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 int 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
 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 int 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
 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 = 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, can copy it */
 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                                 rte_pktmbuf_free(mbuf);
 226                                 break;
 227                         }
 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
 239         return num_rx;
 240 }
 241
 242 static inline void
 243 calculate_timestamp(struct timeval *ts) {
 244         uint64_t cycles;
 245         struct timeval cur_time;
 246
 247         cycles = rte_get_timer_cycles() - start_cycles;
 248         cur_time.tv_sec = cycles / hz;
 249         cur_time.tv_usec = (cycles % hz) * 10e6 / hz;
 250         timeradd(&start_time, &cur_time, ts);
 251 }
 252
 253 /*
 254  * Callback to handle writing packets to a pcap file.
 255  */
 256 static uint16_t
 257 eth_pcap_tx_dumper(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 258 {
 259         unsigned int i;
 260         struct rte_mbuf *mbuf;
 261         struct pcap_tx_queue *dumper_q = queue;
 262         uint16_t num_tx = 0;
 263         uint32_t tx_bytes = 0;
 264         struct pcap_pkthdr header;
 265
 266         if (dumper_q->dumper == NULL || nb_pkts == 0)
 267                 return 0;
 268
 269         /* writes the nb_pkts packets to the previously opened pcap file
 270          * 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. Size (%d) > max jumbo size (%d).\n",
 288                                         mbuf->pkt_len,
 289                                         ETHER_MAX_JUMBO_FRAME_LEN);
 290
 291                                 rte_pktmbuf_free(mbuf);
 292                                 break;
 293                         }
 294                 }
 295
 296                 rte_pktmbuf_free(mbuf);
 297                 num_tx++;
 298                 tx_bytes += mbuf->pkt_len;
 299         }
 300
 301         /*
 302          * Since there's no place to hook a callback when the forwarding
 303          * process stops and to make sure the pcap file is actually written,
 304          * we flush the pcap dumper within each burst.
 305          */
 306         pcap_dump_flush(dumper_q->dumper);
 307         dumper_q->tx_stat.pkts += num_tx;
 308         dumper_q->tx_stat.bytes += tx_bytes;
 309         dumper_q->tx_stat.err_pkts += nb_pkts - num_tx;
 310
 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, struct rte_mbuf **bufs, uint16_t nb_pkts)
 319 {
 320         unsigned int i;
 321         int ret;
 322         struct rte_mbuf *mbuf;
 323         struct pcap_tx_queue *tx_queue = queue;
 324         uint16_t num_tx = 0;
 325         uint32_t tx_bytes = 0;
 326
 327         if (unlikely(nb_pkts == 0 || tx_queue->pcap == NULL))
 328                 return 0;
 329
 330         for (i = 0; i < nb_pkts; i++) {
 331                 mbuf = bufs[i];
 332
 333                 if (likely(mbuf->nb_segs == 1)) {
 334                         ret = pcap_sendpacket(tx_queue->pcap,
 335                                         rte_pktmbuf_mtod(mbuf, u_char *),
 336                                         mbuf->pkt_len);
 337                 } else {
 338                         if (mbuf->pkt_len <= ETHER_MAX_JUMBO_FRAME_LEN) {
 339                                 eth_pcap_gather_data(tx_pcap_data, mbuf);
 340                                 ret = pcap_sendpacket(tx_queue->pcap,
 341                                                 tx_pcap_data, mbuf->pkt_len);
 342                         } else {
 343                                 RTE_LOG(ERR, PMD,
