drivers/net/af_packet/rte_eth_af_packet.c

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