drivers/net/pcap/rte_eth_pcap.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2014 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 #include <rte_mbuf.h>
  37 #include <rte_ethdev.h>
  38 #include <rte_malloc.h>
  39 #include <rte_memcpy.h>
  40 #include <rte_string_fns.h>
  41 #include <rte_cycles.h>
  42 #include <rte_kvargs.h>
  43 #include <rte_dev.h>
  44
  45 #include <net/if.h>
  46
  47 #include <pcap.h>
  48
  49 #define RTE_ETH_PCAP_SNAPSHOT_LEN 65535
  50 #define RTE_ETH_PCAP_SNAPLEN ETHER_MAX_JUMBO_FRAME_LEN
  51 #define RTE_ETH_PCAP_PROMISC 1
  52 #define RTE_ETH_PCAP_TIMEOUT -1
  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 static char errbuf[PCAP_ERRBUF_SIZE];
  62 static unsigned char tx_pcap_data[RTE_ETH_PCAP_SNAPLEN];
  63 static struct timeval start_time;
  64 static uint64_t start_cycles;
  65 static uint64_t hz;
  66
  67 struct pcap_rx_queue {
  68         pcap_t *pcap;
  69         uint8_t in_port;
  70         struct rte_mempool *mb_pool;
  71         volatile unsigned long rx_pkts;
  72         volatile unsigned long err_pkts;
  73         char name[PATH_MAX];
  74         char type[ETH_PCAP_ARG_MAXLEN];
  75 };
  76
  77 struct pcap_tx_queue {
  78         pcap_dumper_t *dumper;
  79         pcap_t *pcap;
  80         volatile unsigned long tx_pkts;
  81         volatile unsigned long err_pkts;
  82         char name[PATH_MAX];
  83         char type[ETH_PCAP_ARG_MAXLEN];
  84 };
  85
  86 struct rx_pcaps {
  87         unsigned num_of_rx;
  88         pcap_t *pcaps[RTE_PMD_RING_MAX_RX_RINGS];
  89         const char *names[RTE_PMD_RING_MAX_RX_RINGS];
  90         const char *types[RTE_PMD_RING_MAX_RX_RINGS];
  91 };
  92
  93 struct tx_pcaps {
  94         unsigned num_of_tx;
  95         pcap_dumper_t *dumpers[RTE_PMD_RING_MAX_TX_RINGS];
  96         pcap_t *pcaps[RTE_PMD_RING_MAX_RX_RINGS];
  97         const char *names[RTE_PMD_RING_MAX_RX_RINGS];
  98         const char *types[RTE_PMD_RING_MAX_RX_RINGS];
  99 };
 100
 101 struct pmd_internals {
 102         struct pcap_rx_queue rx_queue[RTE_PMD_RING_MAX_RX_RINGS];
 103         struct pcap_tx_queue tx_queue[RTE_PMD_RING_MAX_TX_RINGS];
 104         unsigned nb_rx_queues;
 105         unsigned nb_tx_queues;
 106         int if_index;
 107         int single_iface;
 108 };
 109
 110 const char *valid_arguments[] = {
 111         ETH_PCAP_RX_PCAP_ARG,
 112         ETH_PCAP_TX_PCAP_ARG,
 113         ETH_PCAP_RX_IFACE_ARG,
 114         ETH_PCAP_TX_IFACE_ARG,
 115         ETH_PCAP_IFACE_ARG,
 116         NULL
 117 };
 118
 119 static int open_single_tx_pcap(const char *pcap_filename, pcap_dumper_t **dumper);
 120 static int open_single_rx_pcap(const char *pcap_filename, pcap_t **pcap);
 121 static int open_single_iface(const char *iface, pcap_t **pcap);
 122
 123 static struct ether_addr eth_addr = { .addr_bytes = { 0, 0, 0, 0x1, 0x2, 0x3 } };
 124 static const char *drivername = "Pcap PMD";
 125 static struct rte_eth_link pmd_link = {
 126                 .link_speed = 10000,
 127                 .link_duplex = ETH_LINK_FULL_DUPLEX,
 128                 .link_status = 0
 129 };
 130
 131 static int
 132 eth_pcap_rx_jumbo(struct rte_mempool *mb_pool,
 133                   struct rte_mbuf *mbuf,
 134                   const u_char *data,
 135                   uint16_t data_len)
 136 {
 137         struct rte_mbuf *m = mbuf;
 138
 139         /* Copy the first segment. */
 140         uint16_t len = rte_pktmbuf_tailroom(mbuf);
 141
 142         rte_memcpy(rte_pktmbuf_append(mbuf, len), data, len);
 143         data_len -= len;
 144         data += len;
 145
 146         while (data_len > 0) {
 147                 /* Allocate next mbuf and point to that. */
 148                 m->next = rte_pktmbuf_alloc(mb_pool);
 149
 150                 if (unlikely(!m->next))
 151                         return -1;
 152
 153                 m = m->next;
 154
 155                 /* Headroom is not needed in chained mbufs. */
 156                 rte_pktmbuf_prepend(m, rte_pktmbuf_headroom(m));
 157                 m->pkt_len = 0;
 158                 m->data_len = 0;
 159
