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
 436                 goto status_up;
 437         }
 438
 439         /* If not open already, open tx pcaps/dumpers */
 440         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 441                 tx = &internals->tx_queue[i];
 442
 443                 if (!tx->dumper &&
 444                                 strcmp(tx->type, ETH_PCAP_TX_PCAP_ARG) == 0) {
 445                         if (open_single_tx_pcap(tx->name, &tx->dumper) < 0)
 446                                 return -1;
 447                 } else if (!tx->pcap &&
 448                                 strcmp(tx->type, ETH_PCAP_TX_IFACE_ARG) == 0) {
 449                         if (open_single_iface(tx->name, &tx->pcap) < 0)
 450                                 return -1;
 451                 }
 452         }
 453
 454         /* If not open already, open rx pcaps */
 455         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 456                 rx = &internals->rx_queue[i];
 457
 458                 if (rx->pcap != NULL)
 459                         continue;
 460
 461                 if (strcmp(rx->type, ETH_PCAP_RX_PCAP_ARG) == 0) {
 462                         if (open_single_rx_pcap(rx->name, &rx->pcap) < 0)
 463                                 return -1;
 464                 } else if (strcmp(rx->type, ETH_PCAP_RX_IFACE_ARG) == 0) {
 465                         if (open_single_iface(rx->name, &rx->pcap) < 0)
 466                                 return -1;
 467                 }
 468         }
 469
 470 status_up:
 471         for (i = 0; i < dev->data->nb_rx_queues; i++)
 472                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 473
 474         for (i = 0; i < dev->data->nb_tx_queues; i++)
 475                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 476
 477         dev->data->dev_link.link_status = ETH_LINK_UP;
 478
 479         return 0;
 480 }
 481
 482 /*
 483  * This function gets called when the current port gets stopped.
 484  * Is the only place for us to close all the tx streams dumpers.
 485  * If not called the dumpers will be flushed within each tx burst.
 486  */
 487 static void
 488 eth_dev_stop(struct rte_eth_dev *dev)
 489 {
 490         unsigned int i;
 491         struct pmd_internals *internals = dev->data->dev_private;
 492         struct pcap_tx_queue *tx;
 493         struct pcap_rx_queue *rx;
 494
 495         /* Special iface case. Single pcap is open and shared between tx/rx. */
 496         if (internals->single_iface) {
 497                 tx = &internals->tx_queue[0];
 498                 rx = &internals->rx_queue[0];
 499                 pcap_close(tx->pcap);
 500                 tx->pcap = NULL;
 501                 rx->pcap = NULL;
 502                 goto status_down;
 503         }
 504
 505         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 506                 tx = &internals->tx_queue[i];
 507
 508                 if (tx->dumper != NULL) {
 509                         pcap_dump_close(tx->dumper);
 510                         tx->dumper = NULL;
 511                 }
 512
 513                 if (tx->pcap != NULL) {
 514                         pcap_close(tx->pcap);
 515                         tx->pcap = NULL;
 516                 }
 517         }
 518
 519         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 520                 rx = &internals->rx_queue[i];
 521
 522                 if (rx->pcap != NULL) {
 523                         pcap_close(rx->pcap);
 524                         rx->pcap = NULL;
 525                 }
 526         }
 527
 528 status_down:
 529         for (i = 0; i < dev->data->nb_rx_queues; i++)
 530                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 531
 532         for (i = 0; i < dev->data->nb_tx_queues; i++)
 533                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 534
 535         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 536 }
 537
 538 static int
 539 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 540 {
 541         return 0;
 542 }
 543
 544 static void
 545 eth_dev_info(struct rte_eth_dev *dev,
 546                 struct rte_eth_dev_info *dev_info)
