drivers/net/pcap/rte_eth_pcap.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2015 Intel Corporation. All rights reserved.
   5  *   Copyright(c) 2014 6WIND S.A.
   6  *   All rights reserved.
   7  *
   8  *   Redistribution and use in source and binary forms, with or without
   9  *   modification, are permitted provided that the following conditions
  10  *   are met:
  11  *
  12  *     * Redistributions of source code must retain the above copyright
  13  *       notice, this list of conditions and the following disclaimer.
  14  *     * Redistributions in binary form must reproduce the above copyright
  15  *       notice, this list of conditions and the following disclaimer in
  16  *       the documentation and/or other materials provided with the
  17  *       distribution.
  18  *     * Neither the name of Intel Corporation nor the names of its
  19  *       contributors may be used to endorse or promote products derived
  20  *       from this software without specific prior written permission.
  21  *
  22  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  23  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  24  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  25  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  26  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  27  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  28  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  29  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  30  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  31  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  32  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  33  */
  34
  35 #include <time.h>
  36 #include <rte_mbuf.h>
  37 #include <rte_ethdev.h>
  38 #include <rte_malloc.h>
  39 #include <rte_memcpy.h>
  40 #include <rte_string_fns.h>
  41 #include <rte_cycles.h>
  42 #include <rte_kvargs.h>
  43 #include <rte_dev.h>
  44
  45 #include <net/if.h>
  46
  47 #include <pcap.h>
  48
  49 #define RTE_ETH_PCAP_SNAPSHOT_LEN 65535
  50 #define RTE_ETH_PCAP_SNAPLEN ETHER_MAX_JUMBO_FRAME_LEN
  51 #define RTE_ETH_PCAP_PROMISC 1
  52 #define RTE_ETH_PCAP_TIMEOUT -1
  53 #define ETH_PCAP_RX_PCAP_ARG  "rx_pcap"
  54 #define ETH_PCAP_TX_PCAP_ARG  "tx_pcap"
  55 #define ETH_PCAP_RX_IFACE_ARG "rx_iface"
  56 #define ETH_PCAP_TX_IFACE_ARG "tx_iface"
  57 #define ETH_PCAP_IFACE_ARG    "iface"
  58
  59 #define ETH_PCAP_ARG_MAXLEN     64
  60
  61 static char errbuf[PCAP_ERRBUF_SIZE];
  62 static unsigned char tx_pcap_data[RTE_ETH_PCAP_SNAPLEN];
  63 static struct timeval start_time;
  64 static uint64_t start_cycles;
  65 static uint64_t hz;
  66
  67 struct pcap_rx_queue {
  68         pcap_t *pcap;
  69         uint8_t in_port;
  70         struct rte_mempool *mb_pool;
  71         volatile unsigned long rx_pkts;
  72         volatile unsigned long rx_bytes;
  73         volatile unsigned long err_pkts;
  74         char name[PATH_MAX];
  75         char type[ETH_PCAP_ARG_MAXLEN];
  76 };
  77
  78 struct pcap_tx_queue {
  79         pcap_dumper_t *dumper;
  80         pcap_t *pcap;
  81         volatile unsigned long tx_pkts;
  82         volatile unsigned long tx_bytes;
  83         volatile unsigned long err_pkts;
  84         char name[PATH_MAX];
  85         char type[ETH_PCAP_ARG_MAXLEN];
  86 };
  87
  88 struct rx_pcaps {
  89         unsigned num_of_rx;
  90         pcap_t *pcaps[RTE_PMD_RING_MAX_RX_RINGS];
  91         const char *names[RTE_PMD_RING_MAX_RX_RINGS];
  92         const char *types[RTE_PMD_RING_MAX_RX_RINGS];
  93 };
  94
  95 struct tx_pcaps {
  96         unsigned num_of_tx;
  97         pcap_dumper_t *dumpers[RTE_PMD_RING_MAX_TX_RINGS];
  98         pcap_t *pcaps[RTE_PMD_RING_MAX_RX_RINGS];
  99         const char *names[RTE_PMD_RING_MAX_RX_RINGS];
 100         const char *types[RTE_PMD_RING_MAX_RX_RINGS];
 101 };
 102
 103 struct pmd_internals {
 104         struct pcap_rx_queue rx_queue[RTE_PMD_RING_MAX_RX_RINGS];
 105         struct pcap_tx_queue tx_queue[RTE_PMD_RING_MAX_TX_RINGS];
 106         unsigned nb_rx_queues;
 107         unsigned nb_tx_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 = 10000,
 129                 .link_duplex = ETH_LINK_FULL_DUPLEX,
 130                 .link_status = 0
 131 };
 132
 133 static int
 134 eth_pcap_rx_jumbo(struct rte_mempool *mb_pool,
 135                   struct rte_mbuf *mbuf,
 136                   const u_char *data,
 137                   uint16_t data_len)
 138 {
 139         struct rte_mbuf *m = mbuf;
 140
 141         /* Copy the first segment. */
 142         uint16_t len = rte_pktmbuf_tailroom(mbuf);
 143
 144         rte_memcpy(rte_pktmbuf_append(mbuf, len), data, len);
 145         data_len -= len;
 146         data += len;
 147
 148         while (data_len > 0) {
 149                 /* Allocate next mbuf and point to that. */
 150                 m->next = rte_pktmbuf_alloc(mb_pool);
 151
 152                 if (unlikely(!m->next))
 153                         return -1;
 154
 155                 m = m->next;
 156
 157                 /* Headroom is not needed in chained mbufs. */
 158                 rte_pktmbuf_prepend(m, rte_pktmbuf_headroom(m));
 159                 m->pkt_len = 0;
 160                 m->data_len = 0;
 161
 162                 /* Copy next segment. */
 163                 len = RTE_MIN(rte_pktmbuf_tailroom(m), data_len);
 164                 rte_memcpy(rte_pktmbuf_append(m, len), data, len);
 165
 166                 mbuf->nb_segs++;
 167                 data_len -= len;
 168                 data += len;
 169         }
 170
 171         return mbuf->nb_segs;
 172 }
 173
 174 /* Copy data from mbuf chain to a buffer suitable for writing to a PCAP file. */
