linux/drivers/net/macvtap.c
<<
>>
Prefs
   1#include <linux/etherdevice.h>
   2#include <linux/if_macvlan.h>
   3#include <linux/if_vlan.h>
   4#include <linux/interrupt.h>
   5#include <linux/nsproxy.h>
   6#include <linux/compat.h>
   7#include <linux/if_tun.h>
   8#include <linux/module.h>
   9#include <linux/skbuff.h>
  10#include <linux/cache.h>
  11#include <linux/sched.h>
  12#include <linux/types.h>
  13#include <linux/slab.h>
  14#include <linux/init.h>
  15#include <linux/wait.h>
  16#include <linux/cdev.h>
  17#include <linux/idr.h>
  18#include <linux/fs.h>
  19
  20#include <net/net_namespace.h>
  21#include <net/rtnetlink.h>
  22#include <net/sock.h>
  23#include <linux/virtio_net.h>
  24
  25/*
  26 * A macvtap queue is the central object of this driver, it connects
  27 * an open character device to a macvlan interface. There can be
  28 * multiple queues on one interface, which map back to queues
  29 * implemented in hardware on the underlying device.
  30 *
  31 * macvtap_proto is used to allocate queues through the sock allocation
  32 * mechanism.
  33 *
  34 * TODO: multiqueue support is currently not implemented, even though
  35 * macvtap is basically prepared for that. We will need to add this
  36 * here as well as in virtio-net and qemu to get line rate on 10gbit
  37 * adapters from a guest.
  38 */
  39struct macvtap_queue {
  40        struct sock sk;
  41        struct socket sock;
  42        struct socket_wq wq;
  43        int vnet_hdr_sz;
  44        struct macvlan_dev __rcu *vlan;
  45        struct file *file;
  46        unsigned int flags;
  47};
  48
  49static struct proto macvtap_proto = {
  50        .name = "macvtap",
  51        .owner = THIS_MODULE,
  52        .obj_size = sizeof (struct macvtap_queue),
  53};
  54
  55/*
  56 * Variables for dealing with macvtaps device numbers.
  57 */
  58static dev_t macvtap_major;
  59#define MACVTAP_NUM_DEVS (1U << MINORBITS)
  60static DEFINE_MUTEX(minor_lock);
  61static DEFINE_IDR(minor_idr);
  62
  63#define GOODCOPY_LEN 128
  64static struct class *macvtap_class;
  65static struct cdev macvtap_cdev;
  66
  67static const struct proto_ops macvtap_socket_ops;
  68
  69/*
  70 * RCU usage:
  71 * The macvtap_queue and the macvlan_dev are loosely coupled, the
  72 * pointers from one to the other can only be read while rcu_read_lock
  73 * or macvtap_lock is held.
  74 *
  75 * Both the file and the macvlan_dev hold a reference on the macvtap_queue
  76 * through sock_hold(&q->sk). When the macvlan_dev goes away first,
  77 * q->vlan becomes inaccessible. When the files gets closed,
  78 * macvtap_get_queue() fails.
  79 *
  80 * There may still be references to the struct sock inside of the
  81 * queue from outbound SKBs, but these never reference back to the
  82 * file or the dev. The data structure is freed through __sk_free
  83 * when both our references and any pending SKBs are gone.
  84 */
  85static DEFINE_SPINLOCK(macvtap_lock);
  86
  87/*
  88 * get_slot: return a [unused/occupied] slot in vlan->taps[]:
  89 *      - if 'q' is NULL, return the first empty slot;
  90 *      - otherwise, return the slot this pointer occupies.
  91 */
  92static int get_slot(struct macvlan_dev *vlan, struct macvtap_queue *q)
  93{
  94        int i;
  95
  96        for (i = 0; i < MAX_MACVTAP_QUEUES; i++) {
  97                if (rcu_dereference_protected(vlan->taps[i],
  98                                              lockdep_is_held(&macvtap_lock)) == q)
  99                        return i;
 100        }
 101
 102        /* Should never happen */
 103        BUG_ON(1);
 104}
 105
 106static int macvtap_set_queue(struct net_device *dev, struct file *file,
 107                                struct macvtap_queue *q)
 108{
 109        struct macvlan_dev *vlan = netdev_priv(dev);
 110        int index;
 111        int err = -EBUSY;
 112
 113        spin_lock(&macvtap_lock);
 114        if (vlan->numvtaps == MAX_MACVTAP_QUEUES)
 115                goto out;
 116
 117        err = 0;
 118        index = get_slot(vlan, NULL);
 119        rcu_assign_pointer(q->vlan, vlan);
 120        rcu_assign_pointer(vlan->taps[index], q);
 121        sock_hold(&q->sk);
 122
 123        q->file = file;
 124        file->private_data = q;
 125
 126        vlan->numvtaps++;
 127
 128out:
 129        spin_unlock(&macvtap_lock);
 130        return err;
 131}
 132
 133/*
 134 * The file owning the queue got closed, give up both
 135 * the reference that the files holds as well as the
 136 * one from the macvlan_dev if that still exists.
