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 <ethdev_driver.h>
  12 #include <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         uint8_t vlan_strip;
  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;
  81         struct pkt_tx_queue *tx_queue;
  82         uint8_t vlan_strip;
  83 };
  84
  85 static const char *valid_arguments[] = {
  86         ETH_AF_PACKET_IFACE_ARG,
  87         ETH_AF_PACKET_NUM_Q_ARG,
  88         ETH_AF_PACKET_BLOCKSIZE_ARG,
  89         ETH_AF_PACKET_FRAMESIZE_ARG,
  90         ETH_AF_PACKET_FRAMECOUNT_ARG,
  91         ETH_AF_PACKET_QDISC_BYPASS_ARG,
  92         NULL
  93 };
  94
  95 static struct rte_eth_link pmd_link = {
  96         .link_speed = ETH_SPEED_NUM_10G,
  97         .link_duplex = ETH_LINK_FULL_DUPLEX,
  98         .link_status = ETH_LINK_DOWN,
  99         .link_autoneg = ETH_LINK_FIXED,
 100 };
 101
 102 RTE_LOG_REGISTER_DEFAULT(af_packet_logtype, NOTICE);
 103
 104 #define PMD_LOG(level, fmt, args...) \
 105         rte_log(RTE_LOG_ ## level, af_packet_logtype, \
 106                 "%s(): " fmt "\n", __func__, ##args)
 107
 108 #define PMD_LOG_ERRNO(level, fmt, args...) \
 109         rte_log(RTE_LOG_ ## level, af_packet_logtype, \
 110                 "%s(): " fmt ":%s\n", __func__, ##args, strerror(errno))
 111
 112 static uint16_t
 113 eth_af_packet_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 114 {
 115         unsigned i;
 116         struct tpacket2_hdr *ppd;
 117         struct rte_mbuf *mbuf;
 118         uint8_t *pbuf;
 119         struct pkt_rx_queue *pkt_q = queue;
 120         uint16_t num_rx = 0;
 121         unsigned long num_rx_bytes = 0;
 122         unsigned int framecount, framenum;
 123
 124         if (unlikely(nb_pkts == 0))
 125                 return 0;
 126
 127         /*
 128          * Reads the given number of packets from the AF_PACKET socket one by
 129          * one and copies the packet data into a newly allocated mbuf.
 130          */
 131         framecount = pkt_q->framecount;
 132         framenum = pkt_q->framenum;
 133         for (i = 0; i < nb_pkts; i++) {
 134                 /* point at the next incoming frame */
 135                 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
 136                 if ((ppd->tp_status & TP_STATUS_USER) == 0)
 137                         break;
 138
 139                 /* allocate the next mbuf */
 140                 mbuf = rte_pktmbuf_alloc(pkt_q->mb_pool);
 141                 if (unlikely(mbuf == NULL))
 142                         break;
 143
 144                 /* packet will fit in the mbuf, go ahead and receive it */
 145                 rte_pktmbuf_pkt_len(mbuf) = rte_pktmbuf_data_len(mbuf) = ppd->tp_snaplen;
 146                 pbuf = (uint8_t *) ppd + ppd->tp_mac;
 147                 memcpy(rte_pktmbuf_mtod(mbuf, void *), pbuf, rte_pktmbuf_data_len(mbuf));
 148
 149                 /* check for vlan info */
 150                 if (ppd->tp_status & TP_STATUS_VLAN_VALID) {
 151                         mbuf->vlan_tci = ppd->tp_vlan_tci;
 152                         mbuf->ol_flags |= (PKT_RX_VLAN | PKT_RX_VLAN_STRIPPED);
 153
 154                         if (!pkt_q->vlan_strip && rte_vlan_insert(&mbuf))
 155                                 PMD_LOG(ERR, "Failed to reinsert VLAN tag");
 156                 }
 157
 158                 /* release incoming frame and advance ring buffer */
 159                 ppd->tp_status = TP_STATUS_KERNEL;
 160                 if (++framenum >= framecount)
 161                         framenum = 0;
 162                 mbuf->port = pkt_q->in_port;
 163
 164                 /* account for the receive frame */
 165                 bufs[i] = mbuf;
 166                 num_rx++;
 167                 num_rx_bytes += mbuf->pkt_len;
 168         }
 169         pkt_q->framenum = framenum;
 170         pkt_q->rx_pkts += num_rx;
 171         pkt_q->rx_bytes += num_rx_bytes;
 172         return num_rx;
 173 }
 174
 175 /*
 176  * Check if there is an available frame in the ring
 177  */
 178 static inline bool
 179 tx_ring_status_available(uint32_t tp_status)
 180 {
 181         /*
 182          * We eliminate the timestamp status from the packet status.
 183          * This should only matter if timestamping is enabled on the socket,
 184          * but there is a bug in the kernel which is fixed in newer releases.
 185          *
 186          * See the following kernel commit for reference:
 187          *     commit 171c3b151118a2fe0fc1e2a9d1b5a1570cfe82d2
 188          *     net: packetmmap: fix only tx timestamp on request
 189          */
 190         tp_status &= ~(TP_STATUS_TS_SOFTWARE | TP_STATUS_TS_RAW_HARDWARE);
 191
 192         return tp_status == TP_STATUS_AVAILABLE;
 193 }
 194
 195 /*
 196  * Callback to handle sending packets through a real NIC.
