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_TX_IFACE_ARG "tx_iface"
  30 #define ETH_PCAP_IFACE_ARG    "iface"
  31
  32 #define ETH_PCAP_ARG_MAXLEN     64
  33
  34 #define RTE_PMD_PCAP_MAX_QUEUES 16
  35
  36 static char errbuf[PCAP_ERRBUF_SIZE];
  37 static unsigned char tx_pcap_data[RTE_ETH_PCAP_SNAPLEN];
  38 static struct timeval start_time;
  39 static uint64_t start_cycles;
  40 static uint64_t hz;
  41
  42 struct queue_stat {
  43         volatile unsigned long pkts;
  44         volatile unsigned long bytes;
  45         volatile unsigned long err_pkts;
  46 };
  47
  48 struct pcap_rx_queue {
  49         pcap_t *pcap;
  50         uint16_t in_port;
  51         struct rte_mempool *mb_pool;
  52         struct queue_stat rx_stat;
  53         char name[PATH_MAX];
  54         char type[ETH_PCAP_ARG_MAXLEN];
  55 };
  56
  57 struct pcap_tx_queue {
  58         pcap_dumper_t *dumper;
  59         pcap_t *pcap;
  60         struct queue_stat tx_stat;
  61         char name[PATH_MAX];
  62         char type[ETH_PCAP_ARG_MAXLEN];
  63 };
  64
  65 struct pmd_internals {
  66         struct pcap_rx_queue rx_queue[RTE_PMD_PCAP_MAX_QUEUES];
  67         struct pcap_tx_queue tx_queue[RTE_PMD_PCAP_MAX_QUEUES];
  68         int if_index;
  69         int single_iface;
  70 };
  71
  72 struct pmd_devargs {
  73         unsigned int num_of_queue;
  74         struct devargs_queue {
  75                 pcap_dumper_t *dumper;
  76                 pcap_t *pcap;
  77                 const char *name;
  78                 const char *type;
  79         } queue[RTE_PMD_PCAP_MAX_QUEUES];
  80 };
  81
  82 static const char *valid_arguments[] = {
  83         ETH_PCAP_RX_PCAP_ARG,
  84         ETH_PCAP_TX_PCAP_ARG,
  85         ETH_PCAP_RX_IFACE_ARG,
  86         ETH_PCAP_TX_IFACE_ARG,
  87         ETH_PCAP_IFACE_ARG,
  88         NULL
  89 };
  90
  91 static struct ether_addr eth_addr = {
  92         .addr_bytes = { 0, 0, 0, 0x1, 0x2, 0x3 }
  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                 goto status_up;
 434         }
 435
 436         /* If not open already, open tx pcaps/dumpers */
 437         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 438                 tx = &internals->tx_queue[i];
 439
 440                 if (!tx->dumper &&
 441                                 strcmp(tx->type, ETH_PCAP_TX_PCAP_ARG) == 0) {
 442                         if (open_single_tx_pcap(tx->name, &tx->dumper) < 0)
 443                                 return -1;
 444                 } else if (!tx->pcap &&
 445                                 strcmp(tx->type, ETH_PCAP_TX_IFACE_ARG) == 0) {
 446                         if (open_single_iface(tx->name, &tx->pcap) < 0)
 447                                 return -1;
 448                 }
 449         }
 450
 451         /* If not open already, open rx pcaps */
 452         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 453                 rx = &internals->rx_queue[i];
 454
 455                 if (rx->pcap != NULL)
 456                         continue;
 457
 458                 if (strcmp(rx->type, ETH_PCAP_RX_PCAP_ARG) == 0) {
 459                         if (open_single_rx_pcap(rx->name, &rx->pcap) < 0)
 460                                 return -1;
 461                 } else if (strcmp(rx->type, ETH_PCAP_RX_IFACE_ARG) == 0) {
 462                         if (open_single_iface(rx->name, &rx->pcap) < 0)
 463                                 return -1;
 464                 }
 465         }
 466
 467 status_up:
 468         dev->data->dev_link.link_status = ETH_LINK_UP;
 469
 470         return 0;
 471 }
 472
 473 /*
 474  * This function gets called when the current port gets stopped.
 475  * Is the only place for us to close all the tx streams dumpers.
 476  * If not called the dumpers will be flushed within each tx burst.
