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 static uint8_t iface_idx;
  43
  44 struct queue_stat {
  45         volatile unsigned long pkts;
  46         volatile unsigned long bytes;
  47         volatile unsigned long err_pkts;
  48 };
  49
  50 struct pcap_rx_queue {
  51         pcap_t *pcap;
  52         uint16_t in_port;
  53         struct rte_mempool *mb_pool;
  54         struct queue_stat rx_stat;
  55         char name[PATH_MAX];
  56         char type[ETH_PCAP_ARG_MAXLEN];
  57 };
  58
  59 struct pcap_tx_queue {
  60         pcap_dumper_t *dumper;
  61         pcap_t *pcap;
  62         struct queue_stat tx_stat;
  63         char name[PATH_MAX];
  64         char type[ETH_PCAP_ARG_MAXLEN];
  65 };
  66
  67 struct pmd_internals {
  68         struct pcap_rx_queue rx_queue[RTE_PMD_PCAP_MAX_QUEUES];
  69         struct pcap_tx_queue tx_queue[RTE_PMD_PCAP_MAX_QUEUES];
  70         struct ether_addr eth_addr;
  71         int if_index;
  72         int single_iface;
  73 };
  74
  75 struct pmd_devargs {
  76         unsigned int num_of_queue;
  77         struct devargs_queue {
  78                 pcap_dumper_t *dumper;
  79                 pcap_t *pcap;
  80                 const char *name;
  81                 const char *type;
  82         } queue[RTE_PMD_PCAP_MAX_QUEUES];
  83 };
  84
  85 static const char *valid_arguments[] = {
  86         ETH_PCAP_RX_PCAP_ARG,
  87         ETH_PCAP_TX_PCAP_ARG,
  88         ETH_PCAP_RX_IFACE_ARG,
  89         ETH_PCAP_RX_IFACE_IN_ARG,
  90         ETH_PCAP_TX_IFACE_ARG,
  91         ETH_PCAP_IFACE_ARG,
  92         NULL
  93 };
  94
  95 static struct rte_eth_link pmd_link = {
  96                 .link_speed = ETH_SPEED_NUM_10G,
  97                 .link_duplex = ETH_LINK_FULL_DUPLEX,
  98                 .link_status = ETH_LINK_DOWN,
  99                 .link_autoneg = ETH_LINK_FIXED,
 100 };
 101
 102 static int eth_pcap_logtype;
 103
 104 #define PMD_LOG(level, fmt, args...) \
 105         rte_log(RTE_LOG_ ## level, eth_pcap_logtype, \
 106                 "%s(): " fmt "\n", __func__, ##args)
 107
 108 static int
 109 eth_pcap_rx_jumbo(struct rte_mempool *mb_pool, struct rte_mbuf *mbuf,
 110                 const u_char *data, uint16_t data_len)
 111 {
 112         /* Copy the first segment. */
 113         uint16_t len = rte_pktmbuf_tailroom(mbuf);
 114         struct rte_mbuf *m = mbuf;
 115
 116         rte_memcpy(rte_pktmbuf_append(mbuf, len), data, len);
 117         data_len -= len;
 118         data += len;
 119
 120         while (data_len > 0) {
 121                 /* Allocate next mbuf and point to that. */
 122                 m->next = rte_pktmbuf_alloc(mb_pool);
 123
 124                 if (unlikely(!m->next))
 125                         return -1;
 126
 127                 m = m->next;
 128
 129                 /* Headroom is not needed in chained mbufs. */
 130                 rte_pktmbuf_prepend(m, rte_pktmbuf_headroom(m));
 131                 m->pkt_len = 0;
 132                 m->data_len = 0;
 133
 134                 /* Copy next segment. */
 135                 len = RTE_MIN(rte_pktmbuf_tailroom(m), data_len);
 136                 rte_memcpy(rte_pktmbuf_append(m, len), data, len);
 137
 138                 mbuf->nb_segs++;
 139                 data_len -= len;
 140                 data += len;
 141         }
 142
 143         return mbuf->nb_segs;
 144 }
 145
 146 /* Copy data from mbuf chain to a buffer suitable for writing to a PCAP file. */
 147 static void
 148 eth_pcap_gather_data(unsigned char *data, struct rte_mbuf *mbuf)
 149 {
 150         uint16_t data_len = 0;
 151
 152         while (mbuf) {
 153                 rte_memcpy(data + data_len, rte_pktmbuf_mtod(mbuf, void *),
 154                         mbuf->data_len);
 155
 156                 data_len += mbuf->data_len;
 157                 mbuf = mbuf->next;
 158         }
 159 }
 160
 161 static uint16_t
 162 eth_pcap_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 163 {
 164         unsigned int i;
 165         struct pcap_pkthdr header;
 166         const u_char *packet;
 167         struct rte_mbuf *mbuf;
 168         struct pcap_rx_queue *pcap_q = queue;
 169         uint16_t num_rx = 0;
 170         uint16_t buf_size;
 171         uint32_t rx_bytes = 0;
 172
 173         if (unlikely(pcap_q->pcap == NULL || nb_pkts == 0))
 174                 return 0;
 175
 176         /* Reads the given number of packets from the pcap file one by one
 177          * and copies the packet data into a newly allocated mbuf to return.
