drivers/net/af_packet/rte_eth_af_packet.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2014 John W. Linville <linville@tuxdriver.com>
   5  *
   6  *   Originally based upon librte_pmd_pcap code:
   7  *
   8  *   Copyright(c) 2010-2015 Intel Corporation. All rights reserved.
   9  *   Copyright(c) 2014 6WIND S.A.
  10  *   All rights reserved.
  11  *
  12  *   Redistribution and use in source and binary forms, with or without
  13  *   modification, are permitted provided that the following conditions
  14  *   are met:
  15  *
  16  *     * Redistributions of source code must retain the above copyright
  17  *       notice, this list of conditions and the following disclaimer.
  18  *     * Redistributions in binary form must reproduce the above copyright
  19  *       notice, this list of conditions and the following disclaimer in
  20  *       the documentation and/or other materials provided with the
  21  *       distribution.
  22  *     * Neither the name of Intel Corporation nor the names of its
  23  *       contributors may be used to endorse or promote products derived
  24  *       from this software without specific prior written permission.
  25  *
  26  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  27  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  28  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  29  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  30  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  31  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  32  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  33  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  34  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  35  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  36  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  37  */
  38
  39 #include <rte_mbuf.h>
  40 #include <rte_ethdev.h>
  41 #include <rte_ethdev_vdev.h>
  42 #include <rte_malloc.h>
  43 #include <rte_kvargs.h>
  44 #include <rte_vdev.h>
  45
  46 #include <linux/if_ether.h>
  47 #include <linux/if_packet.h>
  48 #include <arpa/inet.h>
  49 #include <net/if.h>
  50 #include <sys/types.h>
  51 #include <sys/socket.h>
  52 #include <sys/ioctl.h>
  53 #include <sys/mman.h>
  54 #include <unistd.h>
  55 #include <poll.h>
  56
  57 #define ETH_AF_PACKET_IFACE_ARG         "iface"
  58 #define ETH_AF_PACKET_NUM_Q_ARG         "qpairs"
  59 #define ETH_AF_PACKET_BLOCKSIZE_ARG     "blocksz"
  60 #define ETH_AF_PACKET_FRAMESIZE_ARG     "framesz"
  61 #define ETH_AF_PACKET_FRAMECOUNT_ARG    "framecnt"
  62
  63 #define DFLT_BLOCK_SIZE         (1 << 12)
  64 #define DFLT_FRAME_SIZE         (1 << 11)
  65 #define DFLT_FRAME_COUNT        (1 << 9)
  66
  67 #define RTE_PMD_AF_PACKET_MAX_RINGS 16
  68
  69 struct pkt_rx_queue {
  70         int sockfd;
  71
  72         struct iovec *rd;
  73         uint8_t *map;
  74         unsigned int framecount;
  75         unsigned int framenum;
  76
  77         struct rte_mempool *mb_pool;
  78         uint8_t in_port;
  79
  80         volatile unsigned long rx_pkts;
  81         volatile unsigned long err_pkts;
  82         volatile unsigned long rx_bytes;
  83 };
  84
  85 struct pkt_tx_queue {
  86         int sockfd;
  87         unsigned int frame_data_size;
  88
  89         struct iovec *rd;
  90         uint8_t *map;
  91         unsigned int framecount;
  92         unsigned int framenum;
  93
  94         volatile unsigned long tx_pkts;
  95         volatile unsigned long err_pkts;
  96         volatile unsigned long tx_bytes;
  97 };
  98
  99 struct pmd_internals {
 100         unsigned nb_queues;
 101
 102         int if_index;
 103         char *if_name;
 104         struct ether_addr eth_addr;
 105
 106         struct tpacket_req req;
 107
 108         struct pkt_rx_queue rx_queue[RTE_PMD_AF_PACKET_MAX_RINGS];
 109         struct pkt_tx_queue tx_queue[RTE_PMD_AF_PACKET_MAX_RINGS];
 110 };
 111
 112 static const char *valid_arguments[] = {
 113         ETH_AF_PACKET_IFACE_ARG,
 114         ETH_AF_PACKET_NUM_Q_ARG,
 115         ETH_AF_PACKET_BLOCKSIZE_ARG,
 116         ETH_AF_PACKET_FRAMESIZE_ARG,
 117         ETH_AF_PACKET_FRAMECOUNT_ARG,
 118         NULL
 119 };
 120
 121 static struct rte_eth_link pmd_link = {
 122         .link_speed = ETH_SPEED_NUM_10G,
 123         .link_duplex = ETH_LINK_FULL_DUPLEX,
 124         .link_status = ETH_LINK_DOWN,
 125         .link_autoneg = ETH_LINK_SPEED_AUTONEG
 126 };
 127
 128 static uint16_t
 129 eth_af_packet_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 130 {
 131         unsigned i;
 132         struct tpacket2_hdr *ppd;
 133         struct rte_mbuf *mbuf;
 134         uint8_t *pbuf;
 135         struct pkt_rx_queue *pkt_q = queue;
 136         uint16_t num_rx = 0;
 137         unsigned long num_rx_bytes = 0;
 138         unsigned int framecount, framenum;
 139
 140         if (unlikely(nb_pkts == 0))
 141                 return 0;
 142
 143         /*
 144          * Reads the given number of packets from the AF_PACKET socket one by
 145          * one and copies the packet data into a newly allocated mbuf.