 197  */
 198 static uint16_t
 199 eth_af_packet_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 200 {
 201         struct tpacket2_hdr *ppd;
 202         struct rte_mbuf *mbuf;
 203         uint8_t *pbuf;
 204         unsigned int framecount, framenum;
 205         struct pollfd pfd;
 206         struct pkt_tx_queue *pkt_q = queue;
 207         uint16_t num_tx = 0;
 208         unsigned long num_tx_bytes = 0;
 209         int i;
 210
 211         if (unlikely(nb_pkts == 0))
 212                 return 0;
 213
 214         memset(&pfd, 0, sizeof(pfd));
 215         pfd.fd = pkt_q->sockfd;
 216         pfd.events = POLLOUT;
 217         pfd.revents = 0;
 218
 219         framecount = pkt_q->framecount;
 220         framenum = pkt_q->framenum;
 221         ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
 222         for (i = 0; i < nb_pkts; i++) {
 223                 mbuf = *bufs++;
 224
 225                 /* drop oversized packets */
 226                 if (mbuf->pkt_len > pkt_q->frame_data_size) {
 227                         rte_pktmbuf_free(mbuf);
 228                         continue;
 229                 }
 230
 231                 /* insert vlan info if necessary */
 232                 if (mbuf->ol_flags & PKT_TX_VLAN_PKT) {
 233                         if (rte_vlan_insert(&mbuf)) {
 234                                 rte_pktmbuf_free(mbuf);
 235                                 continue;
 236                         }
 237                 }
 238
 239                 /* point at the next incoming frame */
 240                 if (!tx_ring_status_available(ppd->tp_status) &&
 241                     poll(&pfd, 1, -1) < 0)
 242                         break;
 243
 244                 /* copy the tx frame data */
 245                 pbuf = (uint8_t *) ppd + TPACKET2_HDRLEN -
 246                         sizeof(struct sockaddr_ll);
 247
 248                 struct rte_mbuf *tmp_mbuf = mbuf;
 249                 while (tmp_mbuf) {
 250                         uint16_t data_len = rte_pktmbuf_data_len(tmp_mbuf);
 251                         memcpy(pbuf, rte_pktmbuf_mtod(tmp_mbuf, void*), data_len);
 252                         pbuf += data_len;
 253                         tmp_mbuf = tmp_mbuf->next;
 254                 }
 255
 256                 ppd->tp_len = mbuf->pkt_len;
 257                 ppd->tp_snaplen = mbuf->pkt_len;
 258
 259                 /* release incoming frame and advance ring buffer */
 260                 ppd->tp_status = TP_STATUS_SEND_REQUEST;
 261                 if (++framenum >= framecount)
 262                         framenum = 0;
 263                 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
 264
 265                 num_tx++;
 266                 num_tx_bytes += mbuf->pkt_len;
 267                 rte_pktmbuf_free(mbuf);
 268         }
 269
 270         /* kick-off transmits */
 271         if (sendto(pkt_q->sockfd, NULL, 0, MSG_DONTWAIT, NULL, 0) == -1 &&
 272                         errno != ENOBUFS && errno != EAGAIN) {
 273                 /*
 274                  * In case of a ENOBUFS/EAGAIN error all of the enqueued
 275                  * packets will be considered successful even though only some
 276                  * are sent.
 277                  */
 278
 279                 num_tx = 0;
 280                 num_tx_bytes = 0;
 281         }
 282
 283         pkt_q->framenum = framenum;
 284         pkt_q->tx_pkts += num_tx;
 285         pkt_q->err_pkts += i - num_tx;
 286         pkt_q->tx_bytes += num_tx_bytes;
 287         return i;
 288 }
 289
 290 static int
 291 eth_dev_start(struct rte_eth_dev *dev)
 292 {
 293         dev->data->dev_link.link_status = ETH_LINK_UP;
 294         return 0;
 295 }
 296
 297 /*
 298  * This function gets called when the current port gets stopped.
