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_malloc.h>
  42 #include <rte_kvargs.h>
  43 #include <rte_vdev.h>
  44
  45 #include <linux/if_ether.h>
  46 #include <linux/if_packet.h>
  47 #include <arpa/inet.h>
  48 #include <net/if.h>
  49 #include <sys/types.h>
  50 #include <sys/socket.h>
  51 #include <sys/ioctl.h>
  52 #include <sys/mman.h>
  53 #include <unistd.h>
  54 #include <poll.h>
  55
  56 #define ETH_AF_PACKET_IFACE_ARG         "iface"
  57 #define ETH_AF_PACKET_NUM_Q_ARG         "qpairs"
  58 #define ETH_AF_PACKET_BLOCKSIZE_ARG     "blocksz"
  59 #define ETH_AF_PACKET_FRAMESIZE_ARG     "framesz"
  60 #define ETH_AF_PACKET_FRAMECOUNT_ARG    "framecnt"
  61
  62 #define DFLT_BLOCK_SIZE         (1 << 12)
  63 #define DFLT_FRAME_SIZE         (1 << 11)
  64 #define DFLT_FRAME_COUNT        (1 << 9)
  65
  66 #define RTE_PMD_AF_PACKET_MAX_RINGS 16
  67
  68 struct pkt_rx_queue {
  69         int sockfd;
  70
  71         struct iovec *rd;
  72         uint8_t *map;
  73         unsigned int framecount;
  74         unsigned int framenum;
  75
  76         struct rte_mempool *mb_pool;
  77         uint8_t in_port;
  78
  79         volatile unsigned long rx_pkts;
  80         volatile unsigned long err_pkts;
  81         volatile unsigned long rx_bytes;
  82 };
  83
  84 struct pkt_tx_queue {
  85         int sockfd;
  86         unsigned int frame_data_size;
  87
  88         struct iovec *rd;
  89         uint8_t *map;
  90         unsigned int framecount;
  91         unsigned int framenum;
  92
  93         volatile unsigned long tx_pkts;
  94         volatile unsigned long err_pkts;
  95         volatile unsigned long tx_bytes;
  96 };
  97
  98 struct pmd_internals {
  99         unsigned nb_queues;
 100
 101         int if_index;
 102         char *if_name;
 103         struct ether_addr eth_addr;
 104
 105         struct tpacket_req req;
 106
 107         struct pkt_rx_queue rx_queue[RTE_PMD_AF_PACKET_MAX_RINGS];
 108         struct pkt_tx_queue tx_queue[RTE_PMD_AF_PACKET_MAX_RINGS];
 109 };
 110
 111 static const char *valid_arguments[] = {
 112         ETH_AF_PACKET_IFACE_ARG,
 113         ETH_AF_PACKET_NUM_Q_ARG,
 114         ETH_AF_PACKET_BLOCKSIZE_ARG,
 115         ETH_AF_PACKET_FRAMESIZE_ARG,
 116         ETH_AF_PACKET_FRAMECOUNT_ARG,
 117         NULL
 118 };
 119
 120 static struct rte_eth_link pmd_link = {
 121         .link_speed = ETH_SPEED_NUM_10G,
 122         .link_duplex = ETH_LINK_FULL_DUPLEX,
 123         .link_status = ETH_LINK_DOWN,
 124         .link_autoneg = ETH_LINK_SPEED_AUTONEG
 125 };
 126
 127 static uint16_t
 128 eth_af_packet_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 129 {
 130         unsigned i;
 131         struct tpacket2_hdr *ppd;
 132         struct rte_mbuf *mbuf;
 133         uint8_t *pbuf;
 134         struct pkt_rx_queue *pkt_q = queue;
 135         uint16_t num_rx = 0;
 136         unsigned long num_rx_bytes = 0;
 137         unsigned int framecount, framenum;
 138
 139         if (unlikely(nb_pkts == 0))
 140                 return 0;
 141
 142         /*
 143          * Reads the given number of packets from the AF_PACKET socket one by
 144          * one and copies the packet data into a newly allocated mbuf.
