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 <sys/types.h>
  23 #include <sys/socket.h>
  24 #include <sys/ioctl.h>
  25 #include <string.h>
  26 #include <sys/mman.h>
  27 #include <unistd.h>
  28 #include <poll.h>
  29
  30 #define ETH_AF_PACKET_IFACE_ARG         "iface"
  31 #define ETH_AF_PACKET_NUM_Q_ARG         "qpairs"
  32 #define ETH_AF_PACKET_BLOCKSIZE_ARG     "blocksz"
  33 #define ETH_AF_PACKET_FRAMESIZE_ARG     "framesz"
  34 #define ETH_AF_PACKET_FRAMECOUNT_ARG    "framecnt"
  35 #define ETH_AF_PACKET_QDISC_BYPASS_ARG  "qdisc_bypass"
  36
  37 #define DFLT_FRAME_SIZE         (1 << 11)
  38 #define DFLT_FRAME_COUNT        (1 << 9)
  39
  40 #define RTE_PMD_AF_PACKET_MAX_RINGS 16
  41
  42 struct pkt_rx_queue {
  43         int sockfd;
  44
  45         struct iovec *rd;
  46         uint8_t *map;
  47         unsigned int framecount;
  48         unsigned int framenum;
  49
  50         struct rte_mempool *mb_pool;
  51         uint16_t in_port;
  52
  53         volatile unsigned long rx_pkts;
  54         volatile unsigned long err_pkts;
  55         volatile unsigned long rx_bytes;
  56 };
  57
  58 struct pkt_tx_queue {
  59         int sockfd;
  60         unsigned int frame_data_size;
  61
  62         struct iovec *rd;
  63         uint8_t *map;
  64         unsigned int framecount;
  65         unsigned int framenum;
  66
  67         volatile unsigned long tx_pkts;
  68         volatile unsigned long err_pkts;
  69         volatile unsigned long tx_bytes;
  70 };
  71
  72 struct pmd_internals {
  73         unsigned nb_queues;
  74
  75         int if_index;
  76         char *if_name;
  77         struct rte_ether_addr eth_addr;
  78
  79         struct tpacket_req req;
  80
  81         struct pkt_rx_queue rx_queue[RTE_PMD_AF_PACKET_MAX_RINGS];
  82         struct pkt_tx_queue tx_queue[RTE_PMD_AF_PACKET_MAX_RINGS];
  83 };
  84
  85 static const char *valid_arguments[] = {
  86         ETH_AF_PACKET_IFACE_ARG,
  87         ETH_AF_PACKET_NUM_Q_ARG,
  88         ETH_AF_PACKET_BLOCKSIZE_ARG,
  89         ETH_AF_PACKET_FRAMESIZE_ARG,
  90         ETH_AF_PACKET_FRAMECOUNT_ARG,
  91         ETH_AF_PACKET_QDISC_BYPASS_ARG,
  92         NULL
  93 };
  94
  95 static struct rte_eth_link pmd_link = {
  96         .link_speed = ETH_SPEED_NUM_10G,
  97         .link_duplex = ETH_LINK_FULL_DUPLEX,
  98         .link_status = ETH_LINK_DOWN,
  99         .link_autoneg = ETH_LINK_FIXED,
 100 };
 101
 102 static int af_packet_logtype;
 103
 104 #define PMD_LOG(level, fmt, args...) \
 105         rte_log(RTE_LOG_ ## level, af_packet_logtype, \
 106                 "%s(): " fmt "\n", __func__, ##args)
 107
 108 #define PMD_LOG_ERRNO(level, fmt, args...) \
 109         rte_log(RTE_LOG_ ## level, af_packet_logtype, \
 110                 "%s(): " fmt ":%s\n", __func__, ##args, strerror(errno))
 111
 112 static uint16_t
 113 eth_af_packet_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 114 {
 115         unsigned i;
 116         struct tpacket2_hdr *ppd;
 117         struct rte_mbuf *mbuf;
 118         uint8_t *pbuf;
 119         struct pkt_rx_queue *pkt_q = queue;
 120         uint16_t num_rx = 0;
 121         unsigned long num_rx_bytes = 0;
 122         unsigned int framecount, framenum;
 123
 124         if (unlikely(nb_pkts == 0))
 125                 return 0;
 126
 127         /*
 128          * Reads the given number of packets from the AF_PACKET socket one by
 129          * one and copies the packet data into a newly allocated mbuf.