 299  */
 300 static int
 301 eth_dev_stop(struct rte_eth_dev *dev)
 302 {
 303         unsigned i;
 304         int sockfd;
 305         struct pmd_internals *internals = dev->data->dev_private;
 306
 307         for (i = 0; i < internals->nb_queues; i++) {
 308                 sockfd = internals->rx_queue[i].sockfd;
 309                 if (sockfd != -1)
 310                         close(sockfd);
 311
 312                 /* Prevent use after free in case tx fd == rx fd */
 313                 if (sockfd != internals->tx_queue[i].sockfd) {
 314                         sockfd = internals->tx_queue[i].sockfd;
 315                         if (sockfd != -1)
 316                                 close(sockfd);
 317                 }
 318
 319                 internals->rx_queue[i].sockfd = -1;
 320                 internals->tx_queue[i].sockfd = -1;
 321         }
 322
 323         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 324         return 0;
 325 }
 326
 327 static int
 328 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 329 {
 330         struct rte_eth_conf *dev_conf = &dev->data->dev_conf;
 331         const struct rte_eth_rxmode *rxmode = &dev_conf->rxmode;
 332         struct pmd_internals *internals = dev->data->dev_private;
 333
 334         internals->vlan_strip = !!(rxmode->offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
 335         return 0;
 336 }
 337
 338 static int
 339 eth_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 340 {
 341         struct pmd_internals *internals = dev->data->dev_private;
 342
 343         dev_info->if_index = internals->if_index;
 344         dev_info->max_mac_addrs = 1;
 345         dev_info->max_rx_pktlen = (uint32_t)ETH_FRAME_LEN;
 346         dev_info->max_rx_queues = (uint16_t)internals->nb_queues;
 347         dev_info->max_tx_queues = (uint16_t)internals->nb_queues;
 348         dev_info->min_rx_bufsize = 0;
 349         dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
 350                 DEV_TX_OFFLOAD_VLAN_INSERT;
 351         dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
 352
 353         return 0;
 354 }
 355
 356 static int
 357 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *igb_stats)
 358 {
 359         unsigned i, imax;
 360         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 361         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 362         const struct pmd_internals *internal = dev->data->dev_private;
 363
 364         imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
 365                 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
 366         for (i = 0; i < imax; i++) {
 367                 igb_stats->q_ipackets[i] = internal->rx_queue[i].rx_pkts;
 368                 igb_stats->q_ibytes[i] = internal->rx_queue[i].rx_bytes;
 369                 rx_total += igb_stats->q_ipackets[i];
 370                 rx_bytes_total += igb_stats->q_ibytes[i];
 371         }
 372
 373         imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
 374                 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
 375         for (i = 0; i < imax; i++) {
 376                 igb_stats->q_opackets[i] = internal->tx_queue[i].tx_pkts;
 377                 igb_stats->q_obytes[i] = internal->tx_queue[i].tx_bytes;
 378                 tx_total += igb_stats->q_opackets[i];
 379                 tx_err_total += internal->tx_queue[i].err_pkts;
 380                 tx_bytes_total += igb_stats->q_obytes[i];
 381         }
 382
 383         igb_stats->ipackets = rx_total;
 384         igb_stats->ibytes = rx_bytes_total;
 385         igb_stats->opackets = tx_total;
 386         igb_stats->oerrors = tx_err_total;
 387         igb_stats->obytes = tx_bytes_total;
 388         return 0;
 389 }
 390
 391 static int
 392 eth_stats_reset(struct rte_eth_dev *dev)
 393 {
 394         unsigned i;
 395         struct pmd_internals *internal = dev->data->dev_private;
 396
 397         for (i = 0; i < internal->nb_queues; i++) {
 398                 internal->rx_queue[i].rx_pkts = 0;
 399                 internal->rx_queue[i].rx_bytes = 0;
 400         }
 401
 402         for (i = 0; i < internal->nb_queues; i++) {
 403                 internal->tx_queue[i].tx_pkts = 0;
 404                 internal->tx_queue[i].err_pkts = 0;
 405                 internal->tx_queue[i].tx_bytes = 0;
 406         }
 407
 408         return 0;
 409 }
 410
 411 static int
 412 eth_dev_close(struct rte_eth_dev *dev)
 413 {
 414         struct pmd_internals *internals;
 415         struct tpacket_req *req;
 416         unsigned int q;
 417
 418         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 419                 return 0;
 420
 421         PMD_LOG(INFO, "Closing AF_PACKET ethdev on NUMA socket %u",
 422                 rte_socket_id());
 423
 424         internals = dev->data->dev_private;
 425         req = &internals->req;
 426         for (q = 0; q < internals->nb_queues; q++) {
 427                 munmap(internals->rx_queue[q].map,
 428                         2 * req->tp_block_size * req->tp_block_nr);
 429                 rte_free(internals->rx_queue[q].rd);
 430                 rte_free(internals->tx_queue[q].rd);
 431         }
 432         free(internals->if_name);
 433         rte_free(internals->rx_queue);
 434         rte_free(internals->tx_queue);
 435
 436         /* mac_addrs must not be freed alone because part of dev_private */
 437         dev->data->mac_addrs = NULL;
 438         return 0;
