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 #include <sys/socket.h>
  11 #include <sys/ioctl.h>
  12 #include <unistd.h>
  13
  14 #if defined(RTE_EXEC_ENV_FREEBSD)
  15 #include <sys/sysctl.h>
  16 #include <net/if_dl.h>
  17 #endif
  18
  19 #include <pcap.h>
  20
  21 #include <rte_cycles.h>
  22 #include <rte_ethdev_driver.h>
  23 #include <rte_ethdev_vdev.h>
  24 #include <rte_kvargs.h>
  25 #include <rte_malloc.h>
  26 #include <rte_mbuf.h>
  27 #include <rte_bus_vdev.h>
  28 #include <rte_string_fns.h>
  29
  30 #define RTE_ETH_PCAP_SNAPSHOT_LEN 65535
  31 #define RTE_ETH_PCAP_SNAPLEN RTE_ETHER_MAX_JUMBO_FRAME_LEN
  32 #define RTE_ETH_PCAP_PROMISC 1
  33 #define RTE_ETH_PCAP_TIMEOUT -1
  34
  35 #define ETH_PCAP_RX_PCAP_ARG  "rx_pcap"
  36 #define ETH_PCAP_TX_PCAP_ARG  "tx_pcap"
  37 #define ETH_PCAP_RX_IFACE_ARG "rx_iface"
  38 #define ETH_PCAP_RX_IFACE_IN_ARG "rx_iface_in"
  39 #define ETH_PCAP_TX_IFACE_ARG "tx_iface"
  40 #define ETH_PCAP_IFACE_ARG    "iface"
  41 #define ETH_PCAP_PHY_MAC_ARG  "phy_mac"
  42 #define ETH_PCAP_INFINITE_RX_ARG  "infinite_rx"
  43
  44 #define ETH_PCAP_ARG_MAXLEN     64
  45
  46 #define RTE_PMD_PCAP_MAX_QUEUES 16
  47
  48 static char errbuf[PCAP_ERRBUF_SIZE];
  49 static struct timeval start_time;
  50 static uint64_t start_cycles;
  51 static uint64_t hz;
  52 static uint8_t iface_idx;
  53
  54 struct queue_stat {
  55         volatile unsigned long pkts;
  56         volatile unsigned long bytes;
  57         volatile unsigned long err_pkts;
  58 };
  59
  60 struct pcap_rx_queue {
  61         uint16_t port_id;
  62         uint16_t queue_id;
  63         struct rte_mempool *mb_pool;
  64         struct queue_stat rx_stat;
  65         char name[PATH_MAX];
  66         char type[ETH_PCAP_ARG_MAXLEN];
  67
  68         /* Contains pre-generated packets to be looped through */
  69         struct rte_ring *pkts;
  70 };
  71
  72 struct pcap_tx_queue {
  73         uint16_t port_id;
  74         uint16_t queue_id;
  75         struct queue_stat tx_stat;
  76         char name[PATH_MAX];
  77         char type[ETH_PCAP_ARG_MAXLEN];
  78 };
  79
  80 struct pmd_internals {
  81         struct pcap_rx_queue rx_queue[RTE_PMD_PCAP_MAX_QUEUES];
  82         struct pcap_tx_queue tx_queue[RTE_PMD_PCAP_MAX_QUEUES];
  83         char devargs[ETH_PCAP_ARG_MAXLEN];
  84         struct rte_ether_addr eth_addr;
  85         int if_index;
  86         int single_iface;
  87         int phy_mac;
  88         unsigned int infinite_rx;
  89 };
  90
  91 struct pmd_process_private {
  92         pcap_t *rx_pcap[RTE_PMD_PCAP_MAX_QUEUES];
  93         pcap_t *tx_pcap[RTE_PMD_PCAP_MAX_QUEUES];
  94         pcap_dumper_t *tx_dumper[RTE_PMD_PCAP_MAX_QUEUES];
  95 };
  96
  97 struct pmd_devargs {
  98         unsigned int num_of_queue;
  99         struct devargs_queue {
 100                 pcap_dumper_t *dumper;
 101                 pcap_t *pcap;
 102                 const char *name;
 103                 const char *type;
 104         } queue[RTE_PMD_PCAP_MAX_QUEUES];
 105         int phy_mac;
 106 };
 107
 108 struct pmd_devargs_all {
 109         struct pmd_devargs rx_queues;
 110         struct pmd_devargs tx_queues;
 111         int single_iface;
 112         unsigned int is_tx_pcap;
 113         unsigned int is_tx_iface;
 114         unsigned int is_rx_pcap;
 115         unsigned int is_rx_iface;
 116         unsigned int infinite_rx;
 117 };
 118
 119 static const char *valid_arguments[] = {
 120         ETH_PCAP_RX_PCAP_ARG,
 121         ETH_PCAP_TX_PCAP_ARG,
 122         ETH_PCAP_RX_IFACE_ARG,
 123         ETH_PCAP_RX_IFACE_IN_ARG,
 124         ETH_PCAP_TX_IFACE_ARG,
 125         ETH_PCAP_IFACE_ARG,
 126         ETH_PCAP_PHY_MAC_ARG,
 127         ETH_PCAP_INFINITE_RX_ARG,
 128         NULL
 129 };
 130
 131 static struct rte_eth_link pmd_link = {
 132                 .link_speed = ETH_SPEED_NUM_10G,
 133                 .link_duplex = ETH_LINK_FULL_DUPLEX,
 134                 .link_status = ETH_LINK_DOWN,
 135                 .link_autoneg = ETH_LINK_FIXED,
 136 };
 137
 138 static int eth_pcap_logtype;
 139
 140 #define PMD_LOG(level, fmt, args...) \
 141         rte_log(RTE_LOG_ ## level, eth_pcap_logtype, \
 142                 "%s(): " fmt "\n", __func__, ##args)
 143
 144 static int
 145 eth_pcap_rx_jumbo(struct rte_mempool *mb_pool, struct rte_mbuf *mbuf,
 146                 const u_char *data, uint16_t data_len)
 147 {
 148         /* Copy the first segment. */
 149         uint16_t len = rte_pktmbuf_tailroom(mbuf);
 150         struct rte_mbuf *m = mbuf;
 151
 152         rte_memcpy(rte_pktmbuf_append(mbuf, len), data, len);
 153         data_len -= len;
 154         data += len;
 155
 156         while (data_len > 0) {
 157                 /* Allocate next mbuf and point to that. */
 158                 m->next = rte_pktmbuf_alloc(mb_pool);
 159
 160                 if (unlikely(!m->next))
 161                         return -1;
 162
 163                 m = m->next;
 164
 165                 /* Headroom is not needed in chained mbufs. */
 166                 rte_pktmbuf_prepend(m, rte_pktmbuf_headroom(m));
 167                 m->pkt_len = 0;
 168                 m->data_len = 0;
 169
 170                 /* Copy next segment. */
 171                 len = RTE_MIN(rte_pktmbuf_tailroom(m), data_len);
 172                 rte_memcpy(rte_pktmbuf_append(m, len), data, len);
 173
 174                 mbuf->nb_segs++;
 175                 data_len -= len;
 176                 data += len;
 177         }
 178
 179         return mbuf->nb_segs;
 180 }
 181
 182 static uint16_t
 183 eth_pcap_rx_infinite(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 184 {
 185         int i;
 186         struct pcap_rx_queue *pcap_q = queue;
 187         uint32_t rx_bytes = 0;
 188
 189         if (unlikely(nb_pkts == 0))
 190                 return 0;
 191
 192         if (rte_pktmbuf_alloc_bulk(pcap_q->mb_pool, bufs, nb_pkts) != 0)
 193                 return 0;
 194
 195         for (i = 0; i < nb_pkts; i++) {
 196                 struct rte_mbuf *pcap_buf;
 197                 int err = rte_ring_dequeue(pcap_q->pkts, (void **)&pcap_buf);
 198                 if (err)
 199                         return i;
 200
 201                 rte_memcpy(rte_pktmbuf_mtod(bufs[i], void *),
 202                                 rte_pktmbuf_mtod(pcap_buf, void *),
 203                                 pcap_buf->data_len);
 204                 bufs[i]->data_len = pcap_buf->data_len;
 205                 bufs[i]->pkt_len = pcap_buf->pkt_len;
 206                 bufs[i]->port = pcap_q->port_id;
 207                 rx_bytes += pcap_buf->data_len;
 208
 209                 /* Enqueue packet back on ring to allow infinite rx. */
 210                 rte_ring_enqueue(pcap_q->pkts, pcap_buf);
 211         }
 212
 213         pcap_q->rx_stat.pkts += i;
 214         pcap_q->rx_stat.bytes += rx_bytes;
 215
 216         return i;
 217 }
 218
 219 static uint16_t
 220 eth_pcap_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 221 {
 222         unsigned int i;
 223         struct pcap_pkthdr header;
 224         struct pmd_process_private *pp;
 225         const u_char *packet;
 226         struct rte_mbuf *mbuf;
 227         struct pcap_rx_queue *pcap_q = queue;
 228         uint16_t num_rx = 0;
 229         uint32_t rx_bytes = 0;
 230         pcap_t *pcap;
 231
 232         pp = rte_eth_devices[pcap_q->port_id].process_private;
 233         pcap = pp->rx_pcap[pcap_q->queue_id];
 234
 235         if (unlikely(pcap == NULL || nb_pkts == 0))
 236                 return 0;
 237
 238         /* Reads the given number of packets from the pcap file one by one
 239          * and copies the packet data into a newly allocated mbuf to return.