 130          */
 131         framecount = pkt_q->framecount;
 132         framenum = pkt_q->framenum;
 133         for (i = 0; i < nb_pkts; i++) {
 134                 /* point at the next incoming frame */
 135                 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
 136                 if ((ppd->tp_status & TP_STATUS_USER) == 0)
 137                         break;
 138
 139                 /* allocate the next mbuf */
 140                 mbuf = rte_pktmbuf_alloc(pkt_q->mb_pool);
 141                 if (unlikely(mbuf == NULL))
 142                         break;
 143
 144                 /* packet will fit in the mbuf, go ahead and receive it */
 145                 rte_pktmbuf_pkt_len(mbuf) = rte_pktmbuf_data_len(mbuf) = ppd->tp_snaplen;
 146                 pbuf = (uint8_t *) ppd + ppd->tp_mac;
 147                 memcpy(rte_pktmbuf_mtod(mbuf, void *), pbuf, rte_pktmbuf_data_len(mbuf));
 148
 149                 /* check for vlan info */
 150                 if (ppd->tp_status & TP_STATUS_VLAN_VALID) {
 151                         mbuf->vlan_tci = ppd->tp_vlan_tci;
 152                         mbuf->ol_flags |= (PKT_RX_VLAN | PKT_RX_VLAN_STRIPPED);
 153                 }
 154
 155                 /* release incoming frame and advance ring buffer */
 156                 ppd->tp_status = TP_STATUS_KERNEL;
 157                 if (++framenum >= framecount)
 158                         framenum = 0;
 159                 mbuf->port = pkt_q->in_port;
 160
 161                 /* account for the receive frame */
 162                 bufs[i] = mbuf;
 163                 num_rx++;
 164                 num_rx_bytes += mbuf->pkt_len;
 165         }
 166         pkt_q->framenum = framenum;
 167         pkt_q->rx_pkts += num_rx;
 168         pkt_q->rx_bytes += num_rx_bytes;
 169         return num_rx;
 170 }
 171
 172 /*
 173  * Callback to handle sending packets through a real NIC.
 174  */
 175 static uint16_t
 176 eth_af_packet_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 177 {
 178         struct tpacket2_hdr *ppd;
 179         struct rte_mbuf *mbuf;
 180         uint8_t *pbuf;
 181         unsigned int framecount, framenum;
 182         struct pollfd pfd;
 183         struct pkt_tx_queue *pkt_q = queue;
 184         uint16_t num_tx = 0;
 185         unsigned long num_tx_bytes = 0;
 186         int i;
 187
 188         if (unlikely(nb_pkts == 0))
 189                 return 0;
 190
 191         memset(&pfd, 0, sizeof(pfd));
 192         pfd.fd = pkt_q->sockfd;
 193         pfd.events = POLLOUT;
 194         pfd.revents = 0;
 195
 196         framecount = pkt_q->framecount;
 197         framenum = pkt_q->framenum;
 198         ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
 199         for (i = 0; i < nb_pkts; i++) {
 200                 mbuf = *bufs++;
 201
 202                 /* drop oversized packets */
 203                 if (mbuf->pkt_len > pkt_q->frame_data_size) {
 204                         rte_pktmbuf_free(mbuf);
 205                         continue;
 206                 }
 207
 208                 /* insert vlan info if necessary */
 209                 if (mbuf->ol_flags & PKT_TX_VLAN_PKT) {
 210                         if (rte_vlan_insert(&mbuf)) {
 211                                 rte_pktmbuf_free(mbuf);
 212                                 continue;
 213                         }
 214                 }
 215
 216                 /* point at the next incoming frame */
 217                 if ((ppd->tp_status != TP_STATUS_AVAILABLE) &&
 218                     (poll(&pfd, 1, -1) < 0))
 219                         break;
 220
 221                 /* copy the tx frame data */
 222                 pbuf = (uint8_t *) ppd + TPACKET2_HDRLEN -
 223                         sizeof(struct sockaddr_ll);
 224
 225                 struct rte_mbuf *tmp_mbuf = mbuf;
 226                 while (tmp_mbuf) {
 227                         uint16_t data_len = rte_pktmbuf_data_len(tmp_mbuf);
 228                         memcpy(pbuf, rte_pktmbuf_mtod(tmp_mbuf, void*), data_len);
 229                         pbuf += data_len;
 230                         tmp_mbuf = tmp_mbuf->next;
 231                 }
 232
 233                 ppd->tp_len = mbuf->pkt_len;
 234                 ppd->tp_snaplen = mbuf->pkt_len;
 235
 236                 /* release incoming frame and advance ring buffer */
 237                 ppd->tp_status = TP_STATUS_SEND_REQUEST;
 238                 if (++framenum >= framecount)
 239                         framenum = 0;
 240                 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
 241
 242                 num_tx++;
 243                 num_tx_bytes += mbuf->pkt_len;
 244                 rte_pktmbuf_free(mbuf);
 245         }
 246
 247         /* kick-off transmits */
 248         if (sendto(pkt_q->sockfd, NULL, 0, MSG_DONTWAIT, NULL, 0) == -1) {
 249                 /* error sending -- no packets transmitted */
 250                 num_tx = 0;
 251                 num_tx_bytes = 0;
 252         }
 253
 254         pkt_q->framenum = framenum;
 255         pkt_q->tx_pkts += num_tx;
 256         pkt_q->err_pkts += i - num_tx;
 257         pkt_q->tx_bytes += num_tx_bytes;
 258         return i;
 259 }
 260
 261 static int
 262 eth_dev_start(struct rte_eth_dev *dev)
 263 {
 264         dev->data->dev_link.link_status = ETH_LINK_UP;
 265         return 0;
 266 }
 267
 268 /*
 269  * This function gets called when the current port gets stopped.
