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 <net/if_arp.h>
  23 #include <sys/types.h>
  24 #include <sys/socket.h>
  25 #include <sys/ioctl.h>
  26 #include <string.h>
  27 #include <sys/mman.h>
  28 #include <unistd.h>
  29 #include <poll.h>
  30
  31 #define ETH_AF_PACKET_IFACE_ARG         "iface"
  32 #define ETH_AF_PACKET_NUM_Q_ARG         "qpairs"
  33 #define ETH_AF_PACKET_BLOCKSIZE_ARG     "blocksz"
  34 #define ETH_AF_PACKET_FRAMESIZE_ARG     "framesz"
  35 #define ETH_AF_PACKET_FRAMECOUNT_ARG    "framecnt"
  36 #define ETH_AF_PACKET_QDISC_BYPASS_ARG  "qdisc_bypass"
  37
  38 #define DFLT_FRAME_SIZE         (1 << 11)
  39 #define DFLT_FRAME_COUNT        (1 << 9)
  40
  41 struct pkt_rx_queue {
  42         int sockfd;
  43
  44         struct iovec *rd;
  45         uint8_t *map;
  46         unsigned int framecount;
  47         unsigned int framenum;
  48
  49         struct rte_mempool *mb_pool;
  50         uint16_t in_port;
  51
  52         volatile unsigned long rx_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 rte_ether_addr eth_addr;
  76
  77         struct tpacket_req req;
  78
  79         struct pkt_rx_queue *rx_queue;
  80         struct pkt_tx_queue *tx_queue;
  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 RTE_LOG_REGISTER(af_packet_logtype, pmd.net.packet, NOTICE);
 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 #define PMD_LOG_ERRNO(level, fmt, args...) \
 107         rte_log(RTE_LOG_ ## level, af_packet_logtype, \
 108                 "%s(): " fmt ":%s\n", __func__, ##args, strerror(errno))
 109
 110 static uint16_t
 111 eth_af_packet_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 112 {
 113         unsigned i;
 114         struct tpacket2_hdr *ppd;
 115         struct rte_mbuf *mbuf;
 116         uint8_t *pbuf;
 117         struct pkt_rx_queue *pkt_q = queue;
 118         uint16_t num_rx = 0;
 119         unsigned long num_rx_bytes = 0;
 120         unsigned int framecount, framenum;
 121
 122         if (unlikely(nb_pkts == 0))
 123                 return 0;
 124
 125         /*
 126          * Reads the given number of packets from the AF_PACKET socket one by
 127          * one and copies the packet data into a newly allocated mbuf.
 128          */
 129         framecount = pkt_q->framecount;
 130         framenum = pkt_q->framenum;
 131         for (i = 0; i < nb_pkts; i++) {
 132                 /* point at the next incoming frame */
 133                 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
 134                 if ((ppd->tp_status & TP_STATUS_USER) == 0)
 135                         break;
 136
 137                 /* allocate the next mbuf */
 138                 mbuf = rte_pktmbuf_alloc(pkt_q->mb_pool);
 139                 if (unlikely(mbuf == NULL))
 140                         break;
 141
 142                 /* packet will fit in the mbuf, go ahead and receive it */
 143                 rte_pktmbuf_pkt_len(mbuf) = rte_pktmbuf_data_len(mbuf) = ppd->tp_snaplen;
 144                 pbuf = (uint8_t *) ppd + ppd->tp_mac;
 145                 memcpy(rte_pktmbuf_mtod(mbuf, void *), pbuf, rte_pktmbuf_data_len(mbuf));
 146
 147                 /* check for vlan info */
 148                 if (ppd->tp_status & TP_STATUS_VLAN_VALID) {
 149                         mbuf->vlan_tci = ppd->tp_vlan_tci;
 150                         mbuf->ol_flags |= (PKT_RX_VLAN | PKT_RX_VLAN_STRIPPED);
 151                 }
 152
 153                 /* release incoming frame and advance ring buffer */
 154                 ppd->tp_status = TP_STATUS_KERNEL;
 155                 if (++framenum >= framecount)
 156                         framenum = 0;
 157                 mbuf->port = pkt_q->in_port;
 158
 159                 /* account for the receive frame */
 160                 bufs[i] = mbuf;
 161                 num_rx++;
 162                 num_rx_bytes += mbuf->pkt_len;
 163         }
 164         pkt_q->framenum = framenum;
 165         pkt_q->rx_pkts += num_rx;
 166         pkt_q->rx_bytes += num_rx_bytes;
 167         return num_rx;
 168 }
 169
 170 /*
 171  * Callback to handle sending packets through a real NIC.
