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                         errno != ENOBUFS && errno != EAGAIN) {
 249                 /*
 250                  * In case of a ENOBUFS/EAGAIN error all of the enqueued
 251                  * packets will be considered successful even though only some
 252                  * are sent.
 253                  */
 254
 255                 num_tx = 0;
 256                 num_tx_bytes = 0;
 257         }
 258
 259         pkt_q->framenum = framenum;
 260         pkt_q->tx_pkts += num_tx;
 261         pkt_q->err_pkts += i - num_tx;
 262         pkt_q->tx_bytes += num_tx_bytes;
 263         return i;
 264 }
 265
 266 static int
 267 eth_dev_start(struct rte_eth_dev *dev)
 268 {
 269         dev->data->dev_link.link_status = ETH_LINK_UP;
 270         return 0;
 271 }
 272
 273 /*
 274  * This function gets called when the current port gets stopped.
 275  */
 276 static void
 277 eth_dev_stop(struct rte_eth_dev *dev)
 278 {
 279         unsigned i;
 280         int sockfd;
 281         struct pmd_internals *internals = dev->data->dev_private;
 282
 283         for (i = 0; i < internals->nb_queues; i++) {
 284                 sockfd = internals->rx_queue[i].sockfd;
 285                 if (sockfd != -1)
 286                         close(sockfd);
 287
 288                 /* Prevent use after free in case tx fd == rx fd */
 289                 if (sockfd != internals->tx_queue[i].sockfd) {
 290                         sockfd = internals->tx_queue[i].sockfd;
 291                         if (sockfd != -1)
 292                                 close(sockfd);
 293                 }
 294
 295                 internals->rx_queue[i].sockfd = -1;
 296                 internals->tx_queue[i].sockfd = -1;
 297         }
 298
 299         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 300 }
 301
 302 static int
 303 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 304 {
 305         return 0;
 306 }
 307
 308 static int
 309 eth_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 310 {
 311         struct pmd_internals *internals = dev->data->dev_private;
 312
 313         dev_info->if_index = internals->if_index;
 314         dev_info->max_mac_addrs = 1;
 315         dev_info->max_rx_pktlen = (uint32_t)ETH_FRAME_LEN;
 316         dev_info->max_rx_queues = (uint16_t)internals->nb_queues;
 317         dev_info->max_tx_queues = (uint16_t)internals->nb_queues;
 318         dev_info->min_rx_bufsize = 0;
 319
 320         return 0;
 321 }
 322
 323 static int
 324 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *igb_stats)
 325 {
 326         unsigned i, imax;
 327         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 328         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 329         const struct pmd_internals *internal = dev->data->dev_private;
 330
 331         imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
 332                 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
 333         for (i = 0; i < imax; i++) {
 334                 igb_stats->q_ipackets[i] = internal->rx_queue[i].rx_pkts;
 335                 igb_stats->q_ibytes[i] = internal->rx_queue[i].rx_bytes;
 336                 rx_total += igb_stats->q_ipackets[i];
 337                 rx_bytes_total += igb_stats->q_ibytes[i];
 338         }
 339
 340         imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
 341                 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
 342         for (i = 0; i < imax; i++) {
 343                 igb_stats->q_opackets[i] = internal->tx_queue[i].tx_pkts;
 344                 igb_stats->q_obytes[i] = internal->tx_queue[i].tx_bytes;
 345                 tx_total += igb_stats->q_opackets[i];
 346                 tx_err_total += internal->tx_queue[i].err_pkts;
 347                 tx_bytes_total += igb_stats->q_obytes[i];
 348         }
 349
 350         igb_stats->ipackets = rx_total;
 351         igb_stats->ibytes = rx_bytes_total;
 352         igb_stats->opackets = tx_total;
 353         igb_stats->oerrors = tx_err_total;
 354         igb_stats->obytes = tx_bytes_total;
 355         return 0;
 356 }
 357
 358 static int
 359 eth_stats_reset(struct rte_eth_dev *dev)
 360 {
 361         unsigned i;
 362         struct pmd_internals *internal = dev->data->dev_private;
 363
 364         for (i = 0; i < internal->nb_queues; i++) {
 365                 internal->rx_queue[i].rx_pkts = 0;
