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 <ethdev_driver.h>
  23 #include <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 static uint64_t timestamp_rx_dynflag;
  55 static int timestamp_dynfield_offset = -1;
  56
  57 struct queue_stat {
  58         volatile unsigned long pkts;
  59         volatile unsigned long bytes;
  60         volatile unsigned long err_pkts;
  61 };
  62
  63 struct pcap_rx_queue {
  64         uint16_t port_id;
  65         uint16_t queue_id;
  66         struct rte_mempool *mb_pool;
  67         struct queue_stat rx_stat;
  68         char name[PATH_MAX];
  69         char type[ETH_PCAP_ARG_MAXLEN];
  70
  71         /* Contains pre-generated packets to be looped through */
  72         struct rte_ring *pkts;
  73 };
  74
  75 struct pcap_tx_queue {
  76         uint16_t port_id;
  77         uint16_t queue_id;
  78         struct queue_stat tx_stat;
  79         char name[PATH_MAX];
  80         char type[ETH_PCAP_ARG_MAXLEN];
  81 };
  82
  83 struct pmd_internals {
  84         struct pcap_rx_queue rx_queue[RTE_PMD_PCAP_MAX_QUEUES];
  85         struct pcap_tx_queue tx_queue[RTE_PMD_PCAP_MAX_QUEUES];
  86         char devargs[ETH_PCAP_ARG_MAXLEN];
  87         struct rte_ether_addr eth_addr;
  88         int if_index;
  89         int single_iface;
  90         int phy_mac;
  91         unsigned int infinite_rx;
  92 };
  93
  94 struct pmd_process_private {
  95         pcap_t *rx_pcap[RTE_PMD_PCAP_MAX_QUEUES];
  96         pcap_t *tx_pcap[RTE_PMD_PCAP_MAX_QUEUES];
  97         pcap_dumper_t *tx_dumper[RTE_PMD_PCAP_MAX_QUEUES];
  98 };
  99
 100 struct pmd_devargs {
 101         unsigned int num_of_queue;
 102         struct devargs_queue {
 103                 pcap_dumper_t *dumper;
 104                 pcap_t *pcap;
 105                 const char *name;
 106                 const char *type;
 107         } queue[RTE_PMD_PCAP_MAX_QUEUES];
 108         int phy_mac;
 109 };
 110
 111 struct pmd_devargs_all {
 112         struct pmd_devargs rx_queues;
 113         struct pmd_devargs tx_queues;
 114         int single_iface;
 115         unsigned int is_tx_pcap;
 116         unsigned int is_tx_iface;
 117         unsigned int is_rx_pcap;
 118         unsigned int is_rx_iface;
 119         unsigned int infinite_rx;
 120 };
 121
 122 static const char *valid_arguments[] = {
 123         ETH_PCAP_RX_PCAP_ARG,
 124         ETH_PCAP_TX_PCAP_ARG,
 125         ETH_PCAP_RX_IFACE_ARG,
 126         ETH_PCAP_RX_IFACE_IN_ARG,
 127         ETH_PCAP_TX_IFACE_ARG,
 128         ETH_PCAP_IFACE_ARG,
 129         ETH_PCAP_PHY_MAC_ARG,
 130         ETH_PCAP_INFINITE_RX_ARG,
 131         NULL
 132 };
 133
 134 static struct rte_eth_link pmd_link = {
 135                 .link_speed = ETH_SPEED_NUM_10G,
 136                 .link_duplex = ETH_LINK_FULL_DUPLEX,
 137                 .link_status = ETH_LINK_DOWN,
 138                 .link_autoneg = ETH_LINK_FIXED,
 139 };
 140
 141 RTE_LOG_REGISTER(eth_pcap_logtype, pmd.net.pcap, NOTICE);
 142
 143 #define PMD_LOG(level, fmt, args...) \
 144         rte_log(RTE_LOG_ ## level, eth_pcap_logtype, \
 145                 "%s(): " fmt "\n", __func__, ##args)
 146
 147 static int
 148 eth_pcap_rx_jumbo(struct rte_mempool *mb_pool, struct rte_mbuf *mbuf,
 149                 const u_char *data, uint16_t data_len)
 150 {
 151         /* Copy the first segment. */
 152         uint16_t len = rte_pktmbuf_tailroom(mbuf);
 153         struct rte_mbuf *m = mbuf;
 154
 155         rte_memcpy(rte_pktmbuf_append(mbuf, len), data, len);
 156         data_len -= len;
 157         data += len;
 158
 159         while (data_len > 0) {
 160                 /* Allocate next mbuf and point to that. */
 161                 m->next = rte_pktmbuf_alloc(mb_pool);
 162
 163                 if (unlikely(!m->next))
 164                         return -1;
 165
 166                 m = m->next;
 167
 168                 /* Headroom is not needed in chained mbufs. */
 169                 rte_pktmbuf_prepend(m, rte_pktmbuf_headroom(m));
 170                 m->pkt_len = 0;
 171                 m->data_len = 0;
 172
 173                 /* Copy next segment. */
 174                 len = RTE_MIN(rte_pktmbuf_tailroom(m), data_len);
 175                 rte_memcpy(rte_pktmbuf_append(m, len), data, len);
 176
 177                 mbuf->nb_segs++;
 178                 data_len -= len;
 179                 data += len;
 180         }
 181
 182         return mbuf->nb_segs;
 183 }
 184
 185 static uint16_t
 186 eth_pcap_rx_infinite(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 187 {
 188         int i;
 189         struct pcap_rx_queue *pcap_q = queue;
 190         uint32_t rx_bytes = 0;
 191
 192         if (unlikely(nb_pkts == 0))
 193                 return 0;
 194
 195         if (rte_pktmbuf_alloc_bulk(pcap_q->mb_pool, bufs, nb_pkts) != 0)
 196                 return 0;
 197
 198         for (i = 0; i < nb_pkts; i++) {
 199                 struct rte_mbuf *pcap_buf;
 200                 int err = rte_ring_dequeue(pcap_q->pkts, (void **)&pcap_buf);
 201                 if (err)
 202                         return i;
 203
 204                 rte_memcpy(rte_pktmbuf_mtod(bufs[i], void *),
 205                                 rte_pktmbuf_mtod(pcap_buf, void *),
 206                                 pcap_buf->data_len);
 207                 bufs[i]->data_len = pcap_buf->data_len;
 208                 bufs[i]->pkt_len = pcap_buf->pkt_len;
 209                 bufs[i]->port = pcap_q->port_id;
 210                 rx_bytes += pcap_buf->data_len;
 211
 212                 /* Enqueue packet back on ring to allow infinite rx. */
 213                 rte_ring_enqueue(pcap_q->pkts, pcap_buf);
 214         }
 215
 216         pcap_q->rx_stat.pkts += i;
 217         pcap_q->rx_stat.bytes += rx_bytes;
 218
 219         return i;
 220 }
 221
 222 static uint16_t
 223 eth_pcap_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 224 {
 225         unsigned int i;
 226         struct pcap_pkthdr header;
 227         struct pmd_process_private *pp;
 228         const u_char *packet;
 229         struct rte_mbuf *mbuf;
 230         struct pcap_rx_queue *pcap_q = queue;
 231         uint16_t num_rx = 0;
 232         uint32_t rx_bytes = 0;
 233         pcap_t *pcap;
 234
 235         pp = rte_eth_devices[pcap_q->port_id].process_private;
 236         pcap = pp->rx_pcap[pcap_q->queue_id];
 237
 238         if (unlikely(pcap == NULL || nb_pkts == 0))
 239                 return 0;
 240
 241         /* Reads the given number of packets from the pcap file one by one
 242          * and copies the packet data into a newly allocated mbuf to return.