 175 static void
 176 eth_pcap_gather_data(unsigned char *data, struct rte_mbuf *mbuf)
 177 {
 178         uint16_t data_len = 0;
 179
 180         while (mbuf) {
 181                 rte_memcpy(data + data_len, rte_pktmbuf_mtod(mbuf, void *),
 182                            mbuf->data_len);
 183
 184                 data_len += mbuf->data_len;
 185                 mbuf = mbuf->next;
 186         }
 187 }
 188
 189 static uint16_t
 190 eth_pcap_rx(void *queue,
 191                 struct rte_mbuf **bufs,
 192                 uint16_t nb_pkts)
 193 {
 194         unsigned i;
 195         struct pcap_pkthdr header;
 196         const u_char *packet;
 197         struct rte_mbuf *mbuf;
 198         struct pcap_rx_queue *pcap_q = queue;
 199         uint16_t num_rx = 0;
 200         uint16_t buf_size;
 201         uint32_t rx_bytes = 0;
 202
 203         if (unlikely(pcap_q->pcap == NULL || nb_pkts == 0))
 204                 return 0;
 205
 206         /* Reads the given number of packets from the pcap file one by one
 207          * and copies the packet data into a newly allocated mbuf to return.
 208          */
 209         for (i = 0; i < nb_pkts; i++) {
 210                 /* Get the next PCAP packet */
 211                 packet = pcap_next(pcap_q->pcap, &header);
 212                 if (unlikely(packet == NULL))
 213                         break;
 214                 else
 215                         mbuf = rte_pktmbuf_alloc(pcap_q->mb_pool);
 216                 if (unlikely(mbuf == NULL))
 217                         break;
 218
 219                 /* Now get the space available for data in the mbuf */
 220                 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(pcap_q->mb_pool) -
 221                                 RTE_PKTMBUF_HEADROOM);
 222
 223                 if (header.len <= buf_size) {
 224                         /* pcap packet will fit in the mbuf, go ahead and copy */
 225                         rte_memcpy(rte_pktmbuf_mtod(mbuf, void *), packet,
 226                                         header.len);
 227                         mbuf->data_len = (uint16_t)header.len;
 228                 } else {
 229                         /* Try read jumbo frame into multi mbufs. */
 230                         if (unlikely(eth_pcap_rx_jumbo(pcap_q->mb_pool,
 231                                                        mbuf,
 232                                                        packet,
 233                                                        header.len) == -1))
 234                                 break;
 235                 }
 236
 237                 mbuf->pkt_len = (uint16_t)header.len;
 238                 mbuf->port = pcap_q->in_port;
 239                 bufs[num_rx] = mbuf;
 240                 num_rx++;
 241                 rx_bytes += header.len;
 242         }
 243         pcap_q->rx_pkts += num_rx;
 244         pcap_q->rx_bytes += rx_bytes;
 245         return num_rx;
 246 }
 247
 248 static inline void
 249 calculate_timestamp(struct timeval *ts) {
 250         uint64_t cycles;
 251         struct timeval cur_time;
 252
 253         cycles = rte_get_timer_cycles() - start_cycles;
 254         cur_time.tv_sec = cycles / hz;
 255         cur_time.tv_usec = (cycles % hz) * 10e6 / hz;
 256         timeradd(&start_time, &cur_time, ts);
 257 }
 258
 259 /*
 260  * Callback to handle writing packets to a pcap file.
 261  */
 262 static uint16_t
 263 eth_pcap_tx_dumper(void *queue,
 264                 struct rte_mbuf **bufs,
 265                 uint16_t nb_pkts)
 266 {
 267         unsigned i;
 268         struct rte_mbuf *mbuf;
 269         struct pcap_tx_queue *dumper_q = queue;
 270         uint16_t num_tx = 0;
 271         uint32_t tx_bytes = 0;
 272         struct pcap_pkthdr header;
 273
 274         if (dumper_q->dumper == NULL || nb_pkts == 0)
 275                 return 0;
 276
 277         /* writes the nb_pkts packets to the previously opened pcap file dumper */
 278         for (i = 0; i < nb_pkts; i++) {
 279                 mbuf = bufs[i];
 280                 calculate_timestamp(&header.ts);
 281                 header.len = mbuf->pkt_len;
 282                 header.caplen = header.len;
 283
 284                 if (likely(mbuf->nb_segs == 1)) {
 285                         pcap_dump((u_char *)dumper_q->dumper, &header,
 286                                   rte_pktmbuf_mtod(mbuf, void*));
 287                 } else {
 288                         if (mbuf->pkt_len <= ETHER_MAX_JUMBO_FRAME_LEN) {
 289                                 eth_pcap_gather_data(tx_pcap_data, mbuf);
 290                                 pcap_dump((u_char *)dumper_q->dumper, &header,
 291                                           tx_pcap_data);
 292                         } else {
 293                                 RTE_LOG(ERR, PMD,
 294                                         "Dropping PCAP packet. "
 295                                         "Size (%d) > max jumbo size (%d).\n",
 296                                         mbuf->pkt_len,
 297                                         ETHER_MAX_JUMBO_FRAME_LEN);
 298
 299                                 rte_pktmbuf_free(mbuf);
 300                                 break;
 301                         }
 302                 }
 303
 304                 rte_pktmbuf_free(mbuf);
 305                 num_tx++;
 306                 tx_bytes += mbuf->pkt_len;
 307         }
 308
 309         /*
 310          * Since there's no place to hook a callback when the forwarding
 311          * process stops and to make sure the pcap file is actually written,
 312          * we flush the pcap dumper within each burst.