 178          */
 179         for (i = 0; i < nb_pkts; i++) {
 180                 /* Get the next PCAP packet */
 181                 packet = pcap_next(pcap_q->pcap, &header);
 182                 if (unlikely(packet == NULL))
 183                         break;
 184
 185                 mbuf = rte_pktmbuf_alloc(pcap_q->mb_pool);
 186                 if (unlikely(mbuf == NULL))
 187                         break;
 188
 189                 /* Now get the space available for data in the mbuf */
 190                 buf_size = rte_pktmbuf_data_room_size(pcap_q->mb_pool) -
 191                                 RTE_PKTMBUF_HEADROOM;
 192
 193                 if (header.caplen <= buf_size) {
 194                         /* pcap packet will fit in the mbuf, can copy it */
 195                         rte_memcpy(rte_pktmbuf_mtod(mbuf, void *), packet,
 196                                         header.caplen);
 197                         mbuf->data_len = (uint16_t)header.caplen;
 198                 } else {
 199                         /* Try read jumbo frame into multi mbufs. */
 200                         if (unlikely(eth_pcap_rx_jumbo(pcap_q->mb_pool,
 201                                                        mbuf,
 202                                                        packet,
 203                                                        header.caplen) == -1)) {
 204                                 rte_pktmbuf_free(mbuf);
 205                                 break;
 206                         }
 207                 }
 208
 209                 mbuf->pkt_len = (uint16_t)header.caplen;
 210                 mbuf->port = pcap_q->in_port;
 211                 bufs[num_rx] = mbuf;
 212                 num_rx++;
 213                 rx_bytes += header.caplen;
 214         }
 215         pcap_q->rx_stat.pkts += num_rx;
 216         pcap_q->rx_stat.bytes += rx_bytes;
 217
 218         return num_rx;
 219 }
 220
 221 static inline void
 222 calculate_timestamp(struct timeval *ts) {
 223         uint64_t cycles;
 224         struct timeval cur_time;
 225
 226         cycles = rte_get_timer_cycles() - start_cycles;
 227         cur_time.tv_sec = cycles / hz;
 228         cur_time.tv_usec = (cycles % hz) * 1e6 / hz;
 229         timeradd(&start_time, &cur_time, ts);
 230 }
 231
 232 /*
 233  * Callback to handle writing packets to a pcap file.
 234  */
 235 static uint16_t
 236 eth_pcap_tx_dumper(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 237 {
 238         unsigned int i;
 239         struct rte_mbuf *mbuf;
 240         struct pcap_tx_queue *dumper_q = queue;
 241         uint16_t num_tx = 0;
 242         uint32_t tx_bytes = 0;
 243         struct pcap_pkthdr header;
 244
 245         if (dumper_q->dumper == NULL || nb_pkts == 0)
 246                 return 0;
 247
 248         /* writes the nb_pkts packets to the previously opened pcap file
 249          * dumper */
 250         for (i = 0; i < nb_pkts; i++) {
 251                 mbuf = bufs[i];
 252                 calculate_timestamp(&header.ts);
 253                 header.len = mbuf->pkt_len;
 254                 header.caplen = header.len;
 255
 256                 if (likely(mbuf->nb_segs == 1)) {
 257                         pcap_dump((u_char *)dumper_q->dumper, &header,
 258                                   rte_pktmbuf_mtod(mbuf, void*));
 259                 } else {
 260                         if (mbuf->pkt_len <= ETHER_MAX_JUMBO_FRAME_LEN) {
 261                                 eth_pcap_gather_data(tx_pcap_data, mbuf);
 262                                 pcap_dump((u_char *)dumper_q->dumper, &header,
 263                                           tx_pcap_data);
 264                         } else {
 265                                 PMD_LOG(ERR,
 266                                         "Dropping PCAP packet. Size (%d) > max jumbo size (%d).",
 267                                         mbuf->pkt_len,
 268                                         ETHER_MAX_JUMBO_FRAME_LEN);
 269
 270                                 rte_pktmbuf_free(mbuf);
 271                                 break;
 272                         }
 273                 }
 274
 275                 num_tx++;
 276                 tx_bytes += mbuf->pkt_len;
 277                 rte_pktmbuf_free(mbuf);
 278         }
 279
 280         /*
 281          * Since there's no place to hook a callback when the forwarding
 282          * process stops and to make sure the pcap file is actually written,
 283          * we flush the pcap dumper within each burst.
 284          */
 285         pcap_dump_flush(dumper_q->dumper);
 286         dumper_q->tx_stat.pkts += num_tx;
 287         dumper_q->tx_stat.bytes += tx_bytes;
 288         dumper_q->tx_stat.err_pkts += nb_pkts - num_tx;
 289
 290         return num_tx;
 291 }
 292
 293 /*
 294  * Callback to handle sending packets through a real NIC.