 137 *
 138 * Using the spinlock makes sure that we don't get
 139 * to the queue again after destroying it.
 140 */
 141static void macvtap_put_queue(struct macvtap_queue *q)
 142{
 143        struct macvlan_dev *vlan;
 144
 145        spin_lock(&macvtap_lock);
 146        vlan = rcu_dereference_protected(q->vlan,
 147                                         lockdep_is_held(&macvtap_lock));
 148        if (vlan) {
 149                int index = get_slot(vlan, q);
 150
 151                RCU_INIT_POINTER(vlan->taps[index], NULL);
 152                RCU_INIT_POINTER(q->vlan, NULL);
 153                sock_put(&q->sk);
 154                --vlan->numvtaps;
 155        }
 156
 157        spin_unlock(&macvtap_lock);
 158
 159        synchronize_rcu();
 160        sock_put(&q->sk);
 161}
 162
 163/*
 164 * Select a queue based on the rxq of the device on which this packet
 165 * arrived. If the incoming device is not mq, calculate a flow hash
 166 * to select a queue. If all fails, find the first available queue.
 167 * Cache vlan->numvtaps since it can become zero during the execution
 168 * of this function.
 169 */
 170static struct macvtap_queue *macvtap_get_queue(struct net_device *dev,
 171                                               struct sk_buff *skb)
 172{
 173        struct macvlan_dev *vlan = netdev_priv(dev);
 174        struct macvtap_queue *tap = NULL;
 175        int numvtaps = vlan->numvtaps;
 176        __u32 rxq;
 177
 178        if (!numvtaps)
 179                goto out;
 180
 181        /* Check if we can use flow to select a queue */
 182        rxq = skb_get_rxhash(skb);
 183        if (rxq) {
 184                tap = rcu_dereference(vlan->taps[rxq % numvtaps]);
 185                if (tap)
 186                        goto out;
 187        }
 188
 189        if (likely(skb_rx_queue_recorded(skb))) {
 190                rxq = skb_get_rx_queue(skb);
 191
 192                while (unlikely(rxq >= numvtaps))
 193                        rxq -= numvtaps;
 194
 195                tap = rcu_dereference(vlan->taps[rxq]);
 196                if (tap)
 197                        goto out;
 198        }
 199
 200        /* Everything failed - find first available queue */
 201        for (rxq = 0; rxq < MAX_MACVTAP_QUEUES; rxq++) {
 202                tap = rcu_dereference(vlan->taps[rxq]);
 203                if (tap)
 204                        break;
 205        }
 206
 207out:
 208        return tap;
 209}
 210
 211/*
 212 * The net_device is going away, give up the reference
 213 * that it holds on all queues and safely set the pointer
 214 * from the queues to NULL.
 215 */
 216static void macvtap_del_queues(struct net_device *dev)
 217{
 218        struct macvlan_dev *vlan = netdev_priv(dev);
 219        struct macvtap_queue *q, *qlist[MAX_MACVTAP_QUEUES];
 220        int i, j = 0;
 221
 222        /* macvtap_put_queue can free some slots, so go through all slots */
 223        spin_lock(&macvtap_lock);
 224        for (i = 0; i < MAX_MACVTAP_QUEUES && vlan->numvtaps; i++) {
 225                q = rcu_dereference_protected(vlan->taps[i],
 226                                              lockdep_is_held(&macvtap_lock));
 227                if (q) {
 228                        qlist[j++] = q;
 229                        RCU_INIT_POINTER(vlan->taps[i], NULL);
 230                        RCU_INIT_POINTER(q->vlan, NULL);
 231                        vlan->numvtaps--;
 232                }
 233        }
 234        BUG_ON(vlan->numvtaps != 0);
 235        /* guarantee that any future macvtap_set_queue will fail */
 236        vlan->numvtaps = MAX_MACVTAP_QUEUES;
 237        spin_unlock(&macvtap_lock);
 238
 239        synchronize_rcu();
 240
 241        for (--j; j >= 0; j--)
 242                sock_put(&qlist[j]->sk);
 243}
 244
 245/*
 246 * Forward happens for data that gets sent from one macvlan
 247 * endpoint to another one in bridge mode. We just take
 248 * the skb and put it into the receive queue.
