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 rx_bytes;
  55 };
  56
  57 struct pkt_tx_queue {
  58         int sockfd;
  59         unsigned int frame_data_size;
  60
  61         struct iovec *rd;
  62         uint8_t *map;
  63         unsigned int framecount;
  64         unsigned int framenum;
  65
  66         volatile unsigned long tx_pkts;
  67         volatile unsigned long err_pkts;
  68         volatile unsigned long tx_bytes;
  69 };
  70
  71 struct pmd_internals {
  72         unsigned nb_queues;
  73
  74         int if_index;
  75         char *if_name;
  76         struct rte_ether_addr eth_addr;
  77
  78         struct tpacket_req req;
  79
  80         struct pkt_rx_queue rx_queue[RTE_PMD_AF_PACKET_MAX_RINGS];
  81         struct pkt_tx_queue tx_queue[RTE_PMD_AF_PACKET_MAX_RINGS];
  82 };
  83
  84 static const char *valid_arguments[] = {
  85         ETH_AF_PACKET_IFACE_ARG,
  86         ETH_AF_PACKET_NUM_Q_ARG,
  87         ETH_AF_PACKET_BLOCKSIZE_ARG,
  88         ETH_AF_PACKET_FRAMESIZE_ARG,
  89         ETH_AF_PACKET_FRAMECOUNT_ARG,
  90         ETH_AF_PACKET_QDISC_BYPASS_ARG,
  91         NULL
  92 };
  93
  94 static struct rte_eth_link pmd_link = {
  95         .link_speed = ETH_SPEED_NUM_10G,
  96         .link_duplex = ETH_LINK_FULL_DUPLEX,
  97         .link_status = ETH_LINK_DOWN,
  98         .link_autoneg = ETH_LINK_FIXED,
  99 };
 100
 101 static int af_packet_logtype;
 102
 103 #define PMD_LOG(level, fmt, args...) \
 104         rte_log(RTE_LOG_ ## level, af_packet_logtype, \
 105                 "%s(): " fmt "\n", __func__, ##args)
 106
 107 #define PMD_LOG_ERRNO(level, fmt, args...) \
 108         rte_log(RTE_LOG_ ## level, af_packet_logtype, \
 109                 "%s(): " fmt ":%s\n", __func__, ##args, strerror(errno))
 110
 111 static uint16_t
 112 eth_af_packet_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 113 {
 114         unsigned i;
 115         struct tpacket2_hdr *ppd;
 116         struct rte_mbuf *mbuf;
 117         uint8_t *pbuf;
 118         struct pkt_rx_queue *pkt_q = queue;
 119         uint16_t num_rx = 0;
 120         unsigned long num_rx_bytes = 0;
 121         unsigned int framecount, framenum;
 122
 123         if (unlikely(nb_pkts == 0))
 124                 return 0;
 125
 126         /*
 127          * Reads the given number of packets from the AF_PACKET socket one by
 128          * one and copies the packet data into a newly allocated mbuf.
 129          */
 130         framecount = pkt_q->framecount;
 131         framenum = pkt_q->framenum;
 132         for (i = 0; i < nb_pkts; i++) {
 133                 /* point at the next incoming frame */
 134                 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
 135                 if ((ppd->tp_status & TP_STATUS_USER) == 0)
 136                         break;
 137
 138                 /* allocate the next mbuf */
 139                 mbuf = rte_pktmbuf_alloc(pkt_q->mb_pool);
 140                 if (unlikely(mbuf == NULL))
 141                         break;
 142
 143                 /* packet will fit in the mbuf, go ahead and receive it */
 144                 rte_pktmbuf_pkt_len(mbuf) = rte_pktmbuf_data_len(mbuf) = ppd->tp_snaplen;
 145                 pbuf = (uint8_t *) ppd + ppd->tp_mac;
 146                 memcpy(rte_pktmbuf_mtod(mbuf, void *), pbuf, rte_pktmbuf_data_len(mbuf));
 147
 148                 /* check for vlan info */
 149                 if (ppd->tp_status & TP_STATUS_VLAN_VALID) {
 150                         mbuf->vlan_tci = ppd->tp_vlan_tci;
 151                         mbuf->ol_flags |= (PKT_RX_VLAN | PKT_RX_VLAN_STRIPPED);
 152                 }
 153
 154                 /* release incoming frame and advance ring buffer */
 155                 ppd->tp_status = TP_STATUS_KERNEL;
 156                 if (++framenum >= framecount)
 157                         framenum = 0;
 158                 mbuf->port = pkt_q->in_port;
 159
 160                 /* account for the receive frame */
 161                 bufs[i] = mbuf;
 162                 num_rx++;
 163                 num_rx_bytes += mbuf->pkt_len;
 164         }
 165         pkt_q->framenum = framenum;
 166         pkt_q->rx_pkts += num_rx;
 167         pkt_q->rx_bytes += num_rx_bytes;
 168         return num_rx;
 169 }
 170
 171 /*
 172  * Callback to handle sending packets through a real NIC.
