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 <rte_ethdev_driver.h>
  12 #include <rte_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(af_packet_logtype, pmd.net.packet, 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 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 int
 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         dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
 319                 DEV_TX_OFFLOAD_VLAN_INSERT;
 320
 321         return 0;
 322 }
 323
 324 static int
 325 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *igb_stats)
 326 {
 327         unsigned i, imax;
 328         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 329         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 330         const struct pmd_internals *internal = dev->data->dev_private;
 331
 332         imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
 333                 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
 334         for (i = 0; i < imax; i++) {
 335                 igb_stats->q_ipackets[i] = internal->rx_queue[i].rx_pkts;
 336                 igb_stats->q_ibytes[i] = internal->rx_queue[i].rx_bytes;
 337                 rx_total += igb_stats->q_ipackets[i];
 338                 rx_bytes_total += igb_stats->q_ibytes[i];
 339         }
 340
 341         imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
 342                 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
 343         for (i = 0; i < imax; i++) {
 344                 igb_stats->q_opackets[i] = internal->tx_queue[i].tx_pkts;
 345                 igb_stats->q_obytes[i] = internal->tx_queue[i].tx_bytes;
 346                 tx_total += igb_stats->q_opackets[i];
 347                 tx_err_total += internal->tx_queue[i].err_pkts;
 348                 tx_bytes_total += igb_stats->q_obytes[i];
 349         }
 350
 351         igb_stats->ipackets = rx_total;
 352         igb_stats->ibytes = rx_bytes_total;
 353         igb_stats->opackets = tx_total;
 354         igb_stats->oerrors = tx_err_total;
 355         igb_stats->obytes = tx_bytes_total;
 356         return 0;
 357 }
 358
 359 static int
 360 eth_stats_reset(struct rte_eth_dev *dev)
 361 {
 362         unsigned i;
 363         struct pmd_internals *internal = dev->data->dev_private;
 364
 365         for (i = 0; i < internal->nb_queues; i++) {
 366                 internal->rx_queue[i].rx_pkts = 0;
 367                 internal->rx_queue[i].rx_bytes = 0;
 368         }
 369
 370         for (i = 0; i < internal->nb_queues; i++) {
 371                 internal->tx_queue[i].tx_pkts = 0;
 372                 internal->tx_queue[i].err_pkts = 0;
 373                 internal->tx_queue[i].tx_bytes = 0;
 374         }
 375
 376         return 0;
 377 }
 378
 379 static int
 380 eth_dev_close(struct rte_eth_dev *dev)
 381 {
 382         struct pmd_internals *internals;
 383         struct tpacket_req *req;
 384         unsigned int q;
 385
 386         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 387                 return 0;
 388
 389         PMD_LOG(INFO, "Closing AF_PACKET ethdev on NUMA socket %u",
 390                 rte_socket_id());
 391
 392         internals = dev->data->dev_private;
 393         req = &internals->req;
 394         for (q = 0; q < internals->nb_queues; q++) {
 395                 munmap(internals->rx_queue[q].map,
 396                         2 * req->tp_block_size * req->tp_block_nr);
 397                 rte_free(internals->rx_queue[q].rd);
 398                 rte_free(internals->tx_queue[q].rd);
 399         }
 400         free(internals->if_name);
 401         rte_free(internals->rx_queue);
 402         rte_free(internals->tx_queue);
 403
 404         /* mac_addrs must not be freed alone because part of dev_private */
 405         dev->data->mac_addrs = NULL;
 406         return 0;
 407 }
 408
 409 static void
 410 eth_queue_release(void *q __rte_unused)
 411 {
 412 }
 413
 414 static int
 415 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 416                 int wait_to_complete __rte_unused)
 417 {
 418         return 0;
 419 }
 420
 421 static int
 422 eth_rx_queue_setup(struct rte_eth_dev *dev,
 423                    uint16_t rx_queue_id,
 424                    uint16_t nb_rx_desc __rte_unused,
 425                    unsigned int socket_id __rte_unused,
 426                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 427                    struct rte_mempool *mb_pool)