 366                 internal->rx_queue[i].rx_bytes = 0;
 367         }
 368
 369         for (i = 0; i < internal->nb_queues; i++) {
 370                 internal->tx_queue[i].tx_pkts = 0;
 371                 internal->tx_queue[i].err_pkts = 0;
 372                 internal->tx_queue[i].tx_bytes = 0;
 373         }
 374
 375         return 0;
 376 }
 377
 378 static void
 379 eth_dev_close(struct rte_eth_dev *dev __rte_unused)
 380 {
 381 }
 382
 383 static void
 384 eth_queue_release(void *q __rte_unused)
 385 {
 386 }
 387
 388 static int
 389 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 390                 int wait_to_complete __rte_unused)
 391 {
 392         return 0;
 393 }
 394
 395 static int
 396 eth_rx_queue_setup(struct rte_eth_dev *dev,
 397                    uint16_t rx_queue_id,
 398                    uint16_t nb_rx_desc __rte_unused,
 399                    unsigned int socket_id __rte_unused,
 400                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 401                    struct rte_mempool *mb_pool)
 402 {
 403         struct pmd_internals *internals = dev->data->dev_private;
 404         struct pkt_rx_queue *pkt_q = &internals->rx_queue[rx_queue_id];
 405         unsigned int buf_size, data_size;
 406
 407         pkt_q->mb_pool = mb_pool;
 408
 409         /* Now get the space available for data in the mbuf */
 410         buf_size = rte_pktmbuf_data_room_size(pkt_q->mb_pool) -
 411                 RTE_PKTMBUF_HEADROOM;
 412         data_size = internals->req.tp_frame_size;
 413         data_size -= TPACKET2_HDRLEN - sizeof(struct sockaddr_ll);
 414
 415         if (data_size > buf_size) {
 416                 PMD_LOG(ERR,
 417                         "%s: %d bytes will not fit in mbuf (%d bytes)",
 418                         dev->device->name, data_size, buf_size);
 419                 return -ENOMEM;
 420         }
 421
 422         dev->data->rx_queues[rx_queue_id] = pkt_q;
 423         pkt_q->in_port = dev->data->port_id;
 424
 425         return 0;
 426 }
 427
 428 static int
 429 eth_tx_queue_setup(struct rte_eth_dev *dev,
 430                    uint16_t tx_queue_id,
 431                    uint16_t nb_tx_desc __rte_unused,
 432                    unsigned int socket_id __rte_unused,
 433                    const struct rte_eth_txconf *tx_conf __rte_unused)
 434 {
 435
 436         struct pmd_internals *internals = dev->data->dev_private;
 437
 438         dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
 439         return 0;
 440 }
 441
 442 static int
 443 eth_dev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 444 {
 445         struct pmd_internals *internals = dev->data->dev_private;
 446         struct ifreq ifr = { .ifr_mtu = mtu };
 447         int ret;
 448         int s;
 449         unsigned int data_size = internals->req.tp_frame_size -
 450                                  TPACKET2_HDRLEN;
 451
 452         if (mtu > data_size)
 453                 return -EINVAL;
 454
 455         s = socket(PF_INET, SOCK_DGRAM, 0);
 456         if (s < 0)
 457                 return -EINVAL;
 458
 459         strlcpy(ifr.ifr_name, internals->if_name, IFNAMSIZ);
 460         ret = ioctl(s, SIOCSIFMTU, &ifr);
 461         close(s);
 462
 463         if (ret < 0)
 464                 return -EINVAL;
 465
 466         return 0;
 467 }
 468
 469 static int
 470 eth_dev_change_flags(char *if_name, uint32_t flags, uint32_t mask)
 471 {
 472         struct ifreq ifr;
 473         int ret = 0;
 474         int s;
 475
 476         s = socket(PF_INET, SOCK_DGRAM, 0);
 477         if (s < 0)
 478                 return -errno;
 479
 480         strlcpy(ifr.ifr_name, if_name, IFNAMSIZ);
 481         if (ioctl(s, SIOCGIFFLAGS, &ifr) < 0) {
 482                 ret = -errno;
 483                 goto out;
 484         }
 485         ifr.ifr_flags &= mask;
 486         ifr.ifr_flags |= flags;
 487         if (ioctl(s, SIOCSIFFLAGS, &ifr) < 0) {
 488                 ret = -errno;
 489                 goto out;
 490         }
 491 out:
 492         close(s);
 493         return ret;
 494 }
 495
 496 static int
 497 eth_dev_promiscuous_enable(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, IFF_PROMISC, ~0);
 502 }
 503
 504 static int
 505 eth_dev_promiscuous_disable(struct rte_eth_dev *dev)
 506 {
 507         struct pmd_internals *internals = dev->data->dev_private;
 508
 509         return eth_dev_change_flags(internals->if_name, 0, ~IFF_PROMISC);
 510 }
 511