 547 {
 548         struct pmd_internals *internals = dev->data->dev_private;
 549
 550         dev_info->if_index = internals->if_index;
 551         dev_info->max_mac_addrs = 1;
 552         dev_info->max_rx_pktlen = (uint32_t) -1;
 553         dev_info->max_rx_queues = dev->data->nb_rx_queues;
 554         dev_info->max_tx_queues = dev->data->nb_tx_queues;
 555         dev_info->min_rx_bufsize = 0;
 556 }
 557
 558 static int
 559 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 560 {
 561         unsigned int i;
 562         unsigned long rx_packets_total = 0, rx_bytes_total = 0;
 563         unsigned long tx_packets_total = 0, tx_bytes_total = 0;
 564         unsigned long tx_packets_err_total = 0;
 565         const struct pmd_internals *internal = dev->data->dev_private;
 566
 567         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 568                         i < dev->data->nb_rx_queues; i++) {
 569                 stats->q_ipackets[i] = internal->rx_queue[i].rx_stat.pkts;
 570                 stats->q_ibytes[i] = internal->rx_queue[i].rx_stat.bytes;
 571                 rx_packets_total += stats->q_ipackets[i];
 572                 rx_bytes_total += stats->q_ibytes[i];
 573         }
 574
 575         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 576                         i < dev->data->nb_tx_queues; i++) {
 577                 stats->q_opackets[i] = internal->tx_queue[i].tx_stat.pkts;
 578                 stats->q_obytes[i] = internal->tx_queue[i].tx_stat.bytes;
 579                 stats->q_errors[i] = internal->tx_queue[i].tx_stat.err_pkts;
 580                 tx_packets_total += stats->q_opackets[i];
 581                 tx_bytes_total += stats->q_obytes[i];
 582                 tx_packets_err_total += stats->q_errors[i];
 583         }
 584
 585         stats->ipackets = rx_packets_total;
 586         stats->ibytes = rx_bytes_total;
 587         stats->opackets = tx_packets_total;
 588         stats->obytes = tx_bytes_total;
 589         stats->oerrors = tx_packets_err_total;
 590
 591         return 0;
 592 }
 593
 594 static void
 595 eth_stats_reset(struct rte_eth_dev *dev)
 596 {
 597         unsigned int i;
 598         struct pmd_internals *internal = dev->data->dev_private;
 599
 600         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 601                 internal->rx_queue[i].rx_stat.pkts = 0;
 602                 internal->rx_queue[i].rx_stat.bytes = 0;
 603         }
 604
 605         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 606                 internal->tx_queue[i].tx_stat.pkts = 0;
 607                 internal->tx_queue[i].tx_stat.bytes = 0;
 608                 internal->tx_queue[i].tx_stat.err_pkts = 0;
 609         }
 610 }
 611
 612 static void
 613 eth_dev_close(struct rte_eth_dev *dev __rte_unused)
 614 {
 615 }
 616
 617 static void
 618 eth_queue_release(void *q __rte_unused)
 619 {
 620 }
 621
 622 static int
 623 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 624                 int wait_to_complete __rte_unused)
 625 {
 626         return 0;
 627 }
 628
 629 static int
 630 eth_rx_queue_setup(struct rte_eth_dev *dev,
 631                 uint16_t rx_queue_id,
 632                 uint16_t nb_rx_desc __rte_unused,
 633                 unsigned int socket_id __rte_unused,
 634                 const struct rte_eth_rxconf *rx_conf __rte_unused,
 635                 struct rte_mempool *mb_pool)
 636 {
 637         struct pmd_internals *internals = dev->data->dev_private;
 638         struct pcap_rx_queue *pcap_q = &internals->rx_queue[rx_queue_id];
 639
 640         pcap_q->mb_pool = mb_pool;
 641         dev->data->rx_queues[rx_queue_id] = pcap_q;
 642         pcap_q->in_port = dev->data->port_id;
 643
 644         return 0;
 645 }
 646
 647 static int
 648 eth_tx_queue_setup(struct rte_eth_dev *dev,
 649                 uint16_t tx_queue_id,
 650                 uint16_t nb_tx_desc __rte_unused,
 651                 unsigned int socket_id __rte_unused,
 652                 const struct rte_eth_txconf *tx_conf __rte_unused)
 653 {
 654         struct pmd_internals *internals = dev->data->dev_private;
 655