 428 {
 429         struct pmd_internals *internals = dev->data->dev_private;
 430         struct pkt_rx_queue *pkt_q = &internals->rx_queue[rx_queue_id];
 431         unsigned int buf_size, data_size;
 432
 433         pkt_q->mb_pool = mb_pool;
 434
 435         /* Now get the space available for data in the mbuf */
 436         buf_size = rte_pktmbuf_data_room_size(pkt_q->mb_pool) -
 437                 RTE_PKTMBUF_HEADROOM;
 438         data_size = internals->req.tp_frame_size;
 439         data_size -= TPACKET2_HDRLEN - sizeof(struct sockaddr_ll);
 440
 441         if (data_size > buf_size) {
 442                 PMD_LOG(ERR,
 443                         "%s: %d bytes will not fit in mbuf (%d bytes)",
 444                         dev->device->name, data_size, buf_size);
 445                 return -ENOMEM;
 446         }
 447
 448         dev->data->rx_queues[rx_queue_id] = pkt_q;
 449         pkt_q->in_port = dev->data->port_id;
 450
 451         return 0;
 452 }
 453
 454 static int
 455 eth_tx_queue_setup(struct rte_eth_dev *dev,
 456                    uint16_t tx_queue_id,
 457                    uint16_t nb_tx_desc __rte_unused,
 458                    unsigned int socket_id __rte_unused,
 459                    const struct rte_eth_txconf *tx_conf __rte_unused)
 460 {
 461
 462         struct pmd_internals *internals = dev->data->dev_private;
 463
 464         dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
 465         return 0;
 466 }
 467
 468 static int
 469 eth_dev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 470 {
 471         struct pmd_internals *internals = dev->data->dev_private;
 472         struct ifreq ifr = { .ifr_mtu = mtu };
 473         int ret;
 474         int s;
 475         unsigned int data_size = internals->req.tp_frame_size -
 476                                  TPACKET2_HDRLEN;
 477
 478         if (mtu > data_size)
 479                 return -EINVAL;
 480
 481         s = socket(PF_INET, SOCK_DGRAM, 0);
 482         if (s < 0)
 483                 return -EINVAL;
 484
 485         strlcpy(ifr.ifr_name, internals->if_name, IFNAMSIZ);
 486         ret = ioctl(s, SIOCSIFMTU, &ifr);
 487         close(s);
 488
 489         if (ret < 0)
 490                 return -EINVAL;
 491
 492         return 0;
 493 }
 494
 495 static int
 496 eth_dev_macaddr_set(struct rte_eth_dev *dev, struct rte_ether_addr *addr)
 497 {
 498         struct pmd_internals *internals = dev->data->dev_private;
 499         struct ifreq ifr = { };
 500         int sockfd = internals->rx_queue[0].sockfd;
 501         int ret;
 502
 503         if (sockfd == -1) {
 504                 PMD_LOG(ERR, "receive socket not found");
 505                 return -EINVAL;
 506         }
 507
 508         strlcpy(ifr.ifr_name, internals->if_name, IFNAMSIZ);
 509         ifr.ifr_hwaddr.sa_family = ARPHRD_ETHER;
 510         memcpy(ifr.ifr_hwaddr.sa_data, addr, sizeof(*addr));
 511         ret = ioctl(sockfd, SIOCSIFHWADDR, &ifr);
 512
 513         if (ret < 0) {
 514                 PMD_LOG_ERRNO(ERR, "ioctl(SIOCSIFHWADDR) failed");
 515                 return -EINVAL;
 516         }
 517
 518         return 0;
 519 }
 520
 521 static int
 522 eth_dev_change_flags(char *if_name, uint32_t flags, uint32_t mask)
 523 {
 524         struct ifreq ifr;
 525         int ret = 0;
 526         int s;
 527
 528         s = socket(PF_INET, SOCK_DGRAM, 0);
 529         if (s < 0)
 530                 return -errno;
 531
 532         strlcpy(ifr.ifr_name, if_name, IFNAMSIZ);
 533         if (ioctl(s, SIOCGIFFLAGS, &ifr) < 0) {
 534                 ret = -errno;
 535                 goto out;
 536         }
 537         ifr.ifr_flags &= mask;
 538         ifr.ifr_flags |= flags;
 539         if (ioctl(s, SIOCSIFFLAGS, &ifr) < 0) {
 540                 ret = -errno;
 541                 goto out;
 542         }
 543 out:
 544         close(s);
 545         return ret;
 546 }
 547
 548 static int
 549 eth_dev_promiscuous_enable(struct rte_eth_dev *dev)
 550 {
 551         struct pmd_internals *internals = dev->data->dev_private;
 552
 553         return eth_dev_change_flags(internals->if_name, IFF_PROMISC, ~0);
 554 }
 555
 556 static int
 557 eth_dev_promiscuous_disable(struct rte_eth_dev *dev)
 558 {
 559         struct pmd_internals *internals = dev->data->dev_private;
 560
 561         return eth_dev_change_flags(internals->if_name, 0, ~IFF_PROMISC);
 562 }
 563
