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 = RTE_ETH_SPEED_NUM_10G,
  97         .link_duplex = RTE_ETH_LINK_FULL_DUPLEX,
  98         .link_status = RTE_ETH_LINK_DOWN,
  99         .link_autoneg = RTE_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 |= (RTE_MBUF_F_RX_VLAN | RTE_MBUF_F_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 & RTE_MBUF_F_TX_VLAN) {
 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                         if (poll(&pfd, 1, -1) < 0)
 242                                 break;
 243
 244                         /* poll() can return POLLERR if the interface is down */
 245                         if (pfd.revents & POLLERR)
 246                                 break;
 247                 }
 248
 249                 /*
 250                  * poll() will almost always return POLLOUT, even if there
 251                  * are no extra buffers available
 252                  *
 253                  * This happens, because packet_poll() calls datagram_poll()
 254                  * which checks the space left in the socket buffer and,
 255                  * in the case of packet_mmap, the default socket buffer length
 256                  * doesn't match the requested size for the tx_ring.
 257                  * As such, there is almost always space left in socket buffer,
 258                  * which doesn't seem to be correlated to the requested size
 259                  * for the tx_ring in packet_mmap.
 260                  *
 261                  * This results in poll() returning POLLOUT.
 262                  */
 263                 if (!tx_ring_status_available(ppd->tp_status))
 264                         break;
 265
 266                 /* copy the tx frame data */
 267                 pbuf = (uint8_t *) ppd + TPACKET2_HDRLEN -
 268                         sizeof(struct sockaddr_ll);
 269
 270                 struct rte_mbuf *tmp_mbuf = mbuf;
 271                 while (tmp_mbuf) {
 272                         uint16_t data_len = rte_pktmbuf_data_len(tmp_mbuf);
 273                         memcpy(pbuf, rte_pktmbuf_mtod(tmp_mbuf, void*), data_len);
 274                         pbuf += data_len;
 275                         tmp_mbuf = tmp_mbuf->next;
 276                 }
 277
 278                 ppd->tp_len = mbuf->pkt_len;
 279                 ppd->tp_snaplen = mbuf->pkt_len;
 280
 281                 /* release incoming frame and advance ring buffer */
 282                 ppd->tp_status = TP_STATUS_SEND_REQUEST;
 283                 if (++framenum >= framecount)
 284                         framenum = 0;
 285                 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
 286
 287                 num_tx++;
 288                 num_tx_bytes += mbuf->pkt_len;
 289                 rte_pktmbuf_free(mbuf);
 290         }
 291
 292         /* kick-off transmits */
 293         if (sendto(pkt_q->sockfd, NULL, 0, MSG_DONTWAIT, NULL, 0) == -1 &&
 294                         errno != ENOBUFS && errno != EAGAIN) {
 295                 /*
 296                  * In case of a ENOBUFS/EAGAIN error all of the enqueued
 297                  * packets will be considered successful even though only some
 298                  * are sent.
 299                  */
 300
 301                 num_tx = 0;
 302                 num_tx_bytes = 0;
 303         }
 304
 305         pkt_q->framenum = framenum;
 306         pkt_q->tx_pkts += num_tx;
 307         pkt_q->err_pkts += i - num_tx;
 308         pkt_q->tx_bytes += num_tx_bytes;
 309         return i;
 310 }
 311
 312 static int
 313 eth_dev_start(struct rte_eth_dev *dev)
 314 {
 315         dev->data->dev_link.link_status = RTE_ETH_LINK_UP;
 316         return 0;
 317 }
 318
 319 /*
 320  * This function gets called when the current port gets stopped.