 240          */
 241         for (i = 0; i < nb_pkts; i++) {
 242                 /* Get the next PCAP packet */
 243                 packet = pcap_next(pcap, &header);
 244                 if (unlikely(packet == NULL))
 245                         break;
 246
 247                 mbuf = rte_pktmbuf_alloc(pcap_q->mb_pool);
 248                 if (unlikely(mbuf == NULL))
 249                         break;
 250
 251                 if (header.caplen <= rte_pktmbuf_tailroom(mbuf)) {
 252                         /* pcap packet will fit in the mbuf, can copy it */
 253                         rte_memcpy(rte_pktmbuf_mtod(mbuf, void *), packet,
 254                                         header.caplen);
 255                         mbuf->data_len = (uint16_t)header.caplen;
 256                 } else {
 257                         /* Try read jumbo frame into multi mbufs. */
 258                         if (unlikely(eth_pcap_rx_jumbo(pcap_q->mb_pool,
 259                                                        mbuf,
 260                                                        packet,
 261                                                        header.caplen) == -1)) {
 262                                 rte_pktmbuf_free(mbuf);
 263                                 break;
 264                         }
 265                 }
 266
 267                 mbuf->pkt_len = (uint16_t)header.caplen;
 268                 mbuf->timestamp = (uint64_t)header.ts.tv_sec * 1000000
 269                                                         + header.ts.tv_usec;
 270                 mbuf->ol_flags |= PKT_RX_TIMESTAMP;
 271                 mbuf->port = pcap_q->port_id;
 272                 bufs[num_rx] = mbuf;
 273                 num_rx++;
 274                 rx_bytes += header.caplen;
 275         }
 276         pcap_q->rx_stat.pkts += num_rx;
 277         pcap_q->rx_stat.bytes += rx_bytes;
 278
 279         return num_rx;
 280 }
 281
 282 static uint16_t
 283 eth_null_rx(void *queue __rte_unused,
 284                 struct rte_mbuf **bufs __rte_unused,
 285                 uint16_t nb_pkts __rte_unused)
 286 {
 287         return 0;
 288 }
 289
 290 static inline void
 291 calculate_timestamp(struct timeval *ts) {
 292         uint64_t cycles;
 293         struct timeval cur_time;
 294
 295         cycles = rte_get_timer_cycles() - start_cycles;
 296         cur_time.tv_sec = cycles / hz;
 297         cur_time.tv_usec = (cycles % hz) * 1e6 / hz;
 298         timeradd(&start_time, &cur_time, ts);
 299 }
 300
 301 /*
 302  * Callback to handle writing packets to a pcap file.
 303  */
 304 static uint16_t
 305 eth_pcap_tx_dumper(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 306 {
 307         unsigned int i;
 308         struct rte_mbuf *mbuf;
 309         struct pmd_process_private *pp;
 310         struct pcap_tx_queue *dumper_q = queue;
 311         uint16_t num_tx = 0;
 312         uint32_t tx_bytes = 0;
 313         struct pcap_pkthdr header;
 314         pcap_dumper_t *dumper;
 315         unsigned char temp_data[RTE_ETH_PCAP_SNAPLEN];
 316         size_t len;
 317
 318         pp = rte_eth_devices[dumper_q->port_id].process_private;
 319         dumper = pp->tx_dumper[dumper_q->queue_id];
 320
 321         if (dumper == NULL || nb_pkts == 0)
 322                 return 0;
 323
 324         /* writes the nb_pkts packets to the previously opened pcap file
 325          * dumper */
 326         for (i = 0; i < nb_pkts; i++) {
 327                 mbuf = bufs[i];
 328                 len = rte_pktmbuf_pkt_len(mbuf);
 329                 if (unlikely(!rte_pktmbuf_is_contiguous(mbuf) &&
 330                                 len > sizeof(temp_data))) {
 331                         PMD_LOG(ERR,
 332                                 "Dropping multi segment PCAP packet. Size (%zd) > max size (%zd).",
 333                                 len, sizeof(temp_data));
 334                         rte_pktmbuf_free(mbuf);
 335                         continue;
 336                 }
 337
 338                 calculate_timestamp(&header.ts);
 339                 header.len = len;
 340                 header.caplen = header.len;
 341                 /* rte_pktmbuf_read() returns a pointer to the data directly
 342                  * in the mbuf (when the mbuf is contiguous) or, otherwise,
 343                  * a pointer to temp_data after copying into it.
 344                  */
 345                 pcap_dump((u_char *)dumper, &header,
 346                         rte_pktmbuf_read(mbuf, 0, len, temp_data));
 347
 348                 num_tx++;
 349                 tx_bytes += len;
 350                 rte_pktmbuf_free(mbuf);
 351         }
 352
 353         /*
 354          * Since there's no place to hook a callback when the forwarding
 355          * process stops and to make sure the pcap file is actually written,
 356          * we flush the pcap dumper within each burst.
 357          */
 358         pcap_dump_flush(dumper);
 359         dumper_q->tx_stat.pkts += num_tx;
 360         dumper_q->tx_stat.bytes += tx_bytes;
 361         dumper_q->tx_stat.err_pkts += nb_pkts - num_tx;
 362
 363         return nb_pkts;
 364 }
 365
 366 /*
 367  * Callback to handle dropping packets in the infinite rx case.
 368  */
 369 static uint16_t
 370 eth_tx_drop(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 371 {
 372         unsigned int i;
 373         uint32_t tx_bytes = 0;
 374         struct pcap_tx_queue *tx_queue = queue;
 375
 376         if (unlikely(nb_pkts == 0))
 377                 return 0;
 378
 379         for (i = 0; i < nb_pkts; i++) {
 380                 tx_bytes += bufs[i]->data_len;
 381                 rte_pktmbuf_free(bufs[i]);
 382         }
 383
 384         tx_queue->tx_stat.pkts += nb_pkts;
 385         tx_queue->tx_stat.bytes += tx_bytes;
 386
 387         return i;
 388 }
 389
 390 /*
 391  * Callback to handle sending packets through a real NIC.