 477  */
 478 static void
 479 eth_dev_stop(struct rte_eth_dev *dev)
 480 {
 481         unsigned int i;
 482         struct pmd_internals *internals = dev->data->dev_private;
 483         struct pcap_tx_queue *tx;
 484         struct pcap_rx_queue *rx;
 485
 486         /* Special iface case. Single pcap is open and shared between tx/rx. */
 487         if (internals->single_iface) {
 488                 tx = &internals->tx_queue[0];
 489                 rx = &internals->rx_queue[0];
 490                 pcap_close(tx->pcap);
 491                 tx->pcap = NULL;
 492                 rx->pcap = NULL;
 493                 goto status_down;
 494         }
 495
 496         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 497                 tx = &internals->tx_queue[i];
 498
 499                 if (tx->dumper != NULL) {
 500                         pcap_dump_close(tx->dumper);
 501                         tx->dumper = NULL;
 502                 }
 503
 504                 if (tx->pcap != NULL) {
 505                         pcap_close(tx->pcap);
 506                         tx->pcap = NULL;
 507                 }
 508         }
 509
 510         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 511                 rx = &internals->rx_queue[i];
 512
 513                 if (rx->pcap != NULL) {
 514                         pcap_close(rx->pcap);
 515                         rx->pcap = NULL;
 516                 }
 517         }
 518
 519 status_down:
 520         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 521 }
 522
 523 static int
 524 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 525 {
 526         return 0;
 527 }
 528
 529 static void
 530 eth_dev_info(struct rte_eth_dev *dev,
 531                 struct rte_eth_dev_info *dev_info)
 532 {
 533         struct pmd_internals *internals = dev->data->dev_private;
 534
 535         dev_info->if_index = internals->if_index;
 536         dev_info->max_mac_addrs = 1;
 537         dev_info->max_rx_pktlen = (uint32_t) -1;
 538         dev_info->max_rx_queues = dev->data->nb_rx_queues;
 539         dev_info->max_tx_queues = dev->data->nb_tx_queues;
 540         dev_info->min_rx_bufsize = 0;
 541 }
 542
 543 static int
 544 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 545 {
 546         unsigned int i;
 547         unsigned long rx_packets_total = 0, rx_bytes_total = 0;
 548         unsigned long tx_packets_total = 0, tx_bytes_total = 0;
 549         unsigned long tx_packets_err_total = 0;
 550         const struct pmd_internals *internal = dev->data->dev_private;
 551
 552         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 553                         i < dev->data->nb_rx_queues; i++) {
 554                 stats->q_ipackets[i] = internal->rx_queue[i].rx_stat.pkts;
 555                 stats->q_ibytes[i] = internal->rx_queue[i].rx_stat.bytes;
 556                 rx_packets_total += stats->q_ipackets[i];
 557                 rx_bytes_total += stats->q_ibytes[i];
 558         }
 559
 560         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 561                         i < dev->data->nb_tx_queues; i++) {
 562                 stats->q_opackets[i] = internal->tx_queue[i].tx_stat.pkts;
 563                 stats->q_obytes[i] = internal->tx_queue[i].tx_stat.bytes;
 564                 stats->q_errors[i] = internal->tx_queue[i].tx_stat.err_pkts;
 565                 tx_packets_total += stats->q_opackets[i];
 566                 tx_bytes_total += stats->q_obytes[i];
 567                 tx_packets_err_total += stats->q_errors[i];
 568         }
 569
 570         stats->ipackets = rx_packets_total;
 571         stats->ibytes = rx_bytes_total;
 572         stats->opackets = tx_packets_total;
 573         stats->obytes = tx_bytes_total;
