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 int
 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         return 0;
 724 }
 725
 726 static void
 727 eth_dev_close(struct rte_eth_dev *dev)
 728 {
 729         unsigned int i;
 730         struct pmd_internals *internals = dev->data->dev_private;
 731
 732         /* Device wide flag, but cleanup must be performed per queue. */
 733         if (internals->infinite_rx) {
 734                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
 735                         struct pcap_rx_queue *pcap_q = &internals->rx_queue[i];
 736                         struct rte_mbuf *pcap_buf;
 737
 738                         while (!rte_ring_dequeue(pcap_q->pkts,
 739                                         (void **)&pcap_buf))
 740                                 rte_pktmbuf_free(pcap_buf);
 741
 742                         rte_ring_free(pcap_q->pkts);
 743                 }
 744         }
 745
 746 }
 747
 748 static void
 749 eth_queue_release(void *q __rte_unused)
 750 {
 751 }
 752
 753 static int
 754 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 755                 int wait_to_complete __rte_unused)
 756 {
 757         return 0;
 758 }
 759
 760 static int
 761 eth_rx_queue_setup(struct rte_eth_dev *dev,
 762                 uint16_t rx_queue_id,
 763                 uint16_t nb_rx_desc __rte_unused,
 764                 unsigned int socket_id __rte_unused,
 765                 const struct rte_eth_rxconf *rx_conf __rte_unused,
 766                 struct rte_mempool *mb_pool)
 767 {
 768         struct pmd_internals *internals = dev->data->dev_private;
 769         struct pcap_rx_queue *pcap_q = &internals->rx_queue[rx_queue_id];
 770
 771         pcap_q->mb_pool = mb_pool;
 772         pcap_q->port_id = dev->data->port_id;
 773         pcap_q->queue_id = rx_queue_id;
 774         dev->data->rx_queues[rx_queue_id] = pcap_q;
 775
 776         if (internals->infinite_rx) {
 777                 struct pmd_process_private *pp;
 778                 char ring_name[NAME_MAX];
 779                 static uint32_t ring_number;
 780                 uint64_t pcap_pkt_count = 0;
 781                 struct rte_mbuf *bufs[1];
 782                 pcap_t **pcap;
 783
 784                 pp = rte_eth_devices[pcap_q->port_id].process_private;
 785                 pcap = &pp->rx_pcap[pcap_q->queue_id];
 786
 787                 if (unlikely(*pcap == NULL))
 788                         return -ENOENT;
 789
 790                 pcap_pkt_count = count_packets_in_pcap(pcap, pcap_q);
 791
 792                 snprintf(ring_name, sizeof(ring_name), "PCAP_RING%" PRIu16,
 793                                 ring_number);
 794
 795                 pcap_q->pkts = rte_ring_create(ring_name,
 796                                 rte_align64pow2(pcap_pkt_count + 1), 0,
 797                                 RING_F_SP_ENQ | RING_F_SC_DEQ);
 798                 ring_number++;
 799                 if (!pcap_q->pkts)
 800                         return -ENOENT;
 801
 802                 /* Fill ring with packets from PCAP file one by one. */
 803                 while (eth_pcap_rx(pcap_q, bufs, 1)) {
 804                         /* Check for multiseg mbufs. */
 805                         if (bufs[0]->nb_segs != 1) {
 806                                 rte_pktmbuf_free(*bufs);
 807
 808                                 while (!rte_ring_dequeue(pcap_q->pkts,
 809                                                 (void **)bufs))
 810                                         rte_pktmbuf_free(*bufs);
 811
 812                                 rte_ring_free(pcap_q->pkts);
 813                                 PMD_LOG(ERR, "Multiseg mbufs are not supported in infinite_rx "
 814                                                 "mode.");
 815                                 return -EINVAL;
 816                         }
 817
 818                         rte_ring_enqueue_bulk(pcap_q->pkts,
 819                                         (void * const *)bufs, 1, NULL);
 820                 }
 821                 /*
 822                  * Reset the stats for this queue since eth_pcap_rx calls above
 823                  * didn't result in the application receiving packets.
