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