drivers/net/pcap/rte_eth_pcap.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2016 Intel Corporation.
   3  * Copyright(c) 2014 6WIND S.A.
   4  * All rights reserved.
   5  */
   6
   7 #include <time.h>
   8
   9 #include <net/if.h>
  10
  11 #include <pcap.h>
  12
  13 #include <rte_cycles.h>
  14 #include <rte_ethdev_driver.h>
  15 #include <rte_ethdev_vdev.h>
  16 #include <rte_kvargs.h>
  17 #include <rte_malloc.h>
  18 #include <rte_mbuf.h>
  19 #include <rte_bus_vdev.h>
  20
  21 #define RTE_ETH_PCAP_SNAPSHOT_LEN 65535
  22 #define RTE_ETH_PCAP_SNAPLEN ETHER_MAX_JUMBO_FRAME_LEN
  23 #define RTE_ETH_PCAP_PROMISC 1
  24 #define RTE_ETH_PCAP_TIMEOUT -1
  25
  26 #define ETH_PCAP_RX_PCAP_ARG  "rx_pcap"
  27 #define ETH_PCAP_TX_PCAP_ARG  "tx_pcap"
  28 #define ETH_PCAP_RX_IFACE_ARG "rx_iface"
  29 #define ETH_PCAP_RX_IFACE_IN_ARG "rx_iface_in"
  30 #define ETH_PCAP_TX_IFACE_ARG "tx_iface"
  31 #define ETH_PCAP_IFACE_ARG    "iface"
  32
  33 #define ETH_PCAP_ARG_MAXLEN     64
  34
  35 #define RTE_PMD_PCAP_MAX_QUEUES 16
  36
  37 static char errbuf[PCAP_ERRBUF_SIZE];
  38 static unsigned char tx_pcap_data[RTE_ETH_PCAP_SNAPLEN];
  39 static struct timeval start_time;
  40 static uint64_t start_cycles;
  41 static uint64_t hz;
  42
  43 struct queue_stat {
  44         volatile unsigned long pkts;
  45         volatile unsigned long bytes;
  46         volatile unsigned long err_pkts;
  47 };
  48
  49 struct pcap_rx_queue {
  50         pcap_t *pcap;
  51         uint16_t in_port;
  52         struct rte_mempool *mb_pool;
  53         struct queue_stat rx_stat;
  54         char name[PATH_MAX];
  55         char type[ETH_PCAP_ARG_MAXLEN];
  56 };
  57
  58 struct pcap_tx_queue {
  59         pcap_dumper_t *dumper;
  60         pcap_t *pcap;
  61         struct queue_stat tx_stat;
  62         char name[PATH_MAX];
  63         char type[ETH_PCAP_ARG_MAXLEN];
  64 };
  65
  66 struct pmd_internals {
  67         struct pcap_rx_queue rx_queue[RTE_PMD_PCAP_MAX_QUEUES];
  68         struct pcap_tx_queue tx_queue[RTE_PMD_PCAP_MAX_QUEUES];
  69         int if_index;
  70         int single_iface;
  71 };
  72
  73 struct pmd_devargs {
  74         unsigned int num_of_queue;
  75         struct devargs_queue {
  76                 pcap_dumper_t *dumper;
  77                 pcap_t *pcap;
  78                 const char *name;
  79                 const char *type;
  80         } queue[RTE_PMD_PCAP_MAX_QUEUES];
  81 };
  82
  83 static const char *valid_arguments[] = {
  84         ETH_PCAP_RX_PCAP_ARG,
  85         ETH_PCAP_TX_PCAP_ARG,
  86         ETH_PCAP_RX_IFACE_ARG,
  87         ETH_PCAP_RX_IFACE_IN_ARG,
  88         ETH_PCAP_TX_IFACE_ARG,
  89         ETH_PCAP_IFACE_ARG,
  90         NULL
  91 };
  92
  93 static struct ether_addr eth_addr = {
  94         .addr_bytes = { 0, 0, 0, 0x1, 0x2, 0x3 }
  95 };
  96
  97 static struct rte_eth_link pmd_link = {
  98                 .link_speed = ETH_SPEED_NUM_10G,
  99                 .link_duplex = ETH_LINK_FULL_DUPLEX,
 100                 .link_status = ETH_LINK_DOWN,
 101                 .link_autoneg = ETH_LINK_FIXED,
 102 };
 103
 104 static int eth_pcap_logtype;
 105
 106 #define PMD_LOG(level, fmt, args...) \
 107         rte_log(RTE_LOG_ ## level, eth_pcap_logtype, \
 108                 "%s(): " fmt "\n", __func__, ##args)
 109
 110 static int
 111 eth_pcap_rx_jumbo(struct rte_mempool *mb_pool, struct rte_mbuf *mbuf,
 112                 const u_char *data, uint16_t data_len)
 113 {
 114         /* Copy the first segment. */
 115         uint16_t len = rte_pktmbuf_tailroom(mbuf);
 116         struct rte_mbuf *m = mbuf;
 117
 118         rte_memcpy(rte_pktmbuf_append(mbuf, len), data, len);
 119         data_len -= len;
 120         data += len;
 121
 122         while (data_len > 0) {
 123                 /* Allocate next mbuf and point to that. */
 124                 m->next = rte_pktmbuf_alloc(mb_pool);
 125
 126                 if (unlikely(!m->next))
 127                         return -1;
 128
 129                 m = m->next;
 130