 295  */
 296 static uint16_t
 297 eth_pcap_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 298 {
 299         unsigned int i;
 300         int ret;
 301         struct rte_mbuf *mbuf;
 302         struct pcap_tx_queue *tx_queue = queue;
 303         uint16_t num_tx = 0;
 304         uint32_t tx_bytes = 0;
 305
 306         if (unlikely(nb_pkts == 0 || tx_queue->pcap == NULL))
 307                 return 0;
 308
 309         for (i = 0; i < nb_pkts; i++) {
 310                 mbuf = bufs[i];
 311
 312                 if (likely(mbuf->nb_segs == 1)) {
 313                         ret = pcap_sendpacket(tx_queue->pcap,
 314                                         rte_pktmbuf_mtod(mbuf, u_char *),
 315                                         mbuf->pkt_len);
 316                 } else {
 317                         if (mbuf->pkt_len <= ETHER_MAX_JUMBO_FRAME_LEN) {
 318                                 eth_pcap_gather_data(tx_pcap_data, mbuf);
 319                                 ret = pcap_sendpacket(tx_queue->pcap,
 320                                                 tx_pcap_data, mbuf->pkt_len);
 321                         } else {
 322                                 PMD_LOG(ERR,
 323                                         "Dropping PCAP packet. Size (%d) > max jumbo size (%d).",
 324                                         mbuf->pkt_len,
 325                                         ETHER_MAX_JUMBO_FRAME_LEN);
 326
 327                                 rte_pktmbuf_free(mbuf);
 328                                 break;
 329                         }
 330                 }
 331
 332                 if (unlikely(ret != 0))
 333                         break;
 334                 num_tx++;
 335                 tx_bytes += mbuf->pkt_len;
 336                 rte_pktmbuf_free(mbuf);
 337         }
 338
 339         tx_queue->tx_stat.pkts += num_tx;
 340         tx_queue->tx_stat.bytes += tx_bytes;
 341         tx_queue->tx_stat.err_pkts += nb_pkts - num_tx;
 342
 343         return num_tx;
 344 }
 345
 346 /*
 347  * pcap_open_live wrapper function
 348  */
 349 static inline int
 350 open_iface_live(const char *iface, pcap_t **pcap) {
 351         *pcap = pcap_open_live(iface, RTE_ETH_PCAP_SNAPLEN,
 352                         RTE_ETH_PCAP_PROMISC, RTE_ETH_PCAP_TIMEOUT, errbuf);
 353
 354         if (*pcap == NULL) {
 355                 PMD_LOG(ERR, "Couldn't open %s: %s", iface, errbuf);
 356                 return -1;
 357         }
 358
 359         return 0;
 360 }
 361
 362 static int
 363 open_single_iface(const char *iface, pcap_t **pcap)
 364 {
 365         if (open_iface_live(iface, pcap) < 0) {
 366                 PMD_LOG(ERR, "Couldn't open interface %s", iface);
 367                 return -1;
 368         }
 369
 370         return 0;
 371 }
 372
 373 static int
 374 open_single_tx_pcap(const char *pcap_filename, pcap_dumper_t **dumper)
 375 {
 376         pcap_t *tx_pcap;
 377
 378         /*
 379          * We need to create a dummy empty pcap_t to use it
 380          * with pcap_dump_open(). We create big enough an Ethernet
 381          * pcap holder.
 382          */
 383         tx_pcap = pcap_open_dead(DLT_EN10MB, RTE_ETH_PCAP_SNAPSHOT_LEN);
 384         if (tx_pcap == NULL) {
 385                 PMD_LOG(ERR, "Couldn't create dead pcap");
 386                 return -1;
 387         }
 388
 389         /* The dumper is created using the previous pcap_t reference */
 390         *dumper = pcap_dump_open(tx_pcap, pcap_filename);
 391         if (*dumper == NULL) {
 392                 pcap_close(tx_pcap);
 393                 PMD_LOG(ERR, "Couldn't open %s for writing.",
 394                         pcap_filename);
 395                 return -1;
 396         }
 397
 398         pcap_close(tx_pcap);
 399         return 0;
 400 }
 401
 402 static int
 403 open_single_rx_pcap(const char *pcap_filename, pcap_t **pcap)
 404 {
 405         *pcap = pcap_open_offline(pcap_filename, errbuf);
 406         if (*pcap == NULL) {
 407                 PMD_LOG(ERR, "Couldn't open %s: %s", pcap_filename,
 408                         errbuf);
 409                 return -1;
 410         }
 411
 412         return 0;
 413 }
 414
 415 static int
 416 eth_dev_start(struct rte_eth_dev *dev)
 417 {
 418         unsigned int i;
 419         struct pmd_internals *internals = dev->data->dev_private;
 420         struct pcap_tx_queue *tx;
 421         struct pcap_rx_queue *rx;
 422
 423         /* Special iface case. Single pcap is open and shared between tx/rx. */
 424         if (internals->single_iface) {
 425                 tx = &internals->tx_queue[0];
 426                 rx = &internals->rx_queue[0];
 427
 428                 if (!tx->pcap && strcmp(tx->type, ETH_PCAP_IFACE_ARG) == 0) {
 429                         if (open_single_iface(tx->name, &tx->pcap) < 0)
 430                                 return -1;
 431                         rx->pcap = tx->pcap;
 432                 }
 433
 434                 goto status_up;
 435         }
 436
 437         /* If not open already, open tx pcaps/dumpers */
 438         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 439                 tx = &internals->tx_queue[i];
 440
 441                 if (!tx->dumper &&