 824                  */
 825                 pcap_q->rx_stat.pkts = 0;
 826                 pcap_q->rx_stat.bytes = 0;
 827         }
 828
 829         return 0;
 830 }
 831
 832 static int
 833 eth_tx_queue_setup(struct rte_eth_dev *dev,
 834                 uint16_t tx_queue_id,
 835                 uint16_t nb_tx_desc __rte_unused,
 836                 unsigned int socket_id __rte_unused,
 837                 const struct rte_eth_txconf *tx_conf __rte_unused)
 838 {
 839         struct pmd_internals *internals = dev->data->dev_private;
 840         struct pcap_tx_queue *pcap_q = &internals->tx_queue[tx_queue_id];
 841
 842         pcap_q->port_id = dev->data->port_id;
 843         pcap_q->queue_id = tx_queue_id;
 844         dev->data->tx_queues[tx_queue_id] = pcap_q;
 845
 846         return 0;
 847 }
 848
 849 static int
 850 eth_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 851 {
 852         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
 853
 854         return 0;
 855 }
 856
 857 static int
 858 eth_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
 859 {
 860         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
 861
 862         return 0;
 863 }
 864
 865 static int
 866 eth_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 867 {
 868         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
 869
 870         return 0;
 871 }
 872
 873 static int
 874 eth_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
 875 {
 876         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
 877
 878         return 0;
 879 }
 880
 881 static const struct eth_dev_ops ops = {
 882         .dev_start = eth_dev_start,
 883         .dev_stop = eth_dev_stop,
 884         .dev_close = eth_dev_close,
 885         .dev_configure = eth_dev_configure,
 886         .dev_infos_get = eth_dev_info,
 887         .rx_queue_setup = eth_rx_queue_setup,
 888         .tx_queue_setup = eth_tx_queue_setup,
 889         .rx_queue_start = eth_rx_queue_start,
 890         .tx_queue_start = eth_tx_queue_start,
 891         .rx_queue_stop = eth_rx_queue_stop,
 892         .tx_queue_stop = eth_tx_queue_stop,
 893         .rx_queue_release = eth_queue_release,
 894         .tx_queue_release = eth_queue_release,
 895         .link_update = eth_link_update,
 896         .stats_get = eth_stats_get,
 897         .stats_reset = eth_stats_reset,
 898 };
 899
 900 static int
 901 add_queue(struct pmd_devargs *pmd, const char *name, const char *type,
 902                 pcap_t *pcap, pcap_dumper_t *dumper)
 903 {
 904         if (pmd->num_of_queue >= RTE_PMD_PCAP_MAX_QUEUES)
 905                 return -1;
 906         if (pcap)
 907                 pmd->queue[pmd->num_of_queue].pcap = pcap;
 908         if (dumper)
 909                 pmd->queue[pmd->num_of_queue].dumper = dumper;
 910         pmd->queue[pmd->num_of_queue].name = name;
 911         pmd->queue[pmd->num_of_queue].type = type;
 912         pmd->num_of_queue++;
 913         return 0;
 914 }
 915
 916 /*
 917  * Function handler that opens the pcap file for reading a stores a
 918  * reference of it for use it later on.
 919  */
 920 static int
 921 open_rx_pcap(const char *key, const char *value, void *extra_args)
 922 {
 923         const char *pcap_filename = value;
 924         struct pmd_devargs *rx = extra_args;
 925         pcap_t *pcap = NULL;
 926
 927         if (open_single_rx_pcap(pcap_filename, &pcap) < 0)
 928                 return -1;
 929
 930         if (add_queue(rx, pcap_filename, key, pcap, NULL) < 0) {
 931                 pcap_close(pcap);
 932                 return -1;
 933         }
 934
 935         return 0;
 936 }
 937
 938 /*
 939  * Opens a pcap file for writing and stores a reference to it
 940  * for use it later on.
