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 #define ETH_AF_PACKET_QDISC_BYPASS_ARG  "qdisc_bypass"
  63
  64 #define DFLT_BLOCK_SIZE         (1 << 12)
  65 #define DFLT_FRAME_SIZE         (1 << 11)
  66 #define DFLT_FRAME_COUNT        (1 << 9)
  67
  68 #define RTE_PMD_AF_PACKET_MAX_RINGS 16
  69
  70 struct pkt_rx_queue {
  71         int sockfd;
  72
  73         struct iovec *rd;
  74         uint8_t *map;
  75         unsigned int framecount;
  76         unsigned int framenum;
  77
  78         struct rte_mempool *mb_pool;
  79         uint16_t in_port;
  80
  81         volatile unsigned long rx_pkts;
  82         volatile unsigned long err_pkts;
  83         volatile unsigned long rx_bytes;
  84 };
  85
  86 struct pkt_tx_queue {
  87         int sockfd;
  88         unsigned int frame_data_size;
  89
  90         struct iovec *rd;
  91         uint8_t *map;
  92         unsigned int framecount;
  93         unsigned int framenum;
  94
  95         volatile unsigned long tx_pkts;
  96         volatile unsigned long err_pkts;
  97         volatile unsigned long tx_bytes;
  98 };
  99
 100 struct pmd_internals {
 101         unsigned nb_queues;
 102
 103         int if_index;
 104         char *if_name;
 105         struct ether_addr eth_addr;
 106
 107         struct tpacket_req req;
 108
 109         struct pkt_rx_queue rx_queue[RTE_PMD_AF_PACKET_MAX_RINGS];
 110         struct pkt_tx_queue tx_queue[RTE_PMD_AF_PACKET_MAX_RINGS];
 111 };
 112
 113 static const char *valid_arguments[] = {
 114         ETH_AF_PACKET_IFACE_ARG,
 115         ETH_AF_PACKET_NUM_Q_ARG,
 116         ETH_AF_PACKET_BLOCKSIZE_ARG,
 117         ETH_AF_PACKET_FRAMESIZE_ARG,
 118         ETH_AF_PACKET_FRAMECOUNT_ARG,
 119         ETH_AF_PACKET_QDISC_BYPASS_ARG,
 120         NULL
 121 };
 122
 123 static struct rte_eth_link pmd_link = {
 124         .link_speed = ETH_SPEED_NUM_10G,
 125         .link_duplex = ETH_LINK_FULL_DUPLEX,
 126         .link_status = ETH_LINK_DOWN,
 127         .link_autoneg = ETH_LINK_SPEED_AUTONEG
 128 };
 129
 130 static uint16_t
 131 eth_af_packet_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 132 {
 133         unsigned i;
 134         struct tpacket2_hdr *ppd;
 135         struct rte_mbuf *mbuf;
 136         uint8_t *pbuf;
 137         struct pkt_rx_queue *pkt_q = queue;
 138         uint16_t num_rx = 0;
 139         unsigned long num_rx_bytes = 0;
 140         unsigned int framecount, framenum;
 141
 142         if (unlikely(nb_pkts == 0))
 143                 return 0;
 144
 145         /*
 146          * Reads the given number of packets from the AF_PACKET socket one by
 147          * one and copies the packet data into a newly allocated mbuf.
 148          */
 149         framecount = pkt_q->framecount;
 150         framenum = pkt_q->framenum;
 151         for (i = 0; i < nb_pkts; i++) {
 152                 /* point at the next incoming frame */
 153                 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
 154                 if ((ppd->tp_status & TP_STATUS_USER) == 0)
 155                         break;
 156
 157                 /* allocate the next mbuf */
 158                 mbuf = rte_pktmbuf_alloc(pkt_q->mb_pool);
 159                 if (unlikely(mbuf == NULL))
 160                         break;
 161
 162                 /* packet will fit in the mbuf, go ahead and receive it */
 163                 rte_pktmbuf_pkt_len(mbuf) = rte_pktmbuf_data_len(mbuf) = ppd->tp_snaplen;
 164                 pbuf = (uint8_t *) ppd + ppd->tp_mac;
 165                 memcpy(rte_pktmbuf_mtod(mbuf, void *), pbuf, rte_pktmbuf_data_len(mbuf));
 166
 167                 /* check for vlan info */
 168                 if (ppd->tp_status & TP_STATUS_VLAN_VALID) {
 169                         mbuf->vlan_tci = ppd->tp_vlan_tci;
 170                         mbuf->ol_flags |= (PKT_RX_VLAN_PKT | PKT_RX_VLAN_STRIPPED);
 171                 }
 172
 173                 /* release incoming frame and advance ring buffer */
 174                 ppd->tp_status = TP_STATUS_KERNEL;
 175                 if (++framenum >= framecount)
 176                         framenum = 0;
 177                 mbuf->port = pkt_q->in_port;
 178
 179                 /* account for the receive frame */
 180                 bufs[i] = mbuf;
 181                 num_rx++;
 182                 num_rx_bytes += mbuf->pkt_len;
 183         }
 184         pkt_q->framenum = framenum;
 185         pkt_q->rx_pkts += num_rx;
 186         pkt_q->rx_bytes += num_rx_bytes;
 187         return num_rx;
 188 }
 189
 190 /*
 191  * Callback to handle sending packets through a real NIC.