 145          */
 146         framecount = pkt_q->framecount;
 147         framenum = pkt_q->framenum;
 148         for (i = 0; i < nb_pkts; i++) {
 149                 /* point at the next incoming frame */
 150                 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
 151                 if ((ppd->tp_status & TP_STATUS_USER) == 0)
 152                         break;
 153
 154                 /* allocate the next mbuf */
 155                 mbuf = rte_pktmbuf_alloc(pkt_q->mb_pool);
 156                 if (unlikely(mbuf == NULL))
 157                         break;
 158
 159                 /* packet will fit in the mbuf, go ahead and receive it */
 160                 rte_pktmbuf_pkt_len(mbuf) = rte_pktmbuf_data_len(mbuf) = ppd->tp_snaplen;
 161                 pbuf = (uint8_t *) ppd + ppd->tp_mac;
 162                 memcpy(rte_pktmbuf_mtod(mbuf, void *), pbuf, rte_pktmbuf_data_len(mbuf));
 163
 164                 /* release incoming frame and advance ring buffer */
 165                 ppd->tp_status = TP_STATUS_KERNEL;
 166                 if (++framenum >= framecount)
 167                         framenum = 0;
 168                 mbuf->port = pkt_q->in_port;
 169
 170                 /* account for the receive frame */
 171                 bufs[i] = mbuf;
 172                 num_rx++;
 173                 num_rx_bytes += mbuf->pkt_len;
 174         }
 175         pkt_q->framenum = framenum;
 176         pkt_q->rx_pkts += num_rx;
 177         pkt_q->rx_bytes += num_rx_bytes;
 178         return num_rx;
 179 }
 180
 181 /*
 182  * Callback to handle sending packets through a real NIC.
 183  */
 184 static uint16_t
 185 eth_af_packet_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 186 {
 187         struct tpacket2_hdr *ppd;
 188         struct rte_mbuf *mbuf;
 189         uint8_t *pbuf;
 190         unsigned int framecount, framenum;
 191         struct pollfd pfd;
 192         struct pkt_tx_queue *pkt_q = queue;
 193         uint16_t num_tx = 0;
 194         unsigned long num_tx_bytes = 0;
 195         int i;
 196
 197         if (unlikely(nb_pkts == 0))
 198                 return 0;
 199
 200         memset(&pfd, 0, sizeof(pfd));
 201         pfd.fd = pkt_q->sockfd;
 202         pfd.events = POLLOUT;
 203         pfd.revents = 0;
 204
 205         framecount = pkt_q->framecount;
 206         framenum = pkt_q->framenum;
 207         ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
 208         for (i = 0; i < nb_pkts; i++) {
 209                 mbuf = *bufs++;
 210
 211                 /* drop oversized packets */
 212                 if (rte_pktmbuf_data_len(mbuf) > pkt_q->frame_data_size) {
 213                         rte_pktmbuf_free(mbuf);
 214                         continue;
 215                 }
 216
 217                 /* point at the next incoming frame */
 218                 if ((ppd->tp_status != TP_STATUS_AVAILABLE) &&
 219                     (poll(&pfd, 1, -1) < 0))
 220                         break;
 221
 222                 /* copy the tx frame data */
 223                 pbuf = (uint8_t *) ppd + TPACKET2_HDRLEN -
 224                         sizeof(struct sockaddr_ll);
 225                 memcpy(pbuf, rte_pktmbuf_mtod(mbuf, void*), rte_pktmbuf_data_len(mbuf));
 226                 ppd->tp_len = ppd->tp_snaplen = rte_pktmbuf_data_len(mbuf);
 227
 228                 /* release incoming frame and advance ring buffer */
 229                 ppd->tp_status = TP_STATUS_SEND_REQUEST;
 230                 if (++framenum >= framecount)
 231                         framenum = 0;
 232                 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
 233
 234                 num_tx++;
 235                 num_tx_bytes += mbuf->pkt_len;
 236                 rte_pktmbuf_free(mbuf);
 237         }
 238
 239         /* kick-off transmits */
 240         if (sendto(pkt_q->sockfd, NULL, 0, MSG_DONTWAIT, NULL, 0) == -1)
 241                 num_tx = 0; /* error sending -- no packets transmitted */
 242
 243         pkt_q->framenum = framenum;
 244         pkt_q->tx_pkts += num_tx;
 245         pkt_q->err_pkts += i - num_tx;
 246         pkt_q->tx_bytes += num_tx_bytes;
 247         return i;
 248 }
 249
 250 static int
 251 eth_dev_start(struct rte_eth_dev *dev)
 252 {
 253         dev->data->dev_link.link_status = ETH_LINK_UP;
 254         return 0;
 255 }
 256
 257 /*
 258  * This function gets called when the current port gets stopped.