 564 static const struct eth_dev_ops ops = {
 565         .dev_start = eth_dev_start,
 566         .dev_stop = eth_dev_stop,
 567         .dev_close = eth_dev_close,
 568         .dev_configure = eth_dev_configure,
 569         .dev_infos_get = eth_dev_info,
 570         .mac_addr_set = eth_dev_macaddr_set,
 571         .mtu_set = eth_dev_mtu_set,
 572         .promiscuous_enable = eth_dev_promiscuous_enable,
 573         .promiscuous_disable = eth_dev_promiscuous_disable,
 574         .rx_queue_setup = eth_rx_queue_setup,
 575         .tx_queue_setup = eth_tx_queue_setup,
 576         .rx_queue_release = eth_queue_release,
 577         .tx_queue_release = eth_queue_release,
 578         .link_update = eth_link_update,
 579         .stats_get = eth_stats_get,
 580         .stats_reset = eth_stats_reset,
 581 };
 582
 583 /*
 584  * Opens an AF_PACKET socket
 585  */
 586 static int
 587 open_packet_iface(const char *key __rte_unused,
 588                   const char *value __rte_unused,
 589                   void *extra_args)
 590 {
 591         int *sockfd = extra_args;
 592
 593         /* Open an AF_PACKET socket... */
 594         *sockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
 595         if (*sockfd == -1) {
 596                 PMD_LOG(ERR, "Could not open AF_PACKET socket");
 597                 return -1;
 598         }
 599
 600         return 0;
 601 }
 602
 603 static int
 604 rte_pmd_init_internals(struct rte_vdev_device *dev,
 605                        const int sockfd,
 606                        const unsigned nb_queues,
 607                        unsigned int blocksize,
 608                        unsigned int blockcnt,
 609                        unsigned int framesize,
 610                        unsigned int framecnt,
 611                        unsigned int qdisc_bypass,
 612                        struct pmd_internals **internals,
 613                        struct rte_eth_dev **eth_dev,
 614                        struct rte_kvargs *kvlist)
 615 {
 616         const char *name = rte_vdev_device_name(dev);
 617         const unsigned int numa_node = dev->device.numa_node;
 618         struct rte_eth_dev_data *data = NULL;
 619         struct rte_kvargs_pair *pair = NULL;
 620         struct ifreq ifr;
 621         size_t ifnamelen;
 622         unsigned k_idx;
 623         struct sockaddr_ll sockaddr;
 624         struct tpacket_req *req;
 625         struct pkt_rx_queue *rx_queue;
 626         struct pkt_tx_queue *tx_queue;
 627         int rc, tpver, discard;
 628         int qsockfd = -1;
 629         unsigned int i, q, rdsize;
 630 #if defined(PACKET_FANOUT)
 631         int fanout_arg;
 632 #endif
 633
 634         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 635                 pair = &kvlist->pairs[k_idx];
 636                 if (strstr(pair->key, ETH_AF_PACKET_IFACE_ARG) != NULL)
 637                         break;
 638         }
 639         if (pair == NULL) {
 640                 PMD_LOG(ERR,
 641                         "%s: no interface specified for AF_PACKET ethdev",
 642                         name);
 643                 return -1;
 644         }
 645
 646         PMD_LOG(INFO,
 647                 "%s: creating AF_PACKET-backed ethdev on numa socket %u",
 648                 name, numa_node);
 649
 650         *internals = rte_zmalloc_socket(name, sizeof(**internals),
 651                                         0, numa_node);
 652         if (*internals == NULL)
 653                 return -1;
 654
 655
 656         (*internals)->rx_queue = rte_calloc_socket("af_packet_rx",
 657                                                 nb_queues,
 658                                                 sizeof(struct pkt_rx_queue),
 659                                                 0, numa_node);
 660         (*internals)->tx_queue = rte_calloc_socket("af_packet_tx",
 661                                                 nb_queues,
 662                                                 sizeof(struct pkt_tx_queue),
 663                                                 0, numa_node);
 664         if (!(*internals)->rx_queue || !(*internals)->tx_queue) {
 665                 goto free_internals;
 666         }
 667
 668         for (q = 0; q < nb_queues; q++) {
 669                 (*internals)->rx_queue[q].map = MAP_FAILED;
 670                 (*internals)->tx_queue[q].map = MAP_FAILED;
 671                 (*internals)->rx_queue[q].sockfd = -1;