 656         dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
 657
 658         return 0;
 659 }
 660
 661 static int
 662 eth_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 663 {
 664         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
 665
 666         return 0;
 667 }
 668
 669 static int
 670 eth_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
 671 {
 672         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
 673
 674         return 0;
 675 }
 676
 677 static int
 678 eth_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 679 {
 680         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
 681
 682         return 0;
 683 }
 684
 685 static int
 686 eth_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
 687 {
 688         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
 689
 690         return 0;
 691 }
 692
 693 static const struct eth_dev_ops ops = {
 694         .dev_start = eth_dev_start,
 695         .dev_stop = eth_dev_stop,
 696         .dev_close = eth_dev_close,
 697         .dev_configure = eth_dev_configure,
 698         .dev_infos_get = eth_dev_info,
 699         .rx_queue_setup = eth_rx_queue_setup,
 700         .tx_queue_setup = eth_tx_queue_setup,
 701         .rx_queue_start = eth_rx_queue_start,
 702         .tx_queue_start = eth_tx_queue_start,
 703         .rx_queue_stop = eth_rx_queue_stop,
 704         .tx_queue_stop = eth_tx_queue_stop,
 705         .rx_queue_release = eth_queue_release,
 706         .tx_queue_release = eth_queue_release,
 707         .link_update = eth_link_update,
 708         .stats_get = eth_stats_get,
 709         .stats_reset = eth_stats_reset,
 710 };
 711
 712 static int
 713 add_queue(struct pmd_devargs *pmd, const char *name, const char *type,
 714                 pcap_t *pcap, pcap_dumper_t *dumper)
 715 {
 716         if (pmd->num_of_queue >= RTE_PMD_PCAP_MAX_QUEUES)
 717                 return -1;
 718         if (pcap)
 719                 pmd->queue[pmd->num_of_queue].pcap = pcap;
 720         if (dumper)
 721                 pmd->queue[pmd->num_of_queue].dumper = dumper;
 722         pmd->queue[pmd->num_of_queue].name = name;
 723         pmd->queue[pmd->num_of_queue].type = type;
 724         pmd->num_of_queue++;
 725         return 0;
 726 }
 727
 728 /*
 729  * Function handler that opens the pcap file for reading a stores a
 730  * reference of it for use it later on.
 731  */
 732 static int
 733 open_rx_pcap(const char *key, const char *value, void *extra_args)
 734 {
 735         const char *pcap_filename = value;
 736         struct pmd_devargs *rx = extra_args;
 737         pcap_t *pcap = NULL;
 738
 739         if (open_single_rx_pcap(pcap_filename, &pcap) < 0)
 740                 return -1;
 741
 742         if (add_queue(rx, pcap_filename, key, pcap, NULL) < 0) {
 743                 pcap_close(pcap);
 744                 return -1;
 745         }
 746
 747         return 0;
 748 }
 749
 750 /*
 751  * Opens a pcap file for writing and stores a reference to it
 752  * for use it later on.
 753  */
 754 static int
 755 open_tx_pcap(const char *key, const char *value, void *extra_args)
 756 {
 757         const char *pcap_filename = value;
 758         struct pmd_devargs *dumpers = extra_args;
 759         pcap_dumper_t *dumper;
 760
 761         if (open_single_tx_pcap(pcap_filename, &dumper) < 0)
 762                 return -1;
 763
 764         if (add_queue(dumpers, pcap_filename, key, NULL, dumper) < 0) {
 765                 pcap_dump_close(dumper);
 766                 return -1;
 767         }
 768
 769         return 0;
 770 }
 771
 772 /*
 773  * Opens an interface for reading and writing
 774  */
 775 static inline int
 776 open_rx_tx_iface(const char *key, const char *value, void *extra_args)
 777 {
 778         const char *iface = value;
 779         struct pmd_devargs *tx = extra_args;
 780         pcap_t *pcap = NULL;
 781
 782         if (open_single_iface(iface, &pcap) < 0)
 783                 return -1;
 784
 785         tx->queue[0].pcap = pcap;
 786         tx->queue[0].name = iface;
 787         tx->queue[0].type = key;
 788
 789         return 0;
 790 }
 791
 792 static inline int
