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