 160                 /* Copy next segment. */
 161                 len = RTE_MIN(rte_pktmbuf_tailroom(m), data_len);
 162                 rte_memcpy(rte_pktmbuf_append(m, len), data, len);
 163
 164                 mbuf->nb_segs++;
 165                 data_len -= len;
 166                 data += len;
 167         }
 168
 169         return mbuf->nb_segs;
 170 }
 171
 172 /* Copy data from mbuf chain to a buffer suitable for writing to a PCAP file. */
 173 static void
 174 eth_pcap_gather_data(unsigned char *data, struct rte_mbuf *mbuf)
 175 {
 176         uint16_t data_len = 0;
 177
 178         while (mbuf) {
 179                 rte_memcpy(data + data_len, rte_pktmbuf_mtod(mbuf, void *),
 180                            mbuf->data_len);
 181
 182                 data_len += mbuf->data_len;
 183                 mbuf = mbuf->next;
 184         }
 185 }
 186
 187 static uint16_t
 188 eth_pcap_rx(void *queue,
 189                 struct rte_mbuf **bufs,
 190                 uint16_t nb_pkts)
 191 {
 192         unsigned i;
 193         struct pcap_pkthdr header;
 194         const u_char *packet;
 195         struct rte_mbuf *mbuf;
 196         struct pcap_rx_queue *pcap_q = queue;
 197         uint16_t num_rx = 0;
 198         uint16_t buf_size;
 199
 200         if (unlikely(pcap_q->pcap == NULL || nb_pkts == 0))
 201                 return 0;
 202
 203         /* Reads the given number of packets from the pcap file one by one
 204          * and copies the packet data into a newly allocated mbuf to return.
 205          */
 206         for (i = 0; i < nb_pkts; i++) {
 207                 /* Get the next PCAP packet */
 208                 packet = pcap_next(pcap_q->pcap, &header);
 209                 if (unlikely(packet == NULL))
 210                         break;
 211                 else
 212                         mbuf = rte_pktmbuf_alloc(pcap_q->mb_pool);
 213                 if (unlikely(mbuf == NULL))
 214                         break;
 215
 216                 /* Now get the space available for data in the mbuf */
 217                 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(pcap_q->mb_pool) -
 218                                 RTE_PKTMBUF_HEADROOM);
 219
 220                 if (header.len <= buf_size) {
 221                         /* pcap packet will fit in the mbuf, go ahead and copy */
 222                         rte_memcpy(rte_pktmbuf_mtod(mbuf, void *), packet,
 223                                         header.len);
 224                         mbuf->data_len = (uint16_t)header.len;
 225                 } else {
 226                         /* Try read jumbo frame into multi mbufs. */
 227                         if (unlikely(eth_pcap_rx_jumbo(pcap_q->mb_pool,
 228                                                        mbuf,
 229                                                        packet,
 230                                                        header.len) == -1))
 231                                 break;
 232                 }
 233
 234                 mbuf->pkt_len = (uint16_t)header.len;
 235                 mbuf->port = pcap_q->in_port;
 236                 bufs[num_rx] = mbuf;
 237                 num_rx++;
 238         }
 239         pcap_q->rx_pkts += num_rx;
 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,
 259                 struct rte_mbuf **bufs,
 260                 uint16_t nb_pkts)
 261 {
 262         unsigned i;
 263         struct rte_mbuf *mbuf;
 264         struct pcap_tx_queue *dumper_q = queue;
 265         uint16_t num_tx = 0;
 266         struct pcap_pkthdr header;
 267
 268         if (dumper_q->dumper == NULL || nb_pkts == 0)
 269                 return 0;
 270
 271         /* writes the nb_pkts packets to the previously opened pcap file 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. "
 289                                         "Size (%d) > max jumbo size (%d).\n",
 290                                         mbuf->pkt_len,
 291                                         ETHER_MAX_JUMBO_FRAME_LEN);
 292
 293                                 rte_pktmbuf_free(mbuf);
 294                                 break;