 439 }
 440
 441 static void
 442 eth_queue_release(void *q __rte_unused)
 443 {
 444 }
 445
 446 static int
 447 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 448                 int wait_to_complete __rte_unused)
 449 {
 450         return 0;
 451 }
 452
 453 static int
 454 eth_rx_queue_setup(struct rte_eth_dev *dev,
 455                    uint16_t rx_queue_id,
 456                    uint16_t nb_rx_desc __rte_unused,
 457                    unsigned int socket_id __rte_unused,
 458                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 459                    struct rte_mempool *mb_pool)
 460 {
 461         struct pmd_internals *internals = dev->data->dev_private;
 462         struct pkt_rx_queue *pkt_q = &internals->rx_queue[rx_queue_id];
 463         unsigned int buf_size, data_size;
 464
 465         pkt_q->mb_pool = mb_pool;
 466
 467         /* Now get the space available for data in the mbuf */
 468         buf_size = rte_pktmbuf_data_room_size(pkt_q->mb_pool) -
 469                 RTE_PKTMBUF_HEADROOM;
 470         data_size = internals->req.tp_frame_size;
 471         data_size -= TPACKET2_HDRLEN - sizeof(struct sockaddr_ll);
 472
 473         if (data_size > buf_size) {
 474                 PMD_LOG(ERR,
 475                         "%s: %d bytes will not fit in mbuf (%d bytes)",
 476                         dev->device->name, data_size, buf_size);
 477                 return -ENOMEM;
 478         }
 479
 480         dev->data->rx_queues[rx_queue_id] = pkt_q;
 481         pkt_q->in_port = dev->data->port_id;
 482         pkt_q->vlan_strip = internals->vlan_strip;
 483
 484         return 0;
 485 }
 486
 487 static int
 488 eth_tx_queue_setup(struct rte_eth_dev *dev,
 489                    uint16_t tx_queue_id,
 490                    uint16_t nb_tx_desc __rte_unused,
 491                    unsigned int socket_id __rte_unused,
 492                    const struct rte_eth_txconf *tx_conf __rte_unused)
 493 {
 494
 495         struct pmd_internals *internals = dev->data->dev_private;
 496
 497         dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
 498         return 0;
 499 }
 500
 501 static int
 502 eth_dev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 503 {
 504         struct pmd_internals *internals = dev->data->dev_private;
 505         struct ifreq ifr = { .ifr_mtu = mtu };
 506         int ret;
 507         int s;
 508         unsigned int data_size = internals->req.tp_frame_size -
 509                                  TPACKET2_HDRLEN;
 510
 511         if (mtu > data_size)
 512                 return -EINVAL;
 513
 514         s = socket(PF_INET, SOCK_DGRAM, 0);
 515         if (s < 0)
 516                 return -EINVAL;
 517
 518         strlcpy(ifr.ifr_name, internals->if_name, IFNAMSIZ);
 519         ret = ioctl(s, SIOCSIFMTU, &ifr);
 520         close(s);
 521
 522         if (ret < 0)
 523                 return -EINVAL;
 524
 525         return 0;
 526 }
 527
 528 static int
 529 eth_dev_macaddr_set(struct rte_eth_dev *dev, struct rte_ether_addr *addr)
 530 {
 531         struct pmd_internals *internals = dev->data->dev_private;
 532         struct ifreq ifr = { };
 533         int sockfd = internals->rx_queue[0].sockfd;
 534         int ret;
 535
 536         if (sockfd == -1) {
 537                 PMD_LOG(ERR, "receive socket not found");
 538                 return -EINVAL;
 539         }
 540
 541         strlcpy(ifr.ifr_name, internals->if_name, IFNAMSIZ);
 542         ifr.ifr_hwaddr.sa_family = ARPHRD_ETHER;
 543         memcpy(ifr.ifr_hwaddr.sa_data, addr, sizeof(*addr));
 544         ret = ioctl(sockfd, SIOCSIFHWADDR, &ifr);
 545
 546         if (ret < 0) {
 547                 PMD_LOG_ERRNO(ERR, "ioctl(SIOCSIFHWADDR) failed");
 548                 return -EINVAL;
 549         }
 550
 551         return 0;
 552 }
 553
 554 static int
 555 eth_dev_change_flags(char *if_name, uint32_t flags, uint32_t mask)
 556 {
 557         struct ifreq ifr;
 558         int ret = 0;
 559         int s;
 560
 561         s = socket(PF_INET, SOCK_DGRAM, 0);
 562         if (s < 0)
 563                 return -errno;
 564
 565         strlcpy(ifr.ifr_name, if_name, IFNAMSIZ);
 566         if (ioctl(s, SIOCGIFFLAGS, &ifr) < 0) {
 567                 ret = -errno;
 568                 goto out;
 569         }
 570         ifr.ifr_flags &= mask;
 571         ifr.ifr_flags |= flags;
 572         if (ioctl(s, SIOCSIFFLAGS, &ifr) < 0) {
 573                 ret = -errno;
 574                 goto out;
 575         }
 576 out:
 577         close(s);
 578         return ret;
 579 }
 580
 581 static int
 582 eth_dev_promiscuous_enable(struct rte_eth_dev *dev)
 583 {
 584         struct pmd_internals *internals = dev->data->dev_private;
 585
 586         return eth_dev_change_flags(internals->if_name, IFF_PROMISC, ~0);
 587 }
 588
 589 static int
 590 eth_dev_promiscuous_disable(struct rte_eth_dev *dev)
 591 {
 592         struct pmd_internals *internals = dev->data->dev_private;
 593
 594         return eth_dev_change_flags(internals->if_name, 0, ~IFF_PROMISC);
 595 }
 596
 597 static const struct eth_dev_ops ops = {
 598         .dev_start = eth_dev_start,
 599         .dev_stop = eth_dev_stop,
 600         .dev_close = eth_dev_close,
 601         .dev_configure = eth_dev_configure,
 602         .dev_infos_get = eth_dev_info,
 603         .mac_addr_set = eth_dev_macaddr_set,
 604         .mtu_set = eth_dev_mtu_set,
 605         .promiscuous_enable = eth_dev_promiscuous_enable,
