev *dev)
 543 {
 544         unsigned i;
 545         struct pmd_internals *internal = dev->data->dev_private;
 546         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 547                 internal->rx_queue[i].rx_stat.pkts = 0;
 548                 internal->rx_queue[i].rx_stat.bytes = 0;
 549         }
 550         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 551                 internal->tx_queue[i].tx_stat.pkts = 0;
 552                 internal->tx_queue[i].tx_stat.bytes = 0;
 553                 internal->tx_queue[i].tx_stat.err_pkts = 0;
 554         }
 555 }
 556
 557 static void
 558 eth_dev_close(struct rte_eth_dev *dev __rte_unused)
 559 {
 560 }
 561
 562 static void
 563 eth_queue_release(void *q __rte_unused)
 564 {
 565 }
 566
 567 static int
 568 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 569                 int wait_to_complete __rte_unused)
 570 {
 571         return 0;
 572 }
 573
 574 static int
 575 eth_rx_queue_setup(struct rte_eth_dev *dev,
 576                 uint16_t rx_queue_id,
 577                 uint16_t nb_rx_desc __rte_unused,
 578                 unsigned int socket_id __rte_unused,
 579                 const struct rte_eth_rxconf *rx_conf __rte_unused,
 580                 struct rte_mempool *mb_pool)
 581 {
 582         struct pmd_internals *internals = dev->data->dev_private;
 583         struct pcap_rx_queue *pcap_q = &internals->rx_queue[rx_queue_id];
 584         pcap_q->mb_pool = mb_pool;
 585         dev->data->rx_queues[rx_queue_id] = pcap_q;
 586         pcap_q->in_port = dev->data->port_id;
 587         return 0;
 588 }
 589
 590 static int
 591 eth_tx_queue_setup(struct rte_eth_dev *dev,
 592                 uint16_t tx_queue_id,
 593                 uint16_t nb_tx_desc __rte_unused,
 594                 unsigned int socket_id __rte_unused,
 595                 const struct rte_eth_txconf *tx_conf __rte_unused)
 596 {
 597
 598         struct pmd_internals *internals = dev->data->dev_private;
 599         dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
 600         return 0;
 601 }
 602
 603 static const struct eth_dev_ops ops = {
 604         .dev_start = eth_dev_start,
 605         .dev_stop =     eth_dev_stop,
 606         .dev_close = eth_dev_close,
 607         .dev_configure = eth_dev_configure,
 608         .dev_infos_get = eth_dev_info,
 609         .rx_queue_setup = eth_rx_queue_setup,
 610         .tx_queue_setup = eth_tx_queue_setup,
 611         .rx_queue_release = eth_queue_release,
 612         .tx_queue_release = eth_queue_release,
 613         .link_update = eth_link_update,
 614         .stats_get = eth_stats_get,
 615         .stats_reset = eth_stats_reset,
 616 };
 617
 618 /*
 619  * Function handler that opens the pcap file for reading a stores a
 620  * reference of it for use it later on.
 621  */
 622 static int
 623 open_rx_pcap(const char *key, const char *value, void *extra_args)
 624 {
 625         unsigned i;
 626         const char *pcap_filename = value;
 627         struct pmd_devargs *rx = extra_args;
 628         pcap_t *pcap = NULL;
 629
 630         for (i = 0; i < rx->num_of_queue; i++) {
 631                 if (open_single_rx_pcap(pcap_filename, &pcap) < 0)
 632                         return -1;
 633
 634                 rx->queue[i].pcap = pcap;
 635                 rx->queue[i].name = pcap_filename;
 636                 rx->queue[i].type = key;
 637         }
 638
 639         return 0;
 640 }
 641
 642 static int
 643 open_single_rx_pcap(const char *pcap_filename, pcap_t **pcap)
 644 {
 645         if ((*pcap = pcap_open_offline(pcap_filename, errbuf)) == NULL) {
 646                 RTE_LOG(ERR, PMD, "Couldn't open %s: %s\n", pcap_filename, errbuf);
 647                 return -1;
 648         }
 649         return 0;
 650 }
 651
 652 /*
 653  * Opens a pcap file for writing and stores a reference to it
 654  * for use it later on.
