drivers/net/af_packet/rte_eth_af_packet.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2014 John W. Linville <linville@tuxdriver.com>
   3  * Originally based upon librte_pmd_pcap code:
   4  * Copyright(c) 2010-2015 Intel Corporation.
   5  * Copyright(c) 2014 6WIND S.A.
   6  * All rights reserved.
   7  */
   8
   9 #include <rte_string_fns.h>
  10 #include <rte_mbuf.h>
  11 #include <rte_ethdev_driver.h>
  12 #include <rte_ethdev_vdev.h>
  13 #include <rte_malloc.h>
  14 #include <rte_kvargs.h>
  15 #include <rte_bus_vdev.h>
  16
  17 #include <errno.h>
  18 #include <linux/if_ether.h>
  19 #include <linux/if_packet.h>
  20 #include <arpa/inet.h>
  21 #include <net/if.h>
  22 #include <sys/types.h>
  23 #include <sys/socket.h>
  24 #include <sys/ioctl.h>
  25 #include <string.h>
  26 #include <sys/mman.h>
  27 #include <unistd.h>
  28 #include <poll.h>
  29
  30 #define ETH_AF_PACKET_IFACE_ARG         "iface"
  31 #define ETH_AF_PACKET_NUM_Q_ARG         "qpairs"
  32 #define ETH_AF_PACKET_BLOCKSIZE_ARG     "blocksz"
  33 #define ETH_AF_PACKET_FRAMESIZE_ARG     "framesz"
  34 #define ETH_AF_PACKET_FRAMECOUNT_ARG    "framecnt"
  35 #define ETH_AF_PACKET_QDISC_BYPASS_ARG  "qdisc_bypass"
  36
  37 #define DFLT_FRAME_SIZE         (1 << 11)
  38 #define DFLT_FRAME_COUNT        (1 << 9)
  39
  40 #define RTE_PMD_AF_PACKET_MAX_RINGS 16
  41
  42 struct pkt_rx_queue {
  43         int sockfd;
  44
  45         struct iovec *rd;
  46         uint8_t *map;
  47         unsigned int framecount;
  48         unsigned int framenum;
  49
  50         struct rte_mempool *mb_pool;
  51         uint16_t in_port;
  52
  53         volatile unsigned long rx_pkts;
  54         volatile unsigned long rx_bytes;
  55 };
  56
  57 struct pkt_tx_queue {
  58         int sockfd;
  59         unsigned int frame_data_size;
  60
  61         struct iovec *rd;
  62         uint8_t *map;
  63         unsigned int framecount;
  64         unsigned int framenum;
  65
  66         volatile unsigned long tx_pkts;
  67         volatile unsigned long err_pkts;
  68         volatile unsigned long tx_bytes;
  69 };
  70
  71 struct pmd_internals {
  72         unsigned nb_queues;
  73
  74         int if_index;
  75         char *if_name;
  76         struct rte_ether_addr eth_addr;
  77
  78         struct tpacket_req req;
  79
  80         struct pkt_rx_queue rx_queue[RTE_PMD_AF_PACKET_MAX_RINGS];
  81         struct pkt_tx_queue tx_queue[RTE_PMD_AF_PACKET_MAX_RINGS];
  82 };
  83
  84 static const char *valid_arguments[] = {
  85         ETH_AF_PACKET_IFACE_ARG,
  86         ETH_AF_PACKET_NUM_Q_ARG,
  87         ETH_AF_PACKET_BLOCKSIZE_ARG,
  88         ETH_AF_PACKET_FRAMESIZE_ARG,
  89         ETH_AF_PACKET_FRAMECOUNT_ARG,
  90         ETH_AF_PACKET_QDISC_BYPASS_ARG,
  91         NULL
  92 };
  93
  94 static struct rte_eth_link pmd_link = {
  95         .link_speed = ETH_SPEED_NUM_10G,
  96         .link_duplex = ETH_LINK_FULL_DUPLEX,
  97         .link_status = ETH_LINK_DOWN,
  98         .link_autoneg = ETH_LINK_FIXED,
  99 };
 100
 101 static int af_packet_logtype;
 102
 103 #define PMD_LOG(level, fmt, args...) \
 104         rte_log(RTE_LOG_ ## level, af_packet_logtype, \
 105                 "%s(): " fmt "\n", __func__, ##args)
 106
 107 #define PMD_LOG_ERRNO(level, fmt, args...) \
 108         rte_log(RTE_LOG_ ## level, af_packet_logtype, \
 109                 "%s(): " fmt ":%s\n", __func__, ##args, strerror(errno))
 110
 111 static uint16_t
 112 eth_af_packet_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 113 {
 114         unsigned i;
 115         struct tpacket2_hdr *ppd;
 116         struct rte_mbuf *mbuf;
 117         uint8_t *pbuf;
 118         struct pkt_rx_queue *pkt_q = queue;
 119         uint16_t num_rx = 0;
 120         unsigned long num_rx_bytes = 0;
 121         unsigned int framecount, framenum;
 122
 123         if (unlikely(nb_pkts == 0))
 124                 return 0;
 125
 126         /*
 127          * Reads the given number of packets from the AF_PACKET socket one by
 128          * one and copies the packet data into a newly allocated mbuf.