 173  */
 174 static uint16_t
 175 eth_af_packet_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 176 {
 177         struct tpacket2_hdr *ppd;
 178         struct rte_mbuf *mbuf;
 179         uint8_t *pbuf;
 180         unsigned int framecount, framenum;
 181         struct pollfd pfd;
 182         struct pkt_tx_queue *pkt_q = queue;
 183         uint16_t num_tx = 0;
 184         unsigned long num_tx_bytes = 0;
 185         int i;
 186
 187         if (unlikely(nb_pkts == 0))
 188                 return 0;
 189
 190         memset(&pfd, 0, sizeof(pfd));
 191         pfd.fd = pkt_q->sockfd;
 192         pfd.events = POLLOUT;
 193         pfd.revents = 0;
 194
 195         framecount = pkt_q->framecount;
 196         framenum = pkt_q->framenum;
 197         ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
 198         for (i = 0; i < nb_pkts; i++) {
 199                 mbuf = *bufs++;
 200
 201                 /* drop oversized packets */
 202                 if (mbuf->pkt_len > pkt_q->frame_data_size) {
 203                         rte_pktmbuf_free(mbuf);
 204                         continue;
 205                 }
 206
 207                 /* insert vlan info if necessary */
 208                 if (mbuf->ol_flags & PKT_TX_VLAN_PKT) {
 209                         if (rte_vlan_insert(&mbuf)) {
 210                                 rte_pktmbuf_free(mbuf);
 211                                 continue;
 212                         }
 213                 }
 214
 215                 /* point at the next incoming frame */
 216                 if ((ppd->tp_status != TP_STATUS_AVAILABLE) &&
 217                     (poll(&pfd, 1, -1) < 0))
 218                         break;
 219
 220                 /* copy the tx frame data */
 221                 pbuf = (uint8_t *) ppd + TPACKET2_HDRLEN -
 222                         sizeof(struct sockaddr_ll);
 223
 224                 struct rte_mbuf *tmp_mbuf = mbuf;
 225                 while (tmp_mbuf) {
 226                         uint16_t data_len = rte_pktmbuf_data_len(tmp_mbuf);
 227                         memcpy(pbuf, rte_pktmbuf_mtod(tmp_mbuf, void*), data_len);
 228                         pbuf += data_len;
 229                         tmp_mbuf = tmp_mbuf->next;
 230                 }
 231
 232                 ppd->tp_len = mbuf->pkt_len;
 233                 ppd->tp_snaplen = mbuf->pkt_len;
 234
 235                 /* release incoming frame and advance ring buffer */
 236                 ppd->tp_status = TP_STATUS_SEND_REQUEST;
 237                 if (++framenum >= framecount)
 238                         framenum = 0;
 239                 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
 240
 241                 num_tx++;
 242                 num_tx_bytes += mbuf->pkt_len;
 243                 rte_pktmbuf_free(mbuf);
 244         }
 245
 246         /* kick-off transmits */
 247         if (sendto(pkt_q->sockfd, NULL, 0, MSG_DONTWAIT, NULL, 0) == -1) {
 248                 /* error sending -- no packets transmitted */
 249                 num_tx = 0;
 250                 num_tx_bytes = 0;
 251         }
 252
 253         pkt_q->framenum = framenum;
 254         pkt_q->tx_pkts += num_tx;
 255         pkt_q->err_pkts += i - num_tx;
 256         pkt_q->tx_bytes += num_tx_bytes;
 257         return i;
 258 }
 259
 260 static int
 261 eth_dev_start(struct rte_eth_dev *dev)
 262 {
 263         dev->data->dev_link.link_status = ETH_LINK_UP;
 264         return 0;
 265 }
 266
 267 /*
 268  * This function gets called when the current port gets stopped.