 574         stats->oerrors = tx_packets_err_total;
 575
 576         return 0;
 577 }
 578
 579 static void
 580 eth_stats_reset(struct rte_eth_dev *dev)
 581 {
 582         unsigned int i;
 583         struct pmd_internals *internal = dev->data->dev_private;
 584
 585         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 586                 internal->rx_queue[i].rx_stat.pkts = 0;
 587                 internal->rx_queue[i].rx_stat.bytes = 0;
 588         }
 589
 590         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 591                 internal->tx_queue[i].tx_stat.pkts = 0;
 592                 internal->tx_queue[i].tx_stat.bytes = 0;
 593                 internal->tx_queue[i].tx_stat.err_pkts = 0;
 594         }
 595 }
 596
 597 static void
 598 eth_dev_close(struct rte_eth_dev *dev __rte_unused)
 599 {
 600 }
 601
 602 static void
 603 eth_queue_release(void *q __rte_unused)
 604 {
 605 }
 606
 607 static int
 608 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 609                 int wait_to_complete __rte_unused)
 610 {
 611         return 0;
 612 }
 613
 614 static int
 615 eth_rx_queue_setup(struct rte_eth_dev *dev,
 616                 uint16_t rx_queue_id,
 617                 uint16_t nb_rx_desc __rte_unused,
 618                 unsigned int socket_id __rte_unused,
 619                 const struct rte_eth_rxconf *rx_conf __rte_unused,
 620                 struct rte_mempool *mb_pool)
 621 {
 622         struct pmd_internals *internals = dev->data->dev_private;
 623         struct pcap_rx_queue *pcap_q = &internals->rx_queue[rx_queue_id];
 624
 625         pcap_q->mb_pool = mb_pool;
 626         dev->data->rx_queues[rx_queue_id] = pcap_q;
 627         pcap_q->in_port = dev->data->port_id;
 628
 629         return 0;
 630 }
 631
 632 static int
 633 eth_tx_queue_setup(struct rte_eth_dev *dev,
 634                 uint16_t tx_queue_id,
 635                 uint16_t nb_tx_desc __rte_unused,
 636                 unsigned int socket_id __rte_unused,
 637                 const struct rte_eth_txconf *tx_conf __rte_unused)
 638 {
 639         struct pmd_internals *internals = dev->data->dev_private;
 640
 641         dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
 642
 643         return 0;
 644 }
 645
 646 static const struct eth_dev_ops ops = {
 647         .dev_start = eth_dev_start,
 648         .dev_stop = eth_dev_stop,
 649         .dev_close = eth_dev_close,
 650         .dev_configure = eth_dev_configure,
 651         .dev_infos_get = eth_dev_info,
 652         .rx_queue_setup = eth_rx_queue_setup,
 653         .tx_queue_setup = eth_tx_queue_setup,
 654         .rx_queue_release = eth_queue_release,
 655         .tx_queue_release = eth_queue_release,
 656         .link_update = eth_link_update,
 657         .stats_get = eth_stats_get,
 658         .stats_reset = eth_stats_reset,
 659 };
 660
 661 /*
 662  * Function handler that opens the pcap file for reading a stores a
 663  * reference of it for use it later on.
 664  */
 665 static int
 666 open_rx_pcap(const char *key, const char *value, void *extra_args)
 667 {
 668         const char *pcap_filename = value;
 669         struct pmd_devargs *rx = extra_args;
 670         pcap_t *pcap = NULL;
 671
 672         if (rx->num_of_queue >= RTE_PMD_PCAP_MAX_QUEUES)
 673                 return -1;
 674         if (open_single_rx_pcap(pcap_filename, &pcap) < 0)
 675                 return -1;
 676
 677         rx->queue[rx->num_of_queue].pcap = pcap;
 678         rx->queue[rx->num_of_queue].name = pcap_filename;
 679         rx->queue[rx->num_of_queue].type = key;
 680         rx->num_of_queue++;
 681
 682         return 0;
 683 }
 684
 685 /*
 686  * Opens a pcap file for writing and stores a reference to it
 687  * for use it later on.
