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_mbuf.h>
  10 #include <rte_ethdev_driver.h>
  11 #include <rte_ethdev_vdev.h>
  12 #include <rte_malloc.h>
  13 #include <rte_kvargs.h>
  14 #include <rte_bus_vdev.h>
  15
  16 #include <linux/if_ether.h>
  17 #include <linux/if_packet.h>
  18 #include <arpa/inet.h>
  19 #include <net/if.h>
  20 #include <sys/types.h>
  21 #include <sys/socket.h>
  22 #include <sys/ioctl.h>
  23 #include <sys/mman.h>
  24 #include <unistd.h>
  25 #include <poll.h>
  26
  27 #define ETH_AF_PACKET_IFACE_ARG         "iface"
  28 #define ETH_AF_PACKET_NUM_Q_ARG         "qpairs"
  29 #define ETH_AF_PACKET_BLOCKSIZE_ARG     "blocksz"
  30 #define ETH_AF_PACKET_FRAMESIZE_ARG     "framesz"
  31 #define ETH_AF_PACKET_FRAMECOUNT_ARG    "framecnt"
  32 #define ETH_AF_PACKET_QDISC_BYPASS_ARG  "qdisc_bypass"
  33
  34 #define DFLT_BLOCK_SIZE         (1 << 12)
  35 #define DFLT_FRAME_SIZE         (1 << 11)
  36 #define DFLT_FRAME_COUNT        (1 << 9)
  37
  38 #define RTE_PMD_AF_PACKET_MAX_RINGS 16
  39
  40 struct pkt_rx_queue {
  41         int sockfd;
  42
  43         struct iovec *rd;
  44         uint8_t *map;
  45         unsigned int framecount;
  46         unsigned int framenum;
  47
  48         struct rte_mempool *mb_pool;
  49         uint16_t in_port;
  50
  51         volatile unsigned long rx_pkts;
  52         volatile unsigned long err_pkts;
  53         volatile unsigned long rx_bytes;
  54 };
  55
  56 struct pkt_tx_queue {
  57         int sockfd;
  58         unsigned int frame_data_size;
  59
  60         struct iovec *rd;
  61         uint8_t *map;
  62         unsigned int framecount;
  63         unsigned int framenum;
  64
  65         volatile unsigned long tx_pkts;
  66         volatile unsigned long err_pkts;
  67         volatile unsigned long tx_bytes;
  68 };
  69
  70 struct pmd_internals {
  71         unsigned nb_queues;
  72
  73         int if_index;
  74         char *if_name;
  75         struct ether_addr eth_addr;
  76
  77         struct tpacket_req req;
  78
  79         struct pkt_rx_queue rx_queue[RTE_PMD_AF_PACKET_MAX_RINGS];
  80         struct pkt_tx_queue tx_queue[RTE_PMD_AF_PACKET_MAX_RINGS];
  81 };
  82
  83 static const char *valid_arguments[] = {
  84         ETH_AF_PACKET_IFACE_ARG,
  85         ETH_AF_PACKET_NUM_Q_ARG,
  86         ETH_AF_PACKET_BLOCKSIZE_ARG,
  87         ETH_AF_PACKET_FRAMESIZE_ARG,
  88         ETH_AF_PACKET_FRAMECOUNT_ARG,
  89         ETH_AF_PACKET_QDISC_BYPASS_ARG,
  90         NULL
  91 };
  92
  93 static struct rte_eth_link pmd_link = {
  94         .link_speed = ETH_SPEED_NUM_10G,
  95         .link_duplex = ETH_LINK_FULL_DUPLEX,
  96         .link_status = ETH_LINK_DOWN,
  97         .link_autoneg = ETH_LINK_FIXED,
  98 };
  99
 100 static int af_packet_logtype;
 101
 102 #define PMD_LOG(level, fmt, args...) \
 103         rte_log(RTE_LOG_ ## level, af_packet_logtype, \
 104                 "%s(): " fmt "\n", __func__, ##args)
 105
 106 static uint16_t
 107 eth_af_packet_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 108 {
 109         unsigned i;
 110         struct tpacket2_hdr *ppd;
 111         struct rte_mbuf *mbuf;
 112         uint8_t *pbuf;
 113         struct pkt_rx_queue *pkt_q = queue;
 114         uint16_t num_rx = 0;
 115         unsigned long num_rx_bytes = 0;
 116         unsigned int framecount, framenum;
 117
 118         if (unlikely(nb_pkts == 0))
 119                 return 0;
 120
 121         /*
 122          * Reads the given number of packets from the AF_PACKET socket one by
 123          * one and copies the packet data into a newly allocated mbuf.