 313          */
 314         pcap_dump_flush(dumper_q->dumper);
 315         dumper_q->tx_pkts += num_tx;
 316         dumper_q->tx_bytes += tx_bytes;
 317         dumper_q->err_pkts += nb_pkts - num_tx;
 318         return num_tx;
 319 }
 320
 321 /*
 322  * Callback to handle sending packets through a real NIC.
 323  */
 324 static uint16_t
 325 eth_pcap_tx(void *queue,
 326                 struct rte_mbuf **bufs,
 327                 uint16_t nb_pkts)
 328 {
 329         unsigned i;
 330         int ret;
 331         struct rte_mbuf *mbuf;
 332         struct pcap_tx_queue *tx_queue = queue;
 333         uint16_t num_tx = 0;
 334         uint32_t tx_bytes = 0;
 335
 336         if (unlikely(nb_pkts == 0 || tx_queue->pcap == NULL))
 337                 return 0;
 338
 339         for (i = 0; i < nb_pkts; i++) {
 340                 mbuf = bufs[i];
 341
 342                 if (likely(mbuf->nb_segs == 1)) {
 343                         ret = pcap_sendpacket(tx_queue->pcap,
 344                                               rte_pktmbuf_mtod(mbuf, u_char *),
 345                                               mbuf->pkt_len);
 346                 } else {
 347                         if (mbuf->pkt_len <= ETHER_MAX_JUMBO_FRAME_LEN) {
 348                                 eth_pcap_gather_data(tx_pcap_data, mbuf);
 349                                 ret = pcap_sendpacket(tx_queue->pcap,
 350                                                       tx_pcap_data,
 351                                                       mbuf->pkt_len);
 352                         } else {
 353                                 RTE_LOG(ERR, PMD,
 354                                         "Dropping PCAP packet. "
 355                                         "Size (%d) > max jumbo size (%d).\n",
 356                                         mbuf->pkt_len,
 357                                         ETHER_MAX_JUMBO_FRAME_LEN);
 358
 359                                 rte_pktmbuf_free(mbuf);
 360                                 break;
 361                         }
 362                 }
 363
 364                 if (unlikely(ret != 0))
 365                         break;
 366                 num_tx++;
 367                 tx_bytes += mbuf->pkt_len;
 368                 rte_pktmbuf_free(mbuf);
 369         }
 370
 371         tx_queue->tx_pkts += num_tx;
 372         tx_queue->tx_bytes += tx_bytes;
 373         tx_queue->err_pkts += nb_pkts - num_tx;
 374         return num_tx;
 375 }
 376
 377 static int
 378 eth_dev_start(struct rte_eth_dev *dev)
 379 {
 380         unsigned i;
 381         struct pmd_internals *internals = dev->data->dev_private;
 382         struct pcap_tx_queue *tx;
 383         struct pcap_rx_queue *rx;
 384
 385         /* Special iface case. Single pcap is open and shared between tx/rx. */
 386         if (internals->single_iface) {
 387                 tx = &internals->tx_queue[0];
 388                 rx = &internals->rx_queue[0];
 389
 390                 if (!tx->pcap && strcmp(tx->type, ETH_PCAP_IFACE_ARG) == 0) {
 391                         if (open_single_iface(tx->name, &tx->pcap) < 0)
 392                                 return -1;
 393                         rx->pcap = tx->pcap;
 394                 }
 395                 goto status_up;
 396         }
 397
 398         /* If not open already, open tx pcaps/dumpers */
 399         for (i = 0; i < internals->nb_tx_queues; i++) {
 400                 tx = &internals->tx_queue[i];
 401
 402                 if (!tx->dumper && strcmp(tx->type, ETH_PCAP_TX_PCAP_ARG) == 0) {
 403                         if (open_single_tx_pcap(tx->name, &tx->dumper) < 0)
 404                                 return -1;
 405                 }
 406
 407                 else if (!tx->pcap && strcmp(tx->type, ETH_PCAP_TX_IFACE_ARG) == 0) {
 408                         if (open_single_iface(tx->name, &tx->pcap) < 0)
 409                                 return -1;
 410                 }
 411         }
 412
 413         /* If not open already, open rx pcaps */
 414         for (i = 0; i < internals->nb_rx_queues; i++) {
 415                 rx = &internals->rx_queue[i];
 416
 417                 if (rx->pcap != NULL)
 418                         continue;
 419
 420                 if (strcmp(rx->type, ETH_PCAP_RX_PCAP_ARG) == 0) {
 421                         if (open_single_rx_pcap(rx->name, &rx->pcap) < 0)
 422                                 return -1;
 423                 }
 424
 425                 else if (strcmp(rx->type, ETH_PCAP_RX_IFACE_ARG) == 0) {
 426                         if (open_single_iface(rx->name, &rx->pcap) < 0)
 427                                 return -1;
 428                 }
 429         }
 430
 431 status_up:
 432
 433         dev->data->dev_link.link_status = 1;
 434         return 0;
 435 }
 436
 437 /*
 438  * This function gets called when the current port gets stopped.