 269  */
 270 static void
 271 eth_dev_stop(struct rte_eth_dev *dev)
 272 {
 273         unsigned i;
 274         int sockfd;
 275         struct pmd_internals *internals = dev->data->dev_private;
 276
 277         for (i = 0; i < internals->nb_queues; i++) {
 278                 sockfd = internals->rx_queue[i].sockfd;
 279                 if (sockfd != -1)
 280                         close(sockfd);
 281
 282                 /* Prevent use after free in case tx fd == rx fd */
 283                 if (sockfd != internals->tx_queue[i].sockfd) {
 284                         sockfd = internals->tx_queue[i].sockfd;
 285                         if (sockfd != -1)
 286                                 close(sockfd);
 287                 }
 288
 289                 internals->rx_queue[i].sockfd = -1;
 290                 internals->tx_queue[i].sockfd = -1;
 291         }
 292
 293         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 294 }
 295
 296 static int
 297 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 298 {
 299         return 0;
 300 }
 301
 302 static int
 303 eth_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 304 {
 305         struct pmd_internals *internals = dev->data->dev_private;
 306
 307         dev_info->if_index = internals->if_index;
 308         dev_info->max_mac_addrs = 1;
 309         dev_info->max_rx_pktlen = (uint32_t)ETH_FRAME_LEN;
 310         dev_info->max_rx_queues = (uint16_t)internals->nb_queues;
 311         dev_info->max_tx_queues = (uint16_t)internals->nb_queues;
 312         dev_info->min_rx_bufsize = 0;
 313
 314         return 0;
 315 }
 316
 317 static int
 318 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *igb_stats)
 319 {
 320         unsigned i, imax;
 321         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 322         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 323         const struct pmd_internals *internal = dev->data->dev_private;
 324
 325         imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
 326                 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
 327         for (i = 0; i < imax; i++) {
 328                 igb_stats->q_ipackets[i] = internal->rx_queue[i].rx_pkts;
 329                 igb_stats->q_ibytes[i] = internal->rx_queue[i].rx_bytes;
 330                 rx_total += igb_stats->q_ipackets[i];
 331                 rx_bytes_total += igb_stats->q_ibytes[i];
 332         }
 333
 334         imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
 335                 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
 336         for (i = 0; i < imax; i++) {
 337                 igb_stats->q_opackets[i] = internal->tx_queue[i].tx_pkts;
 338                 igb_stats->q_obytes[i] = internal->tx_queue[i].tx_bytes;
 339                 tx_total += igb_stats->q_opackets[i];
 340                 tx_err_total += internal->tx_queue[i].err_pkts;
 341                 tx_bytes_total += igb_stats->q_obytes[i];
 342         }
 343
 344         igb_stats->ipackets = rx_total;
 345         igb_stats->ibytes = rx_bytes_total;
 346         igb_stats->opackets = tx_total;
 347         igb_stats->oerrors = tx_err_total;
 348         igb_stats->obytes = tx_bytes_total;
 349         return 0;
 350 }
 351
 352 static void
 353 eth_stats_reset(struct rte_eth_dev *dev)
 354 {
 355         unsigned i;
 356         struct pmd_internals *internal = dev->data->dev_private;
 357
 358         for (i = 0; i < internal->nb_queues; i++) {
 359                 internal->rx_queue[i].rx_pkts = 0;
 360                 internal->rx_queue[i].rx_bytes = 0;
 361         }
 362
 363         for (i = 0; i < internal->nb_queues; i++) {
 364                 internal->tx_queue[i].tx_pkts = 0;
 365                 internal->tx_queue[i].err_pkts = 0;
 366                 internal->tx_queue[i].tx_bytes = 0;
 367         }
 368 }
 369
 370 static void
 371 eth_dev_close(struct rte_eth_dev *dev __rte_unused)
 372 {
 373 }
 374
 375 static void
 376 eth_queue_release(void *q __rte_unused)
 377 {
 378 }
 379
 380 static int
 381 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 382                 int wait_to_complete __rte_unused)
 383 {
 384         return 0;
 385 }
 386
 387 static int
 388 eth_rx_queue_setup(struct rte_eth_dev *dev,
 389                    uint16_t rx_queue_id,
 390                    uint16_t nb_rx_desc __rte_unused,
 391                    unsigned int socket_id __rte_unused,
 392                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 393                    struct rte_mempool *mb_pool)
 394 {
 395         struct pmd_internals *internals = dev->data->dev_private;
 396         struct pkt_rx_queue *pkt_q = &internals->rx_queue[rx_queue_id];
 397         unsigned int buf_size, data_size;
 398
 399         pkt_q->mb_pool = mb_pool;
 400
 401         /* Now get the space available for data in the mbuf */
 402         buf_size = rte_pktmbuf_data_room_size(pkt_q->mb_pool) -
 403                 RTE_PKTMBUF_HEADROOM;
 404         data_size = internals->req.tp_frame_size;