 124          */
 125         framecount = pkt_q->framecount;
 126         framenum = pkt_q->framenum;
 127         for (i = 0; i < nb_pkts; i++) {
 128                 /* point at the next incoming frame */
 129                 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
 130                 if ((ppd->tp_status & TP_STATUS_USER) == 0)
 131                         break;
 132
 133                 /* allocate the next mbuf */
 134                 mbuf = rte_pktmbuf_alloc(pkt_q->mb_pool);
 135                 if (unlikely(mbuf == NULL))
 136                         break;
 137
 138                 /* packet will fit in the mbuf, go ahead and receive it */
 139                 rte_pktmbuf_pkt_len(mbuf) = rte_pktmbuf_data_len(mbuf) = ppd->tp_snaplen;
 140                 pbuf = (uint8_t *) ppd + ppd->tp_mac;
 141                 memcpy(rte_pktmbuf_mtod(mbuf, void *), pbuf, rte_pktmbuf_data_len(mbuf));
 142
 143                 /* check for vlan info */
 144                 if (ppd->tp_status & TP_STATUS_VLAN_VALID) {
 145                         mbuf->vlan_tci = ppd->tp_vlan_tci;
 146                         mbuf->ol_flags |= (PKT_RX_VLAN | PKT_RX_VLAN_STRIPPED);
 147                 }
 148
 149                 /* release incoming frame and advance ring buffer */
 150                 ppd->tp_status = TP_STATUS_KERNEL;
 151                 if (++framenum >= framecount)
 152                         framenum = 0;
 153                 mbuf->port = pkt_q->in_port;
 154
 155                 /* account for the receive frame */
 156                 bufs[i] = mbuf;
 157                 num_rx++;
 158                 num_rx_bytes += mbuf->pkt_len;
 159         }
 160         pkt_q->framenum = framenum;
 161         pkt_q->rx_pkts += num_rx;
 162         pkt_q->rx_bytes += num_rx_bytes;
 163         return num_rx;
 164 }
 165
 166 /*
 167  * Callback to handle sending packets through a real NIC.
 168  */
 169 static uint16_t
 170 eth_af_packet_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 171 {
 172         struct tpacket2_hdr *ppd;
 173         struct rte_mbuf *mbuf;
 174         uint8_t *pbuf;
 175         unsigned int framecount, framenum;
 176         struct pollfd pfd;
 177         struct pkt_tx_queue *pkt_q = queue;
 178         uint16_t num_tx = 0;
 179         unsigned long num_tx_bytes = 0;
 180         int i;
 181
 182         if (unlikely(nb_pkts == 0))
 183                 return 0;
 184
 185         memset(&pfd, 0, sizeof(pfd));
 186         pfd.fd = pkt_q->sockfd;
 187         pfd.events = POLLOUT;
 188         pfd.revents = 0;
 189
 190         framecount = pkt_q->framecount;
 191         framenum = pkt_q->framenum;
 192         ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
 193         for (i = 0; i < nb_pkts; i++) {
 194                 mbuf = *bufs++;
 195
 196                 /* drop oversized packets */
 197                 if (mbuf->pkt_len > pkt_q->frame_data_size) {
 198                         rte_pktmbuf_free(mbuf);
 199                         continue;
 200                 }
 201
 202                 /* insert vlan info if necessary */
 203                 if (mbuf->ol_flags & PKT_TX_VLAN_PKT) {
 204                         if (rte_vlan_insert(&mbuf)) {
 205                                 rte_pktmbuf_free(mbuf);
 206                                 continue;
 207                         }
 208                 }
 209
 210                 /* point at the next incoming frame */
 211                 if ((ppd->tp_status != TP_STATUS_AVAILABLE) &&
 212                     (poll(&pfd, 1, -1) < 0))
 213                         break;
 214
 215                 /* copy the tx frame data */
 216                 pbuf = (uint8_t *) ppd + TPACKET2_HDRLEN -
 217                         sizeof(struct sockaddr_ll);
 218
 219                 struct rte_mbuf *tmp_mbuf = mbuf;
 220                 while (tmp_mbuf) {
 221                         uint16_t data_len = rte_pktmbuf_data_len(tmp_mbuf);
 222                         memcpy(pbuf, rte_pktmbuf_mtod(tmp_mbuf, void*), data_len);
 223                         pbuf += data_len;
 224                         tmp_mbuf = tmp_mbuf->next;
 225                 }
 226
 227                 ppd->tp_len = mbuf->pkt_len;
 228                 ppd->tp_snaplen = mbuf->pkt_len;
 229
 230                 /* release incoming frame and advance ring buffer */
 231                 ppd->tp_status = TP_STATUS_SEND_REQUEST;
 232                 if (++framenum >= framecount)
 233                         framenum = 0;
 234                 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
 235
 236                 num_tx++;
 237                 num_tx_bytes += mbuf->pkt_len;
 238                 rte_pktmbuf_free(mbuf);
 239         }
 240
 241         /* kick-off transmits */
 242         if (sendto(pkt_q->sockfd, NULL, 0, MSG_DONTWAIT, NULL, 0) == -1) {
 243                 /* error sending -- no packets transmitted */
 244                 num_tx = 0;
 245                 num_tx_bytes = 0;
 246         }
 247
 248         pkt_q->framenum = framenum;
 249         pkt_q->tx_pkts += num_tx;
 250         pkt_q->err_pkts += i - num_tx;
 251         pkt_q->tx_bytes += num_tx_bytes;
 252         return i;
 253 }
 254
 255 static int
 256 eth_dev_start(struct rte_eth_dev *dev)
 257 {
 258         dev->data->dev_link.link_status = ETH_LINK_UP;
 259         return 0;
 260 }
 261
 262 /*
 263  * This function gets called when the current port gets stopped.