 295                         }
 296                 }
 297
 298                 rte_pktmbuf_free(mbuf);
 299                 num_tx++;
 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_pkts += num_tx;
 309         dumper_q->err_pkts += nb_pkts - num_tx;
 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,
 318                 struct rte_mbuf **bufs,
 319                 uint16_t nb_pkts)
 320 {
 321         unsigned i;
 322         int ret;
 323         struct rte_mbuf *mbuf;
 324         struct pcap_tx_queue *tx_queue = queue;
 325         uint16_t num_tx = 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,
 342                                                       mbuf->pkt_len);
 343                         } else {
 344                                 RTE_LOG(ERR, PMD,
 345                                         "Dropping PCAP packet. "
 346                                         "Size (%d) > max jumbo size (%d).\n",
 347                                         mbuf->pkt_len,
 348                                         ETHER_MAX_JUMBO_FRAME_LEN);
 349
 350                                 rte_pktmbuf_free(mbuf);
 351                                 break;
 352                         }
 353                 }
 354
 355                 if (unlikely(ret != 0))
 356                         break;
 357                 num_tx++;
 358                 rte_pktmbuf_free(mbuf);
 359         }
 360
 361         tx_queue->tx_pkts += num_tx;
 362         tx_queue->err_pkts += nb_pkts - num_tx;
 363         return num_tx;
 364 }
 365
 366 static int
 367 eth_dev_start(struct rte_eth_dev *dev)
 368 {
 369         unsigned i;
 370         struct pmd_internals *internals = dev->data->dev_private;
 371         struct pcap_tx_queue *tx;
 372         struct pcap_rx_queue *rx;
 373
 374         /* Special iface case. Single pcap is open and shared between tx/rx. */
 375         if (internals->single_iface) {
 376                 tx = &internals->tx_queue[0];
 377                 rx = &internals->rx_queue[0];
 378
 379                 if (!tx->pcap && strcmp(tx->type, ETH_PCAP_IFACE_ARG) == 0) {
 380                         if (open_single_iface(tx->name, &tx->pcap) < 0)
 381                                 return -1;
 382                         rx->pcap = tx->pcap;
 383                 }
 384                 goto status_up;
 385         }
 386
 387         /* If not open already, open tx pcaps/dumpers */
 388         for (i = 0; i < internals->nb_tx_queues; i++) {
 389                 tx = &internals->tx_queue[i];
 390
 391                 if (!tx->dumper && strcmp(tx->type, ETH_PCAP_TX_PCAP_ARG) == 0) {
 392                         if (open_single_tx_pcap(tx->name, &tx->dumper) < 0)
 393                                 return -1;
 394                 }
 395
 396                 else if (!tx->pcap && strcmp(tx->type, ETH_PCAP_TX_IFACE_ARG) == 0) {
 397                         if (open_single_iface(tx->name, &tx->pcap) < 0)
 398                                 return -1;
 399                 }
 400         }
 401
 402         /* If not open already, open rx pcaps */
 403         for (i = 0; i < internals->nb_rx_queues; i++) {
 404                 rx = &internals->rx_queue[i];
 405
 406                 if (rx->pcap != NULL)
 407                         continue;
 408
 409                 if (strcmp(rx->type, ETH_PCAP_RX_PCAP_ARG) == 0) {
 410                         if (open_single_rx_pcap(rx->name, &rx->pcap) < 0)
 411                                 return -1;
 412                 }
 413
 414                 else if (strcmp(rx->type, ETH_PCAP_RX_IFACE_ARG) == 0) {
 415                         if (open_single_iface(rx->name, &rx->pcap) < 0)
 416                                 return -1;
 417                 }
 418         }
 419
 420 status_up:
 421
 422         dev->data->dev_link.link_status = 1;
 423         return 0;
 424 }
 425
 426 /*
 427  * This function gets called when the current port gets stopped.
 428  * Is the only place for us to close all the tx streams dumpers.
 429  * If not called the dumpers will be flushed within each tx burst.