 243          */
 244         for (i = 0; i < nb_pkts; i++) {
 245                 /* Get the next PCAP packet */
 246                 packet = pcap_next(pcap, &header);
 247                 if (unlikely(packet == NULL))
 248                         break;
 249
 250                 mbuf = rte_pktmbuf_alloc(pcap_q->mb_pool);
 251                 if (unlikely(mbuf == NULL))
 252                         break;
 253
 254                 if (header.caplen <= rte_pktmbuf_tailroom(mbuf)) {
 255                         /* pcap packet will fit in the mbuf, can copy it */
 256                         rte_memcpy(rte_pktmbuf_mtod(mbuf, void *), packet,
 257                                         header.caplen);
 258                         mbuf->data_len = (uint16_t)header.caplen;
 259                 } else {
 260                         /* Try read jumbo frame into multi mbufs. */
 261                         if (unlikely(eth_pcap_rx_jumbo(pcap_q->mb_pool,
 262                                                        mbuf,
 263                                                        packet,
 264                                                        header.caplen) == -1)) {
 265                                 rte_pktmbuf_free(mbuf);
 266                                 break;
 267                         }
 268                 }
 269
 270                 mbuf->pkt_len = (uint16_t)header.caplen;
 271                 *RTE_MBUF_DYNFIELD(mbuf, timestamp_dynfield_offset,
 272                         rte_mbuf_timestamp_t *) =
 273                                 (uint64_t)header.ts.tv_sec * 1000000 +
 274                                 header.ts.tv_usec;
 275                 mbuf->ol_flags |= timestamp_rx_dynflag;
 276                 mbuf->port = pcap_q->port_id;
 277                 bufs[num_rx] = mbuf;
 278                 num_rx++;
 279                 rx_bytes += header.caplen;
 280         }
 281         pcap_q->rx_stat.pkts += num_rx;
 282         pcap_q->rx_stat.bytes += rx_bytes;
 283
 284         return num_rx;
 285 }
 286
 287 static uint16_t
 288 eth_null_rx(void *queue __rte_unused,
 289                 struct rte_mbuf **bufs __rte_unused,
 290                 uint16_t nb_pkts __rte_unused)
 291 {
 292         return 0;
 293 }
 294
 295 #define NSEC_PER_SEC    1000000000L
 296
 297 static inline void
 298 calculate_timestamp(struct timeval *ts) {
 299         uint64_t cycles;
 300         struct timeval cur_time;
 301
 302         cycles = rte_get_timer_cycles() - start_cycles;
 303         cur_time.tv_sec = cycles / hz;
 304         cur_time.tv_usec = (cycles % hz) * NSEC_PER_SEC / hz;
 305
 306         ts->tv_sec = start_time.tv_sec + cur_time.tv_sec;
 307         ts->tv_usec = start_time.tv_usec + cur_time.tv_usec;
 308         if (ts->tv_usec >= NSEC_PER_SEC) {
 309                 ts->tv_usec -= NSEC_PER_SEC;
 310                 ts->tv_sec += 1;
 311         }
 312 }
 313
 314 /*
 315  * Callback to handle writing packets to a pcap file.
 316  */
 317 static uint16_t
 318 eth_pcap_tx_dumper(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 319 {
 320         unsigned int i;
 321         struct rte_mbuf *mbuf;
 322         struct pmd_process_private *pp;
 323         struct pcap_tx_queue *dumper_q = queue;
 324         uint16_t num_tx = 0;
 325         uint32_t tx_bytes = 0;
 326         struct pcap_pkthdr header;
 327         pcap_dumper_t *dumper;
 328         unsigned char temp_data[RTE_ETH_PCAP_SNAPLEN];
 329         size_t len, caplen;
 330
 331         pp = rte_eth_devices[dumper_q->port_id].process_private;
 332         dumper = pp->tx_dumper[dumper_q->queue_id];
 333
 334         if (dumper == NULL || nb_pkts == 0)
 335                 return 0;
 336
 337         /* writes the nb_pkts packets to the previously opened pcap file
 338          * dumper */
 339         for (i = 0; i < nb_pkts; i++) {
 340                 mbuf = bufs[i];
 341                 len = caplen = rte_pktmbuf_pkt_len(mbuf);
 342                 if (unlikely(!rte_pktmbuf_is_contiguous(mbuf) &&
 343                                 len > sizeof(temp_data))) {
 344                         caplen = sizeof(temp_data);
 345                 }
 346
 347                 calculate_timestamp(&header.ts);
 348                 header.len = len;
 349                 header.caplen = caplen;
 350                 /* rte_pktmbuf_read() returns a pointer to the data directly
 351                  * in the mbuf (when the mbuf is contiguous) or, otherwise,
 352                  * a pointer to temp_data after copying into it.
 353                  */
 354                 pcap_dump((u_char *)dumper, &header,
 355                         rte_pktmbuf_read(mbuf, 0, caplen, temp_data));
 356
 357                 num_tx++;
 358                 tx_bytes += caplen;
 359                 rte_pktmbuf_free(mbuf);
 360         }
 361
 362         /*
 363          * Since there's no place to hook a callback when the forwarding
 364          * process stops and to make sure the pcap file is actually written,
 365          * we flush the pcap dumper within each burst.
 366          */
 367         pcap_dump_flush(dumper);
 368         dumper_q->tx_stat.pkts += num_tx;
 369         dumper_q->tx_stat.bytes += tx_bytes;
 370         dumper_q->tx_stat.err_pkts += nb_pkts - num_tx;
 371
 372         return nb_pkts;
 373 }
 374
 375 /*
 376  * Callback to handle dropping packets in the infinite rx case.
 377  */
 378 static uint16_t
 379 eth_tx_drop(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 380 {
 381         unsigned int i;
 382         uint32_t tx_bytes = 0;
 383         struct pcap_tx_queue *tx_queue = queue;
 384
 385         if (unlikely(nb_pkts == 0))
 386                 return 0;
 387
 388         for (i = 0; i < nb_pkts; i++) {
 389                 tx_bytes += bufs[i]->pkt_len;
 390                 rte_pktmbuf_free(bufs[i]);
 391         }
 392
 393         tx_queue->tx_stat.pkts += nb_pkts;
 394         tx_queue->tx_stat.bytes += tx_bytes;
 395
 396         return i;
 397 }
 398
 399 /*
 400  * Callback to handle sending packets through a real NIC.
 401  */
 402 static uint16_t
 403 eth_pcap_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 404 {
 405         unsigned int i;
 406         int ret;
 407         struct rte_mbuf *mbuf;
 408         struct pmd_process_private *pp;
 409         struct pcap_tx_queue *tx_queue = queue;
 410         uint16_t num_tx = 0;
 411         uint32_t tx_bytes = 0;
 412         pcap_t *pcap;
 413         unsigned char temp_data[RTE_ETH_PCAP_SNAPLEN];
 414         size_t len;
 415
 416         pp = rte_eth_devices[tx_queue->port_id].process_private;
 417         pcap = pp->tx_pcap[tx_queue->queue_id];
 418
 419         if (unlikely(nb_pkts == 0 || pcap == NULL))
 420                 return 0;
 421
 422         for (i = 0; i < nb_pkts; i++) {
 423                 mbuf = bufs[i];
 424                 len = rte_pktmbuf_pkt_len(mbuf);
 425                 if (unlikely(!rte_pktmbuf_is_contiguous(mbuf) &&
 426                                 len > sizeof(temp_data))) {
 427                         PMD_LOG(ERR,
 428                                 "Dropping multi segment PCAP packet. Size (%zd) > max size (%zd).",
 429                                 len, sizeof(temp_data));
 430                         rte_pktmbuf_free(mbuf);
 431                         continue;
 432                 }
 433
 434                 /* rte_pktmbuf_read() returns a pointer to the data directly
 435                  * in the mbuf (when the mbuf is contiguous) or, otherwise,
 436                  * a pointer to temp_data after copying into it.