 655  */
 656 static int
 657 open_tx_pcap(const char *key, const char *value, void *extra_args)
 658 {
 659         unsigned i;
 660         const char *pcap_filename = value;
 661         struct pmd_devargs *dumpers = extra_args;
 662         pcap_dumper_t *dumper;
 663
 664         for (i = 0; i < dumpers->num_of_queue; i++) {
 665                 if (open_single_tx_pcap(pcap_filename, &dumper) < 0)
 666                         return -1;
 667
 668                 dumpers->queue[i].dumper = dumper;
 669                 dumpers->queue[i].name = pcap_filename;
 670                 dumpers->queue[i].type = key;
 671         }
 672
 673         return 0;
 674 }
 675
 676 static int
 677 open_single_tx_pcap(const char *pcap_filename, pcap_dumper_t **dumper)
 678 {
 679         pcap_t *tx_pcap;
 680         /*
 681          * We need to create a dummy empty pcap_t to use it
 682          * with pcap_dump_open(). We create big enough an Ethernet
 683          * pcap holder.
 684          */
 685
 686         if ((tx_pcap = pcap_open_dead(DLT_EN10MB, RTE_ETH_PCAP_SNAPSHOT_LEN))
 687                         == NULL) {
 688                 RTE_LOG(ERR, PMD, "Couldn't create dead pcap\n");
 689                 return -1;
 690         }
 691
 692         /* The dumper is created using the previous pcap_t reference */
 693         if ((*dumper = pcap_dump_open(tx_pcap, pcap_filename)) == NULL) {
 694                 RTE_LOG(ERR, PMD, "Couldn't open %s for writing.\n", pcap_filename);
 695                 return -1;
 696         }
 697
 698         return 0;
 699 }
 700
 701 /*
 702  * pcap_open_live wrapper function
 703  */
 704 static inline int
 705 open_iface_live(const char *iface, pcap_t **pcap) {
 706         *pcap = pcap_open_live(iface, RTE_ETH_PCAP_SNAPLEN,
 707                         RTE_ETH_PCAP_PROMISC, RTE_ETH_PCAP_TIMEOUT, errbuf);
 708
 709         if (*pcap == NULL) {
 710                 RTE_LOG(ERR, PMD, "Couldn't open %s: %s\n", iface, errbuf);
 711                 return -1;
 712         }
 713         return 0;
 714 }
 715
 716 /*
 717  * Opens an interface for reading and writing
 718  */
 719 static inline int
 720 open_rx_tx_iface(const char *key, const char *value, void *extra_args)
 721 {
 722         const char *iface = value;
 723         struct pmd_devargs *tx = extra_args;
 724         pcap_t *pcap = NULL;
 725
 726         if (open_single_iface(iface, &pcap) < 0)
 727                 return -1;
 728
 729         tx->queue[0].pcap = pcap;
 730         tx->queue[0].name = iface;
 731         tx->queue[0].type = key;
 732
 733         return 0;
 734 }
 735
 736 /*
 737  * Opens a NIC for reading packets from it
 738  */
 739 static inline int
 740 open_rx_iface(const char *key, const char *value, void *extra_args)
 741 {
 742         unsigned i;
 743         const char *iface = value;
 744         struct pmd_devargs *rx = extra_args;
 745         pcap_t *pcap = NULL;
 746
 747         for (i = 0; i < rx->num_of_queue; i++) {
 748                 if (open_single_iface(iface, &pcap) < 0)
 749                         return -1;
 750                 rx->queue[i].pcap = pcap;
 751                 rx->queue[i].name = iface;
 752                 rx->queue[i].type = key;
 753         }
 754
 755         return 0;
 756 }
 757
 758 /*
 759  * Opens a NIC for writing packets to it
 760  */
 761 static int
 762 open_tx_iface(const char *key, const char *value, void *extra_args)
 763 {
 764         unsigned i;
 765         const char *iface = value;
 766         struct pmd_devargs *tx = extra_args;
 767         pcap_t *pcap;
 768
 769         for (i = 0; i < tx->num_of_queue; i++) {
 770                 if (open_single_iface(iface, &pcap) < 0)
 771                         return -1;
 772                 tx->queue[i].pcap = pcap;
 773                 tx->queue[i].name = iface;