 512 static const struct eth_dev_ops ops = {
 513         .dev_start = eth_dev_start,
 514         .dev_stop = eth_dev_stop,
 515         .dev_close = eth_dev_close,
 516         .dev_configure = eth_dev_configure,
 517         .dev_infos_get = eth_dev_info,
 518         .mtu_set = eth_dev_mtu_set,
 519         .promiscuous_enable = eth_dev_promiscuous_enable,
 520         .promiscuous_disable = eth_dev_promiscuous_disable,
 521         .rx_queue_setup = eth_rx_queue_setup,
 522         .tx_queue_setup = eth_tx_queue_setup,
 523         .rx_queue_release = eth_queue_release,
 524         .tx_queue_release = eth_queue_release,
 525         .link_update = eth_link_update,
 526         .stats_get = eth_stats_get,
 527         .stats_reset = eth_stats_reset,
 528 };
 529
 530 /*
 531  * Opens an AF_PACKET socket
 532  */
 533 static int
 534 open_packet_iface(const char *key __rte_unused,
 535                   const char *value __rte_unused,
 536                   void *extra_args)
 537 {
 538         int *sockfd = extra_args;
 539
 540         /* Open an AF_PACKET socket... */
 541         *sockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
 542         if (*sockfd == -1) {
 543                 PMD_LOG(ERR, "Could not open AF_PACKET socket");
 544                 return -1;
 545         }
 546
 547         return 0;
 548 }
 549
 550 static int
 551 rte_pmd_init_internals(struct rte_vdev_device *dev,
 552                        const int sockfd,
 553                        const unsigned nb_queues,
 554                        unsigned int blocksize,
 555                        unsigned int blockcnt,
 556                        unsigned int framesize,
 557                        unsigned int framecnt,
 558                        unsigned int qdisc_bypass,
 559                        struct pmd_internals **internals,
 560                        struct rte_eth_dev **eth_dev,
 561                        struct rte_kvargs *kvlist)
 562 {
 563         const char *name = rte_vdev_device_name(dev);
 564         const unsigned int numa_node = dev->device.numa_node;
 565         struct rte_eth_dev_data *data = NULL;
 566         struct rte_kvargs_pair *pair = NULL;
 567         struct ifreq ifr;
 568         size_t ifnamelen;
 569         unsigned k_idx;
 570         struct sockaddr_ll sockaddr;
 571         struct tpacket_req *req;
 572         struct pkt_rx_queue *rx_queue;
 573         struct pkt_tx_queue *tx_queue;
 574         int rc, tpver, discard;
 575         int qsockfd = -1;
 576         unsigned int i, q, rdsize;
 577 #if defined(PACKET_FANOUT)
 578         int fanout_arg;
 579 #endif
 580
 581         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 582                 pair = &kvlist->pairs[k_idx];
 583                 if (strstr(pair->key, ETH_AF_PACKET_IFACE_ARG) != NULL)
 584                         break;
 585         }
 586         if (pair == NULL) {
 587                 PMD_LOG(ERR,
 588                         "%s: no interface specified for AF_PACKET ethdev",
 589                         name);
 590                 return -1;
 591         }
 592
 593         PMD_LOG(INFO,
 594                 "%s: creating AF_PACKET-backed ethdev on numa socket %u",
 595                 name, numa_node);
 596
 597         *internals = rte_zmalloc_socket(name, sizeof(**internals),
 598                                         0, numa_node);
 599         if (*internals == NULL)
 600                 return -1;
 601
 602         for (q = 0; q < nb_queues; q++) {
 603                 (*internals)->rx_queue[q].map = MAP_FAILED;
 604                 (*internals)->tx_queue[q].map = MAP_FAILED;
 605         }
 606
 607         req = &((*internals)->req);
 608
 609         req->tp_block_size = blocksize;
 610         req->tp_block_nr = blockcnt;
 611         req->tp_frame_size = framesize;
 612         req->tp_frame_nr = framecnt;
 613
 614         ifnamelen = strlen(pair->value);
 615         if (ifnamelen < sizeof(ifr.ifr_name)) {
 616                 memcpy(ifr.ifr_name, pair->value, ifnamelen);
 617                 ifr.ifr_name[ifnamelen] = '\0';
 618         } else {
 619                 PMD_LOG(ERR,
 620                         "%s: I/F name too long (%s)",
 621                         name, pair->value);
 622                 return -1;
 623         }
 624         if (ioctl(sockfd, SIOCGIFINDEX, &ifr) == -1) {
 625                 PMD_LOG_ERRNO(ERR, "%s: ioctl failed (SIOCGIFINDEX)", name);