 430  */
 431 static void
 432 eth_dev_stop(struct rte_eth_dev *dev)
 433 {
 434         unsigned i;
 435         struct pmd_internals *internals = dev->data->dev_private;
 436         struct pcap_tx_queue *tx;
 437         struct pcap_rx_queue *rx;
 438
 439         /* Special iface case. Single pcap is open and shared between tx/rx. */
 440         if (internals->single_iface) {
 441                 tx = &internals->tx_queue[0];
 442                 rx = &internals->rx_queue[0];
 443                 pcap_close(tx->pcap);
 444                 tx->pcap = NULL;
 445                 rx->pcap = NULL;
 446                 goto status_down;
 447         }
 448
 449         for (i = 0; i < internals->nb_tx_queues; i++) {
 450                 tx = &internals->tx_queue[i];
 451
 452                 if (tx->dumper != NULL) {
 453                         pcap_dump_close(tx->dumper);
 454                         tx->dumper = NULL;
 455                 }
 456
 457                 if (tx->pcap != NULL) {
 458                         pcap_close(tx->pcap);
 459                         tx->pcap = NULL;
 460                 }
 461         }
 462
 463         for (i = 0; i < internals->nb_rx_queues; i++) {
 464                 rx = &internals->rx_queue[i];
 465
 466                 if (rx->pcap != NULL) {
 467                         pcap_close(rx->pcap);
 468                         rx->pcap = NULL;
 469                 }
 470         }
 471
 472 status_down:
 473         dev->data->dev_link.link_status = 0;
 474 }
 475
 476 static int
 477 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 478 {
 479         return 0;
 480 }
 481
 482 static void
 483 eth_dev_info(struct rte_eth_dev *dev,
 484                 struct rte_eth_dev_info *dev_info)
 485 {
 486         struct pmd_internals *internals = dev->data->dev_private;
 487         dev_info->driver_name = drivername;
 488         dev_info->if_index = internals->if_index;
 489         dev_info->max_mac_addrs = 1;
 490         dev_info->max_rx_pktlen = (uint32_t) -1;
 491         dev_info->max_rx_queues = (uint16_t)internals->nb_rx_queues;
 492         dev_info->max_tx_queues = (uint16_t)internals->nb_tx_queues;
 493         dev_info->min_rx_bufsize = 0;
 494         dev_info->pci_dev = NULL;
 495 }
 496
 497 static void
 498 eth_stats_get(struct rte_eth_dev *dev,
 499                 struct rte_eth_stats *igb_stats)
 500 {
 501         unsigned i;
 502         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 503         const struct pmd_internals *internal = dev->data->dev_private;
 504
 505         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS && i < internal->nb_rx_queues;
 506                         i++) {
 507                 igb_stats->q_ipackets[i] = internal->rx_queue[i].rx_pkts;
 508                 rx_total += igb_stats->q_ipackets[i];
 509         }
 510
 511         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS && i < internal->nb_tx_queues;
 512                         i++) {
 513                 igb_stats->q_opackets[i] = internal->tx_queue[i].tx_pkts;
 514                 igb_stats->q_errors[i] = internal->tx_queue[i].err_pkts;
 515                 tx_total += igb_stats->q_opackets[i];
 516                 tx_err_total += igb_stats->q_errors[i];
 517         }
 518
 519         igb_stats->ipackets = rx_total;
 520         igb_stats->opackets = tx_total;
 521         igb_stats->oerrors = tx_err_total;
 522 }
 523
 524 static void
 525 eth_stats_reset(struct rte_eth_dev *dev)
 526 {
 527         unsigned i;
 528         struct pmd_internals *internal = dev->data->dev_private;
 529         for (i = 0; i < internal->nb_rx_queues; i++)
 530                 internal->rx_queue[i].rx_pkts = 0;
 531         for (i = 0; i < internal->nb_tx_queues; i++) {
 532                 internal->tx_queue[i].tx_pkts = 0;
 533                 internal->tx_queue[i].err_pkts = 0;
 534         }
 535 }
 536
 537 static void
 538 eth_dev_close(struct rte_eth_dev *dev __rte_unused)
 539 {
 540 }
 541
 542 static void
 543 eth_queue_release(void *q __rte_unused)
 544 {
 545 }
 546
 547 static int
 548 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 549                 int wait_to_complete __rte_unused)
 550 {
 551         return 0;
 552 }
 553
 554 static int
 555 eth_rx_queue_setup(struct rte_eth_dev *dev,
 556                 uint16_t rx_queue_id,
 557                 uint16_t nb_rx_desc __rte_unused,
 558                 unsigned int socket_id __rte_unused,
 559                 const struct rte_eth_rxconf *rx_conf __rte_unused,
 560                 struct rte_mempool *mb_pool)
 561 {
 562         struct pmd_internals *internals = dev->data->dev_private;
 563         struct pcap_rx_queue *pcap_q = &internals->rx_queue[rx_queue_id];
 564         pcap_q->mb_pool = mb_pool;
 565         dev->data->rx_queues[rx_queue_id] = pcap_q;
 566         pcap_q->in_port = dev->data->port_id;
 567         return 0;
 568 }
 569
 570 static int
 571 eth_tx_queue_setup(struct rte_eth_dev *dev,
 572                 uint16_t tx_queue_id,
 573                 uint16_t nb_tx_desc __rte_unused,
 574                 unsigned int socket_id __rte_unused,
 575                 const struct rte_eth_txconf *tx_conf __rte_unused)
 576 {
 577
 578         struct pmd_internals *internals = dev->data->dev_private;
 579         dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
 580         return 0;
 581 }
 582
 583 static const struct eth_dev_ops ops = {
 584         .dev_start = eth_dev_start,
 585         .dev_stop =     eth_dev_stop,
 586         .dev_close = eth_dev_close,
 587         .dev_configure = eth_dev_configure,
 588         .dev_infos_get = eth_dev_info,
 589         .rx_queue_setup = eth_rx_queue_setup,
 590         .tx_queue_setup = eth_tx_queue_setup,
 591         .rx_queue_release = eth_queue_release,
 592         .tx_queue_release = eth_queue_release,
 593         .link_update = eth_link_update,
 594         .stats_get = eth_stats_get,
 595         .stats_reset = eth_stats_reset,
 596 };
 597
 598 static struct eth_driver rte_pcap_pmd = {
 599         .pci_drv = {
 600                 .name = "rte_pcap_pmd",
 601                 .drv_flags = RTE_PCI_DRV_DETACHABLE,
 602         },
 603 };
 604
 605 /*
 606  * Function handler that opens the pcap file for reading a stores a
 607  * reference of it for use it later on.