 672                 (*internals)->tx_queue[q].sockfd = -1;
 673         }
 674
 675         req = &((*internals)->req);
 676
 677         req->tp_block_size = blocksize;
 678         req->tp_block_nr = blockcnt;
 679         req->tp_frame_size = framesize;
 680         req->tp_frame_nr = framecnt;
 681
 682         ifnamelen = strlen(pair->value);
 683         if (ifnamelen < sizeof(ifr.ifr_name)) {
 684                 memcpy(ifr.ifr_name, pair->value, ifnamelen);
 685                 ifr.ifr_name[ifnamelen] = '\0';
 686         } else {
 687                 PMD_LOG(ERR,
 688                         "%s: I/F name too long (%s)",
 689                         name, pair->value);
 690                 goto free_internals;
 691         }
 692         if (ioctl(sockfd, SIOCGIFINDEX, &ifr) == -1) {
 693                 PMD_LOG_ERRNO(ERR, "%s: ioctl failed (SIOCGIFINDEX)", name);
 694                 goto free_internals;
 695         }
 696         (*internals)->if_name = strdup(pair->value);
 697         if ((*internals)->if_name == NULL)
 698                 goto free_internals;
 699         (*internals)->if_index = ifr.ifr_ifindex;
 700
 701         if (ioctl(sockfd, SIOCGIFHWADDR, &ifr) == -1) {
 702                 PMD_LOG_ERRNO(ERR, "%s: ioctl failed (SIOCGIFHWADDR)", name);
 703                 goto free_internals;
 704         }
 705         memcpy(&(*internals)->eth_addr, ifr.ifr_hwaddr.sa_data, ETH_ALEN);
 706
 707         memset(&sockaddr, 0, sizeof(sockaddr));
 708         sockaddr.sll_family = AF_PACKET;
 709         sockaddr.sll_protocol = htons(ETH_P_ALL);
 710         sockaddr.sll_ifindex = (*internals)->if_index;
 711
 712 #if defined(PACKET_FANOUT)
 713         fanout_arg = (getpid() ^ (*internals)->if_index) & 0xffff;
 714         fanout_arg |= (PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_DEFRAG) << 16;
 715 #if defined(PACKET_FANOUT_FLAG_ROLLOVER)
 716         fanout_arg |= PACKET_FANOUT_FLAG_ROLLOVER << 16;
 717 #endif
 718 #endif
 719
 720         for (q = 0; q < nb_queues; q++) {
 721                 /* Open an AF_PACKET socket for this queue... */
 722                 qsockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
 723                 if (qsockfd == -1) {
 724                         PMD_LOG_ERRNO(ERR,
 725                                 "%s: could not open AF_PACKET socket",
 726                                 name);
 727                         goto error;
 728                 }
 729
 730                 tpver = TPACKET_V2;
 731                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_VERSION,
 732                                 &tpver, sizeof(tpver));
 733                 if (rc == -1) {
 734                         PMD_LOG_ERRNO(ERR,
 735                                 "%s: could not set PACKET_VERSION on AF_PACKET socket for %s",
 736                                 name, pair->value);
 737                         goto error;
 738                 }
 739
 740                 discard = 1;
 741                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_LOSS,
 742                                 &discard, sizeof(discard));
 743                 if (rc == -1) {
 744                         PMD_LOG_ERRNO(ERR,
 745                                 "%s: could not set PACKET_LOSS on AF_PACKET socket for %s",
 746                                 name, pair->value);
 747                         goto error;
 748                 }
 749
 750 #if defined(PACKET_QDISC_BYPASS)
 751                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_QDISC_BYPASS,
 752                                 &qdisc_bypass, sizeof(qdisc_bypass));
 753                 if (rc == -1) {
 754                         PMD_LOG_ERRNO(ERR,
 755                                 "%s: could not set PACKET_QDISC_BYPASS on AF_PACKET socket for %s",
 756                                 name, pair->value);
 757                         goto error;
 758                 }
 759 #else
 760                 RTE_SET_USED(qdisc_bypass);
 761 #endif
 762
 763                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_RX_RING, req, sizeof(*req));
 764                 if (rc == -1) {
 765                         PMD_LOG_ERRNO(ERR,
 766                                 "%s: could not set PACKET_RX_RING on AF_PACKET socket for %s",
 767                                 name, pair->value);
 768                         goto error;
 769                 }
 770