 405         data_size -= TPACKET2_HDRLEN - sizeof(struct sockaddr_ll);
 406
 407         if (data_size > buf_size) {
 408                 PMD_LOG(ERR,
 409                         "%s: %d bytes will not fit in mbuf (%d bytes)",
 410                         dev->device->name, data_size, buf_size);
 411                 return -ENOMEM;
 412         }
 413
 414         dev->data->rx_queues[rx_queue_id] = pkt_q;
 415         pkt_q->in_port = dev->data->port_id;
 416
 417         return 0;
 418 }
 419
 420 static int
 421 eth_tx_queue_setup(struct rte_eth_dev *dev,
 422                    uint16_t tx_queue_id,
 423                    uint16_t nb_tx_desc __rte_unused,
 424                    unsigned int socket_id __rte_unused,
 425                    const struct rte_eth_txconf *tx_conf __rte_unused)
 426 {
 427
 428         struct pmd_internals *internals = dev->data->dev_private;
 429
 430         dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
 431         return 0;
 432 }
 433
 434 static int
 435 eth_dev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 436 {
 437         struct pmd_internals *internals = dev->data->dev_private;
 438         struct ifreq ifr = { .ifr_mtu = mtu };
 439         int ret;
 440         int s;
 441         unsigned int data_size = internals->req.tp_frame_size -
 442                                  TPACKET2_HDRLEN;
 443
 444         if (mtu > data_size)
 445                 return -EINVAL;
 446
 447         s = socket(PF_INET, SOCK_DGRAM, 0);
 448         if (s < 0)
 449                 return -EINVAL;
 450
 451         strlcpy(ifr.ifr_name, internals->if_name, IFNAMSIZ);
 452         ret = ioctl(s, SIOCSIFMTU, &ifr);
 453         close(s);
 454
 455         if (ret < 0)
 456                 return -EINVAL;
 457
 458         return 0;
 459 }
 460
 461 static int
 462 eth_dev_change_flags(char *if_name, uint32_t flags, uint32_t mask)
 463 {
 464         struct ifreq ifr;
 465         int ret = 0;
 466         int s;
 467
 468         s = socket(PF_INET, SOCK_DGRAM, 0);
 469         if (s < 0)
 470                 return -errno;
 471
 472         strlcpy(ifr.ifr_name, if_name, IFNAMSIZ);
 473         if (ioctl(s, SIOCGIFFLAGS, &ifr) < 0) {
 474                 ret = -errno;
 475                 goto out;
 476         }
 477         ifr.ifr_flags &= mask;
 478         ifr.ifr_flags |= flags;
 479         if (ioctl(s, SIOCSIFFLAGS, &ifr) < 0) {
 480                 ret = -errno;
 481                 goto out;
 482         }
 483 out:
 484         close(s);
 485         return ret;
 486 }
 487
 488 static int
 489 eth_dev_promiscuous_enable(struct rte_eth_dev *dev)
 490 {
 491         struct pmd_internals *internals = dev->data->dev_private;
 492
 493         return eth_dev_change_flags(internals->if_name, IFF_PROMISC, ~0);
 494 }
 495
 496 static int
 497 eth_dev_promiscuous_disable(struct rte_eth_dev *dev)
 498 {
 499         struct pmd_internals *internals = dev->data->dev_private;
 500
 501         return eth_dev_change_flags(internals->if_name, 0, ~IFF_PROMISC);
 502 }
 503
 504 static const struct eth_dev_ops ops = {
 505         .dev_start = eth_dev_start,
 506         .dev_stop = eth_dev_stop,
 507         .dev_close = eth_dev_close,
 508         .dev_configure = eth_dev_configure,
 509         .dev_infos_get = eth_dev_info,
 510         .mtu_set = eth_dev_mtu_set,
 511         .promiscuous_enable = eth_dev_promiscuous_enable,
 512         .promiscuous_disable = eth_dev_promiscuous_disable,
 513         .rx_queue_setup = eth_rx_queue_setup,
 514         .tx_queue_setup = eth_tx_queue_setup,
 515         .rx_queue_release = eth_queue_release,
 516         .tx_queue_release = eth_queue_release,
 517         .link_update = eth_link_update,
 518         .stats_get = eth_stats_get,
 519         .stats_reset = eth_stats_reset,
 520 };
 521
 522 /*
 523  * Opens an AF_PACKET socket
 524  */
 525 static int
 526 open_packet_iface(const char *key __rte_unused,
 527                   const char *value __rte_unused,
 528                   void *extra_args)
 529 {
 530         int *sockfd = extra_args;
 531
 532         /* Open an AF_PACKET socket... */
 533         *sockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
 534         if (*sockfd == -1) {
 535                 PMD_LOG(ERR, "Could not open AF_PACKET socket");
 536                 return -1;
 537         }
 538
 539         return 0;
 540 }
 541
 542 static int
 543 rte_pmd_init_internals(struct rte_vdev_device *dev,
 544                        const int sockfd,
 545                        const unsigned nb_queues,
 546                        unsigned int blocksize,
 547                        unsigned int blockcnt,
 548                        unsigned int framesize,
 549                        unsigned int framecnt,
 550                        unsigned int qdisc_bypass,
 551                        struct pmd_internals **internals,
 552                        struct rte_eth_dev **eth_dev,
 553                        struct rte_kvargs *kvlist)
 554 {
 555         const char *name = rte_vdev_device_name(dev);