 793 set_iface_direction(const char *iface, pcap_t *pcap,
 794                 pcap_direction_t direction)
 795 {
 796         const char *direction_str = (direction == PCAP_D_IN) ? "IN" : "OUT";
 797         if (pcap_setdirection(pcap, direction) < 0) {
 798                 PMD_LOG(ERR, "Setting %s pcap direction %s failed - %s\n",
 799                                 iface, direction_str, pcap_geterr(pcap));
 800                 return -1;
 801         }
 802         PMD_LOG(INFO, "Setting %s pcap direction %s\n",
 803                         iface, direction_str);
 804         return 0;
 805 }
 806
 807 static inline int
 808 open_iface(const char *key, const char *value, void *extra_args)
 809 {
 810         const char *iface = value;
 811         struct pmd_devargs *pmd = extra_args;
 812         pcap_t *pcap = NULL;
 813
 814         if (open_single_iface(iface, &pcap) < 0)
 815                 return -1;
 816         if (add_queue(pmd, iface, key, pcap, NULL) < 0) {
 817                 pcap_close(pcap);
 818                 return -1;
 819         }
 820
 821         return 0;
 822 }
 823
 824 /*
 825  * Opens a NIC for reading packets from it
 826  */
 827 static inline int
 828 open_rx_iface(const char *key, const char *value, void *extra_args)
 829 {
 830         int ret = open_iface(key, value, extra_args);
 831         if (ret < 0)
 832                 return ret;
 833         if (strcmp(key, ETH_PCAP_RX_IFACE_IN_ARG) == 0) {
 834                 struct pmd_devargs *pmd = extra_args;
 835                 unsigned int qid = pmd->num_of_queue - 1;
 836
 837                 set_iface_direction(pmd->queue[qid].name,
 838                                 pmd->queue[qid].pcap,
 839                                 PCAP_D_IN);
 840         }
 841
 842         return 0;
 843 }
 844
 845 static inline int
 846 rx_iface_args_process(const char *key, const char *value, void *extra_args)
 847 {
 848         if (strcmp(key, ETH_PCAP_RX_IFACE_ARG) == 0 ||
 849                         strcmp(key, ETH_PCAP_RX_IFACE_IN_ARG) == 0)
 850                 return open_rx_iface(key, value, extra_args);
 851
 852         return 0;
 853 }
 854
 855 /*
 856  * Opens a NIC for writing packets to it
 857  */
 858 static int
 859 open_tx_iface(const char *key, const char *value, void *extra_args)
 860 {
 861         return open_iface(key, value, extra_args);
 862 }
 863
 864 static struct rte_vdev_driver pmd_pcap_drv;
 865
 866 static int
 867 pmd_init_internals(struct rte_vdev_device *vdev,
 868                 const unsigned int nb_rx_queues,
 869                 const unsigned int nb_tx_queues,
 870                 struct pmd_internals **internals,
 871                 struct rte_eth_dev **eth_dev)
 872 {
 873         struct rte_eth_dev_data *data;
 874         unsigned int numa_node = vdev->device.numa_node;
 875
 876         PMD_LOG(INFO, "Creating pcap-backed ethdev on numa socket %d",
 877                 numa_node);
 878
 879         /* reserve an ethdev entry */
 880         *eth_dev = rte_eth_vdev_allocate(vdev, sizeof(**internals));
 881         if (!(*eth_dev))
 882                 return -1;
 883
 884         /* now put it all together
 885          * - store queue data in internals,
 886          * - store numa_node info in eth_dev
 887          * - point eth_dev_data to internals
 888          * - and point eth_dev structure to new eth_dev_data structure
 889          */
 890         *internals = (*eth_dev)->data->dev_private;
 891         data = (*eth_dev)->data;
 892         data->nb_rx_queues = (uint16_t)nb_rx_queues;
 893         data->nb_tx_queues = (uint16_t)nb_tx_queues;
 894         data->dev_link = pmd_link;
 895         data->mac_addrs = &eth_addr;
 896
 897         /*
 898          * NOTE: we'll replace the data element, of originally allocated
 899          * eth_dev so the rings are local per-process
 900          */
 901         (*eth_dev)->dev_ops = &ops;
 902
 903         return 0;
 904 }
 905
 906 static int
 907 eth_from_pcaps_common(struct rte_vdev_device *vdev,
 908                 struct pmd_devargs *rx_queues, const unsigned int nb_rx_queues,
 909                 struct pmd_devargs *tx_queues, const unsigned int nb_tx_queues,