 264  */
 265 static void
 266 eth_dev_stop(struct rte_eth_dev *dev)
 267 {
 268         unsigned i;
 269         int sockfd;
 270         struct pmd_internals *internals = dev->data->dev_private;
 271
 272         for (i = 0; i < internals->nb_queues; i++) {
 273                 sockfd = internals->rx_queue[i].sockfd;
 274                 if (sockfd != -1)
 275                         close(sockfd);
 276
 277                 /* Prevent use after free in case tx fd == rx fd */
 278                 if (sockfd != internals->tx_queue[i].sockfd) {
 279                         sockfd = internals->tx_queue[i].sockfd;
 280                         if (sockfd != -1)
 281                                 close(sockfd);
 282                 }
 283
 284                 internals->rx_queue[i].sockfd = -1;
 285                 internals->tx_queue[i].sockfd = -1;
 286         }
 287
 288         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 289 }
 290
 291 static int
 292 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 293 {
 294         return 0;
 295 }
 296
 297 static void
 298 eth_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 299 {
 300         struct pmd_internals *internals = dev->data->dev_private;
 301
 302         dev_info->if_index = internals->if_index;
 303         dev_info->max_mac_addrs = 1;
 304         dev_info->max_rx_pktlen = (uint32_t)ETH_FRAME_LEN;
 305         dev_info->max_rx_queues = (uint16_t)internals->nb_queues;
 306         dev_info->max_tx_queues = (uint16_t)internals->nb_queues;
 307         dev_info->min_rx_bufsize = 0;
 308 }
 309
 310 static int
 311 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *igb_stats)
 312 {
 313         unsigned i, imax;
 314         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 315         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 316         const struct pmd_internals *internal = dev->data->dev_private;
 317
 318         imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
 319                 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
 320         for (i = 0; i < imax; i++) {
 321                 igb_stats->q_ipackets[i] = internal->rx_queue[i].rx_pkts;
 322                 igb_stats->q_ibytes[i] = internal->rx_queue[i].rx_bytes;
 323                 rx_total += igb_stats->q_ipackets[i];
 324                 rx_bytes_total += igb_stats->q_ibytes[i];
 325         }
 326
 327         imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
 328                 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
 329         for (i = 0; i < imax; i++) {
 330                 igb_stats->q_opackets[i] = internal->tx_queue[i].tx_pkts;
 331                 igb_stats->q_errors[i] = internal->tx_queue[i].err_pkts;
 332                 igb_stats->q_obytes[i] = internal->tx_queue[i].tx_bytes;
 333                 tx_total += igb_stats->q_opackets[i];
 334                 tx_err_total += igb_stats->q_errors[i];
 335                 tx_bytes_total += igb_stats->q_obytes[i];
 336         }
 337
 338         igb_stats->ipackets = rx_total;
 339         igb_stats->ibytes = rx_bytes_total;
 340         igb_stats->opackets = tx_total;
 341         igb_stats->oerrors = tx_err_total;
 342         igb_stats->obytes = tx_bytes_total;
 343         return 0;
 344 }
 345
 346 static void
 347 eth_stats_reset(struct rte_eth_dev *dev)
 348 {
 349         unsigned i;
 350         struct pmd_internals *internal = dev->data->dev_private;
 351
 352         for (i = 0; i < internal->nb_queues; i++) {
 353                 internal->rx_queue[i].rx_pkts = 0;
 354                 internal->rx_queue[i].rx_bytes = 0;
 355         }
 356
 357         for (i = 0; i < internal->nb_queues; i++) {
 358                 internal->tx_queue[i].tx_pkts = 0;
 359                 internal->tx_queue[i].err_pkts = 0;
 360                 internal->tx_queue[i].tx_bytes = 0;
 361         }
 362 }
 363
 364 static void
 365 eth_dev_close(struct rte_eth_dev *dev __rte_unused)
 366 {
 367 }
 368
 369 static void
 370 eth_queue_release(void *q __rte_unused)
 371 {
 372 }
 373
 374 static int
 375 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 376                 int wait_to_complete __rte_unused)
 377 {
 378         return 0;
 379 }
 380
 381 static int
 382 eth_rx_queue_setup(struct rte_eth_dev *dev,
 383                    uint16_t rx_queue_id,
 384                    uint16_t nb_rx_desc __rte_unused,
 385                    unsigned int socket_id __rte_unused,
 386                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 387                    struct rte_mempool *mb_pool)
 388 {
 389         struct pmd_internals *internals = dev->data->dev_private;
 390         struct pkt_rx_queue *pkt_q = &internals->rx_queue[rx_queue_id];
 391         unsigned int buf_size, data_size;
 392
 393         pkt_q->mb_pool = mb_pool;
 394
 395         /* Now get the space available for data in the mbuf */
 396         buf_size = rte_pktmbuf_data_room_size(pkt_q->mb_pool) -
 397                 RTE_PKTMBUF_HEADROOM;
