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