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 pcap_rx_queue {
  70         pcap_t *pcap;
  71         uint8_t in_port;
  72         struct rte_mempool *mb_pool;
  73         volatile unsigned long rx_pkts;
  74         volatile unsigned long rx_bytes;
  75         volatile unsigned long err_pkts;
  76         char name[PATH_MAX];
  77         char type[ETH_PCAP_ARG_MAXLEN];
  78 };
  79
  80 struct pcap_tx_queue {
  81         pcap_dumper_t *dumper;
  82         pcap_t *pcap;
  83         volatile unsigned long tx_pkts;
  84         volatile unsigned long tx_bytes;
  85         volatile unsigned long err_pkts;
  86         char name[PATH_MAX];
  87         char type[ETH_PCAP_ARG_MAXLEN];
  88 };
  89
  90 struct rx_pcaps {
  91         unsigned num_of_rx;
  92         pcap_t *pcaps[RTE_PMD_PCAP_MAX_QUEUES];
  93         const char *names[RTE_PMD_PCAP_MAX_QUEUES];
  94         const char *types[RTE_PMD_PCAP_MAX_QUEUES];
  95 };
  96
  97 struct tx_pcaps {
  98         unsigned num_of_tx;
  99         pcap_dumper_t *dumpers[RTE_PMD_PCAP_MAX_QUEUES];
 100         pcap_t *pcaps[RTE_PMD_PCAP_MAX_QUEUES];
 101         const char *names[RTE_PMD_PCAP_MAX_QUEUES];
 102         const char *types[RTE_PMD_PCAP_MAX_QUEUES];
 103 };
 104
 105 struct pmd_internals {
 106         struct pcap_rx_queue rx_queue[RTE_PMD_PCAP_MAX_QUEUES];
 107         struct pcap_tx_queue tx_queue[RTE_PMD_PCAP_MAX_QUEUES];
 108         int if_index;
 109         int single_iface;
 110 };
 111
 112 const char *valid_arguments[] = {
 113         ETH_PCAP_RX_PCAP_ARG,
 114         ETH_PCAP_TX_PCAP_ARG,
 115         ETH_PCAP_RX_IFACE_ARG,
 116         ETH_PCAP_TX_IFACE_ARG,
 117         ETH_PCAP_IFACE_ARG,
 118         NULL
 119 };
 120
 121 static int open_single_tx_pcap(const char *pcap_filename, pcap_dumper_t **dumper);
 122 static int open_single_rx_pcap(const char *pcap_filename, pcap_t **pcap);
 123 static int open_single_iface(const char *iface, pcap_t **pcap);
 124
 125 static struct ether_addr eth_addr = { .addr_bytes = { 0, 0, 0, 0x1, 0x2, 0x3 } };
 126 static const char *drivername = "Pcap PMD";
 127 static struct rte_eth_link pmd_link = {
 128                 .link_speed = ETH_SPEED_NUM_10G,
 129                 .link_duplex = ETH_LINK_FULL_DUPLEX,
 130                 .link_status = ETH_LINK_DOWN,
 131                 .link_autoneg = ETH_LINK_SPEED_FIXED,
 132 };
 133
 134 static int
 135 eth_pcap_rx_jumbo(struct rte_mempool *mb_pool,
 136                   struct rte_mbuf *mbuf,
 137                   const u_char *data,
 138                   uint16_t data_len)
 139 {
 140         struct rte_mbuf *m = mbuf;
 141
 142         /* Copy the first segment. */
 143         uint16_t len = rte_pktmbuf_tailroom(mbuf);
 144
 145         rte_memcpy(rte_pktmbuf_append(mbuf, len), data, len);
 146         data_len -= len;
 147         data += len;
 148
 149         while (data_len > 0) {
 150                 /* Allocate next mbuf and point to that. */
 151                 m->next = rte_pktmbuf_alloc(mb_pool);
 152
 153                 if (unlikely(!m->next))
 154                         return -1;
 155
 156                 m = m->next;
 157
 158                 /* Headroom is not needed in chained mbufs. */
 159                 rte_pktmbuf_prepend(m, rte_pktmbuf_headroom(m));
 160                 m->pkt_len = 0;
 161                 m->data_len = 0;
 162
 163                 /* Copy next segment. */
 164                 len = RTE_MIN(rte_pktmbuf_tailroom(m), data_len);
 165                 rte_memcpy(rte_pktmbuf_append(m, len), data, len);
 166
 167                 mbuf->nb_segs++;
 168                 data_len -= len;
 169                 data += len;
 170         }
 171
 172         return mbuf->nb_segs;
 173 }
 174
 175 /* Copy data from mbuf chain to a buffer suitable for writing to a PCAP file. */
 176 static void
 177 eth_pcap_gather_data(unsigned char *data, struct rte_mbuf *mbuf)
 178 {
 179         uint16_t data_len = 0;
 180
 181         while (mbuf) {
 182                 rte_memcpy(data + data_len, rte_pktmbuf_mtod(mbuf, void *),
 183                            mbuf->data_len);
 184
 185                 data_len += mbuf->data_len;
 186                 mbuf = mbuf->next;
 187         }
 188 }
 189
 190 static uint16_t
 191 eth_pcap_rx(void *queue,
 192                 struct rte_mbuf **bufs,
 193                 uint16_t nb_pkts)
 194 {
 195         unsigned i;
 196         struct pcap_pkthdr header;
 197         const u_char *packet;
 198         struct rte_mbuf *mbuf;
 199         struct pcap_rx_queue *pcap_q = queue;
 200         uint16_t num_rx = 0;
 201         uint16_t buf_size;
 202         uint32_t rx_bytes = 0;
 203
 204         if (unlikely(pcap_q->pcap == NULL || nb_pkts == 0))
 205                 return 0;
 206
 207         /* Reads the given number of packets from the pcap file one by one
 208          * and copies the packet data into a newly allocated mbuf to return.