 439  * Is the only place for us to close all the tx streams dumpers.
 440  * If not called the dumpers will be flushed within each tx burst.
 441  */
 442 static void
 443 eth_dev_stop(struct rte_eth_dev *dev)
 444 {
 445         unsigned i;
 446         struct pmd_internals *internals = dev->data->dev_private;
 447         struct pcap_tx_queue *tx;
 448         struct pcap_rx_queue *rx;
 449
 450         /* Special iface case. Single pcap is open and shared between tx/rx. */
 451         if (internals->single_iface) {
 452                 tx = &internals->tx_queue[0];
 453                 rx = &internals->rx_queue[0];
 454                 pcap_close(tx->pcap);
 455                 tx->pcap = NULL;
 456                 rx->pcap = NULL;
 457                 goto status_down;
 458         }
 459
 460         for (i = 0; i < internals->nb_tx_queues; i++) {
 461                 tx = &internals->tx_queue[i];
 462
 463                 if (tx->dumper != NULL) {
 464                         pcap_dump_close(tx->dumper);
 465                         tx->dumper = NULL;
 466                 }
 467
 468                 if (tx->pcap != NULL) {
 469                         pcap_close(tx->pcap);
 470                         tx->pcap = NULL;
 471                 }
 472         }
 473
 474         for (i = 0; i < internals->nb_rx_queues; i++) {
 475                 rx = &internals->rx_queue[i];
 476
 477                 if (rx->pcap != NULL) {
 478                         pcap_close(rx->pcap);
 479                         rx->pcap = NULL;
 480                 }
 481         }
 482
 483 status_down:
 484         dev->data->dev_link.link_status = 0;
 485 }
 486
 487 static int
 488 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 489 {
 490         return 0;
 491 }
 492
 493 static void
 494 eth_dev_info(struct rte_eth_dev *dev,
 495                 struct rte_eth_dev_info *dev_info)
 496 {
 497         struct pmd_internals *internals = dev->data->dev_private;
 498         dev_info->driver_name = drivername;
 499         dev_info->if_index = internals->if_index;
 500         dev_info->max_mac_addrs = 1;
 501         dev_info->max_rx_pktlen = (uint32_t) -1;
 502         dev_info->max_rx_queues = (uint16_t)internals->nb_rx_queues;
 503         dev_info->max_tx_queues = (uint16_t)internals->nb_tx_queues;
 504         dev_info->min_rx_bufsize = 0;
 505         dev_info->pci_dev = NULL;
 506 }
 507
 508 static void
 509 eth_stats_get(struct rte_eth_dev *dev,
 510                 struct rte_eth_stats *igb_stats)
 511 {
 512         unsigned i;
 513         unsigned long rx_packets_total = 0, rx_bytes_total = 0;
 514         unsigned long tx_packets_total = 0, tx_bytes_total = 0;
 515         unsigned long tx_packets_err_total = 0;
 516         const struct pmd_internals *internal = dev->data->dev_private;
 517
 518         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS && i < internal->nb_rx_queues;
 519                         i++) {
 520                 igb_stats->q_ipackets[i] = internal->rx_queue[i].rx_pkts;
 521                 igb_stats->q_ibytes[i] = internal->rx_queue[i].rx_bytes;
 522                 rx_packets_total += igb_stats->q_ipackets[i];
 523                 rx_bytes_total += igb_stats->q_ibytes[i];
 524         }
 525
 526         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS && i < internal->nb_tx_queues;
 527                         i++) {
 528                 igb_stats->q_opackets[i] = internal->tx_queue[i].tx_pkts;
 529                 igb_stats->q_obytes[i] = internal->tx_queue[i].tx_bytes;
 530                 igb_stats->q_errors[i] = internal->tx_queue[i].err_pkts;
 531                 tx_packets_total += igb_stats->q_opackets[i];
 532                 tx_bytes_total += igb_stats->q_obytes[i];
 533                 tx_packets_err_total += igb_stats->q_errors[i];
 534         }
 535
 536         igb_stats->ipackets = rx_packets_total;
 537         igb_stats->ibytes = rx_bytes_total;
 538         igb_stats->opackets = tx_packets_total;
 539         igb_stats->obytes = tx_bytes_total;
 540         igb_stats->oerrors = tx_packets_err_total;
 541 }
 542
 543 static void
 544 eth_stats_reset(struct rte_eth_dev *dev)
 545 {
 546         unsigned i;
 547         struct pmd_internals *internal = dev->data->dev_private;
 548         for (i = 0; i < internal->nb_rx_queues; i++) {
 549                 internal->rx_queue[i].rx_pkts = 0;
 550                 internal->rx_queue[i].rx_bytes = 0;
 551         }
 552         for (i = 0; i < internal->nb_tx_queues; i++) {
 553                 internal->tx_queue[i].tx_pkts = 0;