 771                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_TX_RING, req, sizeof(*req));
 772                 if (rc == -1) {
 773                         PMD_LOG_ERRNO(ERR,
 774                                 "%s: could not set PACKET_TX_RING on AF_PACKET "
 775                                 "socket for %s", name, pair->value);
 776                         goto error;
 777                 }
 778
 779                 rx_queue = &((*internals)->rx_queue[q]);
 780                 rx_queue->framecount = req->tp_frame_nr;
 781
 782                 rx_queue->map = mmap(NULL, 2 * req->tp_block_size * req->tp_block_nr,
 783                                     PROT_READ | PROT_WRITE, MAP_SHARED | MAP_LOCKED,
 784                                     qsockfd, 0);
 785                 if (rx_queue->map == MAP_FAILED) {
 786                         PMD_LOG_ERRNO(ERR,
 787                                 "%s: call to mmap failed on AF_PACKET socket for %s",
 788                                 name, pair->value);
 789                         goto error;
 790                 }
 791
 792                 /* rdsize is same for both Tx and Rx */
 793                 rdsize = req->tp_frame_nr * sizeof(*(rx_queue->rd));
 794
 795                 rx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
 796                 if (rx_queue->rd == NULL)
 797                         goto error;
 798                 for (i = 0; i < req->tp_frame_nr; ++i) {
 799                         rx_queue->rd[i].iov_base = rx_queue->map + (i * framesize);
 800                         rx_queue->rd[i].iov_len = req->tp_frame_size;
 801                 }
 802                 rx_queue->sockfd = qsockfd;
 803
 804                 tx_queue = &((*internals)->tx_queue[q]);
 805                 tx_queue->framecount = req->tp_frame_nr;
 806                 tx_queue->frame_data_size = req->tp_frame_size;
 807                 tx_queue->frame_data_size -= TPACKET2_HDRLEN -
 808                         sizeof(struct sockaddr_ll);
 809
 810                 tx_queue->map = rx_queue->map + req->tp_block_size * req->tp_block_nr;
 811
 812                 tx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
 813                 if (tx_queue->rd == NULL)
 814                         goto error;
 815                 for (i = 0; i < req->tp_frame_nr; ++i) {
 816                         tx_queue->rd[i].iov_base = tx_queue->map + (i * framesize);
 817                         tx_queue->rd[i].iov_len = req->tp_frame_size;
 818                 }
 819                 tx_queue->sockfd = qsockfd;
 820
 821                 rc = bind(qsockfd, (const struct sockaddr*)&sockaddr, sizeof(sockaddr));
 822                 if (rc == -1) {
 823                         PMD_LOG_ERRNO(ERR,
 824                                 "%s: could not bind AF_PACKET socket to %s",
 825                                 name, pair->value);
 826                         goto error;
 827                 }
 828
 829 #if defined(PACKET_FANOUT)
 830                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_FANOUT,
 831                                 &fanout_arg, sizeof(fanout_arg));
 832                 if (rc == -1) {
 833                         PMD_LOG_ERRNO(ERR,
 834                                 "%s: could not set PACKET_FANOUT on AF_PACKET socket for %s",
 835                                 name, pair->value);
 836                         goto error;
 837                 }
 838 #endif
 839         }
 840
 841         /* reserve an ethdev entry */
 842         *eth_dev = rte_eth_vdev_allocate(dev, 0);
 843         if (*eth_dev == NULL)
 844                 goto error;
 845
 846         /*
 847          * now put it all together
 848          * - store queue data in internals,
 849          * - store numa_node in eth_dev
 850          * - point eth_dev_data to internals
 851          * - and point eth_dev structure to new eth_dev_data structure
 852          */
 853
 854         (*internals)->nb_queues = nb_queues;
 855
 856         data = (*eth_dev)->data;
 857         data->dev_private = *internals;
 858         data->nb_rx_queues = (uint16_t)nb_queues;
 859         data->nb_tx_queues = (uint16_t)nb_queues;
 860         data->dev_link = pmd_link;
 861         data->mac_addrs = &(*internals)->eth_addr;
 862         data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
 863
 864         (*eth_dev)->dev_ops = &ops;
 865
 866         return 0;
 867
 868 error:
 869         if (qsockfd != -1)
