drivers/net/pcap/rte_eth_pcap.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
   5  *   Copyright(c) 2014 6WIND S.A.
   6  *   All rights reserved.
   7  *
   8  *   Redistribution and use in source and binary forms, with or without
   9  *   modification, are permitted provided that the following conditions
  10  *   are met:
  11  *
  12  *     * Redistributions of source code must retain the above copyright
  13  *       notice, this list of conditions and the following disclaimer.
  14  *     * Redistributions in binary form must reproduce the above copyright
  15  *       notice, this list of conditions and the following disclaimer in
  16  *       the documentation and/or other materials provided with the
  17  *       distribution.
  18  *     * Neither the name of Intel Corporation nor the names of its
  19  *       contributors may be used to endorse or promote products derived
  20  *       from this software without specific prior written permission.
  21  *
  22  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  23  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  24  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  25  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  26  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  27  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  28  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  29  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  30  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  31  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  32  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  33  */
  34
  35 #include <time.h>
  36
  37 #include <net/if.h>
  38
  39 #include <pcap.h>
  40
  41 #include <rte_cycles.h>
  42 #include <rte_ethdev.h>
  43 #include <rte_kvargs.h>
  44 #include <rte_malloc.h>
  45 #include <rte_mbuf.h>
  46 #include <rte_vdev.h>
  47
  48 #define RTE_ETH_PCAP_SNAPSHOT_LEN 65535
  49 #define RTE_ETH_PCAP_SNAPLEN ETHER_MAX_JUMBO_FRAME_LEN
  50 #define RTE_ETH_PCAP_PROMISC 1
  51 #define RTE_ETH_PCAP_TIMEOUT -1
  52 #define ETH_PCAP_RX_PCAP_ARG  "rx_pcap"
  53 #define ETH_PCAP_TX_PCAP_ARG  "tx_pcap"
  54 #define ETH_PCAP_RX_IFACE_ARG "rx_iface"
  55 #define ETH_PCAP_TX_IFACE_ARG "tx_iface"
  56 #define ETH_PCAP_IFACE_ARG    "iface"
  57
  58 #define ETH_PCAP_ARG_MAXLEN     64
  59
  60 #define RTE_PMD_PCAP_MAX_QUEUES 16
  61
  62 static char errbuf[PCAP_ERRBUF_SIZE];
  63 static unsigned char tx_pcap_data[RTE_ETH_PCAP_SNAPLEN];
  64 static struct timeval start_time;
  65 static uint64_t start_cycles;
  66 static uint64_t hz;
  67
  68 struct queue_stat {
  69         volatile unsigned long pkts;
  70         volatile unsigned long bytes;
  71         volatile unsigned long err_pkts;
  72 };
  73
  74 struct pcap_rx_queue {
  75         pcap_t *pcap;
  76         uint8_t in_port;
  77         struct rte_mempool *mb_pool;
  78         struct queue_stat rx_stat;
  79         char name[PATH_MAX];
  80         char type[ETH_PCAP_ARG_MAXLEN];
  81 };
  82
  83 struct pcap_tx_queue {
  84         pcap_dumper_t *dumper;
  85         pcap_t *pcap;
  86         struct queue_stat tx_stat;
  87         char name[PATH_MAX];
  88         char type[ETH_PCAP_ARG_MAXLEN];
  89 };
  90
  91 struct pmd_internals {
  92         struct pcap_rx_queue rx_queue[RTE_PMD_PCAP_MAX_QUEUES];
  93         struct pcap_tx_queue tx_queue[RTE_PMD_PCAP_MAX_QUEUES];
  94         int if_index;
  95         int single_iface;
  96 };
  97
  98 struct pmd_devargs {
  99         unsigned num_of_queue;
 100         struct devargs_queue {
 101                 pcap_dumper_t *dumper;
 102                 pcap_t *pcap;
 103                 const char *name;
 104                 const char *type;
 105         } queue[RTE_PMD_PCAP_MAX_QUEUES];
 106 };
 107
 108 static const char *valid_arguments[] = {
 109         ETH_PCAP_RX_PCAP_ARG,
 110         ETH_PCAP_TX_PCAP_ARG,
 111         ETH_PCAP_RX_IFACE_ARG,
 112         ETH_PCAP_TX_IFACE_ARG,
 113         ETH_PCAP_IFACE_ARG,
 114         NULL
 115 };
 116
 117 static struct ether_addr eth_addr = { .addr_bytes = { 0, 0, 0, 0x1, 0x2, 0x3 } };
 118 static const char *drivername = "Pcap PMD";
 119 static struct rte_eth_link pmd_link = {
 120                 .link_speed = ETH_SPEED_NUM_10G,
 121                 .link_duplex = ETH_LINK_FULL_DUPLEX,
 122                 .link_status = ETH_LINK_DOWN,
 123                 .link_autoneg = ETH_LINK_SPEED_FIXED,
 124 };
 125
 126 static int
 127 eth_pcap_rx_jumbo(struct rte_mempool *mb_pool,
 128                   struct rte_mbuf *mbuf,
 129                   const u_char *data,
 130                   uint16_t data_len)