 270  */
 271 static void
 272 eth_dev_stop(struct rte_eth_dev *dev)
 273 {
 274         unsigned i;
 275         int sockfd;
 276         struct pmd_internals *internals = dev->data->dev_private;
 277
 278         for (i = 0; i < internals->nb_queues; i++) {
 279                 sockfd = internals->rx_queue[i].sockfd;
 280                 if (sockfd != -1)
 281                         close(sockfd);
 282
 283                 /* Prevent use after free in case tx fd == rx fd */
 284                 if (sockfd != internals->tx_queue[i].sockfd) {
 285                         sockfd = internals->tx_queue[i].sockfd;
 286                         if (sockfd != -1)
 287                                 close(sockfd);
 288                 }
 289
 290                 internals->rx_queue[i].sockfd = -1;
 291                 internals->tx_queue[i].sockfd = -1;
 292         }
 293
 294         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 295 }
 296
 297 static int
 298 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 299 {
 300         return 0;
 301 }
 302
 303 static void
 304 eth_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 305 {
 306         struct pmd_internals *internals = dev->data->dev_private;
 307
 308         dev_info->if_index = internals->if_index;
 309         dev_info->max_mac_addrs = 1;
 310         dev_info->max_rx_pktlen = (uint32_t)ETH_FRAME_LEN;
 311         dev_info->max_rx_queues = (uint16_t)internals->nb_queues;
 312         dev_info->max_tx_queues = (uint16_t)internals->nb_queues;
 313         dev_info->min_rx_bufsize = 0;
 314 }
 315
 316 static int
 317 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *igb_stats)
 318 {
 319         unsigned i, imax;
 320         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 321         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 322         const struct pmd_internals *internal = dev->data->dev_private;
 323
 324         imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
 325                 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
 326         for (i = 0; i < imax; i++) {
 327                 igb_stats->q_ipackets[i] = internal->rx_queue[i].rx_pkts;
 328                 igb_stats->q_ibytes[i] = internal->rx_queue[i].rx_bytes;
 329                 rx_total += igb_stats->q_ipackets[i];
 330                 rx_bytes_total += igb_stats->q_ibytes[i];
 331         }
 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_opackets[i] = internal->tx_queue[i].tx_pkts;
 337                 igb_stats->q_obytes[i] = internal->tx_queue[i].tx_bytes;
 338                 tx_total += igb_stats->q_opackets[i];
 339                 tx_err_total += internal->tx_queue[i].err_pkts;
 340                 tx_bytes_total += igb_stats->q_obytes[i];
 341         }
 342
 343         igb_stats->ipackets = rx_total;
 344         igb_stats->ibytes = rx_bytes_total;
 345         igb_stats->opackets = tx_total;
 346         igb_stats->oerrors = tx_err_total;
 347         igb_stats->obytes = tx_bytes_total;
 348         return 0;
 349 }
 350
 351 static void
 352 eth_stats_reset(struct rte_eth_dev *dev)
 353 {
 354         unsigned i;
 355         struct pmd_internals *internal = dev->data->dev_private;
 356
 357         for (i = 0; i < internal->nb_queues; i++) {
 358                 internal->rx_queue[i].rx_pkts = 0;
 359                 internal->rx_queue[i].rx_bytes = 0;
 360         }
 361
 362         for (i = 0; i < internal->nb_queues; i++) {
 363                 internal->tx_queue[i].tx_pkts = 0;
 364                 internal->tx_queue[i].err_pkts = 0;
 365                 internal->tx_queue[i].tx_bytes = 0;
 366         }
 367 }
 368
 369 static void
 370 eth_dev_close(struct rte_eth_dev *dev __rte_unused)
 371 {
 372 }
 373
 374 static void
 375 eth_queue_release(void *q __rte_unused)
 376 {
 377 }
 378
 379 static int
 380 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 381                 int wait_to_complete __rte_unused)
 382 {
 383         return 0;
 384 }
 385
 386 static int
 387 eth_rx_queue_setup(struct rte_eth_dev *dev,
 388                    uint16_t rx_queue_id,
 389                    uint16_t nb_rx_desc __rte_unused,
 390                    unsigned int socket_id __rte_unused,
 391                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 392                    struct rte_mempool *mb_pool)
 393 {
 394         struct pmd_internals *internals = dev->data->dev_private;
 395         struct pkt_rx_queue *pkt_q = &internals->rx_queue[rx_queue_id];
 396         unsigned int buf_size, data_size;
 397
 398         pkt_q->mb_pool = mb_pool;
 399
 400         /* Now get the space available for data in the mbuf */
 401         buf_size = rte_pktmbuf_data_room_size(pkt_q->mb_pool) -
 402                 RTE_PKTMBUF_HEADROOM;
 403         data_size = internals->req.tp_frame_size;
 404         data_size -= TPACKET2_HDRLEN - sizeof(struct sockaddr_ll);
 405
 406         if (data_size > buf_size) {