 606         .promiscuous_disable = eth_dev_promiscuous_disable,
 607         .rx_queue_setup = eth_rx_queue_setup,
 608         .tx_queue_setup = eth_tx_queue_setup,
 609         .rx_queue_release = eth_queue_release,
 610         .tx_queue_release = eth_queue_release,
 611         .link_update = eth_link_update,
 612         .stats_get = eth_stats_get,
 613         .stats_reset = eth_stats_reset,
 614 };
 615
 616 /*
 617  * Opens an AF_PACKET socket
 618  */
 619 static int
 620 open_packet_iface(const char *key __rte_unused,
 621                   const char *value __rte_unused,
 622                   void *extra_args)
 623 {
 624         int *sockfd = extra_args;
 625
 626         /* Open an AF_PACKET socket... */
 627         *sockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
 628         if (*sockfd == -1) {
 629                 PMD_LOG(ERR, "Could not open AF_PACKET socket");
 630                 return -1;
 631         }
 632
 633         return 0;
 634 }
 635
 636 static int
 637 rte_pmd_init_internals(struct rte_vdev_device *dev,
 638                        const int sockfd,
 639                        const unsigned nb_queues,
 640                        unsigned int blocksize,
 641                        unsigned int blockcnt,
 642                        unsigned int framesize,
 643                        unsigned int framecnt,
 644                        unsigned int qdisc_bypass,
 645                        struct pmd_internals **internals,
 646                        struct rte_eth_dev **eth_dev,
 647                        struct rte_kvargs *kvlist)
 648 {
 649         const char *name = rte_vdev_device_name(dev);
 650         const unsigned int numa_node = dev->device.numa_node;
 651         struct rte_eth_dev_data *data = NULL;
 652         struct rte_kvargs_pair *pair = NULL;
 653         struct ifreq ifr;
 654         size_t ifnamelen;
 655         unsigned k_idx;
 656         struct sockaddr_ll sockaddr;
 657         struct tpacket_req *req;
 658         struct pkt_rx_queue *rx_queue;
 659         struct pkt_tx_queue *tx_queue;
 660         int rc, tpver, discard;
 661         int qsockfd = -1;
 662         unsigned int i, q, rdsize;
 663 #if defined(PACKET_FANOUT)
 664         int fanout_arg;
 665 #endif
 666
 667         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 668                 pair = &kvlist->pairs[k_idx];
 669                 if (strstr(pair->key, ETH_AF_PACKET_IFACE_ARG) != NULL)
 670                         break;
 671         }
 672         if (pair == NULL) {
 673                 PMD_LOG(ERR,
 674                         "%s: no interface specified for AF_PACKET ethdev",
 675                         name);
 676                 return -1;
 677         }
 678
 679         PMD_LOG(INFO,
 680                 "%s: creating AF_PACKET-backed ethdev on numa socket %u",
 681                 name, numa_node);
 682
 683         *internals = rte_zmalloc_socket(name, sizeof(**internals),
 684                                         0, numa_node);
 685         if (*internals == NULL)
 686                 return -1;
 687
 688
 689         (*internals)->rx_queue = rte_calloc_socket("af_packet_rx",
 690                                                 nb_queues,
 691                                                 sizeof(struct pkt_rx_queue),
 692                                                 0, numa_node);
 693         (*internals)->tx_queue = rte_calloc_socket("af_packet_tx",
 694                                                 nb_queues,
 695                                                 sizeof(struct pkt_tx_queue),
 696                                                 0, numa_node);
 697         if (!(*internals)->rx_queue || !(*internals)->tx_queue) {
 698                 goto free_internals;
 699         }
 700
 701         for (q = 0; q < nb_queues; q++) {
 702                 (*internals)->rx_queue[q].map = MAP_FAILED;
 703                 (*internals)->tx_queue[q].map = MAP_FAILED;
 704                 (*internals)->rx_queue[q].sockfd = -1;
 705                 (*internals)->tx_queue[q].sockfd = -1;
 706         }
 707
 708         req = &((*internals)->req);
 709
 710         req->tp_block_size = blocksize;
 711         req->tp_block_nr = blockcnt;
 712         req->tp_frame_size = framesize;
 713         req->tp_frame_nr = framecnt;
 714
 715         ifnamelen = strlen(pair->value);
 716         if (ifnamelen < sizeof(ifr.ifr_name)) {
 717                 memcpy(ifr.ifr_name, pair->value, ifnamelen);
 718                 ifr.ifr_name[ifnamelen] = '\0';
 719         } else {
 720                 PMD_LOG(ERR,
 721                         "%s: I/F name too long (%s)",
 722                         name, pair->value);
 723                 goto free_internals;
 724         }
 725         if (ioctl(sockfd, SIOCGIFINDEX, &ifr) == -1) {
 726                 PMD_LOG_ERRNO(ERR, "%s: ioctl failed (SIOCGIFINDEX)", name);
 727                 goto free_internals;
 728         }
 729         (*internals)->if_name = strdup(pair->value);
 730         if ((*internals)->if_name == NULL)
 731                 goto free_internals;
 732         (*internals)->if_index = ifr.ifr_ifindex;
 733
 734         if (ioctl(sockfd, SIOCGIFHWADDR, &ifr) == -1) {
 735                 PMD_LOG_ERRNO(ERR, "%s: ioctl failed (SIOCGIFHWADDR)", name);
 736                 goto free_internals;
 737         }
 738         memcpy(&(*internals)->eth_addr, ifr.ifr_hwaddr.sa_data, ETH_ALEN);
