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