 172  */
 173 static uint16_t
 174 eth_af_packet_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 175 {
 176         struct tpacket2_hdr *ppd;
 177         struct rte_mbuf *mbuf;
 178         uint8_t *pbuf;
 179         unsigned int framecount, framenum;
 180         struct pollfd pfd;
 181         struct pkt_tx_queue *pkt_q = queue;
 182         uint16_t num_tx = 0;
 183         unsigned long num_tx_bytes = 0;
 184         int i;
 185
 186         if (unlikely(nb_pkts == 0))
 187                 return 0;
 188
 189         memset(&pfd, 0, sizeof(pfd));
 190         pfd.fd = pkt_q->sockfd;
 191         pfd.events = POLLOUT;
 192         pfd.revents = 0;
 193
 194         framecount = pkt_q->framecount;
 195         framenum = pkt_q->framenum;
 196         ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
 197         for (i = 0; i < nb_pkts; i++) {
 198                 mbuf = *bufs++;
 199
 200                 /* drop oversized packets */
 201                 if (mbuf->pkt_len > pkt_q->frame_data_size) {
 202                         rte_pktmbuf_free(mbuf);
 203                         continue;
 204                 }
 205
 206                 /* insert vlan info if necessary */
 207                 if (mbuf->ol_flags & PKT_TX_VLAN_PKT) {
 208                         if (rte_vlan_insert(&mbuf)) {
 209                                 rte_pktmbuf_free(mbuf);
 210                                 continue;
 211                         }
 212                 }
 213
 214                 /* point at the next incoming frame */
 215                 if ((ppd->tp_status != TP_STATUS_AVAILABLE) &&
 216                     (poll(&pfd, 1, -1) < 0))
 217                         break;
 218
 219                 /* copy the tx frame data */
 220                 pbuf = (uint8_t *) ppd + TPACKET2_HDRLEN -
 221                         sizeof(struct sockaddr_ll);
 222
 223                 struct rte_mbuf *tmp_mbuf = mbuf;
 224                 while (tmp_mbuf) {
 225                         uint16_t data_len = rte_pktmbuf_data_len(tmp_mbuf);
 226                         memcpy(pbuf, rte_pktmbuf_mtod(tmp_mbuf, void*), data_len);
 227                         pbuf += data_len;
 228                         tmp_mbuf = tmp_mbuf->next;
 229                 }
 230
 231                 ppd->tp_len = mbuf->pkt_len;
 232                 ppd->tp_snaplen = mbuf->pkt_len;
 233
 234                 /* release incoming frame and advance ring buffer */
 235                 ppd->tp_status = TP_STATUS_SEND_REQUEST;
 236                 if (++framenum >= framecount)
 237                         framenum = 0;
 238                 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
 239
 240                 num_tx++;
 241                 num_tx_bytes += mbuf->pkt_len;
 242                 rte_pktmbuf_free(mbuf);
 243         }
 244
 245         /* kick-off transmits */
 246         if (sendto(pkt_q->sockfd, NULL, 0, MSG_DONTWAIT, NULL, 0) == -1 &&
 247                         errno != ENOBUFS && errno != EAGAIN) {
 248                 /*
 249                  * In case of a ENOBUFS/EAGAIN error all of the enqueued
 250                  * packets will be considered successful even though only some
 251                  * are sent.
 252                  */
 253
 254                 num_tx = 0;
 255                 num_tx_bytes = 0;
 256         }
 257
 258         pkt_q->framenum = framenum;
 259         pkt_q->tx_pkts += num_tx;
 260         pkt_q->err_pkts += i - num_tx;
 261         pkt_q->tx_bytes += num_tx_bytes;
 262         return i;
 263 }
 264
 265 static int
 266 eth_dev_start(struct rte_eth_dev *dev)
 267 {
 268         dev->data->dev_link.link_status = ETH_LINK_UP;
 269         return 0;
 270 }
 271
 272 /*
 273  * This function gets called when the current port gets stopped.
 274  */
 275 static void
 276 eth_dev_stop(struct rte_eth_dev *dev)
 277 {
 278         unsigned i;
 279         int sockfd;
 280         struct pmd_internals *internals = dev->data->dev_private;
 281
 282         for (i = 0; i < internals->nb_queues; i++) {
 283                 sockfd = internals->rx_queue[i].sockfd;
 284                 if (sockfd != -1)
 285                         close(sockfd);
 286
 287                 /* Prevent use after free in case tx fd == rx fd */
 288                 if (sockfd != internals->tx_queue[i].sockfd) {
 289                         sockfd = internals->tx_queue[i].sockfd;
 290                         if (sockfd != -1)
 291                                 close(sockfd);
 292                 }
 293
 294                 internals->rx_queue[i].sockfd = -1;
 295                 internals->tx_queue[i].sockfd = -1;
 296         }
 297
 298         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 299 }
 300
 301 static int
 302 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 303 {
 304         return 0;
 305 }
 306
 307 static int
 308 eth_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 309 {
 310         struct pmd_internals *internals = dev->data->dev_private;
 311
 312         dev_info->if_index = internals->if_index;
 313         dev_info->max_mac_addrs = 1;