 941  */
 942 static int
 943 open_tx_pcap(const char *key, const char *value, void *extra_args)
 944 {
 945         const char *pcap_filename = value;
 946         struct pmd_devargs *dumpers = extra_args;
 947         pcap_dumper_t *dumper;
 948
 949         if (open_single_tx_pcap(pcap_filename, &dumper) < 0)
 950                 return -1;
 951
 952         if (add_queue(dumpers, pcap_filename, key, NULL, dumper) < 0) {
 953                 pcap_dump_close(dumper);
 954                 return -1;
 955         }
 956
 957         return 0;
 958 }
 959
 960 /*
 961  * Opens an interface for reading and writing
 962  */
 963 static inline int
 964 open_rx_tx_iface(const char *key, const char *value, void *extra_args)
 965 {
 966         const char *iface = value;
 967         struct pmd_devargs *tx = extra_args;
 968         pcap_t *pcap = NULL;
 969
 970         if (open_single_iface(iface, &pcap) < 0)
 971                 return -1;
 972
 973         tx->queue[0].pcap = pcap;
 974         tx->queue[0].name = iface;
 975         tx->queue[0].type = key;
 976
 977         return 0;
 978 }
 979
 980 static inline int
 981 set_iface_direction(const char *iface, pcap_t *pcap,
 982                 pcap_direction_t direction)
 983 {
 984         const char *direction_str = (direction == PCAP_D_IN) ? "IN" : "OUT";
 985         if (pcap_setdirection(pcap, direction) < 0) {
 986                 PMD_LOG(ERR, "Setting %s pcap direction %s failed - %s\n",
 987                                 iface, direction_str, pcap_geterr(pcap));
 988                 return -1;
 989         }
 990         PMD_LOG(INFO, "Setting %s pcap direction %s\n",
 991                         iface, direction_str);
 992         return 0;
 993 }
 994
 995 static inline int
 996 open_iface(const char *key, const char *value, void *extra_args)
 997 {
 998         const char *iface = value;
 999         struct pmd_devargs *pmd = extra_args;
1000         pcap_t *pcap = NULL;
1001
1002         if (open_single_iface(iface, &pcap) < 0)
1003                 return -1;
1004         if (add_queue(pmd, iface, key, pcap, NULL) < 0) {
1005                 pcap_close(pcap);
1006                 return -1;
1007         }
1008
1009         return 0;
1010 }
1011
1012 /*
1013  * Opens a NIC for reading packets from it
1014  */
1015 static inline int
1016 open_rx_iface(const char *key, const char *value, void *extra_args)
1017 {
1018         int ret = open_iface(key, value, extra_args);
1019         if (ret < 0)
1020                 return ret;
1021         if (strcmp(key, ETH_PCAP_RX_IFACE_IN_ARG) == 0) {
1022                 struct pmd_devargs *pmd = extra_args;
1023                 unsigned int qid = pmd->num_of_queue - 1;
1024
1025                 set_iface_direction(pmd->queue[qid].name,
1026                                 pmd->queue[qid].pcap,
1027                                 PCAP_D_IN);
1028         }
1029
1030         return 0;
1031 }
1032
1033 static inline int
1034 rx_iface_args_process(const char *key, const char *value, void *extra_args)
1035 {
1036         if (strcmp(key, ETH_PCAP_RX_IFACE_ARG) == 0 ||
1037                         strcmp(key, ETH_PCAP_RX_IFACE_IN_ARG) == 0)
1038                 return open_rx_iface(key, value, extra_args);
1039
1040         return 0;
1041 }
1042
1043 /*
1044  * Opens a NIC for writing packets to it
1045  */
1046 static int
1047 open_tx_iface(const char *key, const char *value, void *extra_args)
1048 {
1049         return open_iface(key, value, extra_args);
1050 }
1051
1052 static int
1053 select_phy_mac(const char *key __rte_unused, const char *value,
1054                 void *extra_args)
1055 {
1056         if (extra_args) {
1057                 const int phy_mac = atoi(value);
1058                 int *enable_phy_mac = extra_args;
1059
1060                 if (phy_mac)
1061                         *enable_phy_mac = 1;
1062         }
1063         return 0;
1064 }
1065
1066 static int
1067 get_infinite_rx_arg(const char *key __rte_unused,
1068                 const char *value, void *extra_args)
1069 {
1070         if (extra_args) {
1071                 const int infinite_rx = atoi(value);
1072                 int *enable_infinite_rx = extra_args;
1073
1074                 if (infinite_rx > 0)
1075                         *enable_infinite_rx = 1;