 209          */
 210         for (i = 0; i < nb_pkts; i++) {
 211                 /* Get the next PCAP packet */
 212                 packet = pcap_next(pcap_q->pcap, &header);
 213                 if (unlikely(packet == NULL))
 214                         break;
 215                 else
 216                         mbuf = rte_pktmbuf_alloc(pcap_q->mb_pool);
 217                 if (unlikely(mbuf == NULL))
 218                         break;
 219
 220                 /* Now get the space available for data in the mbuf */
 221                 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(pcap_q->mb_pool) -
 222                                 RTE_PKTMBUF_HEADROOM);
 223
 224                 if (header.caplen <= buf_size) {
 225                         /* pcap packet will fit in the mbuf, go ahead and copy */
 226                         rte_memcpy(rte_pktmbuf_mtod(mbuf, void *), packet,
 227                                         header.caplen);
 228                         mbuf->data_len = (uint16_t)header.caplen;
 229                 } else {
 230                         /* Try read jumbo frame into multi mbufs. */
 231                         if (unlikely(eth_pcap_rx_jumbo(pcap_q->mb_pool,
 232                                                        mbuf,
 233                                                        packet,
 234                                                        header.caplen) == -1))
 235                                 break;
 236                 }
 237
 238                 mbuf->pkt_len = (uint16_t)header.caplen;
 239                 mbuf->port = pcap_q->in_port;
 240                 bufs[num_rx] = mbuf;
 241                 num_rx++;
 242                 rx_bytes += header.caplen;
 243         }
 244         pcap_q->rx_pkts += num_rx;
 245         pcap_q->rx_bytes += rx_bytes;
 246         return num_rx;
 247 }
 248
 249 static inline void
 250 calculate_timestamp(struct timeval *ts) {
 251         uint64_t cycles;
 252         struct timeval cur_time;
 253
 254         cycles = rte_get_timer_cycles() - start_cycles;
 255         cur_time.tv_sec = cycles / hz;
 256         cur_time.tv_usec = (cycles % hz) * 10e6 / hz;
 257         timeradd(&start_time, &cur_time, ts);
 258 }
 259
 260 /*
 261  * Callback to handle writing packets to a pcap file.
 262  */
 263 static uint16_t
 264 eth_pcap_tx_dumper(void *queue,
 265                 struct rte_mbuf **bufs,
 266                 uint16_t nb_pkts)
 267 {
 268         unsigned i;
 269         struct rte_mbuf *mbuf;
 270         struct pcap_tx_queue *dumper_q = queue;
 271         uint16_t num_tx = 0;
 272         uint32_t tx_bytes = 0;
 273         struct pcap_pkthdr header;
 274
 275         if (dumper_q->dumper == NULL || nb_pkts == 0)
 276                 return 0;
 277
 278         /* writes the nb_pkts packets to the previously opened pcap file dumper */
 279         for (i = 0; i < nb_pkts; i++) {
 280                 mbuf = bufs[i];
 281                 calculate_timestamp(&header.ts);
 282                 header.len = mbuf->pkt_len;
 283                 header.caplen = header.len;
 284
 285                 if (likely(mbuf->nb_segs == 1)) {
 286                         pcap_dump((u_char *)dumper_q->dumper, &header,
 287                                   rte_pktmbuf_mtod(mbuf, void*));
 288                 } else {
 289                         if (mbuf->pkt_len <= ETHER_MAX_JUMBO_FRAME_LEN) {
 290                                 eth_pcap_gather_data(tx_pcap_data, mbuf);
 291                                 pcap_dump((u_char *)dumper_q->dumper, &header,
 292                                           tx_pcap_data);
 293                         } else {
 294                                 RTE_LOG(ERR, PMD,
 295                                         "Dropping PCAP packet. "
 296                                         "Size (%d) > max jumbo size (%d).\n",
 297                                         mbuf->pkt_len,
 298                                         ETHER_MAX_JUMBO_FRAME_LEN);
 299
 300                                 rte_pktmbuf_free(mbuf);
 301                                 break;
 302                         }
 303                 }
 304
 305                 rte_pktmbuf_free(mbuf);
 306                 num_tx++;
 307                 tx_bytes += mbuf->pkt_len;
 308         }
 309
 310         /*
 311          * Since there's no place to hook a callback when the forwarding
 312          * process stops and to make sure the pcap file is actually written,
 313          * we flush the pcap dumper within each burst.