 344                                         "Dropping PCAP packet. Size (%d) > max jumbo size (%d).\n",
 345                                         mbuf->pkt_len,
 346                                         ETHER_MAX_JUMBO_FRAME_LEN);
 347
 348                                 rte_pktmbuf_free(mbuf);
 349                                 break;
 350                         }
 351                 }
 352
 353                 if (unlikely(ret != 0))
 354                         break;
 355                 num_tx++;
 356                 tx_bytes += mbuf->pkt_len;
 357                 rte_pktmbuf_free(mbuf);
 358         }
 359
 360         tx_queue->tx_stat.pkts += num_tx;
 361         tx_queue->tx_stat.bytes += tx_bytes;
 362         tx_queue->tx_stat.err_pkts += nb_pkts - num_tx;
 363
 364         return num_tx;
 365 }
 366
 367 /*
 368  * pcap_open_live wrapper function
 369  */
 370 static inline int
 371 open_iface_live(const char *iface, pcap_t **pcap) {
 372         *pcap = pcap_open_live(iface, RTE_ETH_PCAP_SNAPLEN,
 373                         RTE_ETH_PCAP_PROMISC, RTE_ETH_PCAP_TIMEOUT, errbuf);
 374
 375         if (*pcap == NULL) {
 376                 RTE_LOG(ERR, PMD, "Couldn't open %s: %s\n", iface, errbuf);
 377                 return -1;
 378         }
 379
 380         return 0;
 381 }
 382
 383 static int
 384 open_single_iface(const char *iface, pcap_t **pcap)
 385 {
 386         if (open_iface_live(iface, pcap) < 0) {
 387                 RTE_LOG(ERR, PMD, "Couldn't open interface %s\n", iface);
 388                 return -1;
 389         }
 390
 391         return 0;
 392 }
 393
 394 static int
 395 open_single_tx_pcap(const char *pcap_filename, pcap_dumper_t **dumper)
 396 {
 397         pcap_t *tx_pcap;
 398
 399         /*
 400          * We need to create a dummy empty pcap_t to use it
 401          * with pcap_dump_open(). We create big enough an Ethernet
 402          * pcap holder.
 403          */
 404         tx_pcap = pcap_open_dead(DLT_EN10MB, RTE_ETH_PCAP_SNAPSHOT_LEN);
 405         if (tx_pcap == NULL) {
 406                 RTE_LOG(ERR, PMD, "Couldn't create dead pcap\n");
 407                 return -1;
 408         }
 409
 410         /* The dumper is created using the previous pcap_t reference */
 411         *dumper = pcap_dump_open(tx_pcap, pcap_filename);
 412         if (*dumper == NULL) {
 413                 RTE_LOG(ERR, PMD, "Couldn't open %s for writing.\n",
 414                         pcap_filename);
 415                 return -1;
 416         }
 417
 418         return 0;
 419 }
 420
 421 static int
 422 open_single_rx_pcap(const char *pcap_filename, pcap_t **pcap)
 423 {
 424         *pcap = pcap_open_offline(pcap_filename, errbuf);
 425         if (*pcap == NULL) {
 426                 RTE_LOG(ERR, PMD, "Couldn't open %s: %s\n", pcap_filename,
 427                         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 int 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 &&
 460                                 strcmp(tx->type, ETH_PCAP_TX_PCAP_ARG) == 0) {
 461                         if (open_single_tx_pcap(tx->name, &tx->dumper) < 0)
 462                                 return -1;
 463                 } else if (!tx->pcap &&
 464                                 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                 } else if (strcmp(rx->type, ETH_PCAP_RX_IFACE_ARG) == 0) {
 481                         if (open_single_iface(rx->name, &rx->pcap) < 0)
 482                                 return -1;
 483                 }
 484         }
 485
 486 status_up:
 487         dev->data->dev_link.link_status = ETH_LINK_UP;
 488
 489         return 0;
 490 }
 491
 492 /*
 493  * This function gets called when the current port gets stopped.
 494  * Is the only place for us to close all the tx streams dumpers.
 495  * If not called the dumpers will be flushed within each tx burst.