 129          */
 130         framecount = pkt_q->framecount;
 131         framenum = pkt_q->framenum;
 132         for (i = 0; i < nb_pkts; i++) {
 133                 /* point at the next incoming frame */
 134                 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
 135                 if ((ppd->tp_status & TP_STATUS_USER) == 0)
 136                         break;
 137
 138                 /* allocate the next mbuf */
 139                 mbuf = rte_pktmbuf_alloc(pkt_q->mb_pool);
 140                 if (unlikely(mbuf == NULL))
 141                         break;
 142
 143                 /* packet will fit in the mbuf, go ahead and receive it */
 144                 rte_pktmbuf_pkt_len(mbuf) = rte_pktmbuf_data_len(mbuf) = ppd->tp_snaplen;
 145                 pbuf = (uint8_t *) ppd + ppd->tp_mac;
 146                 memcpy(rte_pktmbuf_mtod(mbuf, void *), pbuf, rte_pktmbuf_data_len(mbuf));
 147
 148                 /* check for vlan info */
 149                 if (ppd->tp_status & TP_STATUS_VLAN_VALID) {
 150                         mbuf->vlan_tci = ppd->tp_vlan_tci;
 151                         mbuf->ol_flags |= (PKT_RX_VLAN | PKT_RX_VLAN_STRIPPED);
 152                 }
 153
 154                 /* release incoming frame and advance ring buffer */
 155                 ppd->tp_status = TP_STATUS_KERNEL;
 156                 if (++framenum >= framecount)
 157                         framenum = 0;
 158                 mbuf->port = pkt_q->in_port;
 159
 160                 /* account for the receive frame */
 161                 bufs[i] = mbuf;
 162                 num_rx++;
 163                 num_rx_bytes += mbuf->pkt_len;
 164         }
 165         pkt_q->framenum = framenum;
 166         pkt_q->rx_pkts += num_rx;
 167         pkt_q->rx_bytes += num_rx_bytes;
 168         return num_rx;
 169 }
 170
 171 /*
 172  * Callback to handle sending packets through a real NIC.
 173  */
 174 static uint16_t
 175 eth_af_packet_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 176 {
 177         struct tpacket2_hdr *ppd;
 178         struct rte_mbuf *mbuf;
 179         uint8_t *pbuf;
 180         unsigned int framecount, framenum;
 181         struct pollfd pfd;
 182         struct pkt_tx_queue *pkt_q = queue;
 183         uint16_t num_tx = 0;
 184         unsigned long num_tx_bytes = 0;
 185         int i;
 186
 187         if (unlikely(nb_pkts == 0))
 188                 return 0;
 189
 190         memset(&pfd, 0, sizeof(pfd));
 191         pfd.fd = pkt_q->sockfd;
 192         pfd.events = POLLOUT;
 193         pfd.revents = 0;
 194
 195         framecount = pkt_q->framecount;
 196         framenum = pkt_q->framenum;
 197         ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
 198         for (i = 0; i < nb_pkts; i++) {
 199                 mbuf = *bufs++;
 200
 201                 /* drop oversized packets */
 202                 if (mbuf->pkt_len > pkt_q->frame_data_size) {
 203                         rte_pktmbuf_free(mbuf);
 204                         continue;
 205                 }
 206
 207                 /* insert vlan info if necessary */
 208                 if (mbuf->ol_flags & PKT_TX_VLAN_PKT) {
 209                         if (rte_vlan_insert(&mbuf)) {
 210                                 rte_pktmbuf_free(mbuf);
 211                                 continue;
 212                         }
 213                 }
 214
 215                 /* point at the next incoming frame */
 216                 if ((ppd->tp_status != TP_STATUS_AVAILABLE) &&
 217                     (poll(&pfd, 1, -1) < 0))
 218                         break;
 219
 220                 /* copy the tx frame data */
 221                 pbuf = (uint8_t *) ppd + TPACKET2_HDRLEN -
 222                         sizeof(struct sockaddr_ll);
 223
 224                 struct rte_mbuf *tmp_mbuf = mbuf;
 225                 while (tmp_mbuf) {
 226                         uint16_t data_len = rte_pktmbuf_data_len(tmp_mbuf);
 227                         memcpy(pbuf, rte_pktmbuf_mtod(tmp_mbuf, void*), data_len);
 228                         pbuf += data_len;
 229                         tmp_mbuf = tmp_mbuf->next;
 230                 }
 231
 232                 ppd->tp_len = mbuf->pkt_len;
 233                 ppd->tp_snaplen = mbuf->pkt_len;
 234
 235                 /* release incoming frame and advance ring buffer */
 236                 ppd->tp_status = TP_STATUS_SEND_REQUEST;
 237                 if (++framenum >= framecount)
 238                         framenum = 0;
 239                 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
 240
 241                 num_tx++;
 242                 num_tx_bytes += mbuf->pkt_len;
 243                 rte_pktmbuf_free(mbuf);
 244         }
 245
 246         /* kick-off transmits */
 247         if (sendto(pkt_q->sockfd, NULL, 0, MSG_DONTWAIT, NULL, 0) == -1 &&
 248                         errno != ENOBUFS && errno != EAGAIN) {
 249                 /*
 250                  * In case of a ENOBUFS/EAGAIN error all of the enqueued
 251                  * packets will be considered successful even though only some
 252                  * are sent.