 314          */
 315         pcap_dump_flush(dumper_q->dumper);
 316         dumper_q->tx_pkts += num_tx;
 317         dumper_q->tx_bytes += tx_bytes;
 318         dumper_q->err_pkts += nb_pkts - num_tx;
 319         return num_tx;
 320 }
 321
 322 /*
 323  * Callback to handle sending packets through a real NIC.
 324  */
 325 static uint16_t
 326 eth_pcap_tx(void *queue,
 327                 struct rte_mbuf **bufs,
 328                 uint16_t nb_pkts)
 329 {
 330         unsigned i;
 331         int ret;
 332         struct rte_mbuf *mbuf;
 333         struct pcap_tx_queue *tx_queue = queue;
 334         uint16_t num_tx = 0;
 335         uint32_t tx_bytes = 0;
 336
 337         if (unlikely(nb_pkts == 0 || tx_queue->pcap == NULL))
 338                 return 0;
 339
 340         for (i = 0; i < nb_pkts; i++) {
 341                 mbuf = bufs[i];
 342
 343                 if (likely(mbuf->nb_segs == 1)) {
 344                         ret = pcap_sendpacket(tx_queue->pcap,
 345                                               rte_pktmbuf_mtod(mbuf, u_char *),
 346                                               mbuf->pkt_len);
 347                 } else {
 348                         if (mbuf->pkt_len <= ETHER_MAX_JUMBO_FRAME_LEN) {
 349                                 eth_pcap_gather_data(tx_pcap_data, mbuf);
 350                                 ret = pcap_sendpacket(tx_queue->pcap,
 351                                                       tx_pcap_data,
 352                                                       mbuf->pkt_len);
 353                         } else {
 354                                 RTE_LOG(ERR, PMD,
 355                                         "Dropping PCAP packet. "
 356                                         "Size (%d) > max jumbo size (%d).\n",
 357                                         mbuf->pkt_len,
 358                                         ETHER_MAX_JUMBO_FRAME_LEN);
 359
 360                                 rte_pktmbuf_free(mbuf);
 361                                 break;
 362                         }
 363                 }
 364
 365                 if (unlikely(ret != 0))
 366                         break;
 367                 num_tx++;
 368                 tx_bytes += mbuf->pkt_len;
 369                 rte_pktmbuf_free(mbuf);
 370         }
 371
 372         tx_queue->tx_pkts += num_tx;
 373         tx_queue->tx_bytes += tx_bytes;
 374         tx_queue->err_pkts += nb_pkts - num_tx;
 375         return num_tx;
 376 }
 377
 378 static int
 379 eth_dev_start(struct rte_eth_dev *dev)
 380 {
 381         unsigned i;
 382         struct pmd_internals *internals = dev->data->dev_private;
 383         struct pcap_tx_queue *tx;
 384         struct pcap_rx_queue *rx;
 385
 386         /* Special iface case. Single pcap is open and shared between tx/rx. */
 387         if (internals->single_iface) {
 388                 tx = &internals->tx_queue[0];
 389                 rx = &internals->rx_queue[0];
 390
 391                 if (!tx->pcap && strcmp(tx->type, ETH_PCAP_IFACE_ARG) == 0) {
 392                         if (open_single_iface(tx->name, &tx->pcap) < 0)
 393                                 return -1;
 394                         rx->pcap = tx->pcap;
 395                 }
 396                 goto status_up;
 397         }
 398
 399         /* If not open already, open tx pcaps/dumpers */
 400         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 401                 tx = &internals->tx_queue[i];
 402
 403                 if (!tx->dumper && strcmp(tx->type, ETH_PCAP_TX_PCAP_ARG) == 0) {
 404                         if (open_single_tx_pcap(tx->name, &tx->dumper) < 0)
 405                                 return -1;
 406                 }
 407
 408                 else if (!tx->pcap && strcmp(tx->type, ETH_PCAP_TX_IFACE_ARG) == 0) {
 409                         if (open_single_iface(tx->name, &tx->pcap) < 0)
 410                                 return -1;
 411                 }
 412         }
 413
 414         /* If not open already, open rx pcaps */
 415         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 416                 rx = &internals->rx_queue[i];
 417
 418                 if (rx->pcap != NULL)
 419                         continue;
 420
 421                 if (strcmp(rx->type, ETH_PCAP_RX_PCAP_ARG) == 0) {
 422                         if (open_single_rx_pcap(rx->name, &rx->pcap) < 0)
 423                                 return -1;
 424                 }
 425
 426                 else if (strcmp(rx->type, ETH_PCAP_RX_IFACE_ARG) == 0) {
 427                         if (open_single_iface(rx->name, &rx->pcap) < 0)
 428                                 return -1;
 429                 }
 430         }
 431
 432 status_up:
 433
 434         dev->data->dev_link.link_status = ETH_LINK_UP;
 435         return 0;
 436 }
 437
 438 /*
 439  * This function gets called when the current port gets stopped.
 440  * Is the only place for us to close all the tx streams dumpers.