 608  */
 609 static int
 610 open_rx_pcap(const char *key, const char *value, void *extra_args)
 611 {
 612         unsigned i;
 613         const char *pcap_filename = value;
 614         struct rx_pcaps *pcaps = extra_args;
 615         pcap_t *pcap = NULL;
 616
 617         for (i = 0; i < pcaps->num_of_rx; i++) {
 618                 if (open_single_rx_pcap(pcap_filename, &pcap) < 0)
 619                         return -1;
 620
 621                 pcaps->pcaps[i] = pcap;
 622                 pcaps->names[i] = pcap_filename;
 623                 pcaps->types[i] = key;
 624         }
 625
 626         return 0;
 627 }
 628
 629 static int
 630 open_single_rx_pcap(const char *pcap_filename, pcap_t **pcap)
 631 {
 632         if ((*pcap = pcap_open_offline(pcap_filename, errbuf)) == NULL) {
 633                 RTE_LOG(ERR, PMD, "Couldn't open %s: %s\n", pcap_filename, errbuf);
 634                 return -1;
 635         }
 636         return 0;
 637 }
 638
 639 /*
 640  * Opens a pcap file for writing and stores a reference to it
 641  * for use it later on.
 642  */
 643 static int
 644 open_tx_pcap(const char *key, const char *value, void *extra_args)
 645 {
 646         unsigned i;
 647         const char *pcap_filename = value;
 648         struct tx_pcaps *dumpers = extra_args;
 649         pcap_dumper_t *dumper;
 650
 651         for (i = 0; i < dumpers->num_of_tx; i++) {
 652                 if (open_single_tx_pcap(pcap_filename, &dumper) < 0)
 653                         return -1;
 654
 655                 dumpers->dumpers[i] = dumper;
 656                 dumpers->names[i] = pcap_filename;
 657                 dumpers->types[i] = key;
 658         }
 659
 660         return 0;
 661 }
 662
 663 static int
 664 open_single_tx_pcap(const char *pcap_filename, pcap_dumper_t **dumper)
 665 {
 666         pcap_t *tx_pcap;
 667         /*
 668          * We need to create a dummy empty pcap_t to use it
 669          * with pcap_dump_open(). We create big enough an Ethernet
 670          * pcap holder.
 671          */
 672
 673         if ((tx_pcap = pcap_open_dead(DLT_EN10MB, RTE_ETH_PCAP_SNAPSHOT_LEN))
 674                         == NULL) {
 675                 RTE_LOG(ERR, PMD, "Couldn't create dead pcap\n");
 676                 return -1;
 677         }
 678
 679         /* The dumper is created using the previous pcap_t reference */
 680         if ((*dumper = pcap_dump_open(tx_pcap, pcap_filename)) == NULL) {
 681                 RTE_LOG(ERR, PMD, "Couldn't open %s for writing.\n", pcap_filename);
 682                 return -1;
 683         }
 684
 685         return 0;
 686 }
 687
 688 /*
 689  * pcap_open_live wrapper function
 690  */
 691 static inline int
 692 open_iface_live(const char *iface, pcap_t **pcap) {
 693         *pcap = pcap_open_live(iface, RTE_ETH_PCAP_SNAPLEN,
 694                         RTE_ETH_PCAP_PROMISC, RTE_ETH_PCAP_TIMEOUT, errbuf);
 695
 696         if (*pcap == NULL) {
 697                 RTE_LOG(ERR, PMD, "Couldn't open %s: %s\n", iface, errbuf);
 698                 return -1;
 699         }
 700         return 0;
 701 }
 702
 703 /*
 704  * Opens an interface for reading and writing
 705  */
 706 static inline int
 707 open_rx_tx_iface(const char *key, const char *value, void *extra_args)
 708 {
 709         const char *iface = value;
 710         struct rx_pcaps *pcaps = extra_args;
 711         pcap_t *pcap = NULL;
 712
 713         if (open_single_iface(iface, &pcap) < 0)
 714                 return -1;
 715
 716         pcaps->pcaps[0] = pcap;
 717         pcaps->names[0] = iface;
 718         pcaps->types[0] = key;
 719
 720         return 0;
 721 }
 722
 723 /*
 724  * Opens a NIC for reading packets from it
 725  */
 726 static inline int
 727 open_rx_iface(const char *key, const char *value, void *extra_args)
 728 {
 729         unsigned i;
 730         const char *iface = value;
 731         struct rx_pcaps *pcaps = extra_args;
 732         pcap_t *pcap = NULL;
 733
 734         for (i = 0; i < pcaps->num_of_rx; i++) {
 735                 if (open_single_iface(iface, &pcap) < 0)
 736                         return -1;
 737                 pcaps->pcaps[i] = pcap;
 738                 pcaps->names[i] = iface;
 739                 pcaps->types[i] = key;
 740         }
 741
 742         return 0;
 743 }
 744
 745 /*
 746  * Opens a NIC for writing packets to it
 747  */
 748 static int