 626                 return -1;
 627         }
 628         (*internals)->if_name = strdup(pair->value);
 629         if ((*internals)->if_name == NULL)
 630                 return -1;
 631         (*internals)->if_index = ifr.ifr_ifindex;
 632
 633         if (ioctl(sockfd, SIOCGIFHWADDR, &ifr) == -1) {
 634                 PMD_LOG_ERRNO(ERR, "%s: ioctl failed (SIOCGIFHWADDR)", name);
 635                 return -1;
 636         }
 637         memcpy(&(*internals)->eth_addr, ifr.ifr_hwaddr.sa_data, ETH_ALEN);
 638
 639         memset(&sockaddr, 0, sizeof(sockaddr));
 640         sockaddr.sll_family = AF_PACKET;
 641         sockaddr.sll_protocol = htons(ETH_P_ALL);
 642         sockaddr.sll_ifindex = (*internals)->if_index;
 643
 644 #if defined(PACKET_FANOUT)
 645         fanout_arg = (getpid() ^ (*internals)->if_index) & 0xffff;
 646         fanout_arg |= (PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_DEFRAG) << 16;
 647 #if defined(PACKET_FANOUT_FLAG_ROLLOVER)
 648         fanout_arg |= PACKET_FANOUT_FLAG_ROLLOVER << 16;
 649 #endif
 650 #endif
 651
 652         for (q = 0; q < nb_queues; q++) {
 653                 /* Open an AF_PACKET socket for this queue... */
 654                 qsockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
 655                 if (qsockfd == -1) {
 656                         PMD_LOG_ERRNO(ERR,
 657                                 "%s: could not open AF_PACKET socket",
 658                                 name);
 659                         return -1;
 660                 }
 661
 662                 tpver = TPACKET_V2;
 663                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_VERSION,
 664                                 &tpver, sizeof(tpver));
 665                 if (rc == -1) {
 666                         PMD_LOG_ERRNO(ERR,
 667                                 "%s: could not set PACKET_VERSION on AF_PACKET socket for %s",
 668                                 name, pair->value);
 669                         goto error;
 670                 }
 671
 672                 discard = 1;
 673                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_LOSS,
 674                                 &discard, sizeof(discard));
 675                 if (rc == -1) {
 676                         PMD_LOG_ERRNO(ERR,
 677                                 "%s: could not set PACKET_LOSS on AF_PACKET socket for %s",
 678                                 name, pair->value);
 679                         goto error;
 680                 }
 681
 682 #if defined(PACKET_QDISC_BYPASS)
 683                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_QDISC_BYPASS,
 684                                 &qdisc_bypass, sizeof(qdisc_bypass));
 685                 if (rc == -1) {
 686                         PMD_LOG_ERRNO(ERR,
 687                                 "%s: could not set PACKET_QDISC_BYPASS on AF_PACKET socket for %s",
 688                                 name, pair->value);
 689                         goto error;
 690                 }
 691 #else
 692                 RTE_SET_USED(qdisc_bypass);
 693 #endif
 694
 695                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_RX_RING, req, sizeof(*req));
 696                 if (rc == -1) {
 697                         PMD_LOG_ERRNO(ERR,
 698                                 "%s: could not set PACKET_RX_RING on AF_PACKET socket for %s",
 699                                 name, pair->value);
 700                         goto error;
 701                 }
 702
 703                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_TX_RING, req, sizeof(*req));
 704                 if (rc == -1) {
 705                         PMD_LOG_ERRNO(ERR,
 706                                 "%s: could not set PACKET_TX_RING on AF_PACKET "
 707                                 "socket for %s", name, pair->value);
 708                         goto error;
 709                 }
 710
 711                 rx_queue = &((*internals)->rx_queue[q]);
 712                 rx_queue->framecount = req->tp_frame_nr;
 713
 714                 rx_queue->map = mmap(NULL, 2 * req->tp_block_size * req->tp_block_nr,
 715                                     PROT_READ | PROT_WRITE, MAP_SHARED | MAP_LOCKED,
 716                                     qsockfd, 0);
 717                 if (rx_queue->map == MAP_FAILED) {
 718                         PMD_LOG_ERRNO(ERR,
 719                                 "%s: call to mmap failed on AF_PACKET socket for %s",