 253                  */
 254
 255                 num_tx = 0;
 256                 num_tx_bytes = 0;
 257         }
 258
 259         pkt_q->framenum = framenum;
 260         pkt_q->tx_pkts += num_tx;
 261         pkt_q->err_pkts += i - num_tx;
 262         pkt_q->tx_bytes += num_tx_bytes;
 263         return i;
 264 }
 265
 266 static int
 267 eth_dev_start(struct rte_eth_dev *dev)
 268 {
 269         dev->data->dev_link.link_status = ETH_LINK_UP;
 270         return 0;
 271 }
 272
 273 /*
 274  * This function gets called when the current port gets stopped.
 275  */
 276 static void
 277 eth_dev_stop(struct rte_eth_dev *dev)
 278 {
 279         unsigned i;
 280         int sockfd;
 281         struct pmd_internals *internals = dev->data->dev_private;
 282
 283         for (i = 0; i < internals->nb_queues; i++) {
 284                 sockfd = internals->rx_queue[i].sockfd;
 285                 if (sockfd != -1)
 286                         close(sockfd);
 287
 288                 /* Prevent use after free in case tx fd == rx fd */
 289                 if (sockfd != internals->tx_queue[i].sockfd) {
 290                         sockfd = internals->tx_queue[i].sockfd;
 291                         if (sockfd != -1)
 292                                 close(sockfd);
 293                 }
 294
 295                 internals->rx_queue[i].sockfd = -1;
 296                 internals->tx_queue[i].sockfd = -1;
 297         }
 298
 299         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 300 }
 301
 302 static int
 303 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 304 {
 305         return 0;
 306 }
 307
 308 static int
 309 eth_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 310 {
 311         struct pmd_internals *internals = dev->data->dev_private;
 312
 313         dev_info->if_index = internals->if_index;
 314         dev_info->max_mac_addrs = 1;
 315         dev_info->max_rx_pktlen = (uint32_t)ETH_FRAME_LEN;
 316         dev_info->max_rx_queues = (uint16_t)internals->nb_queues;
 317         dev_info->max_tx_queues = (uint16_t)internals->nb_queues;
 318         dev_info->min_rx_bufsize = 0;
 319         dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
 320                 DEV_TX_OFFLOAD_VLAN_INSERT;
 321
 322         return 0;
 323 }
 324
 325 static int
 326 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *igb_stats)
 327 {
 328         unsigned i, imax;
 329         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 330         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 331         const struct pmd_internals *internal = dev->data->dev_private;
 332
 333         imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
 334                 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
 335         for (i = 0; i < imax; i++) {
 336                 igb_stats->q_ipackets[i] = internal->rx_queue[i].rx_pkts;
 337                 igb_stats->q_ibytes[i] = internal->rx_queue[i].rx_bytes;
 338                 rx_total += igb_stats->q_ipackets[i];
 339                 rx_bytes_total += igb_stats->q_ibytes[i];
 340         }
 341
 342         imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
 343                 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
 344         for (i = 0; i < imax; i++) {
 345                 igb_stats->q_opackets[i] = internal->tx_queue[i].tx_pkts;
 346                 igb_stats->q_obytes[i] = internal->tx_queue[i].tx_bytes;
 347                 tx_total += igb_stats->q_opackets[i];
 348                 tx_err_total += internal->tx_queue[i].err_pkts;
 349                 tx_bytes_total += igb_stats->q_obytes[i];
 350         }
 351
 352         igb_stats->ipackets = rx_total;
 353         igb_stats->ibytes = rx_bytes_total;
 354         igb_stats->opackets = tx_total;
 355         igb_stats->oerrors = tx_err_total;
 356         igb_stats->obytes = tx_bytes_total;
 357         return 0;
 358 }
 359
 360 static int
 361 eth_stats_reset(struct rte_eth_dev *dev)
 362 {
 363         unsigned i;
 364         struct pmd_internals *internal = dev->data->dev_private;
 365
 366         for (i = 0; i < internal->nb_queues; i++) {
 367                 internal->rx_queue[i].rx_pkts = 0;
 368                 internal->rx_queue[i].rx_bytes = 0;
 369         }
 370
 371         for (i = 0; i < internal->nb_queues; i++) {
 372                 internal->tx_queue[i].tx_pkts = 0;
 373                 internal->tx_queue[i].err_pkts = 0;
 374                 internal->tx_queue[i].tx_bytes = 0;
 375         }
 376
 377         return 0;
 378 }
 379
 380 static void
 381 eth_dev_close(struct rte_eth_dev *dev __rte_unused)
 382 {
 383 }
 384
 385 static void
 386 eth_queue_release(void *q __rte_unused)
 387 {
 388 }
 389
 390 static int
 391 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 392                 int wait_to_complete __rte_unused)
 393 {
 394         return 0;
 395 }
 396
 397 static int
 398 eth_rx_queue_setup(struct rte_eth_dev *dev,
 399                    uint16_t rx_queue_id,
 400                    uint16_t nb_rx_desc __rte_unused,
 401                    unsigned int socket_id __rte_unused,