1076         }
1077         return 0;
1078 }
1079
1080 static int
1081 pmd_init_internals(struct rte_vdev_device *vdev,
1082                 const unsigned int nb_rx_queues,
1083                 const unsigned int nb_tx_queues,
1084                 struct pmd_internals **internals,
1085                 struct rte_eth_dev **eth_dev)
1086 {
1087         struct rte_eth_dev_data *data;
1088         struct pmd_process_private *pp;
1089         unsigned int numa_node = vdev->device.numa_node;
1090
1091         PMD_LOG(INFO, "Creating pcap-backed ethdev on numa socket %d",
1092                 numa_node);
1093
1094         pp = (struct pmd_process_private *)
1095                 rte_zmalloc(NULL, sizeof(struct pmd_process_private),
1096                                 RTE_CACHE_LINE_SIZE);
1097
1098         if (pp == NULL) {
1099                 PMD_LOG(ERR,
1100                         "Failed to allocate memory for process private");
1101                 return -1;
1102         }
1103
1104         /* reserve an ethdev entry */
1105         *eth_dev = rte_eth_vdev_allocate(vdev, sizeof(**internals));
1106         if (!(*eth_dev)) {
1107                 rte_free(pp);
1108                 return -1;
1109         }
1110         (*eth_dev)->process_private = pp;
1111         /* now put it all together
1112          * - store queue data in internals,
1113          * - store numa_node info in eth_dev
1114          * - point eth_dev_data to internals
1115          * - and point eth_dev structure to new eth_dev_data structure
1116          */
1117         *internals = (*eth_dev)->data->dev_private;
1118         /*
1119          * Interface MAC = 02:70:63:61:70:<iface_idx>
1120          * derived from: 'locally administered':'p':'c':'a':'p':'iface_idx'
1121          * where the middle 4 characters are converted to hex.
1122          */
1123         (*internals)->eth_addr = (struct rte_ether_addr) {
1124                 .addr_bytes = { 0x02, 0x70, 0x63, 0x61, 0x70, iface_idx++ }
1125         };
1126         (*internals)->phy_mac = 0;
1127         data = (*eth_dev)->data;
1128         data->nb_rx_queues = (uint16_t)nb_rx_queues;
1129         data->nb_tx_queues = (uint16_t)nb_tx_queues;
1130         data->dev_link = pmd_link;
1131         data->mac_addrs = &(*internals)->eth_addr;
1132
1133         /*
1134          * NOTE: we'll replace the data element, of originally allocated
1135          * eth_dev so the rings are local per-process
1136          */
1137         (*eth_dev)->dev_ops = &ops;
1138
1139         strlcpy((*internals)->devargs, rte_vdev_device_args(vdev),
1140                         ETH_PCAP_ARG_MAXLEN);
1141
1142         return 0;
1143 }
1144
1145 static int
1146 eth_pcap_update_mac(const char *if_name, struct rte_eth_dev *eth_dev,
1147                 const unsigned int numa_node)
1148 {
1149 #if defined(RTE_EXEC_ENV_LINUX)
1150         void *mac_addrs;
1151         struct ifreq ifr;
1152         int if_fd = socket(AF_INET, SOCK_DGRAM, 0);
1153
1154         if (if_fd == -1)
1155                 return -1;
1156
1157         rte_strscpy(ifr.ifr_name, if_name, sizeof(ifr.ifr_name));
1158         if (ioctl(if_fd, SIOCGIFHWADDR, &ifr)) {
1159                 close(if_fd);
1160                 return -1;
1161         }
1162
1163         mac_addrs = rte_zmalloc_socket(NULL, RTE_ETHER_ADDR_LEN, 0, numa_node);
1164         if (!mac_addrs) {
1165                 close(if_fd);
1166                 return -1;
1167         }
1168
1169         PMD_LOG(INFO, "Setting phy MAC for %s", if_name);
1170         eth_dev->data->mac_addrs = mac_addrs;
1171         rte_memcpy(eth_dev->data->mac_addrs[0].addr_bytes,
1172                         ifr.ifr_hwaddr.sa_data, RTE_ETHER_ADDR_LEN);
1173
1174         close(if_fd);
1175
1176         return 0;
1177
1178 #elif defined(RTE_EXEC_ENV_FREEBSD)
1179         void *mac_addrs;
1180         struct if_msghdr *ifm;
1181         struct sockaddr_dl *sdl;
1182         int mib[6];
1183         size_t len = 0;
1184         char *buf;
1185
1186         mib[0] = CTL_NET;
1187         mib[1] = AF_ROUTE;
1188         mib[2] = 0;
1189         mib[3] = AF_LINK;
1190         mib[4] = NET_RT_IFLIST;
1191         mib[5] = if_nametoindex(if_name);
1192
1193         if (sysctl(mib, 6, NULL, &len, NULL, 0) < 0)
1194                 return -1;