 146          */
 147         framecount = pkt_q->framecount;
 148         framenum = pkt_q->framenum;
 149         for (i = 0; i < nb_pkts; i++) {
 150                 /* point at the next incoming frame */
 151                 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
 152                 if ((ppd->tp_status & TP_STATUS_USER) == 0)
 153                         break;
 154
 155                 /* allocate the next mbuf */
 156                 mbuf = rte_pktmbuf_alloc(pkt_q->mb_pool);
 157                 if (unlikely(mbuf == NULL))
 158                         break;
 159
 160                 /* packet will fit in the mbuf, go ahead and receive it */
 161                 rte_pktmbuf_pkt_len(mbuf) = rte_pktmbuf_data_len(mbuf) = ppd->tp_snaplen;
 162                 pbuf = (uint8_t *) ppd + ppd->tp_mac;
 163                 memcpy(rte_pktmbuf_mtod(mbuf, void *), pbuf, rte_pktmbuf_data_len(mbuf));
 164
 165                 /* check for vlan info */
 166                 if (ppd->tp_status & TP_STATUS_VLAN_VALID) {
 167                         mbuf->vlan_tci = ppd->tp_vlan_tci;
 168                         mbuf->ol_flags |= (PKT_RX_VLAN_PKT | PKT_RX_VLAN_STRIPPED);
 169                 }
 170
 171                 /* release incoming frame and advance ring buffer */
 172                 ppd->tp_status = TP_STATUS_KERNEL;
 173                 if (++framenum >= framecount)
 174                         framenum = 0;
 175                 mbuf->port = pkt_q->in_port;
 176
 177                 /* account for the receive frame */
 178                 bufs[i] = mbuf;
 179                 num_rx++;
 180                 num_rx_bytes += mbuf->pkt_len;
 181         }
 182         pkt_q->framenum = framenum;
 183         pkt_q->rx_pkts += num_rx;
 184         pkt_q->rx_bytes += num_rx_bytes;
 185         return num_rx;
 186 }
 187
 188 /*
 189  * Callback to handle sending packets through a real NIC.
 190  */
 191 static uint16_t
 192 eth_af_packet_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 193 {
 194         struct tpacket2_hdr *ppd;
 195         struct rte_mbuf *mbuf;
 196         uint8_t *pbuf;
 197         unsigned int framecount, framenum;
 198         struct pollfd pfd;
 199         struct pkt_tx_queue *pkt_q = queue;
 200         uint16_t num_tx = 0;
 201         unsigned long num_tx_bytes = 0;
 202         int i;
 203
 204         if (unlikely(nb_pkts == 0))
 205                 return 0;
 206
 207         memset(&pfd, 0, sizeof(pfd));
 208         pfd.fd = pkt_q->sockfd;
 209         pfd.events = POLLOUT;
 210         pfd.revents = 0;
 211
 212         framecount = pkt_q->framecount;
 213         framenum = pkt_q->framenum;
 214         ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
 215         for (i = 0; i < nb_pkts; i++) {
 216                 mbuf = *bufs++;
 217
 218                 /* drop oversized packets */
 219                 if (rte_pktmbuf_data_len(mbuf) > pkt_q->frame_data_size) {
 220                         rte_pktmbuf_free(mbuf);
 221                         continue;
 222                 }
 223
 224                 /* insert vlan info if necessary */
 225                 if (mbuf->ol_flags & PKT_TX_VLAN_PKT) {
 226                         if (rte_vlan_insert(&mbuf)) {
 227                                 rte_pktmbuf_free(mbuf);
 228                                 continue;
 229                         }
 230                 }
 231
 232                 /* point at the next incoming frame */
 233                 if ((ppd->tp_status != TP_STATUS_AVAILABLE) &&
 234                     (poll(&pfd, 1, -1) < 0))
 235                         break;
 236
 237                 /* copy the tx frame data */
 238                 pbuf = (uint8_t *) ppd + TPACKET2_HDRLEN -
 239                         sizeof(struct sockaddr_ll);
 240                 memcpy(pbuf, rte_pktmbuf_mtod(mbuf, void*), rte_pktmbuf_data_len(mbuf));
 241                 ppd->tp_len = ppd->tp_snaplen = rte_pktmbuf_data_len(mbuf);
 242
 243                 /* release incoming frame and advance ring buffer */
 244                 ppd->tp_status = TP_STATUS_SEND_REQUEST;
 245                 if (++framenum >= framecount)
 246                         framenum = 0;
 247                 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
 248
 249                 num_tx++;
 250                 num_tx_bytes += mbuf->pkt_len;
 251                 rte_pktmbuf_free(mbuf);
 252         }
 253
 254         /* kick-off transmits */
 255         if (sendto(pkt_q->sockfd, NULL, 0, MSG_DONTWAIT, NULL, 0) == -1)
 256                 num_tx = 0; /* error sending -- no packets transmitted */
 257
 258         pkt_q->framenum = framenum;
 259         pkt_q->tx_pkts += num_tx;
 260         pkt_q->err_pkts += i - num_tx;
 261         pkt_q->tx_bytes += num_tx_bytes;
 262         return i;
 263 }
 264
 265 static int
 266 eth_dev_start(struct rte_eth_dev *dev)
 267 {
 268         dev->data->dev_link.link_status = ETH_LINK_UP;
 269         return 0;
 270 }
 271
 272 /*
 273  * This function gets called when the current port gets stopped.
