drivers/net/af_packet/rte_eth_af_packet.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2014 John W. Linville <linville@tuxdriver.com>
   3  * Originally based upon librte_pmd_pcap code:
   4  * Copyright(c) 2010-2015 Intel Corporation.
   5  * Copyright(c) 2014 6WIND S.A.
   6  * All rights reserved.
   7  */
   8
   9 #include <rte_string_fns.h>
  10 #include <rte_mbuf.h>
  11 #include <ethdev_driver.h>
  12 #include <ethdev_vdev.h>
  13 #include <rte_malloc.h>
  14 #include <rte_kvargs.h>
  15 #include <rte_bus_vdev.h>
  16
  17 #include <errno.h>
  18 #include <linux/if_ether.h>
  19 #include <linux/if_packet.h>
  20 #include <arpa/inet.h>
  21 #include <net/if.h>
  22 #include <net/if_arp.h>
  23 #include <sys/types.h>
  24 #include <sys/socket.h>
  25 #include <sys/ioctl.h>
  26 #include <string.h>
  27 #include <sys/mman.h>
  28 #include <unistd.h>
  29 #include <poll.h>
  30
  31 #define ETH_AF_PACKET_IFACE_ARG         "iface"
  32 #define ETH_AF_PACKET_NUM_Q_ARG         "qpairs"
  33 #define ETH_AF_PACKET_BLOCKSIZE_ARG     "blocksz"
  34 #define ETH_AF_PACKET_FRAMESIZE_ARG     "framesz"
  35 #define ETH_AF_PACKET_FRAMECOUNT_ARG    "framecnt"
  36 #define ETH_AF_PACKET_QDISC_BYPASS_ARG  "qdisc_bypass"
  37
  38 #define DFLT_FRAME_SIZE         (1 << 11)
  39 #define DFLT_FRAME_COUNT        (1 << 9)
  40
  41 struct pkt_rx_queue {
  42         int sockfd;
  43
  44         struct iovec *rd;
  45         uint8_t *map;
  46         unsigned int framecount;
  47         unsigned int framenum;
  48
  49         struct rte_mempool *mb_pool;
  50         uint16_t in_port;
  51
  52         volatile unsigned long rx_pkts;
  53         volatile unsigned long rx_bytes;
  54 };
  55
  56 struct pkt_tx_queue {
  57         int sockfd;
  58         unsigned int frame_data_size;
  59
  60         struct iovec *rd;
  61         uint8_t *map;
  62         unsigned int framecount;
  63         unsigned int framenum;
  64
  65         volatile unsigned long tx_pkts;
  66         volatile unsigned long err_pkts;
  67         volatile unsigned long tx_bytes;
  68 };
  69
  70 struct pmd_internals {
  71         unsigned nb_queues;
  72
  73         int if_index;
  74         char *if_name;
  75         struct rte_ether_addr eth_addr;
  76
  77         struct tpacket_req req;
  78
  79         struct pkt_rx_queue *rx_queue;
  80         struct pkt_tx_queue *tx_queue;
  81 };
  82
  83 static const char *valid_arguments[] = {
  84         ETH_AF_PACKET_IFACE_ARG,
  85         ETH_AF_PACKET_NUM_Q_ARG,
  86         ETH_AF_PACKET_BLOCKSIZE_ARG,
  87         ETH_AF_PACKET_FRAMESIZE_ARG,
  88         ETH_AF_PACKET_FRAMECOUNT_ARG,
  89         ETH_AF_PACKET_QDISC_BYPASS_ARG,
  90         NULL
  91 };
  92
  93 static struct rte_eth_link pmd_link = {
  94         .link_speed = ETH_SPEED_NUM_10G,
  95         .link_duplex = ETH_LINK_FULL_DUPLEX,
  96         .link_status = ETH_LINK_DOWN,
  97         .link_autoneg = ETH_LINK_FIXED,
  98 };
  99
 100 RTE_LOG_REGISTER_DEFAULT(af_packet_logtype, NOTICE);
 101
 102 #define PMD_LOG(level, fmt, args...) \
 103         rte_log(RTE_LOG_ ## level, af_packet_logtype, \
 104                 "%s(): " fmt "\n", __func__, ##args)
 105
 106 #define PMD_LOG_ERRNO(level, fmt, args...) \
 107         rte_log(RTE_LOG_ ## level, af_packet_logtype, \
 108                 "%s(): " fmt ":%s\n", __func__, ##args, strerror(errno))
 109
 110 static uint16_t
 111 eth_af_packet_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 112 {
 113         unsigned i;
 114         struct tpacket2_hdr *ppd;
 115         struct rte_mbuf *mbuf;
 116         uint8_t *pbuf;
 117         struct pkt_rx_queue *pkt_q = queue;
 118         uint16_t num_rx = 0;
 119         unsigned long num_rx_bytes = 0;
 120         unsigned int framecount, framenum;
 121
 122         if (unlikely(nb_pkts == 0))
 123                 return 0;
 124
 125         /*
 126          * Reads the given number of packets from the AF_PACKET socket one by
 127          * one and copies the packet data into a newly allocated mbuf.