 192  */
 193 static uint16_t
 194 eth_af_packet_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 195 {
 196         struct tpacket2_hdr *ppd;
 197         struct rte_mbuf *mbuf;
 198         uint8_t *pbuf;
 199         unsigned int framecount, framenum;
 200         struct pollfd pfd;
 201         struct pkt_tx_queue *pkt_q = queue;
 202         uint16_t num_tx = 0;
 203         unsigned long num_tx_bytes = 0;
 204         int i;
 205
 206         if (unlikely(nb_pkts == 0))
 207                 return 0;
 208
 209         memset(&pfd, 0, sizeof(pfd));
 210         pfd.fd = pkt_q->sockfd;
 211         pfd.events = POLLOUT;
 212         pfd.revents = 0;
 213
 214         framecount = pkt_q->framecount;
 215         framenum = pkt_q->framenum;
 216         ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
 217         for (i = 0; i < nb_pkts; i++) {
 218                 mbuf = *bufs++;
 219
 220                 /* drop oversized packets */
 221                 if (mbuf->pkt_len > pkt_q->frame_data_size) {
 222                         rte_pktmbuf_free(mbuf);
 223                         continue;
 224                 }
 225
 226                 /* insert vlan info if necessary */
 227                 if (mbuf->ol_flags & PKT_TX_VLAN_PKT) {
 228                         if (rte_vlan_insert(&mbuf)) {
 229                                 rte_pktmbuf_free(mbuf);
 230                                 continue;
 231                         }
 232                 }
 233
 234                 /* point at the next incoming frame */
 235                 if ((ppd->tp_status != TP_STATUS_AVAILABLE) &&
 236                     (poll(&pfd, 1, -1) < 0))
 237                         break;
 238
 239                 /* copy the tx frame data */
 240                 pbuf = (uint8_t *) ppd + TPACKET2_HDRLEN -
 241                         sizeof(struct sockaddr_ll);
 242
 243                 struct rte_mbuf *tmp_mbuf = mbuf;
 244                 while (tmp_mbuf) {
 245                         uint16_t data_len = rte_pktmbuf_data_len(tmp_mbuf);
 246                         memcpy(pbuf, rte_pktmbuf_mtod(tmp_mbuf, void*), data_len);
 247                         pbuf += data_len;
 248                         tmp_mbuf = tmp_mbuf->next;
 249                 }
 250
 251                 ppd->tp_len = mbuf->pkt_len;
 252                 ppd->tp_snaplen = mbuf->pkt_len;
 253
 254                 /* release incoming frame and advance ring buffer */
 255                 ppd->tp_status = TP_STATUS_SEND_REQUEST;
 256                 if (++framenum >= framecount)
 257                         framenum = 0;
 258                 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
 259
 260                 num_tx++;
 261                 num_tx_bytes += mbuf->pkt_len;
 262                 rte_pktmbuf_free(mbuf);
 263         }
 264
 265         /* kick-off transmits */
 266         if (sendto(pkt_q->sockfd, NULL, 0, MSG_DONTWAIT, NULL, 0) == -1) {
 267                 /* error sending -- no packets transmitted */
 268                 num_tx = 0;
 269                 num_tx_bytes = 0;
 270         }
 271
 272         pkt_q->framenum = framenum;
 273         pkt_q->tx_pkts += num_tx;
 274         pkt_q->err_pkts += i - num_tx;
 275         pkt_q->tx_bytes += num_tx_bytes;
 276         return i;
 277 }
 278
 279 static int
 280 eth_dev_start(struct rte_eth_dev *dev)
 281 {
 282         dev->data->dev_link.link_status = ETH_LINK_UP;
 283         return 0;
 284 }
 285
 286 /*
 287  * This function gets called when the current port gets stopped.