 314         dev_info->max_rx_pktlen = (uint32_t)ETH_FRAME_LEN;
 315         dev_info->max_rx_queues = (uint16_t)internals->nb_queues;
 316         dev_info->max_tx_queues = (uint16_t)internals->nb_queues;
 317         dev_info->min_rx_bufsize = 0;
 318         dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
 319                 DEV_TX_OFFLOAD_VLAN_INSERT;
 320
 321         return 0;
 322 }
 323
 324 static int
 325 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *igb_stats)
 326 {
 327         unsigned i, imax;
 328         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 329         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 330         const struct pmd_internals *internal = dev->data->dev_private;
 331
 332         imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
 333                 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
 334         for (i = 0; i < imax; i++) {
 335                 igb_stats->q_ipackets[i] = internal->rx_queue[i].rx_pkts;
 336                 igb_stats->q_ibytes[i] = internal->rx_queue[i].rx_bytes;
 337                 rx_total += igb_stats->q_ipackets[i];
 338                 rx_bytes_total += igb_stats->q_ibytes[i];
 339         }
 340
 341         imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
 342                 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
 343         for (i = 0; i < imax; i++) {
 344                 igb_stats->q_opackets[i] = internal->tx_queue[i].tx_pkts;
 345                 igb_stats->q_obytes[i] = internal->tx_queue[i].tx_bytes;
 346                 tx_total += igb_stats->q_opackets[i];
 347                 tx_err_total += internal->tx_queue[i].err_pkts;
 348                 tx_bytes_total += igb_stats->q_obytes[i];
 349         }
 350
 351         igb_stats->ipackets = rx_total;
 352         igb_stats->ibytes = rx_bytes_total;
 353         igb_stats->opackets = tx_total;
 354         igb_stats->oerrors = tx_err_total;
 355         igb_stats->obytes = tx_bytes_total;
 356         return 0;
 357 }
 358
 359 static int
 360 eth_stats_reset(struct rte_eth_dev *dev)
 361 {
 362         unsigned i;
 363         struct pmd_internals *internal = dev->data->dev_private;
 364
 365         for (i = 0; i < internal->nb_queues; i++) {
 366                 internal->rx_queue[i].rx_pkts = 0;
 367                 internal->rx_queue[i].rx_bytes = 0;
 368         }
 369
 370         for (i = 0; i < internal->nb_queues; i++) {
 371                 internal->tx_queue[i].tx_pkts = 0;
 372                 internal->tx_queue[i].err_pkts = 0;
 373                 internal->tx_queue[i].tx_bytes = 0;
 374         }
 375
 376         return 0;
 377 }
 378
 379 static void
 380 eth_dev_close(struct rte_eth_dev *dev __rte_unused)
 381 {
 382 }
 383
 384 static void
 385 eth_queue_release(void *q __rte_unused)
 386 {
 387 }
 388
 389 static int
 390 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 391                 int wait_to_complete __rte_unused)
 392 {
 393         return 0;
 394 }
 395
 396 static int
 397 eth_rx_queue_setup(struct rte_eth_dev *dev,
 398                    uint16_t rx_queue_id,
 399                    uint16_t nb_rx_desc __rte_unused,
 400                    unsigned int socket_id __rte_unused,
 401                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 402                    struct rte_mempool *mb_pool)
 403 {
 404         struct pmd_internals *internals = dev->data->dev_private;
 405         struct pkt_rx_queue *pkt_q = &internals->rx_queue[rx_queue_id];
 406         unsigned int buf_size, data_size;
 407
 408         pkt_q->mb_pool = mb_pool;
 409
 410         /* Now get the space available for data in the mbuf */
 411         buf_size = rte_pktmbuf_data_room_size(pkt_q->mb_pool) -
 412                 RTE_PKTMBUF_HEADROOM;
 413         data_size = internals->req.tp_frame_size;
 414         data_size -= TPACKET2_HDRLEN - sizeof(struct sockaddr_ll);
 415
 416         if (data_size > buf_size) {
 417                 PMD_LOG(ERR,
 418                         "%s: %d bytes will not fit in mbuf (%d bytes)",
 419                         dev->device->name, data_size, buf_size);
 420                 return -ENOMEM;
 421         }
 422
 423         dev->data->rx_queues[rx_queue_id] = pkt_q;
 424         pkt_q->in_port = dev->data->port_id;
 425
 426         return 0;
 427 }
 428
 429 static int
 430 eth_tx_queue_setup(struct rte_eth_dev *dev,
 431                    uint16_t tx_queue_id,
 432                    uint16_t nb_tx_desc __rte_unused,
 433                    unsigned int socket_id __rte_unused,
 434                    const struct rte_eth_txconf *tx_conf __rte_unused)
 435 {
 436
 437         struct pmd_internals *internals = dev->data->dev_private;
 438
 439         dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
 440         return 0;
 441 }
 442
 443 static int
 444 eth_dev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 445 {
 446         struct pmd_internals *internals = dev->data->dev_private;
 447         struct ifreq ifr = { .ifr_mtu = mtu };
 448         int ret;
 449         int s;
 450         unsigned int data_size = internals->req.tp_frame_size -