 774                 tx->queue[i].type = key;
 775         }
 776
 777         return 0;
 778 }
 779
 780 static int
 781 open_single_iface(const char *iface, pcap_t **pcap)
 782 {
 783         if (open_iface_live(iface, pcap) < 0) {
 784                 RTE_LOG(ERR, PMD, "Couldn't open interface %s\n", iface);
 785                 return -1;
 786         }
 787
 788         return 0;
 789 }
 790
 791 static int
 792 rte_pmd_init_internals(const char *name, const unsigned nb_rx_queues,
 793                 const unsigned nb_tx_queues, struct pmd_internals **internals,
 794                 struct rte_eth_dev **eth_dev)
 795 {
 796         struct rte_eth_dev_data *data = NULL;
 797         unsigned int numa_node = rte_socket_id();
 798
 799         RTE_LOG(INFO, PMD, "Creating pcap-backed ethdev on numa socket %u\n",
 800                 numa_node);
 801
 802         /* now do all data allocation - for eth_dev structure
 803          * and internal (private) data
 804          */
 805         data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
 806         if (data == NULL)
 807                 goto error;
 808
 809         *internals = rte_zmalloc_socket(name, sizeof(**internals), 0,
 810                         numa_node);
 811         if (*internals == NULL)
 812                 goto error;
 813
 814         /* reserve an ethdev entry */
 815         *eth_dev = rte_eth_dev_allocate(name);
 816         if (*eth_dev == NULL)
 817                 goto error;
 818
 819         /* check length of device name */
 820         if ((strlen((*eth_dev)->data->name) + 1) > sizeof(data->name))
 821                 goto error;
 822
 823         /* now put it all together
 824          * - store queue data in internals,
 825          * - store numa_node info in eth_dev
 826          * - point eth_dev_data to internals
 827          * - and point eth_dev structure to new eth_dev_data structure
 828          */
 829         data->dev_private = *internals;
 830         data->port_id = (*eth_dev)->data->port_id;
 831         snprintf(data->name, sizeof(data->name), "%s", (*eth_dev)->data->name);
 832         data->nb_rx_queues = (uint16_t)nb_rx_queues;
 833         data->nb_tx_queues = (uint16_t)nb_tx_queues;
 834         data->dev_link = pmd_link;
 835         data->mac_addrs = &eth_addr;
 836         strncpy(data->name,
 837                 (*eth_dev)->data->name, strlen((*eth_dev)->data->name));
 838
 839         /*
 840          * NOTE: we'll replace the data element, of originally allocated
 841          * eth_dev so the rings are local per-process
 842          */
 843         (*eth_dev)->data = data;
 844         (*eth_dev)->dev_ops = &ops;
 845         (*eth_dev)->driver = NULL;
 846         data->dev_flags = RTE_ETH_DEV_DETACHABLE;
 847         data->kdrv = RTE_KDRV_NONE;
 848         data->drv_name = drivername;
 849         data->numa_node = numa_node;
 850
 851         return 0;
 852
 853 error:
 854         rte_free(data);
 855         rte_free(*internals);
 856
 857         return -1;
 858 }
 859
 860 static int
 861 rte_eth_from_pcaps_common(const char *name, struct pmd_devargs *rx_queues,
 862                 const unsigned nb_rx_queues, struct pmd_devargs *tx_queues,
 863                 const unsigned nb_tx_queues, struct rte_kvargs *kvlist,
 864                 struct pmd_internals **internals, struct rte_eth_dev **eth_dev)
 865 {
 866         struct rte_kvargs_pair *pair = NULL;
 867         unsigned k_idx;
 868         unsigned i;
 869
 870         /* do some parameter checking */
 871         if (rx_queues == NULL && nb_rx_queues > 0)
 872                 return -1;
 873         if (tx_queues == NULL && nb_tx_queues > 0)
 874                 return -1;
 875
 876         if (rte_pmd_init_internals(name, nb_rx_queues, nb_tx_queues, internals,