 442                                 strcmp(tx->type, ETH_PCAP_TX_PCAP_ARG) == 0) {
 443                         if (open_single_tx_pcap(tx->name, &tx->dumper) < 0)
 444                                 return -1;
 445                 } else if (!tx->pcap &&
 446                                 strcmp(tx->type, ETH_PCAP_TX_IFACE_ARG) == 0) {
 447                         if (open_single_iface(tx->name, &tx->pcap) < 0)
 448                                 return -1;
 449                 }
 450         }
 451
 452         /* If not open already, open rx pcaps */
 453         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 454                 rx = &internals->rx_queue[i];
 455
 456                 if (rx->pcap != NULL)
 457                         continue;
 458
 459                 if (strcmp(rx->type, ETH_PCAP_RX_PCAP_ARG) == 0) {
 460                         if (open_single_rx_pcap(rx->name, &rx->pcap) < 0)
 461                                 return -1;
 462                 } else if (strcmp(rx->type, ETH_PCAP_RX_IFACE_ARG) == 0) {
 463                         if (open_single_iface(rx->name, &rx->pcap) < 0)
 464                                 return -1;
 465                 }
 466         }
 467
 468 status_up:
 469         for (i = 0; i < dev->data->nb_rx_queues; i++)
 470                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 471
 472         for (i = 0; i < dev->data->nb_tx_queues; i++)
 473                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 474
 475         dev->data->dev_link.link_status = ETH_LINK_UP;
 476
 477         return 0;
 478 }
 479
 480 /*
 481  * This function gets called when the current port gets stopped.
 482  * Is the only place for us to close all the tx streams dumpers.
 483  * If not called the dumpers will be flushed within each tx burst.
 484  */
 485 static void
 486 eth_dev_stop(struct rte_eth_dev *dev)
 487 {
 488         unsigned int i;
 489         struct pmd_internals *internals = dev->data->dev_private;
 490         struct pcap_tx_queue *tx;
 491         struct pcap_rx_queue *rx;
 492
 493         /* Special iface case. Single pcap is open and shared between tx/rx. */
 494         if (internals->single_iface) {
 495                 tx = &internals->tx_queue[0];
 496                 rx = &internals->rx_queue[0];
 497                 pcap_close(tx->pcap);
 498                 tx->pcap = NULL;
 499                 rx->pcap = NULL;
 500                 goto status_down;
 501         }
 502
 503         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 504                 tx = &internals->tx_queue[i];
 505
 506                 if (tx->dumper != NULL) {
 507                         pcap_dump_close(tx->dumper);
 508                         tx->dumper = NULL;
 509                 }
 510
 511                 if (tx->pcap != NULL) {
 512                         pcap_close(tx->pcap);
 513                         tx->pcap = NULL;
 514                 }
 515         }
 516
 517         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 518                 rx = &internals->rx_queue[i];
 519
 520                 if (rx->pcap != NULL) {
 521                         pcap_close(rx->pcap);
 522                         rx->pcap = NULL;
 523                 }
 524         }
 525
 526 status_down:
 527         for (i = 0; i < dev->data->nb_rx_queues; i++)
 528                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 529
 530         for (i = 0; i < dev->data->nb_tx_queues; i++)
 531                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 532
 533         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 534 }
 535
 536 static int
 537 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 538 {
 539         return 0;
 540 }
 541
 542 static void
 543 eth_dev_info(struct rte_eth_dev *dev,
 544                 struct rte_eth_dev_info *dev_info)
 545 {
 546         struct pmd_internals *internals = dev->data->dev_private;
 547
 548         dev_info->if_index = internals->if_index;
 549         dev_info->max_mac_addrs = 1;
 550         dev_info->max_rx_pktlen = (uint32_t) -1;
 551         dev_info->max_rx_queues = dev->data->nb_rx_queues;
 552         dev_info->max_tx_queues = dev->data->nb_tx_queues;
 553         dev_info->min_rx_bufsize = 0;
 554 }
 555
 556 static int
 557 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 558 {
 559         unsigned int i;
 560         unsigned long rx_packets_total = 0, rx_bytes_total = 0;
 561         unsigned long tx_packets_total = 0, tx_bytes_total = 0;
 562         unsigned long tx_packets_err_total = 0;
 563         const struct pmd_internals *internal = dev->data->dev_private;
 564
 565         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 566                         i < dev->data->nb_rx_queues; i++) {
 567                 stats->q_ipackets[i] = internal->rx_queue[i].rx_stat.pkts;
 568                 stats->q_ibytes[i] = internal->rx_queue[i].rx_stat.bytes;
 569                 rx_packets_total += stats->q_ipackets[i];
 570                 rx_bytes_total += stats->q_ibytes[i];
 571         }
 572
 573         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 574                         i < dev->data->nb_tx_queues; i++) {