 128          */
 129         framecount = pkt_q->framecount;
 130         framenum = pkt_q->framenum;
 131         for (i = 0; i < nb_pkts; i++) {
 132                 /* point at the next incoming frame */
 133                 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
 134                 if ((ppd->tp_status & TP_STATUS_USER) == 0)
 135                         break;
 136
 137                 /* allocate the next mbuf */
 138                 mbuf = rte_pktmbuf_alloc(pkt_q->mb_pool);
 139                 if (unlikely(mbuf == NULL))
 140                         break;
 141
 142                 /* packet will fit in the mbuf, go ahead and receive it */
 143                 rte_pktmbuf_pkt_len(mbuf) = rte_pktmbuf_data_len(mbuf) = ppd->tp_snaplen;
 144                 pbuf = (uint8_t *) ppd + ppd->tp_mac;
 145                 memcpy(rte_pktmbuf_mtod(mbuf, void *), pbuf, rte_pktmbuf_data_len(mbuf));
 146
 147                 /* check for vlan info */
 148                 if (ppd->tp_status & TP_STATUS_VLAN_VALID) {
 149                         mbuf->vlan_tci = ppd->tp_vlan_tci;
 150                         mbuf->ol_flags |= (PKT_RX_VLAN | PKT_RX_VLAN_STRIPPED);
 151                 }
 152
 153                 /* release incoming frame and advance ring buffer */
 154                 ppd->tp_status = TP_STATUS_KERNEL;
 155                 if (++framenum >= framecount)
 156                         framenum = 0;
 157                 mbuf->port = pkt_q->in_port;
 158
 159                 /* account for the receive frame */
 160                 bufs[i] = mbuf;
 161                 num_rx++;
 162                 num_rx_bytes += mbuf->pkt_len;
 163         }
 164         pkt_q->framenum = framenum;
 165         pkt_q->rx_pkts += num_rx;
 166         pkt_q->rx_bytes += num_rx_bytes;
 167         return num_rx;
 168 }
 169
 170 /*
 171  * Callback to handle sending packets through a real NIC.
 172  */
 173 static uint16_t
 174 eth_af_packet_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 175 {
 176         struct tpacket2_hdr *ppd;
 177         struct rte_mbuf *mbuf;
 178         uint8_t *pbuf;
 179         unsigned int framecount, framenum;
 180         struct pollfd pfd;
 181         struct pkt_tx_queue *pkt_q = queue;
 182         uint16_t num_tx = 0;
 183         unsigned long num_tx_bytes = 0;
 184         int i;
 185
 186         if (unlikely(nb_pkts == 0))
 187                 return 0;
 188
 189         memset(&pfd, 0, sizeof(pfd));
 190         pfd.fd = pkt_q->sockfd;
 191         pfd.events = POLLOUT;
 192         pfd.revents = 0;
 193
 194         framecount = pkt_q->framecount;
 195         framenum = pkt_q->framenum;
 196         ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
 197         for (i = 0; i < nb_pkts; i++) {
 198                 mbuf = *bufs++;
 199
 200                 /* drop oversized packets */
 201                 if (mbuf->pkt_len > pkt_q->frame_data_size) {
 202                         rte_pktmbuf_free(mbuf);
 203                         continue;
 204                 }
 205
 206                 /* insert vlan info if necessary */
 207                 if (mbuf->ol_flags & PKT_TX_VLAN_PKT) {
 208                         if (rte_vlan_insert(&mbuf)) {
 209                                 rte_pktmbuf_free(mbuf);
 210                                 continue;
 211                         }
 212                 }
 213
 214                 /* point at the next incoming frame */
 215                 if ((ppd->tp_status != TP_STATUS_AVAILABLE) &&
 216                     (poll(&pfd, 1, -1) < 0))
 217                         break;
 218
 219                 /* copy the tx frame data */
 220                 pbuf = (uint8_t *) ppd + TPACKET2_HDRLEN -
 221                         sizeof(struct sockaddr_ll);
 222
 223                 struct rte_mbuf *tmp_mbuf = mbuf;
 224                 while (tmp_mbuf) {
 225                         uint16_t data_len = rte_pktmbuf_data_len(tmp_mbuf);
 226                         memcpy(pbuf, rte_pktmbuf_mtod(tmp_mbuf, void*), data_len);
 227                         pbuf += data_len;
 228                         tmp_mbuf = tmp_mbuf->next;
 229                 }
 230
 231                 ppd->tp_len = mbuf->pkt_len;
 232                 ppd->tp_snaplen = mbuf->pkt_len;
 233
 234                 /* release incoming frame and advance ring buffer */
 235                 ppd->tp_status = TP_STATUS_SEND_REQUEST;
 236                 if (++framenum >= framecount)
 237                         framenum = 0;
 238                 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
 239
 240                 num_tx++;
 241                 num_tx_bytes += mbuf->pkt_len;
 242                 rte_pktmbuf_free(mbuf);
 243         }
 244
 245         /* kick-off transmits */
 246         if (sendto(pkt_q->sockfd, NULL, 0, MSG_DONTWAIT, NULL, 0) == -1 &&
 247                         errno != ENOBUFS && errno != EAGAIN) {
 248                 /*
 249                  * In case of a ENOBUFS/EAGAIN error all of the enqueued
 250                  * packets will be considered successful even though only some
 251                  * are sent.
