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