 556         const unsigned int numa_node = dev->device.numa_node;
 557         struct rte_eth_dev_data *data = NULL;
 558         struct rte_kvargs_pair *pair = NULL;
 559         struct ifreq ifr;
 560         size_t ifnamelen;
 561         unsigned k_idx;
 562         struct sockaddr_ll sockaddr;
 563         struct tpacket_req *req;
 564         struct pkt_rx_queue *rx_queue;
 565         struct pkt_tx_queue *tx_queue;
 566         int rc, tpver, discard;
 567         int qsockfd = -1;
 568         unsigned int i, q, rdsize;
 569 #if defined(PACKET_FANOUT)
 570         int fanout_arg;
 571 #endif
 572
 573         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 574                 pair = &kvlist->pairs[k_idx];
 575                 if (strstr(pair->key, ETH_AF_PACKET_IFACE_ARG) != NULL)
 576                         break;
 577         }
 578         if (pair == NULL) {
 579                 PMD_LOG(ERR,
 580                         "%s: no interface specified for AF_PACKET ethdev",
 581                         name);
 582                 return -1;
 583         }
 584
 585         PMD_LOG(INFO,
 586                 "%s: creating AF_PACKET-backed ethdev on numa socket %u",
 587                 name, numa_node);
 588
 589         *internals = rte_zmalloc_socket(name, sizeof(**internals),
 590                                         0, numa_node);
 591         if (*internals == NULL)
 592                 return -1;
 593
 594         for (q = 0; q < nb_queues; q++) {
 595                 (*internals)->rx_queue[q].map = MAP_FAILED;
 596                 (*internals)->tx_queue[q].map = MAP_FAILED;
 597         }
 598
 599         req = &((*internals)->req);
 600
 601         req->tp_block_size = blocksize;
 602         req->tp_block_nr = blockcnt;
 603         req->tp_frame_size = framesize;
 604         req->tp_frame_nr = framecnt;
 605
 606         ifnamelen = strlen(pair->value);
 607         if (ifnamelen < sizeof(ifr.ifr_name)) {
 608                 memcpy(ifr.ifr_name, pair->value, ifnamelen);
 609                 ifr.ifr_name[ifnamelen] = '\0';
 610         } else {
 611                 PMD_LOG(ERR,
 612                         "%s: I/F name too long (%s)",
 613                         name, pair->value);
 614                 return -1;
 615         }
 616         if (ioctl(sockfd, SIOCGIFINDEX, &ifr) == -1) {
 617                 PMD_LOG_ERRNO(ERR, "%s: ioctl failed (SIOCGIFINDEX)", name);
 618                 return -1;
 619         }
 620         (*internals)->if_name = strdup(pair->value);
 621         if ((*internals)->if_name == NULL)
 622                 return -1;
 623         (*internals)->if_index = ifr.ifr_ifindex;
 624
 625         if (ioctl(sockfd, SIOCGIFHWADDR, &ifr) == -1) {
 626                 PMD_LOG_ERRNO(ERR, "%s: ioctl failed (SIOCGIFHWADDR)", name);
 627                 return -1;
 628         }
 629         memcpy(&(*internals)->eth_addr, ifr.ifr_hwaddr.sa_data, ETH_ALEN);
 630
 631         memset(&sockaddr, 0, sizeof(sockaddr));
 632         sockaddr.sll_family = AF_PACKET;
 633         sockaddr.sll_protocol = htons(ETH_P_ALL);
 634         sockaddr.sll_ifindex = (*internals)->if_index;
 635
 636 #if defined(PACKET_FANOUT)
 637         fanout_arg = (getpid() ^ (*internals)->if_index) & 0xffff;
 638         fanout_arg |= (PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_DEFRAG) << 16;
 639 #if defined(PACKET_FANOUT_FLAG_ROLLOVER)
 640         fanout_arg |= PACKET_FANOUT_FLAG_ROLLOVER << 16;
 641 #endif
 642 #endif
 643
 644         for (q = 0; q < nb_queues; q++) {
 645                 /* Open an AF_PACKET socket for this queue... */
 646                 qsockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
 647                 if (qsockfd == -1) {
 648                         PMD_LOG_ERRNO(ERR,
 649                                 "%s: could not open AF_PACKET socket",
 650                                 name);
 651                         return -1;
 652                 }
 653
 654                 tpver = TPACKET_V2;
 655                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_VERSION,
 656                                 &tpver, sizeof(tpver));
 657                 if (rc == -1) {
 658                         PMD_LOG_ERRNO(ERR,
 659                                 "%s: could not set PACKET_VERSION on AF_PACKET socket for %s",
 660                                 name, pair->value);
 661                         goto error;
 662                 }
 663
 664                 discard = 1;
 665                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_LOSS,
 666                                 &discard, sizeof(discard));
 667                 if (rc == -1) {
 668                         PMD_LOG_ERRNO(ERR,
 669                                 "%s: could not set PACKET_LOSS on AF_PACKET socket for %s",
 670                                 name, pair->value);
 671                         goto error;
 672                 }
 673
 674 #if defined(PACKET_QDISC_BYPASS)
 675                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_QDISC_BYPASS,