 252                  */
 253
 254                 num_tx = 0;
 255                 num_tx_bytes = 0;
 256         }
 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 int
 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         return 0;
 300 }
 301
 302 static int
 303 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 304 {
 305         return 0;
 306 }
 307
 308 static int
 309 eth_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 310 {
 311         struct pmd_internals *internals = dev->data->dev_private;
 312
 313         dev_info->if_index = internals->if_index;
 314         dev_info->max_mac_addrs = 1;
 315         dev_info->max_rx_pktlen = (uint32_t)ETH_FRAME_LEN;
 316         dev_info->max_rx_queues = (uint16_t)internals->nb_queues;
 317         dev_info->max_tx_queues = (uint16_t)internals->nb_queues;
 318         dev_info->min_rx_bufsize = 0;
 319         dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
 320                 DEV_TX_OFFLOAD_VLAN_INSERT;
 321
 322         return 0;
 323 }
 324
 325 static int
 326 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *igb_stats)
 327 {
 328         unsigned i, imax;
 329         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 330         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 331         const struct pmd_internals *internal = dev->data->dev_private;
 332
 333         imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
 334                 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
 335         for (i = 0; i < imax; i++) {
 336                 igb_stats->q_ipackets[i] = internal->rx_queue[i].rx_pkts;
 337                 igb_stats->q_ibytes[i] = internal->rx_queue[i].rx_bytes;
 338                 rx_total += igb_stats->q_ipackets[i];
 339                 rx_bytes_total += igb_stats->q_ibytes[i];
 340         }
 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_opackets[i] = internal->tx_queue[i].tx_pkts;
 346                 igb_stats->q_obytes[i] = internal->tx_queue[i].tx_bytes;
 347                 tx_total += igb_stats->q_opackets[i];
 348                 tx_err_total += internal->tx_queue[i].err_pkts;
 349                 tx_bytes_total += igb_stats->q_obytes[i];
 350         }
 351
 352         igb_stats->ipackets = rx_total;
 353         igb_stats->ibytes = rx_bytes_total;
 354         igb_stats->opackets = tx_total;
 355         igb_stats->oerrors = tx_err_total;
 356         igb_stats->obytes = tx_bytes_total;
 357         return 0;
 358 }
 359
 360 static int
 361 eth_stats_reset(struct rte_eth_dev *dev)
 362 {
 363         unsigned i;
 364         struct pmd_internals *internal = dev->data->dev_private;
 365
 366         for (i = 0; i < internal->nb_queues; i++) {
 367                 internal->rx_queue[i].rx_pkts = 0;
 368                 internal->rx_queue[i].rx_bytes = 0;
 369         }
 370
 371         for (i = 0; i < internal->nb_queues; i++) {
 372                 internal->tx_queue[i].tx_pkts = 0;
 373                 internal->tx_queue[i].err_pkts = 0;
 374                 internal->tx_queue[i].tx_bytes = 0;
 375         }
 376
 377         return 0;
 378 }
 379
 380 static int
 381 eth_dev_close(struct rte_eth_dev *dev)
 382 {
 383         struct pmd_internals *internals;
 384         struct tpacket_req *req;
 385         unsigned int q;
 386
 387         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 388                 return 0;
 389
 390         PMD_LOG(INFO, "Closing AF_PACKET ethdev on NUMA socket %u",
 391                 rte_socket_id());
 392
 393         internals = dev->data->dev_private;
 394         req = &internals->req;
 395         for (q = 0; q < internals->nb_queues; q++) {
 396                 munmap(internals->rx_queue[q].map,
 397                         2 * req->tp_block_size * req->tp_block_nr);
 398                 rte_free(internals->rx_queue[q].rd);
 399                 rte_free(internals->tx_queue[q].rd);
 400         }
 401         free(internals->if_name);
 402         rte_free(internals->rx_queue);
 403         rte_free(internals->tx_queue);
 404
 405         /* mac_addrs must not be freed alone because part of dev_private */
 406         dev->data->mac_addrs = NULL;
 407         return 0;
 408 }
 409
 410 static void
 411 eth_queue_release(void *q __rte_unused)
 412 {
 413 }
 414
 415 static int
 416 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 417                 int wait_to_complete __rte_unused)
 418 {
 419         return 0;
 420 }
 421
 422 static int
 423 eth_rx_queue_setup(struct rte_eth_dev *dev,
 424                    uint16_t rx_queue_id,
 425                    uint16_t nb_rx_desc __rte_unused,
 426                    unsigned int socket_id __rte_unused,
 427                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 428                    struct rte_mempool *mb_pool)
 429 {
 430         struct pmd_internals *internals = dev->data->dev_private;
 431         struct pkt_rx_queue *pkt_q = &internals->rx_queue[rx_queue_id];
 432         unsigned int buf_size, data_size;
 433
 434         pkt_q->mb_pool = mb_pool;
 435
 436         /* Now get the space available for data in the mbuf */
 437         buf_size = rte_pktmbuf_data_room_size(pkt_q->mb_pool) -
 438                 RTE_PKTMBUF_HEADROOM;