 720                                 name, pair->value);
 721                         goto error;
 722                 }
 723
 724                 /* rdsize is same for both Tx and Rx */
 725                 rdsize = req->tp_frame_nr * sizeof(*(rx_queue->rd));
 726
 727                 rx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
 728                 if (rx_queue->rd == NULL)
 729                         goto error;
 730                 for (i = 0; i < req->tp_frame_nr; ++i) {
 731                         rx_queue->rd[i].iov_base = rx_queue->map + (i * framesize);
 732                         rx_queue->rd[i].iov_len = req->tp_frame_size;
 733                 }
 734                 rx_queue->sockfd = qsockfd;
 735
 736                 tx_queue = &((*internals)->tx_queue[q]);
 737                 tx_queue->framecount = req->tp_frame_nr;
 738                 tx_queue->frame_data_size = req->tp_frame_size;
 739                 tx_queue->frame_data_size -= TPACKET2_HDRLEN -
 740                         sizeof(struct sockaddr_ll);
 741
 742                 tx_queue->map = rx_queue->map + req->tp_block_size * req->tp_block_nr;
 743
 744                 tx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
 745                 if (tx_queue->rd == NULL)
 746                         goto error;
 747                 for (i = 0; i < req->tp_frame_nr; ++i) {
 748                         tx_queue->rd[i].iov_base = tx_queue->map + (i * framesize);
 749                         tx_queue->rd[i].iov_len = req->tp_frame_size;
 750                 }
 751                 tx_queue->sockfd = qsockfd;
 752
 753                 rc = bind(qsockfd, (const struct sockaddr*)&sockaddr, sizeof(sockaddr));
 754                 if (rc == -1) {
 755                         PMD_LOG_ERRNO(ERR,
 756                                 "%s: could not bind AF_PACKET socket to %s",
 757                                 name, pair->value);
 758                         goto error;
 759                 }
 760
 761 #if defined(PACKET_FANOUT)
 762                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_FANOUT,
 763                                 &fanout_arg, sizeof(fanout_arg));
 764                 if (rc == -1) {
 765                         PMD_LOG_ERRNO(ERR,
 766                                 "%s: could not set PACKET_FANOUT on AF_PACKET socket for %s",
 767                                 name, pair->value);
 768                         goto error;
 769                 }
 770 #endif
 771         }
 772
 773         /* reserve an ethdev entry */
 774         *eth_dev = rte_eth_vdev_allocate(dev, 0);
 775         if (*eth_dev == NULL)
 776                 goto error;
 777
 778         /*
 779          * now put it all together
 780          * - store queue data in internals,
 781          * - store numa_node in eth_dev
 782          * - point eth_dev_data to internals
 783          * - and point eth_dev structure to new eth_dev_data structure
 784          */
 785
 786         (*internals)->nb_queues = nb_queues;
 787
 788         data = (*eth_dev)->data;
 789         data->dev_private = *internals;
 790         data->nb_rx_queues = (uint16_t)nb_queues;
 791         data->nb_tx_queues = (uint16_t)nb_queues;
 792         data->dev_link = pmd_link;
 793         data->mac_addrs = &(*internals)->eth_addr;
 794
 795         (*eth_dev)->dev_ops = &ops;
 796
 797         return 0;
 798
 799 error:
 800         if (qsockfd != -1)
 801                 close(qsockfd);
 802         for (q = 0; q < nb_queues; q++) {
 803                 munmap((*internals)->rx_queue[q].map,
 804                        2 * req->tp_block_size * req->tp_block_nr);
 805
 806                 rte_free((*internals)->rx_queue[q].rd);
 807                 rte_free((*internals)->tx_queue[q].rd);
 808                 if (((*internals)->rx_queue[q].sockfd != 0) &&
 809                         ((*internals)->rx_queue[q].sockfd != qsockfd))
 810                         close((*internals)->rx_queue[q].sockfd);
 811         }
 812         free((*internals)->if_name);
 813         rte_free(*internals);
 814         return -1;
 815 }
 816
 817 static int
 818 rte_eth_from_packet(struct rte_vdev_device *dev,
 819                     int const *sockfd,
 820                     struct rte_kvargs *kvlist)
 821 {
 822         const char *name = rte_vdev_device_name(dev);
 823         struct pmd_internals *internals = NULL;