 321  */
 322 static int
 323 eth_dev_stop(struct rte_eth_dev *dev)
 324 {
 325         unsigned i;
 326         int sockfd;
 327         struct pmd_internals *internals = dev->data->dev_private;
 328
 329         for (i = 0; i < internals->nb_queues; i++) {
 330                 sockfd = internals->rx_queue[i].sockfd;
 331                 if (sockfd != -1)
 332                         close(sockfd);
 333
 334                 /* Prevent use after free in case tx fd == rx fd */
 335                 if (sockfd != internals->tx_queue[i].sockfd) {
 336                         sockfd = internals->tx_queue[i].sockfd;
 337                         if (sockfd != -1)
 338                                 close(sockfd);
 339                 }
 340
 341                 internals->rx_queue[i].sockfd = -1;
 342                 internals->tx_queue[i].sockfd = -1;
 343         }
 344
 345         dev->data->dev_link.link_status = RTE_ETH_LINK_DOWN;
 346         return 0;
 347 }
 348
 349 static int
 350 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 351 {
 352         struct rte_eth_conf *dev_conf = &dev->data->dev_conf;
 353         const struct rte_eth_rxmode *rxmode = &dev_conf->rxmode;
 354         struct pmd_internals *internals = dev->data->dev_private;
 355
 356         internals->vlan_strip = !!(rxmode->offloads & RTE_ETH_RX_OFFLOAD_VLAN_STRIP);
 357         return 0;
 358 }
 359
 360 static int
 361 eth_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 362 {
 363         struct pmd_internals *internals = dev->data->dev_private;
 364
 365         dev_info->if_index = internals->if_index;
 366         dev_info->max_mac_addrs = 1;
 367         dev_info->max_rx_pktlen = RTE_ETHER_MAX_LEN;
 368         dev_info->max_rx_queues = (uint16_t)internals->nb_queues;
 369         dev_info->max_tx_queues = (uint16_t)internals->nb_queues;
 370         dev_info->min_rx_bufsize = 0;
 371         dev_info->tx_offload_capa = RTE_ETH_TX_OFFLOAD_MULTI_SEGS |
 372                 RTE_ETH_TX_OFFLOAD_VLAN_INSERT;
 373         dev_info->rx_offload_capa = RTE_ETH_RX_OFFLOAD_VLAN_STRIP;
 374
 375         return 0;
 376 }
 377
 378 static int
 379 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *igb_stats)
 380 {
 381         unsigned i, imax;
 382         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 383         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 384         const struct pmd_internals *internal = dev->data->dev_private;
 385
 386         imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
 387                 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
 388         for (i = 0; i < imax; i++) {
 389                 igb_stats->q_ipackets[i] = internal->rx_queue[i].rx_pkts;
 390                 igb_stats->q_ibytes[i] = internal->rx_queue[i].rx_bytes;
 391                 rx_total += igb_stats->q_ipackets[i];
 392                 rx_bytes_total += igb_stats->q_ibytes[i];
 393         }
 394
 395         imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
 396                 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
 397         for (i = 0; i < imax; i++) {
 398                 igb_stats->q_opackets[i] = internal->tx_queue[i].tx_pkts;
 399                 igb_stats->q_obytes[i] = internal->tx_queue[i].tx_bytes;
 400                 tx_total += igb_stats->q_opackets[i];
 401                 tx_err_total += internal->tx_queue[i].err_pkts;
 402                 tx_bytes_total += igb_stats->q_obytes[i];
 403         }
 404
 405         igb_stats->ipackets = rx_total;
 406         igb_stats->ibytes = rx_bytes_total;
 407         igb_stats->opackets = tx_total;
 408         igb_stats->oerrors = tx_err_total;
 409         igb_stats->obytes = tx_bytes_total;
 410         return 0;
 411 }
 412
 413 static int
 414 eth_stats_reset(struct rte_eth_dev *dev)
 415 {
 416         unsigned i;
 417         struct pmd_internals *internal = dev->data->dev_private;
 418
 419         for (i = 0; i < internal->nb_queues; i++) {
 420                 internal->rx_queue[i].rx_pkts = 0;
 421                 internal->rx_queue[i].rx_bytes = 0;
 422         }
 423
 424         for (i = 0; i < internal->nb_queues; i++) {
 425                 internal->tx_queue[i].tx_pkts = 0;
 426                 internal->tx_queue[i].err_pkts = 0;
 427                 internal->tx_queue[i].tx_bytes = 0;
 428         }
 429
 430         return 0;
 431 }
 432
 433 static int
 434 eth_dev_close(struct rte_eth_dev *dev)
 435 {
 436         struct pmd_internals *internals;
 437         struct tpacket_req *req;
 438         unsigned int q;
 439
 440         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 441                 return 0;