 131 {
 132         struct rte_mbuf *m = mbuf;
 133
 134         /* Copy the first segment. */
 135         uint16_t len = rte_pktmbuf_tailroom(mbuf);
 136
 137         rte_memcpy(rte_pktmbuf_append(mbuf, len), data, len);
 138         data_len -= len;
 139         data += len;
 140
 141         while (data_len > 0) {
 142                 /* Allocate next mbuf and point to that. */
 143                 m->next = rte_pktmbuf_alloc(mb_pool);
 144
 145                 if (unlikely(!m->next))
 146                         return -1;
 147
 148                 m = m->next;
 149
 150                 /* Headroom is not needed in chained mbufs. */
 151                 rte_pktmbuf_prepend(m, rte_pktmbuf_headroom(m));
 152                 m->pkt_len = 0;
 153                 m->data_len = 0;
 154
 155                 /* Copy next segment. */
 156                 len = RTE_MIN(rte_pktmbuf_tailroom(m), data_len);
 157                 rte_memcpy(rte_pktmbuf_append(m, len), data, len);
 158
 159                 mbuf->nb_segs++;
 160                 data_len -= len;
 161                 data += len;
 162         }
 163
 164         return mbuf->nb_segs;
 165 }
 166
 167 /* Copy data from mbuf chain to a buffer suitable for writing to a PCAP file. */
 168 static void
 169 eth_pcap_gather_data(unsigned char *data, struct rte_mbuf *mbuf)
 170 {
 171         uint16_t data_len = 0;
 172
 173         while (mbuf) {
 174                 rte_memcpy(data + data_len, rte_pktmbuf_mtod(mbuf, void *),
 175                            mbuf->data_len);
 176
 177                 data_len += mbuf->data_len;
 178                 mbuf = mbuf->next;
 179         }
 180 }
 181
 182 static uint16_t
 183 eth_pcap_rx(void *queue,
 184                 struct rte_mbuf **bufs,
 185                 uint16_t nb_pkts)
 186 {
 187         unsigned i;
 188         struct pcap_pkthdr header;
 189         const u_char *packet;
 190         struct rte_mbuf *mbuf;
 191         struct pcap_rx_queue *pcap_q = queue;
 192         uint16_t num_rx = 0;
 193         uint16_t buf_size;
 194         uint32_t rx_bytes = 0;
 195
 196         if (unlikely(pcap_q->pcap == NULL || nb_pkts == 0))
 197                 return 0;
 198
 199         /* Reads the given number of packets from the pcap file one by one
 200          * and copies the packet data into a newly allocated mbuf to return.
 201          */
 202         for (i = 0; i < nb_pkts; i++) {
 203                 /* Get the next PCAP packet */
 204                 packet = pcap_next(pcap_q->pcap, &header);
 205                 if (unlikely(packet == NULL))
 206                         break;
 207                 else
 208                         mbuf = rte_pktmbuf_alloc(pcap_q->mb_pool);
 209                 if (unlikely(mbuf == NULL))
 210                         break;
 211
 212                 /* Now get the space available for data in the mbuf */
 213                 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(pcap_q->mb_pool) -
 214                                 RTE_PKTMBUF_HEADROOM);
 215
 216                 if (header.caplen <= buf_size) {
 217                         /* pcap packet will fit in the mbuf, go ahead and copy */
 218                         rte_memcpy(rte_pktmbuf_mtod(mbuf, void *), packet,
 219                                         header.caplen);
 220                         mbuf->data_len = (uint16_t)header.caplen;
 221                 } else {
 222                         /* Try read jumbo frame into multi mbufs. */
 223                         if (unlikely(eth_pcap_rx_jumbo(pcap_q->mb_pool,
 224                                                        mbuf,
 225                                                        packet,
 226                                                        header.caplen) == -1))
 227                                 break;
 228                 }
 229
 230                 mbuf->pkt_len = (uint16_t)header.caplen;
 231                 mbuf->port = pcap_q->in_port;
 232                 bufs[num_rx] = mbuf;
 233                 num_rx++;
 234                 rx_bytes += header.caplen;
 235         }
 236         pcap_q->rx_stat.pkts += num_rx;
 237         pcap_q->rx_stat.bytes += rx_bytes;
 238         return num_rx;
 239 }
 240
 241 static inline void
 242 calculate_timestamp(struct timeval *ts) {
 243         uint64_t cycles;
 244         struct timeval cur_time;
 245
 246         cycles = rte_get_timer_cycles() - start_cycles;
 247         cur_time.tv_sec = cycles / hz;
 248         cur_time.tv_usec = (cycles % hz) * 10e6 / hz;
 249         timeradd(&start_time, &cur_time, ts);
 250 }
 251
 252 /*
 253  * Callback to handle writing packets to a pcap file.