 249 */
 250static int macvtap_forward(struct net_device *dev, struct sk_buff *skb)
 251{
 252        struct macvtap_queue *q = macvtap_get_queue(dev, skb);
 253        if (!q)
 254                goto drop;
 255
 256        if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len)
 257                goto drop;
 258
 259        skb_queue_tail(&q->sk.sk_receive_queue, skb);
 260        wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND);
 261        return NET_RX_SUCCESS;
 262
 263drop:
 264        kfree_skb(skb);
 265        return NET_RX_DROP;
 266}
 267
 268/*
 269 * Receive is for data from the external interface (lowerdev),
 270 * in case of macvtap, we can treat that the same way as
 271 * forward, which macvlan cannot.
 272 */
 273static int macvtap_receive(struct sk_buff *skb)
 274{
 275        skb_push(skb, ETH_HLEN);
 276        return macvtap_forward(skb->dev, skb);
 277}
 278
 279static int macvtap_get_minor(struct macvlan_dev *vlan)
 280{
 281        int retval = -ENOMEM;
 282
 283        mutex_lock(&minor_lock);
 284        retval = idr_alloc(&minor_idr, vlan, 1, MACVTAP_NUM_DEVS, GFP_KERNEL);
 285        if (retval >= 0) {
 286                vlan->minor = retval;
 287        } else if (retval == -ENOSPC) {
 288                printk(KERN_ERR "too many macvtap devices\n");
 289                retval = -EINVAL;
 290        }
 291        mutex_unlock(&minor_lock);
 292        return retval < 0 ? retval : 0;
 293}
 294
 295static void macvtap_free_minor(struct macvlan_dev *vlan)
 296{
 297        mutex_lock(&minor_lock);
 298        if (vlan->minor) {
 299                idr_remove(&minor_idr, vlan->minor);
 300                vlan->minor = 0;
 301        }
 302        mutex_unlock(&minor_lock);
 303}
 304
 305static struct net_device *dev_get_by_macvtap_minor(int minor)
 306{
 307        struct net_device *dev = NULL;
 308        struct macvlan_dev *vlan;
 309
 310        mutex_lock(&minor_lock);
 311        vlan = idr_find(&minor_idr, minor);
 312        if (vlan) {
 313                dev = vlan->dev;
 314                dev_hold(dev);
 315        }
 316        mutex_unlock(&minor_lock);
 317        return dev;
 318}
 319
 320static int macvtap_newlink(struct net *src_net,
 321                           struct net_device *dev,
 322                           struct nlattr *tb[],
 323                           struct nlattr *data[])
 324{
 325        /* Don't put anything that may fail after macvlan_common_newlink
 326         * because we can't undo what it does.