 439         data_size = internals->req.tp_frame_size;
 440         data_size -= TPACKET2_HDRLEN - sizeof(struct sockaddr_ll);
 441
 442         if (data_size > buf_size) {
 443                 PMD_LOG(ERR,
 444                         "%s: %d bytes will not fit in mbuf (%d bytes)",
 445                         dev->device->name, data_size, buf_size);
 446                 return -ENOMEM;
 447         }
 448
 449         dev->data->rx_queues[rx_queue_id] = pkt_q;
 450         pkt_q->in_port = dev->data->port_id;
 451
 452         return 0;
 453 }
 454
 455 static int
 456 eth_tx_queue_setup(struct rte_eth_dev *dev,
 457                    uint16_t tx_queue_id,
 458                    uint16_t nb_tx_desc __rte_unused,
 459                    unsigned int socket_id __rte_unused,
 460                    const struct rte_eth_txconf *tx_conf __rte_unused)
 461 {
 462
 463         struct pmd_internals *internals = dev->data->dev_private;
 464
 465         dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
 466         return 0;
 467 }
 468
 469 static int
 470 eth_dev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 471 {
 472         struct pmd_internals *internals = dev->data->dev_private;
 473         struct ifreq ifr = { .ifr_mtu = mtu };
 474         int ret;
 475         int s;
 476         unsigned int data_size = internals->req.tp_frame_size -
 477                                  TPACKET2_HDRLEN;
 478
 479         if (mtu > data_size)
 480                 return -EINVAL;
 481
 482         s = socket(PF_INET, SOCK_DGRAM, 0);
 483         if (s < 0)
 484                 return -EINVAL;
 485
 486         strlcpy(ifr.ifr_name, internals->if_name, IFNAMSIZ);
 487         ret = ioctl(s, SIOCSIFMTU, &ifr);
 488         close(s);
 489
 490         if (ret < 0)
 491                 return -EINVAL;
 492
 493         return 0;
 494 }
 495
 496 static int
 497 eth_dev_macaddr_set(struct rte_eth_dev *dev, struct rte_ether_addr *addr)
 498 {
 499         struct pmd_internals *internals = dev->data->dev_private;
 500         struct ifreq ifr = { };
 501         int sockfd = internals->rx_queue[0].sockfd;
 502         int ret;
 503
 504         if (sockfd == -1) {
 505                 PMD_LOG(ERR, "receive socket not found");
 506                 return -EINVAL;
 507         }
 508
 509         strlcpy(ifr.ifr_name, internals->if_name, IFNAMSIZ);
 510         ifr.ifr_hwaddr.sa_family = ARPHRD_ETHER;
 511         memcpy(ifr.ifr_hwaddr.sa_data, addr, sizeof(*addr));
 512         ret = ioctl(sockfd, SIOCSIFHWADDR, &ifr);
 513
 514         if (ret < 0) {
 515                 PMD_LOG_ERRNO(ERR, "ioctl(SIOCSIFHWADDR) failed");
 516                 return -EINVAL;
 517         }
 518
 519         return 0;
 520 }
 521
 522 static int
 523 eth_dev_change_flags(char *if_name, uint32_t flags, uint32_t mask)
 524 {
 525         struct ifreq ifr;
 526         int ret = 0;
 527         int s;
 528
 529         s = socket(PF_INET, SOCK_DGRAM, 0);
 530         if (s < 0)
 531                 return -errno;
 532
 533         strlcpy(ifr.ifr_name, if_name, IFNAMSIZ);
 534         if (ioctl(s, SIOCGIFFLAGS, &ifr) < 0) {
 535                 ret = -errno;
 536                 goto out;
 537         }
 538         ifr.ifr_flags &= mask;
 539         ifr.ifr_flags |= flags;
 540         if (ioctl(s, SIOCSIFFLAGS, &ifr) < 0) {
 541                 ret = -errno;
 542                 goto out;
 543         }
 544 out:
 545         close(s);
 546         return ret;
 547 }
 548
 549 static int
 550 eth_dev_promiscuous_enable(struct rte_eth_dev *dev)
 551 {
 552         struct pmd_internals *internals = dev->data->dev_private;
 553
 554         return eth_dev_change_flags(internals->if_name, IFF_PROMISC, ~0);
 555 }
 556
 557 static int
 558 eth_dev_promiscuous_disable(struct rte_eth_dev *dev)
 559 {
 560         struct pmd_internals *internals = dev->data->dev_private;
 561
 562         return eth_dev_change_flags(internals->if_name, 0, ~IFF_PROMISC);
 563 }
 564
 565 static const struct eth_dev_ops ops = {
 566         .dev_start = eth_dev_start,
 567         .dev_stop = eth_dev_stop,
 568         .dev_close = eth_dev_close,
 569         .dev_configure = eth_dev_configure,
 570         .dev_infos_get = eth_dev_info,
 571         .mac_addr_set = eth_dev_macaddr_set,
 572         .mtu_set = eth_dev_mtu_set,
 573         .promiscuous_enable = eth_dev_promiscuous_enable,
 574         .promiscuous_disable = eth_dev_promiscuous_disable,
 575         .rx_queue_setup = eth_rx_queue_setup,
 576         .tx_queue_setup = eth_tx_queue_setup,
 577         .rx_queue_release = eth_queue_release,
 578         .tx_queue_release = eth_queue_release,
 579         .link_update = eth_link_update,
 580         .stats_get = eth_stats_get,
 581         .stats_reset = eth_stats_reset,
 582 };
 583
 584 /*
 585  * Opens an AF_PACKET socket
 586  */
 587 static int
 588 open_packet_iface(const char *key __rte_unused,
 589                   const char *value __rte_unused,
 590                   void *extra_args)
 591 {
 592         int *sockfd = extra_args;
 593
 594         /* Open an AF_PACKET socket... */
 595         *sockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
 596         if (*sockfd == -1) {