 437                  */
 438                 ret = pcap_sendpacket(pcap,
 439                         rte_pktmbuf_read(mbuf, 0, len, temp_data), len);
 440                 if (unlikely(ret != 0))
 441                         break;
 442                 num_tx++;
 443                 tx_bytes += len;
 444                 rte_pktmbuf_free(mbuf);
 445         }
 446
 447         tx_queue->tx_stat.pkts += num_tx;
 448         tx_queue->tx_stat.bytes += tx_bytes;
 449         tx_queue->tx_stat.err_pkts += i - num_tx;
 450
 451         return i;
 452 }
 453
 454 /*
 455  * pcap_open_live wrapper function
 456  */
 457 static inline int
 458 open_iface_live(const char *iface, pcap_t **pcap) {
 459         *pcap = pcap_open_live(iface, RTE_ETH_PCAP_SNAPLEN,
 460                         RTE_ETH_PCAP_PROMISC, RTE_ETH_PCAP_TIMEOUT, errbuf);
 461
 462         if (*pcap == NULL) {
 463                 PMD_LOG(ERR, "Couldn't open %s: %s", iface, errbuf);
 464                 return -1;
 465         }
 466
 467         return 0;
 468 }
 469
 470 static int
 471 open_single_iface(const char *iface, pcap_t **pcap)
 472 {
 473         if (open_iface_live(iface, pcap) < 0) {
 474                 PMD_LOG(ERR, "Couldn't open interface %s", iface);
 475                 return -1;
 476         }
 477
 478         return 0;
 479 }
 480
 481 static int
 482 open_single_tx_pcap(const char *pcap_filename, pcap_dumper_t **dumper)
 483 {
 484         pcap_t *tx_pcap;
 485
 486         /*
 487          * We need to create a dummy empty pcap_t to use it
 488          * with pcap_dump_open(). We create big enough an Ethernet
 489          * pcap holder.
 490          */
 491         tx_pcap = pcap_open_dead_with_tstamp_precision(DLT_EN10MB,
 492                         RTE_ETH_PCAP_SNAPSHOT_LEN, PCAP_TSTAMP_PRECISION_NANO);
 493         if (tx_pcap == NULL) {
 494                 PMD_LOG(ERR, "Couldn't create dead pcap");
 495                 return -1;
 496         }
 497
 498         /* The dumper is created using the previous pcap_t reference */
 499         *dumper = pcap_dump_open(tx_pcap, pcap_filename);
 500         if (*dumper == NULL) {
 501                 pcap_close(tx_pcap);
 502                 PMD_LOG(ERR, "Couldn't open %s for writing.",
 503                         pcap_filename);
 504                 return -1;
 505         }
 506
 507         pcap_close(tx_pcap);
 508         return 0;
 509 }
 510
 511 static int
 512 open_single_rx_pcap(const char *pcap_filename, pcap_t **pcap)
 513 {
 514         *pcap = pcap_open_offline(pcap_filename, errbuf);
 515         if (*pcap == NULL) {
 516                 PMD_LOG(ERR, "Couldn't open %s: %s", pcap_filename,
 517                         errbuf);
 518                 return -1;
 519         }
 520
 521         return 0;
 522 }
 523
 524 static uint64_t
 525 count_packets_in_pcap(pcap_t **pcap, struct pcap_rx_queue *pcap_q)
 526 {
 527         const u_char *packet;
 528         struct pcap_pkthdr header;
 529         uint64_t pcap_pkt_count = 0;
 530
 531         while ((packet = pcap_next(*pcap, &header)))
 532                 pcap_pkt_count++;
 533
 534         /* The pcap is reopened so it can be used as normal later. */
 535         pcap_close(*pcap);
 536         *pcap = NULL;
 537         open_single_rx_pcap(pcap_q->name, pcap);
 538
 539         return pcap_pkt_count;
 540 }
 541
 542 static int
 543 eth_dev_start(struct rte_eth_dev *dev)
 544 {
 545         unsigned int i;
 546         struct pmd_internals *internals = dev->data->dev_private;
 547         struct pmd_process_private *pp = dev->process_private;
 548         struct pcap_tx_queue *tx;
 549         struct pcap_rx_queue *rx;
 550
 551         /* Special iface case. Single pcap is open and shared between tx/rx. */
 552         if (internals->single_iface) {
 553                 tx = &internals->tx_queue[0];
 554                 rx = &internals->rx_queue[0];
 555
 556                 if (!pp->tx_pcap[0] &&
 557                         strcmp(tx->type, ETH_PCAP_IFACE_ARG) == 0) {
 558                         if (open_single_iface(tx->name, &pp->tx_pcap[0]) < 0)
 559                                 return -1;
 560                         pp->rx_pcap[0] = pp->tx_pcap[0];
 561                 }
 562
 563                 goto status_up;
 564         }
 565
 566         /* If not open already, open tx pcaps/dumpers */
 567         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 568                 tx = &internals->tx_queue[i];
 569
 570                 if (!pp->tx_dumper[i] &&
 571                                 strcmp(tx->type, ETH_PCAP_TX_PCAP_ARG) == 0) {
 572                         if (open_single_tx_pcap(tx->name,
 573                                 &pp->tx_dumper[i]) < 0)
 574                                 return -1;
 575                 } else if (!pp->tx_pcap[i] &&
 576                                 strcmp(tx->type, ETH_PCAP_TX_IFACE_ARG) == 0) {
 577                         if (open_single_iface(tx->name, &pp->tx_pcap[i]) < 0)
 578                                 return -1;
 579                 }
 580         }
 581
 582         /* If not open already, open rx pcaps */
 583         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 584                 rx = &internals->rx_queue[i];
 585
 586                 if (pp->rx_pcap[i] != NULL)
 587                         continue;
 588
 589                 if (strcmp(rx->type, ETH_PCAP_RX_PCAP_ARG) == 0) {
 590                         if (open_single_rx_pcap(rx->name, &pp->rx_pcap[i]) < 0)
 591                                 return -1;
 592                 } else if (strcmp(rx->type, ETH_PCAP_RX_IFACE_ARG) == 0) {
 593                         if (open_single_iface(rx->name, &pp->rx_pcap[i]) < 0)
 594                                 return -1;
 595                 }
 596         }
 597
 598 status_up:
 599         for (i = 0; i < dev->data->nb_rx_queues; i++)
 600                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 601
 602         for (i = 0; i < dev->data->nb_tx_queues; i++)
 603                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 604
 605         dev->data->dev_link.link_status = ETH_LINK_UP;
 606
 607         return 0;
 608 }
 609
 610 /*
 611  * This function gets called when the current port gets stopped.
 612  * Is the only place for us to close all the tx streams dumpers.
 613  * If not called the dumpers will be flushed within each tx burst.