 496  */
 497 static void
 498 eth_dev_stop(struct rte_eth_dev *dev)
 499 {
 500         unsigned int i;
 501         struct pmd_internals *internals = dev->data->dev_private;
 502         struct pcap_tx_queue *tx;
 503         struct pcap_rx_queue *rx;
 504
 505         /* Special iface case. Single pcap is open and shared between tx/rx. */
 506         if (internals->single_iface) {
 507                 tx = &internals->tx_queue[0];
 508                 rx = &internals->rx_queue[0];
 509                 pcap_close(tx->pcap);
 510                 tx->pcap = NULL;
 511                 rx->pcap = NULL;
 512                 goto status_down;
 513         }
 514
 515         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 516                 tx = &internals->tx_queue[i];
 517
 518                 if (tx->dumper != NULL) {
 519                         pcap_dump_close(tx->dumper);
 520                         tx->dumper = NULL;
 521                 }
 522
 523                 if (tx->pcap != NULL) {
 524                         pcap_close(tx->pcap);
 525                         tx->pcap = NULL;
 526                 }
 527         }
 528
 529         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 530                 rx = &internals->rx_queue[i];
 531
 532                 if (rx->pcap != NULL) {
 533                         pcap_close(rx->pcap);
 534                         rx->pcap = NULL;
 535                 }
 536         }
 537
 538 status_down:
 539         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 540 }
 541
 542 static int
 543 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 544 {
 545         return 0;
 546 }
 547
 548 static void
 549 eth_dev_info(struct rte_eth_dev *dev,
 550                 struct rte_eth_dev_info *dev_info)
 551 {
 552         struct pmd_internals *internals = dev->data->dev_private;
 553
 554         dev_info->if_index = internals->if_index;
 555         dev_info->max_mac_addrs = 1;
 556         dev_info->max_rx_pktlen = (uint32_t) -1;
 557         dev_info->max_rx_queues = dev->data->nb_rx_queues;
 558         dev_info->max_tx_queues = dev->data->nb_tx_queues;
 559         dev_info->min_rx_bufsize = 0;
 560 }
 561
 562 static void
 563 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 564 {
 565         unsigned int i;
 566         unsigned long rx_packets_total = 0, rx_bytes_total = 0;
 567         unsigned long tx_packets_total = 0, tx_bytes_total = 0;
 568         unsigned long tx_packets_err_total = 0;
 569         const struct pmd_internals *internal = dev->data->dev_private;
 570
 571         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 572                         i < dev->data->nb_rx_queues; i++) {
 573                 stats->q_ipackets[i] = internal->rx_queue[i].rx_stat.pkts;
 574                 stats->q_ibytes[i] = internal->rx_queue[i].rx_stat.bytes;
 575                 rx_packets_total += stats->q_ipackets[i];
 576                 rx_bytes_total += stats->q_ibytes[i];
 577         }
 578
 579         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 580                         i < dev->data->nb_tx_queues; i++) {
 581                 stats->q_opackets[i] = internal->tx_queue[i].tx_stat.pkts;
 582                 stats->q_obytes[i] = internal->tx_queue[i].tx_stat.bytes;
 583                 stats->q_errors[i] = internal->tx_queue[i].tx_stat.err_pkts;
 584                 tx_packets_total += stats->q_opackets[i];
 585                 tx_bytes_total += stats->q_obytes[i];
 586                 tx_packets_err_total += stats->q_errors[i];
 587         }
 588
 589         stats->ipackets = rx_packets_total;
 590         stats->ibytes = rx_bytes_total;
 591         stats->opackets = tx_packets_total;
 592         stats->obytes = tx_bytes_total;
 593         stats->oerrors = tx_packets_err_total;
 594 }
 595
 596 static void
 597 eth_stats_reset(struct rte_eth_dev *dev)
 598 {
 599         unsigned int i;
 600         struct pmd_internals *internal = dev->data->dev_private;
 601