 688  */
 689 static int
 690 open_tx_pcap(const char *key, const char *value, void *extra_args)
 691 {
 692         const char *pcap_filename = value;
 693         struct pmd_devargs *dumpers = extra_args;
 694         pcap_dumper_t *dumper;
 695
 696         if (dumpers->num_of_queue >= RTE_PMD_PCAP_MAX_QUEUES)
 697                 return -1;
 698         if (open_single_tx_pcap(pcap_filename, &dumper) < 0)
 699                 return -1;
 700
 701         dumpers->queue[dumpers->num_of_queue].dumper = dumper;
 702         dumpers->queue[dumpers->num_of_queue].name = pcap_filename;
 703         dumpers->queue[dumpers->num_of_queue].type = key;
 704         dumpers->num_of_queue++;
 705
 706         return 0;
 707 }
 708
 709 /*
 710  * Opens an interface for reading and writing
 711  */
 712 static inline int
 713 open_rx_tx_iface(const char *key, const char *value, void *extra_args)
 714 {
 715         const char *iface = value;
 716         struct pmd_devargs *tx = extra_args;
 717         pcap_t *pcap = NULL;
 718
 719         if (open_single_iface(iface, &pcap) < 0)
 720                 return -1;
 721
 722         tx->queue[0].pcap = pcap;
 723         tx->queue[0].name = iface;
 724         tx->queue[0].type = key;
 725
 726         return 0;
 727 }
 728
 729 /*
 730  * Opens a NIC for reading packets from it
 731  */
 732 static inline int
 733 open_rx_iface(const char *key, const char *value, void *extra_args)
 734 {
 735         const char *iface = value;
 736         struct pmd_devargs *rx = extra_args;
 737         pcap_t *pcap = NULL;
 738
 739         if (rx->num_of_queue >= RTE_PMD_PCAP_MAX_QUEUES)
 740                 return -1;
 741         if (open_single_iface(iface, &pcap) < 0)
 742                 return -1;
 743         rx->queue[rx->num_of_queue].pcap = pcap;
 744         rx->queue[rx->num_of_queue].name = iface;
 745         rx->queue[rx->num_of_queue].type = key;
 746         rx->num_of_queue++;
 747
 748         return 0;
 749 }
 750
 751 /*
 752  * Opens a NIC for writing packets to it
 753  */
 754 static int
 755 open_tx_iface(const char *key, const char *value, void *extra_args)
 756 {
 757         const char *iface = value;
 758         struct pmd_devargs *tx = extra_args;
 759         pcap_t *pcap;
 760
 761         if (tx->num_of_queue >= RTE_PMD_PCAP_MAX_QUEUES)
 762                 return -1;
 763         if (open_single_iface(iface, &pcap) < 0)
 764                 return -1;
 765         tx->queue[tx->num_of_queue].pcap = pcap;
 766         tx->queue[tx->num_of_queue].name = iface;
 767         tx->queue[tx->num_of_queue].type = key;
 768         tx->num_of_queue++;
 769
 770         return 0;
 771 }
 772
 773 static struct rte_vdev_driver pmd_pcap_drv;
 774
 775 static int
 776 pmd_init_internals(struct rte_vdev_device *vdev,
 777                 const unsigned int nb_rx_queues,
 778                 const unsigned int nb_tx_queues,
 779                 struct pmd_internals **internals,
 780                 struct rte_eth_dev **eth_dev)
 781 {
 782         struct rte_eth_dev_data *data;
 783         unsigned int numa_node = vdev->device.numa_node;
 784
 785         PMD_LOG(INFO, "Creating pcap-backed ethdev on numa socket %d",
 786                 numa_node);
 787
 788         /* reserve an ethdev entry */
 789         *eth_dev = rte_eth_vdev_allocate(vdev, sizeof(**internals));
 790         if (!(*eth_dev))
 791                 return -1;
 792
 793         /* now put it all together
 794          * - store queue data in internals,
 795          * - store numa_node info in eth_dev
 796          * - point eth_dev_data to internals
 797          * - and point eth_dev structure to new eth_dev_data structure
 798          */
 799         *internals = (*eth_dev)->data->dev_private;
 800         data = (*eth_dev)->data;
 801         data->nb_rx_queues = (uint16_t)nb_rx_queues;
 802         data->nb_tx_queues = (uint16_t)nb_tx_queues;
 803         data->dev_link = pmd_link;
 804         data->mac_addrs = &eth_addr;
 805
 806         /*
 807          * NOTE: we'll replace the data element, of originally allocated
 808          * eth_dev so the rings are local per-process
 809          */
 810         (*eth_dev)->dev_ops = &ops;
 811
 812         return 0;
 813 }
 814
 815 static int