1195
1196         if (len == 0)
1197                 return -1;
1198
1199         buf = rte_malloc(NULL, len, 0);
1200         if (!buf)
1201                 return -1;
1202
1203         if (sysctl(mib, 6, buf, &len, NULL, 0) < 0) {
1204                 rte_free(buf);
1205                 return -1;
1206         }
1207         ifm = (struct if_msghdr *)buf;
1208         sdl = (struct sockaddr_dl *)(ifm + 1);
1209
1210         mac_addrs = rte_zmalloc_socket(NULL, RTE_ETHER_ADDR_LEN, 0, numa_node);
1211         if (!mac_addrs) {
1212                 rte_free(buf);
1213                 return -1;
1214         }
1215
1216         PMD_LOG(INFO, "Setting phy MAC for %s", if_name);
1217         eth_dev->data->mac_addrs = mac_addrs;
1218         rte_memcpy(eth_dev->data->mac_addrs[0].addr_bytes,
1219                         LLADDR(sdl), RTE_ETHER_ADDR_LEN);
1220
1221         rte_free(buf);
1222
1223         return 0;
1224 #else
1225         return -1;
1226 #endif
1227 }
1228
1229 static int
1230 eth_from_pcaps_common(struct rte_vdev_device *vdev,
1231                 struct pmd_devargs_all *devargs_all,
1232                 struct pmd_internals **internals, struct rte_eth_dev **eth_dev)
1233 {
1234         struct pmd_process_private *pp;
1235         struct pmd_devargs *rx_queues = &devargs_all->rx_queues;
1236         struct pmd_devargs *tx_queues = &devargs_all->tx_queues;
1237         const unsigned int nb_rx_queues = rx_queues->num_of_queue;
1238         const unsigned int nb_tx_queues = tx_queues->num_of_queue;
1239         unsigned int i;
1240
1241         /* do some parameter checking */
1242         if (rx_queues == NULL && nb_rx_queues > 0)
1243                 return -1;
1244         if (tx_queues == NULL && nb_tx_queues > 0)
1245                 return -1;
1246
1247         if (pmd_init_internals(vdev, nb_rx_queues, nb_tx_queues, internals,
1248                         eth_dev) < 0)
1249                 return -1;
1250
1251         pp = (*eth_dev)->process_private;
1252         for (i = 0; i < nb_rx_queues; i++) {
1253                 struct pcap_rx_queue *rx = &(*internals)->rx_queue[i];
1254                 struct devargs_queue *queue = &rx_queues->queue[i];
1255
1256                 pp->rx_pcap[i] = queue->pcap;
1257                 strlcpy(rx->name, queue->name, sizeof(rx->name));
1258                 strlcpy(rx->type, queue->type, sizeof(rx->type));
1259         }
1260
1261         for (i = 0; i < nb_tx_queues; i++) {
1262                 struct pcap_tx_queue *tx = &(*internals)->tx_queue[i];
1263                 struct devargs_queue *queue = &tx_queues->queue[i];
1264
1265                 pp->tx_dumper[i] = queue->dumper;
1266                 pp->tx_pcap[i] = queue->pcap;
1267                 strlcpy(tx->name, queue->name, sizeof(tx->name));
1268                 strlcpy(tx->type, queue->type, sizeof(tx->type));
1269         }
1270
1271         return 0;
1272 }
1273
1274 static int
1275 eth_from_pcaps(struct rte_vdev_device *vdev,
1276                 struct pmd_devargs_all *devargs_all)
1277 {
1278         struct pmd_internals *internals = NULL;
1279         struct rte_eth_dev *eth_dev = NULL;
1280         struct pmd_devargs *rx_queues = &devargs_all->rx_queues;
1281         int single_iface = devargs_all->single_iface;
1282         unsigned int infinite_rx = devargs_all->infinite_rx;
1283         int ret;
1284
1285         ret = eth_from_pcaps_common(vdev, devargs_all, &internals, &eth_dev);
1286
1287         if (ret < 0)
1288                 return ret;
1289
1290         /* store weather we are using a single interface for rx/tx or not */
1291         internals->single_iface = single_iface;
1292
1293         if (single_iface) {
1294                 internals->if_index = if_nametoindex(rx_queues->queue[0].name);
1295
1296                 /* phy_mac arg is applied only only if "iface" devarg is provided */
1297                 if (rx_queues->phy_mac) {
1298                         int ret = eth_pcap_update_mac(rx_queues->queue[0].name,
1299                                         eth_dev, vdev->device.numa_node);
1300                         if (ret == 0)
1301                                 internals->phy_mac = 1;
1302                 }
1303         }
1304
1305         internals->infinite_rx = infinite_rx;