 254  */
 255 static uint16_t
 256 eth_pcap_tx_dumper(void *queue,
 257                 struct rte_mbuf **bufs,
 258                 uint16_t nb_pkts)
 259 {
 260         unsigned i;
 261         struct rte_mbuf *mbuf;
 262         struct pcap_tx_queue *dumper_q = queue;
 263         uint16_t num_tx = 0;
 264         uint32_t tx_bytes = 0;
 265         struct pcap_pkthdr header;
 266
 267         if (dumper_q->dumper == NULL || nb_pkts == 0)
 268                 return 0;
 269
 270         /* writes the nb_pkts packets to the previously opened pcap file dumper */
 271         for (i = 0; i < nb_pkts; i++) {
 272                 mbuf = bufs[i];
 273                 calculate_timestamp(&header.ts);
 274                 header.len = mbuf->pkt_len;
 275                 header.caplen = header.len;
 276
 277                 if (likely(mbuf->nb_segs == 1)) {
 278                         pcap_dump((u_char *)dumper_q->dumper, &header,
 279                                   rte_pktmbuf_mtod(mbuf, void*));
 280                 } else {
 281                         if (mbuf->pkt_len <= ETHER_MAX_JUMBO_FRAME_LEN) {
 282                                 eth_pcap_gather_data(tx_pcap_data, mbuf);
 283                                 pcap_dump((u_char *)dumper_q->dumper, &header,
 284                                           tx_pcap_data);
 285                         } else {
 286                                 RTE_LOG(ERR, PMD,
 287                                         "Dropping PCAP packet. "
 288                                         "Size (%d) > max jumbo size (%d).\n",
 289                                         mbuf->pkt_len,
 290                                         ETHER_MAX_JUMBO_FRAME_LEN);
 291
 292                                 rte_pktmbuf_free(mbuf);
 293                                 break;
 294                         }
 295                 }
 296
 297                 rte_pktmbuf_free(mbuf);
 298                 num_tx++;
 299                 tx_bytes += mbuf->pkt_len;
 300         }
 301
 302         /*
 303          * Since there's no place to hook a callback when the forwarding
 304          * process stops and to make sure the pcap file is actually written,
 305          * we flush the pcap dumper within each burst.
 306          */
 307         pcap_dump_flush(dumper_q->dumper);
 308         dumper_q->tx_stat.pkts += num_tx;
 309         dumper_q->tx_stat.bytes += tx_bytes;
 310         dumper_q->tx_stat.err_pkts += nb_pkts - num_tx;
 311         return num_tx;
 312 }
 313
 314 /*
 315  * Callback to handle sending packets through a real NIC.
 316  */
 317 static uint16_t
 318 eth_pcap_tx(void *queue,
 319                 struct rte_mbuf **bufs,
 320                 uint16_t nb_pkts)
 321 {
 322         unsigned i;
 323         int ret;
 324         struct rte_mbuf *mbuf;
 325         struct pcap_tx_queue *tx_queue = queue;
 326         uint16_t num_tx = 0;
 327         uint32_t tx_bytes = 0;
 328
 329         if (unlikely(nb_pkts == 0 || tx_queue->pcap == NULL))
 330                 return 0;
 331
 332         for (i = 0; i < nb_pkts; i++) {
 333                 mbuf = bufs[i];
 334
 335                 if (likely(mbuf->nb_segs == 1)) {
 336                         ret = pcap_sendpacket(tx_queue->pcap,
 337                                               rte_pktmbuf_mtod(mbuf, u_char *),
 338                                               mbuf->pkt_len);
 339                 } else {
 340                         if (mbuf->pkt_len <= ETHER_MAX_JUMBO_FRAME_LEN) {
 341                                 eth_pcap_gather_data(tx_pcap_data, mbuf);
 342                                 ret = pcap_sendpacket(tx_queue->pcap,
 343                                                       tx_pcap_data,
 344                                                       mbuf->pkt_len);
 345                         } else {
 346                                 RTE_LOG(ERR, PMD,
 347                                         "Dropping PCAP packet. "
 348                                         "Size (%d) > max jumbo size (%d).\n",
 349                                         mbuf->pkt_len,
 350                                         ETHER_MAX_JUMBO_FRAME_LEN);
 351
 352                                 rte_pktmbuf_free(mbuf);
 353                                 break;
 354                         }
 355                 }
 356
 357                 if (unlikely(ret != 0))
 358                         break;
 359                 num_tx++;
 360                 tx_bytes += mbuf->pkt_len;
 361                 rte_pktmbuf_free(mbuf);
 362         }
 363
 364         tx_queue->tx_stat.pkts += num_tx;
 365         tx_queue->tx_stat.bytes += tx_bytes;
 366         tx_queue->tx_stat.err_pkts += nb_pkts - num_tx;
 367         return num_tx;
 368 }
 369
 370 /*
 371  * pcap_open_live wrapper function
 372  */
 373 static inline int
 374 open_iface_live(const char *iface, pcap_t **pcap) {
 375         *pcap = pcap_open_live(iface, RTE_ETH_PCAP_SNAPLEN,
 376                         RTE_ETH_PCAP_PROMISC, RTE_ETH_PCAP_TIMEOUT, errbuf);
 377
 378         if (*pcap == NULL) {
 379                 RTE_LOG(ERR, PMD, "Couldn't open %s: %s\n", iface, errbuf);
 380                 return -1;
 381         }
 382
 383         return 0;
 384 }
 385
 386 static int
 387 open_single_iface(const char *iface, pcap_t **pcap)
 388 {
 389         if (open_iface_live(iface, pcap) < 0) {
 390                 RTE_LOG(ERR, PMD, "Couldn't open interface %s\n", iface);
 391                 return -1;
 392         }
 393
 394         return 0;
 395 }
 396
 397 static int
 398 open_single_tx_pcap(const char *pcap_filename, pcap_dumper_t **dumper)
 399 {
 400         pcap_t *tx_pcap;
 401
 402         /*
 403          * We need to create a dummy empty pcap_t to use it
 404          * with pcap_dump_open(). We create big enough an Ethernet
 405          * pcap holder.