 597                 PMD_LOG(ERR, "Could not open AF_PACKET socket");
 598                 return -1;
 599         }
 600
 601         return 0;
 602 }
 603
 604 static int
 605 rte_pmd_init_internals(struct rte_vdev_device *dev,
 606                        const int sockfd,
 607                        const unsigned nb_queues,
 608                        unsigned int blocksize,
 609                        unsigned int blockcnt,
 610                        unsigned int framesize,
 611                        unsigned int framecnt,
 612                        unsigned int qdisc_bypass,
 613                        struct pmd_internals **internals,
 614                        struct rte_eth_dev **eth_dev,
 615                        struct rte_kvargs *kvlist)
 616 {
 617         const char *name = rte_vdev_device_name(dev);
 618         const unsigned int numa_node = dev->device.numa_node;
 619         struct rte_eth_dev_data *data = NULL;
 620         struct rte_kvargs_pair *pair = NULL;
 621         struct ifreq ifr;
 622         size_t ifnamelen;
 623         unsigned k_idx;
 624         struct sockaddr_ll sockaddr;
 625         struct tpacket_req *req;
 626         struct pkt_rx_queue *rx_queue;
 627         struct pkt_tx_queue *tx_queue;
 628         int rc, tpver, discard;
 629         int qsockfd = -1;
 630         unsigned int i, q, rdsize;
 631 #if defined(PACKET_FANOUT)
 632         int fanout_arg;
 633 #endif
 634
 635         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 636                 pair = &kvlist->pairs[k_idx];
 637                 if (strstr(pair->key, ETH_AF_PACKET_IFACE_ARG) != NULL)
 638                         break;
 639         }
 640         if (pair == NULL) {
 641                 PMD_LOG(ERR,
 642                         "%s: no interface specified for AF_PACKET ethdev",
 643                         name);
 644                 return -1;
 645         }
 646
 647         PMD_LOG(INFO,
 648                 "%s: creating AF_PACKET-backed ethdev on numa socket %u",
 649                 name, numa_node);
 650
 651         *internals = rte_zmalloc_socket(name, sizeof(**internals),
 652                                         0, numa_node);
 653         if (*internals == NULL)
 654                 return -1;
 655
 656
 657         (*internals)->rx_queue = rte_calloc_socket("af_packet_rx",
 658                                                 nb_queues,
 659                                                 sizeof(struct pkt_rx_queue),
 660                                                 0, numa_node);
 661         (*internals)->tx_queue = rte_calloc_socket("af_packet_tx",
 662                                                 nb_queues,
 663                                                 sizeof(struct pkt_tx_queue),
 664                                                 0, numa_node);
 665         if (!(*internals)->rx_queue || !(*internals)->tx_queue) {
 666                 goto free_internals;
 667         }
 668
 669         for (q = 0; q < nb_queues; q++) {
 670                 (*internals)->rx_queue[q].map = MAP_FAILED;
 671                 (*internals)->tx_queue[q].map = MAP_FAILED;
 672                 (*internals)->rx_queue[q].sockfd = -1;
 673                 (*internals)->tx_queue[q].sockfd = -1;
 674         }
 675
 676         req = &((*internals)->req);
 677
 678         req->tp_block_size = blocksize;
 679         req->tp_block_nr = blockcnt;
 680         req->tp_frame_size = framesize;
 681         req->tp_frame_nr = framecnt;
 682
 683         ifnamelen = strlen(pair->value);
 684         if (ifnamelen < sizeof(ifr.ifr_name)) {
 685                 memcpy(ifr.ifr_name, pair->value, ifnamelen);
 686                 ifr.ifr_name[ifnamelen] = '\0';
 687         } else {
 688                 PMD_LOG(ERR,
 689                         "%s: I/F name too long (%s)",
 690                         name, pair->value);
 691                 goto free_internals;
 692         }
 693         if (ioctl(sockfd, SIOCGIFINDEX, &ifr) == -1) {
 694                 PMD_LOG_ERRNO(ERR, "%s: ioctl failed (SIOCGIFINDEX)", name);
 695                 goto free_internals;
 696         }
 697         (*internals)->if_name = strdup(pair->value);
 698         if ((*internals)->if_name == NULL)
 699                 goto free_internals;
 700         (*internals)->if_index = ifr.ifr_ifindex;
 701
 702         if (ioctl(sockfd, SIOCGIFHWADDR, &ifr) == -1) {
 703                 PMD_LOG_ERRNO(ERR, "%s: ioctl failed (SIOCGIFHWADDR)", name);
 704                 goto free_internals;
 705         }
 706         memcpy(&(*internals)->eth_addr, ifr.ifr_hwaddr.sa_data, ETH_ALEN);
 707
 708         memset(&sockaddr, 0, sizeof(sockaddr));
 709         sockaddr.sll_family = AF_PACKET;
 710         sockaddr.sll_protocol = htons(ETH_P_ALL);
 711         sockaddr.sll_ifindex = (*internals)->if_index;
 712
 713 #if defined(PACKET_FANOUT)
 714         fanout_arg = (getpid() ^ (*internals)->if_index) & 0xffff;
 715         fanout_arg |= (PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_DEFRAG) << 16;
 716 #if defined(PACKET_FANOUT_FLAG_ROLLOVER)
 717         fanout_arg |= PACKET_FANOUT_FLAG_ROLLOVER << 16;
 718 #endif
 719 #endif
 720
 721         for (q = 0; q < nb_queues; q++) {
 722                 /* Open an AF_PACKET socket for this queue... */