 288  */
 289 static void
 290 eth_dev_stop(struct rte_eth_dev *dev)
 291 {
 292         unsigned i;
 293         int sockfd;
 294         struct pmd_internals *internals = dev->data->dev_private;
 295
 296         for (i = 0; i < internals->nb_queues; i++) {
 297                 sockfd = internals->rx_queue[i].sockfd;
 298                 if (sockfd != -1)
 299                         close(sockfd);
 300
 301                 /* Prevent use after free in case tx fd == rx fd */
 302                 if (sockfd != internals->tx_queue[i].sockfd) {
 303                         sockfd = internals->tx_queue[i].sockfd;
 304                         if (sockfd != -1)
 305                                 close(sockfd);
 306                 }
 307
 308                 internals->rx_queue[i].sockfd = -1;
 309                 internals->tx_queue[i].sockfd = -1;
 310         }
 311
 312         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 313 }
 314
 315 static int
 316 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 317 {
 318         return 0;
 319 }
 320
 321 static void
 322 eth_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 323 {
 324         struct pmd_internals *internals = dev->data->dev_private;
 325
 326         dev_info->if_index = internals->if_index;
 327         dev_info->max_mac_addrs = 1;
 328         dev_info->max_rx_pktlen = (uint32_t)ETH_FRAME_LEN;
 329         dev_info->max_rx_queues = (uint16_t)internals->nb_queues;
 330         dev_info->max_tx_queues = (uint16_t)internals->nb_queues;
 331         dev_info->min_rx_bufsize = 0;
 332 }
 333
 334 static void
 335 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *igb_stats)
 336 {
 337         unsigned i, imax;
 338         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 339         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 340         const struct pmd_internals *internal = dev->data->dev_private;
 341
 342         imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
 343                 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
 344         for (i = 0; i < imax; i++) {
 345                 igb_stats->q_ipackets[i] = internal->rx_queue[i].rx_pkts;
 346                 igb_stats->q_ibytes[i] = internal->rx_queue[i].rx_bytes;
 347                 rx_total += igb_stats->q_ipackets[i];
 348                 rx_bytes_total += igb_stats->q_ibytes[i];
 349         }
 350
 351         imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
 352                 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
 353         for (i = 0; i < imax; i++) {
 354                 igb_stats->q_opackets[i] = internal->tx_queue[i].tx_pkts;
 355                 igb_stats->q_errors[i] = internal->tx_queue[i].err_pkts;
 356                 igb_stats->q_obytes[i] = internal->tx_queue[i].tx_bytes;
 357                 tx_total += igb_stats->q_opackets[i];
 358                 tx_err_total += igb_stats->q_errors[i];
 359                 tx_bytes_total += igb_stats->q_obytes[i];
 360         }
 361
 362         igb_stats->ipackets = rx_total;
 363         igb_stats->ibytes = rx_bytes_total;
 364         igb_stats->opackets = tx_total;
 365         igb_stats->oerrors = tx_err_total;
 366         igb_stats->obytes = tx_bytes_total;
 367 }
 368
 369 static void
 370 eth_stats_reset(struct rte_eth_dev *dev)
 371 {
 372         unsigned i;
 373         struct pmd_internals *internal = dev->data->dev_private;
 374
 375         for (i = 0; i < internal->nb_queues; i++) {
 376                 internal->rx_queue[i].rx_pkts = 0;
 377                 internal->rx_queue[i].rx_bytes = 0;
 378         }
 379
 380         for (i = 0; i < internal->nb_queues; i++) {
 381                 internal->tx_queue[i].tx_pkts = 0;
 382                 internal->tx_queue[i].err_pkts = 0;
 383                 internal->tx_queue[i].tx_bytes = 0;
 384         }
 385 }
 386
 387 static void
 388 eth_dev_close(struct rte_eth_dev *dev __rte_unused)
 389 {
 390 }
 391
 392 static void
 393 eth_queue_release(void *q __rte_unused)
 394 {
 395 }
 396
 397 static int
 398 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 399                 int wait_to_complete __rte_unused)
 400 {
 401         return 0;
 402 }
 403
 404 static int
 405 eth_rx_queue_setup(struct rte_eth_dev *dev,
 406                    uint16_t rx_queue_id,
 407                    uint16_t nb_rx_desc __rte_unused,
 408                    unsigned int socket_id __rte_unused,
 409                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 410                    struct rte_mempool *mb_pool)
 411 {
 412         struct pmd_internals *internals = dev->data->dev_private;
 413         struct pkt_rx_queue *pkt_q = &internals->rx_queue[rx_queue_id];
 414         unsigned int buf_size, data_size;
 415
 416         pkt_q->mb_pool = mb_pool;
 417
 418         /* Now get the space available for data in the mbuf */
 419         buf_size = rte_pktmbuf_data_room_size(pkt_q->mb_pool) -