 407                 PMD_LOG(ERR,
 408                         "%s: %d bytes will not fit in mbuf (%d bytes)",
 409                         dev->device->name, data_size, buf_size);
 410                 return -ENOMEM;
 411         }
 412
 413         dev->data->rx_queues[rx_queue_id] = pkt_q;
 414         pkt_q->in_port = dev->data->port_id;
 415
 416         return 0;
 417 }
 418
 419 static int
 420 eth_tx_queue_setup(struct rte_eth_dev *dev,
 421                    uint16_t tx_queue_id,
 422                    uint16_t nb_tx_desc __rte_unused,
 423                    unsigned int socket_id __rte_unused,
 424                    const struct rte_eth_txconf *tx_conf __rte_unused)
 425 {
 426
 427         struct pmd_internals *internals = dev->data->dev_private;
 428
 429         dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
 430         return 0;
 431 }
 432
 433 static int
 434 eth_dev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 435 {
 436         struct pmd_internals *internals = dev->data->dev_private;
 437         struct ifreq ifr = { .ifr_mtu = mtu };
 438         int ret;
 439         int s;
 440         unsigned int data_size = internals->req.tp_frame_size -
 441                                  TPACKET2_HDRLEN;
 442
 443         if (mtu > data_size)
 444                 return -EINVAL;
 445
 446         s = socket(PF_INET, SOCK_DGRAM, 0);
 447         if (s < 0)
 448                 return -EINVAL;
 449
 450         strlcpy(ifr.ifr_name, internals->if_name, IFNAMSIZ);
 451         ret = ioctl(s, SIOCSIFMTU, &ifr);
 452         close(s);
 453
 454         if (ret < 0)
 455                 return -EINVAL;
 456
 457         return 0;
 458 }
 459
 460 static void
 461 eth_dev_change_flags(char *if_name, uint32_t flags, uint32_t mask)
 462 {
 463         struct ifreq ifr;
 464         int s;
 465
 466         s = socket(PF_INET, SOCK_DGRAM, 0);
 467         if (s < 0)
 468                 return;
 469
 470         strlcpy(ifr.ifr_name, if_name, IFNAMSIZ);
 471         if (ioctl(s, SIOCGIFFLAGS, &ifr) < 0)
 472                 goto out;
 473         ifr.ifr_flags &= mask;
 474         ifr.ifr_flags |= flags;
 475         if (ioctl(s, SIOCSIFFLAGS, &ifr) < 0)
 476                 goto out;
 477 out:
 478         close(s);
 479 }
 480
 481 static void
 482 eth_dev_promiscuous_enable(struct rte_eth_dev *dev)
 483 {
 484         struct pmd_internals *internals = dev->data->dev_private;
 485
 486         eth_dev_change_flags(internals->if_name, IFF_PROMISC, ~0);
 487 }
 488
 489 static void
 490 eth_dev_promiscuous_disable(struct rte_eth_dev *dev)
 491 {
 492         struct pmd_internals *internals = dev->data->dev_private;
 493
 494         eth_dev_change_flags(internals->if_name, 0, ~IFF_PROMISC);
 495 }
 496
 497 static const struct eth_dev_ops ops = {
 498         .dev_start = eth_dev_start,
 499         .dev_stop = eth_dev_stop,
 500         .dev_close = eth_dev_close,
 501         .dev_configure = eth_dev_configure,
 502         .dev_infos_get = eth_dev_info,
 503         .mtu_set = eth_dev_mtu_set,
 504         .promiscuous_enable = eth_dev_promiscuous_enable,
 505         .promiscuous_disable = eth_dev_promiscuous_disable,
 506         .rx_queue_setup = eth_rx_queue_setup,
 507         .tx_queue_setup = eth_tx_queue_setup,
 508         .rx_queue_release = eth_queue_release,
 509         .tx_queue_release = eth_queue_release,
 510         .link_update = eth_link_update,
 511         .stats_get = eth_stats_get,
 512         .stats_reset = eth_stats_reset,
 513 };
 514
 515 /*
 516  * Opens an AF_PACKET socket
 517  */
 518 static int
 519 open_packet_iface(const char *key __rte_unused,
 520                   const char *value __rte_unused,
 521                   void *extra_args)
 522 {
 523         int *sockfd = extra_args;
 524
 525         /* Open an AF_PACKET socket... */
 526         *sockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
 527         if (*sockfd == -1) {
 528                 PMD_LOG(ERR, "Could not open AF_PACKET socket");
 529                 return -1;
 530         }
 531
 532         return 0;
 533 }
 534
 535 static struct rte_vdev_driver pmd_af_packet_drv;
 536
 537 static int
 538 rte_pmd_init_internals(struct rte_vdev_device *dev,
 539                        const int sockfd,
 540                        const unsigned nb_queues,
 541                        unsigned int blocksize,
 542                        unsigned int blockcnt,
 543                        unsigned int framesize,
 544                        unsigned int framecnt,
 545                        unsigned int qdisc_bypass,
 546                        struct pmd_internals **internals,
 547                        struct rte_eth_dev **eth_dev,
 548                        struct rte_kvargs *kvlist)
 549 {
 550         const char *name = rte_vdev_device_name(dev);
 551         const unsigned int numa_node = dev->device.numa_node;
 552         struct rte_eth_dev_data *data = NULL;
 553         struct rte_kvargs_pair *pair = NULL;