 274  */
 275 static void
 276 eth_dev_stop(struct rte_eth_dev *dev)
 277 {
 278         unsigned i;
 279         int sockfd;
 280         struct pmd_internals *internals = dev->data->dev_private;
 281
 282         for (i = 0; i < internals->nb_queues; i++) {
 283                 sockfd = internals->rx_queue[i].sockfd;
 284                 if (sockfd != -1)
 285                         close(sockfd);
 286
 287                 /* Prevent use after free in case tx fd == rx fd */
 288                 if (sockfd != internals->tx_queue[i].sockfd) {
 289                         sockfd = internals->tx_queue[i].sockfd;
 290                         if (sockfd != -1)
 291                                 close(sockfd);
 292                 }
 293
 294                 internals->rx_queue[i].sockfd = -1;
 295                 internals->tx_queue[i].sockfd = -1;
 296         }
 297
 298         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 299 }
 300
 301 static int
 302 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 303 {
 304         return 0;
 305 }
 306
 307 static void
 308 eth_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 309 {
 310         struct pmd_internals *internals = dev->data->dev_private;
 311
 312         dev_info->if_index = internals->if_index;
 313         dev_info->max_mac_addrs = 1;
 314         dev_info->max_rx_pktlen = (uint32_t)ETH_FRAME_LEN;
 315         dev_info->max_rx_queues = (uint16_t)internals->nb_queues;
 316         dev_info->max_tx_queues = (uint16_t)internals->nb_queues;
 317         dev_info->min_rx_bufsize = 0;
 318 }
 319
 320 static void
 321 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *igb_stats)
 322 {
 323         unsigned i, imax;
 324         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 325         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 326         const struct pmd_internals *internal = dev->data->dev_private;
 327
 328         imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
 329                 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
 330         for (i = 0; i < imax; i++) {
 331                 igb_stats->q_ipackets[i] = internal->rx_queue[i].rx_pkts;
 332                 igb_stats->q_ibytes[i] = internal->rx_queue[i].rx_bytes;
 333                 rx_total += igb_stats->q_ipackets[i];
 334                 rx_bytes_total += igb_stats->q_ibytes[i];
 335         }
 336
 337         imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
 338                 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
 339         for (i = 0; i < imax; i++) {
 340                 igb_stats->q_opackets[i] = internal->tx_queue[i].tx_pkts;
 341                 igb_stats->q_errors[i] = internal->tx_queue[i].err_pkts;
 342                 igb_stats->q_obytes[i] = internal->tx_queue[i].tx_bytes;
 343                 tx_total += igb_stats->q_opackets[i];
 344                 tx_err_total += igb_stats->q_errors[i];
 345                 tx_bytes_total += igb_stats->q_obytes[i];
 346         }
 347
 348         igb_stats->ipackets = rx_total;
 349         igb_stats->ibytes = rx_bytes_total;
 350         igb_stats->opackets = tx_total;
 351         igb_stats->oerrors = tx_err_total;
 352         igb_stats->obytes = tx_bytes_total;
 353 }
 354
 355 static void
 356 eth_stats_reset(struct rte_eth_dev *dev)
 357 {
 358         unsigned i;
 359         struct pmd_internals *internal = dev->data->dev_private;
 360
 361         for (i = 0; i < internal->nb_queues; i++) {
 362                 internal->rx_queue[i].rx_pkts = 0;
 363                 internal->rx_queue[i].rx_bytes = 0;
 364         }
 365
 366         for (i = 0; i < internal->nb_queues; i++) {
 367                 internal->tx_queue[i].tx_pkts = 0;
 368                 internal->tx_queue[i].err_pkts = 0;
 369                 internal->tx_queue[i].tx_bytes = 0;
 370         }
 371 }
 372
 373 static void
 374 eth_dev_close(struct rte_eth_dev *dev __rte_unused)
 375 {
 376 }
 377
 378 static void
 379 eth_queue_release(void *q __rte_unused)
 380 {
 381 }
 382
 383 static int
 384 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 385                 int wait_to_complete __rte_unused)
 386 {
 387         return 0;
 388 }
 389
 390 static int
 391 eth_rx_queue_setup(struct rte_eth_dev *dev,
 392                    uint16_t rx_queue_id,
 393                    uint16_t nb_rx_desc __rte_unused,
 394                    unsigned int socket_id __rte_unused,
 395                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 396                    struct rte_mempool *mb_pool)
 397 {
 398         struct pmd_internals *internals = dev->data->dev_private;
 399         struct pkt_rx_queue *pkt_q = &internals->rx_queue[rx_queue_id];
 400         unsigned int buf_size, data_size;
 401
 402         pkt_q->mb_pool = mb_pool;
 403
 404         /* Now get the space available for data in the mbuf */