 676                                 &qdisc_bypass, sizeof(qdisc_bypass));
 677                 if (rc == -1) {
 678                         PMD_LOG_ERRNO(ERR,
 679                                 "%s: could not set PACKET_QDISC_BYPASS on AF_PACKET socket for %s",
 680                                 name, pair->value);
 681                         goto error;
 682                 }
 683 #else
 684                 RTE_SET_USED(qdisc_bypass);
 685 #endif
 686
 687                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_RX_RING, req, sizeof(*req));
 688                 if (rc == -1) {
 689                         PMD_LOG_ERRNO(ERR,
 690                                 "%s: could not set PACKET_RX_RING on AF_PACKET socket for %s",
 691                                 name, pair->value);
 692                         goto error;
 693                 }
 694
 695                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_TX_RING, req, sizeof(*req));
 696                 if (rc == -1) {
 697                         PMD_LOG_ERRNO(ERR,
 698                                 "%s: could not set PACKET_TX_RING on AF_PACKET "
 699                                 "socket for %s", name, pair->value);
 700                         goto error;
 701                 }
 702
 703                 rx_queue = &((*internals)->rx_queue[q]);
 704                 rx_queue->framecount = req->tp_frame_nr;
 705
 706                 rx_queue->map = mmap(NULL, 2 * req->tp_block_size * req->tp_block_nr,
 707                                     PROT_READ | PROT_WRITE, MAP_SHARED | MAP_LOCKED,
 708                                     qsockfd, 0);
 709                 if (rx_queue->map == MAP_FAILED) {
 710                         PMD_LOG_ERRNO(ERR,
 711                                 "%s: call to mmap failed on AF_PACKET socket for %s",
 712                                 name, pair->value);
 713                         goto error;
 714                 }
 715
 716                 /* rdsize is same for both Tx and Rx */
 717                 rdsize = req->tp_frame_nr * sizeof(*(rx_queue->rd));
 718
 719                 rx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
 720                 if (rx_queue->rd == NULL)
 721                         goto error;
 722                 for (i = 0; i < req->tp_frame_nr; ++i) {
 723                         rx_queue->rd[i].iov_base = rx_queue->map + (i * framesize);
 724                         rx_queue->rd[i].iov_len = req->tp_frame_size;
 725                 }
 726                 rx_queue->sockfd = qsockfd;
 727
 728                 tx_queue = &((*internals)->tx_queue[q]);
 729                 tx_queue->framecount = req->tp_frame_nr;
 730                 tx_queue->frame_data_size = req->tp_frame_size;
 731                 tx_queue->frame_data_size -= TPACKET2_HDRLEN -
 732                         sizeof(struct sockaddr_ll);
 733
 734                 tx_queue->map = rx_queue->map + req->tp_block_size * req->tp_block_nr;
 735
 736                 tx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
 737                 if (tx_queue->rd == NULL)
 738                         goto error;
 739                 for (i = 0; i < req->tp_frame_nr; ++i) {
 740                         tx_queue->rd[i].iov_base = tx_queue->map + (i * framesize);
 741                         tx_queue->rd[i].iov_len = req->tp_frame_size;
 742                 }
 743                 tx_queue->sockfd = qsockfd;
 744
 745                 rc = bind(qsockfd, (const struct sockaddr*)&sockaddr, sizeof(sockaddr));
 746                 if (rc == -1) {
 747                         PMD_LOG_ERRNO(ERR,
 748                                 "%s: could not bind AF_PACKET socket to %s",
 749                                 name, pair->value);
 750                         goto error;
 751                 }
 752
 753 #if defined(PACKET_FANOUT)
 754                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_FANOUT,
 755                                 &fanout_arg, sizeof(fanout_arg));
 756                 if (rc == -1) {
 757                         PMD_LOG_ERRNO(ERR,
 758                                 "%s: could not set PACKET_FANOUT on AF_PACKET socket for %s",
 759                                 name, pair->value);
 760                         goto error;
 761                 }
 762 #endif
 763         }
 764
 765         /* reserve an ethdev entry */
 766         *eth_dev = rte_eth_vdev_allocate(dev, 0);
 767         if (*eth_dev == NULL)
 768                 goto error;
 769
 770         /*
 771          * now put it all together
 772          * - store queue data in internals,
 773          * - store numa_node in eth_dev
 774          * - point eth_dev_data to internals
 775          * - and point eth_dev structure to new eth_dev_data structure
 776          */
 777
 778         (*internals)->nb_queues = nb_queues;
 779
 780         data = (*eth_dev)->data;
 781         data->dev_private = *internals;
 782         data->nb_rx_queues = (uint16_t)nb_queues;
 783         data->nb_tx_queues = (uint16_t)nb_queues;