 259  */
 260 static void
 261 eth_dev_stop(struct rte_eth_dev *dev)
 262 {
 263         unsigned i;
 264         int sockfd;
 265         struct pmd_internals *internals = dev->data->dev_private;
 266
 267         for (i = 0; i < internals->nb_queues; i++) {
 268                 sockfd = internals->rx_queue[i].sockfd;
 269                 if (sockfd != -1)
 270                         close(sockfd);
 271
 272                 /* Prevent use after free in case tx fd == rx fd */
 273                 if (sockfd != internals->tx_queue[i].sockfd) {
 274                         sockfd = internals->tx_queue[i].sockfd;
 275                         if (sockfd != -1)
 276                                 close(sockfd);
 277                 }
 278
 279                 internals->rx_queue[i].sockfd = -1;
 280                 internals->tx_queue[i].sockfd = -1;
 281         }
 282
 283         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 284 }
 285
 286 static int
 287 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 288 {
 289         return 0;
 290 }
 291
 292 static void
 293 eth_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 294 {
 295         struct pmd_internals *internals = dev->data->dev_private;
 296
 297         dev_info->if_index = internals->if_index;
 298         dev_info->max_mac_addrs = 1;
 299         dev_info->max_rx_pktlen = (uint32_t)ETH_FRAME_LEN;
 300         dev_info->max_rx_queues = (uint16_t)internals->nb_queues;
 301         dev_info->max_tx_queues = (uint16_t)internals->nb_queues;
 302         dev_info->min_rx_bufsize = 0;
 303 }
 304
 305 static void
 306 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *igb_stats)
 307 {
 308         unsigned i, imax;
 309         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 310         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 311         const struct pmd_internals *internal = dev->data->dev_private;
 312
 313         imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
 314                 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
 315         for (i = 0; i < imax; i++) {
 316                 igb_stats->q_ipackets[i] = internal->rx_queue[i].rx_pkts;
 317                 igb_stats->q_ibytes[i] = internal->rx_queue[i].rx_bytes;
 318                 rx_total += igb_stats->q_ipackets[i];
 319                 rx_bytes_total += igb_stats->q_ibytes[i];
 320         }
 321
 322         imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
 323                 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
 324         for (i = 0; i < imax; i++) {
 325                 igb_stats->q_opackets[i] = internal->tx_queue[i].tx_pkts;
 326                 igb_stats->q_errors[i] = internal->tx_queue[i].err_pkts;
 327                 igb_stats->q_obytes[i] = internal->tx_queue[i].tx_bytes;
 328                 tx_total += igb_stats->q_opackets[i];
 329                 tx_err_total += igb_stats->q_errors[i];
 330                 tx_bytes_total += igb_stats->q_obytes[i];
 331         }
 332
 333         igb_stats->ipackets = rx_total;
 334         igb_stats->ibytes = rx_bytes_total;
 335         igb_stats->opackets = tx_total;
 336         igb_stats->oerrors = tx_err_total;
 337         igb_stats->obytes = tx_bytes_total;
 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->data->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         strncpy(ifr.ifr_name, internals->if_name, IFNAMSIZ);
 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         strncpy(ifr.ifr_name, if_name, IFNAMSIZ);
 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(const char *name,
 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                        const unsigned numa_node,
 536                        struct pmd_internals **internals,
 537                        struct rte_eth_dev **eth_dev,
 538                        struct rte_kvargs *kvlist)
 539 {
 540         struct rte_eth_dev_data *data = NULL;
 541         struct rte_kvargs_pair *pair = NULL;
 542         struct ifreq ifr;
 543         size_t ifnamelen;
 544         unsigned k_idx;
 545         struct sockaddr_ll sockaddr;
 546         struct tpacket_req *req;
 547         struct pkt_rx_queue *rx_queue;
 548         struct pkt_tx_queue *tx_queue;
 549         int rc, tpver, discard;
 550         int qsockfd = -1;
 551         unsigned int i, q, rdsize;