 406          */
 407         if ((tx_pcap = pcap_open_dead(DLT_EN10MB, RTE_ETH_PCAP_SNAPSHOT_LEN))
 408                         == NULL) {
 409                 RTE_LOG(ERR, PMD, "Couldn't create dead pcap\n");
 410                 return -1;
 411         }
 412
 413         /* The dumper is created using the previous pcap_t reference */
 414         if ((*dumper = pcap_dump_open(tx_pcap, pcap_filename)) == NULL) {
 415                 RTE_LOG(ERR, PMD, "Couldn't open %s for writing.\n",
 416                         pcap_filename);
 417                 return -1;
 418         }
 419
 420         return 0;
 421 }
 422
 423 static int
 424 open_single_rx_pcap(const char *pcap_filename, pcap_t **pcap)
 425 {
 426         if ((*pcap = pcap_open_offline(pcap_filename, errbuf)) == NULL) {
 427                 RTE_LOG(ERR, PMD, "Couldn't open %s: %s\n", pcap_filename, errbuf);
 428                 return -1;
 429         }
 430
 431         return 0;
 432 }
 433
 434 static int
 435 eth_dev_start(struct rte_eth_dev *dev)
 436 {
 437         unsigned i;
 438         struct pmd_internals *internals = dev->data->dev_private;
 439         struct pcap_tx_queue *tx;
 440         struct pcap_rx_queue *rx;
 441
 442         /* Special iface case. Single pcap is open and shared between tx/rx. */
 443         if (internals->single_iface) {
 444                 tx = &internals->tx_queue[0];
 445                 rx = &internals->rx_queue[0];
 446
 447                 if (!tx->pcap && strcmp(tx->type, ETH_PCAP_IFACE_ARG) == 0) {
 448                         if (open_single_iface(tx->name, &tx->pcap) < 0)
 449                                 return -1;
 450                         rx->pcap = tx->pcap;
 451                 }
 452                 goto status_up;
 453         }
 454
 455         /* If not open already, open tx pcaps/dumpers */
 456         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 457                 tx = &internals->tx_queue[i];
 458
 459                 if (!tx->dumper && strcmp(tx->type, ETH_PCAP_TX_PCAP_ARG) == 0) {
 460                         if (open_single_tx_pcap(tx->name, &tx->dumper) < 0)
 461                                 return -1;
 462                 }
 463
 464                 else if (!tx->pcap && strcmp(tx->type, ETH_PCAP_TX_IFACE_ARG) == 0) {
 465                         if (open_single_iface(tx->name, &tx->pcap) < 0)
 466                                 return -1;
 467                 }
 468         }
 469
 470         /* If not open already, open rx pcaps */
 471         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 472                 rx = &internals->rx_queue[i];
 473
 474                 if (rx->pcap != NULL)
 475                         continue;
 476
 477                 if (strcmp(rx->type, ETH_PCAP_RX_PCAP_ARG) == 0) {
 478                         if (open_single_rx_pcap(rx->name, &rx->pcap) < 0)
 479                                 return -1;
 480                 }
 481
 482                 else if (strcmp(rx->type, ETH_PCAP_RX_IFACE_ARG) == 0) {
 483                         if (open_single_iface(rx->name, &rx->pcap) < 0)
 484                                 return -1;
 485                 }
 486         }
 487
 488 status_up:
 489
 490         dev->data->dev_link.link_status = ETH_LINK_UP;
 491         return 0;
 492 }
 493
 494 /*
 495  * This function gets called when the current port gets stopped.
 496  * Is the only place for us to close all the tx streams dumpers.
 497  * If not called the dumpers will be flushed within each tx burst.