 420                 RTE_PKTMBUF_HEADROOM;
 421         data_size = internals->req.tp_frame_size;
 422         data_size -= TPACKET2_HDRLEN - sizeof(struct sockaddr_ll);
 423
 424         if (data_size > buf_size) {
 425                 RTE_LOG(ERR, PMD,
 426                         "%s: %d bytes will not fit in mbuf (%d bytes)\n",
 427                         dev->device->name, data_size, buf_size);
 428                 return -ENOMEM;
 429         }
 430
 431         dev->data->rx_queues[rx_queue_id] = pkt_q;
 432         pkt_q->in_port = dev->data->port_id;
 433
 434         return 0;
 435 }
 436
 437 static int
 438 eth_tx_queue_setup(struct rte_eth_dev *dev,
 439                    uint16_t tx_queue_id,
 440                    uint16_t nb_tx_desc __rte_unused,
 441                    unsigned int socket_id __rte_unused,
 442                    const struct rte_eth_txconf *tx_conf __rte_unused)
 443 {
 444
 445         struct pmd_internals *internals = dev->data->dev_private;
 446
 447         dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
 448         return 0;
 449 }
 450
 451 static int
 452 eth_dev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 453 {
 454         struct pmd_internals *internals = dev->data->dev_private;
 455         struct ifreq ifr = { .ifr_mtu = mtu };
 456         int ret;
 457         int s;
 458         unsigned int data_size = internals->req.tp_frame_size -
 459                                  TPACKET2_HDRLEN -
 460                                  sizeof(struct sockaddr_ll);
 461
 462         if (mtu > data_size)
 463                 return -EINVAL;
 464
 465         s = socket(PF_INET, SOCK_DGRAM, 0);
 466         if (s < 0)
 467                 return -EINVAL;
 468
 469         snprintf(ifr.ifr_name, IFNAMSIZ, "%s", internals->if_name);
 470         ret = ioctl(s, SIOCSIFMTU, &ifr);
 471         close(s);
 472
 473         if (ret < 0)
 474                 return -EINVAL;
 475
 476         return 0;
 477 }
 478
 479 static void
 480 eth_dev_change_flags(char *if_name, uint32_t flags, uint32_t mask)
 481 {
 482         struct ifreq ifr;
 483         int s;
 484
 485         s = socket(PF_INET, SOCK_DGRAM, 0);
 486         if (s < 0)
 487                 return;
 488
 489         snprintf(ifr.ifr_name, IFNAMSIZ, "%s", if_name);
 490         if (ioctl(s, SIOCGIFFLAGS, &ifr) < 0)
 491                 goto out;
 492         ifr.ifr_flags &= mask;
 493         ifr.ifr_flags |= flags;
 494         if (ioctl(s, SIOCSIFFLAGS, &ifr) < 0)
 495                 goto out;
 496 out:
 497         close(s);
 498 }
 499
 500 static void
 501 eth_dev_promiscuous_enable(struct rte_eth_dev *dev)
 502 {
 503         struct pmd_internals *internals = dev->data->dev_private;
 504
 505         eth_dev_change_flags(internals->if_name, IFF_PROMISC, ~0);
 506 }
 507
 508 static void
 509 eth_dev_promiscuous_disable(struct rte_eth_dev *dev)
 510 {
 511         struct pmd_internals *internals = dev->data->dev_private;
 512
 513         eth_dev_change_flags(internals->if_name, 0, ~IFF_PROMISC);
 514 }
 515
 516 static const struct eth_dev_ops ops = {
 517         .dev_start = eth_dev_start,
 518         .dev_stop = eth_dev_stop,
 519         .dev_close = eth_dev_close,
 520         .dev_configure = eth_dev_configure,
 521         .dev_infos_get = eth_dev_info,
 522         .mtu_set = eth_dev_mtu_set,
 523         .promiscuous_enable = eth_dev_promiscuous_enable,
 524         .promiscuous_disable = eth_dev_promiscuous_disable,
 525         .rx_queue_setup = eth_rx_queue_setup,
 526         .tx_queue_setup = eth_tx_queue_setup,
 527         .rx_queue_release = eth_queue_release,
 528         .tx_queue_release = eth_queue_release,
 529         .link_update = eth_link_update,
 530         .stats_get = eth_stats_get,
 531         .stats_reset = eth_stats_reset,
 532 };
 533
 534 /*
 535  * Opens an AF_PACKET socket
 536  */
 537 static int
 538 open_packet_iface(const char *key __rte_unused,
 539                   const char *value __rte_unused,
 540                   void *extra_args)
 541 {
 542         int *sockfd = extra_args;
 543
 544         /* Open an AF_PACKET socket... */
 545         *sockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
 546         if (*sockfd == -1) {
 547                 RTE_LOG(ERR, PMD, "Could not open AF_PACKET socket\n");
 548                 return -1;
 549         }
 550
 551         return 0;
 552 }
 553
 554 static struct rte_vdev_driver pmd_af_packet_drv;
 555
 556 static int
 557 rte_pmd_init_internals(struct rte_vdev_device *dev,
 558                        const int sockfd,
 559                        const unsigned nb_queues,
 560                        unsigned int blocksize,
 561                        unsigned int blockcnt,
 562                        unsigned int framesize,
 563                        unsigned int framecnt,
 564                        unsigned int qdisc_bypass,
 565                        struct pmd_internals **internals,