 441  * If not called the dumpers will be flushed within each tx burst.
 442  */
 443 static void
 444 eth_dev_stop(struct rte_eth_dev *dev)
 445 {
 446         unsigned i;
 447         struct pmd_internals *internals = dev->data->dev_private;
 448         struct pcap_tx_queue *tx;
 449         struct pcap_rx_queue *rx;
 450
 451         /* Special iface case. Single pcap is open and shared between tx/rx. */
 452         if (internals->single_iface) {
 453                 tx = &internals->tx_queue[0];
 454                 rx = &internals->rx_queue[0];
 455                 pcap_close(tx->pcap);
 456                 tx->pcap = NULL;
 457                 rx->pcap = NULL;
 458                 goto status_down;
 459         }
 460
 461         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 462                 tx = &internals->tx_queue[i];
 463
 464                 if (tx->dumper != NULL) {
 465                         pcap_dump_close(tx->dumper);
 466                         tx->dumper = NULL;
 467                 }
 468
 469                 if (tx->pcap != NULL) {
 470                         pcap_close(tx->pcap);
 471                         tx->pcap = NULL;
 472                 }
 473         }
 474
 475         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 476                 rx = &internals->rx_queue[i];
 477
 478                 if (rx->pcap != NULL) {
 479                         pcap_close(rx->pcap);
 480                         rx->pcap = NULL;
 481                 }
 482         }
 483
 484 status_down:
 485         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 486 }
 487
 488 static int
 489 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 490 {
 491         return 0;
 492 }
 493
 494 static void
 495 eth_dev_info(struct rte_eth_dev *dev,
 496                 struct rte_eth_dev_info *dev_info)
 497 {
 498         struct pmd_internals *internals = dev->data->dev_private;
 499         dev_info->driver_name = drivername;
 500         dev_info->if_index = internals->if_index;
 501         dev_info->max_mac_addrs = 1;
 502         dev_info->max_rx_pktlen = (uint32_t) -1;
 503         dev_info->max_rx_queues = dev->data->nb_rx_queues;
 504         dev_info->max_tx_queues = dev->data->nb_tx_queues;
 505         dev_info->min_rx_bufsize = 0;
 506         dev_info->pci_dev = NULL;
 507 }
 508
 509 static void
 510 eth_stats_get(struct rte_eth_dev *dev,
 511                 struct rte_eth_stats *igb_stats)
 512 {
 513         unsigned i;
 514         unsigned long rx_packets_total = 0, rx_bytes_total = 0;
 515         unsigned long tx_packets_total = 0, tx_bytes_total = 0;
 516         unsigned long tx_packets_err_total = 0;
 517         const struct pmd_internals *internal = dev->data->dev_private;
 518
 519         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 520                         i < dev->data->nb_rx_queues; i++) {
 521                 igb_stats->q_ipackets[i] = internal->rx_queue[i].rx_pkts;
 522                 igb_stats->q_ibytes[i] = internal->rx_queue[i].rx_bytes;
 523                 rx_packets_total += igb_stats->q_ipackets[i];
 524                 rx_bytes_total += igb_stats->q_ibytes[i];
 525         }
 526
 527         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 528                         i < dev->data->nb_tx_queues; i++) {
 529                 igb_stats->q_opackets[i] = internal->tx_queue[i].tx_pkts;
 530                 igb_stats->q_obytes[i] = internal->tx_queue[i].tx_bytes;
 531                 igb_stats->q_errors[i] = internal->tx_queue[i].err_pkts;
 532                 tx_packets_total += igb_stats->q_opackets[i];
 533                 tx_bytes_total += igb_stats->q_obytes[i];
 534                 tx_packets_err_total += igb_stats->q_errors[i];
 535         }
 536
 537         igb_stats->ipackets = rx_packets_total;
 538         igb_stats->ibytes = rx_bytes_total;
 539         igb_stats->opackets = tx_packets_total;
 540         igb_stats->obytes = tx_bytes_total;
 541         igb_stats->oerrors = tx_packets_err_total;
 542 }
 543
 544 static void
 545 eth_stats_reset(struct rte_eth_dev *dev)
 546 {
 547         unsigned i;
 548         struct pmd_internals *internal = dev->data->dev_private;
 549         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 550                 internal->rx_queue[i].rx_pkts = 0;
 551                 internal->rx_queue[i].rx_bytes = 0;
 552         }