 498  */
 499 static void
 500 eth_dev_stop(struct rte_eth_dev *dev)
 501 {
 502         unsigned i;
 503         struct pmd_internals *internals = dev->data->dev_private;
 504         struct pcap_tx_queue *tx;
 505         struct pcap_rx_queue *rx;
 506
 507         /* Special iface case. Single pcap is open and shared between tx/rx. */
 508         if (internals->single_iface) {
 509                 tx = &internals->tx_queue[0];
 510                 rx = &internals->rx_queue[0];
 511                 pcap_close(tx->pcap);
 512                 tx->pcap = NULL;
 513                 rx->pcap = NULL;
 514                 goto status_down;
 515         }
 516
 517         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 518                 tx = &internals->tx_queue[i];
 519
 520                 if (tx->dumper != NULL) {
 521                         pcap_dump_close(tx->dumper);
 522                         tx->dumper = NULL;
 523                 }
 524
 525                 if (tx->pcap != NULL) {
 526                         pcap_close(tx->pcap);
 527                         tx->pcap = NULL;
 528                 }
 529         }
 530
 531         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 532                 rx = &internals->rx_queue[i];
 533
 534                 if (rx->pcap != NULL) {
 535                         pcap_close(rx->pcap);
 536                         rx->pcap = NULL;
 537                 }
 538         }
 539
 540 status_down:
 541         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 542 }
 543
 544 static int
 545 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 546 {
 547         return 0;
 548 }
 549
 550 static void
 551 eth_dev_info(struct rte_eth_dev *dev,
 552                 struct rte_eth_dev_info *dev_info)
 553 {
 554         struct pmd_internals *internals = dev->data->dev_private;
 555         dev_info->driver_name = drivername;
 556         dev_info->if_index = internals->if_index;
 557         dev_info->max_mac_addrs = 1;
 558         dev_info->max_rx_pktlen = (uint32_t) -1;
 559         dev_info->max_rx_queues = dev->data->nb_rx_queues;
 560         dev_info->max_tx_queues = dev->data->nb_tx_queues;
 561         dev_info->min_rx_bufsize = 0;
 562         dev_info->pci_dev = NULL;
 563 }
 564
 565 static void
 566 eth_stats_get(struct rte_eth_dev *dev,
 567                 struct rte_eth_stats *stats)
 568 {
 569         unsigned i;
 570         unsigned long rx_packets_total = 0, rx_bytes_total = 0;
 571         unsigned long tx_packets_total = 0, tx_bytes_total = 0;
 572         unsigned long tx_packets_err_total = 0;
 573         const struct pmd_internals *internal = dev->data->dev_private;
 574
 575         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 576                         i < dev->data->nb_rx_queues; i++) {
 577                 stats->q_ipackets[i] = internal->rx_queue[i].rx_stat.pkts;
 578                 stats->q_ibytes[i] = internal->rx_queue[i].rx_stat.bytes;
 579                 rx_packets_total += stats->q_ipackets[i];
 580                 rx_bytes_total += stats->q_ibytes[i];
 581         }
 582
 583         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 584                         i < dev->data->nb_tx_queues; i++) {
 585                 stats->q_opackets[i] = internal->tx_queue[i].tx_stat.pkts;
 586                 stats->q_obytes[i] = internal->tx_queue[i].tx_stat.bytes;
 587                 stats->q_errors[i] = internal->tx_queue[i].tx_stat.err_pkts;
 588                 tx_packets_total += stats->q_opackets[i];
 589                 tx_bytes_total += stats->q_obytes[i];
 590                 tx_packets_err_total += stats->q_errors[i];
 591         }
 592
 593         stats->ipackets = rx_packets_total;
 594         stats->ibytes = rx_bytes_total;
 595         stats->opackets = tx_packets_total;
 596         stats->obytes = tx_bytes_total;
 597         stats->oerrors = tx_packets_err_total;
 598 }
 599
 600 static void
 601 eth_stats_reset(struct rte_eth_dev *dev)
 602 {
 603         unsigned i;
 604         struct pmd_internals *internal = dev->data->dev_private;
 605         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 606                 internal->rx_queue[i].rx_stat.pkts = 0;
 607                 internal->rx_queue[i].rx_stat.bytes = 0;
 608         }
 609         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 610                 internal->tx_queue[i].tx_stat.pkts = 0;
 611                 internal->tx_queue[i].tx_stat.bytes = 0;
 612                 internal->tx_queue[i].tx_stat.err_pkts = 0;
 613         }
 614 }
 615
 616 static void
 617 eth_dev_close(struct rte_eth_dev *dev __rte_unused)
 618 {
 619 }
 620
 621 static void
 622 eth_queue_release(void *q __rte_unused)
 623 {
 624 }
 625
 626 static int
 627 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 628                 int wait_to_complete __rte_unused)
 629 {
 630         return 0;
 631 }
 632
 633 static int
 634 eth_rx_queue_setup(struct rte_eth_dev *dev,
 635                 uint16_t rx_queue_id,
 636                 uint16_t nb_rx_desc __rte_unused,
 637                 unsigned int socket_id __rte_unused,
 638                 const struct rte_eth_rxconf *rx_conf __rte_unused,
 639                 struct rte_mempool *mb_pool)
 640 {
 641         struct pmd_internals *internals = dev->data->dev_private;
 642         struct pcap_rx_queue *pcap_q = &internals->rx_queue[rx_queue_id];
 643         pcap_q->mb_pool = mb_pool;
 644         dev->data->rx_queues[rx_queue_id] = pcap_q;
 645         pcap_q->in_port = dev->data->port_id;
 646         return 0;
 647 }
 648
 649 static int
 650 eth_tx_queue_setup(struct rte_eth_dev *dev,
 651                 uint16_t tx_queue_id,
 652                 uint16_t nb_tx_desc __rte_unused,
 653                 unsigned int socket_id __rte_unused,
 654                 const struct rte_eth_txconf *tx_conf __rte_unused)
 655 {
 656
 657         struct pmd_internals *internals = dev->data->dev_private;
 658         dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
 659         return 0;
 660 }
 661
 662 static const struct eth_dev_ops ops = {
 663         .dev_start = eth_dev_start,
 664         .dev_stop =     eth_dev_stop,
 665         .dev_close = eth_dev_close,
 666         .dev_configure = eth_dev_configure,
 667         .dev_infos_get = eth_dev_info,
 668         .rx_queue_setup = eth_rx_queue_setup,
 669         .tx_queue_setup = eth_tx_queue_setup,
 670         .rx_queue_release = eth_queue_release,
 671         .tx_queue_release = eth_queue_release,
 672         .link_update = eth_link_update,
 673         .stats_get = eth_stats_get,
 674         .stats_reset = eth_stats_reset,
 675 };
 676
 677 /*
 678  * Function handler that opens the pcap file for reading a stores a
 679  * reference of it for use it later on.