 552         int fanout_arg __rte_unused, bypass __rte_unused;
 553
 554         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 555                 pair = &kvlist->pairs[k_idx];
 556                 if (strstr(pair->key, ETH_AF_PACKET_IFACE_ARG) != NULL)
 557                         break;
 558         }
 559         if (pair == NULL) {
 560                 RTE_LOG(ERR, PMD,
 561                         "%s: no interface specified for AF_PACKET ethdev\n",
 562                         name);
 563                 goto error_early;
 564         }
 565
 566         RTE_LOG(INFO, PMD,
 567                 "%s: creating AF_PACKET-backed ethdev on numa socket %u\n",
 568                 name, numa_node);
 569
 570         /*
 571          * now do all data allocation - for eth_dev structure, dummy pci driver
 572          * and internal (private) data
 573          */
 574         data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
 575         if (data == NULL)
 576                 goto error_early;
 577
 578         *internals = rte_zmalloc_socket(name, sizeof(**internals),
 579                                         0, numa_node);
 580         if (*internals == NULL)
 581                 goto error_early;
 582
 583         for (q = 0; q < nb_queues; q++) {
 584                 (*internals)->rx_queue[q].map = MAP_FAILED;
 585                 (*internals)->tx_queue[q].map = MAP_FAILED;
 586         }
 587
 588         req = &((*internals)->req);
 589
 590         req->tp_block_size = blocksize;
 591         req->tp_block_nr = blockcnt;
 592         req->tp_frame_size = framesize;
 593         req->tp_frame_nr = framecnt;
 594
 595         ifnamelen = strlen(pair->value);
 596         if (ifnamelen < sizeof(ifr.ifr_name)) {
 597                 memcpy(ifr.ifr_name, pair->value, ifnamelen);
 598                 ifr.ifr_name[ifnamelen] = '\0';
 599         } else {
 600                 RTE_LOG(ERR, PMD,
 601                         "%s: I/F name too long (%s)\n",
 602                         name, pair->value);
 603                 goto error_early;
 604         }
 605         if (ioctl(sockfd, SIOCGIFINDEX, &ifr) == -1) {
 606                 RTE_LOG(ERR, PMD,
 607                         "%s: ioctl failed (SIOCGIFINDEX)\n",
 608                         name);
 609                 goto error_early;
 610         }
 611         (*internals)->if_name = strdup(pair->value);
 612         (*internals)->if_index = ifr.ifr_ifindex;
 613
 614         if (ioctl(sockfd, SIOCGIFHWADDR, &ifr) == -1) {
 615                 RTE_LOG(ERR, PMD,
 616                         "%s: ioctl failed (SIOCGIFHWADDR)\n",
 617                         name);
 618                 goto error_early;
 619         }
 620         memcpy(&(*internals)->eth_addr, ifr.ifr_hwaddr.sa_data, ETH_ALEN);
 621
 622         memset(&sockaddr, 0, sizeof(sockaddr));
 623         sockaddr.sll_family = AF_PACKET;
 624         sockaddr.sll_protocol = htons(ETH_P_ALL);
 625         sockaddr.sll_ifindex = (*internals)->if_index;
 626
 627 #if defined(PACKET_FANOUT)
 628         fanout_arg = (getpid() ^ (*internals)->if_index) & 0xffff;
 629         fanout_arg |= (PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_DEFRAG) << 16;
 630 #if defined(PACKET_FANOUT_FLAG_ROLLOVER)
 631         fanout_arg |= PACKET_FANOUT_FLAG_ROLLOVER << 16;
 632 #endif
 633 #endif
 634
 635         for (q = 0; q < nb_queues; q++) {
 636                 /* Open an AF_PACKET socket for this queue... */
 637                 qsockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
 638                 if (qsockfd == -1) {
 639                         RTE_LOG(ERR, PMD,
 640                                 "%s: could not open AF_PACKET socket\n",
 641                                 name);
 642                         return -1;
 643                 }
 644
 645                 tpver = TPACKET_V2;
 646                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_VERSION,
 647                                 &tpver, sizeof(tpver));
 648                 if (rc == -1) {
 649                         RTE_LOG(ERR, PMD,
 650                                 "%s: could not set PACKET_VERSION on AF_PACKET "
 651                                 "socket for %s\n", name, pair->value);
 652                         goto error;
 653                 }
 654
 655                 discard = 1;