 910                 struct rte_kvargs *kvlist, struct pmd_internals **internals,
 911                 struct rte_eth_dev **eth_dev)
 912 {
 913         struct rte_kvargs_pair *pair = NULL;
 914         unsigned int k_idx;
 915         unsigned int i;
 916
 917         /* do some parameter checking */
 918         if (rx_queues == NULL && nb_rx_queues > 0)
 919                 return -1;
 920         if (tx_queues == NULL && nb_tx_queues > 0)
 921                 return -1;
 922
 923         if (pmd_init_internals(vdev, nb_rx_queues, nb_tx_queues, internals,
 924                         eth_dev) < 0)
 925                 return -1;
 926
 927         for (i = 0; i < nb_rx_queues; i++) {
 928                 struct pcap_rx_queue *rx = &(*internals)->rx_queue[i];
 929                 struct devargs_queue *queue = &rx_queues->queue[i];
 930
 931                 rx->pcap = queue->pcap;
 932                 snprintf(rx->name, sizeof(rx->name), "%s", queue->name);
 933                 snprintf(rx->type, sizeof(rx->type), "%s", queue->type);
 934         }
 935
 936         for (i = 0; i < nb_tx_queues; i++) {
 937                 struct pcap_tx_queue *tx = &(*internals)->tx_queue[i];
 938                 struct devargs_queue *queue = &tx_queues->queue[i];
 939
 940                 tx->dumper = queue->dumper;
 941                 tx->pcap = queue->pcap;
 942                 snprintf(tx->name, sizeof(tx->name), "%s", queue->name);
 943                 snprintf(tx->type, sizeof(tx->type), "%s", queue->type);
 944         }
 945
 946         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 947                 pair = &kvlist->pairs[k_idx];
 948                 if (strstr(pair->key, ETH_PCAP_IFACE_ARG) != NULL)
 949                         break;
 950         }
 951
 952         if (pair == NULL)
 953                 (*internals)->if_index = 0;
 954         else
 955                 (*internals)->if_index = if_nametoindex(pair->value);
 956
 957         return 0;
 958 }
 959
 960 static int
 961 eth_from_pcaps(struct rte_vdev_device *vdev,
 962                 struct pmd_devargs *rx_queues, const unsigned int nb_rx_queues,
 963                 struct pmd_devargs *tx_queues, const unsigned int nb_tx_queues,
 964                 struct rte_kvargs *kvlist, int single_iface,
 965                 unsigned int using_dumpers)
 966 {
 967         struct pmd_internals *internals = NULL;
 968         struct rte_eth_dev *eth_dev = NULL;
 969         int ret;
 970
 971         ret = eth_from_pcaps_common(vdev, rx_queues, nb_rx_queues,
 972                 tx_queues, nb_tx_queues, kvlist, &internals, &eth_dev);
 973
 974         if (ret < 0)
 975                 return ret;
 976
 977         /* store weather we are using a single interface for rx/tx or not */
 978         internals->single_iface = single_iface;
 979
 980         eth_dev->rx_pkt_burst = eth_pcap_rx;
 981
 982         if (using_dumpers)
 983                 eth_dev->tx_pkt_burst = eth_pcap_tx_dumper;
 984         else
 985                 eth_dev->tx_pkt_burst = eth_pcap_tx;
 986
 987         rte_eth_dev_probing_finish(eth_dev);
 988         return 0;
 989 }
 990
 991 static int
 992 pmd_pcap_probe(struct rte_vdev_device *dev)
 993 {
 994         const char *name;
 995         unsigned int is_rx_pcap = 0, is_tx_pcap = 0;
 996         struct rte_kvargs *kvlist;
 997         struct pmd_devargs pcaps = {0};
 998         struct pmd_devargs dumpers = {0};
 999         struct rte_eth_dev *eth_dev;
1000         int single_iface = 0;
1001         int ret;
1002
1003         name = rte_vdev_device_name(dev);
1004         PMD_LOG(INFO, "Initializing pmd_pcap for %s", name);
1005
1006         gettimeofday(&start_time, NULL);
1007         start_cycles = rte_get_timer_cycles();
1008         hz = rte_get_timer_hz();
1009
1010         if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
1011             strlen(rte_vdev_device_args(dev)) == 0) {
1012                 eth_dev = rte_eth_dev_attach_secondary(name);
1013                 if (!eth_dev) {
1014                         PMD_LOG(ERR, "Failed to probe %s", name);
1015                         return -1;
1016                 }