 680  */
 681 static int
 682 open_rx_pcap(const char *key, const char *value, void *extra_args)
 683 {
 684         unsigned i;
 685         const char *pcap_filename = value;
 686         struct pmd_devargs *rx = extra_args;
 687         pcap_t *pcap = NULL;
 688
 689         for (i = 0; i < rx->num_of_queue; i++) {
 690                 if (open_single_rx_pcap(pcap_filename, &pcap) < 0)
 691                         return -1;
 692
 693                 rx->queue[i].pcap = pcap;
 694                 rx->queue[i].name = pcap_filename;
 695                 rx->queue[i].type = key;
 696         }
 697
 698         return 0;
 699 }
 700
 701 /*
 702  * Opens a pcap file for writing and stores a reference to it
 703  * for use it later on.
 704  */
 705 static int
 706 open_tx_pcap(const char *key, const char *value, void *extra_args)
 707 {
 708         unsigned i;
 709         const char *pcap_filename = value;
 710         struct pmd_devargs *dumpers = extra_args;
 711         pcap_dumper_t *dumper;
 712
 713         for (i = 0; i < dumpers->num_of_queue; i++) {
 714                 if (open_single_tx_pcap(pcap_filename, &dumper) < 0)
 715                         return -1;
 716
 717                 dumpers->queue[i].dumper = dumper;
 718                 dumpers->queue[i].name = pcap_filename;
 719                 dumpers->queue[i].type = key;
 720         }
 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 pmd_devargs *tx = extra_args;
 733         pcap_t *pcap = NULL;
 734
 735         if (open_single_iface(iface, &pcap) < 0)
 736                 return -1;
 737
 738         tx->queue[0].pcap = pcap;
 739         tx->queue[0].name = iface;
 740         tx->queue[0].type = 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 pmd_devargs *rx = extra_args;
 754         pcap_t *pcap = NULL;
 755
 756         for (i = 0; i < rx->num_of_queue; i++) {
 757                 if (open_single_iface(iface, &pcap) < 0)
 758                         return -1;
 759                 rx->queue[i].pcap = pcap;
 760                 rx->queue[i].name = iface;
 761                 rx->queue[i].type = 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 pmd_devargs *tx = extra_args;
 776         pcap_t *pcap;
 777
 778         for (i = 0; i < tx->num_of_queue; i++) {
 779                 if (open_single_iface(iface, &pcap) < 0)
 780                         return -1;
 781                 tx->queue[i].pcap = pcap;
 782                 tx->queue[i].name = iface;
 783                 tx->queue[i].type = key;
 784         }
 785
 786         return 0;
 787 }
 788
 789 static int
 790 rte_pmd_init_internals(const char *name, const unsigned nb_rx_queues,
 791                 const unsigned nb_tx_queues, struct pmd_internals **internals,
 792                 struct rte_eth_dev **eth_dev)
 793 {
 794         struct rte_eth_dev_data *data = NULL;
 795         unsigned int numa_node = rte_socket_id();
 796
 797         RTE_LOG(INFO, PMD, "Creating pcap-backed ethdev on numa socket %u\n",
 798                 numa_node);
 799
 800         /* now do all data allocation - for eth_dev structure
 801          * and internal (private) data
 802          */
 803         data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
 804         if (data == NULL)
 805                 goto error;
 806
 807         *internals = rte_zmalloc_socket(name, sizeof(**internals), 0,
 808                         numa_node);
 809         if (*internals == NULL)
 810                 goto error;
 811
 812         /* reserve an ethdev entry */
 813         *eth_dev = rte_eth_dev_allocate(name);
 814         if (*eth_dev == NULL)
 815                 goto error;
 816
 817         /* check length of device name */
 818         if ((strlen((*eth_dev)->data->name) + 1) > sizeof(data->name))
 819                 goto error;
 820
 821         /* now put it all together
 822          * - store queue data in internals,
 823          * - store numa_node info in eth_dev
 824          * - point eth_dev_data to internals
 825          * - and point eth_dev structure to new eth_dev_data structure
 826          */
 827         data->dev_private = *internals;
 828         data->port_id = (*eth_dev)->data->port_id;
 829         snprintf(data->name, sizeof(data->name), "%s", (*eth_dev)->data->name);
 830         data->nb_rx_queues = (uint16_t)nb_rx_queues;
 831         data->nb_tx_queues = (uint16_t)nb_tx_queues;
 832         data->dev_link = pmd_link;
 833         data->mac_addrs = &eth_addr;