 566                        struct rte_eth_dev **eth_dev,
 567                        struct rte_kvargs *kvlist)
 568 {
 569         const char *name = rte_vdev_device_name(dev);
 570         const unsigned int numa_node = dev->device.numa_node;
 571         struct rte_eth_dev_data *data = NULL;
 572         struct rte_kvargs_pair *pair = NULL;
 573         struct ifreq ifr;
 574         size_t ifnamelen;
 575         unsigned k_idx;
 576         struct sockaddr_ll sockaddr;
 577         struct tpacket_req *req;
 578         struct pkt_rx_queue *rx_queue;
 579         struct pkt_tx_queue *tx_queue;
 580         int rc, tpver, discard;
 581         int qsockfd = -1;
 582         unsigned int i, q, rdsize;
 583 #if defined(PACKET_FANOUT)
 584         int fanout_arg;
 585 #endif
 586
 587         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 588                 pair = &kvlist->pairs[k_idx];
 589                 if (strstr(pair->key, ETH_AF_PACKET_IFACE_ARG) != NULL)
 590                         break;
 591         }
 592         if (pair == NULL) {
 593                 RTE_LOG(ERR, PMD,
 594                         "%s: no interface specified for AF_PACKET ethdev\n",
 595                         name);
 596                 goto error_early;
 597         }
 598
 599         RTE_LOG(INFO, PMD,
 600                 "%s: creating AF_PACKET-backed ethdev on numa socket %u\n",
 601                 name, numa_node);
 602
 603         /*
 604          * now do all data allocation - for eth_dev structure, dummy pci driver
 605          * and internal (private) data
 606          */
 607         data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
 608         if (data == NULL)
 609                 goto error_early;
 610
 611         *internals = rte_zmalloc_socket(name, sizeof(**internals),
 612                                         0, numa_node);
 613         if (*internals == NULL)
 614                 goto error_early;
 615
 616         for (q = 0; q < nb_queues; q++) {
 617                 (*internals)->rx_queue[q].map = MAP_FAILED;
 618                 (*internals)->tx_queue[q].map = MAP_FAILED;
 619         }
 620
 621         req = &((*internals)->req);
 622
 623         req->tp_block_size = blocksize;
 624         req->tp_block_nr = blockcnt;
 625         req->tp_frame_size = framesize;
 626         req->tp_frame_nr = framecnt;
 627
 628         ifnamelen = strlen(pair->value);
 629         if (ifnamelen < sizeof(ifr.ifr_name)) {
 630                 memcpy(ifr.ifr_name, pair->value, ifnamelen);
 631                 ifr.ifr_name[ifnamelen] = '\0';
 632         } else {
 633                 RTE_LOG(ERR, PMD,
 634                         "%s: I/F name too long (%s)\n",
 635                         name, pair->value);
 636                 goto error_early;
 637         }
 638         if (ioctl(sockfd, SIOCGIFINDEX, &ifr) == -1) {
 639                 RTE_LOG(ERR, PMD,
 640                         "%s: ioctl failed (SIOCGIFINDEX)\n",
 641                         name);
 642                 goto error_early;
 643         }
 644         (*internals)->if_name = strdup(pair->value);
 645         if ((*internals)->if_name == NULL)
 646                 goto error_early;
 647         (*internals)->if_index = ifr.ifr_ifindex;
 648
 649         if (ioctl(sockfd, SIOCGIFHWADDR, &ifr) == -1) {
 650                 RTE_LOG(ERR, PMD,
 651                         "%s: ioctl failed (SIOCGIFHWADDR)\n",
 652                         name);
 653                 goto error_early;
 654         }
 655         memcpy(&(*internals)->eth_addr, ifr.ifr_hwaddr.sa_data, ETH_ALEN);
 656
 657         memset(&sockaddr, 0, sizeof(sockaddr));
 658         sockaddr.sll_family = AF_PACKET;
 659         sockaddr.sll_protocol = htons(ETH_P_ALL);
 660         sockaddr.sll_ifindex = (*internals)->if_index;
 661
 662 #if defined(PACKET_FANOUT)
 663         fanout_arg = (getpid() ^ (*internals)->if_index) & 0xffff;
 664         fanout_arg |= (PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_DEFRAG) << 16;
 665 #if defined(PACKET_FANOUT_FLAG_ROLLOVER)
 666         fanout_arg |= PACKET_FANOUT_FLAG_ROLLOVER << 16;
 667 #endif
 668 #endif
 669
 670         for (q = 0; q < nb_queues; q++) {
 671                 /* Open an AF_PACKET socket for this queue... */
 672                 qsockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
 673                 if (qsockfd == -1) {
 674                         RTE_LOG(ERR, PMD,
 675                                 "%s: could not open AF_PACKET socket\n",
 676                                 name);
 677                         return -1;
 678                 }
 679
 680                 tpver = TPACKET_V2;
 681                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_VERSION,
 682                                 &tpver, sizeof(tpver));
 683                 if (rc == -1) {
 684                         RTE_LOG(ERR, PMD,
 685                                 "%s: could not set PACKET_VERSION on AF_PACKET "