 602         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 603                 internal->rx_queue[i].rx_stat.pkts = 0;
 604                 internal->rx_queue[i].rx_stat.bytes = 0;
 605         }
 606
 607         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 608                 internal->tx_queue[i].tx_stat.pkts = 0;
 609                 internal->tx_queue[i].tx_stat.bytes = 0;
 610                 internal->tx_queue[i].tx_stat.err_pkts = 0;
 611         }
 612 }
 613
 614 static void
 615 eth_dev_close(struct rte_eth_dev *dev __rte_unused)
 616 {
 617 }
 618
 619 static void
 620 eth_queue_release(void *q __rte_unused)
 621 {
 622 }
 623
 624 static int
 625 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 626                 int wait_to_complete __rte_unused)
 627 {
 628         return 0;
 629 }
 630
 631 static int
 632 eth_rx_queue_setup(struct rte_eth_dev *dev,
 633                 uint16_t rx_queue_id,
 634                 uint16_t nb_rx_desc __rte_unused,
 635                 unsigned int socket_id __rte_unused,
 636                 const struct rte_eth_rxconf *rx_conf __rte_unused,
 637                 struct rte_mempool *mb_pool)
 638 {
 639         struct pmd_internals *internals = dev->data->dev_private;
 640         struct pcap_rx_queue *pcap_q = &internals->rx_queue[rx_queue_id];
 641
 642         pcap_q->mb_pool = mb_pool;
 643         dev->data->rx_queues[rx_queue_id] = pcap_q;
 644         pcap_q->in_port = dev->data->port_id;
 645
 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         struct pmd_internals *internals = dev->data->dev_private;
 657
 658         dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
 659
 660         return 0;
 661 }
 662
 663 static const struct eth_dev_ops ops = {
 664         .dev_start = eth_dev_start,
 665         .dev_stop = eth_dev_stop,
 666         .dev_close = eth_dev_close,
 667         .dev_configure = eth_dev_configure,
 668         .dev_infos_get = eth_dev_info,
 669         .rx_queue_setup = eth_rx_queue_setup,
 670         .tx_queue_setup = eth_tx_queue_setup,
 671         .rx_queue_release = eth_queue_release,
 672         .tx_queue_release = eth_queue_release,
 673         .link_update = eth_link_update,
 674         .stats_get = eth_stats_get,
 675         .stats_reset = eth_stats_reset,
 676 };
 677
 678 /*
 679  * Function handler that opens the pcap file for reading a stores a
 680  * reference of it for use it later on.
 681  */
 682 static int
 683 open_rx_pcap(const char *key, const char *value, void *extra_args)
 684 {
 685         unsigned int i;
 686         const char *pcap_filename = value;
 687         struct pmd_devargs *rx = extra_args;
 688         pcap_t *pcap = NULL;
 689
 690         for (i = 0; i < rx->num_of_queue; i++) {
 691                 if (open_single_rx_pcap(pcap_filename, &pcap) < 0)
 692                         return -1;
 693
 694                 rx->queue[i].pcap = pcap;
 695                 rx->queue[i].name = pcap_filename;
 696                 rx->queue[i].type = key;
 697         }
 698
 699         return 0;
 700 }
 701
 702 /*
 703  * Opens a pcap file for writing and stores a reference to it
 704  * for use it later on.
 705  */
 706 static int
 707 open_tx_pcap(const char *key, const char *value, void *extra_args)
 708 {
 709         unsigned int i;
 710         const char *pcap_filename = value;
 711         struct pmd_devargs *dumpers = extra_args;
 712         pcap_dumper_t *dumper;
 713
 714         for (i = 0; i < dumpers->num_of_queue; i++) {
 715                 if (open_single_tx_pcap(pcap_filename, &dumper) < 0)
 716                         return -1;
 717
 718                 dumpers->queue[i].dumper = dumper;
 719                 dumpers->queue[i].name = pcap_filename;
 720                 dumpers->queue[i].type = key;
 721         }
 722
 723         return 0;
 724 }
 725
 726 /*
 727  * Opens an interface for reading and writing
 728  */
 729 static inline int
 730 open_rx_tx_iface(const char *key, const char *value, void *extra_args)
 731 {
 732         const char *iface = value;