 870                 close(qsockfd);
 871         for (q = 0; q < nb_queues; q++) {
 872                 if ((*internals)->rx_queue[q].map != MAP_FAILED)
 873                         munmap((*internals)->rx_queue[q].map,
 874                                2 * req->tp_block_size * req->tp_block_nr);
 875
 876                 rte_free((*internals)->rx_queue[q].rd);
 877                 rte_free((*internals)->tx_queue[q].rd);
 878                 if (((*internals)->rx_queue[q].sockfd >= 0) &&
 879                         ((*internals)->rx_queue[q].sockfd != qsockfd))
 880                         close((*internals)->rx_queue[q].sockfd);
 881         }
 882 free_internals:
 883         rte_free((*internals)->rx_queue);
 884         rte_free((*internals)->tx_queue);
 885         free((*internals)->if_name);
 886         rte_free(*internals);
 887         return -1;
 888 }
 889
 890 static int
 891 rte_eth_from_packet(struct rte_vdev_device *dev,
 892                     int const *sockfd,
 893                     struct rte_kvargs *kvlist)
 894 {
 895         const char *name = rte_vdev_device_name(dev);
 896         struct pmd_internals *internals = NULL;
 897         struct rte_eth_dev *eth_dev = NULL;
 898         struct rte_kvargs_pair *pair = NULL;
 899         unsigned k_idx;
 900         unsigned int blockcount;
 901         unsigned int blocksize;
 902         unsigned int framesize = DFLT_FRAME_SIZE;
 903         unsigned int framecount = DFLT_FRAME_COUNT;
 904         unsigned int qpairs = 1;
 905         unsigned int qdisc_bypass = 1;
 906
 907         /* do some parameter checking */
 908         if (*sockfd < 0)
 909                 return -1;
 910
 911         blocksize = getpagesize();
 912
 913         /*
 914          * Walk arguments for configurable settings
 915          */
 916         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 917                 pair = &kvlist->pairs[k_idx];
 918                 if (strstr(pair->key, ETH_AF_PACKET_NUM_Q_ARG) != NULL) {
 919                         qpairs = atoi(pair->value);
 920                         if (qpairs < 1) {
 921                                 PMD_LOG(ERR,
 922                                         "%s: invalid qpairs value",
 923                                         name);
 924                                 return -1;
 925                         }
 926                         continue;
 927                 }
 928                 if (strstr(pair->key, ETH_AF_PACKET_BLOCKSIZE_ARG) != NULL) {
 929                         blocksize = atoi(pair->value);
 930                         if (!blocksize) {
 931                                 PMD_LOG(ERR,
 932                                         "%s: invalid blocksize value",
 933                                         name);
 934                                 return -1;
 935                         }
 936                         continue;
 937                 }
 938                 if (strstr(pair->key, ETH_AF_PACKET_FRAMESIZE_ARG) != NULL) {
 939                         framesize = atoi(pair->value);
 940                         if (!framesize) {
 941                                 PMD_LOG(ERR,
 942                                         "%s: invalid framesize value",
 943                                         name);
 944                                 return -1;
 945                         }
 946                         continue;
 947                 }
 948                 if (strstr(pair->key, ETH_AF_PACKET_FRAMECOUNT_ARG) != NULL) {
 949                         framecount = atoi(pair->value);
 950                         if (!framecount) {
 951                                 PMD_LOG(ERR,
 952                                         "%s: invalid framecount value",
 953                                         name);
 954                                 return -1;
 955                         }
 956                         continue;
 957                 }
 958                 if (strstr(pair->key, ETH_AF_PACKET_QDISC_BYPASS_ARG) != NULL) {
 959                         qdisc_bypass = atoi(pair->value);
 960                         if (qdisc_bypass > 1) {
 961                                 PMD_LOG(ERR,
 962                                         "%s: invalid bypass value",
 963                                         name);
 964                                 return -1;
 965                         }
 966                         continue;