 405         buf_size = rte_pktmbuf_data_room_size(pkt_q->mb_pool) -
 406                 RTE_PKTMBUF_HEADROOM;
 407         data_size = internals->req.tp_frame_size;
 408         data_size -= TPACKET2_HDRLEN - sizeof(struct sockaddr_ll);
 409
 410         if (data_size > buf_size) {
 411                 RTE_LOG(ERR, PMD,
 412                         "%s: %d bytes will not fit in mbuf (%d bytes)\n",
 413                         dev->data->name, data_size, buf_size);
 414                 return -ENOMEM;
 415         }
 416
 417         dev->data->rx_queues[rx_queue_id] = pkt_q;
 418         pkt_q->in_port = dev->data->port_id;
 419
 420         return 0;
 421 }
 422
 423 static int
 424 eth_tx_queue_setup(struct rte_eth_dev *dev,
 425                    uint16_t tx_queue_id,
 426                    uint16_t nb_tx_desc __rte_unused,
 427                    unsigned int socket_id __rte_unused,
 428                    const struct rte_eth_txconf *tx_conf __rte_unused)
 429 {
 430
 431         struct pmd_internals *internals = dev->data->dev_private;
 432
 433         dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
 434         return 0;
 435 }
 436
 437 static int
 438 eth_dev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 439 {
 440         struct pmd_internals *internals = dev->data->dev_private;
 441         struct ifreq ifr = { .ifr_mtu = mtu };
 442         int ret;
 443         int s;
 444         unsigned int data_size = internals->req.tp_frame_size -
 445                                  TPACKET2_HDRLEN -
 446                                  sizeof(struct sockaddr_ll);
 447
 448         if (mtu > data_size)
 449                 return -EINVAL;
 450
 451         s = socket(PF_INET, SOCK_DGRAM, 0);
 452         if (s < 0)
 453                 return -EINVAL;
 454
 455         snprintf(ifr.ifr_name, IFNAMSIZ, "%s", internals->if_name);
 456         ret = ioctl(s, SIOCSIFMTU, &ifr);
 457         close(s);
 458
 459         if (ret < 0)
 460                 return -EINVAL;
 461
 462         return 0;
 463 }
 464
 465 static void
 466 eth_dev_change_flags(char *if_name, uint32_t flags, uint32_t mask)
 467 {
 468         struct ifreq ifr;
 469         int s;
 470
 471         s = socket(PF_INET, SOCK_DGRAM, 0);
 472         if (s < 0)
 473                 return;
 474
 475         snprintf(ifr.ifr_name, IFNAMSIZ, "%s", if_name);
 476         if (ioctl(s, SIOCGIFFLAGS, &ifr) < 0)
 477                 goto out;
 478         ifr.ifr_flags &= mask;
 479         ifr.ifr_flags |= flags;
 480         if (ioctl(s, SIOCSIFFLAGS, &ifr) < 0)
 481                 goto out;
 482 out:
 483         close(s);
 484 }
 485
 486 static void
 487 eth_dev_promiscuous_enable(struct rte_eth_dev *dev)
 488 {
 489         struct pmd_internals *internals = dev->data->dev_private;
 490
 491         eth_dev_change_flags(internals->if_name, IFF_PROMISC, ~0);
 492 }
 493
 494 static void
 495 eth_dev_promiscuous_disable(struct rte_eth_dev *dev)
 496 {
 497         struct pmd_internals *internals = dev->data->dev_private;
 498
 499         eth_dev_change_flags(internals->if_name, 0, ~IFF_PROMISC);
 500 }
 501
 502 static const struct eth_dev_ops ops = {
 503         .dev_start = eth_dev_start,
 504         .dev_stop = eth_dev_stop,
 505         .dev_close = eth_dev_close,
 506         .dev_configure = eth_dev_configure,
 507         .dev_infos_get = eth_dev_info,
 508         .mtu_set = eth_dev_mtu_set,
 509         .promiscuous_enable = eth_dev_promiscuous_enable,
 510         .promiscuous_disable = eth_dev_promiscuous_disable,
 511         .rx_queue_setup = eth_rx_queue_setup,
 512         .tx_queue_setup = eth_tx_queue_setup,
 513         .rx_queue_release = eth_queue_release,
 514         .tx_queue_release = eth_queue_release,
 515         .link_update = eth_link_update,
 516         .stats_get = eth_stats_get,
 517         .stats_reset = eth_stats_reset,
 518 };
 519
 520 /*
 521  * Opens an AF_PACKET socket
 522  */
 523 static int
 524 open_packet_iface(const char *key __rte_unused,
 525                   const char *value __rte_unused,
 526                   void *extra_args)
 527 {
 528         int *sockfd = extra_args;
 529
 530         /* Open an AF_PACKET socket... */
 531         *sockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
 532         if (*sockfd == -1) {
 533                 RTE_LOG(ERR, PMD, "Could not open AF_PACKET socket\n");
 534                 return -1;
 535         }
 536
 537         return 0;
 538 }
 539
 540 static struct rte_vdev_driver pmd_af_packet_drv;
 541
 542 static int
 543 rte_pmd_init_internals(struct rte_vdev_device *dev,
 544                        const int sockfd,
 545                        const unsigned nb_queues,
 546                        unsigned int blocksize,
 547                        unsigned int blockcnt,