 553         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 554                 internal->tx_queue[i].tx_pkts = 0;
 555                 internal->tx_queue[i].tx_bytes = 0;
 556                 internal->tx_queue[i].err_pkts = 0;
 557         }
 558 }
 559
 560 static void
 561 eth_dev_close(struct rte_eth_dev *dev __rte_unused)
 562 {
 563 }
 564
 565 static void
 566 eth_queue_release(void *q __rte_unused)
 567 {
 568 }
 569
 570 static int
 571 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 572                 int wait_to_complete __rte_unused)
 573 {
 574         return 0;
 575 }
 576
 577 static int
 578 eth_rx_queue_setup(struct rte_eth_dev *dev,
 579                 uint16_t rx_queue_id,
 580                 uint16_t nb_rx_desc __rte_unused,
 581                 unsigned int socket_id __rte_unused,
 582                 const struct rte_eth_rxconf *rx_conf __rte_unused,
 583                 struct rte_mempool *mb_pool)
 584 {
 585         struct pmd_internals *internals = dev->data->dev_private;
 586         struct pcap_rx_queue *pcap_q = &internals->rx_queue[rx_queue_id];
 587         pcap_q->mb_pool = mb_pool;
 588         dev->data->rx_queues[rx_queue_id] = pcap_q;
 589         pcap_q->in_port = dev->data->port_id;
 590         return 0;
 591 }
 592
 593 static int
 594 eth_tx_queue_setup(struct rte_eth_dev *dev,
 595                 uint16_t tx_queue_id,
 596                 uint16_t nb_tx_desc __rte_unused,
 597                 unsigned int socket_id __rte_unused,
 598                 const struct rte_eth_txconf *tx_conf __rte_unused)
 599 {
 600
 601         struct pmd_internals *internals = dev->data->dev_private;
 602         dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
 603         return 0;
 604 }
 605
 606 static const struct eth_dev_ops ops = {
 607         .dev_start = eth_dev_start,
 608         .dev_stop =     eth_dev_stop,
 609         .dev_close = eth_dev_close,
 610         .dev_configure = eth_dev_configure,
 611         .dev_infos_get = eth_dev_info,
 612         .rx_queue_setup = eth_rx_queue_setup,
 613         .tx_queue_setup = eth_tx_queue_setup,
 614         .rx_queue_release = eth_queue_release,
 615         .tx_queue_release = eth_queue_release,
 616         .link_update = eth_link_update,
 617         .stats_get = eth_stats_get,
 618         .stats_reset = eth_stats_reset,
 619 };
 620
 621 /*
 622  * Function handler that opens the pcap file for reading a stores a
 623  * reference of it for use it later on.
 624  */
 625 static int
 626 open_rx_pcap(const char *key, const char *value, void *extra_args)
 627 {
 628         unsigned i;
 629         const char *pcap_filename = value;
 630         struct rx_pcaps *pcaps = extra_args;
 631         pcap_t *pcap = NULL;
 632
 633         for (i = 0; i < pcaps->num_of_rx; i++) {
 634                 if (open_single_rx_pcap(pcap_filename, &pcap) < 0)
 635                         return -1;
 636
 637                 pcaps->pcaps[i] = pcap;
 638                 pcaps->names[i] = pcap_filename;
 639                 pcaps->types[i] = key;
 640         }
 641
 642         return 0;
 643 }
 644
 645 static int
 646 open_single_rx_pcap(const char *pcap_filename, pcap_t **pcap)
 647 {
 648         if ((*pcap = pcap_open_offline(pcap_filename, errbuf)) == NULL) {
 649                 RTE_LOG(ERR, PMD, "Couldn't open %s: %s\n", pcap_filename, errbuf);
 650                 return -1;
 651         }
 652         return 0;
 653 }
 654
 655 /*
 656  * Opens a pcap file for writing and stores a reference to it
 657  * for use it later on.
 658  */
 659 static int
 660 open_tx_pcap(const char *key, const char *value, void *extra_args)
 661 {
 662         unsigned i;
 663         const char *pcap_filename = value;
 664         struct tx_pcaps *dumpers = extra_args;
 665         pcap_dumper_t *dumper;
 666
 667         for (i = 0; i < dumpers->num_of_tx; i++) {
 668                 if (open_single_tx_pcap(pcap_filename, &dumper) < 0)
 669                         return -1;
 670
 671                 dumpers->dumpers[i] = dumper;
 672                 dumpers->names[i] = pcap_filename;
 673                 dumpers->types[i] = key;
 674         }
 675
 676         return 0;
 677 }
 678
 679 static int
 680 open_single_tx_pcap(const char *pcap_filename, pcap_dumper_t **dumper)
 681 {
 682         pcap_t *tx_pcap;
 683         /*
 684          * We need to create a dummy empty pcap_t to use it
 685          * with pcap_dump_open(). We create big enough an Ethernet
 686          * pcap holder.