 554                 internal->tx_queue[i].tx_bytes = 0;
 555                 internal->tx_queue[i].err_pkts = 0;
 556         }
 557 }
 558
 559 static void
 560 eth_dev_close(struct rte_eth_dev *dev __rte_unused)
 561 {
 562 }
 563
 564 static void
 565 eth_queue_release(void *q __rte_unused)
 566 {
 567 }
 568
 569 static int
 570 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 571                 int wait_to_complete __rte_unused)
 572 {
 573         return 0;
 574 }
 575
 576 static int
 577 eth_rx_queue_setup(struct rte_eth_dev *dev,
 578                 uint16_t rx_queue_id,
 579                 uint16_t nb_rx_desc __rte_unused,
 580                 unsigned int socket_id __rte_unused,
 581                 const struct rte_eth_rxconf *rx_conf __rte_unused,
 582                 struct rte_mempool *mb_pool)
 583 {
 584         struct pmd_internals *internals = dev->data->dev_private;
 585         struct pcap_rx_queue *pcap_q = &internals->rx_queue[rx_queue_id];
 586         pcap_q->mb_pool = mb_pool;
 587         dev->data->rx_queues[rx_queue_id] = pcap_q;
 588         pcap_q->in_port = dev->data->port_id;
 589         return 0;
 590 }
 591
 592 static int
 593 eth_tx_queue_setup(struct rte_eth_dev *dev,
 594                 uint16_t tx_queue_id,
 595                 uint16_t nb_tx_desc __rte_unused,
 596                 unsigned int socket_id __rte_unused,
 597                 const struct rte_eth_txconf *tx_conf __rte_unused)
 598 {
 599
 600         struct pmd_internals *internals = dev->data->dev_private;
 601         dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
 602         return 0;
 603 }
 604
 605 static const struct eth_dev_ops ops = {
 606         .dev_start = eth_dev_start,
 607         .dev_stop =     eth_dev_stop,
 608         .dev_close = eth_dev_close,
 609         .dev_configure = eth_dev_configure,
 610         .dev_infos_get = eth_dev_info,
 611         .rx_queue_setup = eth_rx_queue_setup,
 612         .tx_queue_setup = eth_tx_queue_setup,
 613         .rx_queue_release = eth_queue_release,
 614         .tx_queue_release = eth_queue_release,
 615         .link_update = eth_link_update,
 616         .stats_get = eth_stats_get,
 617         .stats_reset = eth_stats_reset,
 618 };
 619
 620 static struct eth_driver rte_pcap_pmd = {
 621         .pci_drv = {
 622                 .name = "rte_pcap_pmd",
 623                 .drv_flags = RTE_PCI_DRV_DETACHABLE,
 624         },
 625 };
 626
 627 /*
 628  * Function handler that opens the pcap file for reading a stores a
 629  * reference of it for use it later on.
 630  */
 631 static int
 632 open_rx_pcap(const char *key, const char *value, void *extra_args)
 633 {
 634         unsigned i;
 635         const char *pcap_filename = value;
 636         struct rx_pcaps *pcaps = extra_args;
 637         pcap_t *pcap = NULL;
 638
 639         for (i = 0; i < pcaps->num_of_rx; i++) {
 640                 if (open_single_rx_pcap(pcap_filename, &pcap) < 0)
 641                         return -1;
 642
 643                 pcaps->pcaps[i] = pcap;
 644                 pcaps->names[i] = pcap_filename;
 645                 pcaps->types[i] = key;
 646         }
 647
 648         return 0;
 649 }
 650
 651 static int
 652 open_single_rx_pcap(const char *pcap_filename, pcap_t **pcap)
 653 {
 654         if ((*pcap = pcap_open_offline(pcap_filename, errbuf)) == NULL) {
 655                 RTE_LOG(ERR, PMD, "Couldn't open %s: %s\n", pcap_filename, errbuf);
 656                 return -1;
 657         }
 658         return 0;
 659 }
 660
 661 /*
 662  * Opens a pcap file for writing and stores a reference to it
 663  * for use it later on.
 664  */
 665 static int
 666 open_tx_pcap(const char *key, const char *value, void *extra_args)
 667 {
 668         unsigned i;
 669         const char *pcap_filename = value;
 670         struct tx_pcaps *dumpers = extra_args;
 671         pcap_dumper_t *dumper;
 672
 673         for (i = 0; i < dumpers->num_of_tx; i++) {
 674                 if (open_single_tx_pcap(pcap_filename, &dumper) < 0)
 675                         return -1;
 676
 677                 dumpers->dumpers[i] = dumper;
 678                 dumpers->names[i] = pcap_filename;
 679                 dumpers->types[i] = key;
 680         }
 681
 682         return 0;
 683 }
 684
 685 static int
 686 open_single_tx_pcap(const char *pcap_filename, pcap_dumper_t **dumper)
 687 {
 688         pcap_t *tx_pcap;
 689         /*
 690          * We need to create a dummy empty pcap_t to use it
 691          * with pcap_dump_open(). We create big enough an Ethernet
 692          * pcap holder.