 749 open_tx_iface(const char *key, const char *value, void *extra_args)
 750 {
 751         unsigned i;
 752         const char *iface = value;
 753         struct tx_pcaps *pcaps = extra_args;
 754         pcap_t *pcap;
 755
 756         for (i = 0; i < pcaps->num_of_tx; i++) {
 757                 if (open_single_iface(iface, &pcap) < 0)
 758                         return -1;
 759                 pcaps->pcaps[i] = pcap;
 760                 pcaps->names[i] = iface;
 761                 pcaps->types[i] = key;
 762         }
 763
 764         return 0;
 765 }
 766
 767 static int
 768 open_single_iface(const char *iface, pcap_t **pcap)
 769 {
 770         if (open_iface_live(iface, pcap) < 0) {
 771                 RTE_LOG(ERR, PMD, "Couldn't open interface %s\n", iface);
 772                 return -1;
 773         }
 774
 775         return 0;
 776 }
 777
 778 static int
 779 rte_pmd_init_internals(const char *name, const unsigned nb_rx_queues,
 780                 const unsigned nb_tx_queues,
 781                 const unsigned numa_node,
 782                 struct pmd_internals **internals,
 783                 struct rte_eth_dev **eth_dev,
 784                 struct rte_kvargs *kvlist)
 785 {
 786         struct rte_eth_dev_data *data = NULL;
 787         struct rte_pci_device *pci_dev = NULL;
 788         unsigned k_idx;
 789         struct rte_kvargs_pair *pair = NULL;
 790
 791         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 792                 pair = &kvlist->pairs[k_idx];
 793                 if (strstr(pair->key, ETH_PCAP_IFACE_ARG) != NULL)
 794                         break;
 795         }
 796
 797         RTE_LOG(INFO, PMD,
 798                         "Creating pcap-backed ethdev on numa socket %u\n", numa_node);
 799
 800         /* now do all data allocation - for eth_dev structure, dummy pci driver
 801          * and internal (private) data
 802          */
 803         data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
 804         if (data == NULL)
 805                 goto error;
 806
 807         pci_dev = rte_zmalloc_socket(name, sizeof(*pci_dev), 0, numa_node);
 808         if (pci_dev == NULL)
 809                 goto error;
 810
 811         *internals = rte_zmalloc_socket(name, sizeof(**internals), 0, numa_node);
 812         if (*internals == NULL)
 813                 goto error;
 814
 815         /* reserve an ethdev entry */
 816         *eth_dev = rte_eth_dev_allocate(name, RTE_ETH_DEV_VIRTUAL);
 817         if (*eth_dev == NULL)
 818                 goto error;
 819
 820         /* check length of device name */
 821         if ((strlen((*eth_dev)->data->name) + 1) > sizeof(data->name))
 822                 goto error;
 823
 824         /* now put it all together
 825          * - store queue data in internals,
 826          * - store numa_node info in pci_driver
 827          * - point eth_dev_data to internals and pci_driver
 828          * - and point eth_dev structure to new eth_dev_data structure
 829          */
 830         /* NOTE: we'll replace the data element, of originally allocated eth_dev
 831          * so the rings are local per-process */
 832
 833         (*internals)->nb_rx_queues = nb_rx_queues;
 834         (*internals)->nb_tx_queues = nb_tx_queues;
 835
 836         if (pair == NULL)
 837                 (*internals)->if_index = 0;
 838         else
 839                 (*internals)->if_index = if_nametoindex(pair->value);
 840
 841         pci_dev->numa_node = numa_node;
 842
 843         data->dev_private = *internals;
 844         data->port_id = (*eth_dev)->data->port_id;
 845         snprintf(data->name, sizeof(data->name), "%s", (*eth_dev)->data->name);
 846         data->nb_rx_queues = (uint16_t)nb_rx_queues;
 847         data->nb_tx_queues = (uint16_t)nb_tx_queues;
 848         data->dev_link = pmd_link;
 849         data->mac_addrs = &eth_addr;
 850         strncpy(data->name,
 851                 (*eth_dev)->data->name, strlen((*eth_dev)->data->name));
 852
 853         (*eth_dev)->data = data;
 854         (*eth_dev)->dev_ops = &ops;
 855         (*eth_dev)->pci_dev = pci_dev;
 856         (*eth_dev)->driver = &rte_pcap_pmd;
 857
 858         return 0;
 859
 860 error:
 861         rte_free(data);
 862         rte_free(pci_dev);
 863         rte_free(*internals);
 864
 865         return -1;
 866 }
 867
 868 static int
 869 rte_eth_from_pcaps_n_dumpers(const char *name,