 784         data->dev_link = pmd_link;
 785         data->mac_addrs = &(*internals)->eth_addr;
 786
 787         (*eth_dev)->dev_ops = &ops;
 788
 789         return 0;
 790
 791 error:
 792         if (qsockfd != -1)
 793                 close(qsockfd);
 794         for (q = 0; q < nb_queues; q++) {
 795                 munmap((*internals)->rx_queue[q].map,
 796                        2 * req->tp_block_size * req->tp_block_nr);
 797
 798                 rte_free((*internals)->rx_queue[q].rd);
 799                 rte_free((*internals)->tx_queue[q].rd);
 800                 if (((*internals)->rx_queue[q].sockfd != 0) &&
 801                         ((*internals)->rx_queue[q].sockfd != qsockfd))
 802                         close((*internals)->rx_queue[q].sockfd);
 803         }
 804         free((*internals)->if_name);
 805         rte_free(*internals);
 806         return -1;
 807 }
 808
 809 static int
 810 rte_eth_from_packet(struct rte_vdev_device *dev,
 811                     int const *sockfd,
 812                     struct rte_kvargs *kvlist)
 813 {
 814         const char *name = rte_vdev_device_name(dev);
 815         struct pmd_internals *internals = NULL;
 816         struct rte_eth_dev *eth_dev = NULL;
 817         struct rte_kvargs_pair *pair = NULL;
 818         unsigned k_idx;
 819         unsigned int blockcount;
 820         unsigned int blocksize;
 821         unsigned int framesize = DFLT_FRAME_SIZE;
 822         unsigned int framecount = DFLT_FRAME_COUNT;
 823         unsigned int qpairs = 1;
 824         unsigned int qdisc_bypass = 1;
 825
 826         /* do some parameter checking */
 827         if (*sockfd < 0)
 828                 return -1;
 829
 830         blocksize = getpagesize();
 831
 832         /*
 833          * Walk arguments for configurable settings
 834          */
 835         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 836                 pair = &kvlist->pairs[k_idx];
 837                 if (strstr(pair->key, ETH_AF_PACKET_NUM_Q_ARG) != NULL) {
 838                         qpairs = atoi(pair->value);
 839                         if (qpairs < 1 ||
 840                             qpairs > RTE_PMD_AF_PACKET_MAX_RINGS) {
 841                                 PMD_LOG(ERR,
 842                                         "%s: invalid qpairs value",
 843                                         name);
 844                                 return -1;
 845                         }
 846                         continue;
 847                 }
 848                 if (strstr(pair->key, ETH_AF_PACKET_BLOCKSIZE_ARG) != NULL) {
 849                         blocksize = atoi(pair->value);
 850                         if (!blocksize) {
 851                                 PMD_LOG(ERR,
 852                                         "%s: invalid blocksize value",
 853                                         name);
 854                                 return -1;
 855                         }
 856                         continue;
 857                 }
 858                 if (strstr(pair->key, ETH_AF_PACKET_FRAMESIZE_ARG) != NULL) {
 859                         framesize = atoi(pair->value);
 860                         if (!framesize) {
 861                                 PMD_LOG(ERR,
 862                                         "%s: invalid framesize value",
 863                                         name);
 864                                 return -1;
 865                         }
 866                         continue;
 867                 }
 868                 if (strstr(pair->key, ETH_AF_PACKET_FRAMECOUNT_ARG) != NULL) {
 869                         framecount = atoi(pair->value);
 870                         if (!framecount) {
 871                                 PMD_LOG(ERR,
 872                                         "%s: invalid framecount value",
 873                                         name);
 874                                 return -1;
 875                         }
 876                         continue;
 877                 }
 878                 if (strstr(pair->key, ETH_AF_PACKET_QDISC_BYPASS_ARG) != NULL) {
 879                         qdisc_bypass = atoi(pair->value);
 880                         if (qdisc_bypass > 1) {
 881                                 PMD_LOG(ERR,
 882                                         "%s: invalid bypass value",
 883                                         name);
 884                                 return -1;
 885                         }
 886                         continue;
 887                 }
 888         }
 889
 890         if (framesize > blocksize) {
 891                 PMD_LOG(ERR,
 892                         "%s: AF_PACKET MMAP frame size exceeds block size!",
 893                         name);
 894                 return -1;
 895         }
 896
 897         blockcount = framecount / (blocksize / framesize);
 898         if (!blockcount) {
 899                 PMD_LOG(ERR,
 900                         "%s: invalid AF_PACKET MMAP parameters", name);
 901                 return -1;