 687          */
 688
 689         if ((tx_pcap = pcap_open_dead(DLT_EN10MB, RTE_ETH_PCAP_SNAPSHOT_LEN))
 690                         == NULL) {
 691                 RTE_LOG(ERR, PMD, "Couldn't create dead pcap\n");
 692                 return -1;
 693         }
 694
 695         /* The dumper is created using the previous pcap_t reference */
 696         if ((*dumper = pcap_dump_open(tx_pcap, pcap_filename)) == NULL) {
 697                 RTE_LOG(ERR, PMD, "Couldn't open %s for writing.\n", pcap_filename);
 698                 return -1;
 699         }
 700
 701         return 0;
 702 }
 703
 704 /*
 705  * pcap_open_live wrapper function
 706  */
 707 static inline int
 708 open_iface_live(const char *iface, pcap_t **pcap) {
 709         *pcap = pcap_open_live(iface, RTE_ETH_PCAP_SNAPLEN,
 710                         RTE_ETH_PCAP_PROMISC, RTE_ETH_PCAP_TIMEOUT, errbuf);
 711
 712         if (*pcap == NULL) {
 713                 RTE_LOG(ERR, PMD, "Couldn't open %s: %s\n", iface, errbuf);
 714                 return -1;
 715         }
 716         return 0;
 717 }
 718
 719 /*
 720  * Opens an interface for reading and writing
 721  */
 722 static inline int
 723 open_rx_tx_iface(const char *key, const char *value, void *extra_args)
 724 {
 725         const char *iface = value;
 726         struct rx_pcaps *pcaps = extra_args;
 727         pcap_t *pcap = NULL;
 728
 729         if (open_single_iface(iface, &pcap) < 0)
 730                 return -1;
 731
 732         pcaps->pcaps[0] = pcap;
 733         pcaps->names[0] = iface;
 734         pcaps->types[0] = key;
 735
 736         return 0;
 737 }
 738
 739 /*
 740  * Opens a NIC for reading packets from it
 741  */
 742 static inline int
 743 open_rx_iface(const char *key, const char *value, void *extra_args)
 744 {
 745         unsigned i;
 746         const char *iface = value;
 747         struct rx_pcaps *pcaps = extra_args;
 748         pcap_t *pcap = NULL;
 749
 750         for (i = 0; i < pcaps->num_of_rx; i++) {
 751                 if (open_single_iface(iface, &pcap) < 0)
 752                         return -1;
 753                 pcaps->pcaps[i] = pcap;
 754                 pcaps->names[i] = iface;
 755                 pcaps->types[i] = key;
 756         }
 757
 758         return 0;
 759 }
 760
 761 /*
 762  * Opens a NIC for writing packets to it
 763  */
 764 static int
 765 open_tx_iface(const char *key, const char *value, void *extra_args)
 766 {
 767         unsigned i;
 768         const char *iface = value;
 769         struct tx_pcaps *pcaps = extra_args;
 770         pcap_t *pcap;
 771
 772         for (i = 0; i < pcaps->num_of_tx; i++) {
 773                 if (open_single_iface(iface, &pcap) < 0)
 774                         return -1;
 775                 pcaps->pcaps[i] = pcap;
 776                 pcaps->names[i] = iface;
 777                 pcaps->types[i] = key;
 778         }
 779
 780         return 0;
 781 }
 782
 783 static int
 784 open_single_iface(const char *iface, pcap_t **pcap)
 785 {
 786         if (open_iface_live(iface, pcap) < 0) {
 787                 RTE_LOG(ERR, PMD, "Couldn't open interface %s\n", iface);
 788                 return -1;
 789         }
 790
 791         return 0;
 792 }
 793
 794 static int
 795 rte_pmd_init_internals(const char *name, const unsigned nb_rx_queues,
 796                 const unsigned nb_tx_queues,
 797                 const unsigned numa_node,
 798                 struct pmd_internals **internals,
 799                 struct rte_eth_dev **eth_dev,
 800                 struct rte_kvargs *kvlist)
 801 {
 802         struct rte_eth_dev_data *data = NULL;
 803         unsigned k_idx;
 804         struct rte_kvargs_pair *pair = NULL;
 805
 806         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 807                 pair = &kvlist->pairs[k_idx];
 808                 if (strstr(pair->key, ETH_PCAP_IFACE_ARG) != NULL)
 809                         break;
 810         }
 811
 812         RTE_LOG(INFO, PMD,
 813                         "Creating pcap-backed ethdev on numa socket %u\n", numa_node);
 814
 815         /* now do all data allocation - for eth_dev structure
 816          * and internal (private) data
 817          */
 818         data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
 819         if (data == NULL)
 820                 goto error;
 821
 822         *internals = rte_zmalloc_socket(name, sizeof(**internals), 0, numa_node);
 823         if (*internals == NULL)
 824                 goto error;
 825
 826         /* reserve an ethdev entry */
 827         *eth_dev = rte_eth_dev_allocate(name);
 828         if (*eth_dev == NULL)
 829                 goto error;
 830
 831         /* check length of device name */
 832         if ((strlen((*eth_dev)->data->name) + 1) > sizeof(data->name))
 833                 goto error;
 834
 835         /* now put it all together
 836          * - store queue data in internals,
 837          * - store numa_node info in eth_dev
 838          * - point eth_dev_data to internals