 442
 443         PMD_LOG(INFO, "Closing AF_PACKET ethdev on NUMA socket %u",
 444                 rte_socket_id());
 445
 446         internals = dev->data->dev_private;
 447         req = &internals->req;
 448         for (q = 0; q < internals->nb_queues; q++) {
 449                 munmap(internals->rx_queue[q].map,
 450                         2 * req->tp_block_size * req->tp_block_nr);
 451                 rte_free(internals->rx_queue[q].rd);
 452                 rte_free(internals->tx_queue[q].rd);
 453         }
 454         free(internals->if_name);
 455         rte_free(internals->rx_queue);
 456         rte_free(internals->tx_queue);
 457
 458         /* mac_addrs must not be freed alone because part of dev_private */
 459         dev->data->mac_addrs = NULL;
 460         return 0;
 461 }
 462
 463 static int
 464 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 465                 int wait_to_complete __rte_unused)
 466 {
 467         return 0;
 468 }
 469
 470 static int
 471 eth_rx_queue_setup(struct rte_eth_dev *dev,
 472                    uint16_t rx_queue_id,
 473                    uint16_t nb_rx_desc __rte_unused,
 474                    unsigned int socket_id __rte_unused,
 475                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 476                    struct rte_mempool *mb_pool)
 477 {
 478         struct pmd_internals *internals = dev->data->dev_private;
 479         struct pkt_rx_queue *pkt_q = &internals->rx_queue[rx_queue_id];
 480         unsigned int buf_size, data_size;
 481
 482         pkt_q->mb_pool = mb_pool;
 483
 484         /* Now get the space available for data in the mbuf */
 485         buf_size = rte_pktmbuf_data_room_size(pkt_q->mb_pool) -
 486                 RTE_PKTMBUF_HEADROOM;
 487         data_size = internals->req.tp_frame_size;
 488         data_size -= TPACKET2_HDRLEN - sizeof(struct sockaddr_ll);
 489
 490         if (data_size > buf_size) {
 491                 PMD_LOG(ERR,
 492                         "%s: %d bytes will not fit in mbuf (%d bytes)",
 493                         dev->device->name, data_size, buf_size);
 494                 return -ENOMEM;
 495         }
 496
 497         dev->data->rx_queues[rx_queue_id] = pkt_q;
 498         pkt_q->in_port = dev->data->port_id;
 499         pkt_q->vlan_strip = internals->vlan_strip;
 500
 501         return 0;
 502 }
 503
 504 static int
 505 eth_tx_queue_setup(struct rte_eth_dev *dev,
 506                    uint16_t tx_queue_id,
 507                    uint16_t nb_tx_desc __rte_unused,
 508                    unsigned int socket_id __rte_unused,
 509                    const struct rte_eth_txconf *tx_conf __rte_unused)
 510 {
 511
 512         struct pmd_internals *internals = dev->data->dev_private;
 513
 514         dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
 515         return 0;
 516 }
 517
 518 static int
 519 eth_dev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 520 {
 521         struct pmd_internals *internals = dev->data->dev_private;
 522         struct ifreq ifr = { .ifr_mtu = mtu };
 523         int ret;
 524         int s;
 525         unsigned int data_size = internals->req.tp_frame_size -
 526                                  TPACKET2_HDRLEN;
 527
 528         if (mtu > data_size)
 529                 return -EINVAL;
 530
 531         s = socket(PF_INET, SOCK_DGRAM, 0);
 532         if (s < 0)
 533                 return -EINVAL;
 534
 535         strlcpy(ifr.ifr_name, internals->if_name, IFNAMSIZ);
 536         ret = ioctl(s, SIOCSIFMTU, &ifr);
 537         close(s);
 538
 539         if (ret < 0)
 540                 return -EINVAL;
 541
 542         return 0;
 543 }
 544
 545 static int
 546 eth_dev_macaddr_set(struct rte_eth_dev *dev, struct rte_ether_addr *addr)
 547 {
 548         struct pmd_internals *internals = dev->data->dev_private;
 549         struct ifreq ifr = { };
 550         int sockfd = internals->rx_queue[0].sockfd;
 551         int ret;
 552
 553         if (sockfd == -1) {
 554                 PMD_LOG(ERR, "receive socket not found");
 555                 return -EINVAL;
 556         }
 557
 558         strlcpy(ifr.ifr_name, internals->if_name, IFNAMSIZ);
 559         ifr.ifr_hwaddr.sa_family = ARPHRD_ETHER;
 560         memcpy(ifr.ifr_hwaddr.sa_data, addr, sizeof(*addr));
 561         ret = ioctl(sockfd, SIOCSIFHWADDR, &ifr);
 562
 563         if (ret < 0) {
 564                 PMD_LOG_ERRNO(ERR, "ioctl(SIOCSIFHWADDR) failed");
 565                 return -EINVAL;
 566         }
 567
 568         return 0;
 569 }
 570
 571 static int
 572 eth_dev_change_flags(char *if_name, uint32_t flags, uint32_t mask)
 573 {