 733         struct pmd_devargs *tx = extra_args;
 734         pcap_t *pcap = NULL;
 735
 736         if (open_single_iface(iface, &pcap) < 0)
 737                 return -1;
 738
 739         tx->queue[0].pcap = pcap;
 740         tx->queue[0].name = iface;
 741         tx->queue[0].type = key;
 742
 743         return 0;
 744 }
 745
 746 /*
 747  * Opens a NIC for reading packets from it
 748  */
 749 static inline int
 750 open_rx_iface(const char *key, const char *value, void *extra_args)
 751 {
 752         unsigned int i;
 753         const char *iface = value;
 754         struct pmd_devargs *rx = extra_args;
 755         pcap_t *pcap = NULL;
 756
 757         for (i = 0; i < rx->num_of_queue; i++) {
 758                 if (open_single_iface(iface, &pcap) < 0)
 759                         return -1;
 760                 rx->queue[i].pcap = pcap;
 761                 rx->queue[i].name = iface;
 762                 rx->queue[i].type = key;
 763         }
 764
 765         return 0;
 766 }
 767
 768 /*
 769  * Opens a NIC for writing packets to it
 770  */
 771 static int
 772 open_tx_iface(const char *key, const char *value, void *extra_args)
 773 {
 774         unsigned int i;
 775         const char *iface = value;
 776         struct pmd_devargs *tx = extra_args;
 777         pcap_t *pcap;
 778
 779         for (i = 0; i < tx->num_of_queue; i++) {
 780                 if (open_single_iface(iface, &pcap) < 0)
 781                         return -1;
 782                 tx->queue[i].pcap = pcap;
 783                 tx->queue[i].name = iface;
 784                 tx->queue[i].type = key;
 785         }
 786
 787         return 0;
 788 }
 789
 790 static int
 791 pmd_init_internals(const char *name, const unsigned int nb_rx_queues,
 792                 const unsigned int nb_tx_queues,
 793                 struct pmd_internals **internals,
 794                 struct rte_eth_dev **eth_dev)
 795 {
 796         struct rte_eth_dev_data *data = NULL;
 797         unsigned int numa_node = rte_socket_id();
 798
 799         RTE_LOG(INFO, PMD, "Creating pcap-backed ethdev on numa socket %u\n",
 800                 numa_node);
 801
 802         /* now do all data allocation - for eth_dev structure
 803          * and internal (private) data
 804          */
 805         data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
 806         if (data == NULL)
 807                 goto error;
 808
 809         *internals = rte_zmalloc_socket(name, sizeof(**internals), 0,
 810                         numa_node);
 811         if (*internals == NULL)
 812                 goto error;
 813
 814         /* reserve an ethdev entry */
 815         *eth_dev = rte_eth_dev_allocate(name);
 816         if (*eth_dev == NULL)
 817                 goto error;
 818
 819         /* now put it all together
 820          * - store queue data in internals,
 821          * - store numa_node info in eth_dev
 822          * - point eth_dev_data to internals
 823          * - and point eth_dev structure to new eth_dev_data structure
 824          */
 825         data->dev_private = *internals;
 826         data->port_id = (*eth_dev)->data->port_id;
 827         snprintf(data->name, sizeof(data->name), "%s", (*eth_dev)->data->name);
 828         data->nb_rx_queues = (uint16_t)nb_rx_queues;
 829         data->nb_tx_queues = (uint16_t)nb_tx_queues;
 830         data->dev_link = pmd_link;
 831         data->mac_addrs = &eth_addr;
 832
 833         /*
 834          * NOTE: we'll replace the data element, of originally allocated
 835          * eth_dev so the rings are local per-process
 836          */
 837         (*eth_dev)->data = data;
 838         (*eth_dev)->dev_ops = &ops;
 839         (*eth_dev)->driver = NULL;
 840         data->dev_flags = RTE_ETH_DEV_DETACHABLE;
 841         data->kdrv = RTE_KDRV_NONE;
 842         data->drv_name = "Pcap PMD";
 843         data->numa_node = numa_node;
 844
 845         return 0;
 846
 847 error:
 848         rte_free(data);
 849         rte_free(*internals);
 850
 851         return -1;
 852 }
 853