 723                 qsockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
 724                 if (qsockfd == -1) {
 725                         PMD_LOG_ERRNO(ERR,
 726                                 "%s: could not open AF_PACKET socket",
 727                                 name);
 728                         goto error;
 729                 }
 730
 731                 tpver = TPACKET_V2;
 732                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_VERSION,
 733                                 &tpver, sizeof(tpver));
 734                 if (rc == -1) {
 735                         PMD_LOG_ERRNO(ERR,
 736                                 "%s: could not set PACKET_VERSION on AF_PACKET socket for %s",
 737                                 name, pair->value);
 738                         goto error;
 739                 }
 740
 741                 discard = 1;
 742                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_LOSS,
 743                                 &discard, sizeof(discard));
 744                 if (rc == -1) {
 745                         PMD_LOG_ERRNO(ERR,
 746                                 "%s: could not set PACKET_LOSS on AF_PACKET socket for %s",
 747                                 name, pair->value);
 748                         goto error;
 749                 }
 750
 751                 if (qdisc_bypass) {
 752 #if defined(PACKET_QDISC_BYPASS)
 753                         rc = setsockopt(qsockfd, SOL_PACKET, PACKET_QDISC_BYPASS,
 754                                         &qdisc_bypass, sizeof(qdisc_bypass));
 755                         if (rc == -1) {
 756                                 PMD_LOG_ERRNO(ERR,
 757                                         "%s: could not set PACKET_QDISC_BYPASS on AF_PACKET socket for %s",
 758                                         name, pair->value);
 759                                 goto error;
 760                         }
 761 #endif
 762                 }
 763
 764                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_RX_RING, req, sizeof(*req));
 765                 if (rc == -1) {
 766                         PMD_LOG_ERRNO(ERR,
 767                                 "%s: could not set PACKET_RX_RING on AF_PACKET socket for %s",
 768                                 name, pair->value);
 769                         goto error;
 770                 }
 771
 772                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_TX_RING, req, sizeof(*req));
 773                 if (rc == -1) {
 774                         PMD_LOG_ERRNO(ERR,
 775                                 "%s: could not set PACKET_TX_RING on AF_PACKET "
 776                                 "socket for %s", name, pair->value);
 777                         goto error;
 778                 }
 779
 780                 rx_queue = &((*internals)->rx_queue[q]);
 781                 rx_queue->framecount = req->tp_frame_nr;
 782
 783                 rx_queue->map = mmap(NULL, 2 * req->tp_block_size * req->tp_block_nr,
 784                                     PROT_READ | PROT_WRITE, MAP_SHARED | MAP_LOCKED,
 785                                     qsockfd, 0);
 786                 if (rx_queue->map == MAP_FAILED) {
 787                         PMD_LOG_ERRNO(ERR,
 788                                 "%s: call to mmap failed on AF_PACKET socket for %s",
 789                                 name, pair->value);
 790                         goto error;
 791                 }
 792
 793                 /* rdsize is same for both Tx and Rx */
 794                 rdsize = req->tp_frame_nr * sizeof(*(rx_queue->rd));
 795
 796                 rx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
 797                 if (rx_queue->rd == NULL)
 798                         goto error;
 799                 for (i = 0; i < req->tp_frame_nr; ++i) {
 800                         rx_queue->rd[i].iov_base = rx_queue->map + (i * framesize);
 801                         rx_queue->rd[i].iov_len = req->tp_frame_size;
 802                 }
 803                 rx_queue->sockfd = qsockfd;
 804
 805                 tx_queue = &((*internals)->tx_queue[q]);
 806                 tx_queue->framecount = req->tp_frame_nr;
 807                 tx_queue->frame_data_size = req->tp_frame_size;
 808                 tx_queue->frame_data_size -= TPACKET2_HDRLEN -
 809                         sizeof(struct sockaddr_ll);
 810
 811                 tx_queue->map = rx_queue->map + req->tp_block_size * req->tp_block_nr;
 812
 813                 tx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
 814                 if (tx_queue->rd == NULL)
 815                         goto error;
 816                 for (i = 0; i < req->tp_frame_nr; ++i) {
 817                         tx_queue->rd[i].iov_base = tx_queue->map + (i * framesize);
 818                         tx_queue->rd[i].iov_len = req->tp_frame_size;
 819                 }
 820                 tx_queue->sockfd = qsockfd;
 821
 822                 rc = bind(qsockfd, (const struct sockaddr*)&sockaddr, sizeof(sockaddr));
 823                 if (rc == -1) {
 824                         PMD_LOG_ERRNO(ERR,
 825                                 "%s: could not bind AF_PACKET socket to %s",