 574         struct ifreq ifr;
 575         int ret = 0;
 576         int s;
 577
 578         s = socket(PF_INET, SOCK_DGRAM, 0);
 579         if (s < 0)
 580                 return -errno;
 581
 582         strlcpy(ifr.ifr_name, if_name, IFNAMSIZ);
 583         if (ioctl(s, SIOCGIFFLAGS, &ifr) < 0) {
 584                 ret = -errno;
 585                 goto out;
 586         }
 587         ifr.ifr_flags &= mask;
 588         ifr.ifr_flags |= flags;
 589         if (ioctl(s, SIOCSIFFLAGS, &ifr) < 0) {
 590                 ret = -errno;
 591                 goto out;
 592         }
 593 out:
 594         close(s);
 595         return ret;
 596 }
 597
 598 static int
 599 eth_dev_promiscuous_enable(struct rte_eth_dev *dev)
 600 {
 601         struct pmd_internals *internals = dev->data->dev_private;
 602
 603         return eth_dev_change_flags(internals->if_name, IFF_PROMISC, ~0);
 604 }
 605
 606 static int
 607 eth_dev_promiscuous_disable(struct rte_eth_dev *dev)
 608 {
 609         struct pmd_internals *internals = dev->data->dev_private;
 610
 611         return eth_dev_change_flags(internals->if_name, 0, ~IFF_PROMISC);
 612 }
 613
 614 static const struct eth_dev_ops ops = {
 615         .dev_start = eth_dev_start,
 616         .dev_stop = eth_dev_stop,
 617         .dev_close = eth_dev_close,
 618         .dev_configure = eth_dev_configure,
 619         .dev_infos_get = eth_dev_info,
 620         .mac_addr_set = eth_dev_macaddr_set,
 621         .mtu_set = eth_dev_mtu_set,
 622         .promiscuous_enable = eth_dev_promiscuous_enable,
 623         .promiscuous_disable = eth_dev_promiscuous_disable,
 624         .rx_queue_setup = eth_rx_queue_setup,
 625         .tx_queue_setup = eth_tx_queue_setup,
 626         .link_update = eth_link_update,
 627         .stats_get = eth_stats_get,
 628         .stats_reset = eth_stats_reset,
 629 };
 630
 631 /*
 632  * Opens an AF_PACKET socket
 633  */
 634 static int
 635 open_packet_iface(const char *key __rte_unused,
 636                   const char *value __rte_unused,
 637                   void *extra_args)
 638 {
 639         int *sockfd = extra_args;
 640
 641         /* Open an AF_PACKET socket... */
 642         *sockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
 643         if (*sockfd == -1) {
 644                 PMD_LOG(ERR, "Could not open AF_PACKET socket");
 645                 return -1;
 646         }
 647
 648         return 0;
 649 }
 650
 651 static int
 652 rte_pmd_init_internals(struct rte_vdev_device *dev,
 653                        const int sockfd,
 654                        const unsigned nb_queues,
 655                        unsigned int blocksize,
 656                        unsigned int blockcnt,
 657                        unsigned int framesize,
 658                        unsigned int framecnt,
 659                        unsigned int qdisc_bypass,
 660                        struct pmd_internals **internals,
 661                        struct rte_eth_dev **eth_dev,
 662                        struct rte_kvargs *kvlist)
 663 {
 664         const char *name = rte_vdev_device_name(dev);
 665         const unsigned int numa_node = dev->device.numa_node;
 666         struct rte_eth_dev_data *data = NULL;
 667         struct rte_kvargs_pair *pair = NULL;
 668         struct ifreq ifr;
 669         size_t ifnamelen;
 670         unsigned k_idx;
 671         struct sockaddr_ll sockaddr;
 672         struct tpacket_req *req;
 673         struct pkt_rx_queue *rx_queue;
 674         struct pkt_tx_queue *tx_queue;
 675         int rc, tpver, discard;
 676         int qsockfd = -1;
 677         unsigned int i, q, rdsize;
 678 #if defined(PACKET_FANOUT)
 679         int fanout_arg;
 680 #endif
 681
 682         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 683                 pair = &kvlist->pairs[k_idx];
 684                 if (strstr(pair->key, ETH_AF_PACKET_IFACE_ARG) != NULL)
 685                         break;
 686         }
 687         if (pair == NULL) {
 688                 PMD_LOG(ERR,
 689                         "%s: no interface specified for AF_PACKET ethdev",
 690                         name);
 691                 return -1;
 692         }
 693
 694         PMD_LOG(INFO,
 695                 "%s: creating AF_PACKET-backed ethdev on numa socket %u",
 696                 name, numa_node);
 697
 698         *internals = rte_zmalloc_socket(name, sizeof(**internals),
 699                                         0, numa_node);
 700         if (*internals == NULL)
 701                 return -1;
 702
 703
 704         (*internals)->rx_queue = rte_calloc_socket("af_packet_rx",
 705                                                 nb_queues,