 693          */
 694
 695         if ((tx_pcap = pcap_open_dead(DLT_EN10MB, RTE_ETH_PCAP_SNAPSHOT_LEN))
 696                         == NULL) {
 697                 RTE_LOG(ERR, PMD, "Couldn't create dead pcap\n");
 698                 return -1;
 699         }
 700
 701         /* The dumper is created using the previous pcap_t reference */
 702         if ((*dumper = pcap_dump_open(tx_pcap, pcap_filename)) == NULL) {
 703                 RTE_LOG(ERR, PMD, "Couldn't open %s for writing.\n", pcap_filename);
 704                 return -1;
 705         }
 706
 707         return 0;
 708 }
 709
 710 /*
 711  * pcap_open_live wrapper function
 712  */
 713 static inline int
 714 open_iface_live(const char *iface, pcap_t **pcap) {
 715         *pcap = pcap_open_live(iface, RTE_ETH_PCAP_SNAPLEN,
 716                         RTE_ETH_PCAP_PROMISC, RTE_ETH_PCAP_TIMEOUT, errbuf);
 717
 718         if (*pcap == NULL) {
 719                 RTE_LOG(ERR, PMD, "Couldn't open %s: %s\n", iface, errbuf);
 720                 return -1;
 721         }
 722         return 0;
 723 }
 724
 725 /*
 726  * Opens an interface for reading and writing
 727  */
 728 static inline int
 729 open_rx_tx_iface(const char *key, const char *value, void *extra_args)
 730 {
 731         const char *iface = value;
 732         struct rx_pcaps *pcaps = extra_args;
 733         pcap_t *pcap = NULL;
 734
 735         if (open_single_iface(iface, &pcap) < 0)
 736                 return -1;
 737
 738         pcaps->pcaps[0] = pcap;
 739         pcaps->names[0] = iface;
 740         pcaps->types[0] = key;
 741
 742         return 0;
 743 }
 744
 745 /*
 746  * Opens a NIC for reading packets from it
 747  */
 748 static inline int
 749 open_rx_iface(const char *key, const char *value, void *extra_args)
 750 {
 751         unsigned i;
 752         const char *iface = value;
 753         struct rx_pcaps *pcaps = extra_args;
 754         pcap_t *pcap = NULL;
 755
 756         for (i = 0; i < pcaps->num_of_rx; i++) {
 757                 if (open_single_iface(iface, &pcap) < 0)
 758                         return -1;
 759                 pcaps->pcaps[i] = pcap;
 760                 pcaps->names[i] = iface;
 761                 pcaps->types[i] = key;
 762         }
 763
 764         return 0;
 765 }
 766
 767 /*
 768  * Opens a NIC for writing packets to it
 769  */
 770 static int
 771 open_tx_iface(const char *key, const char *value, void *extra_args)
 772 {
 773         unsigned i;
 774         const char *iface = value;
 775         struct tx_pcaps *pcaps = extra_args;
 776         pcap_t *pcap;
 777
 778         for (i = 0; i < pcaps->num_of_tx; i++) {
 779                 if (open_single_iface(iface, &pcap) < 0)
 780                         return -1;
 781                 pcaps->pcaps[i] = pcap;
 782                 pcaps->names[i] = iface;
 783                 pcaps->types[i] = key;
 784         }
 785
 786         return 0;
 787 }
 788
 789 static int
 790 open_single_iface(const char *iface, pcap_t **pcap)
 791 {
 792         if (open_iface_live(iface, pcap) < 0) {
 793                 RTE_LOG(ERR, PMD, "Couldn't open interface %s\n", iface);
 794                 return -1;
 795         }
 796
 797         return 0;
 798 }
 799
 800 static int
 801 rte_pmd_init_internals(const char *name, const unsigned nb_rx_queues,
 802                 const unsigned nb_tx_queues,
 803                 const unsigned numa_node,
 804                 struct pmd_internals **internals,
 805                 struct rte_eth_dev **eth_dev,
 806                 struct rte_kvargs *kvlist)
 807 {
 808         struct rte_eth_dev_data *data = NULL;
 809         struct rte_pci_device *pci_dev = NULL;
 810         unsigned k_idx;
 811         struct rte_kvargs_pair *pair = NULL;
 812
 813         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 814                 pair = &kvlist->pairs[k_idx];
 815                 if (strstr(pair->key, ETH_PCAP_IFACE_ARG) != NULL)
 816                         break;
 817         }
 818
 819         RTE_LOG(INFO, PMD,
 820                         "Creating pcap-backed ethdev on numa socket %u\n", numa_node);
 821
 822         /* now do all data allocation - for eth_dev structure, dummy pci driver
 823          * and internal (private) data
 824          */
 825         data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
 826         if (data == NULL)
 827                 goto error;
 828
 829         pci_dev = rte_zmalloc_socket(name, sizeof(*pci_dev), 0, numa_node);
 830         if (pci_dev == NULL)
 831                 goto error;
 832
 833         *internals = rte_zmalloc_socket(name, sizeof(**internals), 0, numa_node);
 834         if (*internals == NULL)
 835                 goto error;
 836
 837         /* reserve an ethdev entry */
 838         *eth_dev = rte_eth_dev_allocate(name, RTE_ETH_DEV_VIRTUAL);
 839         if (*eth_dev == NULL)
 840                 goto error;
 841
 842         /* check length of device name */
 843         if ((strlen((*eth_dev)->data->name) + 1) > sizeof(data->name))