 451                                  TPACKET2_HDRLEN;
 452
 453         if (mtu > data_size)
 454                 return -EINVAL;
 455
 456         s = socket(PF_INET, SOCK_DGRAM, 0);
 457         if (s < 0)
 458                 return -EINVAL;
 459
 460         strlcpy(ifr.ifr_name, internals->if_name, IFNAMSIZ);
 461         ret = ioctl(s, SIOCSIFMTU, &ifr);
 462         close(s);
 463
 464         if (ret < 0)
 465                 return -EINVAL;
 466
 467         return 0;
 468 }
 469
 470 static int
 471 eth_dev_macaddr_set(struct rte_eth_dev *dev, struct rte_ether_addr *addr)
 472 {
 473         struct pmd_internals *internals = dev->data->dev_private;
 474         struct ifreq ifr = { };
 475         int sockfd = internals->rx_queue[0].sockfd;
 476         int ret;
 477
 478         if (sockfd == -1) {
 479                 PMD_LOG(ERR, "receive socket not found");
 480                 return -EINVAL;
 481         }
 482
 483         strlcpy(ifr.ifr_name, internals->if_name, IFNAMSIZ);
 484         ifr.ifr_hwaddr.sa_family = ARPHRD_ETHER;
 485         memcpy(ifr.ifr_hwaddr.sa_data, addr, sizeof(*addr));
 486         ret = ioctl(sockfd, SIOCSIFHWADDR, &ifr);
 487
 488         if (ret < 0) {
 489                 PMD_LOG_ERRNO(ERR, "ioctl(SIOCSIFHWADDR) failed");
 490                 return -EINVAL;
 491         }
 492
 493         return 0;
 494 }
 495
 496 static int
 497 eth_dev_change_flags(char *if_name, uint32_t flags, uint32_t mask)
 498 {
 499         struct ifreq ifr;
 500         int ret = 0;
 501         int s;
 502
 503         s = socket(PF_INET, SOCK_DGRAM, 0);
 504         if (s < 0)
 505                 return -errno;
 506
 507         strlcpy(ifr.ifr_name, if_name, IFNAMSIZ);
 508         if (ioctl(s, SIOCGIFFLAGS, &ifr) < 0) {
 509                 ret = -errno;
 510                 goto out;
 511         }
 512         ifr.ifr_flags &= mask;
 513         ifr.ifr_flags |= flags;
 514         if (ioctl(s, SIOCSIFFLAGS, &ifr) < 0) {
 515                 ret = -errno;
 516                 goto out;
 517         }
 518 out:
 519         close(s);
 520         return ret;
 521 }
 522
 523 static int
 524 eth_dev_promiscuous_enable(struct rte_eth_dev *dev)
 525 {
 526         struct pmd_internals *internals = dev->data->dev_private;
 527
 528         return eth_dev_change_flags(internals->if_name, IFF_PROMISC, ~0);
 529 }
 530
 531 static int
 532 eth_dev_promiscuous_disable(struct rte_eth_dev *dev)
 533 {
 534         struct pmd_internals *internals = dev->data->dev_private;
 535
 536         return eth_dev_change_flags(internals->if_name, 0, ~IFF_PROMISC);
 537 }
 538
 539 static const struct eth_dev_ops ops = {
 540         .dev_start = eth_dev_start,
 541         .dev_stop = eth_dev_stop,
 542         .dev_close = eth_dev_close,
 543         .dev_configure = eth_dev_configure,
 544         .dev_infos_get = eth_dev_info,
 545         .mac_addr_set = eth_dev_macaddr_set,
 546         .mtu_set = eth_dev_mtu_set,
 547         .promiscuous_enable = eth_dev_promiscuous_enable,
 548         .promiscuous_disable = eth_dev_promiscuous_disable,
 549         .rx_queue_setup = eth_rx_queue_setup,
 550         .tx_queue_setup = eth_tx_queue_setup,
 551         .rx_queue_release = eth_queue_release,
 552         .tx_queue_release = eth_queue_release,
 553         .link_update = eth_link_update,
 554         .stats_get = eth_stats_get,
 555         .stats_reset = eth_stats_reset,
 556 };
 557
 558 /*
 559  * Opens an AF_PACKET socket
 560  */
 561 static int
 562 open_packet_iface(const char *key __rte_unused,
 563                   const char *value __rte_unused,
 564                   void *extra_args)
 565 {
 566         int *sockfd = extra_args;
 567
 568         /* Open an AF_PACKET socket... */
 569         *sockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
 570         if (*sockfd == -1) {
 571                 PMD_LOG(ERR, "Could not open AF_PACKET socket");
 572                 return -1;
 573         }
 574
 575         return 0;
 576 }
 577
 578 static int
 579 rte_pmd_init_internals(struct rte_vdev_device *dev,
 580                        const int sockfd,
 581                        const unsigned nb_queues,
 582                        unsigned int blocksize,
 583                        unsigned int blockcnt,
 584                        unsigned int framesize,
 585                        unsigned int framecnt,
 586                        unsigned int qdisc_bypass,
 587                        struct pmd_internals **internals,
 588                        struct rte_eth_dev **eth_dev,
 589                        struct rte_kvargs *kvlist)
 590 {
 591         const char *name = rte_vdev_device_name(dev);
 592         const unsigned int numa_node = dev->device.numa_node;
 593         struct rte_eth_dev_data *data = NULL;
 594         struct rte_kvargs_pair *pair = NULL;
 595         struct ifreq ifr;
 596         size_t ifnamelen;
 597         unsigned k_idx;
 598         struct sockaddr_ll sockaddr;
 599         struct tpacket_req *req;
 600         struct pkt_rx_queue *rx_queue;
 601         struct pkt_tx_queue *tx_queue;