 706                                                 sizeof(struct pkt_rx_queue),
 707                                                 0, numa_node);
 708         (*internals)->tx_queue = rte_calloc_socket("af_packet_tx",
 709                                                 nb_queues,
 710                                                 sizeof(struct pkt_tx_queue),
 711                                                 0, numa_node);
 712         if (!(*internals)->rx_queue || !(*internals)->tx_queue) {
 713                 goto free_internals;
 714         }
 715
 716         for (q = 0; q < nb_queues; q++) {
 717                 (*internals)->rx_queue[q].map = MAP_FAILED;
 718                 (*internals)->tx_queue[q].map = MAP_FAILED;
 719                 (*internals)->rx_queue[q].sockfd = -1;
 720                 (*internals)->tx_queue[q].sockfd = -1;
 721         }
 722
 723         req = &((*internals)->req);
 724
 725         req->tp_block_size = blocksize;
 726         req->tp_block_nr = blockcnt;
 727         req->tp_frame_size = framesize;
 728         req->tp_frame_nr = framecnt;
 729
 730         ifnamelen = strlen(pair->value);
 731         if (ifnamelen < sizeof(ifr.ifr_name)) {
 732                 memcpy(ifr.ifr_name, pair->value, ifnamelen);
 733                 ifr.ifr_name[ifnamelen] = '\0';
 734         } else {
 735                 PMD_LOG(ERR,
 736                         "%s: I/F name too long (%s)",
 737                         name, pair->value);
 738                 goto free_internals;
 739         }
 740         if (ioctl(sockfd, SIOCGIFINDEX, &ifr) == -1) {
 741                 PMD_LOG_ERRNO(ERR, "%s: ioctl failed (SIOCGIFINDEX)", name);
 742                 goto free_internals;
 743         }
 744         (*internals)->if_name = strdup(pair->value);
 745         if ((*internals)->if_name == NULL)
 746                 goto free_internals;
 747         (*internals)->if_index = ifr.ifr_ifindex;
 748
 749         if (ioctl(sockfd, SIOCGIFHWADDR, &ifr) == -1) {
 750                 PMD_LOG_ERRNO(ERR, "%s: ioctl failed (SIOCGIFHWADDR)", name);
 751                 goto free_internals;
 752         }
 753         memcpy(&(*internals)->eth_addr, ifr.ifr_hwaddr.sa_data, ETH_ALEN);
 754
 755         memset(&sockaddr, 0, sizeof(sockaddr));
 756         sockaddr.sll_family = AF_PACKET;
 757         sockaddr.sll_protocol = htons(ETH_P_ALL);
 758         sockaddr.sll_ifindex = (*internals)->if_index;
 759
 760 #if defined(PACKET_FANOUT)
 761         fanout_arg = (getpid() ^ (*internals)->if_index) & 0xffff;
 762         fanout_arg |= (PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_DEFRAG) << 16;
 763 #if defined(PACKET_FANOUT_FLAG_ROLLOVER)
 764         fanout_arg |= PACKET_FANOUT_FLAG_ROLLOVER << 16;
 765 #endif
 766 #endif
 767
 768         for (q = 0; q < nb_queues; q++) {
 769                 /* Open an AF_PACKET socket for this queue... */
 770                 qsockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
 771                 if (qsockfd == -1) {
 772                         PMD_LOG_ERRNO(ERR,
 773                                 "%s: could not open AF_PACKET socket",
 774                                 name);
 775                         goto error;
 776                 }
 777
 778                 tpver = TPACKET_V2;
 779                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_VERSION,
 780                                 &tpver, sizeof(tpver));
 781                 if (rc == -1) {
 782                         PMD_LOG_ERRNO(ERR,
 783                                 "%s: could not set PACKET_VERSION on AF_PACKET socket for %s",
 784                                 name, pair->value);
 785                         goto error;
 786                 }
 787
 788                 discard = 1;
 789                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_LOSS,
 790                                 &discard, sizeof(discard));
 791                 if (rc == -1) {
 792                         PMD_LOG_ERRNO(ERR,
 793                                 "%s: could not set PACKET_LOSS on AF_PACKET socket for %s",
 794                                 name, pair->value);
 795                         goto error;
 796                 }
 797
 798                 if (qdisc_bypass) {
 799 #if defined(PACKET_QDISC_BYPASS)
 800                         rc = setsockopt(qsockfd, SOL_PACKET, PACKET_QDISC_BYPASS,
 801                                         &qdisc_bypass, sizeof(qdisc_bypass));
 802                         if (rc == -1) {
 803                                 PMD_LOG_ERRNO(ERR,
 804                                         "%s: could not set PACKET_QDISC_BYPASS on AF_PACKET socket for %s",