 548                        unsigned int framesize,
 549                        unsigned int framecnt,
 550                        struct pmd_internals **internals,
 551                        struct rte_eth_dev **eth_dev,
 552                        struct rte_kvargs *kvlist)
 553 {
 554         const char *name = rte_vdev_device_name(dev);
 555         const unsigned int numa_node = dev->device.numa_node;
 556         struct rte_eth_dev_data *data = NULL;
 557         struct rte_kvargs_pair *pair = NULL;
 558         struct ifreq ifr;
 559         size_t ifnamelen;
 560         unsigned k_idx;
 561         struct sockaddr_ll sockaddr;
 562         struct tpacket_req *req;
 563         struct pkt_rx_queue *rx_queue;
 564         struct pkt_tx_queue *tx_queue;
 565         int rc, tpver, discard;
 566         int qsockfd = -1;
 567         unsigned int i, q, rdsize;
 568 #if defined(PACKET_FANOUT)
 569         int fanout_arg;
 570 #endif
 571 #if defined(PACKET_QDISC_BYPASS)
 572         int bypass;
 573 #endif
 574
 575         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 576                 pair = &kvlist->pairs[k_idx];
 577                 if (strstr(pair->key, ETH_AF_PACKET_IFACE_ARG) != NULL)
 578                         break;
 579         }
 580         if (pair == NULL) {
 581                 RTE_LOG(ERR, PMD,
 582                         "%s: no interface specified for AF_PACKET ethdev\n",
 583                         name);
 584                 goto error_early;
 585         }
 586
 587         RTE_LOG(INFO, PMD,
 588                 "%s: creating AF_PACKET-backed ethdev on numa socket %u\n",
 589                 name, numa_node);
 590
 591         /*
 592          * now do all data allocation - for eth_dev structure, dummy pci driver
 593          * and internal (private) data
 594          */
 595         data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
 596         if (data == NULL)
 597                 goto error_early;
 598
 599         *internals = rte_zmalloc_socket(name, sizeof(**internals),
 600                                         0, numa_node);
 601         if (*internals == NULL)
 602                 goto error_early;
 603
 604         for (q = 0; q < nb_queues; q++) {
 605                 (*internals)->rx_queue[q].map = MAP_FAILED;
 606                 (*internals)->tx_queue[q].map = MAP_FAILED;
 607         }
 608
 609         req = &((*internals)->req);
 610
 611         req->tp_block_size = blocksize;
 612         req->tp_block_nr = blockcnt;
 613         req->tp_frame_size = framesize;
 614         req->tp_frame_nr = framecnt;
 615
 616         ifnamelen = strlen(pair->value);
 617         if (ifnamelen < sizeof(ifr.ifr_name)) {
 618                 memcpy(ifr.ifr_name, pair->value, ifnamelen);
 619                 ifr.ifr_name[ifnamelen] = '\0';
 620         } else {
 621                 RTE_LOG(ERR, PMD,
 622                         "%s: I/F name too long (%s)\n",
 623                         name, pair->value);
 624                 goto error_early;
 625         }
 626         if (ioctl(sockfd, SIOCGIFINDEX, &ifr) == -1) {
 627                 RTE_LOG(ERR, PMD,
 628                         "%s: ioctl failed (SIOCGIFINDEX)\n",
 629                         name);
 630                 goto error_early;
 631         }
 632         (*internals)->if_name = strdup(pair->value);
 633         if ((*internals)->if_name == NULL)
 634                 goto error_early;
 635         (*internals)->if_index = ifr.ifr_ifindex;
 636
 637         if (ioctl(sockfd, SIOCGIFHWADDR, &ifr) == -1) {
 638                 RTE_LOG(ERR, PMD,
 639                         "%s: ioctl failed (SIOCGIFHWADDR)\n",
 640                         name);
 641                 goto error_early;
 642         }
 643         memcpy(&(*internals)->eth_addr, ifr.ifr_hwaddr.sa_data, ETH_ALEN);
 644
 645         memset(&sockaddr, 0, sizeof(sockaddr));
 646         sockaddr.sll_family = AF_PACKET;
 647         sockaddr.sll_protocol = htons(ETH_P_ALL);
 648         sockaddr.sll_ifindex = (*internals)->if_index;
 649
 650 #if defined(PACKET_FANOUT)
 651         fanout_arg = (getpid() ^ (*internals)->if_index) & 0xffff;
 652         fanout_arg |= (PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_DEFRAG) << 16;
 653 #if defined(PACKET_FANOUT_FLAG_ROLLOVER)
 654         fanout_arg |= PACKET_FANOUT_FLAG_ROLLOVER << 16;
 655 #endif
 656 #endif
 657
 658         for (q = 0; q < nb_queues; q++) {
 659                 /* Open an AF_PACKET socket for this queue... */
 660                 qsockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
 661                 if (qsockfd == -1) {
 662                         RTE_LOG(ERR, PMD,
 663                                 "%s: could not open AF_PACKET socket\n",
 664                                 name);
 665                         return -1;
 666                 }
 667