 816 eth_from_pcaps_common(struct rte_vdev_device *vdev,
 817                 struct pmd_devargs *rx_queues, const unsigned int nb_rx_queues,
 818                 struct pmd_devargs *tx_queues, const unsigned int nb_tx_queues,
 819                 struct rte_kvargs *kvlist, struct pmd_internals **internals,
 820                 struct rte_eth_dev **eth_dev)
 821 {
 822         struct rte_kvargs_pair *pair = NULL;
 823         unsigned int k_idx;
 824         unsigned int i;
 825
 826         /* do some parameter checking */
 827         if (rx_queues == NULL && nb_rx_queues > 0)
 828                 return -1;
 829         if (tx_queues == NULL && nb_tx_queues > 0)
 830                 return -1;
 831
 832         if (pmd_init_internals(vdev, nb_rx_queues, nb_tx_queues, internals,
 833                         eth_dev) < 0)
 834                 return -1;
 835
 836         for (i = 0; i < nb_rx_queues; i++) {
 837                 struct pcap_rx_queue *rx = &(*internals)->rx_queue[i];
 838                 struct devargs_queue *queue = &rx_queues->queue[i];
 839
 840                 rx->pcap = queue->pcap;
 841                 snprintf(rx->name, sizeof(rx->name), "%s", queue->name);
 842                 snprintf(rx->type, sizeof(rx->type), "%s", queue->type);
 843         }
 844
 845         for (i = 0; i < nb_tx_queues; i++) {
 846                 struct pcap_tx_queue *tx = &(*internals)->tx_queue[i];
 847                 struct devargs_queue *queue = &tx_queues->queue[i];
 848
 849                 tx->dumper = queue->dumper;
 850                 tx->pcap = queue->pcap;
 851                 snprintf(tx->name, sizeof(tx->name), "%s", queue->name);
 852                 snprintf(tx->type, sizeof(tx->type), "%s", queue->type);
 853         }
 854
 855         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 856                 pair = &kvlist->pairs[k_idx];
 857                 if (strstr(pair->key, ETH_PCAP_IFACE_ARG) != NULL)
 858                         break;
 859         }
 860
 861         if (pair == NULL)
 862                 (*internals)->if_index = 0;
 863         else
 864                 (*internals)->if_index = if_nametoindex(pair->value);
 865
 866         return 0;
 867 }
 868
 869 static int
 870 eth_from_pcaps(struct rte_vdev_device *vdev,
 871                 struct pmd_devargs *rx_queues, const unsigned int nb_rx_queues,
 872                 struct pmd_devargs *tx_queues, const unsigned int nb_tx_queues,
 873                 struct rte_kvargs *kvlist, int single_iface,
 874                 unsigned int using_dumpers)
 875 {
 876         struct pmd_internals *internals = NULL;
 877         struct rte_eth_dev *eth_dev = NULL;
 878         int ret;
 879
 880         ret = eth_from_pcaps_common(vdev, rx_queues, nb_rx_queues,
 881                 tx_queues, nb_tx_queues, kvlist, &internals, &eth_dev);
 882
 883         if (ret < 0)
 884                 return ret;
 885
 886         /* store weather we are using a single interface for rx/tx or not */
 887         internals->single_iface = single_iface;
 888
 889         eth_dev->rx_pkt_burst = eth_pcap_rx;
 890
 891         if (using_dumpers)
 892                 eth_dev->tx_pkt_burst = eth_pcap_tx_dumper;
 893         else
 894                 eth_dev->tx_pkt_burst = eth_pcap_tx;
 895
 896         rte_eth_dev_probing_finish(eth_dev);
 897         return 0;
 898 }
 899
 900 static int
 901 pmd_pcap_probe(struct rte_vdev_device *dev)
 902 {
 903         const char *name;
 904         unsigned int is_rx_pcap = 0, is_tx_pcap = 0;
 905         struct rte_kvargs *kvlist;
 906         struct pmd_devargs pcaps = {0};
 907         struct pmd_devargs dumpers = {0};
 908         struct rte_eth_dev *eth_dev;
 909         int single_iface = 0;
 910         int ret;
 911
 912         name = rte_vdev_device_name(dev);
 913         PMD_LOG(INFO, "Initializing pmd_pcap for %s", name);
 914
 915         gettimeofday(&start_time, NULL);
 916         start_cycles = rte_get_timer_cycles();
 917         hz = rte_get_timer_hz();
 918
 919         if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
 920             strlen(rte_vdev_device_args(dev)) == 0) {
 921                 eth_dev = rte_eth_dev_attach_secondary(name);
 922                 if (!eth_dev) {