 844                 goto error;
 845
 846         /* now put it all together
 847          * - store queue data in internals,
 848          * - store numa_node info in eth_dev
 849          * - point eth_dev_data to internals
 850          * - and point eth_dev structure to new eth_dev_data structure
 851          */
 852         /* NOTE: we'll replace the data element, of originally allocated eth_dev
 853          * so the rings are local per-process */
 854
 855         (*internals)->nb_rx_queues = nb_rx_queues;
 856         (*internals)->nb_tx_queues = nb_tx_queues;
 857
 858         if (pair == NULL)
 859                 (*internals)->if_index = 0;
 860         else
 861                 (*internals)->if_index = if_nametoindex(pair->value);
 862
 863         pci_dev->numa_node = numa_node;
 864
 865         data->dev_private = *internals;
 866         data->port_id = (*eth_dev)->data->port_id;
 867         snprintf(data->name, sizeof(data->name), "%s", (*eth_dev)->data->name);
 868         data->nb_rx_queues = (uint16_t)nb_rx_queues;
 869         data->nb_tx_queues = (uint16_t)nb_tx_queues;
 870         data->dev_link = pmd_link;
 871         data->mac_addrs = &eth_addr;
 872         strncpy(data->name,
 873                 (*eth_dev)->data->name, strlen((*eth_dev)->data->name));
 874
 875         (*eth_dev)->data = data;
 876         (*eth_dev)->dev_ops = &ops;
 877         (*eth_dev)->pci_dev = pci_dev;
 878         (*eth_dev)->driver = &rte_pcap_pmd;
 879         (*eth_dev)->data->dev_flags = RTE_ETH_DEV_DETACHABLE;
 880         (*eth_dev)->driver = NULL;
 881         (*eth_dev)->data->kdrv = RTE_KDRV_NONE;
 882         (*eth_dev)->data->drv_name = drivername;
 883         (*eth_dev)->data->numa_node = numa_node;
 884
 885         return 0;
 886
 887 error:
 888         rte_free(data);
 889         rte_free(pci_dev);
 890         rte_free(*internals);
 891
 892         return -1;
 893 }
 894
 895 static int
 896 rte_eth_from_pcaps_n_dumpers(const char *name,
 897                 struct rx_pcaps *rx_queues,
 898                 const unsigned nb_rx_queues,
 899                 struct tx_pcaps *tx_queues,
 900                 const unsigned nb_tx_queues,
 901                 const unsigned numa_node,
 902                 struct rte_kvargs *kvlist)
 903 {
 904         struct pmd_internals *internals = NULL;
 905         struct rte_eth_dev *eth_dev = NULL;
 906         unsigned i;
 907
 908         /* do some parameter checking */
 909         if (rx_queues == NULL && nb_rx_queues > 0)
 910                 return -1;
 911         if (tx_queues == NULL && nb_tx_queues > 0)
 912                 return -1;
 913
 914         if (rte_pmd_init_internals(name, nb_rx_queues, nb_tx_queues, numa_node,
 915                         &internals, &eth_dev, kvlist) < 0)
 916                 return -1;
 917
 918         for (i = 0; i < nb_rx_queues; i++) {
 919                 internals->rx_queue[i].pcap = rx_queues->pcaps[i];
 920                 snprintf(internals->rx_queue[i].name,
 921                         sizeof(internals->rx_queue[i].name), "%s",
 922                         rx_queues->names[i]);
 923                 snprintf(internals->rx_queue[i].type,
 924                         sizeof(internals->rx_queue[i].type), "%s",
 925                         rx_queues->types[i]);
 926         }
 927         for (i = 0; i < nb_tx_queues; i++) {
 928                 internals->tx_queue[i].dumper = tx_queues->dumpers[i];
 929                 snprintf(internals->tx_queue[i].name,
 930                         sizeof(internals->tx_queue[i].name), "%s",
 931                         tx_queues->names[i]);
 932                 snprintf(internals->tx_queue[i].type,
 933                         sizeof(internals->tx_queue[i].type), "%s",
 934                         tx_queues->types[i]);
 935         }
 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         unsigned i;
 959
 960         /* do some parameter checking */
 961         if (rx_queues == NULL && nb_rx_queues > 0)
 962                 return -1;
 963         if (tx_queues == NULL && nb_tx_queues > 0)
 964                 return -1;
 965
 966         if (rte_pmd_init_internals(name, nb_rx_queues, nb_tx_queues, numa_node,
 967                         &internals, &eth_dev, kvlist) < 0)
 968                 return -1;
 969
 970         for (i = 0; i < nb_rx_queues; i++) {
 971                 internals->rx_queue[i].pcap = rx_queues->pcaps[i];
 972                 snprintf(internals->rx_queue[i].name,
 973                         sizeof(internals->rx_queue[i].name), "%s",
 974                         rx_queues->names[i]);
 975                 snprintf(internals->rx_queue[i].type,
 976                         sizeof(internals->rx_queue[i].type), "%s",
 977                         rx_queues->types[i]);
 978         }
 979         for (i = 0; i < nb_tx_queues; i++) {
 980                 internals->tx_queue[i].dumper = tx_queues->dumpers[i];
 981                 snprintf(internals->tx_queue[i].name,
 982                         sizeof(internals->tx_queue[i].name), "%s",