 854 static int
 855 eth_from_pcaps_common(const char *name, struct pmd_devargs *rx_queues,
 856                 const unsigned int nb_rx_queues, struct pmd_devargs *tx_queues,
 857                 const unsigned int nb_tx_queues, struct rte_kvargs *kvlist,
 858                 struct pmd_internals **internals, struct rte_eth_dev **eth_dev)
 859 {
 860         struct rte_kvargs_pair *pair = NULL;
 861         unsigned int k_idx;
 862         unsigned int i;
 863
 864         /* do some parameter checking */
 865         if (rx_queues == NULL && nb_rx_queues > 0)
 866                 return -1;
 867         if (tx_queues == NULL && nb_tx_queues > 0)
 868                 return -1;
 869
 870         if (pmd_init_internals(name, nb_rx_queues, nb_tx_queues, internals,
 871                         eth_dev) < 0)
 872                 return -1;
 873
 874         for (i = 0; i < nb_rx_queues; i++) {
 875                 struct pcap_rx_queue *rx = &(*internals)->rx_queue[i];
 876                 struct devargs_queue *queue = &rx_queues->queue[i];
 877
 878                 rx->pcap = queue->pcap;
 879                 snprintf(rx->name, sizeof(rx->name), "%s", queue->name);
 880                 snprintf(rx->type, sizeof(rx->type), "%s", queue->type);
 881         }
 882
 883         for (i = 0; i < nb_tx_queues; i++) {
 884                 struct pcap_tx_queue *tx = &(*internals)->tx_queue[i];
 885                 struct devargs_queue *queue = &tx_queues->queue[i];
 886
 887                 tx->dumper = queue->dumper;
 888                 tx->pcap = queue->pcap;
 889                 snprintf(tx->name, sizeof(tx->name), "%s", queue->name);
 890                 snprintf(tx->type, sizeof(tx->type), "%s", queue->type);
 891         }
 892
 893         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 894                 pair = &kvlist->pairs[k_idx];
 895                 if (strstr(pair->key, ETH_PCAP_IFACE_ARG) != NULL)
 896                         break;
 897         }
 898
 899         if (pair == NULL)
 900                 (*internals)->if_index = 0;
 901         else
 902                 (*internals)->if_index = if_nametoindex(pair->value);
 903
 904         return 0;
 905 }
 906
 907 static int
 908 eth_from_pcaps(const char *name, struct pmd_devargs *rx_queues,
 909                 const unsigned int nb_rx_queues, struct pmd_devargs *tx_queues,
 910                 const unsigned int nb_tx_queues, struct rte_kvargs *kvlist,
 911                 int single_iface, unsigned int using_dumpers)
 912 {
 913         struct pmd_internals *internals = NULL;
 914         struct rte_eth_dev *eth_dev = NULL;
 915         int ret;
 916
 917         ret = eth_from_pcaps_common(name, rx_queues, nb_rx_queues,
 918                 tx_queues, nb_tx_queues, kvlist, &internals, &eth_dev);
 919
 920         if (ret < 0)
 921                 return ret;
 922
 923         /* store weather we are using a single interface for rx/tx or not */
 924         internals->single_iface = single_iface;
 925
 926         eth_dev->rx_pkt_burst = eth_pcap_rx;
 927
 928         if (using_dumpers)
 929                 eth_dev->tx_pkt_burst = eth_pcap_tx_dumper;
 930         else
 931                 eth_dev->tx_pkt_burst = eth_pcap_tx;
 932
 933         return 0;
 934 }
 935
 936 static int
 937 pmd_pcap_probe(const char *name, const char *params)
 938 {
 939         unsigned int is_rx_pcap = 0, is_tx_pcap = 0;
 940         struct rte_kvargs *kvlist;
 941         struct pmd_devargs pcaps = {0};
 942         struct pmd_devargs dumpers = {0};
 943         int single_iface = 0;
 944         int ret;
 945
 946         RTE_LOG(INFO, PMD, "Initializing pmd_pcap for %s\n", name);
 947
 948         gettimeofday(&start_time, NULL);
 949         start_cycles = rte_get_timer_cycles();
 950         hz = rte_get_timer_hz();
 951
 952         kvlist = rte_kvargs_parse(params, valid_arguments);
 953         if (kvlist == NULL)
 954                 return -1;
 955
 956         /*
 957          * If iface argument is passed we open the NICs and use them for