 834         strncpy(data->name,
 835                 (*eth_dev)->data->name, strlen((*eth_dev)->data->name));
 836
 837         /*
 838          * NOTE: we'll replace the data element, of originally allocated
 839          * eth_dev so the rings are local per-process
 840          */
 841         (*eth_dev)->data = data;
 842         (*eth_dev)->dev_ops = &ops;
 843         (*eth_dev)->driver = NULL;
 844         data->dev_flags = RTE_ETH_DEV_DETACHABLE;
 845         data->kdrv = RTE_KDRV_NONE;
 846         data->drv_name = drivername;
 847         data->numa_node = numa_node;
 848
 849         return 0;
 850
 851 error:
 852         rte_free(data);
 853         rte_free(*internals);
 854
 855         return -1;
 856 }
 857
 858 static int
 859 rte_eth_from_pcaps_common(const char *name, struct pmd_devargs *rx_queues,
 860                 const unsigned nb_rx_queues, struct pmd_devargs *tx_queues,
 861                 const unsigned nb_tx_queues, struct rte_kvargs *kvlist,
 862                 struct pmd_internals **internals, struct rte_eth_dev **eth_dev)
 863 {
 864         struct rte_kvargs_pair *pair = NULL;
 865         unsigned k_idx;
 866         unsigned i;
 867
 868         /* do some parameter checking */
 869         if (rx_queues == NULL && nb_rx_queues > 0)
 870                 return -1;
 871         if (tx_queues == NULL && nb_tx_queues > 0)
 872                 return -1;
 873
 874         if (rte_pmd_init_internals(name, nb_rx_queues, nb_tx_queues, internals,
 875                         eth_dev) < 0)
 876                 return -1;
 877
 878         for (i = 0; i < nb_rx_queues; i++) {
 879                 (*internals)->rx_queue[i].pcap = rx_queues->queue[i].pcap;
 880                 snprintf((*internals)->rx_queue[i].name,
 881                         sizeof((*internals)->rx_queue[i].name), "%s",
 882                         rx_queues->queue[i].name);
 883                 snprintf((*internals)->rx_queue[i].type,
 884                         sizeof((*internals)->rx_queue[i].type), "%s",
 885                         rx_queues->queue[i].type);
 886         }
 887         for (i = 0; i < nb_tx_queues; i++) {
 888                 (*internals)->tx_queue[i].dumper = tx_queues->queue[i].dumper;
 889                 snprintf((*internals)->tx_queue[i].name,
 890                         sizeof((*internals)->tx_queue[i].name), "%s",
 891                         tx_queues->queue[i].name);
 892                 snprintf((*internals)->tx_queue[i].type,
 893                         sizeof((*internals)->tx_queue[i].type), "%s",
 894                         tx_queues->queue[i].type);
 895         }
 896
 897         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 898                 pair = &kvlist->pairs[k_idx];
 899                 if (strstr(pair->key, ETH_PCAP_IFACE_ARG) != NULL)
 900                         break;
 901         }
 902
 903         if (pair == NULL)
 904                 (*internals)->if_index = 0;
 905         else
 906                 (*internals)->if_index = if_nametoindex(pair->value);
 907
 908         return 0;
 909 }
 910
 911 static int
 912 rte_eth_from_pcaps(const char *name, struct pmd_devargs *rx_queues,
 913                 const unsigned nb_rx_queues, struct pmd_devargs *tx_queues,
 914                 const unsigned nb_tx_queues, struct rte_kvargs *kvlist,
 915                 int single_iface, unsigned int using_dumpers)
 916 {
 917         struct pmd_internals *internals = NULL;
 918         struct rte_eth_dev *eth_dev = NULL;
 919         int ret;
 920
 921         ret = rte_eth_from_pcaps_common(name, rx_queues, nb_rx_queues,
 922                 tx_queues, nb_tx_queues, kvlist, &internals, &eth_dev);
 923
 924         if (ret < 0)
 925                 return ret;
 926
 927         /* store wether we are using a single interface for rx/tx or not */
 928         internals->single_iface = single_iface;
 929
 930         eth_dev->rx_pkt_burst = eth_pcap_rx;
 931
 932         if (using_dumpers)
 933                 eth_dev->tx_pkt_burst = eth_pcap_tx_dumper;
 934         else
 935                 eth_dev->tx_pkt_burst = eth_pcap_tx;
 936
 937         return 0;
 938 }
 939
 940
 941 static int
 942 rte_pmd_pcap_devinit(const char *name, const char *params)
 943 {
 944         unsigned int is_rx_pcap = 0, is_tx_pcap = 0;
 945         struct rte_kvargs *kvlist;
 946         struct pmd_devargs pcaps = {0};
 947         struct pmd_devargs dumpers = {0};
 948         int single_iface = 0;