 327         */
 328        return macvlan_common_newlink(src_net, dev, tb, data,
 329                                      macvtap_receive, macvtap_forward);
 330}
 331
 332static void macvtap_dellink(struct net_device *dev,
 333                            struct list_head *head)
 334{
 335        macvtap_del_queues(dev);
 336        macvlan_dellink(dev, head);
 337}
 338
 339static void macvtap_setup(struct net_device *dev)
 340{
 341        macvlan_common_setup(dev);
 342        dev->tx_queue_len = TUN_READQ_SIZE;
 343}
 344
 345static struct rtnl_link_ops macvtap_link_ops __read_mostly = {
 346        .kind           = "macvtap",
 347        .setup          = macvtap_setup,
 348        .newlink        = macvtap_newlink,
 349        .dellink        = macvtap_dellink,
 350};
 351
 352
 353static void macvtap_sock_write_space(struct sock *sk)
 354{
 355        wait_queue_head_t *wqueue;
 356
 357        if (!sock_writeable(sk) ||
 358            !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
 359                return;
 360
 361        wqueue = sk_sleep(sk);
 362        if (wqueue && waitqueue_active(wqueue))
 363                wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND);
 364}
 365
 366static void macvtap_sock_destruct(struct sock *sk)
 367{
 368        skb_queue_purge(&sk->sk_receive_queue);
 369}
 370
 371static int macvtap_open(struct inode *inode, struct file *file)
 372{
 373        struct net *net = current->nsproxy->net_ns;
 374        struct net_device *dev = dev_get_by_macvtap_minor(iminor(inode));
 375        struct macvtap_queue *q;
 376        int err;
 377
 378        err = -ENODEV;
 379        if (!dev)
 380                goto out;
 381
 382        err = -ENOMEM;
 383        q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
 384                                             &macvtap_proto);
 385        if (!q)
 386                goto out;
 387
 388        q->sock.wq = &q->wq;
 389        init_waitqueue_head(&q->wq.wait);
 390        q->sock.type = SOCK_RAW;
 391        q->sock.state = SS_CONNECTED;
 392        q->sock.file = file;
 393        q->sock.ops = &macvtap_socket_ops;
 394        sock_init_data(&q->sock, &q->sk);
 395        q->sk.sk_write_space = macvtap_sock_write_space;
 396        q->sk.sk_destruct = macvtap_sock_destruct;
 397        q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
 398        q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
 399
 400        /*
 401         * so far only KVM virtio_net uses macvtap, enable zero copy between
 402         * guest kernel and host kernel when lower device supports zerocopy
 403         *
 404         * The macvlan supports zerocopy iff the lower device supports zero
 405         * copy so we don't have to look at the lower device directly.
 406         */
 407        if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG))
 408                sock_set_flag(&q->sk, SOCK_ZEROCOPY);
 409
 410        err = macvtap_set_queue(dev, file, q);
 411        if (err)
 412                sock_put(&q->sk);
 413
 414out:
 415        if (dev)
 416                dev_put(dev);
 417
 418        return err;
 419}
 420
 421static int macvtap_release(struct inode *inode, struct file *file)
 422{
 423        struct macvtap_queue *q = file->private_data;
 424        macvtap_put_queue(q);
 425        return 0;
 426}
 427
 428static unsigned int macvtap_poll(struct file *file, poll_table * wait)
 429{
 430        struct macvtap_queue *q = file->private_data;
 431        unsigned int mask = POLLERR;
 432
 433        if (!q)
 434                goto out;
 435
 436        mask = 0;
 437        poll_wait(file, &q->wq.wait, wait);
 438
 439        if (!skb_queue_empty(&q->sk.sk_receive_queue))
 440                mask |= POLLIN | POLLRDNORM;
 441
 442        if (sock_writeable(&q->sk) ||
 443            (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) &&
 444             sock_writeable(&q->sk)))
 445                mask |= POLLOUT | POLLWRNORM;
 446
 447out:
 448        return mask;
 449}
 450
 451static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad,
 452                                                size_t len, size_t linear,
 453                                                int noblock, int *err)
 454{
 455        struct sk_buff *skb;
 456
 457        /* Under a page?  Don't bother with paged skb. */
 458        if (prepad + len < PAGE_SIZE || !linear)
 459                linear = len;
 460
 461        skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
 462                                   err);
 463        if (!skb)
 464                return NULL;
 465
 466        skb_reserve(skb, prepad);
 467        skb_put(skb, linear);
 468        skb->data_len = len - linear;
 469        skb->len += len - linear;
 470
 471        return skb;
 472}
 473
 474/* set skb frags from iovec, this can move to core network code for reuse */
 475static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
 476                                  int offset, size_t count)
 477{
 478        int len = iov_length(from, count) - offset;
 479        int copy = skb_headlen(skb);
 480        int size, offset1 = 0;
 481        int i = 0;
 482
 483        /* Skip over from offset */
 484        while (count && (offset >= from->iov_len)) {