 958          * reading / writing
 959          */
 960         if (rte_kvargs_count(kvlist, ETH_PCAP_IFACE_ARG) == 1) {
 961
 962                 ret = rte_kvargs_process(kvlist, ETH_PCAP_IFACE_ARG,
 963                                 &open_rx_tx_iface, &pcaps);
 964
 965                 if (ret < 0)
 966                         goto free_kvlist;
 967
 968                 dumpers.queue[0] = pcaps.queue[0];
 969
 970                 single_iface = 1;
 971                 pcaps.num_of_queue = 1;
 972                 dumpers.num_of_queue = 1;
 973
 974                 goto create_eth;
 975         }
 976
 977         /*
 978          * We check whether we want to open a RX stream from a real NIC or a
 979          * pcap file
 980          */
 981         pcaps.num_of_queue = rte_kvargs_count(kvlist, ETH_PCAP_RX_PCAP_ARG);
 982         if (pcaps.num_of_queue)
 983                 is_rx_pcap = 1;
 984         else
 985                 pcaps.num_of_queue = rte_kvargs_count(kvlist,
 986                                 ETH_PCAP_RX_IFACE_ARG);
 987
 988         if (pcaps.num_of_queue > RTE_PMD_PCAP_MAX_QUEUES)
 989                 pcaps.num_of_queue = RTE_PMD_PCAP_MAX_QUEUES;
 990
 991         if (is_rx_pcap)
 992                 ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_PCAP_ARG,
 993                                 &open_rx_pcap, &pcaps);
 994         else
 995                 ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_IFACE_ARG,
 996                                 &open_rx_iface, &pcaps);
 997
 998         if (ret < 0)
 999                 goto free_kvlist;
1000
1001         /*
1002          * We check whether we want to open a TX stream to a real NIC or a
1003          * pcap file
1004          */
1005         dumpers.num_of_queue = rte_kvargs_count(kvlist, ETH_PCAP_TX_PCAP_ARG);
1006         if (dumpers.num_of_queue)
1007                 is_tx_pcap = 1;
1008         else
1009                 dumpers.num_of_queue = rte_kvargs_count(kvlist,
1010                                 ETH_PCAP_TX_IFACE_ARG);
1011
1012         if (dumpers.num_of_queue > RTE_PMD_PCAP_MAX_QUEUES)
1013                 dumpers.num_of_queue = RTE_PMD_PCAP_MAX_QUEUES;
1014
1015         if (is_tx_pcap)
1016                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_PCAP_ARG,
1017                                 &open_tx_pcap, &dumpers);
1018         else
1019                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_IFACE_ARG,
1020                                 &open_tx_iface, &dumpers);
1021
1022         if (ret < 0)
1023                 goto free_kvlist;
1024
1025 create_eth:
1026         ret = eth_from_pcaps(name, &pcaps, pcaps.num_of_queue, &dumpers,
1027                 dumpers.num_of_queue, kvlist, single_iface, is_tx_pcap);
1028
1029 free_kvlist:
1030         rte_kvargs_free(kvlist);
1031
1032         return ret;
1033 }
1034
1035 static int
1036 pmd_pcap_remove(const char *name)
1037 {
1038         struct rte_eth_dev *eth_dev = NULL;
1039
1040         RTE_LOG(INFO, PMD, "Closing pcap ethdev on numa socket %u\n",
1041                         rte_socket_id());
1042
1043         if (name == NULL)
1044                 return -1;
1045
1046         /* reserve an ethdev entry */
1047         eth_dev = rte_eth_dev_allocated(name);
1048         if (eth_dev == NULL)
1049                 return -1;
1050
1051         rte_free(eth_dev->data->dev_private);
1052         rte_free(eth_dev->data);
1053
1054         rte_eth_dev_release_port(eth_dev);
1055
1056         return 0;
1057 }
1058
1059 static struct rte_vdev_driver pmd_pcap_drv = {
1060         .probe = pmd_pcap_probe,
1061         .remove = pmd_pcap_remove,
1062 };
1063
1064 RTE_PMD_REGISTER_VDEV(net_pcap, pmd_pcap_drv);
1065 RTE_PMD_REGISTER_ALIAS(net_pcap, eth_pcap);
1066 RTE_PMD_REGISTER_PARAM_STRING(net_pcap,
1067         ETH_PCAP_RX_PCAP_ARG "=<string> "
1068         ETH_PCAP_TX_PCAP_ARG "=<string> "
1069         ETH_PCAP_RX_IFACE_ARG "=<ifc> "
1070         ETH_PCAP_TX_IFACE_ARG "=<ifc> "
1071         ETH_PCAP_IFACE_ARG "=<ifc>");