 826                                 name, pair->value);
 827                         goto error;
 828                 }
 829
 830 #if defined(PACKET_FANOUT)
 831                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_FANOUT,
 832                                 &fanout_arg, sizeof(fanout_arg));
 833                 if (rc == -1) {
 834                         PMD_LOG_ERRNO(ERR,
 835                                 "%s: could not set PACKET_FANOUT on AF_PACKET socket for %s",
 836                                 name, pair->value);
 837                         goto error;
 838                 }
 839 #endif
 840         }
 841
 842         /* reserve an ethdev entry */
 843         *eth_dev = rte_eth_vdev_allocate(dev, 0);
 844         if (*eth_dev == NULL)
 845                 goto error;
 846
 847         /*
 848          * now put it all together
 849          * - store queue data in internals,
 850          * - store numa_node in eth_dev
 851          * - point eth_dev_data to internals
 852          * - and point eth_dev structure to new eth_dev_data structure
 853          */
 854
 855         (*internals)->nb_queues = nb_queues;
 856
 857         data = (*eth_dev)->data;
 858         data->dev_private = *internals;
 859         data->nb_rx_queues = (uint16_t)nb_queues;
 860         data->nb_tx_queues = (uint16_t)nb_queues;
 861         data->dev_link = pmd_link;
 862         data->mac_addrs = &(*internals)->eth_addr;
 863         data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
 864
 865         (*eth_dev)->dev_ops = &ops;
 866
 867         return 0;
 868
 869 error:
 870         if (qsockfd != -1)
 871                 close(qsockfd);
 872         for (q = 0; q < nb_queues; q++) {
 873                 if ((*internals)->rx_queue[q].map != MAP_FAILED)
 874                         munmap((*internals)->rx_queue[q].map,
 875                                2 * req->tp_block_size * req->tp_block_nr);
 876
 877                 rte_free((*internals)->rx_queue[q].rd);
 878                 rte_free((*internals)->tx_queue[q].rd);
 879                 if (((*internals)->rx_queue[q].sockfd >= 0) &&
 880                         ((*internals)->rx_queue[q].sockfd != qsockfd))
 881                         close((*internals)->rx_queue[q].sockfd);
 882         }
 883 free_internals:
 884         rte_free((*internals)->rx_queue);
 885         rte_free((*internals)->tx_queue);
 886         free((*internals)->if_name);
 887         rte_free(*internals);
 888         return -1;
 889 }
 890
 891 static int
 892 rte_eth_from_packet(struct rte_vdev_device *dev,
 893                     int const *sockfd,
 894                     struct rte_kvargs *kvlist)
 895 {
 896         const char *name = rte_vdev_device_name(dev);
 897         struct pmd_internals *internals = NULL;
 898         struct rte_eth_dev *eth_dev = NULL;
 899         struct rte_kvargs_pair *pair = NULL;
 900         unsigned k_idx;
 901         unsigned int blockcount;
 902         unsigned int blocksize;
 903         unsigned int framesize = DFLT_FRAME_SIZE;
 904         unsigned int framecount = DFLT_FRAME_COUNT;
 905         unsigned int qpairs = 1;
 906         unsigned int qdisc_bypass = 1;
 907
 908         /* do some parameter checking */
 909         if (*sockfd < 0)
 910                 return -1;
 911
 912         blocksize = getpagesize();
 913
 914         /*
 915          * Walk arguments for configurable settings
 916          */
 917         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 918                 pair = &kvlist->pairs[k_idx];
 919                 if (strstr(pair->key, ETH_AF_PACKET_NUM_Q_ARG) != NULL) {
 920                         qpairs = atoi(pair->value);
 921                         if (qpairs < 1) {
 922                                 PMD_LOG(ERR,
 923                                         "%s: invalid qpairs value",
 924                                         name);
 925                                 return -1;
 926                         }
 927                         continue;
 928                 }
 929                 if (strstr(pair->key, ETH_AF_PACKET_BLOCKSIZE_ARG) != NULL) {
 930                         blocksize = atoi(pair->value);
 931                         if (!blocksize) {
 932                                 PMD_LOG(ERR,
 933                                         "%s: invalid blocksize value",
 934                                         name);
 935                                 return -1;
 936                         }
 937                         continue;
 938                 }
 939                 if (strstr(pair->key, ETH_AF_PACKET_FRAMESIZE_ARG) != NULL) {
 940                         framesize = atoi(pair->value);
 941                         if (!framesize) {
 942                                 PMD_LOG(ERR,
 943                                         "%s: invalid framesize value",
 944                                         name);
 945                                 return -1;
 946                         }
 947                         continue;
 948                 }
 949                 if (strstr(pair->key, ETH_AF_PACKET_FRAMECOUNT_ARG) != NULL) {
 950                         framecount = atoi(pair->value);
 951                         if (!framecount) {