 485                offset -= from->iov_len;
 486                ++from;
 487                --count;
 488        }
 489
 490        /* copy up to skb headlen */
 491        while (count && (copy > 0)) {
 492                size = min_t(unsigned int, copy, from->iov_len - offset);
 493                if (copy_from_user(skb->data + offset1, from->iov_base + offset,
 494                                   size))
 495                        return -EFAULT;
 496                if (copy > size) {
 497                        ++from;
 498                        --count;
 499                        offset = 0;
 500                } else
 501                        offset += size;
 502                copy -= size;
 503                offset1 += size;
 504        }
 505
 506        if (len == offset1)
 507                return 0;
 508
 509        while (count--) {
 510                struct page *page[MAX_SKB_FRAGS];
 511                int num_pages;
 512                unsigned long base;
 513                unsigned long truesize;
 514
 515                len = from->iov_len - offset;
 516                if (!len) {
 517                        offset = 0;
 518                        ++from;
 519                        continue;
 520                }
 521                base = (unsigned long)from->iov_base + offset;
 522                size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
 523                if (i + size > MAX_SKB_FRAGS)
 524                        return -EMSGSIZE;
 525                num_pages = get_user_pages_fast(base, size, 0, &page[i]);
 526                if (num_pages != size) {
 527                        int j;
 528
 529                        for (j = 0; j < num_pages; j++)
 530                                put_page(page[i + j]);
 531                        return -EFAULT;
 532                }
 533                truesize = size * PAGE_SIZE;
 534                skb->data_len += len;
 535                skb->len += len;
 536                skb->truesize += truesize;
 537                atomic_add(truesize, &skb->sk->sk_wmem_alloc);
 538                while (len) {
 539                        int off = base & ~PAGE_MASK;
 540                        int size = min_t(int, len, PAGE_SIZE - off);
 541                        __skb_fill_page_desc(skb, i, page[i], off, size);
 542                        skb_shinfo(skb)->nr_frags++;
 543                        /* increase sk_wmem_alloc */
 544                        base += size;
 545                        len -= size;
 546                        i++;
 547                }
 548                offset = 0;
 549                ++from;
 550        }
 551        return 0;
 552}
 553
 554/*
 555 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
 556 * be shared with the tun/tap driver.
 557 */
 558static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb,
 559                                     struct virtio_net_hdr *vnet_hdr)
 560{
 561        unsigned short gso_type = 0;
 562        if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
 563                switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
 564                case VIRTIO_NET_HDR_GSO_TCPV4:
 565                        gso_type = SKB_GSO_TCPV4;
 566                        break;
 567                case VIRTIO_NET_HDR_GSO_TCPV6:
 568                        gso_type = SKB_GSO_TCPV6;
 569                        break;
 570                case VIRTIO_NET_HDR_GSO_UDP:
 571                        gso_type = SKB_GSO_UDP;
 572                        break;
 573                default:
 574                        return -EINVAL;
 575                }
 576
 577                if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
 578                        gso_type |= SKB_GSO_TCP_ECN;
 579
 580                if (vnet_hdr->gso_size == 0)
 581                        return -EINVAL;
 582        }
 583
 584        if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
 585                if (!skb_partial_csum_set(skb, vnet_hdr->csum_start,
 586                                          vnet_hdr->csum_offset))
 587                        return -EINVAL;
 588        }
 589
 590        if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
 591                skb_shinfo(skb)->gso_size = vnet_hdr->gso_size;
 592                skb_shinfo(skb)->gso_type = gso_type;
 593
 594                /* Header must be checked, and gso_segs computed. */
 595                skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
 596                skb_shinfo(skb)->gso_segs = 0;
 597        }
 598        return 0;
 599}
 600
 601static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb,
 602                                   struct virtio_net_hdr *vnet_hdr)
 603{
 604        memset(vnet_hdr, 0, sizeof(*vnet_hdr));
 605
 606        if (skb_is_gso(skb)) {
 607                struct skb_shared_info *sinfo = skb_shinfo(skb);
 608
 609                /* This is a hint as to how much should be linear. */
 610                vnet_hdr->hdr_len = skb_headlen(skb);
 611                vnet_hdr->gso_size = sinfo->gso_size;
 612                if (sinfo->gso_type & SKB_GSO_TCPV4)
 613                        vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
 614                else if (sinfo->gso_type & SKB_GSO_TCPV6)