 668                 tpver = TPACKET_V2;
 669                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_VERSION,
 670                                 &tpver, sizeof(tpver));
 671                 if (rc == -1) {
 672                         RTE_LOG(ERR, PMD,
 673                                 "%s: could not set PACKET_VERSION on AF_PACKET "
 674                                 "socket for %s\n", name, pair->value);
 675                         goto error;
 676                 }
 677
 678                 discard = 1;
 679                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_LOSS,
 680                                 &discard, sizeof(discard));
 681                 if (rc == -1) {
 682                         RTE_LOG(ERR, PMD,
 683                                 "%s: could not set PACKET_LOSS on "
 684                                 "AF_PACKET socket for %s\n", name, pair->value);
 685                         goto error;
 686                 }
 687
 688 #if defined(PACKET_QDISC_BYPASS)
 689                 bypass = 1;
 690                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_QDISC_BYPASS,
 691                                 &bypass, sizeof(bypass));
 692                 if (rc == -1) {
 693                         RTE_LOG(ERR, PMD,
 694                                 "%s: could not set PACKET_QDISC_BYPASS "
 695                                 "on AF_PACKET socket for %s\n", name,
 696                                 pair->value);
 697                         goto error;
 698                 }
 699 #endif
 700
 701                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_RX_RING, req, sizeof(*req));
 702                 if (rc == -1) {
 703                         RTE_LOG(ERR, PMD,
 704                                 "%s: could not set PACKET_RX_RING on AF_PACKET "
 705                                 "socket for %s\n", name, pair->value);
 706                         goto error;
 707                 }
 708
 709                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_TX_RING, req, sizeof(*req));
 710                 if (rc == -1) {
 711                         RTE_LOG(ERR, PMD,
 712                                 "%s: could not set PACKET_TX_RING on AF_PACKET "
 713                                 "socket for %s\n", name, pair->value);
 714                         goto error;
 715                 }
 716
 717                 rx_queue = &((*internals)->rx_queue[q]);
 718                 rx_queue->framecount = req->tp_frame_nr;
 719
 720                 rx_queue->map = mmap(NULL, 2 * req->tp_block_size * req->tp_block_nr,
 721                                     PROT_READ | PROT_WRITE, MAP_SHARED | MAP_LOCKED,
 722                                     qsockfd, 0);
 723                 if (rx_queue->map == MAP_FAILED) {
 724                         RTE_LOG(ERR, PMD,
 725                                 "%s: call to mmap failed on AF_PACKET socket for %s\n",
 726                                 name, pair->value);
 727                         goto error;
 728                 }
 729
 730                 /* rdsize is same for both Tx and Rx */
 731                 rdsize = req->tp_frame_nr * sizeof(*(rx_queue->rd));
 732
 733                 rx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
 734                 if (rx_queue->rd == NULL)
 735                         goto error;
 736                 for (i = 0; i < req->tp_frame_nr; ++i) {
 737                         rx_queue->rd[i].iov_base = rx_queue->map + (i * framesize);
 738                         rx_queue->rd[i].iov_len = req->tp_frame_size;
 739                 }
 740                 rx_queue->sockfd = qsockfd;
 741
 742                 tx_queue = &((*internals)->tx_queue[q]);
 743                 tx_queue->framecount = req->tp_frame_nr;
 744                 tx_queue->frame_data_size = req->tp_frame_size;
 745                 tx_queue->frame_data_size -= TPACKET2_HDRLEN -
 746                         sizeof(struct sockaddr_ll);
 747
 748                 tx_queue->map = rx_queue->map + req->tp_block_size * req->tp_block_nr;
 749
 750                 tx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
 751                 if (tx_queue->rd == NULL)
 752                         goto error;
 753                 for (i = 0; i < req->tp_frame_nr; ++i) {
 754                         tx_queue->rd[i].iov_base = tx_queue->map + (i * framesize);
 755                         tx_queue->rd[i].iov_len = req->tp_frame_size;
 756                 }
 757                 tx_queue->sockfd = qsockfd;
 758
 759                 rc = bind(qsockfd, (const struct sockaddr*)&sockaddr, sizeof(sockaddr));
 760                 if (rc == -1) {
 761                         RTE_LOG(ERR, PMD,
 762                                 "%s: could not bind AF_PACKET socket to %s\n",
 763                                 name, pair->value);