 805                                         name, pair->value);
 806                                 goto error;
 807                         }
 808 #endif
 809                 }
 810
 811                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_RX_RING, req, sizeof(*req));
 812                 if (rc == -1) {
 813                         PMD_LOG_ERRNO(ERR,
 814                                 "%s: could not set PACKET_RX_RING on AF_PACKET socket for %s",
 815                                 name, pair->value);
 816                         goto error;
 817                 }
 818
 819                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_TX_RING, req, sizeof(*req));
 820                 if (rc == -1) {
 821                         PMD_LOG_ERRNO(ERR,
 822                                 "%s: could not set PACKET_TX_RING on AF_PACKET "
 823                                 "socket for %s", name, pair->value);
 824                         goto error;
 825                 }
 826
 827                 rx_queue = &((*internals)->rx_queue[q]);
 828                 rx_queue->framecount = req->tp_frame_nr;
 829
 830                 rx_queue->map = mmap(NULL, 2 * req->tp_block_size * req->tp_block_nr,
 831                                     PROT_READ | PROT_WRITE, MAP_SHARED | MAP_LOCKED,
 832                                     qsockfd, 0);
 833                 if (rx_queue->map == MAP_FAILED) {
 834                         PMD_LOG_ERRNO(ERR,
 835                                 "%s: call to mmap failed on AF_PACKET socket for %s",
 836                                 name, pair->value);
 837                         goto error;
 838                 }
 839
 840                 /* rdsize is same for both Tx and Rx */
 841                 rdsize = req->tp_frame_nr * sizeof(*(rx_queue->rd));
 842
 843                 rx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
 844                 if (rx_queue->rd == NULL)
 845                         goto error;
 846                 for (i = 0; i < req->tp_frame_nr; ++i) {
 847                         rx_queue->rd[i].iov_base = rx_queue->map + (i * framesize);
 848                         rx_queue->rd[i].iov_len = req->tp_frame_size;
 849                 }
 850                 rx_queue->sockfd = qsockfd;
 851
 852                 tx_queue = &((*internals)->tx_queue[q]);
 853                 tx_queue->framecount = req->tp_frame_nr;
 854                 tx_queue->frame_data_size = req->tp_frame_size;
 855                 tx_queue->frame_data_size -= TPACKET2_HDRLEN -
 856                         sizeof(struct sockaddr_ll);
 857
 858                 tx_queue->map = rx_queue->map + req->tp_block_size * req->tp_block_nr;
 859
 860                 tx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
 861                 if (tx_queue->rd == NULL)
 862                         goto error;
 863                 for (i = 0; i < req->tp_frame_nr; ++i) {
 864                         tx_queue->rd[i].iov_base = tx_queue->map + (i * framesize);
 865                         tx_queue->rd[i].iov_len = req->tp_frame_size;
 866                 }
 867                 tx_queue->sockfd = qsockfd;
 868
 869                 rc = bind(qsockfd, (const struct sockaddr*)&sockaddr, sizeof(sockaddr));
 870                 if (rc == -1) {
 871                         PMD_LOG_ERRNO(ERR,
 872                                 "%s: could not bind AF_PACKET socket to %s",
 873                                 name, pair->value);
 874                         goto error;
 875                 }
 876
 877 #if defined(PACKET_FANOUT)
 878                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_FANOUT,
 879                                 &fanout_arg, sizeof(fanout_arg));
 880                 if (rc == -1) {
 881                         PMD_LOG_ERRNO(ERR,
 882                                 "%s: could not set PACKET_FANOUT on AF_PACKET socket for %s",
 883                                 name, pair->value);
 884                         goto error;
 885                 }
 886 #endif
 887         }
 888
 889         /* reserve an ethdev entry */
 890         *eth_dev = rte_eth_vdev_allocate(dev, 0);
 891         if (*eth_dev == NULL)
 892                 goto error;
 893
 894         /*
 895          * now put it all together
 896          * - store queue data in internals,
 897          * - store numa_node in eth_dev
 898          * - point eth_dev_data to internals
 899          * - and point eth_dev structure to new eth_dev_data structure
 900          */
 901
 902         (*internals)->nb_queues = nb_queues;
 903
 904         data = (*eth_dev)->data;