 877                         eth_dev) < 0)
 878                 return -1;
 879
 880         for (i = 0; i < nb_rx_queues; i++) {
 881                 (*internals)->rx_queue[i].pcap = rx_queues->queue[i].pcap;
 882                 snprintf((*internals)->rx_queue[i].name,
 883                         sizeof((*internals)->rx_queue[i].name), "%s",
 884                         rx_queues->queue[i].name);
 885                 snprintf((*internals)->rx_queue[i].type,
 886                         sizeof((*internals)->rx_queue[i].type), "%s",
 887                         rx_queues->queue[i].type);
 888         }
 889         for (i = 0; i < nb_tx_queues; i++) {
 890                 (*internals)->tx_queue[i].dumper = tx_queues->queue[i].dumper;
 891                 snprintf((*internals)->tx_queue[i].name,
 892                         sizeof((*internals)->tx_queue[i].name), "%s",
 893                         tx_queues->queue[i].name);
 894                 snprintf((*internals)->tx_queue[i].type,
 895                         sizeof((*internals)->tx_queue[i].type), "%s",
 896                         tx_queues->queue[i].type);
 897         }
 898
 899         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 900                 pair = &kvlist->pairs[k_idx];
 901                 if (strstr(pair->key, ETH_PCAP_IFACE_ARG) != NULL)
 902                         break;
 903         }
 904
 905         if (pair == NULL)
 906                 (*internals)->if_index = 0;
 907         else
 908                 (*internals)->if_index = if_nametoindex(pair->value);
 909
 910         return 0;
 911 }
 912
 913 static int
 914 rte_eth_from_pcaps(const char *name, struct pmd_devargs *rx_queues,
 915                 const unsigned nb_rx_queues, struct pmd_devargs *tx_queues,
 916                 const unsigned nb_tx_queues, struct rte_kvargs *kvlist,
 917                 int single_iface, unsigned int using_dumpers)
 918 {
 919         struct pmd_internals *internals = NULL;
 920         struct rte_eth_dev *eth_dev = NULL;
 921         int ret;
 922
 923         ret = rte_eth_from_pcaps_common(name, rx_queues, nb_rx_queues,
 924                 tx_queues, nb_tx_queues, kvlist, &internals, &eth_dev);
 925
 926         if (ret < 0)
 927                 return ret;
 928
 929         /* store wether we are using a single interface for rx/tx or not */
 930         internals->single_iface = single_iface;
 931
 932         eth_dev->rx_pkt_burst = eth_pcap_rx;
 933
 934         if (using_dumpers)
 935                 eth_dev->tx_pkt_burst = eth_pcap_tx_dumper;
 936         else
 937                 eth_dev->tx_pkt_burst = eth_pcap_tx;
 938
 939         return 0;
 940 }
 941
 942
 943 static int
 944 rte_pmd_pcap_devinit(const char *name, const char *params)
 945 {
 946         unsigned int is_rx_pcap = 0, is_tx_pcap = 0;
 947         struct rte_kvargs *kvlist;
 948         struct pmd_devargs pcaps = {0};
 949         struct pmd_devargs dumpers = {0};
 950         int single_iface = 0;
 951         int ret;
 952
 953         RTE_LOG(INFO, PMD, "Initializing pmd_pcap for %s\n", name);
 954
 955         gettimeofday(&start_time, NULL);
 956         start_cycles = rte_get_timer_cycles();
 957         hz = rte_get_timer_hz();
 958
 959         kvlist = rte_kvargs_parse(params, valid_arguments);
 960         if (kvlist == NULL)
 961                 return -1;
 962
 963         /*
 964          * If iface argument is passed we open the NICs and use them for
 965          * reading / writing
 966          */
 967         if (rte_kvargs_count(kvlist, ETH_PCAP_IFACE_ARG) == 1) {
 968
 969                 ret = rte_kvargs_process(kvlist, ETH_PCAP_IFACE_ARG,
 970                                 &open_rx_tx_iface, &pcaps);
 971
 972                 if (ret < 0)
 973                         goto free_kvlist;