 392  */
 393 static uint16_t
 394 eth_pcap_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 395 {
 396         unsigned int i;
 397         int ret;
 398         struct rte_mbuf *mbuf;
 399         struct pmd_process_private *pp;
 400         struct pcap_tx_queue *tx_queue = queue;
 401         uint16_t num_tx = 0;
 402         uint32_t tx_bytes = 0;
 403         pcap_t *pcap;
 404         unsigned char temp_data[RTE_ETH_PCAP_SNAPLEN];
 405         size_t len;
 406
 407         pp = rte_eth_devices[tx_queue->port_id].process_private;
 408         pcap = pp->tx_pcap[tx_queue->queue_id];
 409
 410         if (unlikely(nb_pkts == 0 || pcap == NULL))
 411                 return 0;
 412
 413         for (i = 0; i < nb_pkts; i++) {
 414                 mbuf = bufs[i];
 415                 len = rte_pktmbuf_pkt_len(mbuf);
 416                 if (unlikely(!rte_pktmbuf_is_contiguous(mbuf) &&
 417                                 len > sizeof(temp_data))) {
 418                         PMD_LOG(ERR,
 419                                 "Dropping multi segment PCAP packet. Size (%zd) > max size (%zd).",
 420                                 len, sizeof(temp_data));
 421                         rte_pktmbuf_free(mbuf);
 422                         continue;
 423                 }
 424
 425                 /* rte_pktmbuf_read() returns a pointer to the data directly
 426                  * in the mbuf (when the mbuf is contiguous) or, otherwise,
 427                  * a pointer to temp_data after copying into it.
 428                  */
 429                 ret = pcap_sendpacket(pcap,
 430                         rte_pktmbuf_read(mbuf, 0, len, temp_data), len);
 431                 if (unlikely(ret != 0))
 432                         break;
 433                 num_tx++;
 434                 tx_bytes += len;
 435                 rte_pktmbuf_free(mbuf);
 436         }
 437
 438         tx_queue->tx_stat.pkts += num_tx;
 439         tx_queue->tx_stat.bytes += tx_bytes;
 440         tx_queue->tx_stat.err_pkts += i - num_tx;
 441
 442         return i;
 443 }
 444
 445 /*
 446  * pcap_open_live wrapper function
 447  */
 448 static inline int
 449 open_iface_live(const char *iface, pcap_t **pcap) {
 450         *pcap = pcap_open_live(iface, RTE_ETH_PCAP_SNAPLEN,
 451                         RTE_ETH_PCAP_PROMISC, RTE_ETH_PCAP_TIMEOUT, errbuf);
 452
 453         if (*pcap == NULL) {
 454                 PMD_LOG(ERR, "Couldn't open %s: %s", iface, errbuf);
 455                 return -1;
 456         }
 457
 458         return 0;
 459 }
 460
 461 static int
 462 open_single_iface(const char *iface, pcap_t **pcap)
 463 {
 464         if (open_iface_live(iface, pcap) < 0) {
 465                 PMD_LOG(ERR, "Couldn't open interface %s", iface);
 466                 return -1;
 467         }
 468
 469         return 0;
 470 }
 471
 472 static int
 473 open_single_tx_pcap(const char *pcap_filename, pcap_dumper_t **dumper)
 474 {
 475         pcap_t *tx_pcap;
 476
 477         /*
 478          * We need to create a dummy empty pcap_t to use it
 479          * with pcap_dump_open(). We create big enough an Ethernet
 480          * pcap holder.
 481          */
 482         tx_pcap = pcap_open_dead(DLT_EN10MB, RTE_ETH_PCAP_SNAPSHOT_LEN);
 483         if (tx_pcap == NULL) {
 484                 PMD_LOG(ERR, "Couldn't create dead pcap");
 485                 return -1;
 486         }
 487
 488         /* The dumper is created using the previous pcap_t reference */
 489         *dumper = pcap_dump_open(tx_pcap, pcap_filename);
 490         if (*dumper == NULL) {
 491                 pcap_close(tx_pcap);
 492                 PMD_LOG(ERR, "Couldn't open %s for writing.",
 493                         pcap_filename);
 494                 return -1;
 495         }
 496
 497         pcap_close(tx_pcap);
 498         return 0;
 499 }
 500
 501 static int
 502 open_single_rx_pcap(const char *pcap_filename, pcap_t **pcap)
 503 {
 504         *pcap = pcap_open_offline(pcap_filename, errbuf);
 505         if (*pcap == NULL) {
 506                 PMD_LOG(ERR, "Couldn't open %s: %s", pcap_filename,
 507                         errbuf);
 508                 return -1;
 509         }
 510
 511         return 0;
 512 }
 513
 514 static uint64_t
 515 count_packets_in_pcap(pcap_t **pcap, struct pcap_rx_queue *pcap_q)
 516 {
 517         const u_char *packet;
 518         struct pcap_pkthdr header;
 519         uint64_t pcap_pkt_count = 0;
 520
 521         while ((packet = pcap_next(*pcap, &header)))
 522                 pcap_pkt_count++;
 523
 524         /* The pcap is reopened so it can be used as normal later. */
 525         pcap_close(*pcap);
 526         *pcap = NULL;
 527         open_single_rx_pcap(pcap_q->name, pcap);
 528
 529         return pcap_pkt_count;
 530 }
 531
 532 static int
 533 eth_dev_start(struct rte_eth_dev *dev)
 534 {
 535         unsigned int i;
 536         struct pmd_internals *internals = dev->data->dev_private;
 537         struct pmd_process_private *pp = dev->process_private;
 538         struct pcap_tx_queue *tx;
 539         struct pcap_rx_queue *rx;
 540
 541         /* Special iface case. Single pcap is open and shared between tx/rx. */
 542         if (internals->single_iface) {
 543                 tx = &internals->tx_queue[0];
 544                 rx = &internals->rx_queue[0];
 545
 546                 if (!pp->tx_pcap[0] &&
 547                         strcmp(tx->type, ETH_PCAP_IFACE_ARG) == 0) {
 548                         if (open_single_iface(tx->name, &pp->tx_pcap[0]) < 0)
 549                                 return -1;
 550                         pp->rx_pcap[0] = pp->tx_pcap[0];
 551                 }
 552
 553                 goto status_up;
 554         }
 555
 556         /* If not open already, open tx pcaps/dumpers */
 557         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 558                 tx = &internals->tx_queue[i];
 559
 560                 if (!pp->tx_dumper[i] &&
 561                                 strcmp(tx->type, ETH_PCAP_TX_PCAP_ARG) == 0) {
 562                         if (open_single_tx_pcap(tx->name,
 563                                 &pp->tx_dumper[i]) < 0)
 564                                 return -1;
 565                 } else if (!pp->tx_pcap[i] &&
 566                                 strcmp(tx->type, ETH_PCAP_TX_IFACE_ARG) == 0) {
 567                         if (open_single_iface(tx->name, &pp->tx_pcap[i]) < 0)
 568                                 return -1;
 569                 }
 570         }
 571
 572         /* If not open already, open rx pcaps */
 573         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 574                 rx = &internals->rx_queue[i];
 575
 576                 if (pp->rx_pcap[i] != NULL)
 577                         continue;
 578
 579                 if (strcmp(rx->type, ETH_PCAP_RX_PCAP_ARG) == 0) {
 580                         if (open_single_rx_pcap(rx->name, &pp->rx_pcap[i]) < 0)
 581                                 return -1;
 582                 } else if (strcmp(rx->type, ETH_PCAP_RX_IFACE_ARG) == 0) {
 583                         if (open_single_iface(rx->name, &pp->rx_pcap[i]) < 0)
 584                                 return -1;
 585                 }
 586         }
 587
 588 status_up:
 589         for (i = 0; i < dev->data->nb_rx_queues; i++)
 590                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 591
 592         for (i = 0; i < dev->data->nb_tx_queues; i++)
 593                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 594
 595         dev->data->dev_link.link_status = ETH_LINK_UP;
 596
 597         return 0;
 598 }
 599
 600 /*
 601  * This function gets called when the current port gets stopped.