1306         /* Assign rx ops. */
1307         if (infinite_rx)
1308                 eth_dev->rx_pkt_burst = eth_pcap_rx_infinite;
1309         else if (devargs_all->is_rx_pcap || devargs_all->is_rx_iface ||
1310                         single_iface)
1311                 eth_dev->rx_pkt_burst = eth_pcap_rx;
1312         else
1313                 eth_dev->rx_pkt_burst = eth_null_rx;
1314
1315         /* Assign tx ops. */
1316         if (devargs_all->is_tx_pcap)
1317                 eth_dev->tx_pkt_burst = eth_pcap_tx_dumper;
1318         else if (devargs_all->is_tx_iface || single_iface)
1319                 eth_dev->tx_pkt_burst = eth_pcap_tx;
1320         else
1321                 eth_dev->tx_pkt_burst = eth_tx_drop;
1322
1323         rte_eth_dev_probing_finish(eth_dev);
1324         return 0;
1325 }
1326
1327 static int
1328 pmd_pcap_probe(struct rte_vdev_device *dev)
1329 {
1330         const char *name;
1331         struct rte_kvargs *kvlist;
1332         struct pmd_devargs pcaps = {0};
1333         struct pmd_devargs dumpers = {0};
1334         struct rte_eth_dev *eth_dev =  NULL;
1335         struct pmd_internals *internal;
1336         int ret = 0;
1337
1338         struct pmd_devargs_all devargs_all = {
1339                 .single_iface = 0,
1340                 .is_tx_pcap = 0,
1341                 .is_tx_iface = 0,
1342                 .infinite_rx = 0,
1343         };
1344
1345         name = rte_vdev_device_name(dev);
1346         PMD_LOG(INFO, "Initializing pmd_pcap for %s", name);
1347
1348         gettimeofday(&start_time, NULL);
1349         start_cycles = rte_get_timer_cycles();
1350         hz = rte_get_timer_hz();
1351
1352         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1353                 eth_dev = rte_eth_dev_attach_secondary(name);
1354                 if (!eth_dev) {
1355                         PMD_LOG(ERR, "Failed to probe %s", name);
1356                         return -1;
1357                 }
1358
1359                 internal = eth_dev->data->dev_private;
1360
1361                 kvlist = rte_kvargs_parse(internal->devargs, valid_arguments);
1362                 if (kvlist == NULL)
1363                         return -1;
1364         } else {
1365                 kvlist = rte_kvargs_parse(rte_vdev_device_args(dev),
1366                                 valid_arguments);
1367                 if (kvlist == NULL)
1368                         return -1;
1369         }
1370
1371         /*
1372          * If iface argument is passed we open the NICs and use them for
1373          * reading / writing
1374          */
1375         if (rte_kvargs_count(kvlist, ETH_PCAP_IFACE_ARG) == 1) {
1376
1377                 ret = rte_kvargs_process(kvlist, ETH_PCAP_IFACE_ARG,
1378                                 &open_rx_tx_iface, &pcaps);
1379                 if (ret < 0)
1380                         goto free_kvlist;
1381
1382                 dumpers.queue[0] = pcaps.queue[0];
1383
1384                 ret = rte_kvargs_process(kvlist, ETH_PCAP_PHY_MAC_ARG,
1385                                 &select_phy_mac, &pcaps.phy_mac);
1386                 if (ret < 0)
1387                         goto free_kvlist;
1388
1389                 dumpers.phy_mac = pcaps.phy_mac;
1390
1391                 devargs_all.single_iface = 1;
1392                 pcaps.num_of_queue = 1;
1393                 dumpers.num_of_queue = 1;
1394
1395                 goto create_eth;
1396         }
1397
1398         /*
1399          * We check whether we want to open a RX stream from a real NIC, a
1400          * pcap file or open a dummy RX stream
1401          */
1402         devargs_all.is_rx_pcap =
1403                 rte_kvargs_count(kvlist, ETH_PCAP_RX_PCAP_ARG) ? 1 : 0;
1404         devargs_all.is_rx_iface =
1405                 rte_kvargs_count(kvlist, ETH_PCAP_RX_IFACE_ARG) ? 1 : 0;
1406         pcaps.num_of_queue = 0;
1407
1408         devargs_all.is_tx_pcap =
1409                 rte_kvargs_count(kvlist, ETH_PCAP_TX_PCAP_ARG) ? 1 : 0;
1410         devargs_all.is_tx_iface =
1411                 rte_kvargs_count(kvlist, ETH_PCAP_TX_IFACE_ARG) ? 1 : 0;
1412         dumpers.num_of_queue = 0;
1413
1414         if (devargs_all.is_rx_pcap) {
1415                 /*
1416                  * We check whether we want to infinitely rx the pcap file.