 398         data_size = internals->req.tp_frame_size;
 399         data_size -= TPACKET2_HDRLEN - sizeof(struct sockaddr_ll);
 400
 401         if (data_size > buf_size) {
 402                 PMD_LOG(ERR,
 403                         "%s: %d bytes will not fit in mbuf (%d bytes)",
 404                         dev->device->name, data_size, buf_size);
 405                 return -ENOMEM;
 406         }
 407
 408         dev->data->rx_queues[rx_queue_id] = pkt_q;
 409         pkt_q->in_port = dev->data->port_id;
 410
 411         return 0;
 412 }
 413
 414 static int
 415 eth_tx_queue_setup(struct rte_eth_dev *dev,
 416                    uint16_t tx_queue_id,
 417                    uint16_t nb_tx_desc __rte_unused,
 418                    unsigned int socket_id __rte_unused,
 419                    const struct rte_eth_txconf *tx_conf __rte_unused)
 420 {
 421
 422         struct pmd_internals *internals = dev->data->dev_private;
 423
 424         dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
 425         return 0;
 426 }
 427
 428 static int
 429 eth_dev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 430 {
 431         struct pmd_internals *internals = dev->data->dev_private;
 432         struct ifreq ifr = { .ifr_mtu = mtu };
 433         int ret;
 434         int s;
 435         unsigned int data_size = internals->req.tp_frame_size -
 436                                  TPACKET2_HDRLEN;
 437
 438         if (mtu > data_size)
 439                 return -EINVAL;
 440
 441         s = socket(PF_INET, SOCK_DGRAM, 0);
 442         if (s < 0)
 443                 return -EINVAL;
 444
 445         snprintf(ifr.ifr_name, IFNAMSIZ, "%s", internals->if_name);
 446         ret = ioctl(s, SIOCSIFMTU, &ifr);
 447         close(s);
 448
 449         if (ret < 0)
 450                 return -EINVAL;
 451
 452         return 0;
 453 }
 454
 455 static void
 456 eth_dev_change_flags(char *if_name, uint32_t flags, uint32_t mask)
 457 {
 458         struct ifreq ifr;
 459         int s;
 460
 461         s = socket(PF_INET, SOCK_DGRAM, 0);
 462         if (s < 0)
 463                 return;
 464
 465         snprintf(ifr.ifr_name, IFNAMSIZ, "%s", if_name);
 466         if (ioctl(s, SIOCGIFFLAGS, &ifr) < 0)
 467                 goto out;
 468         ifr.ifr_flags &= mask;
 469         ifr.ifr_flags |= flags;
 470         if (ioctl(s, SIOCSIFFLAGS, &ifr) < 0)
 471                 goto out;
 472 out:
 473         close(s);
 474 }
 475
 476 static void
 477 eth_dev_promiscuous_enable(struct rte_eth_dev *dev)
 478 {
 479         struct pmd_internals *internals = dev->data->dev_private;
 480
 481         eth_dev_change_flags(internals->if_name, IFF_PROMISC, ~0);
 482 }
 483
 484 static void
 485 eth_dev_promiscuous_disable(struct rte_eth_dev *dev)
 486 {
 487         struct pmd_internals *internals = dev->data->dev_private;
 488
 489         eth_dev_change_flags(internals->if_name, 0, ~IFF_PROMISC);
 490 }
 491
 492 static const struct eth_dev_ops ops = {
 493         .dev_start = eth_dev_start,
 494         .dev_stop = eth_dev_stop,
 495         .dev_close = eth_dev_close,
 496         .dev_configure = eth_dev_configure,
 497         .dev_infos_get = eth_dev_info,
 498         .mtu_set = eth_dev_mtu_set,
 499         .promiscuous_enable = eth_dev_promiscuous_enable,
 500         .promiscuous_disable = eth_dev_promiscuous_disable,
 501         .rx_queue_setup = eth_rx_queue_setup,
 502         .tx_queue_setup = eth_tx_queue_setup,
 503         .rx_queue_release = eth_queue_release,
 504         .tx_queue_release = eth_queue_release,
 505         .link_update = eth_link_update,
 506         .stats_get = eth_stats_get,
 507         .stats_reset = eth_stats_reset,
 508 };
 509
 510 /*
 511  * Opens an AF_PACKET socket
 512  */
 513 static int
 514 open_packet_iface(const char *key __rte_unused,
 515                   const char *value __rte_unused,
 516                   void *extra_args)
 517 {
 518         int *sockfd = extra_args;
 519
 520         /* Open an AF_PACKET socket... */
 521         *sockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
 522         if (*sockfd == -1) {
 523                 PMD_LOG(ERR, "Could not open AF_PACKET socket");
 524                 return -1;
 525         }
 526
 527         return 0;
 528 }
 529
 530 static struct rte_vdev_driver pmd_af_packet_drv;
 531
 532 static int
 533 rte_pmd_init_internals(struct rte_vdev_device *dev,
 534                        const int sockfd,
 535                        const unsigned nb_queues,
 536                        unsigned int blocksize,
 537                        unsigned int blockcnt,
 538                        unsigned int framesize,
 539                        unsigned int framecnt,
 540                        unsigned int qdisc_bypass,
 541                        struct pmd_internals **internals,
 542                        struct rte_eth_dev **eth_dev,
 543                        struct rte_kvargs *kvlist)
 544 {
 545         const char *name = rte_vdev_device_name(dev);