 602  * Is the only place for us to close all the tx streams dumpers.
 603  * If not called the dumpers will be flushed within each tx burst.
 604  */
 605 static void
 606 eth_dev_stop(struct rte_eth_dev *dev)
 607 {
 608         unsigned int i;
 609         struct pmd_internals *internals = dev->data->dev_private;
 610         struct pmd_process_private *pp = dev->process_private;
 611
 612         /* Special iface case. Single pcap is open and shared between tx/rx. */
 613         if (internals->single_iface) {
 614                 pcap_close(pp->tx_pcap[0]);
 615                 pp->tx_pcap[0] = NULL;
 616                 pp->rx_pcap[0] = NULL;
 617                 goto status_down;
 618         }
 619
 620         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 621                 if (pp->tx_dumper[i] != NULL) {
 622                         pcap_dump_close(pp->tx_dumper[i]);
 623                         pp->tx_dumper[i] = NULL;
 624                 }
 625
 626                 if (pp->tx_pcap[i] != NULL) {
 627                         pcap_close(pp->tx_pcap[i]);
 628                         pp->tx_pcap[i] = NULL;
 629                 }
 630         }
 631
 632         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 633                 if (pp->rx_pcap[i] != NULL) {
 634                         pcap_close(pp->rx_pcap[i]);
 635                         pp->rx_pcap[i] = NULL;
 636                 }
 637         }
 638
 639 status_down:
 640         for (i = 0; i < dev->data->nb_rx_queues; i++)
 641                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 642
 643         for (i = 0; i < dev->data->nb_tx_queues; i++)
 644                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 645
 646         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 647 }
 648
 649 static int
 650 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 651 {
 652         return 0;
 653 }
 654
 655 static int
 656 eth_dev_info(struct rte_eth_dev *dev,
 657                 struct rte_eth_dev_info *dev_info)
 658 {
 659         struct pmd_internals *internals = dev->data->dev_private;
 660
 661         dev_info->if_index = internals->if_index;
 662         dev_info->max_mac_addrs = 1;
 663         dev_info->max_rx_pktlen = (uint32_t) -1;
 664         dev_info->max_rx_queues = dev->data->nb_rx_queues;
 665         dev_info->max_tx_queues = dev->data->nb_tx_queues;
 666         dev_info->min_rx_bufsize = 0;
 667
 668         return 0;
 669 }
 670
 671 static int
 672 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 673 {
 674         unsigned int i;
 675         unsigned long rx_packets_total = 0, rx_bytes_total = 0;
 676         unsigned long tx_packets_total = 0, tx_bytes_total = 0;
 677         unsigned long tx_packets_err_total = 0;
 678         const struct pmd_internals *internal = dev->data->dev_private;
 679
 680         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 681                         i < dev->data->nb_rx_queues; i++) {
 682                 stats->q_ipackets[i] = internal->rx_queue[i].rx_stat.pkts;
 683                 stats->q_ibytes[i] = internal->rx_queue[i].rx_stat.bytes;
 684                 rx_packets_total += stats->q_ipackets[i];
 685                 rx_bytes_total += stats->q_ibytes[i];
 686         }
 687
 688         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 689                         i < dev->data->nb_tx_queues; i++) {
 690                 stats->q_opackets[i] = internal->tx_queue[i].tx_stat.pkts;
 691                 stats->q_obytes[i] = internal->tx_queue[i].tx_stat.bytes;
 692                 tx_packets_total += stats->q_opackets[i];
 693                 tx_bytes_total += stats->q_obytes[i];
 694                 tx_packets_err_total += internal->tx_queue[i].tx_stat.err_pkts;
 695         }
 696
 697         stats->ipackets = rx_packets_total;
 698         stats->ibytes = rx_bytes_total;
 699         stats->opackets = tx_packets_total;
 700         stats->obytes = tx_bytes_total;
 701         stats->oerrors = tx_packets_err_total;
 702
 703         return 0;
 704 }
 705
 706 static void
 707 eth_stats_reset(struct rte_eth_dev *dev)
 708 {
 709         unsigned int i;
 710         struct pmd_internals *internal = dev->data->dev_private;
 711
 712         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 713                 internal->rx_queue[i].rx_stat.pkts = 0;
 714                 internal->rx_queue[i].rx_stat.bytes = 0;
 715         }
 716
 717         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 718                 internal->tx_queue[i].tx_stat.pkts = 0;
 719                 internal->tx_queue[i].tx_stat.bytes = 0;
 720                 internal->tx_queue[i].tx_stat.err_pkts = 0;
 721         }
 722 }
 723
 724 static void
 725 eth_dev_close(struct rte_eth_dev *dev)
 726 {
 727         unsigned int i;
 728         struct pmd_internals *internals = dev->data->dev_private;
 729
 730         /* Device wide flag, but cleanup must be performed per queue. */
 731         if (internals->infinite_rx) {
 732                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
 733                         struct pcap_rx_queue *pcap_q = &internals->rx_queue[i];
 734                         struct rte_mbuf *pcap_buf;
 735
 736                         while (!rte_ring_dequeue(pcap_q->pkts,
 737                                         (void **)&pcap_buf))
 738                                 rte_pktmbuf_free(pcap_buf);
 739
 740                         rte_ring_free(pcap_q->pkts);
 741                 }
 742         }
 743
 744 }
 745
 746 static void
 747 eth_queue_release(void *q __rte_unused)
 748 {
 749 }
 750
 751 static int
 752 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 753                 int wait_to_complete __rte_unused)
 754 {
 755         return 0;
 756 }
 757
 758 static int
 759 eth_rx_queue_setup(struct rte_eth_dev *dev,
 760                 uint16_t rx_queue_id,
 761                 uint16_t nb_rx_desc __rte_unused,
 762                 unsigned int socket_id __rte_unused,
 763                 const struct rte_eth_rxconf *rx_conf __rte_unused,
 764                 struct rte_mempool *mb_pool)
 765 {
 766         struct pmd_internals *internals = dev->data->dev_private;
 767         struct pcap_rx_queue *pcap_q = &internals->rx_queue[rx_queue_id];
 768
 769         pcap_q->mb_pool = mb_pool;
 770         pcap_q->port_id = dev->data->port_id;
 771         pcap_q->queue_id = rx_queue_id;
 772         dev->data->rx_queues[rx_queue_id] = pcap_q;
 773
 774         if (internals->infinite_rx) {
 775                 struct pmd_process_private *pp;
 776                 char ring_name[NAME_MAX];
 777                 static uint32_t ring_number;
 778                 uint64_t pcap_pkt_count = 0;
 779                 struct rte_mbuf *bufs[1];
 780                 pcap_t **pcap;
 781
 782                 pp = rte_eth_devices[pcap_q->port_id].process_private;
 783                 pcap = &pp->rx_pcap[pcap_q->queue_id];
 784
 785                 if (unlikely(*pcap == NULL))
 786                         return -ENOENT;
 787
 788                 pcap_pkt_count = count_packets_in_pcap(pcap, pcap_q);
 789
 790                 snprintf(ring_name, sizeof(ring_name), "PCAP_RING%" PRIu16,
 791                                 ring_number);
 792
 793                 pcap_q->pkts = rte_ring_create(ring_name,
 794                                 rte_align64pow2(pcap_pkt_count + 1), 0,
 795                                 RING_F_SP_ENQ | RING_F_SC_DEQ);
 796                 ring_number++;
 797                 if (!pcap_q->pkts)
 798                         return -ENOENT;
 799
 800                 /* Fill ring with packets from PCAP file one by one. */
 801                 while (eth_pcap_rx(pcap_q, bufs, 1)) {
 802                         /* Check for multiseg mbufs. */
 803                         if (bufs[0]->nb_segs != 1) {
 804                                 rte_pktmbuf_free(*bufs);
 805
 806                                 while (!rte_ring_dequeue(pcap_q->pkts,
 807                                                 (void **)bufs))
 808                                         rte_pktmbuf_free(*bufs);
 809
 810                                 rte_ring_free(pcap_q->pkts);
 811                                 PMD_LOG(ERR, "Multiseg mbufs are not supported in infinite_rx "
 812                                                 "mode.");
 813                                 return -EINVAL;
 814                         }
 815
 816                         rte_ring_enqueue_bulk(pcap_q->pkts,
 817                                         (void * const *)bufs, 1, NULL);
 818                 }
 819                 /*
 820                  * Reset the stats for this queue since eth_pcap_rx calls above
 821                  * didn't result in the application receiving packets.