 575                 stats->q_opackets[i] = internal->tx_queue[i].tx_stat.pkts;
 576                 stats->q_obytes[i] = internal->tx_queue[i].tx_stat.bytes;
 577                 stats->q_errors[i] = internal->tx_queue[i].tx_stat.err_pkts;
 578                 tx_packets_total += stats->q_opackets[i];
 579                 tx_bytes_total += stats->q_obytes[i];
 580                 tx_packets_err_total += stats->q_errors[i];
 581         }
 582
 583         stats->ipackets = rx_packets_total;
 584         stats->ibytes = rx_bytes_total;
 585         stats->opackets = tx_packets_total;
 586         stats->obytes = tx_bytes_total;
 587         stats->oerrors = tx_packets_err_total;
 588
 589         return 0;
 590 }
 591
 592 static void
 593 eth_stats_reset(struct rte_eth_dev *dev)
 594 {
 595         unsigned int i;
 596         struct pmd_internals *internal = dev->data->dev_private;
 597
 598         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 599                 internal->rx_queue[i].rx_stat.pkts = 0;
 600                 internal->rx_queue[i].rx_stat.bytes = 0;
 601         }
 602
 603         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 604                 internal->tx_queue[i].tx_stat.pkts = 0;
 605                 internal->tx_queue[i].tx_stat.bytes = 0;
 606                 internal->tx_queue[i].tx_stat.err_pkts = 0;
 607         }
 608 }
 609
 610 static void
 611 eth_dev_close(struct rte_eth_dev *dev __rte_unused)
 612 {
 613 }
 614
 615 static void
 616 eth_queue_release(void *q __rte_unused)
 617 {
 618 }
 619
 620 static int
 621 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 622                 int wait_to_complete __rte_unused)
 623 {
 624         return 0;
 625 }
 626
 627 static int
 628 eth_rx_queue_setup(struct rte_eth_dev *dev,
 629                 uint16_t rx_queue_id,
 630                 uint16_t nb_rx_desc __rte_unused,
 631                 unsigned int socket_id __rte_unused,
 632                 const struct rte_eth_rxconf *rx_conf __rte_unused,
 633                 struct rte_mempool *mb_pool)
 634 {
 635         struct pmd_internals *internals = dev->data->dev_private;
 636         struct pcap_rx_queue *pcap_q = &internals->rx_queue[rx_queue_id];
 637
 638         pcap_q->mb_pool = mb_pool;
 639         dev->data->rx_queues[rx_queue_id] = pcap_q;
 640         pcap_q->in_port = dev->data->port_id;
 641
 642         return 0;
 643 }
 644
 645 static int
 646 eth_tx_queue_setup(struct rte_eth_dev *dev,
 647                 uint16_t tx_queue_id,
 648                 uint16_t nb_tx_desc __rte_unused,
 649                 unsigned int socket_id __rte_unused,
 650                 const struct rte_eth_txconf *tx_conf __rte_unused)
 651 {
 652         struct pmd_internals *internals = dev->data->dev_private;
 653
 654         dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
 655
 656         return 0;
 657 }
 658
 659 static int
 660 eth_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 661 {
 662         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
 663
 664         return 0;
 665 }
 666
 667 static int
 668 eth_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
 669 {
 670         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
 671
 672         return 0;
 673 }
 674
 675 static int
 676 eth_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 677 {
 678         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
 679
 680         return 0;
 681 }
 682
 683 static int
 684 eth_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
 685 {
 686         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
 687
 688         return 0;
 689 }
 690
 691 static const struct eth_dev_ops ops = {
 692         .dev_start = eth_dev_start,
 693         .dev_stop = eth_dev_stop,
 694         .dev_close = eth_dev_close,
 695         .dev_configure = eth_dev_configure,
 696         .dev_infos_get = eth_dev_info,
 697         .rx_queue_setup = eth_rx_queue_setup,
 698         .tx_queue_setup = eth_tx_queue_setup,
 699         .rx_queue_start = eth_rx_queue_start,
 700         .tx_queue_start = eth_tx_queue_start,
 701         .rx_queue_stop = eth_rx_queue_stop,
 702         .tx_queue_stop = eth_tx_queue_stop,
 703         .rx_queue_release = eth_queue_release,
 704         .tx_queue_release = eth_queue_release,
 705         .link_update = eth_link_update,
 706         .stats_get = eth_stats_get,
 707         .stats_reset = eth_stats_reset,
 708 };
 709
 710 static int
 711 add_queue(struct pmd_devargs *pmd, const char *name, const char *type,
 712                 pcap_t *pcap, pcap_dumper_t *dumper)
 713 {
 714         if (pmd->num_of_queue >= RTE_PMD_PCAP_MAX_QUEUES)
 715                 return -1;
 716         if (pcap)
 717                 pmd->queue[pmd->num_of_queue].pcap = pcap;
 718         if (dumper)
 719                 pmd->queue[pmd->num_of_queue].dumper = dumper;
 720         pmd->queue[pmd->num_of_queue].name = name;
 721         pmd->queue[pmd->num_of_queue].type = type;
 722         pmd->num_of_queue++;
 723         return 0;
 724 }
 725
 726 /*
 727  * Function handler that opens the pcap file for reading a stores a
 728  * reference of it for use it later on.