 546         const unsigned int numa_node = dev->device.numa_node;
 547         struct rte_eth_dev_data *data = NULL;
 548         struct rte_kvargs_pair *pair = NULL;
 549         struct ifreq ifr;
 550         size_t ifnamelen;
 551         unsigned k_idx;
 552         struct sockaddr_ll sockaddr;
 553         struct tpacket_req *req;
 554         struct pkt_rx_queue *rx_queue;
 555         struct pkt_tx_queue *tx_queue;
 556         int rc, tpver, discard;
 557         int qsockfd = -1;
 558         unsigned int i, q, rdsize;
 559 #if defined(PACKET_FANOUT)
 560         int fanout_arg;
 561 #endif
 562
 563         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 564                 pair = &kvlist->pairs[k_idx];
 565                 if (strstr(pair->key, ETH_AF_PACKET_IFACE_ARG) != NULL)
 566                         break;
 567         }
 568         if (pair == NULL) {
 569                 PMD_LOG(ERR,
 570                         "%s: no interface specified for AF_PACKET ethdev",
 571                         name);
 572                 return -1;
 573         }
 574
 575         PMD_LOG(INFO,
 576                 "%s: creating AF_PACKET-backed ethdev on numa socket %u",
 577                 name, numa_node);
 578
 579         *internals = rte_zmalloc_socket(name, sizeof(**internals),
 580                                         0, numa_node);
 581         if (*internals == NULL)
 582                 return -1;
 583
 584         for (q = 0; q < nb_queues; q++) {
 585                 (*internals)->rx_queue[q].map = MAP_FAILED;
 586                 (*internals)->tx_queue[q].map = MAP_FAILED;
 587         }
 588
 589         req = &((*internals)->req);
 590
 591         req->tp_block_size = blocksize;
 592         req->tp_block_nr = blockcnt;
 593         req->tp_frame_size = framesize;
 594         req->tp_frame_nr = framecnt;
 595
 596         ifnamelen = strlen(pair->value);
 597         if (ifnamelen < sizeof(ifr.ifr_name)) {
 598                 memcpy(ifr.ifr_name, pair->value, ifnamelen);
 599                 ifr.ifr_name[ifnamelen] = '\0';
 600         } else {
 601                 PMD_LOG(ERR,
 602                         "%s: I/F name too long (%s)",
 603                         name, pair->value);
 604                 return -1;
 605         }
 606         if (ioctl(sockfd, SIOCGIFINDEX, &ifr) == -1) {
 607                 PMD_LOG(ERR,
 608                         "%s: ioctl failed (SIOCGIFINDEX)",
 609                         name);
 610                 return -1;
 611         }
 612         (*internals)->if_name = strdup(pair->value);
 613         if ((*internals)->if_name == NULL)
 614                 return -1;
 615         (*internals)->if_index = ifr.ifr_ifindex;
 616
 617         if (ioctl(sockfd, SIOCGIFHWADDR, &ifr) == -1) {
 618                 PMD_LOG(ERR,
 619                         "%s: ioctl failed (SIOCGIFHWADDR)",
 620                         name);
 621                 return -1;
 622         }
 623         memcpy(&(*internals)->eth_addr, ifr.ifr_hwaddr.sa_data, ETH_ALEN);
 624
 625         memset(&sockaddr, 0, sizeof(sockaddr));
 626         sockaddr.sll_family = AF_PACKET;
 627         sockaddr.sll_protocol = htons(ETH_P_ALL);
 628         sockaddr.sll_ifindex = (*internals)->if_index;
 629
 630 #if defined(PACKET_FANOUT)
 631         fanout_arg = (getpid() ^ (*internals)->if_index) & 0xffff;
 632         fanout_arg |= (PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_DEFRAG) << 16;
 633 #if defined(PACKET_FANOUT_FLAG_ROLLOVER)
 634         fanout_arg |= PACKET_FANOUT_FLAG_ROLLOVER << 16;
 635 #endif
 636 #endif
 637
 638         for (q = 0; q < nb_queues; q++) {
 639                 /* Open an AF_PACKET socket for this queue... */
 640                 qsockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
 641                 if (qsockfd == -1) {
 642                         PMD_LOG(ERR,
 643                                 "%s: could not open AF_PACKET socket",
 644                                 name);
 645                         return -1;
 646                 }
 647
 648                 tpver = TPACKET_V2;
 649                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_VERSION,
 650                                 &tpver, sizeof(tpver));
 651                 if (rc == -1) {
 652                         PMD_LOG(ERR,
 653                                 "%s: could not set PACKET_VERSION on AF_PACKET socket for %s",
 654                                 name, pair->value);
 655                         goto error;
 656                 }
 657
 658                 discard = 1;
 659                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_LOSS,
 660                                 &discard, sizeof(discard));
 661                 if (rc == -1) {
 662                         PMD_LOG(ERR,
 663                                 "%s: could not set PACKET_LOSS on AF_PACKET socket for %s",
 664                                 name, pair->value);
 665                         goto error;