 686                                 "socket for %s\n", name, pair->value);
 687                         goto error;
 688                 }
 689
 690                 discard = 1;
 691                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_LOSS,
 692                                 &discard, sizeof(discard));
 693                 if (rc == -1) {
 694                         RTE_LOG(ERR, PMD,
 695                                 "%s: could not set PACKET_LOSS on "
 696                                 "AF_PACKET socket for %s\n", name, pair->value);
 697                         goto error;
 698                 }
 699
 700 #if defined(PACKET_QDISC_BYPASS)
 701                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_QDISC_BYPASS,
 702                                 &qdisc_bypass, sizeof(qdisc_bypass));
 703                 if (rc == -1) {
 704                         RTE_LOG(ERR, PMD,
 705                                 "%s: could not set PACKET_QDISC_BYPASS "
 706                                 "on AF_PACKET socket for %s\n", name,
 707                                 pair->value);
 708                         goto error;
 709                 }
 710 #else
 711                 RTE_SET_USED(qdisc_bypass);
 712 #endif
 713
 714                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_RX_RING, req, sizeof(*req));
 715                 if (rc == -1) {
 716                         RTE_LOG(ERR, PMD,
 717                                 "%s: could not set PACKET_RX_RING on AF_PACKET "
 718                                 "socket for %s\n", name, pair->value);
 719                         goto error;
 720                 }
 721
 722                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_TX_RING, req, sizeof(*req));
 723                 if (rc == -1) {
 724                         RTE_LOG(ERR, PMD,
 725                                 "%s: could not set PACKET_TX_RING on AF_PACKET "
 726                                 "socket for %s\n", name, pair->value);
 727                         goto error;
 728                 }
 729
 730                 rx_queue = &((*internals)->rx_queue[q]);
 731                 rx_queue->framecount = req->tp_frame_nr;
 732
 733                 rx_queue->map = mmap(NULL, 2 * req->tp_block_size * req->tp_block_nr,
 734                                     PROT_READ | PROT_WRITE, MAP_SHARED | MAP_LOCKED,
 735                                     qsockfd, 0);
 736                 if (rx_queue->map == MAP_FAILED) {
 737                         RTE_LOG(ERR, PMD,
 738                                 "%s: call to mmap failed on AF_PACKET socket for %s\n",
 739                                 name, pair->value);
 740                         goto error;
 741                 }
 742
 743                 /* rdsize is same for both Tx and Rx */
 744                 rdsize = req->tp_frame_nr * sizeof(*(rx_queue->rd));
 745
 746                 rx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
 747                 if (rx_queue->rd == NULL)
 748                         goto error;
 749                 for (i = 0; i < req->tp_frame_nr; ++i) {
 750                         rx_queue->rd[i].iov_base = rx_queue->map + (i * framesize);
 751                         rx_queue->rd[i].iov_len = req->tp_frame_size;
 752                 }
 753                 rx_queue->sockfd = qsockfd;
 754
 755                 tx_queue = &((*internals)->tx_queue[q]);
 756                 tx_queue->framecount = req->tp_frame_nr;
 757                 tx_queue->frame_data_size = req->tp_frame_size;
 758                 tx_queue->frame_data_size -= TPACKET2_HDRLEN -
 759                         sizeof(struct sockaddr_ll);
 760
 761                 tx_queue->map = rx_queue->map + req->tp_block_size * req->tp_block_nr;
 762
 763                 tx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
 764                 if (tx_queue->rd == NULL)
 765                         goto error;
 766                 for (i = 0; i < req->tp_frame_nr; ++i) {
 767                         tx_queue->rd[i].iov_base = tx_queue->map + (i * framesize);
 768                         tx_queue->rd[i].iov_len = req->tp_frame_size;
 769                 }
 770                 tx_queue->sockfd = qsockfd;
 771
 772                 rc = bind(qsockfd, (const struct sockaddr*)&sockaddr, sizeof(sockaddr));
 773                 if (rc == -1) {
 774                         RTE_LOG(ERR, PMD,
 775                                 "%s: could not bind AF_PACKET socket to %s\n",
 776                                 name, pair->value);
 777                         goto error;
 778                 }
 779
 780 #if defined(PACKET_FANOUT)
 781                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_FANOUT,
 782                                 &fanout_arg, sizeof(fanout_arg));
 783                 if (rc == -1) {
 784                         RTE_LOG(ERR, PMD,
 785                                 "%s: could not set PACKET_FANOUT on AF_PACKET socket "