 870                 struct rx_pcaps *rx_queues,
 871                 const unsigned nb_rx_queues,
 872                 struct tx_pcaps *tx_queues,
 873                 const unsigned nb_tx_queues,
 874                 const unsigned numa_node,
 875                 struct rte_kvargs *kvlist)
 876 {
 877         struct pmd_internals *internals = NULL;
 878         struct rte_eth_dev *eth_dev = NULL;
 879         unsigned i;
 880
 881         /* do some parameter checking */
 882         if (rx_queues == NULL && nb_rx_queues > 0)
 883                 return -1;
 884         if (tx_queues == NULL && nb_tx_queues > 0)
 885                 return -1;
 886
 887         if (rte_pmd_init_internals(name, nb_rx_queues, nb_tx_queues, numa_node,
 888                         &internals, &eth_dev, kvlist) < 0)
 889                 return -1;
 890
 891         for (i = 0; i < nb_rx_queues; i++) {
 892                 internals->rx_queue[i].pcap = rx_queues->pcaps[i];
 893                 snprintf(internals->rx_queue[i].name,
 894                         sizeof(internals->rx_queue[i].name), "%s",
 895                         rx_queues->names[i]);
 896                 snprintf(internals->rx_queue[i].type,
 897                         sizeof(internals->rx_queue[i].type), "%s",
 898                         rx_queues->types[i]);
 899         }
 900         for (i = 0; i < nb_tx_queues; i++) {
 901                 internals->tx_queue[i].dumper = tx_queues->dumpers[i];
 902                 snprintf(internals->tx_queue[i].name,
 903                         sizeof(internals->tx_queue[i].name), "%s",
 904                         tx_queues->names[i]);
 905                 snprintf(internals->tx_queue[i].type,
 906                         sizeof(internals->tx_queue[i].type), "%s",
 907                         tx_queues->types[i]);
 908         }
 909
 910         /* using multiple pcaps/interfaces */
 911         internals->single_iface = 0;
 912
 913         eth_dev->rx_pkt_burst = eth_pcap_rx;
 914         eth_dev->tx_pkt_burst = eth_pcap_tx_dumper;
 915
 916         return 0;
 917 }
 918
 919 static int
 920 rte_eth_from_pcaps(const char *name,
 921                 struct rx_pcaps *rx_queues,
 922                 const unsigned nb_rx_queues,
 923                 struct tx_pcaps *tx_queues,
 924                 const unsigned nb_tx_queues,
 925                 const unsigned numa_node,
 926                 struct rte_kvargs *kvlist,
 927                 int single_iface)
 928 {
 929         struct pmd_internals *internals = NULL;
 930         struct rte_eth_dev *eth_dev = NULL;
 931         unsigned i;
 932
 933         /* do some parameter checking */
 934         if (rx_queues == NULL && nb_rx_queues > 0)
 935                 return -1;
 936         if (tx_queues == NULL && nb_tx_queues > 0)
 937                 return -1;
 938
 939         if (rte_pmd_init_internals(name, nb_rx_queues, nb_tx_queues, numa_node,
 940                         &internals, &eth_dev, kvlist) < 0)
 941                 return -1;
 942
 943         for (i = 0; i < nb_rx_queues; i++) {
 944                 internals->rx_queue[i].pcap = rx_queues->pcaps[i];
 945                 snprintf(internals->rx_queue[i].name,
 946                         sizeof(internals->rx_queue[i].name), "%s",
 947                         rx_queues->names[i]);
 948                 snprintf(internals->rx_queue[i].type,
 949                         sizeof(internals->rx_queue[i].type), "%s",
 950                         rx_queues->types[i]);
 951         }
 952         for (i = 0; i < nb_tx_queues; i++) {
 953                 internals->tx_queue[i].dumper = tx_queues->dumpers[i];
 954                 snprintf(internals->tx_queue[i].name,
 955                         sizeof(internals->tx_queue[i].name), "%s",
 956                         tx_queues->names[i]);
 957                 snprintf(internals->tx_queue[i].type,
 958                         sizeof(internals->tx_queue[i].type), "%s",
 959                         tx_queues->types[i]);
 960         }
 961
 962         /* store wether we are using a single interface for rx/tx or not */
 963         internals->single_iface = single_iface;
 964
 965         eth_dev->rx_pkt_burst = eth_pcap_rx;
 966         eth_dev->tx_pkt_burst = eth_pcap_tx;
 967
 968         return 0;
 969 }
 970
 971
 972 static int
 973 rte_pmd_pcap_devinit(const char *name, const char *params)
 974 {