 131                 /* Headroom is not needed in chained mbufs. */
 132                 rte_pktmbuf_prepend(m, rte_pktmbuf_headroom(m));
 133                 m->pkt_len = 0;
 134                 m->data_len = 0;
 135
 136                 /* Copy next segment. */
 137                 len = RTE_MIN(rte_pktmbuf_tailroom(m), data_len);
 138                 rte_memcpy(rte_pktmbuf_append(m, len), data, len);
 139
 140                 mbuf->nb_segs++;
 141                 data_len -= len;
 142                 data += len;
 143         }
 144
 145         return mbuf->nb_segs;
 146 }
 147
 148 /* Copy data from mbuf chain to a buffer suitable for writing to a PCAP file. */
 149 static void
 150 eth_pcap_gather_data(unsigned char *data, struct rte_mbuf *mbuf)
 151 {
 152         uint16_t data_len = 0;
 153
 154         while (mbuf) {
 155                 rte_memcpy(data + data_len, rte_pktmbuf_mtod(mbuf, void *),
 156                         mbuf->data_len);
 157
 158                 data_len += mbuf->data_len;
 159                 mbuf = mbuf->next;
 160         }
 161 }
 162
 163 static uint16_t
 164 eth_pcap_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 165 {
 166         unsigned int i;
 167         struct pcap_pkthdr header;
 168         const u_char *packet;
 169         struct rte_mbuf *mbuf;
 170         struct pcap_rx_queue *pcap_q = queue;
 171         uint16_t num_rx = 0;
 172         uint16_t buf_size;
 173         uint32_t rx_bytes = 0;
 174
 175         if (unlikely(pcap_q->pcap == NULL || nb_pkts == 0))
 176                 return 0;
 177
 178         /* Reads the given number of packets from the pcap file one by one
 179          * and copies the packet data into a newly allocated mbuf to return.
 180          */
 181         for (i = 0; i < nb_pkts; i++) {
 182                 /* Get the next PCAP packet */
 183                 packet = pcap_next(pcap_q->pcap, &header);
 184                 if (unlikely(packet == NULL))
 185                         break;
 186
 187                 mbuf = rte_pktmbuf_alloc(pcap_q->mb_pool);
 188                 if (unlikely(mbuf == NULL))
 189                         break;
 190
 191                 /* Now get the space available for data in the mbuf */
 192                 buf_size = rte_pktmbuf_data_room_size(pcap_q->mb_pool) -
 193                                 RTE_PKTMBUF_HEADROOM;
 194
 195                 if (header.caplen <= buf_size) {
 196                         /* pcap packet will fit in the mbuf, can copy it */
 197                         rte_memcpy(rte_pktmbuf_mtod(mbuf, void *), packet,
 198                                         header.caplen);
 199                         mbuf->data_len = (uint16_t)header.caplen;
 200                 } else {
 201                         /* Try read jumbo frame into multi mbufs. */
 202                         if (unlikely(eth_pcap_rx_jumbo(pcap_q->mb_pool,
 203                                                        mbuf,
 204                                                        packet,
 205                                                        header.caplen) == -1)) {
 206                                 rte_pktmbuf_free(mbuf);
 207                                 break;
 208                         }
 209                 }
 210
 211                 mbuf->pkt_len = (uint16_t)header.caplen;
 212                 mbuf->port = pcap_q->in_port;
 213                 bufs[num_rx] = mbuf;
 214                 num_rx++;
 215                 rx_bytes += header.caplen;
 216         }
 217         pcap_q->rx_stat.pkts += num_rx;
 218         pcap_q->rx_stat.bytes += rx_bytes;
 219
 220         return num_rx;
 221 }
 222
 223 static inline void
 224 calculate_timestamp(struct timeval *ts) {
 225         uint64_t cycles;
 226         struct timeval cur_time;
 227
 228         cycles = rte_get_timer_cycles() - start_cycles;
 229         cur_time.tv_sec = cycles / hz;
 230         cur_time.tv_usec = (cycles % hz) * 1e6 / hz;
 231         timeradd(&start_time, &cur_time, ts);
 232 }
 233
 234 /*
 235  * Callback to handle writing packets to a pcap file.