 666                 }
 667
 668 #if defined(PACKET_QDISC_BYPASS)
 669                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_QDISC_BYPASS,
 670                                 &qdisc_bypass, sizeof(qdisc_bypass));
 671                 if (rc == -1) {
 672                         PMD_LOG(ERR,
 673                                 "%s: could not set PACKET_QDISC_BYPASS on AF_PACKET socket for %s",
 674                                 name, pair->value);
 675                         goto error;
 676                 }
 677 #else
 678                 RTE_SET_USED(qdisc_bypass);
 679 #endif
 680
 681                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_RX_RING, req, sizeof(*req));
 682                 if (rc == -1) {
 683                         PMD_LOG(ERR,
 684                                 "%s: could not set PACKET_RX_RING on AF_PACKET socket for %s",
 685                                 name, pair->value);
 686                         goto error;
 687                 }
 688
 689                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_TX_RING, req, sizeof(*req));
 690                 if (rc == -1) {
 691                         PMD_LOG(ERR,
 692                                 "%s: could not set PACKET_TX_RING on AF_PACKET "
 693                                 "socket for %s", name, pair->value);
 694                         goto error;
 695                 }
 696
 697                 rx_queue = &((*internals)->rx_queue[q]);
 698                 rx_queue->framecount = req->tp_frame_nr;
 699
 700                 rx_queue->map = mmap(NULL, 2 * req->tp_block_size * req->tp_block_nr,
 701                                     PROT_READ | PROT_WRITE, MAP_SHARED | MAP_LOCKED,
 702                                     qsockfd, 0);
 703                 if (rx_queue->map == MAP_FAILED) {
 704                         PMD_LOG(ERR,
 705                                 "%s: call to mmap failed on AF_PACKET socket for %s",
 706                                 name, pair->value);
 707                         goto error;
 708                 }
 709
 710                 /* rdsize is same for both Tx and Rx */
 711                 rdsize = req->tp_frame_nr * sizeof(*(rx_queue->rd));
 712
 713                 rx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
 714                 if (rx_queue->rd == NULL)
 715                         goto error;
 716                 for (i = 0; i < req->tp_frame_nr; ++i) {
 717                         rx_queue->rd[i].iov_base = rx_queue->map + (i * framesize);
 718                         rx_queue->rd[i].iov_len = req->tp_frame_size;
 719                 }
 720                 rx_queue->sockfd = qsockfd;
 721
 722                 tx_queue = &((*internals)->tx_queue[q]);
 723                 tx_queue->framecount = req->tp_frame_nr;
 724                 tx_queue->frame_data_size = req->tp_frame_size;
 725                 tx_queue->frame_data_size -= TPACKET2_HDRLEN -
 726                         sizeof(struct sockaddr_ll);
 727
 728                 tx_queue->map = rx_queue->map + req->tp_block_size * req->tp_block_nr;
 729
 730                 tx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
 731                 if (tx_queue->rd == NULL)
 732                         goto error;
 733                 for (i = 0; i < req->tp_frame_nr; ++i) {
 734                         tx_queue->rd[i].iov_base = tx_queue->map + (i * framesize);
 735                         tx_queue->rd[i].iov_len = req->tp_frame_size;
 736                 }
 737                 tx_queue->sockfd = qsockfd;
 738
 739                 rc = bind(qsockfd, (const struct sockaddr*)&sockaddr, sizeof(sockaddr));
 740                 if (rc == -1) {
 741                         PMD_LOG(ERR,
 742                                 "%s: could not bind AF_PACKET socket to %s",
 743                                 name, pair->value);
 744                         goto error;
 745                 }
 746
 747 #if defined(PACKET_FANOUT)
 748                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_FANOUT,
 749                                 &fanout_arg, sizeof(fanout_arg));
 750                 if (rc == -1) {
 751                         PMD_LOG(ERR,
 752                                 "%s: could not set PACKET_FANOUT on AF_PACKET socket "
 753                                 "for %s", name, pair->value);
 754                         goto error;
 755                 }
 756 #endif
 757         }
 758
 759         /* reserve an ethdev entry */
 760         *eth_dev = rte_eth_vdev_allocate(dev, 0);
 761         if (*eth_dev == NULL)
 762                 goto error;
 763
 764         /*
 765          * now put it all together
 766          * - store queue data in internals,
 767          * - store numa_node in eth_dev
 768          * - point eth_dev_data to internals
 769          * - and point eth_dev structure to new eth_dev_data structure
 770          */
 771
 772         (*internals)->nb_queues = nb_queues;
 773
 774         data = (*eth_dev)->data;