 614  */
 615 static int
 616 eth_dev_stop(struct rte_eth_dev *dev)
 617 {
 618         unsigned int i;
 619         struct pmd_internals *internals = dev->data->dev_private;
 620         struct pmd_process_private *pp = dev->process_private;
 621
 622         /* Special iface case. Single pcap is open and shared between tx/rx. */
 623         if (internals->single_iface) {
 624                 pcap_close(pp->tx_pcap[0]);
 625                 pp->tx_pcap[0] = NULL;
 626                 pp->rx_pcap[0] = NULL;
 627                 goto status_down;
 628         }
 629
 630         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 631                 if (pp->tx_dumper[i] != NULL) {
 632                         pcap_dump_close(pp->tx_dumper[i]);
 633                         pp->tx_dumper[i] = NULL;
 634                 }
 635
 636                 if (pp->tx_pcap[i] != NULL) {
 637                         pcap_close(pp->tx_pcap[i]);
 638                         pp->tx_pcap[i] = NULL;
 639                 }
 640         }
 641
 642         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 643                 if (pp->rx_pcap[i] != NULL) {
 644                         pcap_close(pp->rx_pcap[i]);
 645                         pp->rx_pcap[i] = NULL;
 646                 }
 647         }
 648
 649 status_down:
 650         for (i = 0; i < dev->data->nb_rx_queues; i++)
 651                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 652
 653         for (i = 0; i < dev->data->nb_tx_queues; i++)
 654                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 655
 656         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 657
 658         return 0;
 659 }
 660
 661 static int
 662 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 663 {
 664         return 0;
 665 }
 666
 667 static int
 668 eth_dev_info(struct rte_eth_dev *dev,
 669                 struct rte_eth_dev_info *dev_info)
 670 {
 671         struct pmd_internals *internals = dev->data->dev_private;
 672
 673         dev_info->if_index = internals->if_index;
 674         dev_info->max_mac_addrs = 1;
 675         dev_info->max_rx_pktlen = (uint32_t) -1;
 676         dev_info->max_rx_queues = dev->data->nb_rx_queues;
 677         dev_info->max_tx_queues = dev->data->nb_tx_queues;
 678         dev_info->min_rx_bufsize = 0;
 679
 680         return 0;
 681 }
 682
 683 static int
 684 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 685 {
 686         unsigned int i;
 687         unsigned long rx_packets_total = 0, rx_bytes_total = 0;
 688         unsigned long tx_packets_total = 0, tx_bytes_total = 0;
 689         unsigned long tx_packets_err_total = 0;
 690         const struct pmd_internals *internal = dev->data->dev_private;
 691
 692         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 693                         i < dev->data->nb_rx_queues; i++) {
 694                 stats->q_ipackets[i] = internal->rx_queue[i].rx_stat.pkts;
 695                 stats->q_ibytes[i] = internal->rx_queue[i].rx_stat.bytes;
 696                 rx_packets_total += stats->q_ipackets[i];
 697                 rx_bytes_total += stats->q_ibytes[i];
 698         }
 699
 700         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
 701                         i < dev->data->nb_tx_queues; i++) {
 702                 stats->q_opackets[i] = internal->tx_queue[i].tx_stat.pkts;
 703                 stats->q_obytes[i] = internal->tx_queue[i].tx_stat.bytes;
 704                 tx_packets_total += stats->q_opackets[i];
 705                 tx_bytes_total += stats->q_obytes[i];
 706                 tx_packets_err_total += internal->tx_queue[i].tx_stat.err_pkts;
 707         }
 708
 709         stats->ipackets = rx_packets_total;
 710         stats->ibytes = rx_bytes_total;
 711         stats->opackets = tx_packets_total;
 712         stats->obytes = tx_bytes_total;
 713         stats->oerrors = tx_packets_err_total;
 714
 715         return 0;
 716 }
 717
 718 static int
 719 eth_stats_reset(struct rte_eth_dev *dev)
 720 {
 721         unsigned int i;
 722         struct pmd_internals *internal = dev->data->dev_private;
 723
 724         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 725                 internal->rx_queue[i].rx_stat.pkts = 0;
 726                 internal->rx_queue[i].rx_stat.bytes = 0;
 727         }
 728
 729         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 730                 internal->tx_queue[i].tx_stat.pkts = 0;
 731                 internal->tx_queue[i].tx_stat.bytes = 0;
 732                 internal->tx_queue[i].tx_stat.err_pkts = 0;
 733         }
 734
 735         return 0;
 736 }
 737
 738 static inline void
 739 infinite_rx_ring_free(struct rte_ring *pkts)
 740 {
 741         struct rte_mbuf *bufs;
 742
 743         while (!rte_ring_dequeue(pkts, (void **)&bufs))
 744                 rte_pktmbuf_free(bufs);
 745
 746         rte_ring_free(pkts);
 747 }
 748
 749 static int
 750 eth_dev_close(struct rte_eth_dev *dev)
 751 {
 752         unsigned int i;
 753         struct pmd_internals *internals = dev->data->dev_private;
 754
 755         PMD_LOG(INFO, "Closing pcap ethdev on NUMA socket %d",
 756                         rte_socket_id());
 757
 758         rte_free(dev->process_private);
 759
 760         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 761                 return 0;
 762
 763         /* Device wide flag, but cleanup must be performed per queue. */
 764         if (internals->infinite_rx) {
 765                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
 766                         struct pcap_rx_queue *pcap_q = &internals->rx_queue[i];
 767
 768                         /*
 769                          * 'pcap_q->pkts' can be NULL if 'eth_dev_close()'
 770                          * called before 'eth_rx_queue_setup()' has been called
 771                          */
 772                         if (pcap_q->pkts == NULL)
 773                                 continue;
 774
 775                         infinite_rx_ring_free(pcap_q->pkts);
 776                 }
 777         }
 778
 779         if (internals->phy_mac == 0)
 780                 /* not dynamically allocated, must not be freed */
 781                 dev->data->mac_addrs = NULL;
 782
 783         return 0;
 784 }
 785
 786 static void
 787 eth_queue_release(void *q __rte_unused)
 788 {
 789 }
 790
 791 static int
 792 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 793                 int wait_to_complete __rte_unused)
 794 {
 795         return 0;
 796 }
 797
 798 static int
 799 eth_rx_queue_setup(struct rte_eth_dev *dev,
 800                 uint16_t rx_queue_id,
 801                 uint16_t nb_rx_desc __rte_unused,
 802                 unsigned int socket_id __rte_unused,
 803                 const struct rte_eth_rxconf *rx_conf __rte_unused,
 804                 struct rte_mempool *mb_pool)
 805 {
 806         struct pmd_internals *internals = dev->data->dev_private;
 807         struct pcap_rx_queue *pcap_q = &internals->rx_queue[rx_queue_id];
 808
 809         pcap_q->mb_pool = mb_pool;
 810         pcap_q->port_id = dev->data->port_id;
 811         pcap_q->queue_id = rx_queue_id;
 812         dev->data->rx_queues[rx_queue_id] = pcap_q;
 813
 814         if (internals->infinite_rx) {
 815                 struct pmd_process_private *pp;
 816                 char ring_name[NAME_MAX];
 817                 static uint32_t ring_number;
 818                 uint64_t pcap_pkt_count = 0;
 819                 struct rte_mbuf *bufs[1];
 820                 pcap_t **pcap;
 821
 822                 pp = rte_eth_devices[pcap_q->port_id].process_private;
 823                 pcap = &pp->rx_pcap[pcap_q->queue_id];
 824
 825                 if (unlikely(*pcap == NULL))
 826                         return -ENOENT;
 827
 828                 pcap_pkt_count = count_packets_in_pcap(pcap, pcap_q);
 829
 830                 snprintf(ring_name, sizeof(ring_name), "PCAP_RING%" PRIu16,
 831                                 ring_number);
 832
 833                 pcap_q->pkts = rte_ring_create(ring_name,
 834                                 rte_align64pow2(pcap_pkt_count + 1), 0,
 835                                 RING_F_SP_ENQ | RING_F_SC_DEQ);
 836                 ring_number++;
 837                 if (!pcap_q->pkts)
 838                         return -ENOENT;
 839
 840                 /* Fill ring with packets from PCAP file one by one. */
 841                 while (eth_pcap_rx(pcap_q, bufs, 1)) {
 842                         /* Check for multiseg mbufs. */
 843                         if (bufs[0]->nb_segs != 1) {
 844                                 infinite_rx_ring_free(pcap_q->pkts);
 845                                 PMD_LOG(ERR,
 846                                         "Multiseg mbufs are not supported in infinite_rx mode.");
 847                                 return -EINVAL;
 848                         }
 849
 850                         rte_ring_enqueue_bulk(pcap_q->pkts,
 851                                         (void * const *)bufs, 1, NULL);
 852                 }
 853
 854                 if (rte_ring_count(pcap_q->pkts) < pcap_pkt_count) {
 855                         infinite_rx_ring_free(pcap_q->pkts);
 856                         PMD_LOG(ERR,
 857                                 "Not enough mbufs to accommodate packets in pcap file. "
 858                                 "At least %" PRIu64 " mbufs per queue is required.",
 859                                 pcap_pkt_count);
 860                         return -EINVAL;
 861                 }
 862
 863                 /*
 864                  * Reset the stats for this queue since eth_pcap_rx calls above
 865                  * didn't result in the application receiving packets.