 905         data->dev_private = *internals;
 906         data->nb_rx_queues = (uint16_t)nb_queues;
 907         data->nb_tx_queues = (uint16_t)nb_queues;
 908         data->dev_link = pmd_link;
 909         data->mac_addrs = &(*internals)->eth_addr;
 910         data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
 911
 912         (*eth_dev)->dev_ops = &ops;
 913
 914         return 0;
 915
 916 error:
 917         if (qsockfd != -1)
 918                 close(qsockfd);
 919         for (q = 0; q < nb_queues; q++) {
 920                 if ((*internals)->rx_queue[q].map != MAP_FAILED)
 921                         munmap((*internals)->rx_queue[q].map,
 922                                2 * req->tp_block_size * req->tp_block_nr);
 923
 924                 rte_free((*internals)->rx_queue[q].rd);
 925                 rte_free((*internals)->tx_queue[q].rd);
 926                 if (((*internals)->rx_queue[q].sockfd >= 0) &&
 927                         ((*internals)->rx_queue[q].sockfd != qsockfd))
 928                         close((*internals)->rx_queue[q].sockfd);
 929         }
 930 free_internals:
 931         rte_free((*internals)->rx_queue);
 932         rte_free((*internals)->tx_queue);
 933         free((*internals)->if_name);
 934         rte_free(*internals);
 935         return -1;
 936 }
 937
 938 static int
 939 rte_eth_from_packet(struct rte_vdev_device *dev,
 940                     int const *sockfd,
 941                     struct rte_kvargs *kvlist)
 942 {
 943         const char *name = rte_vdev_device_name(dev);
 944         struct pmd_internals *internals = NULL;
 945         struct rte_eth_dev *eth_dev = NULL;
 946         struct rte_kvargs_pair *pair = NULL;
 947         unsigned k_idx;
 948         unsigned int blockcount;
 949         unsigned int blocksize;
 950         unsigned int framesize = DFLT_FRAME_SIZE;
 951         unsigned int framecount = DFLT_FRAME_COUNT;
 952         unsigned int qpairs = 1;
 953         unsigned int qdisc_bypass = 1;
 954
 955         /* do some parameter checking */
 956         if (*sockfd < 0)
 957                 return -1;
 958
 959         blocksize = getpagesize();
 960
 961         /*
 962          * Walk arguments for configurable settings
 963          */
 964         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 965                 pair = &kvlist->pairs[k_idx];
 966                 if (strstr(pair->key, ETH_AF_PACKET_NUM_Q_ARG) != NULL) {
 967                         qpairs = atoi(pair->value);
 968                         if (qpairs < 1) {
 969                                 PMD_LOG(ERR,
 970                                         "%s: invalid qpairs value",
 971                                         name);
 972                                 return -1;
 973                         }
 974                         continue;
 975                 }
 976                 if (strstr(pair->key, ETH_AF_PACKET_BLOCKSIZE_ARG) != NULL) {
 977                         blocksize = atoi(pair->value);
 978                         if (!blocksize) {
 979                                 PMD_LOG(ERR,
 980                                         "%s: invalid blocksize value",
 981                                         name);
 982                                 return -1;
 983                         }
 984                         continue;
 985                 }
 986                 if (strstr(pair->key, ETH_AF_PACKET_FRAMESIZE_ARG) != NULL) {
 987                         framesize = atoi(pair->value);
 988                         if (!framesize) {
 989                                 PMD_LOG(ERR,
 990                                         "%s: invalid framesize value",
 991                                         name);
 992                                 return -1;
 993                         }
 994                         continue;
 995                 }
 996                 if (strstr(pair->key, ETH_AF_PACKET_FRAMECOUNT_ARG) != NULL) {
 997                         framecount = atoi(pair->value);
 998                         if (!framecount) {
 999                                 PMD_LOG(ERR,
1000                                         "%s: invalid framecount value",
1001                                         name);
1002                                 return -1;
1003                         }
1004                         continue;
1005                 }
1006                 if (strstr(pair->key, ETH_AF_PACKET_QDISC_BYPASS_ARG) != NULL) {
1007                         qdisc_bypass = atoi(pair->value);
1008                         if (qdisc_bypass > 1) {
1009                                 PMD_LOG(ERR,