 729  */
 730 static int
 731 open_rx_pcap(const char *key, const char *value, void *extra_args)
 732 {
 733         const char *pcap_filename = value;
 734         struct pmd_devargs *rx = extra_args;
 735         pcap_t *pcap = NULL;
 736
 737         if (open_single_rx_pcap(pcap_filename, &pcap) < 0)
 738                 return -1;
 739
 740         if (add_queue(rx, pcap_filename, key, pcap, NULL) < 0) {
 741                 pcap_close(pcap);
 742                 return -1;
 743         }
 744
 745         return 0;
 746 }
 747
 748 /*
 749  * Opens a pcap file for writing and stores a reference to it
 750  * for use it later on.
 751  */
 752 static int
 753 open_tx_pcap(const char *key, const char *value, void *extra_args)
 754 {
 755         const char *pcap_filename = value;
 756         struct pmd_devargs *dumpers = extra_args;
 757         pcap_dumper_t *dumper;
 758
 759         if (open_single_tx_pcap(pcap_filename, &dumper) < 0)
 760                 return -1;
 761
 762         if (add_queue(dumpers, pcap_filename, key, NULL, dumper) < 0) {
 763                 pcap_dump_close(dumper);
 764                 return -1;
 765         }
 766
 767         return 0;
 768 }
 769
 770 /*
 771  * Opens an interface for reading and writing
 772  */
 773 static inline int
 774 open_rx_tx_iface(const char *key, const char *value, void *extra_args)
 775 {
 776         const char *iface = value;
 777         struct pmd_devargs *tx = extra_args;
 778         pcap_t *pcap = NULL;
 779
 780         if (open_single_iface(iface, &pcap) < 0)
 781                 return -1;
 782
 783         tx->queue[0].pcap = pcap;
 784         tx->queue[0].name = iface;
 785         tx->queue[0].type = key;
 786
 787         return 0;
 788 }
 789
 790 static inline int
 791 set_iface_direction(const char *iface, pcap_t *pcap,
 792                 pcap_direction_t direction)
 793 {
 794         const char *direction_str = (direction == PCAP_D_IN) ? "IN" : "OUT";
 795         if (pcap_setdirection(pcap, direction) < 0) {
 796                 PMD_LOG(ERR, "Setting %s pcap direction %s failed - %s\n",
 797                                 iface, direction_str, pcap_geterr(pcap));
 798                 return -1;
 799         }
 800         PMD_LOG(INFO, "Setting %s pcap direction %s\n",
 801                         iface, direction_str);
 802         return 0;
 803 }
 804
 805 static inline int
 806 open_iface(const char *key, const char *value, void *extra_args)
 807 {
 808         const char *iface = value;
 809         struct pmd_devargs *pmd = extra_args;
 810         pcap_t *pcap = NULL;
 811
 812         if (open_single_iface(iface, &pcap) < 0)
 813                 return -1;
 814         if (add_queue(pmd, iface, key, pcap, NULL) < 0) {
 815                 pcap_close(pcap);
 816                 return -1;
 817         }
 818
 819         return 0;
 820 }
 821
 822 /*
 823  * Opens a NIC for reading packets from it
 824  */
 825 static inline int
 826 open_rx_iface(const char *key, const char *value, void *extra_args)
 827 {
 828         int ret = open_iface(key, value, extra_args);
 829         if (ret < 0)
 830                 return ret;
 831         if (strcmp(key, ETH_PCAP_RX_IFACE_IN_ARG) == 0) {
 832                 struct pmd_devargs *pmd = extra_args;
 833                 unsigned int qid = pmd->num_of_queue - 1;
 834
 835                 set_iface_direction(pmd->queue[qid].name,
 836                                 pmd->queue[qid].pcap,
 837                                 PCAP_D_IN);
 838         }
 839
 840         return 0;
 841 }
 842
 843 static inline int
 844 rx_iface_args_process(const char *key, const char *value, void *extra_args)
 845 {
 846         if (strcmp(key, ETH_PCAP_RX_IFACE_ARG) == 0 ||
 847                         strcmp(key, ETH_PCAP_RX_IFACE_IN_ARG) == 0)
 848                 return open_rx_iface(key, value, extra_args);
 849
 850         return 0;
 851 }
 852
 853 /*
 854  * Opens a NIC for writing packets to it
 855  */
 856 static int
 857 open_tx_iface(const char *key, const char *value, void *extra_args)
 858 {
 859         return open_iface(key, value, extra_args);
 860 }
 861
 862 static struct rte_vdev_driver pmd_pcap_drv;
 863
 864 static int
 865 pmd_init_internals(struct rte_vdev_device *vdev,
 866                 const unsigned int nb_rx_queues,
 867                 const unsigned int nb_tx_queues,
 868                 struct pmd_internals **internals,
 869                 struct rte_eth_dev **eth_dev)
 870 {
 871         struct rte_eth_dev_data *data;
 872         unsigned int numa_node = vdev->device.numa_node;
 873
 874         PMD_LOG(INFO, "Creating pcap-backed ethdev on numa socket %d",
 875                 numa_node);
 876
 877         /* reserve an ethdev entry */
 878         *eth_dev = rte_eth_vdev_allocate(vdev, sizeof(**internals));
 879         if (!(*eth_dev))
 880                 return -1;
 881
 882         /* now put it all together
 883          * - store queue data in internals,
 884          * - store numa_node info in eth_dev
 885          * - point eth_dev_data to internals
 886          * - and point eth_dev structure to new eth_dev_data structure
 887          */
 888         *internals = (*eth_dev)->data->dev_private;
 889         /*
 890          * Interface MAC = 02:70:63:61:70:<iface_idx>
 891          * derived from: 'locally administered':'p':'c':'a':'p':'iface_idx'
 892          * where the middle 4 characters are converted to hex.