 975         unsigned numa_node, using_dumpers = 0;
 976         int ret;
 977         struct rte_kvargs *kvlist;
 978         struct rx_pcaps pcaps;
 979         struct tx_pcaps dumpers;
 980
 981         RTE_LOG(INFO, PMD, "Initializing pmd_pcap for %s\n", name);
 982
 983         numa_node = rte_socket_id();
 984
 985         gettimeofday(&start_time, NULL);
 986         start_cycles = rte_get_timer_cycles();
 987         hz = rte_get_timer_hz();
 988
 989         kvlist = rte_kvargs_parse(params, valid_arguments);
 990         if (kvlist == NULL)
 991                 return -1;
 992
 993         /*
 994          * If iface argument is passed we open the NICs and use them for
 995          * reading / writing
 996          */
 997         if (rte_kvargs_count(kvlist, ETH_PCAP_IFACE_ARG) == 1) {
 998
 999                 ret = rte_kvargs_process(kvlist, ETH_PCAP_IFACE_ARG,
1000                                 &open_rx_tx_iface, &pcaps);
1001                 if (ret < 0)
1002                         goto free_kvlist;
1003                 dumpers.pcaps[0] = pcaps.pcaps[0];
1004                 dumpers.names[0] = pcaps.names[0];
1005                 dumpers.types[0] = pcaps.types[0];
1006                 ret = rte_eth_from_pcaps(name, &pcaps, 1, &dumpers, 1,
1007                                 numa_node, kvlist, 1);
1008                 goto free_kvlist;
1009         }
1010
1011         /*
1012          * We check whether we want to open a RX stream from a real NIC or a
1013          * pcap file
1014          */
1015         if ((pcaps.num_of_rx = rte_kvargs_count(kvlist, ETH_PCAP_RX_PCAP_ARG))) {
1016                 ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_PCAP_ARG,
1017                                 &open_rx_pcap, &pcaps);
1018         } else {
1019                 pcaps.num_of_rx = rte_kvargs_count(kvlist,
1020                                 ETH_PCAP_RX_IFACE_ARG);
1021                 ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_IFACE_ARG,
1022                                 &open_rx_iface, &pcaps);
1023         }
1024
1025         if (ret < 0)
1026                 goto free_kvlist;
1027
1028         /*
1029          * We check whether we want to open a TX stream to a real NIC or a
1030          * pcap file
1031          */
1032         if ((dumpers.num_of_tx = rte_kvargs_count(kvlist,
1033                         ETH_PCAP_TX_PCAP_ARG))) {
1034                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_PCAP_ARG,
1035                                 &open_tx_pcap, &dumpers);
1036                 using_dumpers = 1;
1037         } else {
1038                 dumpers.num_of_tx = rte_kvargs_count(kvlist,
1039                                 ETH_PCAP_TX_IFACE_ARG);
1040                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_IFACE_ARG,
1041                                 &open_tx_iface, &dumpers);
1042         }
1043
1044         if (ret < 0)
1045                 goto free_kvlist;
1046
1047         if (using_dumpers)
1048                 ret = rte_eth_from_pcaps_n_dumpers(name, &pcaps, pcaps.num_of_rx,
1049                                 &dumpers, dumpers.num_of_tx, numa_node, kvlist);
1050         else
1051                 ret = rte_eth_from_pcaps(name, &pcaps, pcaps.num_of_rx, &dumpers,
1052                         dumpers.num_of_tx, numa_node, kvlist, 0);
1053
1054 free_kvlist:
1055         rte_kvargs_free(kvlist);
1056         return ret;
1057 }
1058
1059 static int
1060 rte_pmd_pcap_devuninit(const char *name)
1061 {
1062         struct rte_eth_dev *eth_dev = NULL;
1063
1064         RTE_LOG(INFO, PMD, "Closing pcap ethdev on numa socket %u\n",
1065                         rte_socket_id());
1066
1067         if (name == NULL)
1068                 return -1;
1069
1070         /* reserve an ethdev entry */
1071         eth_dev = rte_eth_dev_allocated(name);
1072         if (eth_dev == NULL)
1073                 return -1;
1074
1075         rte_free(eth_dev->data->dev_private);
1076         rte_free(eth_dev->data);
1077         rte_free(eth_dev->pci_dev);
1078
1079         rte_eth_dev_release_port(eth_dev);
1080
1081         return 0;
1082 }
1083
1084 static struct rte_driver pmd_pcap_drv = {
1085         .name = "eth_pcap",
1086         .type = PMD_VDEV,
1087         .init = rte_pmd_pcap_devinit,
1088         .uninit = rte_pmd_pcap_devuninit,
1089 };
1090
1091 PMD_REGISTER_DRIVER(pmd_pcap_drv);