1417                  */
1418                 unsigned int infinite_rx_arg_cnt = rte_kvargs_count(kvlist,
1419                                 ETH_PCAP_INFINITE_RX_ARG);
1420
1421                 if (infinite_rx_arg_cnt == 1) {
1422                         ret = rte_kvargs_process(kvlist,
1423                                         ETH_PCAP_INFINITE_RX_ARG,
1424                                         &get_infinite_rx_arg,
1425                                         &devargs_all.infinite_rx);
1426                         if (ret < 0)
1427                                 goto free_kvlist;
1428                         PMD_LOG(INFO, "infinite_rx has been %s for %s",
1429                                         devargs_all.infinite_rx ? "enabled" : "disabled",
1430                                         name);
1431
1432                 } else if (infinite_rx_arg_cnt > 1) {
1433                         PMD_LOG(WARNING, "infinite_rx has not been enabled since the "
1434                                         "argument has been provided more than once "
1435                                         "for %s", name);
1436                 }
1437
1438                 ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_PCAP_ARG,
1439                                 &open_rx_pcap, &pcaps);
1440         } else if (devargs_all.is_rx_iface) {
1441                 ret = rte_kvargs_process(kvlist, NULL,
1442                                 &rx_iface_args_process, &pcaps);
1443         } else if (devargs_all.is_tx_iface || devargs_all.is_tx_pcap) {
1444                 unsigned int i;
1445
1446                 /* Count number of tx queue args passed before dummy rx queue
1447                  * creation so a dummy rx queue can be created for each tx queue
1448                  */
1449                 unsigned int num_tx_queues =
1450                         (rte_kvargs_count(kvlist, ETH_PCAP_TX_PCAP_ARG) +
1451                         rte_kvargs_count(kvlist, ETH_PCAP_TX_IFACE_ARG));
1452
1453                 PMD_LOG(INFO, "Creating null rx queue since no rx queues were provided.");
1454
1455                 /* Creating a dummy rx queue for each tx queue passed */
1456                 for (i = 0; i < num_tx_queues; i++)
1457                         ret = add_queue(&pcaps, "dummy_rx", "rx_null", NULL,
1458                                         NULL);
1459         } else {
1460                 PMD_LOG(ERR, "Error - No rx or tx queues provided");
1461                 ret = -ENOENT;
1462         }
1463         if (ret < 0)
1464                 goto free_kvlist;
1465
1466         /*
1467          * We check whether we want to open a TX stream to a real NIC,
1468          * a pcap file, or drop packets on tx
1469          */
1470         if (devargs_all.is_tx_pcap) {
1471                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_PCAP_ARG,
1472                                 &open_tx_pcap, &dumpers);
1473         } else if (devargs_all.is_tx_iface) {
1474                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_IFACE_ARG,
1475                                 &open_tx_iface, &dumpers);
1476         } else {
1477                 unsigned int i;
1478
1479                 PMD_LOG(INFO, "Dropping packets on tx since no tx queues were provided.");
1480
1481                 /* Add 1 dummy queue per rxq which counts and drops packets. */
1482                 for (i = 0; i < pcaps.num_of_queue; i++)
1483                         ret = add_queue(&dumpers, "dummy_tx", "tx_drop", NULL,
1484                                         NULL);
1485         }
1486
1487         if (ret < 0)
1488                 goto free_kvlist;
1489
1490 create_eth:
1491         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1492                 struct pmd_process_private *pp;