 602         int rc, tpver, discard;
 603         int qsockfd = -1;
 604         unsigned int i, q, rdsize;
 605 #if defined(PACKET_FANOUT)
 606         int fanout_arg;
 607 #endif
 608
 609         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 610                 pair = &kvlist->pairs[k_idx];
 611                 if (strstr(pair->key, ETH_AF_PACKET_IFACE_ARG) != NULL)
 612                         break;
 613         }
 614         if (pair == NULL) {
 615                 PMD_LOG(ERR,
 616                         "%s: no interface specified for AF_PACKET ethdev",
 617                         name);
 618                 return -1;
 619         }
 620
 621         PMD_LOG(INFO,
 622                 "%s: creating AF_PACKET-backed ethdev on numa socket %u",
 623                 name, numa_node);
 624
 625         *internals = rte_zmalloc_socket(name, sizeof(**internals),
 626                                         0, numa_node);
 627         if (*internals == NULL)
 628                 return -1;
 629
 630
 631         (*internals)->rx_queue = rte_calloc_socket("af_packet_rx",
 632                                                 nb_queues,
 633                                                 sizeof(struct pkt_rx_queue),
 634                                                 0, numa_node);
 635         (*internals)->tx_queue = rte_calloc_socket("af_packet_tx",
 636                                                 nb_queues,
 637                                                 sizeof(struct pkt_tx_queue),
 638                                                 0, numa_node);
 639         if (!(*internals)->rx_queue || !(*internals)->tx_queue) {
 640                 rte_free((*internals)->rx_queue);
 641                 rte_free((*internals)->tx_queue);
 642                 return -1;
 643         }
 644
 645         for (q = 0; q < nb_queues; q++) {
 646                 (*internals)->rx_queue[q].map = MAP_FAILED;
 647                 (*internals)->tx_queue[q].map = MAP_FAILED;
 648                 (*internals)->rx_queue[q].sockfd = -1;
 649                 (*internals)->tx_queue[q].sockfd = -1;
 650         }
 651
 652         req = &((*internals)->req);
 653
 654         req->tp_block_size = blocksize;
 655         req->tp_block_nr = blockcnt;
 656         req->tp_frame_size = framesize;
 657         req->tp_frame_nr = framecnt;
 658
 659         ifnamelen = strlen(pair->value);
 660         if (ifnamelen < sizeof(ifr.ifr_name)) {
 661                 memcpy(ifr.ifr_name, pair->value, ifnamelen);
 662                 ifr.ifr_name[ifnamelen] = '\0';
 663         } else {
 664                 PMD_LOG(ERR,
 665                         "%s: I/F name too long (%s)",
 666                         name, pair->value);
 667                 return -1;
 668         }
 669         if (ioctl(sockfd, SIOCGIFINDEX, &ifr) == -1) {
 670                 PMD_LOG_ERRNO(ERR, "%s: ioctl failed (SIOCGIFINDEX)", name);
 671                 return -1;
 672         }
 673         (*internals)->if_name = strdup(pair->value);
 674         if ((*internals)->if_name == NULL)
 675                 return -1;
 676         (*internals)->if_index = ifr.ifr_ifindex;
 677
 678         if (ioctl(sockfd, SIOCGIFHWADDR, &ifr) == -1) {
 679                 PMD_LOG_ERRNO(ERR, "%s: ioctl failed (SIOCGIFHWADDR)", name);
 680                 return -1;
 681         }
 682         memcpy(&(*internals)->eth_addr, ifr.ifr_hwaddr.sa_data, ETH_ALEN);
 683
 684         memset(&sockaddr, 0, sizeof(sockaddr));
 685         sockaddr.sll_family = AF_PACKET;
 686         sockaddr.sll_protocol = htons(ETH_P_ALL);
 687         sockaddr.sll_ifindex = (*internals)->if_index;
 688
 689 #if defined(PACKET_FANOUT)
 690         fanout_arg = (getpid() ^ (*internals)->if_index) & 0xffff;
 691         fanout_arg |= (PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_DEFRAG) << 16;
 692 #if defined(PACKET_FANOUT_FLAG_ROLLOVER)
 693         fanout_arg |= PACKET_FANOUT_FLAG_ROLLOVER << 16;
 694 #endif
 695 #endif
 696
 697         for (q = 0; q < nb_queues; q++) {
 698                 /* Open an AF_PACKET socket for this queue... */
 699                 qsockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
 700                 if (qsockfd == -1) {
 701                         PMD_LOG_ERRNO(ERR,
 702                                 "%s: could not open AF_PACKET socket",
 703                                 name);
 704                         return -1;
 705                 }
 706
 707                 tpver = TPACKET_V2;
 708                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_VERSION,
 709                                 &tpver, sizeof(tpver));
 710                 if (rc == -1) {
 711                         PMD_LOG_ERRNO(ERR,
 712                                 "%s: could not set PACKET_VERSION on AF_PACKET socket for %s",
 713                                 name, pair->value);
 714                         goto error;
 715                 }
 716
 717                 discard = 1;
 718                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_LOSS,
 719                                 &discard, sizeof(discard));