 902         }
 903
 904         PMD_LOG(INFO, "%s: AF_PACKET MMAP parameters:", name);
 905         PMD_LOG(INFO, "%s:\tblock size %d", name, blocksize);
 906         PMD_LOG(INFO, "%s:\tblock count %d", name, blockcount);
 907         PMD_LOG(INFO, "%s:\tframe size %d", name, framesize);
 908         PMD_LOG(INFO, "%s:\tframe count %d", name, framecount);
 909
 910         if (rte_pmd_init_internals(dev, *sockfd, qpairs,
 911                                    blocksize, blockcount,
 912                                    framesize, framecount,
 913                                    qdisc_bypass,
 914                                    &internals, &eth_dev,
 915                                    kvlist) < 0)
 916                 return -1;
 917
 918         eth_dev->rx_pkt_burst = eth_af_packet_rx;
 919         eth_dev->tx_pkt_burst = eth_af_packet_tx;
 920
 921         rte_eth_dev_probing_finish(eth_dev);
 922         return 0;
 923 }
 924
 925 static int
 926 rte_pmd_af_packet_probe(struct rte_vdev_device *dev)
 927 {
 928         int ret = 0;
 929         struct rte_kvargs *kvlist;
 930         int sockfd = -1;
 931         struct rte_eth_dev *eth_dev;
 932         const char *name = rte_vdev_device_name(dev);
 933
 934         PMD_LOG(INFO, "Initializing pmd_af_packet for %s", name);
 935
 936         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
 937                 eth_dev = rte_eth_dev_attach_secondary(name);
 938                 if (!eth_dev) {
 939                         PMD_LOG(ERR, "Failed to probe %s", name);
 940                         return -1;
 941                 }
 942                 /* TODO: request info from primary to set up Rx and Tx */
 943                 eth_dev->dev_ops = &ops;
 944                 eth_dev->device = &dev->device;
 945                 rte_eth_dev_probing_finish(eth_dev);
 946                 return 0;
 947         }
 948
 949         kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_arguments);
 950         if (kvlist == NULL) {
 951                 ret = -1;
 952                 goto exit;
 953         }
 954
 955         /*
 956          * If iface argument is passed we open the NICs and use them for
 957          * reading / writing
 958          */
 959         if (rte_kvargs_count(kvlist, ETH_AF_PACKET_IFACE_ARG) == 1) {
 960
 961                 ret = rte_kvargs_process(kvlist, ETH_AF_PACKET_IFACE_ARG,
 962                                          &open_packet_iface, &sockfd);
 963                 if (ret < 0)
 964                         goto exit;
 965         }
 966
 967         if (dev->device.numa_node == SOCKET_ID_ANY)
 968                 dev->device.numa_node = rte_socket_id();
 969
 970         ret = rte_eth_from_packet(dev, &sockfd, kvlist);
 971         close(sockfd); /* no longer needed */
 972
 973 exit:
 974         rte_kvargs_free(kvlist);
 975         return ret;
 976 }
 977
 978 static int
 979 rte_pmd_af_packet_remove(struct rte_vdev_device *dev)
 980 {
 981         struct rte_eth_dev *eth_dev = NULL;
 982         struct pmd_internals *internals;
 983         struct tpacket_req *req;
 984         unsigned q;
 985
 986         PMD_LOG(INFO, "Closing AF_PACKET ethdev on numa socket %u",
 987                 rte_socket_id());
 988
 989         if (dev == NULL)
 990                 return -1;
 991
 992         /* find the ethdev entry */
 993         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
 994         if (eth_dev == NULL)
 995                 return -1;
 996
 997         /* mac_addrs must not be freed alone because part of dev_private */
 998         eth_dev->data->mac_addrs = NULL;
 999
1000         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1001                 return rte_eth_dev_release_port(eth_dev);
1002
1003         internals = eth_dev->data->dev_private;
1004         req = &internals->req;
1005         for (q = 0; q < internals->nb_queues; q++) {
1006                 munmap(internals->rx_queue[q].map,
1007                         2 * req->tp_block_size * req->tp_block_nr);
1008                 rte_free(internals->rx_queue[q].rd);
1009                 rte_free(internals->tx_queue[q].rd);
1010         }
1011         free(internals->if_name);
1012
1013         rte_eth_dev_release_port(eth_dev);
1014
1015         return 0;
1016 }
1017
1018 static struct rte_vdev_driver pmd_af_packet_drv = {
1019         .probe = rte_pmd_af_packet_probe,
1020         .remove = rte_pmd_af_packet_remove,
1021 };
1022
1023 RTE_PMD_REGISTER_VDEV(net_af_packet, pmd_af_packet_drv);
1024 RTE_PMD_REGISTER_ALIAS(net_af_packet, eth_af_packet);
1025 RTE_PMD_REGISTER_PARAM_STRING(net_af_packet,
1026         "iface=<string> "
1027         "qpairs=<int> "
1028         "blocksz=<int> "
1029         "framesz=<int> "
1030         "framecnt=<int> "
1031         "qdisc_bypass=<0|1>");
1032
1033 RTE_INIT(af_packet_init_log)
1034 {
1035         af_packet_logtype = rte_log_register("pmd.net.packet");
1036         if (af_packet_logtype >= 0)
1037                 rte_log_set_level(af_packet_logtype, RTE_LOG_NOTICE);
1038 }