 764                         goto error;
 765                 }
 766
 767 #if defined(PACKET_FANOUT)
 768                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_FANOUT,
 769                                 &fanout_arg, sizeof(fanout_arg));
 770                 if (rc == -1) {
 771                         RTE_LOG(ERR, PMD,
 772                                 "%s: could not set PACKET_FANOUT on AF_PACKET socket "
 773                                 "for %s\n", name, pair->value);
 774                         goto error;
 775                 }
 776 #endif
 777         }
 778
 779         /* reserve an ethdev entry */
 780         *eth_dev = rte_eth_vdev_allocate(dev, 0);
 781         if (*eth_dev == NULL)
 782                 goto error;
 783
 784         /*
 785          * now put it all together
 786          * - store queue data in internals,
 787          * - store numa_node in eth_dev
 788          * - point eth_dev_data to internals
 789          * - and point eth_dev structure to new eth_dev_data structure
 790          */
 791
 792         (*internals)->nb_queues = nb_queues;
 793
 794         rte_memcpy(data, (*eth_dev)->data, sizeof(*data));
 795         data->dev_private = *internals;
 796         data->nb_rx_queues = (uint16_t)nb_queues;
 797         data->nb_tx_queues = (uint16_t)nb_queues;
 798         data->dev_link = pmd_link;
 799         data->mac_addrs = &(*internals)->eth_addr;
 800
 801         (*eth_dev)->data = data;
 802         (*eth_dev)->dev_ops = &ops;
 803         (*eth_dev)->data->dev_flags = RTE_ETH_DEV_DETACHABLE;
 804
 805         return 0;
 806
 807 error:
 808         if (qsockfd != -1)
 809                 close(qsockfd);
 810         for (q = 0; q < nb_queues; q++) {
 811                 munmap((*internals)->rx_queue[q].map,
 812                        2 * req->tp_block_size * req->tp_block_nr);
 813
 814                 rte_free((*internals)->rx_queue[q].rd);
 815                 rte_free((*internals)->tx_queue[q].rd);
 816                 if (((*internals)->rx_queue[q].sockfd != 0) &&
 817                         ((*internals)->rx_queue[q].sockfd != qsockfd))
 818                         close((*internals)->rx_queue[q].sockfd);
 819         }
 820         free((*internals)->if_name);
 821         rte_free(*internals);
 822 error_early:
 823         rte_free(data);
 824         return -1;
 825 }
 826
 827 static int
 828 rte_eth_from_packet(struct rte_vdev_device *dev,
 829                     int const *sockfd,
 830                     struct rte_kvargs *kvlist)
 831 {
 832         const char *name = rte_vdev_device_name(dev);
 833         struct pmd_internals *internals = NULL;
 834         struct rte_eth_dev *eth_dev = NULL;
 835         struct rte_kvargs_pair *pair = NULL;
 836         unsigned k_idx;
 837         unsigned int blockcount;
 838         unsigned int blocksize = DFLT_BLOCK_SIZE;
 839         unsigned int framesize = DFLT_FRAME_SIZE;
 840         unsigned int framecount = DFLT_FRAME_COUNT;
 841         unsigned int qpairs = 1;
 842
 843         /* do some parameter checking */
 844         if (*sockfd < 0)
 845                 return -1;
 846
 847         /*
 848          * Walk arguments for configurable settings
 849          */
 850         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 851                 pair = &kvlist->pairs[k_idx];
 852                 if (strstr(pair->key, ETH_AF_PACKET_NUM_Q_ARG) != NULL) {
 853                         qpairs = atoi(pair->value);
 854                         if (qpairs < 1 ||
 855                             qpairs > RTE_PMD_AF_PACKET_MAX_RINGS) {
 856                                 RTE_LOG(ERR, PMD,
 857                                         "%s: invalid qpairs value\n",
 858                                         name);
 859                                 return -1;
 860                         }
 861                         continue;
 862                 }
 863                 if (strstr(pair->key, ETH_AF_PACKET_BLOCKSIZE_ARG) != NULL) {
 864                         blocksize = atoi(pair->value);
 865                         if (!blocksize) {
 866                                 RTE_LOG(ERR, PMD,
 867                                         "%s: invalid blocksize value\n",
 868                                         name);
 869                                 return -1;
 870                         }
 871                         continue;
 872                 }
 873                 if (strstr(pair->key, ETH_AF_PACKET_FRAMESIZE_ARG) != NULL) {
 874                         framesize = atoi(pair->value);
 875                         if (!framesize) {
 876                                 RTE_LOG(ERR, PMD,
 877                                         "%s: invalid framesize value\n",
 878                                         name);
 879                                 return -1;