 822                  */
 823                 pcap_q->rx_stat.pkts = 0;
 824                 pcap_q->rx_stat.bytes = 0;
 825         }
 826
 827         return 0;
 828 }
 829
 830 static int
 831 eth_tx_queue_setup(struct rte_eth_dev *dev,
 832                 uint16_t tx_queue_id,
 833                 uint16_t nb_tx_desc __rte_unused,
 834                 unsigned int socket_id __rte_unused,
 835                 const struct rte_eth_txconf *tx_conf __rte_unused)
 836 {
 837         struct pmd_internals *internals = dev->data->dev_private;
 838         struct pcap_tx_queue *pcap_q = &internals->tx_queue[tx_queue_id];
 839
 840         pcap_q->port_id = dev->data->port_id;
 841         pcap_q->queue_id = tx_queue_id;
 842         dev->data->tx_queues[tx_queue_id] = pcap_q;
 843
 844         return 0;
 845 }
 846
 847 static int
 848 eth_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 849 {
 850         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
 851
 852         return 0;
 853 }
 854
 855 static int
 856 eth_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
 857 {
 858         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
 859
 860         return 0;
 861 }
 862
 863 static int
 864 eth_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 865 {
 866         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
 867
 868         return 0;
 869 }
 870
 871 static int
 872 eth_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
 873 {
 874         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
 875
 876         return 0;
 877 }
 878
 879 static const struct eth_dev_ops ops = {
 880         .dev_start = eth_dev_start,
 881         .dev_stop = eth_dev_stop,
 882         .dev_close = eth_dev_close,
 883         .dev_configure = eth_dev_configure,
 884         .dev_infos_get = eth_dev_info,
 885         .rx_queue_setup = eth_rx_queue_setup,
 886         .tx_queue_setup = eth_tx_queue_setup,
 887         .rx_queue_start = eth_rx_queue_start,
 888         .tx_queue_start = eth_tx_queue_start,
 889         .rx_queue_stop = eth_rx_queue_stop,
 890         .tx_queue_stop = eth_tx_queue_stop,
 891         .rx_queue_release = eth_queue_release,
 892         .tx_queue_release = eth_queue_release,
 893         .link_update = eth_link_update,
 894         .stats_get = eth_stats_get,
 895         .stats_reset = eth_stats_reset,
 896 };
 897
 898 static int
 899 add_queue(struct pmd_devargs *pmd, const char *name, const char *type,
 900                 pcap_t *pcap, pcap_dumper_t *dumper)
 901 {
 902         if (pmd->num_of_queue >= RTE_PMD_PCAP_MAX_QUEUES)
 903                 return -1;
 904         if (pcap)
 905                 pmd->queue[pmd->num_of_queue].pcap = pcap;
 906         if (dumper)
 907                 pmd->queue[pmd->num_of_queue].dumper = dumper;
 908         pmd->queue[pmd->num_of_queue].name = name;
 909         pmd->queue[pmd->num_of_queue].type = type;
 910         pmd->num_of_queue++;
 911         return 0;
 912 }
 913
 914 /*
 915  * Function handler that opens the pcap file for reading a stores a
 916  * reference of it for use it later on.
 917  */
 918 static int
 919 open_rx_pcap(const char *key, const char *value, void *extra_args)
 920 {
 921         const char *pcap_filename = value;
 922         struct pmd_devargs *rx = extra_args;
 923         pcap_t *pcap = NULL;
 924
 925         if (open_single_rx_pcap(pcap_filename, &pcap) < 0)
 926                 return -1;
 927
 928         if (add_queue(rx, pcap_filename, key, pcap, NULL) < 0) {
 929                 pcap_close(pcap);
 930                 return -1;
 931         }
 932
 933         return 0;
 934 }
 935
 936 /*
 937  * Opens a pcap file for writing and stores a reference to it
 938  * for use it later on.
 939  */
 940 static int
 941 open_tx_pcap(const char *key, const char *value, void *extra_args)
 942 {
 943         const char *pcap_filename = value;
 944         struct pmd_devargs *dumpers = extra_args;
 945         pcap_dumper_t *dumper;
 946
 947         if (open_single_tx_pcap(pcap_filename, &dumper) < 0)
 948                 return -1;
 949
 950         if (add_queue(dumpers, pcap_filename, key, NULL, dumper) < 0) {
 951                 pcap_dump_close(dumper);
 952                 return -1;
 953         }
 954
 955         return 0;
 956 }
 957
 958 /*
 959  * Opens an interface for reading and writing
 960  */
 961 static inline int
 962 open_rx_tx_iface(const char *key, const char *value, void *extra_args)
 963 {
 964         const char *iface = value;
 965         struct pmd_devargs *tx = extra_args;
 966         pcap_t *pcap = NULL;
 967
 968         if (open_single_iface(iface, &pcap) < 0)
 969                 return -1;
 970
 971         tx->queue[0].pcap = pcap;
 972         tx->queue[0].name = iface;
 973         tx->queue[0].type = key;
 974
 975         return 0;
 976 }
 977
 978 static inline int
 979 set_iface_direction(const char *iface, pcap_t *pcap,
 980                 pcap_direction_t direction)
 981 {
 982         const char *direction_str = (direction == PCAP_D_IN) ? "IN" : "OUT";
 983         if (pcap_setdirection(pcap, direction) < 0) {
 984                 PMD_LOG(ERR, "Setting %s pcap direction %s failed - %s\n",
 985                                 iface, direction_str, pcap_geterr(pcap));
 986                 return -1;
 987         }
 988         PMD_LOG(INFO, "Setting %s pcap direction %s\n",
 989                         iface, direction_str);
 990         return 0;
 991 }
 992
 993 static inline int
 994 open_iface(const char *key, const char *value, void *extra_args)
 995 {
 996         const char *iface = value;
 997         struct pmd_devargs *pmd = extra_args;
 998         pcap_t *pcap = NULL;
 999
1000         if (open_single_iface(iface, &pcap) < 0)
1001                 return -1;
1002         if (add_queue(pmd, iface, key, pcap, NULL) < 0) {
1003                 pcap_close(pcap);
1004                 return -1;
1005         }
1006
1007         return 0;
1008 }
1009
1010 /*
1011  * Opens a NIC for reading packets from it
1012  */
1013 static inline int
1014 open_rx_iface(const char *key, const char *value, void *extra_args)
1015 {
1016         int ret = open_iface(key, value, extra_args);
1017         if (ret < 0)
1018                 return ret;
1019         if (strcmp(key, ETH_PCAP_RX_IFACE_IN_ARG) == 0) {
1020                 struct pmd_devargs *pmd = extra_args;
1021                 unsigned int qid = pmd->num_of_queue - 1;
1022
1023                 set_iface_direction(pmd->queue[qid].name,
1024                                 pmd->queue[qid].pcap,
1025                                 PCAP_D_IN);
1026         }
1027
1028         return 0;
1029 }
1030
1031 static inline int
1032 rx_iface_args_process(const char *key, const char *value, void *extra_args)
1033 {
1034         if (strcmp(key, ETH_PCAP_RX_IFACE_ARG) == 0 ||
1035                         strcmp(key, ETH_PCAP_RX_IFACE_IN_ARG) == 0)
1036                 return open_rx_iface(key, value, extra_args);