 236  */
 237 static uint16_t
 238 eth_pcap_tx_dumper(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 239 {
 240         unsigned int i;
 241         struct rte_mbuf *mbuf;
 242         struct pcap_tx_queue *dumper_q = queue;
 243         uint16_t num_tx = 0;
 244         uint32_t tx_bytes = 0;
 245         struct pcap_pkthdr header;
 246
 247         if (dumper_q->dumper == NULL || nb_pkts == 0)
 248                 return 0;
 249
 250         /* writes the nb_pkts packets to the previously opened pcap file
 251          * dumper */
 252         for (i = 0; i < nb_pkts; i++) {
 253                 mbuf = bufs[i];
 254                 calculate_timestamp(&header.ts);
 255                 header.len = mbuf->pkt_len;
 256                 header.caplen = header.len;
 257
 258                 if (likely(mbuf->nb_segs == 1)) {
 259                         pcap_dump((u_char *)dumper_q->dumper, &header,
 260                                   rte_pktmbuf_mtod(mbuf, void*));
 261                 } else {
 262                         if (mbuf->pkt_len <= ETHER_MAX_JUMBO_FRAME_LEN) {
 263                                 eth_pcap_gather_data(tx_pcap_data, mbuf);
 264                                 pcap_dump((u_char *)dumper_q->dumper, &header,
 265                                           tx_pcap_data);
 266                         } else {
 267                                 PMD_LOG(ERR,
 268                                         "Dropping PCAP packet. Size (%d) > max jumbo size (%d).",
 269                                         mbuf->pkt_len,
 270                                         ETHER_MAX_JUMBO_FRAME_LEN);
 271
 272                                 rte_pktmbuf_free(mbuf);
 273                                 break;
 274                         }
 275                 }
 276
 277                 num_tx++;
 278                 tx_bytes += mbuf->pkt_len;
 279                 rte_pktmbuf_free(mbuf);
 280         }
 281
 282         /*
 283          * Since there's no place to hook a callback when the forwarding
 284          * process stops and to make sure the pcap file is actually written,
 285          * we flush the pcap dumper within each burst.
 286          */
 287         pcap_dump_flush(dumper_q->dumper);
 288         dumper_q->tx_stat.pkts += num_tx;
 289         dumper_q->tx_stat.bytes += tx_bytes;
 290         dumper_q->tx_stat.err_pkts += nb_pkts - num_tx;
 291
 292         return num_tx;
 293 }
 294
 295 /*
 296  * Callback to handle sending packets through a real NIC.
 297  */
 298 static uint16_t
 299 eth_pcap_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 300 {
 301         unsigned int i;
 302         int ret;
 303         struct rte_mbuf *mbuf;
 304         struct pcap_tx_queue *tx_queue = queue;
 305         uint16_t num_tx = 0;
 306         uint32_t tx_bytes = 0;
 307
 308         if (unlikely(nb_pkts == 0 || tx_queue->pcap == NULL))
 309                 return 0;
 310
 311         for (i = 0; i < nb_pkts; i++) {
 312                 mbuf = bufs[i];
 313
 314                 if (likely(mbuf->nb_segs == 1)) {
 315                         ret = pcap_sendpacket(tx_queue->pcap,
 316                                         rte_pktmbuf_mtod(mbuf, u_char *),
 317                                         mbuf->pkt_len);
 318                 } else {
 319                         if (mbuf->pkt_len <= ETHER_MAX_JUMBO_FRAME_LEN) {
 320                                 eth_pcap_gather_data(tx_pcap_data, mbuf);
 321                                 ret = pcap_sendpacket(tx_queue->pcap,
 322                                                 tx_pcap_data, mbuf->pkt_len);
 323                         } else {
 324                                 PMD_LOG(ERR,
 325                                         "Dropping PCAP packet. Size (%d) > max jumbo size (%d).",
 326                                         mbuf->pkt_len,
 327                                         ETHER_MAX_JUMBO_FRAME_LEN);
 328
 329                                 rte_pktmbuf_free(mbuf);
 330                                 break;
 331                         }
 332                 }
 333
 334                 if (unlikely(ret != 0))
 335                         break;
 336                 num_tx++;
 337                 tx_bytes += mbuf->pkt_len;
 338                 rte_pktmbuf_free(mbuf);
 339         }
 340
 341         tx_queue->tx_stat.pkts += num_tx;
 342         tx_queue->tx_stat.bytes += tx_bytes;
 343         tx_queue->tx_stat.err_pkts += nb_pkts - num_tx;
 344
 345         return num_tx;
 346 }
 347
 348 /*
 349  * pcap_open_live wrapper function
 350  */
 351 static inline int
 352 open_iface_live(const char *iface, pcap_t **pcap) {
 353         *pcap = pcap_open_live(iface, RTE_ETH_PCAP_SNAPLEN,
 354                         RTE_ETH_PCAP_PROMISC, RTE_ETH_PCAP_TIMEOUT, errbuf);
 355
 356         if (*pcap == NULL) {
 357                 PMD_LOG(ERR, "Couldn't open %s: %s", iface, errbuf);
 358                 return -1;
 359         }
 360
 361         return 0;
 362 }
 363
 364 static int
 365 open_single_iface(const char *iface, pcap_t **pcap)
 366 {
 367         if (open_iface_live(iface, pcap) < 0) {
 368                 PMD_LOG(ERR, "Couldn't open interface %s", iface);
 369                 return -1;
 370         }
 371
 372         return 0;
 373 }
 374
 375 static int
 376 open_single_tx_pcap(const char *pcap_filename, pcap_dumper_t **dumper)
 377 {
 378         pcap_t *tx_pcap;
 379
 380         /*
 381          * We need to create a dummy empty pcap_t to use it
 382          * with pcap_dump_open(). We create big enough an Ethernet
 383          * pcap holder.