 656                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_LOSS,
 657                                 &discard, sizeof(discard));
 658                 if (rc == -1) {
 659                         RTE_LOG(ERR, PMD,
 660                                 "%s: could not set PACKET_LOSS on "
 661                                 "AF_PACKET socket for %s\n", name, pair->value);
 662                         goto error;
 663                 }
 664
 665 #if defined(PACKET_QDISC_BYPASS)
 666                 bypass = 1;
 667                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_QDISC_BYPASS,
 668                                 &bypass, sizeof(bypass));
 669                 if (rc == -1) {
 670                         RTE_LOG(ERR, PMD,
 671                                 "%s: could not set PACKET_QDISC_BYPASS "
 672                                 "on AF_PACKET socket for %s\n", name,
 673                                 pair->value);
 674                         goto error;
 675                 }
 676 #endif
 677
 678                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_RX_RING, req, sizeof(*req));
 679                 if (rc == -1) {
 680                         RTE_LOG(ERR, PMD,
 681                                 "%s: could not set PACKET_RX_RING on AF_PACKET "
 682                                 "socket for %s\n", name, pair->value);
 683                         goto error;
 684                 }
 685
 686                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_TX_RING, req, sizeof(*req));
 687                 if (rc == -1) {
 688                         RTE_LOG(ERR, PMD,
 689                                 "%s: could not set PACKET_TX_RING on AF_PACKET "
 690                                 "socket for %s\n", name, pair->value);
 691                         goto error;
 692                 }
 693
 694                 rx_queue = &((*internals)->rx_queue[q]);
 695                 rx_queue->framecount = req->tp_frame_nr;
 696
 697                 rx_queue->map = mmap(NULL, 2 * req->tp_block_size * req->tp_block_nr,
 698                                     PROT_READ | PROT_WRITE, MAP_SHARED | MAP_LOCKED,
 699                                     qsockfd, 0);
 700                 if (rx_queue->map == MAP_FAILED) {
 701                         RTE_LOG(ERR, PMD,
 702                                 "%s: call to mmap failed on AF_PACKET socket for %s\n",
 703                                 name, pair->value);
 704                         goto error;
 705                 }
 706
 707                 /* rdsize is same for both Tx and Rx */
 708                 rdsize = req->tp_frame_nr * sizeof(*(rx_queue->rd));
 709
 710                 rx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
 711                 if (rx_queue->rd == NULL)
 712                         goto error;
 713                 for (i = 0; i < req->tp_frame_nr; ++i) {
 714                         rx_queue->rd[i].iov_base = rx_queue->map + (i * framesize);
 715                         rx_queue->rd[i].iov_len = req->tp_frame_size;
 716                 }
 717                 rx_queue->sockfd = qsockfd;
 718
 719                 tx_queue = &((*internals)->tx_queue[q]);
 720                 tx_queue->framecount = req->tp_frame_nr;
 721                 tx_queue->frame_data_size = req->tp_frame_size;
 722                 tx_queue->frame_data_size -= TPACKET2_HDRLEN -
 723                         sizeof(struct sockaddr_ll);
 724
 725                 tx_queue->map = rx_queue->map + req->tp_block_size * req->tp_block_nr;
 726
 727                 tx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
 728                 if (tx_queue->rd == NULL)
 729                         goto error;
 730                 for (i = 0; i < req->tp_frame_nr; ++i) {
 731                         tx_queue->rd[i].iov_base = tx_queue->map + (i * framesize);
 732                         tx_queue->rd[i].iov_len = req->tp_frame_size;
 733                 }
 734                 tx_queue->sockfd = qsockfd;
 735
 736                 rc = bind(qsockfd, (const struct sockaddr*)&sockaddr, sizeof(sockaddr));
 737                 if (rc == -1) {
 738                         RTE_LOG(ERR, PMD,
 739                                 "%s: could not bind AF_PACKET socket to %s\n",
 740                                 name, pair->value);