 720                 if (rc == -1) {
 721                         PMD_LOG_ERRNO(ERR,
 722                                 "%s: could not set PACKET_LOSS on AF_PACKET socket for %s",
 723                                 name, pair->value);
 724                         goto error;
 725                 }
 726
 727 #if defined(PACKET_QDISC_BYPASS)
 728                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_QDISC_BYPASS,
 729                                 &qdisc_bypass, sizeof(qdisc_bypass));
 730                 if (rc == -1) {
 731                         PMD_LOG_ERRNO(ERR,
 732                                 "%s: could not set PACKET_QDISC_BYPASS on AF_PACKET socket for %s",
 733                                 name, pair->value);
 734                         goto error;
 735                 }
 736 #else
 737                 RTE_SET_USED(qdisc_bypass);
 738 #endif
 739
 740                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_RX_RING, req, sizeof(*req));
 741                 if (rc == -1) {
 742                         PMD_LOG_ERRNO(ERR,
 743                                 "%s: could not set PACKET_RX_RING on AF_PACKET socket for %s",
 744                                 name, pair->value);
 745                         goto error;
 746                 }
 747
 748                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_TX_RING, req, sizeof(*req));
 749                 if (rc == -1) {
 750                         PMD_LOG_ERRNO(ERR,
 751                                 "%s: could not set PACKET_TX_RING on AF_PACKET "
 752                                 "socket for %s", name, pair->value);
 753                         goto error;
 754                 }
 755
 756                 rx_queue = &((*internals)->rx_queue[q]);
 757                 rx_queue->framecount = req->tp_frame_nr;
 758
 759                 rx_queue->map = mmap(NULL, 2 * req->tp_block_size * req->tp_block_nr,
 760                                     PROT_READ | PROT_WRITE, MAP_SHARED | MAP_LOCKED,
 761                                     qsockfd, 0);
 762                 if (rx_queue->map == MAP_FAILED) {
 763                         PMD_LOG_ERRNO(ERR,
 764                                 "%s: call to mmap failed on AF_PACKET socket for %s",
 765                                 name, pair->value);
 766                         goto error;
 767                 }
 768
 769                 /* rdsize is same for both Tx and Rx */
 770                 rdsize = req->tp_frame_nr * sizeof(*(rx_queue->rd));
 771
 772                 rx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
 773                 if (rx_queue->rd == NULL)
 774                         goto error;
 775                 for (i = 0; i < req->tp_frame_nr; ++i) {
 776                         rx_queue->rd[i].iov_base = rx_queue->map + (i * framesize);
 777                         rx_queue->rd[i].iov_len = req->tp_frame_size;
 778                 }
 779                 rx_queue->sockfd = qsockfd;
 780
 781                 tx_queue = &((*internals)->tx_queue[q]);
 782                 tx_queue->framecount = req->tp_frame_nr;
 783                 tx_queue->frame_data_size = req->tp_frame_size;
 784                 tx_queue->frame_data_size -= TPACKET2_HDRLEN -
 785                         sizeof(struct sockaddr_ll);
 786
 787                 tx_queue->map = rx_queue->map + req->tp_block_size * req->tp_block_nr;
 788
 789                 tx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
 790                 if (tx_queue->rd == NULL)
 791                         goto error;
 792                 for (i = 0; i < req->tp_frame_nr; ++i) {
 793                         tx_queue->rd[i].iov_base = tx_queue->map + (i * framesize);
 794                         tx_queue->rd[i].iov_len = req->tp_frame_size;
 795                 }
 796                 tx_queue->sockfd = qsockfd;
 797
 798                 rc = bind(qsockfd, (const struct sockaddr*)&sockaddr, sizeof(sockaddr));
 799                 if (rc == -1) {
 800                         PMD_LOG_ERRNO(ERR,
 801                                 "%s: could not bind AF_PACKET socket to %s",
 802                                 name, pair->value);
 803                         goto error;
 804                 }
 805
 806 #if defined(PACKET_FANOUT)
 807                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_FANOUT,
 808                                 &fanout_arg, sizeof(fanout_arg));
 809                 if (rc == -1) {
 810                         PMD_LOG_ERRNO(ERR,
 811                                 "%s: could not set PACKET_FANOUT on AF_PACKET socket for %s",
 812                                 name, pair->value);
 813                         goto error;
 814                 }
 815 #endif
 816         }
 817
 818         /* reserve an ethdev entry */
 819         *eth_dev = rte_eth_vdev_allocate(dev, 0);
 820         if (*eth_dev == NULL)
 821                 goto error;
 822
 823         /*
 824          * now put it all together
 825          * - store queue data in internals,
 826          * - store numa_node in eth_dev