 554         struct ifreq ifr;
 555         size_t ifnamelen;
 556         unsigned k_idx;
 557         struct sockaddr_ll sockaddr;
 558         struct tpacket_req *req;
 559         struct pkt_rx_queue *rx_queue;
 560         struct pkt_tx_queue *tx_queue;
 561         int rc, tpver, discard;
 562         int qsockfd = -1;
 563         unsigned int i, q, rdsize;
 564 #if defined(PACKET_FANOUT)
 565         int fanout_arg;
 566 #endif
 567
 568         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 569                 pair = &kvlist->pairs[k_idx];
 570                 if (strstr(pair->key, ETH_AF_PACKET_IFACE_ARG) != NULL)
 571                         break;
 572         }
 573         if (pair == NULL) {
 574                 PMD_LOG(ERR,
 575                         "%s: no interface specified for AF_PACKET ethdev",
 576                         name);
 577                 return -1;
 578         }
 579
 580         PMD_LOG(INFO,
 581                 "%s: creating AF_PACKET-backed ethdev on numa socket %u",
 582                 name, numa_node);
 583
 584         *internals = rte_zmalloc_socket(name, sizeof(**internals),
 585                                         0, numa_node);
 586         if (*internals == NULL)
 587                 return -1;
 588
 589         for (q = 0; q < nb_queues; q++) {
 590                 (*internals)->rx_queue[q].map = MAP_FAILED;
 591                 (*internals)->tx_queue[q].map = MAP_FAILED;
 592         }
 593
 594         req = &((*internals)->req);
 595
 596         req->tp_block_size = blocksize;
 597         req->tp_block_nr = blockcnt;
 598         req->tp_frame_size = framesize;
 599         req->tp_frame_nr = framecnt;
 600
 601         ifnamelen = strlen(pair->value);
 602         if (ifnamelen < sizeof(ifr.ifr_name)) {
 603                 memcpy(ifr.ifr_name, pair->value, ifnamelen);
 604                 ifr.ifr_name[ifnamelen] = '\0';
 605         } else {
 606                 PMD_LOG(ERR,
 607                         "%s: I/F name too long (%s)",
 608                         name, pair->value);
 609                 return -1;
 610         }
 611         if (ioctl(sockfd, SIOCGIFINDEX, &ifr) == -1) {
 612                 PMD_LOG_ERRNO(ERR, "%s: ioctl failed (SIOCGIFINDEX)", name);
 613                 return -1;
 614         }
 615         (*internals)->if_name = strdup(pair->value);
 616         if ((*internals)->if_name == NULL)
 617                 return -1;
 618         (*internals)->if_index = ifr.ifr_ifindex;
 619
 620         if (ioctl(sockfd, SIOCGIFHWADDR, &ifr) == -1) {
 621                 PMD_LOG_ERRNO(ERR, "%s: ioctl failed (SIOCGIFHWADDR)", name);
 622                 return -1;
 623         }
 624         memcpy(&(*internals)->eth_addr, ifr.ifr_hwaddr.sa_data, ETH_ALEN);
 625
 626         memset(&sockaddr, 0, sizeof(sockaddr));
 627         sockaddr.sll_family = AF_PACKET;
 628         sockaddr.sll_protocol = htons(ETH_P_ALL);
 629         sockaddr.sll_ifindex = (*internals)->if_index;
 630
 631 #if defined(PACKET_FANOUT)
 632         fanout_arg = (getpid() ^ (*internals)->if_index) & 0xffff;
 633         fanout_arg |= (PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_DEFRAG) << 16;
 634 #if defined(PACKET_FANOUT_FLAG_ROLLOVER)
 635         fanout_arg |= PACKET_FANOUT_FLAG_ROLLOVER << 16;
 636 #endif
 637 #endif
 638
 639         for (q = 0; q < nb_queues; q++) {
 640                 /* Open an AF_PACKET socket for this queue... */
 641                 qsockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
 642                 if (qsockfd == -1) {
 643                         PMD_LOG_ERRNO(ERR,
 644                                 "%s: could not open AF_PACKET socket",
 645                                 name);
 646                         return -1;
 647                 }
 648
 649                 tpver = TPACKET_V2;
 650                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_VERSION,
 651                                 &tpver, sizeof(tpver));
 652                 if (rc == -1) {
 653                         PMD_LOG_ERRNO(ERR,
 654                                 "%s: could not set PACKET_VERSION on AF_PACKET socket for %s",
 655                                 name, pair->value);
 656                         goto error;
 657                 }
 658
 659                 discard = 1;
 660                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_LOSS,
 661                                 &discard, sizeof(discard));
 662                 if (rc == -1) {
 663                         PMD_LOG_ERRNO(ERR,
 664                                 "%s: could not set PACKET_LOSS on AF_PACKET socket for %s",
 665                                 name, pair->value);
 666                         goto error;
 667                 }
 668
 669 #if defined(PACKET_QDISC_BYPASS)
 670                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_QDISC_BYPASS,
 671                                 &qdisc_bypass, sizeof(qdisc_bypass));