 839          * - and point eth_dev structure to new eth_dev_data structure
 840          */
 841         /* NOTE: we'll replace the data element, of originally allocated eth_dev
 842          * so the rings are local per-process */
 843
 844         if (pair == NULL)
 845                 (*internals)->if_index = 0;
 846         else
 847                 (*internals)->if_index = if_nametoindex(pair->value);
 848
 849         data->dev_private = *internals;
 850         data->port_id = (*eth_dev)->data->port_id;
 851         snprintf(data->name, sizeof(data->name), "%s", (*eth_dev)->data->name);
 852         data->nb_rx_queues = (uint16_t)nb_rx_queues;
 853         data->nb_tx_queues = (uint16_t)nb_tx_queues;
 854         data->dev_link = pmd_link;
 855         data->mac_addrs = &eth_addr;
 856         strncpy(data->name,
 857                 (*eth_dev)->data->name, strlen((*eth_dev)->data->name));
 858
 859         (*eth_dev)->data = data;
 860         (*eth_dev)->dev_ops = &ops;
 861         (*eth_dev)->driver = NULL;
 862         data->dev_flags = RTE_ETH_DEV_DETACHABLE;
 863         data->kdrv = RTE_KDRV_NONE;
 864         data->drv_name = drivername;
 865         data->numa_node = numa_node;
 866
 867         return 0;
 868
 869 error:
 870         rte_free(data);
 871         rte_free(*internals);
 872
 873         return -1;
 874 }
 875
 876 static int
 877 rte_eth_from_pcaps_common(const char *name, struct rx_pcaps *rx_queues,
 878                 const unsigned nb_rx_queues, struct tx_pcaps *tx_queues,
 879                 const unsigned nb_tx_queues, const unsigned numa_node,
 880                 struct rte_kvargs *kvlist, struct pmd_internals **internals,
 881                 struct rte_eth_dev **eth_dev)
 882 {
 883         unsigned i;
 884
 885         /* do some parameter checking */
 886         if (rx_queues == NULL && nb_rx_queues > 0)
 887                 return -1;
 888         if (tx_queues == NULL && nb_tx_queues > 0)
 889                 return -1;
 890
 891         if (rte_pmd_init_internals(name, nb_rx_queues, nb_tx_queues, numa_node,
 892                         internals, eth_dev, kvlist) < 0)
 893                 return -1;
 894
 895         for (i = 0; i < nb_rx_queues; i++) {
 896                 (*internals)->rx_queue[i].pcap = rx_queues->pcaps[i];
 897                 snprintf((*internals)->rx_queue[i].name,
 898                         sizeof((*internals)->rx_queue[i].name), "%s",
 899                         rx_queues->names[i]);
 900                 snprintf((*internals)->rx_queue[i].type,
 901                         sizeof((*internals)->rx_queue[i].type), "%s",
 902                         rx_queues->types[i]);
 903         }
 904         for (i = 0; i < nb_tx_queues; i++) {
 905                 (*internals)->tx_queue[i].dumper = tx_queues->dumpers[i];
 906                 snprintf((*internals)->tx_queue[i].name,
 907                         sizeof((*internals)->tx_queue[i].name), "%s",
 908                         tx_queues->names[i]);
 909                 snprintf((*internals)->tx_queue[i].type,
 910                         sizeof((*internals)->tx_queue[i].type), "%s",
 911                         tx_queues->types[i]);
 912         }
 913
 914         return 0;
 915 }
 916
 917 static int
 918 rte_eth_from_pcaps_n_dumpers(const char *name,
 919                 struct rx_pcaps *rx_queues,
 920                 const unsigned nb_rx_queues,
 921                 struct tx_pcaps *tx_queues,
 922                 const unsigned nb_tx_queues,
 923                 const unsigned numa_node,
 924                 struct rte_kvargs *kvlist)
 925 {
 926         struct pmd_internals *internals = NULL;
 927         struct rte_eth_dev *eth_dev = NULL;
 928         int ret;
 929
 930         ret = rte_eth_from_pcaps_common(name, rx_queues, nb_rx_queues,
 931                         tx_queues, nb_tx_queues, numa_node, kvlist,
 932                         &internals, &eth_dev);
 933
 934         if (ret < 0)
 935                 return ret;
 936
 937         /* using multiple pcaps/interfaces */
 938         internals->single_iface = 0;
 939
 940         eth_dev->rx_pkt_burst = eth_pcap_rx;
 941         eth_dev->tx_pkt_burst = eth_pcap_tx_dumper;
 942
 943         return 0;
 944 }
 945
 946 static int
 947 rte_eth_from_pcaps(const char *name,
 948                 struct rx_pcaps *rx_queues,
 949                 const unsigned nb_rx_queues,
 950                 struct tx_pcaps *tx_queues,
 951                 const unsigned nb_tx_queues,
 952                 const unsigned numa_node,
 953                 struct rte_kvargs *kvlist,
 954                 int single_iface)
 955 {
 956         struct pmd_internals *internals = NULL;
 957         struct rte_eth_dev *eth_dev = NULL;
 958         int ret;
 959
 960         ret = rte_eth_from_pcaps_common(name, rx_queues, nb_rx_queues,
 961                         tx_queues, nb_tx_queues, numa_node, kvlist,
 962                         &internals, &eth_dev);
 963
 964         if (ret < 0)