 615                        vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
 616                else if (sinfo->gso_type & SKB_GSO_UDP)
 617                        vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
 618                else
 619                        BUG();
 620                if (sinfo->gso_type & SKB_GSO_TCP_ECN)
 621                        vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
 622        } else
 623                vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
 624
 625        if (skb->ip_summed == CHECKSUM_PARTIAL) {
 626                vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 627                vnet_hdr->csum_start = skb_checksum_start_offset(skb);
 628                vnet_hdr->csum_offset = skb->csum_offset;
 629        } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
 630                vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID;
 631        } /* else everything is zero */
 632
 633        return 0;
 634}
 635
 636
 637/* Get packet from user space buffer */
 638static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m,
 639                                const struct iovec *iv, unsigned long total_len,
 640                                size_t count, int noblock)
 641{
 642        struct sk_buff *skb;
 643        struct macvlan_dev *vlan;
 644        unsigned long len = total_len;
 645        int err;
 646        struct virtio_net_hdr vnet_hdr = { 0 };
 647        int vnet_hdr_len = 0;
 648        int copylen = 0;
 649        bool zerocopy = false;
 650
 651        if (q->flags & IFF_VNET_HDR) {
 652                vnet_hdr_len = q->vnet_hdr_sz;
 653
 654                err = -EINVAL;
 655                if (len < vnet_hdr_len)
 656                        goto err;
 657                len -= vnet_hdr_len;
 658
 659                err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0,
 660                                           sizeof(vnet_hdr));
 661                if (err < 0)
 662                        goto err;
 663                if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
 664                     vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
 665                                                        vnet_hdr.hdr_len)
 666                        vnet_hdr.hdr_len = vnet_hdr.csum_start +
 667                                                vnet_hdr.csum_offset + 2;
 668                err = -EINVAL;
 669                if (vnet_hdr.hdr_len > len)
 670                        goto err;
 671        }
 672
 673        err = -EINVAL;
 674        if (unlikely(len < ETH_HLEN))
 675                goto err;
 676
 677        err = -EMSGSIZE;
 678        if (unlikely(count > UIO_MAXIOV))
 679                goto err;
 680
 681        if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY))
 682                zerocopy = true;
 683
 684        if (zerocopy) {
 685                /* Userspace may produce vectors with count greater than
 686                 * MAX_SKB_FRAGS, so we need to linearize parts of the skb
 687                 * to let the rest of data to be fit in the frags.
 688                 */
 689                if (count > MAX_SKB_FRAGS) {
 690                        copylen = iov_length(iv, count - MAX_SKB_FRAGS);
 691                        if (copylen < vnet_hdr_len)
 692                                copylen = 0;
 693                        else
 694                                copylen -= vnet_hdr_len;
 695                }
 696                /* There are 256 bytes to be copied in skb, so there is enough
 697                 * room for skb expand head in case it is used.
 698                 * The rest buffer is mapped from userspace.
 699                 */
 700                if (copylen < vnet_hdr.hdr_len)
 701                        copylen = vnet_hdr.hdr_len;
 702                if (!copylen)
 703                        copylen = GOODCOPY_LEN;
 704        } else
 705                copylen = len;
 706
 707        skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, copylen,
 708                                vnet_hdr.hdr_len, noblock, &err);
 709        if (!skb)
 710                goto err;
 711
 712        if (zerocopy)
 713                err = zerocopy_sg_from_iovec(skb, iv, vnet_hdr_len, count);
 714        else
 715                err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len,
 716                                                   len);
 717        if (err)
 718                goto err_kfree;
 719
 720        skb_set_network_header(skb, ETH_HLEN);
 721        skb_reset_mac_header(skb);
 722        skb->protocol = eth_hdr(skb)->h_proto;
 723
 724        if (vnet_hdr_len) {
 725                err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr);
 726                if (err)
 727                        goto err_kfree;
 728        }
 729
 730        skb_probe_transport_header(skb, ETH_HLEN);
 731
 732        rcu_read_lock_bh();
 733        vlan = rcu_dereference_bh(q->vlan);
 734        /* copy skb_ubuf_info for callback when skb has no error */
 735        if (zerocopy) {
 736                skb_shinfo(skb)->destructor_arg = m->msg_control;
 737                skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
 738                skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