 827          * - point eth_dev_data to internals
 828          * - and point eth_dev structure to new eth_dev_data structure
 829          */
 830
 831         (*internals)->nb_queues = nb_queues;
 832
 833         data = (*eth_dev)->data;
 834         data->dev_private = *internals;
 835         data->nb_rx_queues = (uint16_t)nb_queues;
 836         data->nb_tx_queues = (uint16_t)nb_queues;
 837         data->dev_link = pmd_link;
 838         data->mac_addrs = &(*internals)->eth_addr;
 839
 840         (*eth_dev)->dev_ops = &ops;
 841
 842         return 0;
 843
 844 error:
 845         if (qsockfd != -1)
 846                 close(qsockfd);
 847         for (q = 0; q < nb_queues; q++) {
 848                 munmap((*internals)->rx_queue[q].map,
 849                        2 * req->tp_block_size * req->tp_block_nr);
 850
 851                 rte_free((*internals)->rx_queue[q].rd);
 852                 rte_free((*internals)->tx_queue[q].rd);
 853                 if (((*internals)->rx_queue[q].sockfd >= 0) &&
 854                         ((*internals)->rx_queue[q].sockfd != qsockfd))
 855                         close((*internals)->rx_queue[q].sockfd);
 856         }
 857         free((*internals)->if_name);
 858         rte_free(*internals);
 859         return -1;
 860 }
 861
 862 static int
 863 rte_eth_from_packet(struct rte_vdev_device *dev,
 864                     int const *sockfd,
 865                     struct rte_kvargs *kvlist)
 866 {
 867         const char *name = rte_vdev_device_name(dev);
 868         struct pmd_internals *internals = NULL;
 869         struct rte_eth_dev *eth_dev = NULL;
 870         struct rte_kvargs_pair *pair = NULL;
 871         unsigned k_idx;
 872         unsigned int blockcount;
 873         unsigned int blocksize;
 874         unsigned int framesize = DFLT_FRAME_SIZE;
 875         unsigned int framecount = DFLT_FRAME_COUNT;
 876         unsigned int qpairs = 1;
 877         unsigned int qdisc_bypass = 1;
 878
 879         /* do some parameter checking */
 880         if (*sockfd < 0)
 881                 return -1;
 882
 883         blocksize = getpagesize();
 884
 885         /*
 886          * Walk arguments for configurable settings
 887          */
 888         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 889                 pair = &kvlist->pairs[k_idx];
 890                 if (strstr(pair->key, ETH_AF_PACKET_NUM_Q_ARG) != NULL) {
 891                         qpairs = atoi(pair->value);
 892                         if (qpairs < 1) {
 893                                 PMD_LOG(ERR,
 894                                         "%s: invalid qpairs value",
 895                                         name);
 896                                 return -1;
 897                         }
 898                         continue;
 899                 }
 900                 if (strstr(pair->key, ETH_AF_PACKET_BLOCKSIZE_ARG) != NULL) {
 901                         blocksize = atoi(pair->value);
 902                         if (!blocksize) {
 903                                 PMD_LOG(ERR,
 904                                         "%s: invalid blocksize value",
 905                                         name);
 906                                 return -1;
 907                         }
 908                         continue;
 909                 }
 910                 if (strstr(pair->key, ETH_AF_PACKET_FRAMESIZE_ARG) != NULL) {
 911                         framesize = atoi(pair->value);
 912                         if (!framesize) {
 913                                 PMD_LOG(ERR,
 914                                         "%s: invalid framesize value",
 915                                         name);
 916                                 return -1;
 917                         }
 918                         continue;
 919                 }
 920                 if (strstr(pair->key, ETH_AF_PACKET_FRAMECOUNT_ARG) != NULL) {
 921                         framecount = atoi(pair->value);
 922                         if (!framecount) {
 923                                 PMD_LOG(ERR,
 924                                         "%s: invalid framecount value",
 925                                         name);
 926                                 return -1;
 927                         }
 928                         continue;
 929                 }
 930                 if (strstr(pair->key, ETH_AF_PACKET_QDISC_BYPASS_ARG) != NULL) {
 931                         qdisc_bypass = atoi(pair->value);
 932                         if (qdisc_bypass > 1) {
 933                                 PMD_LOG(ERR,
 934                                         "%s: invalid bypass value",
 935                                         name);
 936                                 return -1;
 937                         }
 938                         continue;
 939                 }
 940         }
 941
 942         if (framesize > blocksize) {
 943                 PMD_LOG(ERR,
 944                         "%s: AF_PACKET MMAP frame size exceeds block size!",
 945                         name);
 946                 return -1;
 947         }
 948