 741                         goto error;
 742                 }
 743
 744 #if defined(PACKET_FANOUT)
 745                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_FANOUT,
 746                                 &fanout_arg, sizeof(fanout_arg));
 747                 if (rc == -1) {
 748                         RTE_LOG(ERR, PMD,
 749                                 "%s: could not set PACKET_FANOUT on AF_PACKET socket "
 750                                 "for %s\n", name, pair->value);
 751                         goto error;
 752                 }
 753 #endif
 754         }
 755
 756         /* reserve an ethdev entry */
 757         *eth_dev = rte_eth_dev_allocate(name);
 758         if (*eth_dev == NULL)
 759                 goto error;
 760
 761         /*
 762          * now put it all together
 763          * - store queue data in internals,
 764          * - store numa_node in eth_dev
 765          * - point eth_dev_data to internals
 766          * - and point eth_dev structure to new eth_dev_data structure
 767          */
 768
 769         (*internals)->nb_queues = nb_queues;
 770
 771         data->dev_private = *internals;
 772         data->port_id = (*eth_dev)->data->port_id;
 773         data->nb_rx_queues = (uint16_t)nb_queues;
 774         data->nb_tx_queues = (uint16_t)nb_queues;
 775         data->dev_link = pmd_link;
 776         data->mac_addrs = &(*internals)->eth_addr;
 777         strncpy(data->name,
 778                 (*eth_dev)->data->name, strlen((*eth_dev)->data->name));
 779
 780         (*eth_dev)->data = data;
 781         (*eth_dev)->dev_ops = &ops;
 782         (*eth_dev)->driver = NULL;
 783         (*eth_dev)->data->dev_flags = RTE_ETH_DEV_DETACHABLE;
 784         (*eth_dev)->data->drv_name = pmd_af_packet_drv.driver.name;
 785         (*eth_dev)->data->kdrv = RTE_KDRV_NONE;
 786         (*eth_dev)->data->numa_node = numa_node;
 787
 788         return 0;
 789
 790 error:
 791         if (qsockfd != -1)
 792                 close(qsockfd);
 793         for (q = 0; q < nb_queues; q++) {
 794                 munmap((*internals)->rx_queue[q].map,
 795                        2 * req->tp_block_size * req->tp_block_nr);
 796
 797                 rte_free((*internals)->rx_queue[q].rd);
 798                 rte_free((*internals)->tx_queue[q].rd);
 799                 if (((*internals)->rx_queue[q].sockfd != 0) &&
 800                         ((*internals)->rx_queue[q].sockfd != qsockfd))
 801                         close((*internals)->rx_queue[q].sockfd);
 802         }
 803         free((*internals)->if_name);
 804         rte_free(*internals);
 805 error_early:
 806         rte_free(data);
 807         return -1;
 808 }
 809
 810 static int
 811 rte_eth_from_packet(const char *name,
 812                     int const *sockfd,
 813                     const unsigned numa_node,
 814                     struct rte_kvargs *kvlist)
 815 {
 816         struct pmd_internals *internals = NULL;
 817         struct rte_eth_dev *eth_dev = NULL;
 818         struct rte_kvargs_pair *pair = NULL;
 819         unsigned k_idx;
 820         unsigned int blockcount;
 821         unsigned int blocksize = DFLT_BLOCK_SIZE;
 822         unsigned int framesize = DFLT_FRAME_SIZE;
 823         unsigned int framecount = DFLT_FRAME_COUNT;
 824         unsigned int qpairs = 1;
 825
 826         /* do some parameter checking */
 827         if (*sockfd < 0)
 828                 return -1;
 829
 830         /*
 831          * Walk arguments for configurable settings
 832          */
 833         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 834                 pair = &kvlist->pairs[k_idx];
 835                 if (strstr(pair->key, ETH_AF_PACKET_NUM_Q_ARG) != NULL) {
 836                         qpairs = atoi(pair->value);
 837                         if (qpairs < 1 ||
 838                             qpairs > RTE_PMD_AF_PACKET_MAX_RINGS) {
 839                                 RTE_LOG(ERR, PMD,
 840                                         "%s: invalid qpairs value\n",
 841                                         name);
 842                                 return -1;
 843                         }
 844                         continue;
 845                 }
 846                 if (strstr(pair->key, ETH_AF_PACKET_BLOCKSIZE_ARG) != NULL) {