 866                  */
 867                 pcap_q->rx_stat.pkts = 0;
 868                 pcap_q->rx_stat.bytes = 0;
 869         }
 870
 871         return 0;
 872 }
 873
 874 static int
 875 eth_tx_queue_setup(struct rte_eth_dev *dev,
 876                 uint16_t tx_queue_id,
 877                 uint16_t nb_tx_desc __rte_unused,
 878                 unsigned int socket_id __rte_unused,
 879                 const struct rte_eth_txconf *tx_conf __rte_unused)
 880 {
 881         struct pmd_internals *internals = dev->data->dev_private;
 882         struct pcap_tx_queue *pcap_q = &internals->tx_queue[tx_queue_id];
 883
 884         pcap_q->port_id = dev->data->port_id;
 885         pcap_q->queue_id = tx_queue_id;
 886         dev->data->tx_queues[tx_queue_id] = pcap_q;
 887
 888         return 0;
 889 }
 890
 891 static int
 892 eth_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 893 {
 894         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
 895
 896         return 0;
 897 }
 898
 899 static int
 900 eth_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
 901 {
 902         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
 903
 904         return 0;
 905 }
 906
 907 static int
 908 eth_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 909 {
 910         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
 911
 912         return 0;
 913 }
 914
 915 static int
 916 eth_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
 917 {
 918         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
 919
 920         return 0;
 921 }
 922
 923 static const struct eth_dev_ops ops = {
 924         .dev_start = eth_dev_start,
 925         .dev_stop = eth_dev_stop,
 926         .dev_close = eth_dev_close,
 927         .dev_configure = eth_dev_configure,
 928         .dev_infos_get = eth_dev_info,
 929         .rx_queue_setup = eth_rx_queue_setup,
 930         .tx_queue_setup = eth_tx_queue_setup,
 931         .rx_queue_start = eth_rx_queue_start,
 932         .tx_queue_start = eth_tx_queue_start,
 933         .rx_queue_stop = eth_rx_queue_stop,
 934         .tx_queue_stop = eth_tx_queue_stop,
 935         .rx_queue_release = eth_queue_release,
 936         .tx_queue_release = eth_queue_release,
 937         .link_update = eth_link_update,
 938         .stats_get = eth_stats_get,
 939         .stats_reset = eth_stats_reset,
 940 };
 941
 942 static int
 943 add_queue(struct pmd_devargs *pmd, const char *name, const char *type,
 944                 pcap_t *pcap, pcap_dumper_t *dumper)
 945 {
 946         if (pmd->num_of_queue >= RTE_PMD_PCAP_MAX_QUEUES)
 947                 return -1;
 948         if (pcap)
 949                 pmd->queue[pmd->num_of_queue].pcap = pcap;
 950         if (dumper)
 951                 pmd->queue[pmd->num_of_queue].dumper = dumper;
 952         pmd->queue[pmd->num_of_queue].name = name;
 953         pmd->queue[pmd->num_of_queue].type = type;
 954         pmd->num_of_queue++;
 955         return 0;
 956 }
 957
 958 /*
 959  * Function handler that opens the pcap file for reading a stores a
 960  * reference of it for use it later on.
 961  */
 962 static int
 963 open_rx_pcap(const char *key, const char *value, void *extra_args)
 964 {
 965         const char *pcap_filename = value;
 966         struct pmd_devargs *rx = extra_args;
 967         pcap_t *pcap = NULL;
 968
 969         if (open_single_rx_pcap(pcap_filename, &pcap) < 0)
 970                 return -1;
 971
 972         if (add_queue(rx, pcap_filename, key, pcap, NULL) < 0) {
 973                 pcap_close(pcap);
 974                 return -1;
 975         }
 976
 977         return 0;
 978 }
 979
 980 /*
 981  * Opens a pcap file for writing and stores a reference to it
 982  * for use it later on.
 983  */
 984 static int
 985 open_tx_pcap(const char *key, const char *value, void *extra_args)
 986 {
 987         const char *pcap_filename = value;
 988         struct pmd_devargs *dumpers = extra_args;
 989         pcap_dumper_t *dumper;
 990
 991         if (open_single_tx_pcap(pcap_filename, &dumper) < 0)
 992                 return -1;
 993
 994         if (add_queue(dumpers, pcap_filename, key, NULL, dumper) < 0) {
 995                 pcap_dump_close(dumper);
 996                 return -1;
 997         }
 998
 999         return 0;
1000 }
1001
1002 /*
1003  * Opens an interface for reading and writing
1004  */
1005 static inline int
1006 open_rx_tx_iface(const char *key, const char *value, void *extra_args)
1007 {
1008         const char *iface = value;
1009         struct pmd_devargs *tx = extra_args;
1010         pcap_t *pcap = NULL;
1011
1012         if (open_single_iface(iface, &pcap) < 0)
1013                 return -1;
1014
1015         tx->queue[0].pcap = pcap;
1016         tx->queue[0].name = iface;
1017         tx->queue[0].type = key;
1018
1019         return 0;
1020 }
1021
1022 static inline int
1023 set_iface_direction(const char *iface, pcap_t *pcap,
1024                 pcap_direction_t direction)
1025 {
1026         const char *direction_str = (direction == PCAP_D_IN) ? "IN" : "OUT";
1027         if (pcap_setdirection(pcap, direction) < 0) {
1028                 PMD_LOG(ERR, "Setting %s pcap direction %s failed - %s\n",
1029                                 iface, direction_str, pcap_geterr(pcap));
1030                 return -1;
1031         }
1032         PMD_LOG(INFO, "Setting %s pcap direction %s\n",
1033                         iface, direction_str);
1034         return 0;
1035 }
1036
1037 static inline int
1038 open_iface(const char *key, const char *value, void *extra_args)
1039 {
1040         const char *iface = value;
1041         struct pmd_devargs *pmd = extra_args;
1042         pcap_t *pcap = NULL;
1043
1044         if (open_single_iface(iface, &pcap) < 0)
1045                 return -1;
1046         if (add_queue(pmd, iface, key, pcap, NULL) < 0) {
1047                 pcap_close(pcap);
1048                 return -1;
1049         }
1050
1051         return 0;
1052 }
1053
1054 /*
1055  * Opens a NIC for reading packets from it
1056  */
1057 static inline int
1058 open_rx_iface(const char *key, const char *value, void *extra_args)
1059 {
1060         int ret = open_iface(key, value, extra_args);
1061         if (ret < 0)
1062                 return ret;
1063         if (strcmp(key, ETH_PCAP_RX_IFACE_IN_ARG) == 0) {
1064                 struct pmd_devargs *pmd = extra_args;
1065                 unsigned int qid = pmd->num_of_queue - 1;
1066
1067                 set_iface_direction(pmd->queue[qid].name,
1068                                 pmd->queue[qid].pcap,
1069                                 PCAP_D_IN);
1070         }
1071
1072         return 0;
1073 }
1074
1075 static inline int
1076 rx_iface_args_process(const char *key, const char *value, void *extra_args)
1077 {
1078         if (strcmp(key, ETH_PCAP_RX_IFACE_ARG) == 0 ||
1079                         strcmp(key, ETH_PCAP_RX_IFACE_IN_ARG) == 0)
1080                 return open_rx_iface(key, value, extra_args);