 672                 if (rc == -1) {
 673                         PMD_LOG_ERRNO(ERR,
 674                                 "%s: could not set PACKET_QDISC_BYPASS on AF_PACKET socket for %s",
 675                                 name, pair->value);
 676                         goto error;
 677                 }
 678 #else
 679                 RTE_SET_USED(qdisc_bypass);
 680 #endif
 681
 682                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_RX_RING, req, sizeof(*req));
 683                 if (rc == -1) {
 684                         PMD_LOG_ERRNO(ERR,
 685                                 "%s: could not set PACKET_RX_RING on AF_PACKET socket for %s",
 686                                 name, pair->value);
 687                         goto error;
 688                 }
 689
 690                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_TX_RING, req, sizeof(*req));
 691                 if (rc == -1) {
 692                         PMD_LOG_ERRNO(ERR,
 693                                 "%s: could not set PACKET_TX_RING on AF_PACKET "
 694                                 "socket for %s", name, pair->value);
 695                         goto error;
 696                 }
 697
 698                 rx_queue = &((*internals)->rx_queue[q]);
 699                 rx_queue->framecount = req->tp_frame_nr;
 700
 701                 rx_queue->map = mmap(NULL, 2 * req->tp_block_size * req->tp_block_nr,
 702                                     PROT_READ | PROT_WRITE, MAP_SHARED | MAP_LOCKED,
 703                                     qsockfd, 0);
 704                 if (rx_queue->map == MAP_FAILED) {
 705                         PMD_LOG_ERRNO(ERR,
 706                                 "%s: call to mmap failed on AF_PACKET socket for %s",
 707                                 name, pair->value);
 708                         goto error;
 709                 }
 710
 711                 /* rdsize is same for both Tx and Rx */
 712                 rdsize = req->tp_frame_nr * sizeof(*(rx_queue->rd));
 713
 714                 rx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
 715                 if (rx_queue->rd == NULL)
 716                         goto error;
 717                 for (i = 0; i < req->tp_frame_nr; ++i) {
 718                         rx_queue->rd[i].iov_base = rx_queue->map + (i * framesize);
 719                         rx_queue->rd[i].iov_len = req->tp_frame_size;
 720                 }
 721                 rx_queue->sockfd = qsockfd;
 722
 723                 tx_queue = &((*internals)->tx_queue[q]);
 724                 tx_queue->framecount = req->tp_frame_nr;
 725                 tx_queue->frame_data_size = req->tp_frame_size;
 726                 tx_queue->frame_data_size -= TPACKET2_HDRLEN -
 727                         sizeof(struct sockaddr_ll);
 728
 729                 tx_queue->map = rx_queue->map + req->tp_block_size * req->tp_block_nr;
 730
 731                 tx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
 732                 if (tx_queue->rd == NULL)
 733                         goto error;
 734                 for (i = 0; i < req->tp_frame_nr; ++i) {
 735                         tx_queue->rd[i].iov_base = tx_queue->map + (i * framesize);
 736                         tx_queue->rd[i].iov_len = req->tp_frame_size;
 737                 }
 738                 tx_queue->sockfd = qsockfd;
 739
 740                 rc = bind(qsockfd, (const struct sockaddr*)&sockaddr, sizeof(sockaddr));
 741                 if (rc == -1) {
 742                         PMD_LOG_ERRNO(ERR,
 743                                 "%s: could not bind AF_PACKET socket to %s",
 744                                 name, pair->value);
 745                         goto error;
 746                 }
 747
 748 #if defined(PACKET_FANOUT)
 749                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_FANOUT,
 750                                 &fanout_arg, sizeof(fanout_arg));
 751                 if (rc == -1) {
 752                         PMD_LOG_ERRNO(ERR,
 753                                 "%s: could not set PACKET_FANOUT on AF_PACKET socket for %s",
 754                                 name, pair->value);
 755                         goto error;
 756                 }
 757 #endif
 758         }
 759
 760         /* reserve an ethdev entry */
 761         *eth_dev = rte_eth_vdev_allocate(dev, 0);
 762         if (*eth_dev == NULL)
 763                 goto error;
 764
 765         /*
 766          * now put it all together
 767          * - store queue data in internals,
 768          * - store numa_node in eth_dev
 769          * - point eth_dev_data to internals
 770          * - and point eth_dev structure to new eth_dev_data structure
 771          */
 772
 773         (*internals)->nb_queues = nb_queues;
 774
 775         data = (*eth_dev)->data;
 776         data->dev_private = *internals;
 777         data->nb_rx_queues = (uint16_t)nb_queues;