 384          */
 385         tx_pcap = pcap_open_dead(DLT_EN10MB, RTE_ETH_PCAP_SNAPSHOT_LEN);
 386         if (tx_pcap == NULL) {
 387                 PMD_LOG(ERR, "Couldn't create dead pcap");
 388                 return -1;
 389         }
 390
 391         /* The dumper is created using the previous pcap_t reference */
 392         *dumper = pcap_dump_open(tx_pcap, pcap_filename);
 393         if (*dumper == NULL) {
 394                 pcap_close(tx_pcap);
 395                 PMD_LOG(ERR, "Couldn't open %s for writing.",
 396                         pcap_filename);
 397                 return -1;
 398         }
 399
 400         pcap_close(tx_pcap);
 401         return 0;
 402 }
 403
 404 static int
 405 open_single_rx_pcap(const char *pcap_filename, pcap_t **pcap)
 406 {
 407         *pcap = pcap_open_offline(pcap_filename, errbuf);
 408         if (*pcap == NULL) {
 409                 PMD_LOG(ERR, "Couldn't open %s: %s", pcap_filename,
 410                         errbuf);
 411                 return -1;
 412         }
 413
 414         return 0;
 415 }
 416
 417 static int
 418 eth_dev_start(struct rte_eth_dev *dev)
 419 {
 420         unsigned int i;
 421         struct pmd_internals *internals = dev->data->dev_private;
 422         struct pcap_tx_queue *tx;
 423         struct pcap_rx_queue *rx;
 424
 425         /* Special iface case. Single pcap is open and shared between tx/rx. */
 426         if (internals->single_iface) {
 427                 tx = &internals->tx_queue[0];
 428                 rx = &internals->rx_queue[0];
 429
 430                 if (!tx->pcap && strcmp(tx->type, ETH_PCAP_IFACE_ARG) == 0) {
 431                         if (open_single_iface(tx->name, &tx->pcap) < 0)
 432                                 return -1;
 433                         rx->pcap = tx->pcap;
 434                 }
 435                 goto status_up;
 436         }
 437
 438         /* If not open already, open tx pcaps/dumpers */
 439         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 440                 tx = &internals->tx_queue[i];
 441
 442                 if (!tx->dumper &&
 443                                 strcmp(tx->type, ETH_PCAP_TX_PCAP_ARG) == 0) {
 444                         if (open_single_tx_pcap(tx->name, &tx->dumper) < 0)
 445                                 return -1;
 446                 } else if (!tx->pcap &&
 447                                 strcmp(tx->type, ETH_PCAP_TX_IFACE_ARG) == 0) {
 448                         if (open_single_iface(tx->name, &tx->pcap) < 0)
 449                                 return -1;
 450                 }
 451         }
 452
 453         /* If not open already, open rx pcaps */
 454         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 455                 rx = &internals->rx_queue[i];
 456
 457                 if (rx->pcap != NULL)
 458                         continue;
 459
 460                 if (strcmp(rx->type, ETH_PCAP_RX_PCAP_ARG) == 0) {
 461                         if (open_single_rx_pcap(rx->name, &rx->pcap) < 0)
 462                                 return -1;
 463                 } else if (strcmp(rx->type, ETH_PCAP_RX_IFACE_ARG) == 0) {
 464                         if (open_single_iface(rx->name, &rx->pcap) < 0)
 465                                 return -1;
 466                 }
 467         }
 468
 469 status_up:
 470         dev->data->dev_link.link_status = ETH_LINK_UP;
 471
 472         return 0;
 473 }
 474
 475 /*
 476  * This function gets called when the current port gets stopped.
 477  * Is the only place for us to close all the tx streams dumpers.
 478  * If not called the dumpers will be flushed within each tx burst.
 479  */
 480 static void
 481 eth_dev_stop(struct rte_eth_dev *dev)
 482 {
 483         unsigned int i;
 484         struct pmd_internals *internals = dev->data->dev_private;
 485         struct pcap_tx_queue *tx;
 486         struct pcap_rx_queue *rx;
 487
 488         /* Special iface case. Single pcap is open and shared between tx/rx. */
 489         if (internals->single_iface) {
 490                 tx = &internals->tx_queue[0];
 491                 rx = &internals->rx_queue[0];
 492                 pcap_close(tx->pcap);
 493                 tx->pcap = NULL;
 494                 rx->pcap = NULL;
 495                 goto status_down;
 496         }
 497
 498         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 499                 tx = &internals->tx_queue[i];
 500
 501                 if (tx->dumper != NULL) {
 502                         pcap_dump_close(tx->dumper);
 503                         tx->dumper = NULL;
 504                 }
 505
 506                 if (tx->pcap != NULL) {
 507                         pcap_close(tx->pcap);
 508                         tx->pcap = NULL;
 509                 }
 510         }
 511
 512         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 513                 rx = &internals->rx_queue[i];
 514
 515                 if (rx->pcap != NULL) {
 516                         pcap_close(rx->pcap);
 517                         rx->pcap = NULL;
 518                 }
 519         }
 520
 521 status_down:
 522         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 523 }
 524
 525 static int
 526 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 527 {
 528         return 0;
 529 }
 530
 531 static void
 532 eth_dev_info(struct rte_eth_dev *dev,
 533                 struct rte_eth_dev_info *dev_info)
 534 {
 535         struct pmd_internals *internals = dev->data->dev_private;
 536
 537         dev_info->if_index = internals->if_index;
 538         dev_info->max_mac_addrs = 1;
 539         dev_info->max_rx_pktlen = (uint32_t) -1;
 540         dev_info->max_rx_queues = dev->data->nb_rx_queues;
 541         dev_info->max_tx_queues = dev->data->nb_tx_queues;