1081
1082         return 0;
1083 }
1084
1085 /*
1086  * Opens a NIC for writing packets to it
1087  */
1088 static int
1089 open_tx_iface(const char *key, const char *value, void *extra_args)
1090 {
1091         return open_iface(key, value, extra_args);
1092 }
1093
1094 static int
1095 select_phy_mac(const char *key __rte_unused, const char *value,
1096                 void *extra_args)
1097 {
1098         if (extra_args) {
1099                 const int phy_mac = atoi(value);
1100                 int *enable_phy_mac = extra_args;
1101
1102                 if (phy_mac)
1103                         *enable_phy_mac = 1;
1104         }
1105         return 0;
1106 }
1107
1108 static int
1109 get_infinite_rx_arg(const char *key __rte_unused,
1110                 const char *value, void *extra_args)
1111 {
1112         if (extra_args) {
1113                 const int infinite_rx = atoi(value);
1114                 int *enable_infinite_rx = extra_args;
1115
1116                 if (infinite_rx > 0)
1117                         *enable_infinite_rx = 1;
1118         }
1119         return 0;
1120 }
1121
1122 static int
1123 pmd_init_internals(struct rte_vdev_device *vdev,
1124                 const unsigned int nb_rx_queues,
1125                 const unsigned int nb_tx_queues,
1126                 struct pmd_internals **internals,
1127                 struct rte_eth_dev **eth_dev)
1128 {
1129         struct rte_eth_dev_data *data;
1130         struct pmd_process_private *pp;
1131         unsigned int numa_node = vdev->device.numa_node;
1132
1133         PMD_LOG(INFO, "Creating pcap-backed ethdev on numa socket %d",
1134                 numa_node);
1135
1136         pp = (struct pmd_process_private *)
1137                 rte_zmalloc(NULL, sizeof(struct pmd_process_private),
1138                                 RTE_CACHE_LINE_SIZE);
1139
1140         if (pp == NULL) {
1141                 PMD_LOG(ERR,
1142                         "Failed to allocate memory for process private");
1143                 return -1;
1144         }
1145
1146         /* reserve an ethdev entry */
1147         *eth_dev = rte_eth_vdev_allocate(vdev, sizeof(**internals));
1148         if (!(*eth_dev)) {
1149                 rte_free(pp);
1150                 return -1;
1151         }
1152         (*eth_dev)->process_private = pp;
1153         /* now put it all together
1154          * - store queue data in internals,
1155          * - store numa_node info in eth_dev
1156          * - point eth_dev_data to internals
1157          * - and point eth_dev structure to new eth_dev_data structure
1158          */
1159         *internals = (*eth_dev)->data->dev_private;
1160         /*
1161          * Interface MAC = 02:70:63:61:70:<iface_idx>
1162          * derived from: 'locally administered':'p':'c':'a':'p':'iface_idx'
1163          * where the middle 4 characters are converted to hex.
1164          */
1165         (*internals)->eth_addr = (struct rte_ether_addr) {
1166                 .addr_bytes = { 0x02, 0x70, 0x63, 0x61, 0x70, iface_idx++ }
1167         };
1168         (*internals)->phy_mac = 0;
1169         data = (*eth_dev)->data;
1170         data->nb_rx_queues = (uint16_t)nb_rx_queues;
1171         data->nb_tx_queues = (uint16_t)nb_tx_queues;
1172         data->dev_link = pmd_link;
1173         data->mac_addrs = &(*internals)->eth_addr;
1174         data->promiscuous = 1;
1175         data->all_multicast = 1;
1176         data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
1177
1178         /*
1179          * NOTE: we'll replace the data element, of originally allocated
1180          * eth_dev so the rings are local per-process
1181          */
1182         (*eth_dev)->dev_ops = &ops;
1183
1184         strlcpy((*internals)->devargs, rte_vdev_device_args(vdev),
1185                         ETH_PCAP_ARG_MAXLEN);
1186
1187         return 0;
1188 }
1189
1190 static int
1191 eth_pcap_update_mac(const char *if_name, struct rte_eth_dev *eth_dev,
1192                 const unsigned int numa_node)
1193 {
1194 #if defined(RTE_EXEC_ENV_LINUX)
1195         void *mac_addrs;
1196         struct ifreq ifr;
1197         int if_fd = socket(AF_INET, SOCK_DGRAM, 0);
1198
1199         if (if_fd == -1)
1200                 return -1;
1201
1202         rte_strscpy(ifr.ifr_name, if_name, sizeof(ifr.ifr_name));
1203         if (ioctl(if_fd, SIOCGIFHWADDR, &ifr)) {
1204                 close(if_fd);
1205                 return -1;
1206         }
1207
1208         mac_addrs = rte_zmalloc_socket(NULL, RTE_ETHER_ADDR_LEN, 0, numa_node);
1209         if (!mac_addrs) {
1210                 close(if_fd);
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                         ifr.ifr_hwaddr.sa_data, RTE_ETHER_ADDR_LEN);
1218
1219         close(if_fd);
1220
1221         return 0;
1222
1223 #elif defined(RTE_EXEC_ENV_FREEBSD)
1224         void *mac_addrs;
1225         struct if_msghdr *ifm;
1226         struct sockaddr_dl *sdl;
1227         int mib[6];
1228         size_t len = 0;
1229         char *buf;
1230
1231         mib[0] = CTL_NET;
1232         mib[1] = AF_ROUTE;
1233         mib[2] = 0;
1234         mib[3] = AF_LINK;
1235         mib[4] = NET_RT_IFLIST;
1236         mib[5] = if_nametoindex(if_name);
1237
1238         if (sysctl(mib, 6, NULL, &len, NULL, 0) < 0)
1239                 return -1;
1240
1241         if (len == 0)
1242                 return -1;
1243
1244         buf = rte_malloc(NULL, len, 0);
1245         if (!buf)
1246                 return -1;
1247
1248         if (sysctl(mib, 6, buf, &len, NULL, 0) < 0) {
1249                 rte_free(buf);
1250                 return -1;
1251         }
1252         ifm = (struct if_msghdr *)buf;
1253         sdl = (struct sockaddr_dl *)(ifm + 1);
1254
1255         mac_addrs = rte_zmalloc_socket(NULL, RTE_ETHER_ADDR_LEN, 0, numa_node);
1256         if (!mac_addrs) {
1257                 rte_free(buf);
1258                 return -1;
1259         }
1260
1261         PMD_LOG(INFO, "Setting phy MAC for %s", if_name);
1262         eth_dev->data->mac_addrs = mac_addrs;
1263         rte_memcpy(eth_dev->data->mac_addrs[0].addr_bytes,
1264                         LLADDR(sdl), RTE_ETHER_ADDR_LEN);
1265
1266         rte_free(buf);
1267
1268         return 0;
1269 #else
1270         return -1;
1271 #endif
1272 }
1273
1274 static int
1275 eth_from_pcaps_common(struct rte_vdev_device *vdev,
1276                 struct pmd_devargs_all *devargs_all,
1277                 struct pmd_internals **internals, struct rte_eth_dev **eth_dev)
1278 {
1279         struct pmd_process_private *pp;
1280         struct pmd_devargs *rx_queues = &devargs_all->rx_queues;
1281         struct pmd_devargs *tx_queues = &devargs_all->tx_queues;
1282         const unsigned int nb_rx_queues = rx_queues->num_of_queue;
1283         const unsigned int nb_tx_queues = tx_queues->num_of_queue;
1284         unsigned int i;
1285
1286         if (pmd_init_internals(vdev, nb_rx_queues, nb_tx_queues, internals,
1287                         eth_dev) < 0)