 775         data->dev_private = *internals;
 776         data->nb_rx_queues = (uint16_t)nb_queues;
 777         data->nb_tx_queues = (uint16_t)nb_queues;
 778         data->dev_link = pmd_link;
 779         data->mac_addrs = &(*internals)->eth_addr;
 780
 781         (*eth_dev)->dev_ops = &ops;
 782
 783         return 0;
 784
 785 error:
 786         if (qsockfd != -1)
 787                 close(qsockfd);
 788         for (q = 0; q < nb_queues; q++) {
 789                 munmap((*internals)->rx_queue[q].map,
 790                        2 * req->tp_block_size * req->tp_block_nr);
 791
 792                 rte_free((*internals)->rx_queue[q].rd);
 793                 rte_free((*internals)->tx_queue[q].rd);
 794                 if (((*internals)->rx_queue[q].sockfd != 0) &&
 795                         ((*internals)->rx_queue[q].sockfd != qsockfd))
 796                         close((*internals)->rx_queue[q].sockfd);
 797         }
 798         free((*internals)->if_name);
 799         rte_free(*internals);
 800         return -1;
 801 }
 802
 803 static int
 804 rte_eth_from_packet(struct rte_vdev_device *dev,
 805                     int const *sockfd,
 806                     struct rte_kvargs *kvlist)
 807 {
 808         const char *name = rte_vdev_device_name(dev);
 809         struct pmd_internals *internals = NULL;
 810         struct rte_eth_dev *eth_dev = NULL;
 811         struct rte_kvargs_pair *pair = NULL;
 812         unsigned k_idx;
 813         unsigned int blockcount;
 814         unsigned int blocksize = DFLT_BLOCK_SIZE;
 815         unsigned int framesize = DFLT_FRAME_SIZE;
 816         unsigned int framecount = DFLT_FRAME_COUNT;
 817         unsigned int qpairs = 1;
 818         unsigned int qdisc_bypass = 1;
 819
 820         /* do some parameter checking */
 821         if (*sockfd < 0)
 822                 return -1;
 823
 824         /*
 825          * Walk arguments for configurable settings
 826          */
 827         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 828                 pair = &kvlist->pairs[k_idx];
 829                 if (strstr(pair->key, ETH_AF_PACKET_NUM_Q_ARG) != NULL) {
 830                         qpairs = atoi(pair->value);
 831                         if (qpairs < 1 ||
 832                             qpairs > RTE_PMD_AF_PACKET_MAX_RINGS) {
 833                                 PMD_LOG(ERR,
 834                                         "%s: invalid qpairs value",
 835                                         name);
 836                                 return -1;
 837                         }
 838                         continue;
 839                 }
 840                 if (strstr(pair->key, ETH_AF_PACKET_BLOCKSIZE_ARG) != NULL) {
 841                         blocksize = atoi(pair->value);
 842                         if (!blocksize) {
 843                                 PMD_LOG(ERR,
 844                                         "%s: invalid blocksize value",
 845                                         name);
 846                                 return -1;
 847                         }
 848                         continue;
 849                 }
 850                 if (strstr(pair->key, ETH_AF_PACKET_FRAMESIZE_ARG) != NULL) {
 851                         framesize = atoi(pair->value);
 852                         if (!framesize) {
 853                                 PMD_LOG(ERR,
 854                                         "%s: invalid framesize value",
 855                                         name);
 856                                 return -1;
 857                         }
 858                         continue;
 859                 }
 860                 if (strstr(pair->key, ETH_AF_PACKET_FRAMECOUNT_ARG) != NULL) {
 861                         framecount = atoi(pair->value);
 862                         if (!framecount) {
 863                                 PMD_LOG(ERR,
 864                                         "%s: invalid framecount value",
 865                                         name);
 866                                 return -1;
 867                         }
 868                         continue;
 869                 }
 870                 if (strstr(pair->key, ETH_AF_PACKET_QDISC_BYPASS_ARG) != NULL) {
 871                         qdisc_bypass = atoi(pair->value);
 872                         if (qdisc_bypass > 1) {
 873                                 PMD_LOG(ERR,
 874                                         "%s: invalid bypass value",
 875                                         name);
 876                                 return -1;
 877                         }
 878                         continue;
 879                 }
 880         }
 881
 882         if (framesize > blocksize) {
 883                 PMD_LOG(ERR,
 884                         "%s: AF_PACKET MMAP frame size exceeds block size!",
 885                         name);
 886                 return -1;
 887         }
 888
 889         blockcount = framecount / (blocksize / framesize);