 893          */
 894         (*internals)->eth_addr = (struct ether_addr) {
 895                 .addr_bytes = { 0x02, 0x70, 0x63, 0x61, 0x70, iface_idx++ }
 896         };
 897         data = (*eth_dev)->data;
 898         data->nb_rx_queues = (uint16_t)nb_rx_queues;
 899         data->nb_tx_queues = (uint16_t)nb_tx_queues;
 900         data->dev_link = pmd_link;
 901         data->mac_addrs = &(*internals)->eth_addr;
 902
 903         /*
 904          * NOTE: we'll replace the data element, of originally allocated
 905          * eth_dev so the rings are local per-process
 906          */
 907         (*eth_dev)->dev_ops = &ops;
 908
 909         return 0;
 910 }
 911
 912 static int
 913 eth_from_pcaps_common(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, struct pmd_internals **internals,
 917                 struct rte_eth_dev **eth_dev)
 918 {
 919         struct rte_kvargs_pair *pair = NULL;
 920         unsigned int k_idx;
 921         unsigned int i;
 922
 923         /* do some parameter checking */
 924         if (rx_queues == NULL && nb_rx_queues > 0)
 925                 return -1;
 926         if (tx_queues == NULL && nb_tx_queues > 0)
 927                 return -1;
 928
 929         if (pmd_init_internals(vdev, nb_rx_queues, nb_tx_queues, internals,
 930                         eth_dev) < 0)
 931                 return -1;
 932
 933         for (i = 0; i < nb_rx_queues; i++) {
 934                 struct pcap_rx_queue *rx = &(*internals)->rx_queue[i];
 935                 struct devargs_queue *queue = &rx_queues->queue[i];
 936
 937                 rx->pcap = queue->pcap;
 938                 snprintf(rx->name, sizeof(rx->name), "%s", queue->name);
 939                 snprintf(rx->type, sizeof(rx->type), "%s", queue->type);
 940         }
 941
 942         for (i = 0; i < nb_tx_queues; i++) {
 943                 struct pcap_tx_queue *tx = &(*internals)->tx_queue[i];
 944                 struct devargs_queue *queue = &tx_queues->queue[i];
 945
 946                 tx->dumper = queue->dumper;
 947                 tx->pcap = queue->pcap;
 948                 snprintf(tx->name, sizeof(tx->name), "%s", queue->name);
 949                 snprintf(tx->type, sizeof(tx->type), "%s", queue->type);
 950         }
 951
 952         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 953                 pair = &kvlist->pairs[k_idx];
 954                 if (strstr(pair->key, ETH_PCAP_IFACE_ARG) != NULL)
 955                         break;
 956         }
 957
 958         if (pair == NULL)
 959                 (*internals)->if_index = 0;
 960         else
 961                 (*internals)->if_index = if_nametoindex(pair->value);
 962
 963         return 0;
 964 }
 965
 966 static int
 967 eth_from_pcaps(struct rte_vdev_device *vdev,
 968                 struct pmd_devargs *rx_queues, const unsigned int nb_rx_queues,
 969                 struct pmd_devargs *tx_queues, const unsigned int nb_tx_queues,
 970                 struct rte_kvargs *kvlist, int single_iface,
 971                 unsigned int using_dumpers)
 972 {
 973         struct pmd_internals *internals = NULL;
 974         struct rte_eth_dev *eth_dev = NULL;
 975         int ret;
 976
 977         ret = eth_from_pcaps_common(vdev, rx_queues, nb_rx_queues,
 978                 tx_queues, nb_tx_queues, kvlist, &internals, &eth_dev);
 979
 980         if (ret < 0)
 981                 return ret;
 982
 983         /* store weather we are using a single interface for rx/tx or not */
 984         internals->single_iface = single_iface;
 985
 986         eth_dev->rx_pkt_burst = eth_pcap_rx;
 987
 988         if (using_dumpers)
 989                 eth_dev->tx_pkt_burst = eth_pcap_tx_dumper;
 990         else
 991                 eth_dev->tx_pkt_burst = eth_pcap_tx;
 992
 993         rte_eth_dev_probing_finish(eth_dev);
 994         return 0;
 995 }
 996
 997 static int
 998 pmd_pcap_probe(struct rte_vdev_device *dev)
 999 {
1000         const char *name;
1001         unsigned int is_rx_pcap = 0, is_tx_pcap = 0;
1002         struct rte_kvargs *kvlist;
1003         struct pmd_devargs pcaps = {0};
1004         struct pmd_devargs dumpers = {0};
1005         struct rte_eth_dev *eth_dev;
1006         int single_iface = 0;
1007         int ret;
1008
1009         name = rte_vdev_device_name(dev);
1010         PMD_LOG(INFO, "Initializing pmd_pcap for %s", name);
1011
1012         gettimeofday(&start_time, NULL);
1013         start_cycles = rte_get_timer_cycles();
1014         hz = rte_get_timer_hz();
1015