1037
1038         return 0;
1039 }
1040
1041 /*
1042  * Opens a NIC for writing packets to it
1043  */
1044 static int
1045 open_tx_iface(const char *key, const char *value, void *extra_args)
1046 {
1047         return open_iface(key, value, extra_args);
1048 }
1049
1050 static int
1051 select_phy_mac(const char *key __rte_unused, const char *value,
1052                 void *extra_args)
1053 {
1054         if (extra_args) {
1055                 const int phy_mac = atoi(value);
1056                 int *enable_phy_mac = extra_args;
1057
1058                 if (phy_mac)
1059                         *enable_phy_mac = 1;
1060         }
1061         return 0;
1062 }
1063
1064 static int
1065 get_infinite_rx_arg(const char *key __rte_unused,
1066                 const char *value, void *extra_args)
1067 {
1068         if (extra_args) {
1069                 const int infinite_rx = atoi(value);
1070                 int *enable_infinite_rx = extra_args;
1071
1072                 if (infinite_rx > 0)
1073                         *enable_infinite_rx = 1;
1074         }
1075         return 0;
1076 }
1077
1078 static int
1079 pmd_init_internals(struct rte_vdev_device *vdev,
1080                 const unsigned int nb_rx_queues,
1081                 const unsigned int nb_tx_queues,
1082                 struct pmd_internals **internals,
1083                 struct rte_eth_dev **eth_dev)
1084 {
1085         struct rte_eth_dev_data *data;
1086         struct pmd_process_private *pp;
1087         unsigned int numa_node = vdev->device.numa_node;
1088
1089         PMD_LOG(INFO, "Creating pcap-backed ethdev on numa socket %d",
1090                 numa_node);
1091
1092         pp = (struct pmd_process_private *)
1093                 rte_zmalloc(NULL, sizeof(struct pmd_process_private),
1094                                 RTE_CACHE_LINE_SIZE);
1095
1096         if (pp == NULL) {
1097                 PMD_LOG(ERR,
1098                         "Failed to allocate memory for process private");
1099                 return -1;
1100         }
1101
1102         /* reserve an ethdev entry */
1103         *eth_dev = rte_eth_vdev_allocate(vdev, sizeof(**internals));
1104         if (!(*eth_dev)) {
1105                 rte_free(pp);
1106                 return -1;
1107         }
1108         (*eth_dev)->process_private = pp;
1109         /* now put it all together
1110          * - store queue data in internals,
1111          * - store numa_node info in eth_dev
1112          * - point eth_dev_data to internals
1113          * - and point eth_dev structure to new eth_dev_data structure
1114          */
1115         *internals = (*eth_dev)->data->dev_private;
1116         /*
1117          * Interface MAC = 02:70:63:61:70:<iface_idx>
1118          * derived from: 'locally administered':'p':'c':'a':'p':'iface_idx'
1119          * where the middle 4 characters are converted to hex.
1120          */
1121         (*internals)->eth_addr = (struct rte_ether_addr) {
1122                 .addr_bytes = { 0x02, 0x70, 0x63, 0x61, 0x70, iface_idx++ }
1123         };
1124         (*internals)->phy_mac = 0;
1125         data = (*eth_dev)->data;
1126         data->nb_rx_queues = (uint16_t)nb_rx_queues;
1127         data->nb_tx_queues = (uint16_t)nb_tx_queues;
1128         data->dev_link = pmd_link;
1129         data->mac_addrs = &(*internals)->eth_addr;
1130
1131         /*
1132          * NOTE: we'll replace the data element, of originally allocated
1133          * eth_dev so the rings are local per-process
1134          */
1135         (*eth_dev)->dev_ops = &ops;
1136
1137         strlcpy((*internals)->devargs, rte_vdev_device_args(vdev),
1138                         ETH_PCAP_ARG_MAXLEN);
1139
1140         return 0;
1141 }
1142
1143 static int
1144 eth_pcap_update_mac(const char *if_name, struct rte_eth_dev *eth_dev,
1145                 const unsigned int numa_node)
1146 {
1147 #if defined(RTE_EXEC_ENV_LINUX)
1148         void *mac_addrs;
1149         struct ifreq ifr;
1150         int if_fd = socket(AF_INET, SOCK_DGRAM, 0);
1151
1152         if (if_fd == -1)
1153                 return -1;
1154
1155         rte_strscpy(ifr.ifr_name, if_name, sizeof(ifr.ifr_name));
1156         if (ioctl(if_fd, SIOCGIFHWADDR, &ifr)) {
1157                 close(if_fd);
1158                 return -1;
1159         }
1160
1161         mac_addrs = rte_zmalloc_socket(NULL, RTE_ETHER_ADDR_LEN, 0, numa_node);
1162         if (!mac_addrs) {
1163                 close(if_fd);
1164                 return -1;
1165         }
1166
1167         PMD_LOG(INFO, "Setting phy MAC for %s", if_name);
1168         eth_dev->data->mac_addrs = mac_addrs;
1169         rte_memcpy(eth_dev->data->mac_addrs[0].addr_bytes,
1170                         ifr.ifr_hwaddr.sa_data, RTE_ETHER_ADDR_LEN);
1171
1172         close(if_fd);
1173
1174         return 0;
1175
1176 #elif defined(RTE_EXEC_ENV_FREEBSD)
1177         void *mac_addrs;
1178         struct if_msghdr *ifm;
1179         struct sockaddr_dl *sdl;
1180         int mib[6];
1181         size_t len = 0;
1182         char *buf;
1183
1184         mib[0] = CTL_NET;
1185         mib[1] = AF_ROUTE;
1186         mib[2] = 0;
1187         mib[3] = AF_LINK;
1188         mib[4] = NET_RT_IFLIST;
1189         mib[5] = if_nametoindex(if_name);
1190
1191         if (sysctl(mib, 6, NULL, &len, NULL, 0) < 0)
1192                 return -1;
1193
1194         if (len == 0)
1195                 return -1;
1196
1197         buf = rte_malloc(NULL, len, 0);
1198         if (!buf)
1199                 return -1;
1200
1201         if (sysctl(mib, 6, buf, &len, NULL, 0) < 0) {
1202                 rte_free(buf);
1203                 return -1;
1204         }
1205         ifm = (struct if_msghdr *)buf;
1206         sdl = (struct sockaddr_dl *)(ifm + 1);
1207
1208         mac_addrs = rte_zmalloc_socket(NULL, RTE_ETHER_ADDR_LEN, 0, numa_node);
1209         if (!mac_addrs) {
1210                 rte_free(buf);
1211                 return -1;
1212         }
1213
1214         PMD_LOG(INFO, "Setting phy MAC for %s", if_name);
1215         eth_dev->data->mac_addrs = mac_addrs;
1216         rte_memcpy(eth_dev->data->mac_addrs[0].addr_bytes,
1217                         LLADDR(sdl), RTE_ETHER_ADDR_LEN);
1218
1219         rte_free(buf);
1220
1221         return 0;
1222 #else
1223         return -1;
1224 #endif
1225 }
1226
1227 static int
1228 eth_from_pcaps_common(struct rte_vdev_device *vdev,
1229                 struct pmd_devargs_all *devargs_all,
1230                 struct pmd_internals **internals, struct rte_eth_dev **eth_dev)
1231 {
1232         struct pmd_process_private *pp;
1233         struct pmd_devargs *rx_queues = &devargs_all->rx_queues;
1234         struct pmd_devargs *tx_queues = &devargs_all->tx_queues;
1235         const unsigned int nb_rx_queues = rx_queues->num_of_queue;
1236         const unsigned int nb_tx_queues = tx_queues->num_of_queue;
1237         unsigned int i;
1238
1239         /* do some parameter checking */