 890         if (!blockcount) {
 891                 PMD_LOG(ERR,
 892                         "%s: invalid AF_PACKET MMAP parameters", name);
 893                 return -1;
 894         }
 895
 896         PMD_LOG(INFO, "%s: AF_PACKET MMAP parameters:", name);
 897         PMD_LOG(INFO, "%s:\tblock size %d", name, blocksize);
 898         PMD_LOG(INFO, "%s:\tblock count %d", name, blockcount);
 899         PMD_LOG(INFO, "%s:\tframe size %d", name, framesize);
 900         PMD_LOG(INFO, "%s:\tframe count %d", name, framecount);
 901
 902         if (rte_pmd_init_internals(dev, *sockfd, qpairs,
 903                                    blocksize, blockcount,
 904                                    framesize, framecount,
 905                                    qdisc_bypass,
 906                                    &internals, &eth_dev,
 907                                    kvlist) < 0)
 908                 return -1;
 909
 910         eth_dev->rx_pkt_burst = eth_af_packet_rx;
 911         eth_dev->tx_pkt_burst = eth_af_packet_tx;
 912
 913         rte_eth_dev_probing_finish(eth_dev);
 914         return 0;
 915 }
 916
 917 static int
 918 rte_pmd_af_packet_probe(struct rte_vdev_device *dev)
 919 {
 920         int ret = 0;
 921         struct rte_kvargs *kvlist;
 922         int sockfd = -1;
 923         struct rte_eth_dev *eth_dev;
 924         const char *name = rte_vdev_device_name(dev);
 925
 926         PMD_LOG(INFO, "Initializing pmd_af_packet for %s", name);
 927
 928         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
 929                 eth_dev = rte_eth_dev_attach_secondary(name);
 930                 if (!eth_dev) {
 931                         PMD_LOG(ERR, "Failed to probe %s", name);
 932                         return -1;
 933                 }
 934                 /* TODO: request info from primary to set up Rx and Tx */
 935                 eth_dev->dev_ops = &ops;
 936                 eth_dev->device = &dev->device;
 937                 rte_eth_dev_probing_finish(eth_dev);
 938                 return 0;
 939         }
 940
 941         kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_arguments);
 942         if (kvlist == NULL) {
 943                 ret = -1;
 944                 goto exit;
 945         }
 946
 947         /*
 948          * If iface argument is passed we open the NICs and use them for
 949          * reading / writing
 950          */
 951         if (rte_kvargs_count(kvlist, ETH_AF_PACKET_IFACE_ARG) == 1) {
 952
 953                 ret = rte_kvargs_process(kvlist, ETH_AF_PACKET_IFACE_ARG,
 954                                          &open_packet_iface, &sockfd);
 955                 if (ret < 0)
 956                         goto exit;
 957         }
 958
 959         if (dev->device.numa_node == SOCKET_ID_ANY)
 960                 dev->device.numa_node = rte_socket_id();
 961
 962         ret = rte_eth_from_packet(dev, &sockfd, kvlist);
 963         close(sockfd); /* no longer needed */
 964
 965 exit:
 966         rte_kvargs_free(kvlist);
 967         return ret;
 968 }
 969
 970 static int
 971 rte_pmd_af_packet_remove(struct rte_vdev_device *dev)
 972 {
 973         struct rte_eth_dev *eth_dev = NULL;
 974         struct pmd_internals *internals;
 975         unsigned q;
 976
 977         PMD_LOG(INFO, "Closing AF_PACKET ethdev on numa socket %u",
 978                 rte_socket_id());
 979
 980         if (dev == NULL)
 981                 return -1;
 982
 983         /* find the ethdev entry */
 984         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
 985         if (eth_dev == NULL)
 986                 return -1;
 987
 988         /* mac_addrs must not be freed alone because part of dev_private */
 989         eth_dev->data->mac_addrs = NULL;
 990
 991         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 992                 return rte_eth_dev_release_port(eth_dev);
 993
 994         internals = eth_dev->data->dev_private;
 995         for (q = 0; q < internals->nb_queues; q++) {
 996                 rte_free(internals->rx_queue[q].rd);
 997                 rte_free(internals->tx_queue[q].rd);
 998         }
 999         free(internals->if_name);
1000
1001         rte_eth_dev_release_port(eth_dev);
1002
1003         return 0;
1004 }
1005
1006 static struct rte_vdev_driver pmd_af_packet_drv = {
1007         .probe = rte_pmd_af_packet_probe,
1008         .remove = rte_pmd_af_packet_remove,
1009 };
1010
1011 RTE_PMD_REGISTER_VDEV(net_af_packet, pmd_af_packet_drv);
1012 RTE_PMD_REGISTER_ALIAS(net_af_packet, eth_af_packet);
1013 RTE_PMD_REGISTER_PARAM_STRING(net_af_packet,
1014         "iface=<string> "
1015         "qpairs=<int> "
1016         "blocksz=<int> "
1017         "framesz=<int> "
1018         "framecnt=<int> "
1019         "qdisc_bypass=<0|1>");
1020
1021 RTE_INIT(af_packet_init_log)
1022 {
1023         af_packet_logtype = rte_log_register("pmd.net.packet");
1024         if (af_packet_logtype >= 0)
1025                 rte_log_set_level(af_packet_logtype, RTE_LOG_NOTICE);
1026 }