1288                 return -1;
1289
1290         pp = (*eth_dev)->process_private;
1291         for (i = 0; i < nb_rx_queues; i++) {
1292                 struct pcap_rx_queue *rx = &(*internals)->rx_queue[i];
1293                 struct devargs_queue *queue = &rx_queues->queue[i];
1294
1295                 pp->rx_pcap[i] = queue->pcap;
1296                 strlcpy(rx->name, queue->name, sizeof(rx->name));
1297                 strlcpy(rx->type, queue->type, sizeof(rx->type));
1298         }
1299
1300         for (i = 0; i < nb_tx_queues; i++) {
1301                 struct pcap_tx_queue *tx = &(*internals)->tx_queue[i];
1302                 struct devargs_queue *queue = &tx_queues->queue[i];
1303
1304                 pp->tx_dumper[i] = queue->dumper;
1305                 pp->tx_pcap[i] = queue->pcap;
1306                 strlcpy(tx->name, queue->name, sizeof(tx->name));
1307                 strlcpy(tx->type, queue->type, sizeof(tx->type));
1308         }
1309
1310         return 0;
1311 }
1312
1313 static int
1314 eth_from_pcaps(struct rte_vdev_device *vdev,
1315                 struct pmd_devargs_all *devargs_all)
1316 {
1317         struct pmd_internals *internals = NULL;
1318         struct rte_eth_dev *eth_dev = NULL;
1319         struct pmd_devargs *rx_queues = &devargs_all->rx_queues;
1320         int single_iface = devargs_all->single_iface;
1321         unsigned int infinite_rx = devargs_all->infinite_rx;
1322         int ret;
1323
1324         ret = eth_from_pcaps_common(vdev, devargs_all, &internals, &eth_dev);
1325
1326         if (ret < 0)
1327                 return ret;
1328
1329         /* store weather we are using a single interface for rx/tx or not */
1330         internals->single_iface = single_iface;
1331
1332         if (single_iface) {
1333                 internals->if_index = if_nametoindex(rx_queues->queue[0].name);
1334
1335                 /* phy_mac arg is applied only only if "iface" devarg is provided */
1336                 if (rx_queues->phy_mac) {
1337                         if (eth_pcap_update_mac(rx_queues->queue[0].name,
1338                                         eth_dev, vdev->device.numa_node) == 0)
1339                                 internals->phy_mac = 1;
1340                 }
1341         }
1342
1343         internals->infinite_rx = infinite_rx;
1344         /* Assign rx ops. */
1345         if (infinite_rx)
1346                 eth_dev->rx_pkt_burst = eth_pcap_rx_infinite;
1347         else if (devargs_all->is_rx_pcap || devargs_all->is_rx_iface ||
1348                         single_iface)
1349                 eth_dev->rx_pkt_burst = eth_pcap_rx;
1350         else
1351                 eth_dev->rx_pkt_burst = eth_null_rx;
1352
1353         /* Assign tx ops. */
1354         if (devargs_all->is_tx_pcap)
1355                 eth_dev->tx_pkt_burst = eth_pcap_tx_dumper;
1356         else if (devargs_all->is_tx_iface || single_iface)
1357                 eth_dev->tx_pkt_burst = eth_pcap_tx;
1358         else
1359                 eth_dev->tx_pkt_burst = eth_tx_drop;
1360
1361         rte_eth_dev_probing_finish(eth_dev);
1362         return 0;
1363 }
1364
1365 static int
1366 pmd_pcap_probe(struct rte_vdev_device *dev)
1367 {
1368         const char *name;
1369         struct rte_kvargs *kvlist;
1370         struct pmd_devargs pcaps = {0};
1371         struct pmd_devargs dumpers = {0};
1372         struct rte_eth_dev *eth_dev =  NULL;
1373         struct pmd_internals *internal;
1374         int ret = 0;
1375
1376         struct pmd_devargs_all devargs_all = {
1377                 .single_iface = 0,
1378                 .is_tx_pcap = 0,
1379                 .is_tx_iface = 0,
1380                 .infinite_rx = 0,
1381         };
1382
1383         name = rte_vdev_device_name(dev);
1384         PMD_LOG(INFO, "Initializing pmd_pcap for %s", name);
1385
1386         gettimeofday(&start_time, NULL);
1387         start_cycles = rte_get_timer_cycles();
1388         hz = rte_get_timer_hz();
1389
1390         ret = rte_mbuf_dyn_rx_timestamp_register(&timestamp_dynfield_offset,
1391                         &timestamp_rx_dynflag);
1392         if (ret != 0) {
1393                 PMD_LOG(ERR, "Failed to register Rx timestamp field/flag");
1394                 return -1;
1395         }
1396
1397         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1398                 eth_dev = rte_eth_dev_attach_secondary(name);
1399                 if (!eth_dev) {
1400                         PMD_LOG(ERR, "Failed to probe %s", name);
1401                         return -1;
1402                 }
1403
1404                 internal = eth_dev->data->dev_private;
1405
1406                 kvlist = rte_kvargs_parse(internal->devargs, valid_arguments);
1407                 if (kvlist == NULL)
1408                         return -1;
1409         } else {
1410                 kvlist = rte_kvargs_parse(rte_vdev_device_args(dev),
1411                                 valid_arguments);
1412                 if (kvlist == NULL)
1413                         return -1;
1414         }
1415
1416         /*
1417          * If iface argument is passed we open the NICs and use them for
1418          * reading / writing
1419          */
1420         if (rte_kvargs_count(kvlist, ETH_PCAP_IFACE_ARG) == 1) {
1421
1422                 ret = rte_kvargs_process(kvlist, ETH_PCAP_IFACE_ARG,
1423                                 &open_rx_tx_iface, &pcaps);
1424                 if (ret < 0)
1425                         goto free_kvlist;
1426
1427                 dumpers.queue[0] = pcaps.queue[0];
1428
1429                 ret = rte_kvargs_process(kvlist, ETH_PCAP_PHY_MAC_ARG,
1430                                 &select_phy_mac, &pcaps.phy_mac);
1431                 if (ret < 0)
1432                         goto free_kvlist;
1433
1434                 dumpers.phy_mac = pcaps.phy_mac;
1435
1436                 devargs_all.single_iface = 1;
1437                 pcaps.num_of_queue = 1;
1438                 dumpers.num_of_queue = 1;
1439
1440                 goto create_eth;
1441         }
1442
1443         /*
1444          * We check whether we want to open a RX stream from a real NIC, a
1445          * pcap file or open a dummy RX stream
1446          */
1447         devargs_all.is_rx_pcap =
1448                 rte_kvargs_count(kvlist, ETH_PCAP_RX_PCAP_ARG) ? 1 : 0;
1449         devargs_all.is_rx_iface =
1450                 (rte_kvargs_count(kvlist, ETH_PCAP_RX_IFACE_ARG) +
1451                  rte_kvargs_count(kvlist, ETH_PCAP_RX_IFACE_IN_ARG)) ? 1 : 0;
1452         pcaps.num_of_queue = 0;
1453
1454         devargs_all.is_tx_pcap =
1455                 rte_kvargs_count(kvlist, ETH_PCAP_TX_PCAP_ARG) ? 1 : 0;
1456         devargs_all.is_tx_iface =
1457                 rte_kvargs_count(kvlist, ETH_PCAP_TX_IFACE_ARG) ? 1 : 0;
1458         dumpers.num_of_queue = 0;
1459
1460         if (devargs_all.is_rx_pcap) {
1461                 /*
1462                  * We check whether we want to infinitely rx the pcap file.