 402                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 403                    struct rte_mempool *mb_pool)
 404 {
 405         struct pmd_internals *internals = dev->data->dev_private;
 406         struct pkt_rx_queue *pkt_q = &internals->rx_queue[rx_queue_id];
 407         unsigned int buf_size, data_size;
 408
 409         pkt_q->mb_pool = mb_pool;
 410
 411         /* Now get the space available for data in the mbuf */
 412         buf_size = rte_pktmbuf_data_room_size(pkt_q->mb_pool) -
 413                 RTE_PKTMBUF_HEADROOM;
 414         data_size = internals->req.tp_frame_size;
 415         data_size -= TPACKET2_HDRLEN - sizeof(struct sockaddr_ll);
 416
 417         if (data_size > buf_size) {
 418                 PMD_LOG(ERR,
 419                         "%s: %d bytes will not fit in mbuf (%d bytes)",
 420                         dev->device->name, data_size, buf_size);
 421                 return -ENOMEM;
 422         }
 423
 424         dev->data->rx_queues[rx_queue_id] = pkt_q;
 425         pkt_q->in_port = dev->data->port_id;
 426
 427         return 0;
 428 }
 429
 430 static int
 431 eth_tx_queue_setup(struct rte_eth_dev *dev,
 432                    uint16_t tx_queue_id,
 433                    uint16_t nb_tx_desc __rte_unused,
 434                    unsigned int socket_id __rte_unused,
 435                    const struct rte_eth_txconf *tx_conf __rte_unused)
 436 {
 437
 438         struct pmd_internals *internals = dev->data->dev_private;
 439
 440         dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
 441         return 0;
 442 }
 443
 444 static int
 445 eth_dev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 446 {
 447         struct pmd_internals *internals = dev->data->dev_private;
 448         struct ifreq ifr = { .ifr_mtu = mtu };
 449         int ret;
 450         int s;
 451         unsigned int data_size = internals->req.tp_frame_size -
 452                                  TPACKET2_HDRLEN;
 453
 454         if (mtu > data_size)
 455                 return -EINVAL;
 456
 457         s = socket(PF_INET, SOCK_DGRAM, 0);
 458         if (s < 0)
 459                 return -EINVAL;
 460
 461         strlcpy(ifr.ifr_name, internals->if_name, IFNAMSIZ);
 462         ret = ioctl(s, SIOCSIFMTU, &ifr);
 463         close(s);
 464
 465         if (ret < 0)
 466                 return -EINVAL;
 467
 468         return 0;
 469 }
 470
 471 static int
 472 eth_dev_change_flags(char *if_name, uint32_t flags, uint32_t mask)
 473 {
 474         struct ifreq ifr;
 475         int ret = 0;
 476         int s;
 477
 478         s = socket(PF_INET, SOCK_DGRAM, 0);
 479         if (s < 0)
 480                 return -errno;
 481
 482         strlcpy(ifr.ifr_name, if_name, IFNAMSIZ);
 483         if (ioctl(s, SIOCGIFFLAGS, &ifr) < 0) {
 484                 ret = -errno;
 485                 goto out;
 486         }
 487         ifr.ifr_flags &= mask;
 488         ifr.ifr_flags |= flags;
 489         if (ioctl(s, SIOCSIFFLAGS, &ifr) < 0) {
 490                 ret = -errno;
 491                 goto out;
 492         }
 493 out:
 494         close(s);
 495         return ret;
 496 }
 497
 498 static int
 499 eth_dev_promiscuous_enable(struct rte_eth_dev *dev)
 500 {
 501         struct pmd_internals *internals = dev->data->dev_private;
 502
 503         return eth_dev_change_flags(internals->if_name, IFF_PROMISC, ~0);
 504 }
 505
 506 static int
 507 eth_dev_promiscuous_disable(struct rte_eth_dev *dev)
 508 {
 509         struct pmd_internals *internals = dev->data->dev_private;
 510
 511         return eth_dev_change_flags(internals->if_name, 0, ~IFF_PROMISC);
 512 }
 513
 514 static const struct eth_dev_ops ops = {
 515         .dev_start = eth_dev_start,
 516         .dev_stop = eth_dev_stop,
 517         .dev_close = eth_dev_close,
 518         .dev_configure = eth_dev_configure,
 519         .dev_infos_get = eth_dev_info,
 520         .mtu_set = eth_dev_mtu_set,
 521         .promiscuous_enable = eth_dev_promiscuous_enable,
 522         .promiscuous_disable = eth_dev_promiscuous_disable,
 523         .rx_queue_setup = eth_rx_queue_setup,
 524         .tx_queue_setup = eth_tx_queue_setup,
 525         .rx_queue_release = eth_queue_release,
 526         .tx_queue_release = eth_queue_release,
 527         .link_update = eth_link_update,
 528         .stats_get = eth_stats_get,
 529         .stats_reset = eth_stats_reset,
 530 };
 531
 532 /*
 533  * Opens an AF_PACKET socket
 534  */
 535 static int
 536 open_packet_iface(const char *key __rte_unused,
 537                   const char *value __rte_unused,
 538                   void *extra_args)
 539 {
 540         int *sockfd = extra_args;
 541
 542         /* Open an AF_PACKET socket... */
 543         *sockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
 544         if (*sockfd == -1) {
 545                 PMD_LOG(ERR, "Could not open AF_PACKET socket");
 546                 return -1;
 547         }
 548
 549         return 0;
 550 }
 551
 552 static int
 553 rte_pmd_init_internals(struct rte_vdev_device *dev,
 554                        const int sockfd,
 555                        const unsigned nb_queues,
 556                        unsigned int blocksize,