 983                         tx_queues->names[i]);
 984                 snprintf(internals->tx_queue[i].type,
 985                         sizeof(internals->tx_queue[i].type), "%s",
 986                         tx_queues->types[i]);
 987         }
 988
 989         /* store wether we are using a single interface for rx/tx or not */
 990         internals->single_iface = single_iface;
 991
 992         eth_dev->rx_pkt_burst = eth_pcap_rx;
 993         eth_dev->tx_pkt_burst = eth_pcap_tx;
 994
 995         return 0;
 996 }
 997
 998
 999 static int
1000 rte_pmd_pcap_devinit(const char *name, const char *params)
1001 {
1002         unsigned numa_node, using_dumpers = 0;
1003         int ret;
1004         struct rte_kvargs *kvlist;
1005         struct rx_pcaps pcaps;
1006         struct tx_pcaps dumpers;
1007
1008         RTE_LOG(INFO, PMD, "Initializing pmd_pcap for %s\n", name);
1009
1010         numa_node = rte_socket_id();
1011
1012         gettimeofday(&start_time, NULL);
1013         start_cycles = rte_get_timer_cycles();
1014         hz = rte_get_timer_hz();
1015
1016         kvlist = rte_kvargs_parse(params, valid_arguments);
1017         if (kvlist == NULL)
1018                 return -1;
1019
1020         /*
1021          * If iface argument is passed we open the NICs and use them for
1022          * reading / writing
1023          */
1024         if (rte_kvargs_count(kvlist, ETH_PCAP_IFACE_ARG) == 1) {
1025
1026                 ret = rte_kvargs_process(kvlist, ETH_PCAP_IFACE_ARG,
1027                                 &open_rx_tx_iface, &pcaps);
1028                 if (ret < 0)
1029                         goto free_kvlist;
1030                 dumpers.pcaps[0] = pcaps.pcaps[0];
1031                 dumpers.names[0] = pcaps.names[0];
1032                 dumpers.types[0] = pcaps.types[0];
1033                 ret = rte_eth_from_pcaps(name, &pcaps, 1, &dumpers, 1,
1034                                 numa_node, kvlist, 1);
1035                 goto free_kvlist;
1036         }
1037
1038         /*
1039          * We check whether we want to open a RX stream from a real NIC or a
1040          * pcap file
1041          */
1042         if ((pcaps.num_of_rx = rte_kvargs_count(kvlist, ETH_PCAP_RX_PCAP_ARG))) {
1043                 ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_PCAP_ARG,
1044                                 &open_rx_pcap, &pcaps);
1045         } else {
1046                 pcaps.num_of_rx = rte_kvargs_count(kvlist,
1047                                 ETH_PCAP_RX_IFACE_ARG);
1048                 ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_IFACE_ARG,
1049                                 &open_rx_iface, &pcaps);
1050         }
1051
1052         if (ret < 0)
1053                 goto free_kvlist;
1054
1055         /*
1056          * We check whether we want to open a TX stream to a real NIC or a
1057          * pcap file
1058          */
1059         if ((dumpers.num_of_tx = rte_kvargs_count(kvlist,
1060                         ETH_PCAP_TX_PCAP_ARG))) {
1061                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_PCAP_ARG,
1062                                 &open_tx_pcap, &dumpers);
1063                 using_dumpers = 1;
1064         } else {
1065                 dumpers.num_of_tx = rte_kvargs_count(kvlist,
1066                                 ETH_PCAP_TX_IFACE_ARG);
1067                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_IFACE_ARG,
1068                                 &open_tx_iface, &dumpers);
1069         }
1070
1071         if (ret < 0)
1072                 goto free_kvlist;
1073
1074         if (using_dumpers)
1075                 ret = rte_eth_from_pcaps_n_dumpers(name, &pcaps, pcaps.num_of_rx,
1076                                 &dumpers, dumpers.num_of_tx, numa_node, kvlist);
1077         else
1078                 ret = rte_eth_from_pcaps(name, &pcaps, pcaps.num_of_rx, &dumpers,
1079                         dumpers.num_of_tx, numa_node, kvlist, 0);
1080
1081 free_kvlist:
1082         rte_kvargs_free(kvlist);
1083         return ret;
1084 }
1085
1086 static int
1087 rte_pmd_pcap_devuninit(const char *name)
1088 {
1089         struct rte_eth_dev *eth_dev = NULL;
1090
1091         RTE_LOG(INFO, PMD, "Closing pcap ethdev on numa socket %u\n",
1092                         rte_socket_id());
1093
1094         if (name == NULL)
1095                 return -1;
1096
1097         /* reserve an ethdev entry */
1098         eth_dev = rte_eth_dev_allocated(name);
1099         if (eth_dev == NULL)
1100                 return -1;
1101
1102         rte_free(eth_dev->data->dev_private);
1103         rte_free(eth_dev->data);
1104         rte_free(eth_dev->pci_dev);
1105
1106         rte_eth_dev_release_port(eth_dev);
1107
1108         return 0;
1109 }
1110
1111 static struct rte_driver pmd_pcap_drv = {
1112         .name = "eth_pcap",
1113         .type = PMD_VDEV,
1114         .init = rte_pmd_pcap_devinit,
1115         .uninit = rte_pmd_pcap_devuninit,
1116 };
1117
1118 PMD_REGISTER_DRIVER(pmd_pcap_drv);