 542         dev_info->min_rx_bufsize = 0;
 543         dev_info->rx_offload_capa = DEV_RX_OFFLOAD_CRC_STRIP;
 544 }
 545
 546 static int
 547 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 548 {
 549         unsigned int i;
 550         unsigned long rx_packets_total = 0, rx_bytes_total = 0;
 551         unsigned long tx_packets_total = 0, tx_bytes_total = 0;
 552         unsigned long tx_packets_err_total = 0;
 553         const struct pmd_internals *internal = dev->data->dev_private;
 554
 555         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 556                         i < dev->data->nb_rx_queues; i++) {
 557                 stats->q_ipackets[i] = internal->rx_queue[i].rx_stat.pkts;
 558                 stats->q_ibytes[i] = internal->rx_queue[i].rx_stat.bytes;
 559                 rx_packets_total += stats->q_ipackets[i];
 560                 rx_bytes_total += stats->q_ibytes[i];
 561         }
 562
 563         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 564                         i < dev->data->nb_tx_queues; i++) {
 565                 stats->q_opackets[i] = internal->tx_queue[i].tx_stat.pkts;
 566                 stats->q_obytes[i] = internal->tx_queue[i].tx_stat.bytes;
 567                 stats->q_errors[i] = internal->tx_queue[i].tx_stat.err_pkts;
 568                 tx_packets_total += stats->q_opackets[i];
 569                 tx_bytes_total += stats->q_obytes[i];
 570                 tx_packets_err_total += stats->q_errors[i];
 571         }
 572
 573         stats->ipackets = rx_packets_total;
 574         stats->ibytes = rx_bytes_total;
 575         stats->opackets = tx_packets_total;
 576         stats->obytes = tx_bytes_total;
 577         stats->oerrors = tx_packets_err_total;
 578
 579         return 0;
 580 }
 581
 582 static void
 583 eth_stats_reset(struct rte_eth_dev *dev)
 584 {
 585         unsigned int i;
 586         struct pmd_internals *internal = dev->data->dev_private;
 587
 588         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 589                 internal->rx_queue[i].rx_stat.pkts = 0;
 590                 internal->rx_queue[i].rx_stat.bytes = 0;
 591         }
 592
 593         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 594                 internal->tx_queue[i].tx_stat.pkts = 0;
 595                 internal->tx_queue[i].tx_stat.bytes = 0;
 596                 internal->tx_queue[i].tx_stat.err_pkts = 0;
 597         }
 598 }
 599
 600 static void
 601 eth_dev_close(struct rte_eth_dev *dev __rte_unused)
 602 {
 603 }
 604
 605 static void
 606 eth_queue_release(void *q __rte_unused)
 607 {
 608 }
 609
 610 static int
 611 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 612                 int wait_to_complete __rte_unused)
 613 {
 614         return 0;
 615 }
 616
 617 static int
 618 eth_rx_queue_setup(struct rte_eth_dev *dev,
 619                 uint16_t rx_queue_id,
 620                 uint16_t nb_rx_desc __rte_unused,
 621                 unsigned int socket_id __rte_unused,
 622                 const struct rte_eth_rxconf *rx_conf __rte_unused,
 623                 struct rte_mempool *mb_pool)
 624 {
 625         struct pmd_internals *internals = dev->data->dev_private;
 626         struct pcap_rx_queue *pcap_q = &internals->rx_queue[rx_queue_id];
 627
 628         pcap_q->mb_pool = mb_pool;
 629         dev->data->rx_queues[rx_queue_id] = pcap_q;
 630         pcap_q->in_port = dev->data->port_id;
 631
 632         return 0;
 633 }
 634
 635 static int
 636 eth_tx_queue_setup(struct rte_eth_dev *dev,
 637                 uint16_t tx_queue_id,
 638                 uint16_t nb_tx_desc __rte_unused,
 639                 unsigned int socket_id __rte_unused,
 640                 const struct rte_eth_txconf *tx_conf __rte_unused)
 641 {
 642         struct pmd_internals *internals = dev->data->dev_private;
 643
 644         dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
 645
 646         return 0;
 647 }
 648
 649 static const struct eth_dev_ops ops = {
 650         .dev_start = eth_dev_start,
 651         .dev_stop = eth_dev_stop,
 652         .dev_close = eth_dev_close,
 653         .dev_configure = eth_dev_configure,
 654         .dev_infos_get = eth_dev_info,
 655         .rx_queue_setup = eth_rx_queue_setup,
 656         .tx_queue_setup = eth_tx_queue_setup,
 657         .rx_queue_release = eth_queue_release,
 658         .tx_queue_release = eth_queue_release,
 659         .link_update = eth_link_update,
 660         .stats_get = eth_stats_get,
 661         .stats_reset = eth_stats_reset,
 662 };
 663
 664 static int
 665 add_queue(struct pmd_devargs *pmd, const char *name, const char *type,
 666                 pcap_t *pcap, pcap_dumper_t *dumper)
 667 {
 668         if (pmd->num_of_queue >= RTE_PMD_PCAP_MAX_QUEUES)
 669                 return -1;
 670         if (pcap)
 671                 pmd->queue[pmd->num_of_queue].pcap = pcap;
 672         if (dumper)
 673                 pmd->queue[pmd->num_of_queue].dumper = dumper;
 674         pmd->queue[pmd->num_of_queue].name = name;
 675         pmd->queue[pmd->num_of_queue].type = type;
 676         pmd->num_of_queue++;
 677         return 0;
 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         const char *pcap_filename = value;
 688         struct pmd_devargs *rx = extra_args;
 689         pcap_t *pcap = NULL;
 690
 691         if (open_single_rx_pcap(pcap_filename, &pcap) < 0)
 692                 return -1;
 693
 694         if (add_queue(rx, pcap_filename, key, pcap, NULL) < 0) {
 695                 pcap_close(pcap);
 696                 return -1;
 697         }
 698
 699         return 0;
 700 }
 701
 702 /*
 703  * Opens a pcap file for writing and stores a reference to it
 704  * for use it later on.