 786                                 "for %s\n", name, pair->value);
 787                         goto error;
 788                 }
 789 #endif
 790         }
 791
 792         /* reserve an ethdev entry */
 793         *eth_dev = rte_eth_vdev_allocate(dev, 0);
 794         if (*eth_dev == NULL)
 795                 goto error;
 796
 797         /*
 798          * now put it all together
 799          * - store queue data in internals,
 800          * - store numa_node in eth_dev
 801          * - point eth_dev_data to internals
 802          * - and point eth_dev structure to new eth_dev_data structure
 803          */
 804
 805         (*internals)->nb_queues = nb_queues;
 806
 807         rte_memcpy(data, (*eth_dev)->data, sizeof(*data));
 808         data->dev_private = *internals;
 809         data->nb_rx_queues = (uint16_t)nb_queues;
 810         data->nb_tx_queues = (uint16_t)nb_queues;
 811         data->dev_link = pmd_link;
 812         data->mac_addrs = &(*internals)->eth_addr;
 813
 814         (*eth_dev)->data = data;
 815         (*eth_dev)->dev_ops = &ops;
 816         (*eth_dev)->data->dev_flags = RTE_ETH_DEV_DETACHABLE;
 817
 818         return 0;
 819
 820 error:
 821         if (qsockfd != -1)
 822                 close(qsockfd);
 823         for (q = 0; q < nb_queues; q++) {
 824                 munmap((*internals)->rx_queue[q].map,
 825                        2 * req->tp_block_size * req->tp_block_nr);
 826
 827                 rte_free((*internals)->rx_queue[q].rd);
 828                 rte_free((*internals)->tx_queue[q].rd);
 829                 if (((*internals)->rx_queue[q].sockfd != 0) &&
 830                         ((*internals)->rx_queue[q].sockfd != qsockfd))
 831                         close((*internals)->rx_queue[q].sockfd);
 832         }
 833         free((*internals)->if_name);
 834         rte_free(*internals);
 835 error_early:
 836         rte_free(data);
 837         return -1;
 838 }
 839
 840 static int
 841 rte_eth_from_packet(struct rte_vdev_device *dev,
 842                     int const *sockfd,
 843                     struct rte_kvargs *kvlist)
 844 {
 845         const char *name = rte_vdev_device_name(dev);
 846         struct pmd_internals *internals = NULL;
 847         struct rte_eth_dev *eth_dev = NULL;
 848         struct rte_kvargs_pair *pair = NULL;
 849         unsigned k_idx;
 850         unsigned int blockcount;
 851         unsigned int blocksize = DFLT_BLOCK_SIZE;
 852         unsigned int framesize = DFLT_FRAME_SIZE;
 853         unsigned int framecount = DFLT_FRAME_COUNT;
 854         unsigned int qpairs = 1;
 855         unsigned int qdisc_bypass = 1;
 856
 857         /* do some parameter checking */
 858         if (*sockfd < 0)
 859                 return -1;
 860
 861         /*
 862          * Walk arguments for configurable settings
 863          */
 864         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 865                 pair = &kvlist->pairs[k_idx];
 866                 if (strstr(pair->key, ETH_AF_PACKET_NUM_Q_ARG) != NULL) {
 867                         qpairs = atoi(pair->value);
 868                         if (qpairs < 1 ||
 869                             qpairs > RTE_PMD_AF_PACKET_MAX_RINGS) {
 870                                 RTE_LOG(ERR, PMD,
 871                                         "%s: invalid qpairs value\n",
 872                                         name);
 873                                 return -1;
 874                         }
 875                         continue;
 876                 }
 877                 if (strstr(pair->key, ETH_AF_PACKET_BLOCKSIZE_ARG) != NULL) {
 878                         blocksize = atoi(pair->value);
 879                         if (!blocksize) {
 880                                 RTE_LOG(ERR, PMD,
 881                                         "%s: invalid blocksize value\n",
 882                                         name);
 883                                 return -1;
 884                         }
 885                         continue;
 886                 }
 887                 if (strstr(pair->key, ETH_AF_PACKET_FRAMESIZE_ARG) != NULL) {
 888                         framesize = atoi(pair->value);
 889                         if (!framesize) {
 890                                 RTE_LOG(ERR, PMD,
 891                                         "%s: invalid framesize value\n",
 892                                         name);
 893                                 return -1;
 894                         }
 895                         continue;
 896                 }
 897                 if (strstr(pair->key, ETH_AF_PACKET_FRAMECOUNT_ARG) != NULL) {
 898                         framecount = atoi(pair->value);
 899                         if (!framecount) {
 900                                 RTE_LOG(ERR, PMD,
 901                                         "%s: invalid framecount value\n",
 902                                         name);