 967                 }
 968         }
 969
 970         if (framesize > blocksize) {
 971                 PMD_LOG(ERR,
 972                         "%s: AF_PACKET MMAP frame size exceeds block size!",
 973                         name);
 974                 return -1;
 975         }
 976
 977         blockcount = framecount / (blocksize / framesize);
 978         if (!blockcount) {
 979                 PMD_LOG(ERR,
 980                         "%s: invalid AF_PACKET MMAP parameters", name);
 981                 return -1;
 982         }
 983
 984         PMD_LOG(INFO, "%s: AF_PACKET MMAP parameters:", name);
 985         PMD_LOG(INFO, "%s:\tblock size %d", name, blocksize);
 986         PMD_LOG(INFO, "%s:\tblock count %d", name, blockcount);
 987         PMD_LOG(INFO, "%s:\tframe size %d", name, framesize);
 988         PMD_LOG(INFO, "%s:\tframe count %d", name, framecount);
 989
 990         if (rte_pmd_init_internals(dev, *sockfd, qpairs,
 991                                    blocksize, blockcount,
 992                                    framesize, framecount,
 993                                    qdisc_bypass,
 994                                    &internals, &eth_dev,
 995                                    kvlist) < 0)
 996                 return -1;
 997
 998         eth_dev->rx_pkt_burst = eth_af_packet_rx;
 999         eth_dev->tx_pkt_burst = eth_af_packet_tx;
1000
1001         rte_eth_dev_probing_finish(eth_dev);
1002         return 0;
1003 }
1004
1005 static int
1006 rte_pmd_af_packet_probe(struct rte_vdev_device *dev)
1007 {
1008         int ret = 0;
1009         struct rte_kvargs *kvlist;
1010         int sockfd = -1;
1011         struct rte_eth_dev *eth_dev;
1012         const char *name = rte_vdev_device_name(dev);
1013
1014         PMD_LOG(INFO, "Initializing pmd_af_packet for %s", name);
1015
1016         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1017                 eth_dev = rte_eth_dev_attach_secondary(name);
1018                 if (!eth_dev) {
1019                         PMD_LOG(ERR, "Failed to probe %s", name);
1020                         return -1;
1021                 }
1022                 /* TODO: request info from primary to set up Rx and Tx */
1023                 eth_dev->dev_ops = &ops;
1024                 eth_dev->device = &dev->device;
1025                 rte_eth_dev_probing_finish(eth_dev);
1026                 return 0;
1027         }
1028
1029         kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_arguments);
1030         if (kvlist == NULL) {
1031                 ret = -1;
1032                 goto exit;
1033         }
1034
1035         /*
1036          * If iface argument is passed we open the NICs and use them for
1037          * reading / writing
1038          */
1039         if (rte_kvargs_count(kvlist, ETH_AF_PACKET_IFACE_ARG) == 1) {
1040
1041                 ret = rte_kvargs_process(kvlist, ETH_AF_PACKET_IFACE_ARG,
1042                                          &open_packet_iface, &sockfd);
1043                 if (ret < 0)
1044                         goto exit;
1045         }
1046
1047         if (dev->device.numa_node == SOCKET_ID_ANY)
1048                 dev->device.numa_node = rte_socket_id();
1049
1050         ret = rte_eth_from_packet(dev, &sockfd, kvlist);
1051         close(sockfd); /* no longer needed */
1052
1053 exit:
1054         rte_kvargs_free(kvlist);
1055         return ret;
1056 }
1057
1058 static int
1059 rte_pmd_af_packet_remove(struct rte_vdev_device *dev)
1060 {
1061         struct rte_eth_dev *eth_dev;
1062
1063         if (dev == NULL)
1064                 return -1;
1065
1066         /* find the ethdev entry */
1067         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1068         if (eth_dev == NULL)
1069                 return 0; /* port already released */
1070
1071         eth_dev_close(eth_dev);
1072         rte_eth_dev_release_port(eth_dev);
1073
1074         return 0;
1075 }
1076
1077 static struct rte_vdev_driver pmd_af_packet_drv = {
1078         .probe = rte_pmd_af_packet_probe,
1079         .remove = rte_pmd_af_packet_remove,
1080 };
1081
1082 RTE_PMD_REGISTER_VDEV(net_af_packet, pmd_af_packet_drv);
1083 RTE_PMD_REGISTER_ALIAS(net_af_packet, eth_af_packet);
1084 RTE_PMD_REGISTER_PARAM_STRING(net_af_packet,
1085         "iface=<string> "
1086         "qpairs=<int> "
1087         "blocksz=<int> "
1088         "framesz=<int> "
1089         "framecnt=<int> "
1090         "qdisc_bypass=<0|1>");