 739        }
 740        if (vlan)
 741                macvlan_start_xmit(skb, vlan->dev);
 742        else
 743                kfree_skb(skb);
 744        rcu_read_unlock_bh();
 745
 746        return total_len;
 747
 748err_kfree:
 749        kfree_skb(skb);
 750
 751err:
 752        rcu_read_lock_bh();
 753        vlan = rcu_dereference_bh(q->vlan);
 754        if (vlan)
 755                vlan->dev->stats.tx_dropped++;
 756        rcu_read_unlock_bh();
 757
 758        return err;
 759}
 760
 761static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv,
 762                                 unsigned long count, loff_t pos)
 763{
 764        struct file *file = iocb->ki_filp;
 765        ssize_t result = -ENOLINK;
 766        struct macvtap_queue *q = file->private_data;
 767
 768        result = macvtap_get_user(q, NULL, iv, iov_length(iv, count), count,
 769                                  file->f_flags & O_NONBLOCK);
 770        return result;
 771}
 772
 773/* Put packet to the user space buffer */
 774static ssize_t macvtap_put_user(struct macvtap_queue *q,
 775                                const struct sk_buff *skb,
 776                                const struct iovec *iv, int len)
 777{
 778        struct macvlan_dev *vlan;
 779        int ret;
 780        int vnet_hdr_len = 0;
 781        int vlan_offset = 0;
 782        int copied;
 783
 784        if (q->flags & IFF_VNET_HDR) {
 785                struct virtio_net_hdr vnet_hdr;
 786                vnet_hdr_len = q->vnet_hdr_sz;
 787                if ((len -= vnet_hdr_len) < 0)
 788                        return -EINVAL;
 789
 790                ret = macvtap_skb_to_vnet_hdr(skb, &vnet_hdr);
 791                if (ret)
 792                        return ret;
 793
 794                if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, sizeof(vnet_hdr)))
 795                        return -EFAULT;
 796        }
 797        copied = vnet_hdr_len;
 798
 799        if (!vlan_tx_tag_present(skb))
 800                len = min_t(int, skb->len, len);
 801        else {
 802                int copy;
 803                struct {
 804                        __be16 h_vlan_proto;
 805                        __be16 h_vlan_TCI;
 806                } veth;
 807                veth.h_vlan_proto = htons(ETH_P_8021Q);
 808                veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb));
 809
 810                vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
 811                len = min_t(int, skb->len + VLAN_HLEN, len);
 812
 813                copy = min_t(int, vlan_offset, len);
 814                ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy);
 815                len -= copy;
 816                copied += copy;
 817                if (ret || !len)
 818                        goto done;
 819
 820                copy = min_t(int, sizeof(veth), len);
 821                ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy);
 822                len -= copy;
 823                copied += copy;
 824                if (ret || !len)
 825                        goto done;
 826        }
 827
 828        ret = skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len);
 829        copied += len;
 830
 831done:
 832        rcu_read_lock_bh();
 833        vlan = rcu_dereference_bh(q->vlan);
 834        if (vlan)
 835                macvlan_count_rx(vlan, copied - vnet_hdr_len, ret == 0, 0);
 836        rcu_read_unlock_bh();
 837
 838        return ret ? ret : copied;
 839}
 840
 841static ssize_t macvtap_do_read(struct macvtap_queue *q, struct kiocb *iocb,
 842                               const struct iovec *iv, unsigned long len,
 843                               int noblock)
 844{
 845        DEFINE_WAIT(wait);
 846        struct sk_buff *skb;
 847        ssize_t ret = 0;
 848
 849        while (len) {
 850                prepare_to_wait(sk_sleep(&q->sk), &wait, TASK_INTERRUPTIBLE);
 851
 852                /* Read frames from the queue */
 853                skb = skb_dequeue(&q->sk.sk_receive_queue);
 854                if (!skb) {
 855                        if (noblock) {
 856                                ret = -EAGAIN;
 857                                break;
 858                        }
 859                        if (signal_pending(current)) {
 860                                ret = -ERESTARTSYS;
 861                                break;
 862                        }
 863                        /* Nothing to read, let's sleep */
 864                        schedule();
 865                        continue;
 866                }
 867                ret = macvtap_put_user(q, skb, iv, len);
 868                kfree_skb(skb);
 869                break;
 870        }
 871
 872        finish_wait(sk_sleep(&q->sk), &wait);
 873        return ret;
 874}
 875
 876static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv,
 877                                unsigned long count, loff_t pos)
 878{
 879        struct file *file = iocb->ki_filp;
 880        struct macvtap_queue *q = file->private_data;
 881        ssize_t len, ret = 0;
 882
 883        len = iov_length(iv, count);
 884        if (len < 0) {
 885                ret = -EINVAL;
 886                goto out;
 887        }
 888