1463                  */
1464                 unsigned int infinite_rx_arg_cnt = rte_kvargs_count(kvlist,
1465                                 ETH_PCAP_INFINITE_RX_ARG);
1466
1467                 if (infinite_rx_arg_cnt == 1) {
1468                         ret = rte_kvargs_process(kvlist,
1469                                         ETH_PCAP_INFINITE_RX_ARG,
1470                                         &get_infinite_rx_arg,
1471                                         &devargs_all.infinite_rx);
1472                         if (ret < 0)
1473                                 goto free_kvlist;
1474                         PMD_LOG(INFO, "infinite_rx has been %s for %s",
1475                                         devargs_all.infinite_rx ? "enabled" : "disabled",
1476                                         name);
1477
1478                 } else if (infinite_rx_arg_cnt > 1) {
1479                         PMD_LOG(WARNING, "infinite_rx has not been enabled since the "
1480                                         "argument has been provided more than once "
1481                                         "for %s", name);
1482                 }
1483
1484                 ret = rte_kvargs_process(kvlist, ETH_PCAP_RX_PCAP_ARG,
1485                                 &open_rx_pcap, &pcaps);
1486         } else if (devargs_all.is_rx_iface) {
1487                 ret = rte_kvargs_process(kvlist, NULL,
1488                                 &rx_iface_args_process, &pcaps);
1489         } else if (devargs_all.is_tx_iface || devargs_all.is_tx_pcap) {
1490                 unsigned int i;
1491
1492                 /* Count number of tx queue args passed before dummy rx queue
1493                  * creation so a dummy rx queue can be created for each tx queue
1494                  */
1495                 unsigned int num_tx_queues =
1496                         (rte_kvargs_count(kvlist, ETH_PCAP_TX_PCAP_ARG) +
1497                         rte_kvargs_count(kvlist, ETH_PCAP_TX_IFACE_ARG));
1498
1499                 PMD_LOG(INFO, "Creating null rx queue since no rx queues were provided.");
1500
1501                 /* Creating a dummy rx queue for each tx queue passed */
1502                 for (i = 0; i < num_tx_queues; i++)
1503                         ret = add_queue(&pcaps, "dummy_rx", "rx_null", NULL,
1504                                         NULL);
1505         } else {
1506                 PMD_LOG(ERR, "Error - No rx or tx queues provided");
1507                 ret = -ENOENT;
1508         }
1509         if (ret < 0)
1510                 goto free_kvlist;
1511
1512         /*
1513          * We check whether we want to open a TX stream to a real NIC,
1514          * a pcap file, or drop packets on tx
1515          */
1516         if (devargs_all.is_tx_pcap) {
1517                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_PCAP_ARG,
1518                                 &open_tx_pcap, &dumpers);
1519         } else if (devargs_all.is_tx_iface) {
1520                 ret = rte_kvargs_process(kvlist, ETH_PCAP_TX_IFACE_ARG,
1521                                 &open_tx_iface, &dumpers);
1522         } else {
1523                 unsigned int i;
1524
1525                 PMD_LOG(INFO, "Dropping packets on tx since no tx queues were provided.");
1526
1527                 /* Add 1 dummy queue per rxq which counts and drops packets. */
1528                 for (i = 0; i < pcaps.num_of_queue; i++)
1529                         ret = add_queue(&dumpers, "dummy_tx", "tx_drop", NULL,
1530                                         NULL);
1531         }
1532
1533         if (ret < 0)
1534                 goto free_kvlist;
1535
1536 create_eth:
1537         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1538                 struct pmd_process_private *pp;
1539                 unsigned int i;
1540
1541                 internal = eth_dev->data->dev_private;
1542                         pp = (struct pmd_process_private *)
1543                                 rte_zmalloc(NULL,
1544                                         sizeof(struct pmd_process_private),
1545                                         RTE_CACHE_LINE_SIZE);
1546
1547                 if (pp == NULL) {
1548                         PMD_LOG(ERR,
1549                                 "Failed to allocate memory for process private");
1550                         ret = -1;
1551                         goto free_kvlist;
1552                 }
1553
1554                 eth_dev->dev_ops = &ops;
1555                 eth_dev->device = &dev->device;
1556
1557                 /* setup process private */
1558                 for (i = 0; i < pcaps.num_of_queue; i++)
1559                         pp->rx_pcap[i] = pcaps.queue[i].pcap;
1560
1561                 for (i = 0; i < dumpers.num_of_queue; i++) {
1562                         pp->tx_dumper[i] = dumpers.queue[i].dumper;
1563                         pp->tx_pcap[i] = dumpers.queue[i].pcap;
1564                 }
1565
1566                 eth_dev->process_private = pp;
1567                 eth_dev->rx_pkt_burst = eth_pcap_rx;
1568                 if (devargs_all.is_tx_pcap)
1569                         eth_dev->tx_pkt_burst = eth_pcap_tx_dumper;
1570                 else
1571                         eth_dev->tx_pkt_burst = eth_pcap_tx;
1572
1573                 rte_eth_dev_probing_finish(eth_dev);
1574                 goto free_kvlist;
1575         }
1576
1577         devargs_all.rx_queues = pcaps;
1578         devargs_all.tx_queues = dumpers;
1579
1580         ret = eth_from_pcaps(dev, &devargs_all);
1581
1582 free_kvlist:
1583         rte_kvargs_free(kvlist);
1584
1585         return ret;
1586 }
1587
1588 static int
1589 pmd_pcap_remove(struct rte_vdev_device *dev)
1590 {
1591         struct rte_eth_dev *eth_dev = NULL;
1592
1593         if (!dev)
1594                 return -1;
1595
1596         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1597         if (eth_dev == NULL)
1598                 return 0; /* port already released */
1599
1600         eth_dev_close(eth_dev);
1601         rte_eth_dev_release_port(eth_dev);
1602
1603         return 0;
1604 }
1605
1606 static struct rte_vdev_driver pmd_pcap_drv = {
1607         .probe = pmd_pcap_probe,
1608         .remove = pmd_pcap_remove,
1609 };
1610
1611 RTE_PMD_REGISTER_VDEV(net_pcap, pmd_pcap_drv);
1612 RTE_PMD_REGISTER_ALIAS(net_pcap, eth_pcap);
1613 RTE_PMD_REGISTER_PARAM_STRING(net_pcap,
1614         ETH_PCAP_RX_PCAP_ARG "=<string> "
1615         ETH_PCAP_TX_PCAP_ARG "=<string> "
1616         ETH_PCAP_RX_IFACE_ARG "=<ifc> "
1617         ETH_PCAP_RX_IFACE_IN_ARG "=<ifc> "
1618         ETH_PCAP_TX_IFACE_ARG "=<ifc> "
1619         ETH_PCAP_IFACE_ARG "=<ifc> "
1620         ETH_PCAP_PHY_MAC_ARG "=<int>"
1621         ETH_PCAP_INFINITE_RX_ARG "=<0|1>");