 705  */
 706 static int
 707 open_tx_pcap(const char *key, const char *value, void *extra_args)
 708 {
 709         const char *pcap_filename = value;
 710         struct pmd_devargs *dumpers = extra_args;
 711         pcap_dumper_t *dumper;
 712
 713         if (open_single_tx_pcap(pcap_filename, &dumper) < 0)
 714                 return -1;
 715
 716         if (add_queue(dumpers, pcap_filename, key, NULL, dumper) < 0) {
 717                 pcap_dump_close(dumper);
 718                 return -1;
 719         }
 720
 721         return 0;
 722 }
 723
 724 /*
 725  * Opens an interface for reading and writing
 726  */
 727 static inline int
 728 open_rx_tx_iface(const char *key, const char *value, void *extra_args)
 729 {
 730         const char *iface = value;
 731         struct pmd_devargs *tx = extra_args;
 732         pcap_t *pcap = NULL;
 733
 734         if (open_single_iface(iface, &pcap) < 0)
 735                 return -1;
 736
 737         tx->queue[0].pcap = pcap;
 738         tx->queue[0].name = iface;
 739         tx->queue[0].type = key;
 740
 741         return 0;
 742 }
 743
 744 static inline int
 745 set_iface_direction(const char *iface, pcap_t *pcap,
 746                 pcap_direction_t direction)
 747 {
 748         const char *direction_str = (direction == PCAP_D_IN) ? "IN" : "OUT";
 749         if (pcap_setdirection(pcap, direction) < 0) {
 750                 PMD_LOG(ERR, "Setting %s pcap direction %s failed - %s\n",
 751                                 iface, direction_str, pcap_geterr(pcap));
 752                 return -1;
 753         }
 754         PMD_LOG(INFO, "Setting %s pcap direction %s\n",
 755                         iface, direction_str);
 756         return 0;
 757 }
 758
 759 static inline int
 760 open_iface(const char *key, const char *value, void *extra_args)
 761 {
 762         const char *iface = value;
 763         struct pmd_devargs *pmd = extra_args;
 764         pcap_t *pcap = NULL;
 765
 766         if (open_single_iface(iface, &pcap) < 0)
 767                 return -1;
 768         if (add_queue(pmd, iface, key, pcap, NULL) < 0) {
 769                 pcap_close(pcap);
 770                 return -1;
 771         }
 772
 773         return 0;
 774 }
 775
 776 /*
 777  * Opens a NIC for reading packets from it
 778  */
 779 static inline int
 780 open_rx_iface(const char *key, const char *value, void *extra_args)
 781 {
 782         int ret = open_iface(key, value, extra_args);
 783         if (ret < 0)
 784                 return ret;
 785         if (strcmp(key, ETH_PCAP_RX_IFACE_IN_ARG) == 0) {
 786                 struct pmd_devargs *pmd = extra_args;
 787                 unsigned int qid = pmd->num_of_queue - 1;
 788
 789                 set_iface_direction(pmd->queue[qid].name,
 790                                 pmd->queue[qid].pcap,
 791                                 PCAP_D_IN);
 792         }
 793
 794         return 0;
 795 }
 796
 797 static inline int
 798 rx_iface_args_process(const char *key, const char *value, void *extra_args)
 799 {
 800         if (strcmp(key, ETH_PCAP_RX_IFACE_ARG) == 0 ||
 801                         strcmp(key, ETH_PCAP_RX_IFACE_IN_ARG) == 0)
 802                 return open_rx_iface(key, value, extra_args);
 803
 804         return 0;
 805 }
 806
 807 /*
 808  * Opens a NIC for writing packets to it
 809  */
 810 static int
 811 open_tx_iface(const char *key, const char *value, void *extra_args)
 812 {
 813         return open_iface(key, value, extra_args);
 814 }
 815
 816 static struct rte_vdev_driver pmd_pcap_drv;
 817
 818 static int
 819 pmd_init_internals(struct rte_vdev_device *vdev,
 820                 const unsigned int nb_rx_queues,
 821                 const unsigned int nb_tx_queues,
 822                 struct pmd_internals **internals,
 823                 struct rte_eth_dev **eth_dev)
 824 {
 825         struct rte_eth_dev_data *data;
 826         unsigned int numa_node = vdev->device.numa_node;
 827
 828         PMD_LOG(INFO, "Creating pcap-backed ethdev on numa socket %d",
 829                 numa_node);
 830
 831         /* reserve an ethdev entry */
 832         *eth_dev = rte_eth_vdev_allocate(vdev, sizeof(**internals));
 833         if (!(*eth_dev))
 834                 return -1;
 835
 836         /* now put it all together
 837          * - store queue data in internals,
 838          * - store numa_node info in eth_dev
 839          * - point eth_dev_data to internals
 840          * - and point eth_dev structure to new eth_dev_data structure
 841          */