1016         if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
1017             strlen(rte_vdev_device_args(dev)) == 0) {
1018                 eth_dev = rte_eth_dev_attach_secondary(name);
1019                 if (!eth_dev) {
1020                         PMD_LOG(ERR, "Failed to probe %s", name);
1021                         return -1;
1022                 }
1023                 /* TODO: request info from primary to set up Rx and Tx */
1024                 eth_dev->dev_ops = &ops;
1025                 eth_dev->device = &dev->device;
1026                 rte_eth_dev_probing_finish(eth_dev);
1027                 return 0;
1028         }
1029
1030         kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_arguments);
1031         if (kvlist == NULL)
1032                 return -1;
1033
1034         /*
1035          * If iface argument is passed we open the NICs and use them for
1036          * reading / writing
1037          */
1038         if (rte_kvargs_count(kvlist, ETH_PCAP_IFACE_ARG) == 1) {
1039
1040                 ret = rte_kvargs_process(kvlist, ETH_PCAP_IFACE_ARG,
1041                                 &open_rx_tx_iface, &pcaps);
1042
1043                 if (ret < 0)
1044                         goto free_kvlist;
1045
1046                 dumpers.queue[0] = pcaps.queue[0];
1047
1048                 single_iface = 1;
1049                 pcaps.num_of_queue = 1;
1050                 dumpers.num_of_queue = 1;
1051
1052                 goto create_eth;
1053         }
1054
1055         /*
1056          * We check whether we want to open a RX stream from a real NIC or a
1057          * pcap file
1058          */
1059         is_rx_pcap = rte_kvargs_count(kvlist, ETH_PCAP_RX_PCAP_ARG) ? 1 : 0;
1060         pcaps.num_of_queue = 0;
1061
1062         if (is_rx_pcap) {
1063                 ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_PCAP_ARG,
1064                                 &open_rx_pcap, &pcaps);
1065         } else {
1066                 ret = rte_kvargs_process(kvlist, NULL,
1067                                 &rx_iface_args_process, &pcaps);
1068         }
1069
1070         if (ret < 0)
1071                 goto free_kvlist;
1072
1073         /*
1074          * We check whether we want to open a TX stream to a real NIC or a
1075          * pcap file
1076          */
1077         is_tx_pcap = rte_kvargs_count(kvlist, ETH_PCAP_TX_PCAP_ARG) ? 1 : 0;
1078         dumpers.num_of_queue = 0;
1079
1080         if (is_tx_pcap)
1081                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_PCAP_ARG,
1082                                 &open_tx_pcap, &dumpers);
1083         else
1084                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_IFACE_ARG,
1085                                 &open_tx_iface, &dumpers);
1086
1087         if (ret < 0)
1088                 goto free_kvlist;
1089
1090 create_eth:
1091         ret = eth_from_pcaps(dev, &pcaps, pcaps.num_of_queue, &dumpers,
1092                 dumpers.num_of_queue, kvlist, single_iface, is_tx_pcap);
1093
1094 free_kvlist:
1095         rte_kvargs_free(kvlist);
1096
1097         return ret;
1098 }
1099
1100 static int
1101 pmd_pcap_remove(struct rte_vdev_device *dev)
1102 {
1103         struct rte_eth_dev *eth_dev = NULL;
1104
1105         PMD_LOG(INFO, "Closing pcap ethdev on numa socket %d",
1106                         rte_socket_id());
1107
1108         if (!dev)
1109                 return -1;
1110
1111         /* reserve an ethdev entry */
1112         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1113         if (eth_dev == NULL)
1114                 return -1;
1115
1116         rte_free(eth_dev->data->dev_private);
1117
1118         rte_eth_dev_release_port(eth_dev);
1119
1120         return 0;
1121 }
1122
1123 static struct rte_vdev_driver pmd_pcap_drv = {
1124         .probe = pmd_pcap_probe,
1125         .remove = pmd_pcap_remove,
1126 };
1127
1128 RTE_PMD_REGISTER_VDEV(net_pcap, pmd_pcap_drv);
1129 RTE_PMD_REGISTER_ALIAS(net_pcap, eth_pcap);
1130 RTE_PMD_REGISTER_PARAM_STRING(net_pcap,
1131         ETH_PCAP_RX_PCAP_ARG "=<string> "
1132         ETH_PCAP_TX_PCAP_ARG "=<string> "
1133         ETH_PCAP_RX_IFACE_ARG "=<ifc> "
1134         ETH_PCAP_RX_IFACE_IN_ARG "=<ifc> "
1135         ETH_PCAP_TX_IFACE_ARG "=<ifc> "
1136         ETH_PCAP_IFACE_ARG "=<ifc>");
1137
1138 RTE_INIT(eth_pcap_init_log)
1139 {
1140         eth_pcap_logtype = rte_log_register("pmd.net.pcap");
1141         if (eth_pcap_logtype >= 0)
1142                 rte_log_set_level(eth_pcap_logtype, RTE_LOG_NOTICE);
1143 }