 847                         blocksize = atoi(pair->value);
 848                         if (!blocksize) {
 849                                 RTE_LOG(ERR, PMD,
 850                                         "%s: invalid blocksize value\n",
 851                                         name);
 852                                 return -1;
 853                         }
 854                         continue;
 855                 }
 856                 if (strstr(pair->key, ETH_AF_PACKET_FRAMESIZE_ARG) != NULL) {
 857                         framesize = atoi(pair->value);
 858                         if (!framesize) {
 859                                 RTE_LOG(ERR, PMD,
 860                                         "%s: invalid framesize value\n",
 861                                         name);
 862                                 return -1;
 863                         }
 864                         continue;
 865                 }
 866                 if (strstr(pair->key, ETH_AF_PACKET_FRAMECOUNT_ARG) != NULL) {
 867                         framecount = atoi(pair->value);
 868                         if (!framecount) {
 869                                 RTE_LOG(ERR, PMD,
 870                                         "%s: invalid framecount 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(name, *sockfd, qpairs,
 899                                    blocksize, blockcount,
 900                                    framesize, framecount,
 901                                    numa_node, &internals, &eth_dev,
 902                                    kvlist) < 0)
 903                 return -1;
 904
 905         eth_dev->rx_pkt_burst = eth_af_packet_rx;
 906         eth_dev->tx_pkt_burst = eth_af_packet_tx;
 907
 908         return 0;
 909 }
 910
 911 static int
 912 rte_pmd_af_packet_probe(const char *name, const char *params)
 913 {
 914         unsigned numa_node;
 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", name);
 920
 921         numa_node = rte_socket_id();
 922
 923         kvlist = rte_kvargs_parse(params, valid_arguments);
 924         if (kvlist == NULL) {
 925                 ret = -1;
 926                 goto exit;
 927         }
 928
 929         /*
 930          * If iface argument is passed we open the NICs and use them for
 931          * reading / writing
 932          */
 933         if (rte_kvargs_count(kvlist, ETH_AF_PACKET_IFACE_ARG) == 1) {
 934
 935                 ret = rte_kvargs_process(kvlist, ETH_AF_PACKET_IFACE_ARG,
 936                                          &open_packet_iface, &sockfd);
 937                 if (ret < 0)
 938                         goto exit;
 939         }
 940
 941         ret = rte_eth_from_packet(name, &sockfd, numa_node, kvlist);
 942         close(sockfd); /* no longer needed */
 943
 944 exit:
 945         rte_kvargs_free(kvlist);
 946         return ret;
 947 }
 948
 949 static int
 950 rte_pmd_af_packet_remove(const char *name)
 951 {
 952         struct rte_eth_dev *eth_dev = NULL;
 953         struct pmd_internals *internals;
 954         unsigned q;
 955
 956         RTE_LOG(INFO, PMD, "Closing AF_PACKET ethdev on numa socket %u\n",
 957                         rte_socket_id());
 958
 959         if (name == NULL)
 960                 return -1;
 961
 962         /* find the ethdev entry */
 963         eth_dev = rte_eth_dev_allocated(name);
 964         if (eth_dev == NULL)
 965                 return -1;
 966
 967         internals = eth_dev->data->dev_private;
 968         for (q = 0; q < internals->nb_queues; q++) {
 969                 rte_free(internals->rx_queue[q].rd);
 970                 rte_free(internals->tx_queue[q].rd);
 971         }
 972         free(internals->if_name);
 973
 974         rte_free(eth_dev->data->dev_private);
 975         rte_free(eth_dev->data);
 976
 977         rte_eth_dev_release_port(eth_dev);
 978
 979         return 0;
 980 }
 981
 982 static struct rte_vdev_driver pmd_af_packet_drv = {
 983         .probe = rte_pmd_af_packet_probe,
 984         .remove = rte_pmd_af_packet_remove,
 985 };
 986
 987 RTE_PMD_REGISTER_VDEV(net_af_packet, pmd_af_packet_drv);
 988 RTE_PMD_REGISTER_ALIAS(net_af_packet, eth_af_packet);
 989 RTE_PMD_REGISTER_PARAM_STRING(net_af_packet,
 990         "iface=<string> "
 991         "qpairs=<int> "
 992         "blocksz=<int> "
 993         "framesz=<int> "
 994         "framecnt=<int>");