 974                 dumpers.queue[0].pcap = pcaps.queue[0].pcap;
 975                 dumpers.queue[0].name = pcaps.queue[0].name;
 976                 dumpers.queue[0].type = pcaps.queue[0].type;
 977
 978                 single_iface = 1;
 979
 980                 ret = rte_eth_from_pcaps(name, &pcaps, 1, &dumpers, 1,
 981                         kvlist, single_iface, is_tx_pcap);
 982
 983                 goto free_kvlist;
 984         }
 985
 986         /*
 987          * We check whether we want to open a RX stream from a real NIC or a
 988          * pcap file
 989          */
 990         pcaps.num_of_queue = rte_kvargs_count(kvlist, ETH_PCAP_RX_PCAP_ARG);
 991         if (pcaps.num_of_queue)
 992                 is_rx_pcap = 1;
 993         else
 994                 pcaps.num_of_queue = rte_kvargs_count(kvlist,
 995                                 ETH_PCAP_RX_IFACE_ARG);
 996
 997         if (pcaps.num_of_queue > RTE_PMD_PCAP_MAX_QUEUES)
 998                 pcaps.num_of_queue = RTE_PMD_PCAP_MAX_QUEUES;
 999
1000         if (is_rx_pcap)
1001                 ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_PCAP_ARG,
1002                                 &open_rx_pcap, &pcaps);
1003         else
1004                 ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_IFACE_ARG,
1005                                 &open_rx_iface, &pcaps);
1006
1007         if (ret < 0)
1008                 goto free_kvlist;
1009
1010         /*
1011          * We check whether we want to open a TX stream to a real NIC or a
1012          * pcap file
1013          */
1014         dumpers.num_of_queue = rte_kvargs_count(kvlist, ETH_PCAP_TX_PCAP_ARG);
1015         if (dumpers.num_of_queue)
1016                 is_tx_pcap = 1;
1017         else
1018                 dumpers.num_of_queue = rte_kvargs_count(kvlist,
1019                                 ETH_PCAP_TX_IFACE_ARG);
1020
1021         if (dumpers.num_of_queue > RTE_PMD_PCAP_MAX_QUEUES)
1022                 dumpers.num_of_queue = RTE_PMD_PCAP_MAX_QUEUES;
1023
1024         if (is_tx_pcap)
1025                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_PCAP_ARG,
1026                                 &open_tx_pcap, &dumpers);
1027         else
1028                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_IFACE_ARG,
1029                                 &open_tx_iface, &dumpers);
1030
1031         if (ret < 0)
1032                 goto free_kvlist;
1033
1034         ret = rte_eth_from_pcaps(name, &pcaps, pcaps.num_of_queue, &dumpers,
1035                 dumpers.num_of_queue, kvlist, single_iface, is_tx_pcap);
1036
1037 free_kvlist:
1038         rte_kvargs_free(kvlist);
1039         return ret;
1040 }
1041
1042 static int
1043 rte_pmd_pcap_devuninit(const char *name)
1044 {
1045         struct rte_eth_dev *eth_dev = NULL;
1046
1047         RTE_LOG(INFO, PMD, "Closing pcap ethdev on numa socket %u\n",
1048                         rte_socket_id());
1049
1050         if (name == NULL)
1051                 return -1;
1052
1053         /* reserve an ethdev entry */
1054         eth_dev = rte_eth_dev_allocated(name);
1055         if (eth_dev == NULL)
1056                 return -1;
1057
1058         rte_free(eth_dev->data->dev_private);
1059         rte_free(eth_dev->data);
1060
1061         rte_eth_dev_release_port(eth_dev);
1062
1063         return 0;
1064 }
1065
1066 static struct rte_vdev_driver pmd_pcap_drv = {
1067         .init = rte_pmd_pcap_devinit,
1068         .uninit = rte_pmd_pcap_devuninit,
1069 };
1070
1071 DRIVER_REGISTER_VDEV(net_pcap, pmd_pcap_drv);
1072 DRIVER_REGISTER_PARAM_STRING(net_pcap,
1073         ETH_PCAP_RX_PCAP_ARG "=<string> "
1074         ETH_PCAP_TX_PCAP_ARG "=<string> "
1075         ETH_PCAP_RX_IFACE_ARG "=<ifc> "
1076         ETH_PCAP_TX_IFACE_ARG "=<ifc> "
1077         ETH_PCAP_IFACE_ARG "=<ifc>");