1240         if (rx_queues == NULL && nb_rx_queues > 0)
1241                 return -1;
1242         if (tx_queues == NULL && nb_tx_queues > 0)
1243                 return -1;
1244
1245         if (pmd_init_internals(vdev, nb_rx_queues, nb_tx_queues, internals,
1246                         eth_dev) < 0)
1247                 return -1;
1248
1249         pp = (*eth_dev)->process_private;
1250         for (i = 0; i < nb_rx_queues; i++) {
1251                 struct pcap_rx_queue *rx = &(*internals)->rx_queue[i];
1252                 struct devargs_queue *queue = &rx_queues->queue[i];
1253
1254                 pp->rx_pcap[i] = queue->pcap;
1255                 strlcpy(rx->name, queue->name, sizeof(rx->name));
1256                 strlcpy(rx->type, queue->type, sizeof(rx->type));
1257         }
1258
1259         for (i = 0; i < nb_tx_queues; i++) {
1260                 struct pcap_tx_queue *tx = &(*internals)->tx_queue[i];
1261                 struct devargs_queue *queue = &tx_queues->queue[i];
1262
1263                 pp->tx_dumper[i] = queue->dumper;
1264                 pp->tx_pcap[i] = queue->pcap;
1265                 strlcpy(tx->name, queue->name, sizeof(tx->name));
1266                 strlcpy(tx->type, queue->type, sizeof(tx->type));
1267         }
1268
1269         return 0;
1270 }
1271
1272 static int
1273 eth_from_pcaps(struct rte_vdev_device *vdev,
1274                 struct pmd_devargs_all *devargs_all)
1275 {
1276         struct pmd_internals *internals = NULL;
1277         struct rte_eth_dev *eth_dev = NULL;
1278         struct pmd_devargs *rx_queues = &devargs_all->rx_queues;
1279         int single_iface = devargs_all->single_iface;
1280         unsigned int infinite_rx = devargs_all->infinite_rx;
1281         int ret;
1282
1283         ret = eth_from_pcaps_common(vdev, devargs_all, &internals, &eth_dev);
1284
1285         if (ret < 0)
1286                 return ret;
1287
1288         /* store weather we are using a single interface for rx/tx or not */
1289         internals->single_iface = single_iface;
1290
1291         if (single_iface) {
1292                 internals->if_index = if_nametoindex(rx_queues->queue[0].name);
1293
1294                 /* phy_mac arg is applied only only if "iface" devarg is provided */
1295                 if (rx_queues->phy_mac) {
1296                         int ret = eth_pcap_update_mac(rx_queues->queue[0].name,
1297                                         eth_dev, vdev->device.numa_node);
1298                         if (ret == 0)
1299                                 internals->phy_mac = 1;
1300                 }
1301         }
1302
1303         internals->infinite_rx = infinite_rx;
1304         /* Assign rx ops. */
1305         if (infinite_rx)
1306                 eth_dev->rx_pkt_burst = eth_pcap_rx_infinite;
1307         else if (devargs_all->is_rx_pcap || devargs_all->is_rx_iface ||
1308                         single_iface)
1309                 eth_dev->rx_pkt_burst = eth_pcap_rx;
1310         else
1311                 eth_dev->rx_pkt_burst = eth_null_rx;
1312
1313         /* Assign tx ops. */
1314         if (devargs_all->is_tx_pcap)
1315                 eth_dev->tx_pkt_burst = eth_pcap_tx_dumper;
1316         else if (devargs_all->is_tx_iface || single_iface)
1317                 eth_dev->tx_pkt_burst = eth_pcap_tx;
1318         else
1319                 eth_dev->tx_pkt_burst = eth_tx_drop;
1320
1321         rte_eth_dev_probing_finish(eth_dev);
1322         return 0;
1323 }
1324
1325 static int
1326 pmd_pcap_probe(struct rte_vdev_device *dev)
1327 {
1328         const char *name;
1329         struct rte_kvargs *kvlist;
1330         struct pmd_devargs pcaps = {0};
1331         struct pmd_devargs dumpers = {0};
1332         struct rte_eth_dev *eth_dev =  NULL;
1333         struct pmd_internals *internal;
1334         int ret = 0;
1335
1336         struct pmd_devargs_all devargs_all = {
1337                 .single_iface = 0,
1338                 .is_tx_pcap = 0,
1339                 .is_tx_iface = 0,
1340                 .infinite_rx = 0,
1341         };
1342
1343         name = rte_vdev_device_name(dev);
1344         PMD_LOG(INFO, "Initializing pmd_pcap for %s", name);
1345
1346         gettimeofday(&start_time, NULL);
1347         start_cycles = rte_get_timer_cycles();
1348         hz = rte_get_timer_hz();
1349
1350         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1351                 eth_dev = rte_eth_dev_attach_secondary(name);
1352                 if (!eth_dev) {
1353                         PMD_LOG(ERR, "Failed to probe %s", name);
1354                         return -1;
1355                 }
1356
1357                 internal = eth_dev->data->dev_private;
1358
1359                 kvlist = rte_kvargs_parse(internal->devargs, valid_arguments);
1360                 if (kvlist == NULL)
1361                         return -1;
1362         } else {
1363                 kvlist = rte_kvargs_parse(rte_vdev_device_args(dev),
1364                                 valid_arguments);
1365                 if (kvlist == NULL)
1366                         return -1;
1367         }
1368
1369         /*
1370          * If iface argument is passed we open the NICs and use them for
1371          * reading / writing
1372          */
1373         if (rte_kvargs_count(kvlist, ETH_PCAP_IFACE_ARG) == 1) {
1374
1375                 ret = rte_kvargs_process(kvlist, ETH_PCAP_IFACE_ARG,
1376                                 &open_rx_tx_iface, &pcaps);
1377                 if (ret < 0)
1378                         goto free_kvlist;
1379
1380                 dumpers.queue[0] = pcaps.queue[0];
1381
1382                 ret = rte_kvargs_process(kvlist, ETH_PCAP_PHY_MAC_ARG,
1383                                 &select_phy_mac, &pcaps.phy_mac);
1384                 if (ret < 0)
1385                         goto free_kvlist;
1386
1387                 dumpers.phy_mac = pcaps.phy_mac;
1388
1389                 devargs_all.single_iface = 1;
1390                 pcaps.num_of_queue = 1;
1391                 dumpers.num_of_queue = 1;
1392
1393                 goto create_eth;
1394         }
1395
1396         /*
1397          * We check whether we want to open a RX stream from a real NIC, a
1398          * pcap file or open a dummy RX stream
1399          */
1400         devargs_all.is_rx_pcap =
1401                 rte_kvargs_count(kvlist, ETH_PCAP_RX_PCAP_ARG) ? 1 : 0;
1402         devargs_all.is_rx_iface =
1403                 rte_kvargs_count(kvlist, ETH_PCAP_RX_IFACE_ARG) ? 1 : 0;
1404         pcaps.num_of_queue = 0;
1405
1406         devargs_all.is_tx_pcap =
1407                 rte_kvargs_count(kvlist, ETH_PCAP_TX_PCAP_ARG) ? 1 : 0;
1408         devargs_all.is_tx_iface =
1409                 rte_kvargs_count(kvlist, ETH_PCAP_TX_IFACE_ARG) ? 1 : 0;
1410         dumpers.num_of_queue = 0;
1411
1412         if (devargs_all.is_rx_pcap) {
1413                 /*
1414                  * We check whether we want to infinitely rx the pcap file.