 739
 740         memset(&sockaddr, 0, sizeof(sockaddr));
 741         sockaddr.sll_family = AF_PACKET;
 742         sockaddr.sll_protocol = htons(ETH_P_ALL);
 743         sockaddr.sll_ifindex = (*internals)->if_index;
 744
 745 #if defined(PACKET_FANOUT)
 746         fanout_arg = (getpid() ^ (*internals)->if_index) & 0xffff;
 747         fanout_arg |= (PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_DEFRAG) << 16;
 748 #if defined(PACKET_FANOUT_FLAG_ROLLOVER)
 749         fanout_arg |= PACKET_FANOUT_FLAG_ROLLOVER << 16;
 750 #endif
 751 #endif
 752
 753         for (q = 0; q < nb_queues; q++) {
 754                 /* Open an AF_PACKET socket for this queue... */
 755                 qsockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
 756                 if (qsockfd == -1) {
 757                         PMD_LOG_ERRNO(ERR,
 758                                 "%s: could not open AF_PACKET socket",
 759                                 name);
 760                         goto error;
 761                 }
 762
 763                 tpver = TPACKET_V2;
 764                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_VERSION,
 765                                 &tpver, sizeof(tpver));
 766                 if (rc == -1) {
 767                         PMD_LOG_ERRNO(ERR,
 768                                 "%s: could not set PACKET_VERSION on AF_PACKET socket for %s",
 769                                 name, pair->value);
 770                         goto error;
 771                 }
 772
 773                 discard = 1;
 774                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_LOSS,
 775                                 &discard, sizeof(discard));
 776                 if (rc == -1) {
 777                         PMD_LOG_ERRNO(ERR,
 778                                 "%s: could not set PACKET_LOSS on AF_PACKET socket for %s",
 779                                 name, pair->value);
 780                         goto error;
 781                 }
 782
 783                 if (qdisc_bypass) {
 784 #if defined(PACKET_QDISC_BYPASS)
 785                         rc = setsockopt(qsockfd, SOL_PACKET, PACKET_QDISC_BYPASS,
 786                                         &qdisc_bypass, sizeof(qdisc_bypass));
 787                         if (rc == -1) {
 788                                 PMD_LOG_ERRNO(ERR,
 789                                         "%s: could not set PACKET_QDISC_BYPASS on AF_PACKET socket for %s",
 790                                         name, pair->value);
 791                                 goto error;
 792                         }
 793 #endif
 794                 }
 795
 796                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_RX_RING, req, sizeof(*req));
 797                 if (rc == -1) {
 798                         PMD_LOG_ERRNO(ERR,
 799                                 "%s: could not set PACKET_RX_RING on AF_PACKET socket for %s",
 800                                 name, pair->value);
 801                         goto error;
 802                 }
 803
 804                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_TX_RING, req, sizeof(*req));
 805                 if (rc == -1) {
 806                         PMD_LOG_ERRNO(ERR,
 807                                 "%s: could not set PACKET_TX_RING on AF_PACKET "
 808                                 "socket for %s", name, pair->value);
 809                         goto error;
 810                 }
 811
 812                 rx_queue = &((*internals)->rx_queue[q]);
 813                 rx_queue->framecount = req->tp_frame_nr;
 814
 815                 rx_queue->map = mmap(NULL, 2 * req->tp_block_size * req->tp_block_nr,
 816                                     PROT_READ | PROT_WRITE, MAP_SHARED | MAP_LOCKED,
 817                                     qsockfd, 0);
 818                 if (rx_queue->map == MAP_FAILED) {
 819                         PMD_LOG_ERRNO(ERR,
 820                                 "%s: call to mmap failed on AF_PACKET socket for %s",
 821                                 name, pair->value);
 822                         goto error;
 823                 }
 824
 825                 /* rdsize is same for both Tx and Rx */
 826                 rdsize = req->tp_frame_nr * sizeof(*(rx_queue->rd));
 827
 828                 rx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
 829                 if (rx_queue->rd == NULL)
 830                         goto error;
 831                 for (i = 0; i < req->tp_frame_nr; ++i) {
 832                         rx_queue->rd[i].iov_base = rx_queue->map + (i * framesize);
 833                         rx_queue->rd[i].iov_len = req->tp_frame_size;
 834                 }
 835                 rx_queue->sockfd = qsockfd;
 836
 837                 tx_queue = &((*internals)->tx_queue[q]);
 838                 tx_queue->framecount = req->tp_frame_nr;
 839                 tx_queue->frame_data_size = req->tp_frame_size;
 840                 tx_queue->frame_data_size -= TPACKET2_HDRLEN -
 841                         sizeof(struct sockaddr_ll);
 842
 843                 tx_queue->map = rx_queue->map + req->tp_block_size * req->tp_block_nr;
 844
 845                 tx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
 846                 if (tx_queue->rd == NULL)
 847                         goto error;
 848                 for (i = 0; i < req->tp_frame_nr; ++i) {
 849                         tx_queue->rd[i].iov_base = tx_queue->map + (i * framesize);
