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