1415                  */
1416                 unsigned int infinite_rx_arg_cnt = rte_kvargs_count(kvlist,
1417                                 ETH_PCAP_INFINITE_RX_ARG);
1418
1419                 if (infinite_rx_arg_cnt == 1) {
1420                         ret = rte_kvargs_process(kvlist,
1421                                         ETH_PCAP_INFINITE_RX_ARG,
1422                                         &get_infinite_rx_arg,
1423                                         &devargs_all.infinite_rx);
1424                         if (ret < 0)
1425                                 goto free_kvlist;
1426                         PMD_LOG(INFO, "infinite_rx has been %s for %s",
1427                                         devargs_all.infinite_rx ? "enabled" : "disabled",
1428                                         name);
1429
1430                 } else if (infinite_rx_arg_cnt > 1) {
1431                         PMD_LOG(WARNING, "infinite_rx has not been enabled since the "
1432                                         "argument has been provided more than once "
1433                                         "for %s", name);
1434                 }
1435
1436                 ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_PCAP_ARG,
1437                                 &open_rx_pcap, &pcaps);
1438         } else if (devargs_all.is_rx_iface) {
1439                 ret = rte_kvargs_process(kvlist, NULL,
1440                                 &rx_iface_args_process, &pcaps);
1441         } else if (devargs_all.is_tx_iface || devargs_all.is_tx_pcap) {
1442                 unsigned int i;
1443
1444                 /* Count number of tx queue args passed before dummy rx queue
1445                  * creation so a dummy rx queue can be created for each tx queue
1446                  */
1447                 unsigned int num_tx_queues =
1448                         (rte_kvargs_count(kvlist, ETH_PCAP_TX_PCAP_ARG) +
1449                         rte_kvargs_count(kvlist, ETH_PCAP_TX_IFACE_ARG));
1450
1451                 PMD_LOG(INFO, "Creating null rx queue since no rx queues were provided.");
1452
1453                 /* Creating a dummy rx queue for each tx queue passed */
1454                 for (i = 0; i < num_tx_queues; i++)
1455                         ret = add_queue(&pcaps, "dummy_rx", "rx_null", NULL,
1456                                         NULL);
1457         } else {
1458                 PMD_LOG(ERR, "Error - No rx or tx queues provided");
1459                 ret = -ENOENT;
1460         }
1461         if (ret < 0)
1462                 goto free_kvlist;
1463
1464         /*
1465          * We check whether we want to open a TX stream to a real NIC,
1466          * a pcap file, or drop packets on tx
1467          */
1468         if (devargs_all.is_tx_pcap) {
1469                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_PCAP_ARG,
1470                                 &open_tx_pcap, &dumpers);
1471         } else if (devargs_all.is_tx_iface) {
1472                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_IFACE_ARG,
1473                                 &open_tx_iface, &dumpers);
1474         } else {
1475                 unsigned int i;
1476
1477                 PMD_LOG(INFO, "Dropping packets on tx since no tx queues were provided.");
1478
1479                 /* Add 1 dummy queue per rxq which counts and drops packets. */
1480                 for (i = 0; i < pcaps.num_of_queue; i++)
1481                         ret = add_queue(&dumpers, "dummy_tx", "tx_drop", NULL,
1482                                         NULL);
1483         }
1484
1485         if (ret < 0)
1486                 goto free_kvlist;
1487
1488 create_eth:
1489         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1490                 struct pmd_process_private *pp;
1491                 unsigned int i;
1492
1493                 internal = eth_dev->data->dev_private;
1494                         pp = (struct pmd_process_private *)
1495                                 rte_zmalloc(NULL,
1496                                         sizeof(struct pmd_process_private),
1497                                         RTE_CACHE_LINE_SIZE);
1498
1499                 if (pp == NULL) {
1500                         PMD_LOG(ERR,
1501                                 "Failed to allocate memory for process private");
1502                         ret = -1;
1503                         goto free_kvlist;
1504                 }
1505
1506                 eth_dev->dev_ops = &ops;
1507                 eth_dev->device = &dev->device;
1508
1509                 /* setup process private */
1510                 for (i = 0; i < pcaps.num_of_queue; i++)
1511                         pp->rx_pcap[i] = pcaps.queue[i].pcap;
1512
1513                 for (i = 0; i < dumpers.num_of_queue; i++) {
1514                         pp->tx_dumper[i] = dumpers.queue[i].dumper;
1515                         pp->tx_pcap[i] = dumpers.queue[i].pcap;
1516                 }
1517
1518                 eth_dev->process_private = pp;
1519                 eth_dev->rx_pkt_burst = eth_pcap_rx;
1520                 if (devargs_all.is_tx_pcap)
1521                         eth_dev->tx_pkt_burst = eth_pcap_tx_dumper;
1522                 else
1523                         eth_dev->tx_pkt_burst = eth_pcap_tx;
1524
1525                 rte_eth_dev_probing_finish(eth_dev);
1526                 goto free_kvlist;
1527         }
1528
1529         devargs_all.rx_queues = pcaps;
1530         devargs_all.tx_queues = dumpers;
1531
1532         ret = eth_from_pcaps(dev, &devargs_all);
1533
1534 free_kvlist:
1535         rte_kvargs_free(kvlist);
1536
1537         return ret;
1538 }
1539
1540 static int
1541 pmd_pcap_remove(struct rte_vdev_device *dev)
1542 {
1543         struct pmd_internals *internals = NULL;
1544         struct rte_eth_dev *eth_dev = NULL;
1545
1546         PMD_LOG(INFO, "Closing pcap ethdev on numa socket %d",
1547                         rte_socket_id());
1548
1549         if (!dev)
1550                 return -1;
1551
1552         /* reserve an ethdev entry */
1553         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1554         if (eth_dev == NULL)
1555                 return -1;
1556
1557         if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
1558                 internals = eth_dev->data->dev_private;
1559                 if (internals != NULL && internals->phy_mac == 0)
1560                         /* not dynamically allocated, must not be freed */
1561                         eth_dev->data->mac_addrs = NULL;
1562         }
1563
1564         eth_dev_close(eth_dev);
1565
1566         rte_free(eth_dev->process_private);
1567         rte_eth_dev_release_port(eth_dev);
1568
1569         return 0;
1570 }
1571
1572 static struct rte_vdev_driver pmd_pcap_drv = {
1573         .probe = pmd_pcap_probe,
1574         .remove = pmd_pcap_remove,
1575 };
1576
1577 RTE_PMD_REGISTER_VDEV(net_pcap, pmd_pcap_drv);
1578 RTE_PMD_REGISTER_ALIAS(net_pcap, eth_pcap);
1579 RTE_PMD_REGISTER_PARAM_STRING(net_pcap,
1580         ETH_PCAP_RX_PCAP_ARG "=<string> "
1581         ETH_PCAP_TX_PCAP_ARG "=<string> "
1582         ETH_PCAP_RX_IFACE_ARG "=<ifc> "
1583         ETH_PCAP_RX_IFACE_IN_ARG "=<ifc> "
1584         ETH_PCAP_TX_IFACE_ARG "=<ifc> "
1585         ETH_PCAP_IFACE_ARG "=<ifc> "
1586         ETH_PCAP_PHY_MAC_ARG "=<int>"
1587         ETH_PCAP_INFINITE_RX_ARG "=<0|1>");
1588
1589 RTE_INIT(eth_pcap_init_log)
1590 {
1591         eth_pcap_logtype = rte_log_register("pmd.net.pcap");
1592         if (eth_pcap_logtype >= 0)
1593                 rte_log_set_level(eth_pcap_logtype, RTE_LOG_NOTICE);
1594 }