1010                                         "%s: invalid bypass value",
1011                                         name);
1012                                 return -1;
1013                         }
1014                         continue;
1015                 }
1016         }
1017
1018         if (framesize > blocksize) {
1019                 PMD_LOG(ERR,
1020                         "%s: AF_PACKET MMAP frame size exceeds block size!",
1021                         name);
1022                 return -1;
1023         }
1024
1025         blockcount = framecount / (blocksize / framesize);
1026         if (!blockcount) {
1027                 PMD_LOG(ERR,
1028                         "%s: invalid AF_PACKET MMAP parameters", name);
1029                 return -1;
1030         }
1031
1032         PMD_LOG(INFO, "%s: AF_PACKET MMAP parameters:", name);
1033         PMD_LOG(INFO, "%s:\tblock size %d", name, blocksize);
1034         PMD_LOG(INFO, "%s:\tblock count %d", name, blockcount);
1035         PMD_LOG(INFO, "%s:\tframe size %d", name, framesize);
1036         PMD_LOG(INFO, "%s:\tframe count %d", name, framecount);
1037
1038         if (rte_pmd_init_internals(dev, *sockfd, qpairs,
1039                                    blocksize, blockcount,
1040                                    framesize, framecount,
1041                                    qdisc_bypass,
1042                                    &internals, &eth_dev,
1043                                    kvlist) < 0)
1044                 return -1;
1045
1046         eth_dev->rx_pkt_burst = eth_af_packet_rx;
1047         eth_dev->tx_pkt_burst = eth_af_packet_tx;
1048
1049         rte_eth_dev_probing_finish(eth_dev);
1050         return 0;
1051 }
1052
1053 static int
1054 rte_pmd_af_packet_probe(struct rte_vdev_device *dev)
1055 {
1056         int ret = 0;
1057         struct rte_kvargs *kvlist;
1058         int sockfd = -1;
1059         struct rte_eth_dev *eth_dev;
1060         const char *name = rte_vdev_device_name(dev);
1061
1062         PMD_LOG(INFO, "Initializing pmd_af_packet for %s", name);
1063
1064         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1065                 eth_dev = rte_eth_dev_attach_secondary(name);
1066                 if (!eth_dev) {
1067                         PMD_LOG(ERR, "Failed to probe %s", name);
1068                         return -1;
1069                 }
1070                 /* TODO: request info from primary to set up Rx and Tx */
1071                 eth_dev->dev_ops = &ops;
1072                 eth_dev->device = &dev->device;
1073                 rte_eth_dev_probing_finish(eth_dev);
1074                 return 0;
1075         }
1076
1077         kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_arguments);
1078         if (kvlist == NULL) {
1079                 ret = -1;
1080                 goto exit;
1081         }
1082
1083         /*
1084          * If iface argument is passed we open the NICs and use them for
1085          * reading / writing
1086          */
1087         if (rte_kvargs_count(kvlist, ETH_AF_PACKET_IFACE_ARG) == 1) {
1088
1089                 ret = rte_kvargs_process(kvlist, ETH_AF_PACKET_IFACE_ARG,
1090                                          &open_packet_iface, &sockfd);
1091                 if (ret < 0)
1092                         goto exit;
1093         }
1094
1095         if (dev->device.numa_node == SOCKET_ID_ANY)
1096                 dev->device.numa_node = rte_socket_id();
1097
1098         ret = rte_eth_from_packet(dev, &sockfd, kvlist);
1099         close(sockfd); /* no longer needed */
1100
1101 exit:
1102         rte_kvargs_free(kvlist);
1103         return ret;
1104 }
1105
1106 static int
1107 rte_pmd_af_packet_remove(struct rte_vdev_device *dev)
1108 {
1109         struct rte_eth_dev *eth_dev;
1110
1111         if (dev == NULL)
1112                 return -1;
1113
1114         /* find the ethdev entry */
1115         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1116         if (eth_dev == NULL)
1117                 return 0; /* port already released */
1118
1119         eth_dev_close(eth_dev);
1120         rte_eth_dev_release_port(eth_dev);
1121
1122         return 0;
1123 }
1124
1125 static struct rte_vdev_driver pmd_af_packet_drv = {
1126         .probe = rte_pmd_af_packet_probe,
1127         .remove = rte_pmd_af_packet_remove,
1128 };
1129
1130 RTE_PMD_REGISTER_VDEV(net_af_packet, pmd_af_packet_drv);
1131 RTE_PMD_REGISTER_ALIAS(net_af_packet, eth_af_packet);
1132 RTE_PMD_REGISTER_PARAM_STRING(net_af_packet,
1133         "iface=<string> "
1134         "qpairs=<int> "
1135         "blocksz=<int> "
1136         "framesz=<int> "
1137         "framecnt=<int> "
1138         "qdisc_bypass=<0|1>");