1017                 /* TODO: request info from primary to set up Rx and Tx */
1018                 eth_dev->dev_ops = &ops;
1019                 eth_dev->device = &dev->device;
1020                 rte_eth_dev_probing_finish(eth_dev);
1021                 return 0;
1022         }
1023
1024         kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_arguments);
1025         if (kvlist == NULL)
1026                 return -1;
1027
1028         /*
1029          * If iface argument is passed we open the NICs and use them for
1030          * reading / writing
1031          */
1032         if (rte_kvargs_count(kvlist, ETH_PCAP_IFACE_ARG) == 1) {
1033
1034                 ret = rte_kvargs_process(kvlist, ETH_PCAP_IFACE_ARG,
1035                                 &open_rx_tx_iface, &pcaps);
1036
1037                 if (ret < 0)
1038                         goto free_kvlist;
1039
1040                 dumpers.queue[0] = pcaps.queue[0];
1041
1042                 single_iface = 1;
1043                 pcaps.num_of_queue = 1;
1044                 dumpers.num_of_queue = 1;
1045
1046                 goto create_eth;
1047         }
1048
1049         /*
1050          * We check whether we want to open a RX stream from a real NIC or a
1051          * pcap file
1052          */
1053         is_rx_pcap = rte_kvargs_count(kvlist, ETH_PCAP_RX_PCAP_ARG) ? 1 : 0;
1054         pcaps.num_of_queue = 0;
1055
1056         if (is_rx_pcap) {
1057                 ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_PCAP_ARG,
1058                                 &open_rx_pcap, &pcaps);
1059         } else {
1060                 ret = rte_kvargs_process(kvlist, NULL,
1061                                 &rx_iface_args_process, &pcaps);
1062         }
1063
1064         if (ret < 0)
1065                 goto free_kvlist;
1066
1067         /*
1068          * We check whether we want to open a TX stream to a real NIC or a
1069          * pcap file
1070          */
1071         is_tx_pcap = rte_kvargs_count(kvlist, ETH_PCAP_TX_PCAP_ARG) ? 1 : 0;
1072         dumpers.num_of_queue = 0;
1073
1074         if (is_tx_pcap)
1075                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_PCAP_ARG,
1076                                 &open_tx_pcap, &dumpers);
1077         else
1078                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_IFACE_ARG,
1079                                 &open_tx_iface, &dumpers);
1080
1081         if (ret < 0)
1082                 goto free_kvlist;
1083
1084 create_eth:
1085         ret = eth_from_pcaps(dev, &pcaps, pcaps.num_of_queue, &dumpers,
1086                 dumpers.num_of_queue, kvlist, single_iface, is_tx_pcap);
1087
1088 free_kvlist:
1089         rte_kvargs_free(kvlist);
1090
1091         return ret;
1092 }
1093
1094 static int
1095 pmd_pcap_remove(struct rte_vdev_device *dev)
1096 {
1097         struct rte_eth_dev *eth_dev = NULL;
1098
1099         PMD_LOG(INFO, "Closing pcap ethdev on numa socket %d",
1100                         rte_socket_id());
1101
1102         if (!dev)
1103                 return -1;
1104
1105         /* reserve an ethdev entry */
1106         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1107         if (eth_dev == NULL)
1108                 return -1;
1109
1110         rte_free(eth_dev->data->dev_private);
1111
1112         rte_eth_dev_release_port(eth_dev);
1113
1114         return 0;
1115 }
1116
1117 static struct rte_vdev_driver pmd_pcap_drv = {
1118         .probe = pmd_pcap_probe,
1119         .remove = pmd_pcap_remove,
1120 };
1121
1122 RTE_PMD_REGISTER_VDEV(net_pcap, pmd_pcap_drv);
1123 RTE_PMD_REGISTER_ALIAS(net_pcap, eth_pcap);
1124 RTE_PMD_REGISTER_PARAM_STRING(net_pcap,
1125         ETH_PCAP_RX_PCAP_ARG "=<string> "
1126         ETH_PCAP_TX_PCAP_ARG "=<string> "
1127         ETH_PCAP_RX_IFACE_ARG "=<ifc> "
1128         ETH_PCAP_RX_IFACE_IN_ARG "=<ifc> "
1129         ETH_PCAP_TX_IFACE_ARG "=<ifc> "
1130         ETH_PCAP_IFACE_ARG "=<ifc>");
1131
1132 RTE_INIT(eth_pcap_init_log)
1133 {
1134         eth_pcap_logtype = rte_log_register("pmd.net.pcap");
1135         if (eth_pcap_logtype >= 0)
1136                 rte_log_set_level(eth_pcap_logtype, RTE_LOG_NOTICE);
1137 }