 880                         }
 881                         continue;
 882                 }
 883                 if (strstr(pair->key, ETH_AF_PACKET_FRAMECOUNT_ARG) != NULL) {
 884                         framecount = atoi(pair->value);
 885                         if (!framecount) {
 886                                 RTE_LOG(ERR, PMD,
 887                                         "%s: invalid framecount value\n",
 888                                         name);
 889                                 return -1;
 890                         }
 891                         continue;
 892                 }
 893         }
 894
 895         if (framesize > blocksize) {
 896                 RTE_LOG(ERR, PMD,
 897                         "%s: AF_PACKET MMAP frame size exceeds block size!\n",
 898                         name);
 899                 return -1;
 900         }
 901
 902         blockcount = framecount / (blocksize / framesize);
 903         if (!blockcount) {
 904                 RTE_LOG(ERR, PMD,
 905                         "%s: invalid AF_PACKET MMAP parameters\n", name);
 906                 return -1;
 907         }
 908
 909         RTE_LOG(INFO, PMD, "%s: AF_PACKET MMAP parameters:\n", name);
 910         RTE_LOG(INFO, PMD, "%s:\tblock size %d\n", name, blocksize);
 911         RTE_LOG(INFO, PMD, "%s:\tblock count %d\n", name, blockcount);
 912         RTE_LOG(INFO, PMD, "%s:\tframe size %d\n", name, framesize);
 913         RTE_LOG(INFO, PMD, "%s:\tframe count %d\n", name, framecount);
 914
 915         if (rte_pmd_init_internals(dev, *sockfd, qpairs,
 916                                    blocksize, blockcount,
 917                                    framesize, framecount,
 918                                    &internals, &eth_dev,
 919                                    kvlist) < 0)
 920                 return -1;
 921
 922         eth_dev->rx_pkt_burst = eth_af_packet_rx;
 923         eth_dev->tx_pkt_burst = eth_af_packet_tx;
 924
 925         return 0;
 926 }
 927
 928 static int
 929 rte_pmd_af_packet_probe(struct rte_vdev_device *dev)
 930 {
 931         int ret = 0;
 932         struct rte_kvargs *kvlist;
 933         int sockfd = -1;
 934
 935         RTE_LOG(INFO, PMD, "Initializing pmd_af_packet for %s\n",
 936                 rte_vdev_device_name(dev));
 937
 938         kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_arguments);
 939         if (kvlist == NULL) {
 940                 ret = -1;
 941                 goto exit;
 942         }
 943
 944         /*
 945          * If iface argument is passed we open the NICs and use them for
 946          * reading / writing
 947          */
 948         if (rte_kvargs_count(kvlist, ETH_AF_PACKET_IFACE_ARG) == 1) {
 949
 950                 ret = rte_kvargs_process(kvlist, ETH_AF_PACKET_IFACE_ARG,
 951                                          &open_packet_iface, &sockfd);
 952                 if (ret < 0)
 953                         goto exit;
 954         }
 955
 956         if (dev->device.numa_node == SOCKET_ID_ANY)
 957                 dev->device.numa_node = rte_socket_id();
 958
 959         ret = rte_eth_from_packet(dev, &sockfd, kvlist);
 960         close(sockfd); /* no longer needed */
 961
 962 exit:
 963         rte_kvargs_free(kvlist);
 964         return ret;
 965 }
 966
 967 static int
 968 rte_pmd_af_packet_remove(struct rte_vdev_device *dev)
 969 {
 970         struct rte_eth_dev *eth_dev = NULL;
 971         struct pmd_internals *internals;
 972         unsigned q;
 973
 974         RTE_LOG(INFO, PMD, "Closing AF_PACKET ethdev on numa socket %u\n",
 975                         rte_socket_id());
 976
 977         if (dev == NULL)
 978                 return -1;
 979
 980         /* find the ethdev entry */
 981         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
 982         if (eth_dev == NULL)
 983                 return -1;
 984
 985         internals = eth_dev->data->dev_private;
 986         for (q = 0; q < internals->nb_queues; q++) {
 987                 rte_free(internals->rx_queue[q].rd);
 988                 rte_free(internals->tx_queue[q].rd);
 989         }
 990         free(internals->if_name);
 991
 992         rte_free(eth_dev->data->dev_private);
 993         rte_free(eth_dev->data);
 994
 995         rte_eth_dev_release_port(eth_dev);
 996
 997         return 0;
 998 }
 999
1000 static struct rte_vdev_driver pmd_af_packet_drv = {
1001         .probe = rte_pmd_af_packet_probe,
1002         .remove = rte_pmd_af_packet_remove,
1003 };
1004
1005 RTE_PMD_REGISTER_VDEV(net_af_packet, pmd_af_packet_drv);
1006 RTE_PMD_REGISTER_ALIAS(net_af_packet, eth_af_packet);
1007 RTE_PMD_REGISTER_PARAM_STRING(net_af_packet,
1008         "iface=<string> "
1009         "qpairs=<int> "
1010         "blocksz=<int> "
1011         "framesz=<int> "
1012         "framecnt=<int>");