 557                        unsigned int blockcnt,
 558                        unsigned int framesize,
 559                        unsigned int framecnt,
 560                        unsigned int qdisc_bypass,
 561                        struct pmd_internals **internals,
 562                        struct rte_eth_dev **eth_dev,
 563                        struct rte_kvargs *kvlist)
 564 {
 565         const char *name = rte_vdev_device_name(dev);
 566         const unsigned int numa_node = dev->device.numa_node;
 567         struct rte_eth_dev_data *data = NULL;
 568         struct rte_kvargs_pair *pair = NULL;
 569         struct ifreq ifr;
 570         size_t ifnamelen;
 571         unsigned k_idx;
 572         struct sockaddr_ll sockaddr;
 573         struct tpacket_req *req;
 574         struct pkt_rx_queue *rx_queue;
 575         struct pkt_tx_queue *tx_queue;
 576         int rc, tpver, discard;
 577         int qsockfd = -1;
 578         unsigned int i, q, rdsize;
 579 #if defined(PACKET_FANOUT)
 580         int fanout_arg;
 581 #endif
 582
 583         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 584                 pair = &kvlist->pairs[k_idx];
 585                 if (strstr(pair->key, ETH_AF_PACKET_IFACE_ARG) != NULL)
 586                         break;
 587         }
 588         if (pair == NULL) {
 589                 PMD_LOG(ERR,
 590                         "%s: no interface specified for AF_PACKET ethdev",
 591                         name);
 592                 return -1;
 593         }
 594
 595         PMD_LOG(INFO,
 596                 "%s: creating AF_PACKET-backed ethdev on numa socket %u",
 597                 name, numa_node);
 598
 599         *internals = rte_zmalloc_socket(name, sizeof(**internals),
 600                                         0, numa_node);
 601         if (*internals == NULL)
 602                 return -1;
 603
 604         for (q = 0; q < nb_queues; q++) {
 605                 (*internals)->rx_queue[q].map = MAP_FAILED;
 606                 (*internals)->tx_queue[q].map = MAP_FAILED;
 607         }
 608
 609         req = &((*internals)->req);
 610
 611         req->tp_block_size = blocksize;
 612         req->tp_block_nr = blockcnt;
 613         req->tp_frame_size = framesize;
 614         req->tp_frame_nr = framecnt;
 615
 616         ifnamelen = strlen(pair->value);
 617         if (ifnamelen < sizeof(ifr.ifr_name)) {
 618                 memcpy(ifr.ifr_name, pair->value, ifnamelen);
 619                 ifr.ifr_name[ifnamelen] = '\0';
 620         } else {
 621                 PMD_LOG(ERR,
 622                         "%s: I/F name too long (%s)",
 623                         name, pair->value);
 624                 return -1;
 625         }
 626         if (ioctl(sockfd, SIOCGIFINDEX, &ifr) == -1) {
 627                 PMD_LOG_ERRNO(ERR, "%s: ioctl failed (SIOCGIFINDEX)", name);
 628                 return -1;
 629         }
 630         (*internals)->if_name = strdup(pair->value);
 631         if ((*internals)->if_name == NULL)
 632                 return -1;
 633         (*internals)->if_index = ifr.ifr_ifindex;
 634
 635         if (ioctl(sockfd, SIOCGIFHWADDR, &ifr) == -1) {
 636                 PMD_LOG_ERRNO(ERR, "%s: ioctl failed (SIOCGIFHWADDR)", name);
 637                 return -1;
 638         }
 639         memcpy(&(*internals)->eth_addr, ifr.ifr_hwaddr.sa_data, ETH_ALEN);
 640
 641         memset(&sockaddr, 0, sizeof(sockaddr));
 642         sockaddr.sll_family = AF_PACKET;
 643         sockaddr.sll_protocol = htons(ETH_P_ALL);
 644         sockaddr.sll_ifindex = (*internals)->if_index;
 645
 646 #if defined(PACKET_FANOUT)
 647         fanout_arg = (getpid() ^ (*internals)->if_index) & 0xffff;
 648         fanout_arg |= (PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_DEFRAG) << 16;
 649 #if defined(PACKET_FANOUT_FLAG_ROLLOVER)
 650         fanout_arg |= PACKET_FANOUT_FLAG_ROLLOVER << 16;
 651 #endif
 652 #endif
 653
 654         for (q = 0; q < nb_queues; q++) {
 655                 /* Open an AF_PACKET socket for this queue... */
 656                 qsockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
 657                 if (qsockfd == -1) {
 658                         PMD_LOG_ERRNO(ERR,
 659                                 "%s: could not open AF_PACKET socket",
 660                                 name);
 661                         return -1;
 662                 }
 663
 664                 tpver = TPACKET_V2;
 665                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_VERSION,
 666                                 &tpver, sizeof(tpver));
 667                 if (rc == -1) {
 668                         PMD_LOG_ERRNO(ERR,
 669                                 "%s: could not set PACKET_VERSION on AF_PACKET socket for %s",
 670                                 name, pair->value);
 671                         goto error;
 672                 }
 673
 674                 discard = 1;
 675                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_LOSS,
 676                                 &discard, sizeof(discard));
 677                 if (rc == -1) {
 678                         PMD_LOG_ERRNO(ERR,
