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