 952                                 PMD_LOG(ERR,
 953                                         "%s: invalid framecount value",
 954                                         name);
 955                                 return -1;
 956                         }
 957                         continue;
 958                 }
 959                 if (strstr(pair->key, ETH_AF_PACKET_QDISC_BYPASS_ARG) != NULL) {
 960                         qdisc_bypass = atoi(pair->value);
 961                         if (qdisc_bypass > 1) {
 962                                 PMD_LOG(ERR,
 963                                         "%s: invalid bypass value",
 964                                         name);
 965                                 return -1;
 966                         }
 967                         continue;
 968                 }
 969         }
 970
 971         if (framesize > blocksize) {
 972                 PMD_LOG(ERR,
 973                         "%s: AF_PACKET MMAP frame size exceeds block size!",
 974                         name);
 975                 return -1;
 976         }
 977
 978         blockcount = framecount / (blocksize / framesize);
 979         if (!blockcount) {
 980                 PMD_LOG(ERR,
 981                         "%s: invalid AF_PACKET MMAP parameters", name);
 982                 return -1;
 983         }
 984
 985         PMD_LOG(INFO, "%s: AF_PACKET MMAP parameters:", name);
 986         PMD_LOG(INFO, "%s:\tblock size %d", name, blocksize);
 987         PMD_LOG(INFO, "%s:\tblock count %d", name, blockcount);
 988         PMD_LOG(INFO, "%s:\tframe size %d", name, framesize);
 989         PMD_LOG(INFO, "%s:\tframe count %d", name, framecount);
 990
 991         if (rte_pmd_init_internals(dev, *sockfd, qpairs,
 992                                    blocksize, blockcount,
 993                                    framesize, framecount,
 994                                    qdisc_bypass,
 995                                    &internals, &eth_dev,
 996                                    kvlist) < 0)
 997                 return -1;
 998
 999         eth_dev->rx_pkt_burst = eth_af_packet_rx;
1000         eth_dev->tx_pkt_burst = eth_af_packet_tx;
1001
1002         rte_eth_dev_probing_finish(eth_dev);
1003         return 0;
1004 }
1005
1006 static int
1007 rte_pmd_af_packet_probe(struct rte_vdev_device *dev)
1008 {
1009         int ret = 0;
1010         struct rte_kvargs *kvlist;
1011         int sockfd = -1;
1012         struct rte_eth_dev *eth_dev;
1013         const char *name = rte_vdev_device_name(dev);
1014
1015         PMD_LOG(INFO, "Initializing pmd_af_packet for %s", name);
1016
1017         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1018                 eth_dev = rte_eth_dev_attach_secondary(name);
1019                 if (!eth_dev) {
1020                         PMD_LOG(ERR, "Failed to probe %s", name);
1021                         return -1;
1022                 }
1023                 /* TODO: request info from primary to set up Rx and Tx */
1024                 eth_dev->dev_ops = &ops;
1025                 eth_dev->device = &dev->device;
1026                 rte_eth_dev_probing_finish(eth_dev);
1027                 return 0;
1028         }
1029
1030         kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_arguments);
1031         if (kvlist == NULL) {
1032                 ret = -1;
1033                 goto exit;
1034         }
1035
1036         /*
1037          * If iface argument is passed we open the NICs and use them for
1038          * reading / writing
1039          */
1040         if (rte_kvargs_count(kvlist, ETH_AF_PACKET_IFACE_ARG) == 1) {
1041
1042                 ret = rte_kvargs_process(kvlist, ETH_AF_PACKET_IFACE_ARG,
1043                                          &open_packet_iface, &sockfd);
1044                 if (ret < 0)
1045                         goto exit;
1046         }
1047
1048         if (dev->device.numa_node == SOCKET_ID_ANY)
1049                 dev->device.numa_node = rte_socket_id();
1050
1051         ret = rte_eth_from_packet(dev, &sockfd, kvlist);
1052         close(sockfd); /* no longer needed */
1053
1054 exit:
1055         rte_kvargs_free(kvlist);
1056         return ret;
1057 }
1058
1059 static int
1060 rte_pmd_af_packet_remove(struct rte_vdev_device *dev)
1061 {
1062         struct rte_eth_dev *eth_dev;
1063
1064         if (dev == NULL)
1065                 return -1;
1066
1067         /* find the ethdev entry */
1068         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1069         if (eth_dev == NULL)
1070                 return 0; /* port already released */
1071
1072         eth_dev_close(eth_dev);
1073         rte_eth_dev_release_port(eth_dev);
1074
1075         return 0;
1076 }
1077
1078 static struct rte_vdev_driver pmd_af_packet_drv = {
1079         .probe = rte_pmd_af_packet_probe,
1080         .remove = rte_pmd_af_packet_remove,
1081 };
1082
1083 RTE_PMD_REGISTER_VDEV(net_af_packet, pmd_af_packet_drv);
1084 RTE_PMD_REGISTER_ALIAS(net_af_packet, eth_af_packet);
1085 RTE_PMD_REGISTER_PARAM_STRING(net_af_packet,
1086         "iface=<string> "
1087         "qpairs=<int> "
1088         "blocksz=<int> "
1089         "framesz=<int> "
1090         "framecnt=<int> "
1091         "qdisc_bypass=<0|1>");