 679                                 "%s: could not set PACKET_LOSS on AF_PACKET socket for %s",
 680                                 name, pair->value);
 681                         goto error;
 682                 }
 683
 684 #if defined(PACKET_QDISC_BYPASS)
 685                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_QDISC_BYPASS,
 686                                 &qdisc_bypass, sizeof(qdisc_bypass));
 687                 if (rc == -1) {
 688                         PMD_LOG_ERRNO(ERR,
 689                                 "%s: could not set PACKET_QDISC_BYPASS on AF_PACKET socket for %s",
 690                                 name, pair->value);
 691                         goto error;
 692                 }
 693 #else
 694                 RTE_SET_USED(qdisc_bypass);
 695 #endif
 696
 697                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_RX_RING, req, sizeof(*req));
 698                 if (rc == -1) {
 699                         PMD_LOG_ERRNO(ERR,
 700                                 "%s: could not set PACKET_RX_RING on AF_PACKET socket for %s",
 701                                 name, pair->value);
 702                         goto error;
 703                 }
 704
 705                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_TX_RING, req, sizeof(*req));
 706                 if (rc == -1) {
 707                         PMD_LOG_ERRNO(ERR,
 708                                 "%s: could not set PACKET_TX_RING on AF_PACKET "
 709                                 "socket for %s", name, pair->value);
 710                         goto error;
 711                 }
 712
 713                 rx_queue = &((*internals)->rx_queue[q]);
 714                 rx_queue->framecount = req->tp_frame_nr;
 715
 716                 rx_queue->map = mmap(NULL, 2 * req->tp_block_size * req->tp_block_nr,
 717                                     PROT_READ | PROT_WRITE, MAP_SHARED | MAP_LOCKED,
 718                                     qsockfd, 0);
 719                 if (rx_queue->map == MAP_FAILED) {
 720                         PMD_LOG_ERRNO(ERR,
 721                                 "%s: call to mmap failed on AF_PACKET socket for %s",
 722                                 name, pair->value);
 723                         goto error;
 724                 }
 725
 726                 /* rdsize is same for both Tx and Rx */
 727                 rdsize = req->tp_frame_nr * sizeof(*(rx_queue->rd));
 728
 729                 rx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
 730                 if (rx_queue->rd == NULL)
 731                         goto error;
 732                 for (i = 0; i < req->tp_frame_nr; ++i) {
 733                         rx_queue->rd[i].iov_base = rx_queue->map + (i * framesize);
 734                         rx_queue->rd[i].iov_len = req->tp_frame_size;
 735                 }
 736                 rx_queue->sockfd = qsockfd;
 737
 738                 tx_queue = &((*internals)->tx_queue[q]);
 739                 tx_queue->framecount = req->tp_frame_nr;
 740                 tx_queue->frame_data_size = req->tp_frame_size;
 741                 tx_queue->frame_data_size -= TPACKET2_HDRLEN -
 742                         sizeof(struct sockaddr_ll);
 743
 744                 tx_queue->map = rx_queue->map + req->tp_block_size * req->tp_block_nr;
 745
 746                 tx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
 747                 if (tx_queue->rd == NULL)
 748                         goto error;
 749                 for (i = 0; i < req->tp_frame_nr; ++i) {
 750                         tx_queue->rd[i].iov_base = tx_queue->map + (i * framesize);
 751                         tx_queue->rd[i].iov_len = req->tp_frame_size;
 752                 }
 753                 tx_queue->sockfd = qsockfd;
 754
 755                 rc = bind(qsockfd, (const struct sockaddr*)&sockaddr, sizeof(sockaddr));
 756                 if (rc == -1) {
 757                         PMD_LOG_ERRNO(ERR,
 758                                 "%s: could not bind AF_PACKET socket to %s",
 759                                 name, pair->value);
 760                         goto error;
 761                 }
 762
 763 #if defined(PACKET_FANOUT)
 764                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_FANOUT,
 765                                 &fanout_arg, sizeof(fanout_arg));
 766                 if (rc == -1) {
 767                         PMD_LOG_ERRNO(ERR,
 768                                 "%s: could not set PACKET_FANOUT on AF_PACKET socket for %s",
 769                                 name, pair->value);
 770                         goto error;
 771                 }
 772 #endif
 773         }
 774
 775         /* reserve an ethdev entry */
 776         *eth_dev = rte_eth_vdev_allocate(dev, 0);
 777         if (*eth_dev == NULL)
 778                 goto error;
 779
 780         /*
 781          * now put it all together
 782          * - store queue data in internals,
 783          * - store numa_node in eth_dev
 784          * - point eth_dev_data to internals
 785          * - and point eth_dev structure to new eth_dev_data structure
 786          */