 850                         tx_queue->rd[i].iov_len = req->tp_frame_size;
 851                 }
 852                 tx_queue->sockfd = qsockfd;
 853
 854                 rc = bind(qsockfd, (const struct sockaddr*)&sockaddr, sizeof(sockaddr));
 855                 if (rc == -1) {
 856                         PMD_LOG_ERRNO(ERR,
 857                                 "%s: could not bind AF_PACKET socket to %s",
 858                                 name, pair->value);
 859                         goto error;
 860                 }
 861
 862 #if defined(PACKET_FANOUT)
 863                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_FANOUT,
 864                                 &fanout_arg, sizeof(fanout_arg));
 865                 if (rc == -1) {
 866                         PMD_LOG_ERRNO(ERR,
 867                                 "%s: could not set PACKET_FANOUT on AF_PACKET socket for %s",
 868                                 name, pair->value);
 869                         goto error;
 870                 }
 871 #endif
 872         }
 873
 874         /* reserve an ethdev entry */
 875         *eth_dev = rte_eth_vdev_allocate(dev, 0);
 876         if (*eth_dev == NULL)
 877                 goto error;
 878
 879         /*
 880          * now put it all together
 881          * - store queue data in internals,
 882          * - store numa_node in eth_dev
 883          * - point eth_dev_data to internals
 884          * - and point eth_dev structure to new eth_dev_data structure
 885          */
 886
 887         (*internals)->nb_queues = nb_queues;
 888
 889         data = (*eth_dev)->data;
 890         data->dev_private = *internals;
 891         data->nb_rx_queues = (uint16_t)nb_queues;
 892         data->nb_tx_queues = (uint16_t)nb_queues;
 893         data->dev_link = pmd_link;
 894         data->mac_addrs = &(*internals)->eth_addr;
 895         data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
 896
 897         (*eth_dev)->dev_ops = &ops;
 898
 899         return 0;
 900
 901 error:
 902         if (qsockfd != -1)
 903                 close(qsockfd);
 904         for (q = 0; q < nb_queues; q++) {
 905                 if ((*internals)->rx_queue[q].map != MAP_FAILED)
 906                         munmap((*internals)->rx_queue[q].map,
 907                                2 * req->tp_block_size * req->tp_block_nr);
 908
 909                 rte_free((*internals)->rx_queue[q].rd);
 910                 rte_free((*internals)->tx_queue[q].rd);
 911                 if (((*internals)->rx_queue[q].sockfd >= 0) &&
 912                         ((*internals)->rx_queue[q].sockfd != qsockfd))
 913                         close((*internals)->rx_queue[q].sockfd);
 914         }
 915 free_internals:
 916         rte_free((*internals)->rx_queue);
 917         rte_free((*internals)->tx_queue);
 918         free((*internals)->if_name);
 919         rte_free(*internals);
 920         return -1;
 921 }
 922
 923 static int
 924 rte_eth_from_packet(struct rte_vdev_device *dev,
 925                     int const *sockfd,
 926                     struct rte_kvargs *kvlist)
 927 {
 928         const char *name = rte_vdev_device_name(dev);
 929         struct pmd_internals *internals = NULL;
 930         struct rte_eth_dev *eth_dev = NULL;
 931         struct rte_kvargs_pair *pair = NULL;
 932         unsigned k_idx;
 933         unsigned int blockcount;
 934         unsigned int blocksize;
 935         unsigned int framesize = DFLT_FRAME_SIZE;
 936         unsigned int framecount = DFLT_FRAME_COUNT;
 937         unsigned int qpairs = 1;
 938         unsigned int qdisc_bypass = 1;
 939
 940         /* do some parameter checking */
 941         if (*sockfd < 0)
 942                 return -1;
 943
 944         blocksize = getpagesize();
 945
 946         /*
 947          * Walk arguments for configurable settings
 948          */
 949         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 950                 pair = &kvlist->pairs[k_idx];
 951                 if (strstr(pair->key, ETH_AF_PACKET_NUM_Q_ARG) != NULL) {
 952                         qpairs = atoi(pair->value);
 953                         if (qpairs < 1) {
 954                                 PMD_LOG(ERR,
 955                                         "%s: invalid qpairs value",
 956                                         name);
 957                                 return -1;
 958                         }
 959                         continue;
 960                 }
 961                 if (strstr(pair->key, ETH_AF_PACKET_BLOCKSIZE_ARG) != NULL) {
 962                         blocksize = atoi(pair->value);
 963                         if (!blocksize) {
 964                                 PMD_LOG(ERR,
 965                                         "%s: invalid blocksize value",
 966                                         name);
 967                                 return -1;
 968                         }
 969                         continue;
 970                 }
 971                 if (strstr(pair->key, ETH_AF_PACKET_FRAMESIZE_ARG) != NULL) {
 972                         framesize = atoi(pair->value);
 973                         if (!framesize) {
 974                                 PMD_LOG(ERR,
 975                                         "%s: invalid framesize value",
 976                                         name);
 977                                 return -1;
 978                         }
 979                         continue;