 778         data->nb_tx_queues = (uint16_t)nb_queues;
 779         data->dev_link = pmd_link;
 780         data->mac_addrs = &(*internals)->eth_addr;
 781
 782         (*eth_dev)->dev_ops = &ops;
 783
 784         return 0;
 785
 786 error:
 787         if (qsockfd != -1)
 788                 close(qsockfd);
 789         for (q = 0; q < nb_queues; q++) {
 790                 munmap((*internals)->rx_queue[q].map,
 791                        2 * req->tp_block_size * req->tp_block_nr);
 792
 793                 rte_free((*internals)->rx_queue[q].rd);
 794                 rte_free((*internals)->tx_queue[q].rd);
 795                 if (((*internals)->rx_queue[q].sockfd != 0) &&
 796                         ((*internals)->rx_queue[q].sockfd != qsockfd))
 797                         close((*internals)->rx_queue[q].sockfd);
 798         }
 799         free((*internals)->if_name);
 800         rte_free(*internals);
 801         return -1;
 802 }
 803
 804 static int
 805 rte_eth_from_packet(struct rte_vdev_device *dev,
 806                     int const *sockfd,
 807                     struct rte_kvargs *kvlist)
 808 {
 809         const char *name = rte_vdev_device_name(dev);
 810         struct pmd_internals *internals = NULL;
 811         struct rte_eth_dev *eth_dev = NULL;
 812         struct rte_kvargs_pair *pair = NULL;
 813         unsigned k_idx;
 814         unsigned int blockcount;
 815         unsigned int blocksize;
 816         unsigned int framesize = DFLT_FRAME_SIZE;
 817         unsigned int framecount = DFLT_FRAME_COUNT;
 818         unsigned int qpairs = 1;
 819         unsigned int qdisc_bypass = 1;
 820
 821         /* do some parameter checking */
 822         if (*sockfd < 0)
 823                 return -1;
 824
 825         blocksize = getpagesize();
 826
 827         /*
 828          * Walk arguments for configurable settings
 829          */
 830         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 831                 pair = &kvlist->pairs[k_idx];
 832                 if (strstr(pair->key, ETH_AF_PACKET_NUM_Q_ARG) != NULL) {
 833                         qpairs = atoi(pair->value);
 834                         if (qpairs < 1 ||
 835                             qpairs > RTE_PMD_AF_PACKET_MAX_RINGS) {
 836                                 PMD_LOG(ERR,
 837                                         "%s: invalid qpairs value",
 838                                         name);
 839                                 return -1;
 840                         }
 841                         continue;
 842                 }
 843                 if (strstr(pair->key, ETH_AF_PACKET_BLOCKSIZE_ARG) != NULL) {
 844                         blocksize = atoi(pair->value);
 845                         if (!blocksize) {
 846                                 PMD_LOG(ERR,
 847                                         "%s: invalid blocksize value",
 848                                         name);
 849                                 return -1;
 850                         }
 851                         continue;
 852                 }
 853                 if (strstr(pair->key, ETH_AF_PACKET_FRAMESIZE_ARG) != NULL) {
 854                         framesize = atoi(pair->value);
 855                         if (!framesize) {
 856                                 PMD_LOG(ERR,
 857                                         "%s: invalid framesize value",
 858                                         name);
 859                                 return -1;
 860                         }
 861                         continue;
 862                 }
 863                 if (strstr(pair->key, ETH_AF_PACKET_FRAMECOUNT_ARG) != NULL) {
 864                         framecount = atoi(pair->value);
 865                         if (!framecount) {
 866                                 PMD_LOG(ERR,
 867                                         "%s: invalid framecount value",
 868                                         name);
 869                                 return -1;
 870                         }
 871                         continue;
 872                 }
 873                 if (strstr(pair->key, ETH_AF_PACKET_QDISC_BYPASS_ARG) != NULL) {
 874                         qdisc_bypass = atoi(pair->value);
 875                         if (qdisc_bypass > 1) {
 876                                 PMD_LOG(ERR,
 877                                         "%s: invalid bypass value",
 878                                         name);
 879                                 return -1;
 880                         }
 881                         continue;
 882                 }
 883         }
 884
 885         if (framesize > blocksize) {
 886                 PMD_LOG(ERR,
 887                         "%s: AF_PACKET MMAP frame size exceeds block size!",
 888                         name);
 889                 return -1;
 890         }
 891
 892         blockcount = framecount / (blocksize / framesize);
 893         if (!blockcount) {
 894                 PMD_LOG(ERR,