 824         struct rte_eth_dev *eth_dev = NULL;
 825         struct rte_kvargs_pair *pair = NULL;
 826         unsigned k_idx;
 827         unsigned int blockcount;
 828         unsigned int blocksize;
 829         unsigned int framesize = DFLT_FRAME_SIZE;
 830         unsigned int framecount = DFLT_FRAME_COUNT;
 831         unsigned int qpairs = 1;
 832         unsigned int qdisc_bypass = 1;
 833
 834         /* do some parameter checking */
 835         if (*sockfd < 0)
 836                 return -1;
 837
 838         blocksize = getpagesize();
 839
 840         /*
 841          * Walk arguments for configurable settings
 842          */
 843         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 844                 pair = &kvlist->pairs[k_idx];
 845                 if (strstr(pair->key, ETH_AF_PACKET_NUM_Q_ARG) != NULL) {
 846                         qpairs = atoi(pair->value);
 847                         if (qpairs < 1 ||
 848                             qpairs > RTE_PMD_AF_PACKET_MAX_RINGS) {
 849                                 PMD_LOG(ERR,
 850                                         "%s: invalid qpairs value",
 851                                         name);
 852                                 return -1;
 853                         }
 854                         continue;
 855                 }
 856                 if (strstr(pair->key, ETH_AF_PACKET_BLOCKSIZE_ARG) != NULL) {
 857                         blocksize = atoi(pair->value);
 858                         if (!blocksize) {
 859                                 PMD_LOG(ERR,
 860                                         "%s: invalid blocksize value",
 861                                         name);
 862                                 return -1;
 863                         }
 864                         continue;
 865                 }
 866                 if (strstr(pair->key, ETH_AF_PACKET_FRAMESIZE_ARG) != NULL) {
 867                         framesize = atoi(pair->value);
 868                         if (!framesize) {
 869                                 PMD_LOG(ERR,
 870                                         "%s: invalid framesize value",
 871                                         name);
 872                                 return -1;
 873                         }
 874                         continue;
 875                 }
 876                 if (strstr(pair->key, ETH_AF_PACKET_FRAMECOUNT_ARG) != NULL) {
 877                         framecount = atoi(pair->value);
 878                         if (!framecount) {
 879                                 PMD_LOG(ERR,
 880                                         "%s: invalid framecount value",
 881                                         name);
 882                                 return -1;
 883                         }
 884                         continue;
 885                 }
 886                 if (strstr(pair->key, ETH_AF_PACKET_QDISC_BYPASS_ARG) != NULL) {
 887                         qdisc_bypass = atoi(pair->value);
 888                         if (qdisc_bypass > 1) {
 889                                 PMD_LOG(ERR,
 890                                         "%s: invalid bypass value",
 891                                         name);
 892                                 return -1;
 893                         }
 894                         continue;
 895                 }
 896         }
 897
 898         if (framesize > blocksize) {
 899                 PMD_LOG(ERR,
 900                         "%s: AF_PACKET MMAP frame size exceeds block size!",
 901                         name);
 902                 return -1;
 903         }
 904
 905         blockcount = framecount / (blocksize / framesize);
 906         if (!blockcount) {
 907                 PMD_LOG(ERR,
 908                         "%s: invalid AF_PACKET MMAP parameters", name);
 909                 return -1;
 910         }
 911
 912         PMD_LOG(INFO, "%s: AF_PACKET MMAP parameters:", name);
 913         PMD_LOG(INFO, "%s:\tblock size %d", name, blocksize);
 914         PMD_LOG(INFO, "%s:\tblock count %d", name, blockcount);
 915         PMD_LOG(INFO, "%s:\tframe size %d", name, framesize);
 916         PMD_LOG(INFO, "%s:\tframe count %d", name, framecount);
 917
 918         if (rte_pmd_init_internals(dev, *sockfd, qpairs,
 919                                    blocksize, blockcount,
 920                                    framesize, framecount,
 921                                    qdisc_bypass,