 980                 }
 981                 if (strstr(pair->key, ETH_AF_PACKET_FRAMECOUNT_ARG) != NULL) {
 982                         framecount = atoi(pair->value);
 983                         if (!framecount) {
 984                                 PMD_LOG(ERR,
 985                                         "%s: invalid framecount value",
 986                                         name);
 987                                 return -1;
 988                         }
 989                         continue;
 990                 }
 991                 if (strstr(pair->key, ETH_AF_PACKET_QDISC_BYPASS_ARG) != NULL) {
 992                         qdisc_bypass = atoi(pair->value);
 993                         if (qdisc_bypass > 1) {
 994                                 PMD_LOG(ERR,
 995                                         "%s: invalid bypass value",
 996                                         name);
 997                                 return -1;
 998                         }
 999                         continue;
1000                 }
1001         }
1002
1003         if (framesize > blocksize) {
1004                 PMD_LOG(ERR,
1005                         "%s: AF_PACKET MMAP frame size exceeds block size!",
1006                         name);
1007                 return -1;
1008         }
1009
1010         blockcount = framecount / (blocksize / framesize);
1011         if (!blockcount) {
1012                 PMD_LOG(ERR,
1013                         "%s: invalid AF_PACKET MMAP parameters", name);
1014                 return -1;
1015         }
1016
1017         PMD_LOG(INFO, "%s: AF_PACKET MMAP parameters:", name);
1018         PMD_LOG(INFO, "%s:\tblock size %d", name, blocksize);
1019         PMD_LOG(INFO, "%s:\tblock count %d", name, blockcount);
1020         PMD_LOG(INFO, "%s:\tframe size %d", name, framesize);
1021         PMD_LOG(INFO, "%s:\tframe count %d", name, framecount);
1022
1023         if (rte_pmd_init_internals(dev, *sockfd, qpairs,
1024                                    blocksize, blockcount,
1025                                    framesize, framecount,
1026                                    qdisc_bypass,
1027                                    &internals, &eth_dev,
1028                                    kvlist) < 0)
1029                 return -1;
1030
1031         eth_dev->rx_pkt_burst = eth_af_packet_rx;
1032         eth_dev->tx_pkt_burst = eth_af_packet_tx;
1033
1034         rte_eth_dev_probing_finish(eth_dev);
1035         return 0;
1036 }
1037
1038 static int
1039 rte_pmd_af_packet_probe(struct rte_vdev_device *dev)
1040 {
1041         int ret = 0;
1042         struct rte_kvargs *kvlist;
1043         int sockfd = -1;
1044         struct rte_eth_dev *eth_dev;
1045         const char *name = rte_vdev_device_name(dev);
1046
1047         PMD_LOG(INFO, "Initializing pmd_af_packet for %s", name);
1048
1049         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1050                 eth_dev = rte_eth_dev_attach_secondary(name);
1051                 if (!eth_dev) {
1052                         PMD_LOG(ERR, "Failed to probe %s", name);
1053                         return -1;
1054                 }
1055                 /* TODO: request info from primary to set up Rx and Tx */
1056                 eth_dev->dev_ops = &ops;
1057                 eth_dev->device = &dev->device;
1058                 rte_eth_dev_probing_finish(eth_dev);
1059                 return 0;
1060         }
1061
1062         kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_arguments);
1063         if (kvlist == NULL) {
1064                 ret = -1;
1065                 goto exit;
1066         }
1067
1068         /*
1069          * If iface argument is passed we open the NICs and use them for
1070          * reading / writing
1071          */
1072         if (rte_kvargs_count(kvlist, ETH_AF_PACKET_IFACE_ARG) == 1) {
1073
1074                 ret = rte_kvargs_process(kvlist, ETH_AF_PACKET_IFACE_ARG,
1075                                          &open_packet_iface, &sockfd);
1076                 if (ret < 0)
1077                         goto exit;
1078         }
1079
1080         if (dev->device.numa_node == SOCKET_ID_ANY)
1081                 dev->device.numa_node = rte_socket_id();
1082
1083         ret = rte_eth_from_packet(dev, &sockfd, kvlist);
1084         close(sockfd); /* no longer needed */
1085
1086 exit:
1087         rte_kvargs_free(kvlist);
1088         return ret;
1089 }
1090
1091 static int
1092 rte_pmd_af_packet_remove(struct rte_vdev_device *dev)
1093 {
1094         struct rte_eth_dev *eth_dev;
1095
1096         if (dev == NULL)
1097                 return -1;
1098
1099         /* find the ethdev entry */
1100         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1101         if (eth_dev == NULL)
1102                 return 0; /* port already released */
1103
1104         eth_dev_close(eth_dev);
1105         rte_eth_dev_release_port(eth_dev);
1106
1107         return 0;
1108 }
1109
1110 static struct rte_vdev_driver pmd_af_packet_drv = {
1111         .probe = rte_pmd_af_packet_probe,
1112         .remove = rte_pmd_af_packet_remove,
1113 };
1114
1115 RTE_PMD_REGISTER_VDEV(net_af_packet, pmd_af_packet_drv);
1116 RTE_PMD_REGISTER_ALIAS(net_af_packet, eth_af_packet);
1117 RTE_PMD_REGISTER_PARAM_STRING(net_af_packet,
1118         "iface=<string> "
1119         "qpairs=<int> "
1120         "blocksz=<int> "
1121         "framesz=<int> "
1122         "framecnt=<int> "
1123         "qdisc_bypass=<0|1>");