1493                 unsigned int i;
1494
1495                 internal = eth_dev->data->dev_private;
1496                         pp = (struct pmd_process_private *)
1497                                 rte_zmalloc(NULL,
1498                                         sizeof(struct pmd_process_private),
1499                                         RTE_CACHE_LINE_SIZE);
1500
1501                 if (pp == NULL) {
1502                         PMD_LOG(ERR,
1503                                 "Failed to allocate memory for process private");
1504                         ret = -1;
1505                         goto free_kvlist;
1506                 }
1507
1508                 eth_dev->dev_ops = &ops;
1509                 eth_dev->device = &dev->device;
1510
1511                 /* setup process private */
1512                 for (i = 0; i < pcaps.num_of_queue; i++)
1513                         pp->rx_pcap[i] = pcaps.queue[i].pcap;
1514
1515                 for (i = 0; i < dumpers.num_of_queue; i++) {
1516                         pp->tx_dumper[i] = dumpers.queue[i].dumper;
1517                         pp->tx_pcap[i] = dumpers.queue[i].pcap;
1518                 }
1519
1520                 eth_dev->process_private = pp;
1521                 eth_dev->rx_pkt_burst = eth_pcap_rx;
1522                 if (devargs_all.is_tx_pcap)
1523                         eth_dev->tx_pkt_burst = eth_pcap_tx_dumper;
1524                 else
1525                         eth_dev->tx_pkt_burst = eth_pcap_tx;
1526
1527                 rte_eth_dev_probing_finish(eth_dev);
1528                 goto free_kvlist;
1529         }
1530
1531         devargs_all.rx_queues = pcaps;
1532         devargs_all.tx_queues = dumpers;
1533
1534         ret = eth_from_pcaps(dev, &devargs_all);
1535
1536 free_kvlist:
1537         rte_kvargs_free(kvlist);
1538
1539         return ret;
1540 }
1541
1542 static int
1543 pmd_pcap_remove(struct rte_vdev_device *dev)
1544 {
1545         struct pmd_internals *internals = NULL;
1546         struct rte_eth_dev *eth_dev = NULL;
1547
1548         PMD_LOG(INFO, "Closing pcap ethdev on numa socket %d",
1549                         rte_socket_id());
1550
1551         if (!dev)
1552                 return -1;
1553
1554         /* reserve an ethdev entry */
1555         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1556         if (eth_dev == NULL)
1557                 return -1;
1558
1559         if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
1560                 internals = eth_dev->data->dev_private;
1561                 if (internals != NULL && internals->phy_mac == 0)
1562                         /* not dynamically allocated, must not be freed */
1563                         eth_dev->data->mac_addrs = NULL;
1564         }
1565
1566         eth_dev_close(eth_dev);
1567
1568         rte_free(eth_dev->process_private);
1569         rte_eth_dev_release_port(eth_dev);
1570
1571         return 0;
1572 }
1573
1574 static struct rte_vdev_driver pmd_pcap_drv = {
1575         .probe = pmd_pcap_probe,
1576         .remove = pmd_pcap_remove,
1577 };
1578
1579 RTE_PMD_REGISTER_VDEV(net_pcap, pmd_pcap_drv);
1580 RTE_PMD_REGISTER_ALIAS(net_pcap, eth_pcap);
1581 RTE_PMD_REGISTER_PARAM_STRING(net_pcap,
1582         ETH_PCAP_RX_PCAP_ARG "=<string> "
1583         ETH_PCAP_TX_PCAP_ARG "=<string> "
1584         ETH_PCAP_RX_IFACE_ARG "=<ifc> "
1585         ETH_PCAP_RX_IFACE_IN_ARG "=<ifc> "
1586         ETH_PCAP_TX_IFACE_ARG "=<ifc> "
1587         ETH_PCAP_IFACE_ARG "=<ifc> "
1588         ETH_PCAP_PHY_MAC_ARG "=<int>"
1589         ETH_PCAP_INFINITE_RX_ARG "=<0|1>");
1590
1591 RTE_INIT(eth_pcap_init_log)
1592 {
1593         eth_pcap_logtype = rte_log_register("pmd.net.pcap");
1594         if (eth_pcap_logtype >= 0)
1595                 rte_log_set_level(eth_pcap_logtype, RTE_LOG_NOTICE);
1596 }