 903                                 return -1;
 904                         }
 905                         continue;
 906                 }
 907                 if (strstr(pair->key, ETH_AF_PACKET_QDISC_BYPASS_ARG) != NULL) {
 908                         qdisc_bypass = atoi(pair->value);
 909                         if (qdisc_bypass > 1) {
 910                                 RTE_LOG(ERR, PMD,
 911                                         "%s: invalid bypass value\n",
 912                                         name);
 913                                 return -1;
 914                         }
 915                         continue;
 916                 }
 917         }
 918
 919         if (framesize > blocksize) {
 920                 RTE_LOG(ERR, PMD,
 921                         "%s: AF_PACKET MMAP frame size exceeds block size!\n",
 922                         name);
 923                 return -1;
 924         }
 925
 926         blockcount = framecount / (blocksize / framesize);
 927         if (!blockcount) {
 928                 RTE_LOG(ERR, PMD,
 929                         "%s: invalid AF_PACKET MMAP parameters\n", name);
 930                 return -1;
 931         }
 932
 933         RTE_LOG(INFO, PMD, "%s: AF_PACKET MMAP parameters:\n", name);
 934         RTE_LOG(INFO, PMD, "%s:\tblock size %d\n", name, blocksize);
 935         RTE_LOG(INFO, PMD, "%s:\tblock count %d\n", name, blockcount);
 936         RTE_LOG(INFO, PMD, "%s:\tframe size %d\n", name, framesize);
 937         RTE_LOG(INFO, PMD, "%s:\tframe count %d\n", name, framecount);
 938
 939         if (rte_pmd_init_internals(dev, *sockfd, qpairs,
 940                                    blocksize, blockcount,
 941                                    framesize, framecount,
 942                                    qdisc_bypass,
 943                                    &internals, &eth_dev,
 944                                    kvlist) < 0)
 945                 return -1;
 946
 947         eth_dev->rx_pkt_burst = eth_af_packet_rx;
 948         eth_dev->tx_pkt_burst = eth_af_packet_tx;
 949
 950         return 0;
 951 }
 952
 953 static int
 954 rte_pmd_af_packet_probe(struct rte_vdev_device *dev)
 955 {
 956         int ret = 0;
 957         struct rte_kvargs *kvlist;
 958         int sockfd = -1;
 959
 960         RTE_LOG(INFO, PMD, "Initializing pmd_af_packet for %s\n",
 961                 rte_vdev_device_name(dev));
 962
 963         kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_arguments);
 964         if (kvlist == NULL) {
 965                 ret = -1;
 966                 goto exit;
 967         }
 968
 969         /*
 970          * If iface argument is passed we open the NICs and use them for
 971          * reading / writing
 972          */
 973         if (rte_kvargs_count(kvlist, ETH_AF_PACKET_IFACE_ARG) == 1) {
 974
 975                 ret = rte_kvargs_process(kvlist, ETH_AF_PACKET_IFACE_ARG,
 976                                          &open_packet_iface, &sockfd);
 977                 if (ret < 0)
 978                         goto exit;
 979         }
 980
 981         if (dev->device.numa_node == SOCKET_ID_ANY)
 982                 dev->device.numa_node = rte_socket_id();
 983
 984         ret = rte_eth_from_packet(dev, &sockfd, kvlist);
 985         close(sockfd); /* no longer needed */
 986
 987 exit:
 988         rte_kvargs_free(kvlist);
 989         return ret;
 990 }
 991
 992 static int
 993 rte_pmd_af_packet_remove(struct rte_vdev_device *dev)
 994 {
 995         struct rte_eth_dev *eth_dev = NULL;
 996         struct pmd_internals *internals;
 997         unsigned q;
 998
 999         RTE_LOG(INFO, PMD, "Closing AF_PACKET ethdev on numa socket %u\n",
1000                         rte_socket_id());
1001
1002         if (dev == NULL)
1003                 return -1;
1004
1005         /* find the ethdev entry */
1006         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1007         if (eth_dev == NULL)
1008                 return -1;
1009
1010         internals = eth_dev->data->dev_private;
1011         for (q = 0; q < internals->nb_queues; q++) {
1012                 rte_free(internals->rx_queue[q].rd);
1013                 rte_free(internals->tx_queue[q].rd);
1014         }
1015         free(internals->if_name);
1016
1017         rte_free(eth_dev->data->dev_private);
1018         rte_free(eth_dev->data);
1019
1020         rte_eth_dev_release_port(eth_dev);
1021
1022         return 0;
1023 }
1024
1025 static struct rte_vdev_driver pmd_af_packet_drv = {
1026         .probe = rte_pmd_af_packet_probe,
1027         .remove = rte_pmd_af_packet_remove,
1028 };
1029
1030 RTE_PMD_REGISTER_VDEV(net_af_packet, pmd_af_packet_drv);
1031 RTE_PMD_REGISTER_ALIAS(net_af_packet, eth_af_packet);
1032 RTE_PMD_REGISTER_PARAM_STRING(net_af_packet,
1033         "iface=<string> "
1034         "qpairs=<int> "
1035         "blocksz=<int> "
1036         "framesz=<int> "
1037         "framecnt=<int> "
1038         "qdisc_bypass=<0|1>");