 842         *internals = (*eth_dev)->data->dev_private;
 843         data = (*eth_dev)->data;
 844         data->nb_rx_queues = (uint16_t)nb_rx_queues;
 845         data->nb_tx_queues = (uint16_t)nb_tx_queues;
 846         data->dev_link = pmd_link;
 847         data->mac_addrs = &eth_addr;
 848
 849         /*
 850          * NOTE: we'll replace the data element, of originally allocated
 851          * eth_dev so the rings are local per-process
 852          */
 853         (*eth_dev)->dev_ops = &ops;
 854
 855         return 0;
 856 }
 857
 858 static int
 859 eth_from_pcaps_common(struct rte_vdev_device *vdev,
 860                 struct pmd_devargs *rx_queues, const unsigned int nb_rx_queues,
 861                 struct pmd_devargs *tx_queues, const unsigned int nb_tx_queues,
 862                 struct rte_kvargs *kvlist, struct pmd_internals **internals,
 863                 struct rte_eth_dev **eth_dev)
 864 {
 865         struct rte_kvargs_pair *pair = NULL;
 866         unsigned int k_idx;
 867         unsigned int i;
 868
 869         /* do some parameter checking */
 870         if (rx_queues == NULL && nb_rx_queues > 0)
 871                 return -1;
 872         if (tx_queues == NULL && nb_tx_queues > 0)
 873                 return -1;
 874
 875         if (pmd_init_internals(vdev, nb_rx_queues, nb_tx_queues, internals,
 876                         eth_dev) < 0)
 877                 return -1;
 878
 879         for (i = 0; i < nb_rx_queues; i++) {
 880                 struct pcap_rx_queue *rx = &(*internals)->rx_queue[i];
 881                 struct devargs_queue *queue = &rx_queues->queue[i];
 882
 883                 rx->pcap = queue->pcap;
 884                 snprintf(rx->name, sizeof(rx->name), "%s", queue->name);
 885                 snprintf(rx->type, sizeof(rx->type), "%s", queue->type);
 886         }
 887
 888         for (i = 0; i < nb_tx_queues; i++) {
 889                 struct pcap_tx_queue *tx = &(*internals)->tx_queue[i];
 890                 struct devargs_queue *queue = &tx_queues->queue[i];
 891
 892                 tx->dumper = queue->dumper;
 893                 tx->pcap = queue->pcap;
 894                 snprintf(tx->name, sizeof(tx->name), "%s", queue->name);
 895                 snprintf(tx->type, sizeof(tx->type), "%s", queue->type);
 896         }
 897
 898         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 899                 pair = &kvlist->pairs[k_idx];
 900                 if (strstr(pair->key, ETH_PCAP_IFACE_ARG) != NULL)
 901                         break;
 902         }
 903
 904         if (pair == NULL)
 905                 (*internals)->if_index = 0;
 906         else
 907                 (*internals)->if_index = if_nametoindex(pair->value);
 908
 909         return 0;
 910 }
 911
 912 static int
 913 eth_from_pcaps(struct rte_vdev_device *vdev,
 914                 struct pmd_devargs *rx_queues, const unsigned int nb_rx_queues,
 915                 struct pmd_devargs *tx_queues, const unsigned int nb_tx_queues,
 916                 struct rte_kvargs *kvlist, int single_iface,
 917                 unsigned int using_dumpers)
 918 {
 919         struct pmd_internals *internals = NULL;
 920         struct rte_eth_dev *eth_dev = NULL;
 921         int ret;
 922
 923         ret = eth_from_pcaps_common(vdev, rx_queues, nb_rx_queues,
 924                 tx_queues, nb_tx_queues, kvlist, &internals, &eth_dev);
 925
 926         if (ret < 0)
 927                 return ret;
 928
 929         /* store weather we are using a single interface for rx/tx or not */
 930         internals->single_iface = single_iface;
 931
 932         eth_dev->rx_pkt_burst = eth_pcap_rx;
 933
 934         if (using_dumpers)
 935                 eth_dev->tx_pkt_burst = eth_pcap_tx_dumper;
 936         else
 937                 eth_dev->tx_pkt_burst = eth_pcap_tx;
 938
 939         rte_eth_dev_probing_finish(eth_dev);
 940         return 0;
 941 }
 942
 943 static int
 944 pmd_pcap_probe(struct rte_vdev_device *dev)
 945 {
 946         const char *name;
 947         unsigned int is_rx_pcap = 0, is_tx_pcap = 0;
 948         struct rte_kvargs *kvlist;
 949         struct pmd_devargs pcaps = {0};
 950         struct pmd_devargs dumpers = {0};
 951         struct rte_eth_dev *eth_dev;
 952         int single_iface = 0;
 953         int ret;
 954
 955         name = rte_vdev_device_name(dev);
 956         PMD_LOG(INFO, "Initializing pmd_pcap for %s", name);
 957
 958         gettimeofday(&start_time, NULL);
 959         start_cycles = rte_get_timer_cycles();
 960         hz = rte_get_timer_hz();
