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