 949         int ret;
 950
 951         RTE_LOG(INFO, PMD, "Initializing pmd_pcap for %s\n", name);
 952
 953         gettimeofday(&start_time, NULL);
 954         start_cycles = rte_get_timer_cycles();
 955         hz = rte_get_timer_hz();
 956
 957         kvlist = rte_kvargs_parse(params, valid_arguments);
 958         if (kvlist == NULL)
 959                 return -1;
 960
 961         /*
 962          * If iface argument is passed we open the NICs and use them for
 963          * reading / writing
 964          */
 965         if (rte_kvargs_count(kvlist, ETH_PCAP_IFACE_ARG) == 1) {
 966
 967                 ret = rte_kvargs_process(kvlist, ETH_PCAP_IFACE_ARG,
 968                                 &open_rx_tx_iface, &pcaps);
 969
 970                 if (ret < 0)
 971                         goto free_kvlist;
 972
 973                 dumpers.queue[0] = pcaps.queue[0];
 974
 975                 single_iface = 1;
 976                 pcaps.num_of_queue = 1;
 977                 dumpers.num_of_queue = 1;
 978
 979                 goto create_eth;
 980         }
 981
 982         /*
 983          * We check whether we want to open a RX stream from a real NIC or a
 984          * pcap file
 985          */
 986         pcaps.num_of_queue = rte_kvargs_count(kvlist, ETH_PCAP_RX_PCAP_ARG);
 987         if (pcaps.num_of_queue)
 988                 is_rx_pcap = 1;
 989         else
 990                 pcaps.num_of_queue = rte_kvargs_count(kvlist,
 991                                 ETH_PCAP_RX_IFACE_ARG);
 992
 993         if (pcaps.num_of_queue > RTE_PMD_PCAP_MAX_QUEUES)
 994                 pcaps.num_of_queue = RTE_PMD_PCAP_MAX_QUEUES;
 995
 996         if (is_rx_pcap)
 997                 ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_PCAP_ARG,
 998                                 &open_rx_pcap, &pcaps);
 999         else
1000                 ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_IFACE_ARG,
1001                                 &open_rx_iface, &pcaps);
1002
1003         if (ret < 0)
1004                 goto free_kvlist;
1005
1006         /*
1007          * We check whether we want to open a TX stream to a real NIC or a
1008          * pcap file
1009          */
1010         dumpers.num_of_queue = rte_kvargs_count(kvlist, ETH_PCAP_TX_PCAP_ARG);
1011         if (dumpers.num_of_queue)
1012                 is_tx_pcap = 1;
1013         else
1014                 dumpers.num_of_queue = rte_kvargs_count(kvlist,
1015                                 ETH_PCAP_TX_IFACE_ARG);
1016
1017         if (dumpers.num_of_queue > RTE_PMD_PCAP_MAX_QUEUES)
1018                 dumpers.num_of_queue = RTE_PMD_PCAP_MAX_QUEUES;
1019
1020         if (is_tx_pcap)
1021                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_PCAP_ARG,
1022                                 &open_tx_pcap, &dumpers);
1023         else
1024                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_IFACE_ARG,
1025                                 &open_tx_iface, &dumpers);
1026
1027         if (ret < 0)
1028                 goto free_kvlist;
1029
1030 create_eth:
1031         ret = rte_eth_from_pcaps(name, &pcaps, pcaps.num_of_queue, &dumpers,
1032                 dumpers.num_of_queue, kvlist, single_iface, is_tx_pcap);
1033
1034 free_kvlist:
1035         rte_kvargs_free(kvlist);
1036         return ret;
1037 }
1038
1039 static int
1040 rte_pmd_pcap_devuninit(const char *name)
1041 {
1042         struct rte_eth_dev *eth_dev = NULL;
1043
1044         RTE_LOG(INFO, PMD, "Closing pcap ethdev on numa socket %u\n",
1045                         rte_socket_id());
1046
1047         if (name == NULL)
1048                 return -1;
1049
1050         /* reserve an ethdev entry */
1051         eth_dev = rte_eth_dev_allocated(name);
1052         if (eth_dev == NULL)
1053                 return -1;
1054
1055         rte_free(eth_dev->data->dev_private);
1056         rte_free(eth_dev->data);
1057
1058         rte_eth_dev_release_port(eth_dev);
1059
1060         return 0;
1061 }
1062
1063 static struct rte_vdev_driver pmd_pcap_drv = {
1064         .init = rte_pmd_pcap_devinit,
1065         .uninit = rte_pmd_pcap_devuninit,
1066 };
1067
1068 DRIVER_REGISTER_VDEV(net_pcap, pmd_pcap_drv);
1069 DRIVER_REGISTER_PARAM_STRING(net_pcap,
1070         ETH_PCAP_RX_PCAP_ARG "=<string> "
1071         ETH_PCAP_TX_PCAP_ARG "=<string> "
1072         ETH_PCAP_RX_IFACE_ARG "=<ifc> "
1073         ETH_PCAP_TX_IFACE_ARG "=<ifc> "
1074         ETH_PCAP_IFACE_ARG "=<ifc>");