 787
 788         (*internals)->nb_queues = nb_queues;
 789
 790         data = (*eth_dev)->data;
 791         data->dev_private = *internals;
 792         data->nb_rx_queues = (uint16_t)nb_queues;
 793         data->nb_tx_queues = (uint16_t)nb_queues;
 794         data->dev_link = pmd_link;
 795         data->mac_addrs = &(*internals)->eth_addr;
 796
 797         (*eth_dev)->dev_ops = &ops;
 798
 799         return 0;
 800
 801 error:
 802         if (qsockfd != -1)
 803                 close(qsockfd);
 804         for (q = 0; q < nb_queues; q++) {
 805                 munmap((*internals)->rx_queue[q].map,
 806                        2 * req->tp_block_size * req->tp_block_nr);
 807
 808                 rte_free((*internals)->rx_queue[q].rd);
 809                 rte_free((*internals)->tx_queue[q].rd);
 810                 if (((*internals)->rx_queue[q].sockfd != 0) &&
 811                         ((*internals)->rx_queue[q].sockfd != qsockfd))
 812                         close((*internals)->rx_queue[q].sockfd);
 813         }
 814         free((*internals)->if_name);
 815         rte_free(*internals);
 816         return -1;
 817 }
 818
 819 static int
 820 rte_eth_from_packet(struct rte_vdev_device *dev,
 821                     int const *sockfd,
 822                     struct rte_kvargs *kvlist)
 823 {
 824         const char *name = rte_vdev_device_name(dev);
 825         struct pmd_internals *internals = NULL;
 826         struct rte_eth_dev *eth_dev = NULL;
 827         struct rte_kvargs_pair *pair = NULL;
 828         unsigned k_idx;
 829         unsigned int blockcount;
 830         unsigned int blocksize;
 831         unsigned int framesize = DFLT_FRAME_SIZE;
 832         unsigned int framecount = DFLT_FRAME_COUNT;
 833         unsigned int qpairs = 1;
 834         unsigned int qdisc_bypass = 1;
 835
 836         /* do some parameter checking */
 837         if (*sockfd < 0)
 838                 return -1;
 839
 840         blocksize = getpagesize();
 841
 842         /*
 843          * Walk arguments for configurable settings
 844          */
 845         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 846                 pair = &kvlist->pairs[k_idx];
 847                 if (strstr(pair->key, ETH_AF_PACKET_NUM_Q_ARG) != NULL) {
 848                         qpairs = atoi(pair->value);
 849                         if (qpairs < 1 ||
 850                             qpairs > RTE_PMD_AF_PACKET_MAX_RINGS) {
 851                                 PMD_LOG(ERR,
 852                                         "%s: invalid qpairs value",
 853                                         name);
 854                                 return -1;
 855                         }
 856                         continue;
 857                 }
 858                 if (strstr(pair->key, ETH_AF_PACKET_BLOCKSIZE_ARG) != NULL) {
 859                         blocksize = atoi(pair->value);
 860                         if (!blocksize) {
 861                                 PMD_LOG(ERR,
 862                                         "%s: invalid blocksize value",
 863                                         name);
 864                                 return -1;
 865                         }
 866                         continue;
 867                 }
 868                 if (strstr(pair->key, ETH_AF_PACKET_FRAMESIZE_ARG) != NULL) {
 869                         framesize = atoi(pair->value);
 870                         if (!framesize) {
 871                                 PMD_LOG(ERR,
 872                                         "%s: invalid framesize value",
 873                                         name);
 874                                 return -1;
 875                         }
 876                         continue;
 877                 }
 878                 if (strstr(pair->key, ETH_AF_PACKET_FRAMECOUNT_ARG) != NULL) {
 879                         framecount = atoi(pair->value);
 880                         if (!framecount) {
 881                                 PMD_LOG(ERR,
 882                                         "%s: invalid framecount value",
 883                                         name);
 884                                 return -1;
 885                         }
 886                         continue;
 887                 }
 888                 if (strstr(pair->key, ETH_AF_PACKET_QDISC_BYPASS_ARG) != NULL) {
 889                         qdisc_bypass = atoi(pair->value);
 890                         if (qdisc_bypass > 1) {
 891                                 PMD_LOG(ERR,
 892                                         "%s: invalid bypass value",
 893                                         name);
 894                                 return -1;
 895                         }
 896                         continue;
 897                 }
 898         }
 899
 900         if (framesize > blocksize) {
 901                 PMD_LOG(ERR,
 902                         "%s: AF_PACKET MMAP frame size exceeds block size!",
 903                         name);
 904                 return -1;
 905         }
 906
 907         blockcount = framecount / (blocksize / framesize);
 908         if (!blockcount) {
 909                 PMD_LOG(ERR,