 895                         "%s: invalid AF_PACKET MMAP parameters", name);
 896                 return -1;
 897         }
 898
 899         PMD_LOG(INFO, "%s: AF_PACKET MMAP parameters:", name);
 900         PMD_LOG(INFO, "%s:\tblock size %d", name, blocksize);
 901         PMD_LOG(INFO, "%s:\tblock count %d", name, blockcount);
 902         PMD_LOG(INFO, "%s:\tframe size %d", name, framesize);
 903         PMD_LOG(INFO, "%s:\tframe count %d", name, framecount);
 904
 905         if (rte_pmd_init_internals(dev, *sockfd, qpairs,
 906                                    blocksize, blockcount,
 907                                    framesize, framecount,
 908                                    qdisc_bypass,
 909                                    &internals, &eth_dev,
 910                                    kvlist) < 0)
 911                 return -1;
 912
 913         eth_dev->rx_pkt_burst = eth_af_packet_rx;
 914         eth_dev->tx_pkt_burst = eth_af_packet_tx;
 915
 916         rte_eth_dev_probing_finish(eth_dev);
 917         return 0;
 918 }
 919
 920 static int
 921 rte_pmd_af_packet_probe(struct rte_vdev_device *dev)
 922 {
 923         int ret = 0;
 924         struct rte_kvargs *kvlist;
 925         int sockfd = -1;
 926         struct rte_eth_dev *eth_dev;
 927         const char *name = rte_vdev_device_name(dev);
 928
 929         PMD_LOG(INFO, "Initializing pmd_af_packet for %s", name);
 930
 931         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
 932                 eth_dev = rte_eth_dev_attach_secondary(name);
 933                 if (!eth_dev) {
 934                         PMD_LOG(ERR, "Failed to probe %s", name);
 935                         return -1;
 936                 }
 937                 /* TODO: request info from primary to set up Rx and Tx */
 938                 eth_dev->dev_ops = &ops;
 939                 eth_dev->device = &dev->device;
 940                 rte_eth_dev_probing_finish(eth_dev);
 941                 return 0;
 942         }
 943
 944         kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_arguments);
 945         if (kvlist == NULL) {
 946                 ret = -1;
 947                 goto exit;
 948         }
 949
 950         /*
 951          * If iface argument is passed we open the NICs and use them for
 952          * reading / writing
 953          */
 954         if (rte_kvargs_count(kvlist, ETH_AF_PACKET_IFACE_ARG) == 1) {
 955
 956                 ret = rte_kvargs_process(kvlist, ETH_AF_PACKET_IFACE_ARG,
 957                                          &open_packet_iface, &sockfd);
 958                 if (ret < 0)
 959                         goto exit;
 960         }
 961
 962         if (dev->device.numa_node == SOCKET_ID_ANY)
 963                 dev->device.numa_node = rte_socket_id();
 964
 965         ret = rte_eth_from_packet(dev, &sockfd, kvlist);
 966         close(sockfd); /* no longer needed */
 967
 968 exit:
 969         rte_kvargs_free(kvlist);
 970         return ret;
 971 }
 972
 973 static int
 974 rte_pmd_af_packet_remove(struct rte_vdev_device *dev)
 975 {
 976         struct rte_eth_dev *eth_dev = NULL;
 977         struct pmd_internals *internals;
 978         unsigned q;
 979
 980         PMD_LOG(INFO, "Closing AF_PACKET ethdev on numa socket %u",
 981                 rte_socket_id());
 982
 983         if (dev == NULL)
 984                 return -1;
 985
 986         /* find the ethdev entry */
 987         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
 988         if (eth_dev == NULL)
 989                 return -1;
 990
 991         /* mac_addrs must not be freed alone because part of dev_private */
 992         eth_dev->data->mac_addrs = NULL;
 993
 994         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 995                 return rte_eth_dev_release_port(eth_dev);
 996
 997         internals = eth_dev->data->dev_private;
 998         for (q = 0; q < internals->nb_queues; q++) {
 999                 rte_free(internals->rx_queue[q].rd);
1000                 rte_free(internals->tx_queue[q].rd);
1001         }
1002         free(internals->if_name);
1003
1004         rte_eth_dev_release_port(eth_dev);
1005
1006         return 0;
1007 }
1008
1009 static struct rte_vdev_driver pmd_af_packet_drv = {
1010         .probe = rte_pmd_af_packet_probe,
1011         .remove = rte_pmd_af_packet_remove,
1012 };
1013
1014 RTE_PMD_REGISTER_VDEV(net_af_packet, pmd_af_packet_drv);
1015 RTE_PMD_REGISTER_ALIAS(net_af_packet, eth_af_packet);
1016 RTE_PMD_REGISTER_PARAM_STRING(net_af_packet,
1017         "iface=<string> "
1018         "qpairs=<int> "
1019         "blocksz=<int> "
1020         "framesz=<int> "
1021         "framecnt=<int> "
1022         "qdisc_bypass=<0|1>");
1023
1024 RTE_INIT(af_packet_init_log)
1025 {
1026         af_packet_logtype = rte_log_register("pmd.net.packet");
1027         if (af_packet_logtype >= 0)
1028                 rte_log_set_level(af_packet_logtype, RTE_LOG_NOTICE);
1029 }