 949         blockcount = framecount / (blocksize / framesize);
 950         if (!blockcount) {
 951                 PMD_LOG(ERR,
 952                         "%s: invalid AF_PACKET MMAP parameters", name);
 953                 return -1;
 954         }
 955
 956         PMD_LOG(INFO, "%s: AF_PACKET MMAP parameters:", name);
 957         PMD_LOG(INFO, "%s:\tblock size %d", name, blocksize);
 958         PMD_LOG(INFO, "%s:\tblock count %d", name, blockcount);
 959         PMD_LOG(INFO, "%s:\tframe size %d", name, framesize);
 960         PMD_LOG(INFO, "%s:\tframe count %d", name, framecount);
 961
 962         if (rte_pmd_init_internals(dev, *sockfd, qpairs,
 963                                    blocksize, blockcount,
 964                                    framesize, framecount,
 965                                    qdisc_bypass,
 966                                    &internals, &eth_dev,
 967                                    kvlist) < 0)
 968                 return -1;
 969
 970         eth_dev->rx_pkt_burst = eth_af_packet_rx;
 971         eth_dev->tx_pkt_burst = eth_af_packet_tx;
 972
 973         rte_eth_dev_probing_finish(eth_dev);
 974         return 0;
 975 }
 976
 977 static int
 978 rte_pmd_af_packet_probe(struct rte_vdev_device *dev)
 979 {
 980         int ret = 0;
 981         struct rte_kvargs *kvlist;
 982         int sockfd = -1;
 983         struct rte_eth_dev *eth_dev;
 984         const char *name = rte_vdev_device_name(dev);
 985
 986         PMD_LOG(INFO, "Initializing pmd_af_packet for %s", name);
 987
 988         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
 989                 eth_dev = rte_eth_dev_attach_secondary(name);
 990                 if (!eth_dev) {
 991                         PMD_LOG(ERR, "Failed to probe %s", name);
 992                         return -1;
 993                 }
 994                 /* TODO: request info from primary to set up Rx and Tx */
 995                 eth_dev->dev_ops = &ops;
 996                 eth_dev->device = &dev->device;
 997                 rte_eth_dev_probing_finish(eth_dev);
 998                 return 0;
 999         }
1000
1001         kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_arguments);
1002         if (kvlist == NULL) {
1003                 ret = -1;
1004                 goto exit;
1005         }
1006
1007         /*
1008          * If iface argument is passed we open the NICs and use them for
1009          * reading / writing
1010          */
1011         if (rte_kvargs_count(kvlist, ETH_AF_PACKET_IFACE_ARG) == 1) {
1012
1013                 ret = rte_kvargs_process(kvlist, ETH_AF_PACKET_IFACE_ARG,
1014                                          &open_packet_iface, &sockfd);
1015                 if (ret < 0)
1016                         goto exit;
1017         }
1018
1019         if (dev->device.numa_node == SOCKET_ID_ANY)
1020                 dev->device.numa_node = rte_socket_id();
1021
1022         ret = rte_eth_from_packet(dev, &sockfd, kvlist);
1023         close(sockfd); /* no longer needed */
1024
1025 exit:
1026         rte_kvargs_free(kvlist);
1027         return ret;
1028 }
1029
1030 static int
1031 rte_pmd_af_packet_remove(struct rte_vdev_device *dev)
1032 {
1033         struct rte_eth_dev *eth_dev = NULL;
1034         struct pmd_internals *internals;
1035         struct tpacket_req *req;
1036         unsigned q;
1037
1038         PMD_LOG(INFO, "Closing AF_PACKET ethdev on numa socket %u",
1039                 rte_socket_id());
1040
1041         if (dev == NULL)
1042                 return -1;
1043
1044         /* find the ethdev entry */
1045         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1046         if (eth_dev == NULL)
1047                 return -1;
1048
1049         /* mac_addrs must not be freed alone because part of dev_private */
1050         eth_dev->data->mac_addrs = NULL;
1051
1052         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1053                 return rte_eth_dev_release_port(eth_dev);
1054
1055         internals = eth_dev->data->dev_private;
1056         req = &internals->req;
1057         for (q = 0; q < internals->nb_queues; q++) {
1058                 munmap(internals->rx_queue[q].map,
1059                         2 * req->tp_block_size * req->tp_block_nr);
1060                 rte_free(internals->rx_queue[q].rd);
1061                 rte_free(internals->tx_queue[q].rd);
1062         }
1063         free(internals->if_name);
1064         rte_free(internals->rx_queue);
1065         rte_free(internals->tx_queue);
1066
1067         rte_eth_dev_release_port(eth_dev);
1068
1069         return 0;
1070 }
1071
1072 static struct rte_vdev_driver pmd_af_packet_drv = {
1073         .probe = rte_pmd_af_packet_probe,
1074         .remove = rte_pmd_af_packet_remove,
1075 };
1076
1077 RTE_PMD_REGISTER_VDEV(net_af_packet, pmd_af_packet_drv);
1078 RTE_PMD_REGISTER_ALIAS(net_af_packet, eth_af_packet);
1079 RTE_PMD_REGISTER_PARAM_STRING(net_af_packet,
1080         "iface=<string> "
1081         "qpairs=<int> "
1082         "blocksz=<int> "
1083         "framesz=<int> "
1084         "framecnt=<int> "
1085         "qdisc_bypass=<0|1>");