 910                         "%s: invalid AF_PACKET MMAP parameters", name);
 911                 return -1;
 912         }
 913
 914         PMD_LOG(INFO, "%s: AF_PACKET MMAP parameters:", name);
 915         PMD_LOG(INFO, "%s:\tblock size %d", name, blocksize);
 916         PMD_LOG(INFO, "%s:\tblock count %d", name, blockcount);
 917         PMD_LOG(INFO, "%s:\tframe size %d", name, framesize);
 918         PMD_LOG(INFO, "%s:\tframe count %d", name, framecount);
 919
 920         if (rte_pmd_init_internals(dev, *sockfd, qpairs,
 921                                    blocksize, blockcount,
 922                                    framesize, framecount,
 923                                    qdisc_bypass,
 924                                    &internals, &eth_dev,
 925                                    kvlist) < 0)
 926                 return -1;
 927
 928         eth_dev->rx_pkt_burst = eth_af_packet_rx;
 929         eth_dev->tx_pkt_burst = eth_af_packet_tx;
 930
 931         rte_eth_dev_probing_finish(eth_dev);
 932         return 0;
 933 }
 934
 935 static int
 936 rte_pmd_af_packet_probe(struct rte_vdev_device *dev)
 937 {
 938         int ret = 0;
 939         struct rte_kvargs *kvlist;
 940         int sockfd = -1;
 941         struct rte_eth_dev *eth_dev;
 942         const char *name = rte_vdev_device_name(dev);
 943
 944         PMD_LOG(INFO, "Initializing pmd_af_packet for %s", name);
 945
 946         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
 947                 eth_dev = rte_eth_dev_attach_secondary(name);
 948                 if (!eth_dev) {
 949                         PMD_LOG(ERR, "Failed to probe %s", name);
 950                         return -1;
 951                 }
 952                 /* TODO: request info from primary to set up Rx and Tx */
 953                 eth_dev->dev_ops = &ops;
 954                 eth_dev->device = &dev->device;
 955                 rte_eth_dev_probing_finish(eth_dev);
 956                 return 0;
 957         }
 958
 959         kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_arguments);
 960         if (kvlist == NULL) {
 961                 ret = -1;
 962                 goto exit;
 963         }
 964
 965         /*
 966          * If iface argument is passed we open the NICs and use them for
 967          * reading / writing
 968          */
 969         if (rte_kvargs_count(kvlist, ETH_AF_PACKET_IFACE_ARG) == 1) {
 970
 971                 ret = rte_kvargs_process(kvlist, ETH_AF_PACKET_IFACE_ARG,
 972                                          &open_packet_iface, &sockfd);
 973                 if (ret < 0)
 974                         goto exit;
 975         }
 976
 977         if (dev->device.numa_node == SOCKET_ID_ANY)
 978                 dev->device.numa_node = rte_socket_id();
 979
 980         ret = rte_eth_from_packet(dev, &sockfd, kvlist);
 981         close(sockfd); /* no longer needed */
 982
 983 exit:
 984         rte_kvargs_free(kvlist);
 985         return ret;
 986 }
 987
 988 static int
 989 rte_pmd_af_packet_remove(struct rte_vdev_device *dev)
 990 {
 991         struct rte_eth_dev *eth_dev = NULL;
 992         struct pmd_internals *internals;
 993         struct tpacket_req *req;
 994         unsigned q;
 995
 996         PMD_LOG(INFO, "Closing AF_PACKET ethdev on numa socket %u",
 997                 rte_socket_id());
 998
 999         if (dev == NULL)
1000                 return -1;
1001
1002         /* find the ethdev entry */
1003         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1004         if (eth_dev == NULL)
1005                 return -1;
1006
1007         /* mac_addrs must not be freed alone because part of dev_private */
1008         eth_dev->data->mac_addrs = NULL;
1009
1010         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1011                 return rte_eth_dev_release_port(eth_dev);
1012
1013         internals = eth_dev->data->dev_private;
1014         req = &internals->req;
1015         for (q = 0; q < internals->nb_queues; q++) {
1016                 munmap(internals->rx_queue[q].map,
1017                         2 * req->tp_block_size * req->tp_block_nr);
1018                 rte_free(internals->rx_queue[q].rd);
1019                 rte_free(internals->tx_queue[q].rd);
1020         }
1021         free(internals->if_name);
1022
1023         rte_eth_dev_release_port(eth_dev);
1024
1025         return 0;
1026 }
1027
1028 static struct rte_vdev_driver pmd_af_packet_drv = {
1029         .probe = rte_pmd_af_packet_probe,
1030         .remove = rte_pmd_af_packet_remove,
1031 };
1032
1033 RTE_PMD_REGISTER_VDEV(net_af_packet, pmd_af_packet_drv);
1034 RTE_PMD_REGISTER_ALIAS(net_af_packet, eth_af_packet);
1035 RTE_PMD_REGISTER_PARAM_STRING(net_af_packet,
1036         "iface=<string> "
1037         "qpairs=<int> "
1038         "blocksz=<int> "
1039         "framesz=<int> "
1040         "framecnt=<int> "
1041         "qdisc_bypass=<0|1>");
1042
1043 RTE_INIT(af_packet_init_log)
1044 {
1045         af_packet_logtype = rte_log_register("pmd.net.packet");
1046         if (af_packet_logtype >= 0)
1047                 rte_log_set_level(af_packet_logtype, RTE_LOG_NOTICE);
1048 }