 965                 return ret;
 966
 967         /* store wether we are using a single interface for rx/tx or not */
 968         internals->single_iface = single_iface;
 969
 970         eth_dev->rx_pkt_burst = eth_pcap_rx;
 971         eth_dev->tx_pkt_burst = eth_pcap_tx;
 972
 973         return 0;
 974 }
 975
 976
 977 static int
 978 rte_pmd_pcap_devinit(const char *name, const char *params)
 979 {
 980         unsigned numa_node, using_dumpers = 0;
 981         int ret;
 982         struct rte_kvargs *kvlist;
 983         struct rx_pcaps pcaps = {0};
 984         struct tx_pcaps dumpers = {0};
 985
 986         RTE_LOG(INFO, PMD, "Initializing pmd_pcap for %s\n", name);
 987
 988         numa_node = rte_socket_id();
 989
 990         gettimeofday(&start_time, NULL);
 991         start_cycles = rte_get_timer_cycles();
 992         hz = rte_get_timer_hz();
 993
 994         kvlist = rte_kvargs_parse(params, valid_arguments);
 995         if (kvlist == NULL)
 996                 return -1;
 997
 998         /*
 999          * If iface argument is passed we open the NICs and use them for
1000          * reading / writing
1001          */
1002         if (rte_kvargs_count(kvlist, ETH_PCAP_IFACE_ARG) == 1) {
1003
1004                 ret = rte_kvargs_process(kvlist, ETH_PCAP_IFACE_ARG,
1005                                 &open_rx_tx_iface, &pcaps);
1006                 if (ret < 0)
1007                         goto free_kvlist;
1008                 dumpers.pcaps[0] = pcaps.pcaps[0];
1009                 dumpers.names[0] = pcaps.names[0];
1010                 dumpers.types[0] = pcaps.types[0];
1011                 ret = rte_eth_from_pcaps(name, &pcaps, 1, &dumpers, 1,
1012                                 numa_node, kvlist, 1);
1013                 goto free_kvlist;
1014         }
1015
1016         /*
1017          * We check whether we want to open a RX stream from a real NIC or a
1018          * pcap file
1019          */
1020         if ((pcaps.num_of_rx = rte_kvargs_count(kvlist, ETH_PCAP_RX_PCAP_ARG))) {
1021                 ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_PCAP_ARG,
1022                                 &open_rx_pcap, &pcaps);
1023         } else {
1024                 pcaps.num_of_rx = rte_kvargs_count(kvlist,
1025                                 ETH_PCAP_RX_IFACE_ARG);
1026                 ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_IFACE_ARG,
1027                                 &open_rx_iface, &pcaps);
1028         }
1029
1030         if (ret < 0)
1031                 goto free_kvlist;
1032
1033         /*
1034          * We check whether we want to open a TX stream to a real NIC or a
1035          * pcap file
1036          */
1037         if ((dumpers.num_of_tx = rte_kvargs_count(kvlist,
1038                         ETH_PCAP_TX_PCAP_ARG))) {
1039                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_PCAP_ARG,
1040                                 &open_tx_pcap, &dumpers);
1041                 using_dumpers = 1;
1042         } else {
1043                 dumpers.num_of_tx = rte_kvargs_count(kvlist,
1044                                 ETH_PCAP_TX_IFACE_ARG);
1045                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_IFACE_ARG,
1046                                 &open_tx_iface, &dumpers);
1047         }
1048
1049         if (ret < 0)
1050                 goto free_kvlist;
1051
1052         if (using_dumpers)
1053                 ret = rte_eth_from_pcaps_n_dumpers(name, &pcaps, pcaps.num_of_rx,
1054                                 &dumpers, dumpers.num_of_tx, numa_node, kvlist);
1055         else
1056                 ret = rte_eth_from_pcaps(name, &pcaps, pcaps.num_of_rx, &dumpers,
1057                         dumpers.num_of_tx, numa_node, kvlist, 0);
1058
1059 free_kvlist:
1060         rte_kvargs_free(kvlist);
1061         return ret;
1062 }
1063
1064 static int
1065 rte_pmd_pcap_devuninit(const char *name)
1066 {
1067         struct rte_eth_dev *eth_dev = NULL;
1068
1069         RTE_LOG(INFO, PMD, "Closing pcap ethdev on numa socket %u\n",
1070                         rte_socket_id());
1071
1072         if (name == NULL)
1073                 return -1;
1074
1075         /* reserve an ethdev entry */
1076         eth_dev = rte_eth_dev_allocated(name);
1077         if (eth_dev == NULL)
1078                 return -1;
1079
1080         rte_free(eth_dev->data->dev_private);
1081         rte_free(eth_dev->data);
1082
1083         rte_eth_dev_release_port(eth_dev);
1084
1085         return 0;
1086 }
1087
1088 static struct rte_vdev_driver pmd_pcap_drv = {
1089         .init = rte_pmd_pcap_devinit,
1090         .uninit = rte_pmd_pcap_devuninit,
1091 };
1092
1093 DRIVER_REGISTER_VDEV(net_pcap, pmd_pcap_drv);
1094 DRIVER_REGISTER_PARAM_STRING(net_pcap,
1095         ETH_PCAP_RX_PCAP_ARG "=<string> "
1096         ETH_PCAP_TX_PCAP_ARG "=<string> "
1097         ETH_PCAP_RX_IFACE_ARG "=<ifc> "
1098         ETH_PCAP_TX_IFACE_ARG "=<ifc> "
1099         ETH_PCAP_IFACE_ARG "=<ifc>");