 923                         PMD_LOG(ERR, "Failed to probe %s", name);
 924                         return -1;
 925                 }
 926                 /* TODO: request info from primary to set up Rx and Tx */
 927                 eth_dev->dev_ops = &ops;
 928                 rte_eth_dev_probing_finish(eth_dev);
 929                 return 0;
 930         }
 931
 932         kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_arguments);
 933         if (kvlist == NULL)
 934                 return -1;
 935
 936         /*
 937          * If iface argument is passed we open the NICs and use them for
 938          * reading / writing
 939          */
 940         if (rte_kvargs_count(kvlist, ETH_PCAP_IFACE_ARG) == 1) {
 941
 942                 ret = rte_kvargs_process(kvlist, ETH_PCAP_IFACE_ARG,
 943                                 &open_rx_tx_iface, &pcaps);
 944
 945                 if (ret < 0)
 946                         goto free_kvlist;
 947
 948                 dumpers.queue[0] = pcaps.queue[0];
 949
 950                 single_iface = 1;
 951                 pcaps.num_of_queue = 1;
 952                 dumpers.num_of_queue = 1;
 953
 954                 goto create_eth;
 955         }
 956
 957         /*
 958          * We check whether we want to open a RX stream from a real NIC or a
 959          * pcap file
 960          */
 961         is_rx_pcap = rte_kvargs_count(kvlist, ETH_PCAP_RX_PCAP_ARG) ? 1 : 0;
 962         pcaps.num_of_queue = 0;
 963
 964         if (is_rx_pcap)
 965                 ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_PCAP_ARG,
 966                                 &open_rx_pcap, &pcaps);
 967         else
 968                 ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_IFACE_ARG,
 969                                 &open_rx_iface, &pcaps);
 970
 971         if (ret < 0)
 972                 goto free_kvlist;
 973
 974         /*
 975          * We check whether we want to open a TX stream to a real NIC or a
 976          * pcap file
 977          */
 978         is_tx_pcap = rte_kvargs_count(kvlist, ETH_PCAP_TX_PCAP_ARG) ? 1 : 0;
 979         dumpers.num_of_queue = 0;
 980
 981         if (is_tx_pcap)
 982                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_PCAP_ARG,
 983                                 &open_tx_pcap, &dumpers);
 984         else
 985                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_IFACE_ARG,
 986                                 &open_tx_iface, &dumpers);
 987
 988         if (ret < 0)
 989                 goto free_kvlist;
 990
 991 create_eth:
 992         ret = eth_from_pcaps(dev, &pcaps, pcaps.num_of_queue, &dumpers,
 993                 dumpers.num_of_queue, kvlist, single_iface, is_tx_pcap);
 994
 995 free_kvlist:
 996         rte_kvargs_free(kvlist);
 997
 998         return ret;
 999 }
1000
1001 static int
1002 pmd_pcap_remove(struct rte_vdev_device *dev)
1003 {
1004         struct rte_eth_dev *eth_dev = NULL;
1005
1006         PMD_LOG(INFO, "Closing pcap ethdev on numa socket %d",
1007                         rte_socket_id());
1008
1009         if (!dev)
1010                 return -1;
1011
1012         /* reserve an ethdev entry */
1013         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1014         if (eth_dev == NULL)
1015                 return -1;
1016
1017         rte_free(eth_dev->data->dev_private);
1018
1019         rte_eth_dev_release_port(eth_dev);
1020
1021         return 0;
1022 }
1023
1024 static struct rte_vdev_driver pmd_pcap_drv = {
1025         .probe = pmd_pcap_probe,
1026         .remove = pmd_pcap_remove,
1027 };
1028
1029 RTE_PMD_REGISTER_VDEV(net_pcap, pmd_pcap_drv);
1030 RTE_PMD_REGISTER_ALIAS(net_pcap, eth_pcap);
1031 RTE_PMD_REGISTER_PARAM_STRING(net_pcap,
1032         ETH_PCAP_RX_PCAP_ARG "=<string> "
1033         ETH_PCAP_TX_PCAP_ARG "=<string> "
1034         ETH_PCAP_RX_IFACE_ARG "=<ifc> "
1035         ETH_PCAP_TX_IFACE_ARG "=<ifc> "
1036         ETH_PCAP_IFACE_ARG "=<ifc>");
1037
1038 RTE_INIT(eth_pcap_init_log);
1039 static void
1040 eth_pcap_init_log(void)
1041 {
1042         eth_pcap_logtype = rte_log_register("pmd.net.pcap");
1043         if (eth_pcap_logtype >= 0)
1044                 rte_log_set_level(eth_pcap_logtype, RTE_LOG_NOTICE);
1045 }