 961
 962         if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
 963             strlen(rte_vdev_device_args(dev)) == 0) {
 964                 eth_dev = rte_eth_dev_attach_secondary(name);
 965                 if (!eth_dev) {
 966                         PMD_LOG(ERR, "Failed to probe %s", name);
 967                         return -1;
 968                 }
 969                 /* TODO: request info from primary to set up Rx and Tx */
 970                 eth_dev->dev_ops = &ops;
 971                 rte_eth_dev_probing_finish(eth_dev);
 972                 return 0;
 973         }
 974
 975         kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_arguments);
 976         if (kvlist == NULL)
 977                 return -1;
 978
 979         /*
 980          * If iface argument is passed we open the NICs and use them for
 981          * reading / writing
 982          */
 983         if (rte_kvargs_count(kvlist, ETH_PCAP_IFACE_ARG) == 1) {
 984
 985                 ret = rte_kvargs_process(kvlist, ETH_PCAP_IFACE_ARG,
 986                                 &open_rx_tx_iface, &pcaps);
 987
 988                 if (ret < 0)
 989                         goto free_kvlist;
 990
 991                 dumpers.queue[0] = pcaps.queue[0];
 992
 993                 single_iface = 1;
 994                 pcaps.num_of_queue = 1;
 995                 dumpers.num_of_queue = 1;
 996
 997                 goto create_eth;
 998         }
 999
1000         /*
1001          * We check whether we want to open a RX stream from a real NIC or a
1002          * pcap file
1003          */
1004         is_rx_pcap = rte_kvargs_count(kvlist, ETH_PCAP_RX_PCAP_ARG) ? 1 : 0;
1005         pcaps.num_of_queue = 0;
1006
1007         if (is_rx_pcap) {
1008                 ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_PCAP_ARG,
1009                                 &open_rx_pcap, &pcaps);
1010         } else {
1011                 ret = rte_kvargs_process(kvlist, NULL,
1012                                 &rx_iface_args_process, &pcaps);
1013         }
1014
1015         if (ret < 0)
1016                 goto free_kvlist;
1017
1018         /*
1019          * We check whether we want to open a TX stream to a real NIC or a
1020          * pcap file
1021          */
1022         is_tx_pcap = rte_kvargs_count(kvlist, ETH_PCAP_TX_PCAP_ARG) ? 1 : 0;
1023         dumpers.num_of_queue = 0;
1024
1025         if (is_tx_pcap)
1026                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_PCAP_ARG,
1027                                 &open_tx_pcap, &dumpers);
1028         else
1029                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_IFACE_ARG,
1030                                 &open_tx_iface, &dumpers);
1031
1032         if (ret < 0)
1033                 goto free_kvlist;
1034
1035 create_eth:
1036         ret = eth_from_pcaps(dev, &pcaps, pcaps.num_of_queue, &dumpers,
1037                 dumpers.num_of_queue, kvlist, single_iface, is_tx_pcap);
1038
1039 free_kvlist:
1040         rte_kvargs_free(kvlist);
1041
1042         return ret;
1043 }
1044
1045 static int
1046 pmd_pcap_remove(struct rte_vdev_device *dev)
1047 {
1048         struct rte_eth_dev *eth_dev = NULL;
1049
1050         PMD_LOG(INFO, "Closing pcap ethdev on numa socket %d",
1051                         rte_socket_id());
1052
1053         if (!dev)
1054                 return -1;
1055
1056         /* reserve an ethdev entry */
1057         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1058         if (eth_dev == NULL)
1059                 return -1;
1060
1061         rte_free(eth_dev->data->dev_private);
1062
1063         rte_eth_dev_release_port(eth_dev);
1064
1065         return 0;
1066 }
1067
1068 static struct rte_vdev_driver pmd_pcap_drv = {
1069         .probe = pmd_pcap_probe,
1070         .remove = pmd_pcap_remove,
1071 };
1072
1073 RTE_PMD_REGISTER_VDEV(net_pcap, pmd_pcap_drv);
1074 RTE_PMD_REGISTER_ALIAS(net_pcap, eth_pcap);
1075 RTE_PMD_REGISTER_PARAM_STRING(net_pcap,
1076         ETH_PCAP_RX_PCAP_ARG "=<string> "
1077         ETH_PCAP_TX_PCAP_ARG "=<string> "
1078         ETH_PCAP_RX_IFACE_ARG "=<ifc> "
1079         ETH_PCAP_RX_IFACE_IN_ARG "=<ifc> "
1080         ETH_PCAP_TX_IFACE_ARG "=<ifc> "
1081         ETH_PCAP_IFACE_ARG "=<ifc>");
1082
1083 RTE_INIT(eth_pcap_init_log);
1084 static void
1085 eth_pcap_init_log(void)
1086 {
1087         eth_pcap_logtype = rte_log_register("pmd.net.pcap");
1088         if (eth_pcap_logtype >= 0)
1089                 rte_log_set_level(eth_pcap_logtype, RTE_LOG_NOTICE);
1090 }