 922                                    &internals, &eth_dev,
 923                                    kvlist) < 0)
 924                 return -1;
 925
 926         eth_dev->rx_pkt_burst = eth_af_packet_rx;
 927         eth_dev->tx_pkt_burst = eth_af_packet_tx;
 928
 929         rte_eth_dev_probing_finish(eth_dev);
 930         return 0;
 931 }
 932
 933 static int
 934 rte_pmd_af_packet_probe(struct rte_vdev_device *dev)
 935 {
 936         int ret = 0;
 937         struct rte_kvargs *kvlist;
 938         int sockfd = -1;
 939         struct rte_eth_dev *eth_dev;
 940         const char *name = rte_vdev_device_name(dev);
 941
 942         PMD_LOG(INFO, "Initializing pmd_af_packet for %s", name);
 943
 944         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
 945                 eth_dev = rte_eth_dev_attach_secondary(name);
 946                 if (!eth_dev) {
 947                         PMD_LOG(ERR, "Failed to probe %s", name);
 948                         return -1;
 949                 }
 950                 /* TODO: request info from primary to set up Rx and Tx */
 951                 eth_dev->dev_ops = &ops;
 952                 eth_dev->device = &dev->device;
 953                 rte_eth_dev_probing_finish(eth_dev);
 954                 return 0;
 955         }
 956
 957         kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_arguments);
 958         if (kvlist == NULL) {
 959                 ret = -1;
 960                 goto exit;
 961         }
 962
 963         /*
 964          * If iface argument is passed we open the NICs and use them for
 965          * reading / writing
 966          */
 967         if (rte_kvargs_count(kvlist, ETH_AF_PACKET_IFACE_ARG) == 1) {
 968
 969                 ret = rte_kvargs_process(kvlist, ETH_AF_PACKET_IFACE_ARG,
 970                                          &open_packet_iface, &sockfd);
 971                 if (ret < 0)
 972                         goto exit;
 973         }
 974
 975         if (dev->device.numa_node == SOCKET_ID_ANY)
 976                 dev->device.numa_node = rte_socket_id();
 977
 978         ret = rte_eth_from_packet(dev, &sockfd, kvlist);
 979         close(sockfd); /* no longer needed */
 980
 981 exit:
 982         rte_kvargs_free(kvlist);
 983         return ret;
 984 }
 985
 986 static int
 987 rte_pmd_af_packet_remove(struct rte_vdev_device *dev)
 988 {
 989         struct rte_eth_dev *eth_dev = NULL;
 990         struct pmd_internals *internals;
 991         struct tpacket_req *req;
 992         unsigned q;
 993
 994         PMD_LOG(INFO, "Closing AF_PACKET ethdev on numa socket %u",
 995                 rte_socket_id());
 996
 997         if (dev == NULL)
 998                 return -1;
 999
1000         /* find the ethdev entry */
1001         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1002         if (eth_dev == NULL)
1003                 return -1;
1004
1005         /* mac_addrs must not be freed alone because part of dev_private */
1006         eth_dev->data->mac_addrs = NULL;
1007
1008         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1009                 return rte_eth_dev_release_port(eth_dev);
1010
1011         internals = eth_dev->data->dev_private;
1012         req = &internals->req;
1013         for (q = 0; q < internals->nb_queues; q++) {
1014                 munmap(internals->rx_queue[q].map,
1015                         2 * req->tp_block_size * req->tp_block_nr);
1016                 rte_free(internals->rx_queue[q].rd);
1017                 rte_free(internals->tx_queue[q].rd);
1018         }
1019         free(internals->if_name);
1020
1021         rte_eth_dev_release_port(eth_dev);
1022
1023         return 0;
1024 }
1025
1026 static struct rte_vdev_driver pmd_af_packet_drv = {
1027         .probe = rte_pmd_af_packet_probe,
1028         .remove = rte_pmd_af_packet_remove,
1029 };
1030
1031 RTE_PMD_REGISTER_VDEV(net_af_packet, pmd_af_packet_drv);
1032 RTE_PMD_REGISTER_ALIAS(net_af_packet, eth_af_packet);
1033 RTE_PMD_REGISTER_PARAM_STRING(net_af_packet,
1034         "iface=<string> "
1035         "qpairs=<int> "
1036         "blocksz=<int> "
1037         "framesz=<int> "
1038         "framecnt=<int> "
1039         "qdisc_bypass=<0|1>");
1040
1041 RTE_INIT(af_packet_init_log)
1042 {
1043         af_packet_logtype = rte_log_register("pmd.net.packet");
1044         if (af_packet_logtype >= 0)
1045                 rte_log_set_level(af_packet_logtype, RTE_LOG_NOTICE);
1046 }