 889        ret = macvtap_do_read(q, iocb, iv, len, file->f_flags & O_NONBLOCK);
 890        ret = min_t(ssize_t, ret, len); /* XXX copied from tun.c. Why? */
 891out:
 892        return ret;
 893}
 894
 895/*
 896 * provide compatibility with generic tun/tap interface
 897 */
 898static long macvtap_ioctl(struct file *file, unsigned int cmd,
 899                          unsigned long arg)
 900{
 901        struct macvtap_queue *q = file->private_data;
 902        struct macvlan_dev *vlan;
 903        void __user *argp = (void __user *)arg;
 904        struct ifreq __user *ifr = argp;
 905        unsigned int __user *up = argp;
 906        unsigned int u;
 907        int __user *sp = argp;
 908        int s;
 909        int ret;
 910
 911        switch (cmd) {
 912        case TUNSETIFF:
 913                /* ignore the name, just look at flags */
 914                if (get_user(u, &ifr->ifr_flags))
 915                        return -EFAULT;
 916
 917                ret = 0;
 918                if ((u & ~IFF_VNET_HDR) != (IFF_NO_PI | IFF_TAP))
 919                        ret = -EINVAL;
 920                else
 921                        q->flags = u;
 922
 923                return ret;
 924
 925        case TUNGETIFF:
 926                rcu_read_lock_bh();
 927                vlan = rcu_dereference_bh(q->vlan);
 928                if (vlan)
 929                        dev_hold(vlan->dev);
 930                rcu_read_unlock_bh();
 931
 932                if (!vlan)
 933                        return -ENOLINK;
 934
 935                ret = 0;
 936                if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) ||
 937                    put_user(q->flags, &ifr->ifr_flags))
 938                        ret = -EFAULT;
 939                dev_put(vlan->dev);
 940                return ret;
 941
 942        case TUNGETFEATURES:
 943                if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR, up))
 944                        return -EFAULT;
 945                return 0;
 946
 947        case TUNSETSNDBUF:
 948                if (get_user(u, up))
 949                        return -EFAULT;
 950
 951                q->sk.sk_sndbuf = u;
 952                return 0;
 953
 954        case TUNGETVNETHDRSZ:
 955                s = q->vnet_hdr_sz;
 956                if (put_user(s, sp))
 957                        return -EFAULT;
 958                return 0;
 959
 960        case TUNSETVNETHDRSZ:
 961                if (get_user(s, sp))
 962                        return -EFAULT;
 963                if (s < (int)sizeof(struct virtio_net_hdr))
 964                        return -EINVAL;
 965
 966                q->vnet_hdr_sz = s;
 967                return 0;
 968
 969        case TUNSETOFFLOAD:
 970                /* let the user check for future flags */
 971                if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
 972                            TUN_F_TSO_ECN | TUN_F_UFO))
 973                        return -EINVAL;
 974
 975                /* TODO: only accept frames with the features that
 976                         got enabled for forwarded frames */
 977                if (!(q->flags & IFF_VNET_HDR))
 978                        return  -EINVAL;
 979                return 0;
 980
 981        default:
 982                return -EINVAL;
 983        }
 984}
 985
 986#ifdef CONFIG_COMPAT
 987static long macvtap_compat_ioctl(struct file *file, unsigned int cmd,
 988                                 unsigned long arg)
 989{
 990        return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
 991}
 992#endif
 993
 994static const struct file_operations macvtap_fops = {
 995        .owner          = THIS_MODULE,
 996        .open           = macvtap_open,
 997        .release        = macvtap_release,
 998        .aio_read       = macvtap_aio_read,
 999        .aio_write      = macvtap_aio_write,
1000        .poll           = macvtap_poll,
1001        .llseek         = no_llseek,
1002        .unlocked_ioctl = macvtap_ioctl,
1003#ifdef CONFIG_COMPAT
1004        .compat_ioctl   = macvtap_compat_ioctl,
1005#endif
1006};
1007
1008static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock,
1009                           struct msghdr *m, size_t total_len)
1010{
1011        struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1012        return macvtap_get_user(q, m, m->msg_iov, total_len, m->msg_iovlen,
1013                            m->msg_flags & MSG_DONTWAIT);
1014}
1015
1016static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock,
1017                           struct msghdr *m, size_t total_len,
1018                           int flags)
1019{
1020        struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1021        int ret;
1022        if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1023                return -EINVAL;
1024        ret = macvtap_do_read(q, iocb, m->msg_iov, total_len,
1025                          flags & MSG_DONTWAIT);
1026        if (ret > total_len) {
1027                m->msg_flags |= MSG_TRUNC;
1028                ret = flags & MSG_TRUNC ? ret : total_len;
1029        }
1030        return ret;
1031}
1032
1033/* Ops structure to mimic raw sockets with tun */
1034static const struct proto_ops macvtap_socket_ops = {
1035        .sendmsg = macvtap_sendmsg,
1036        .recvmsg = macvtap_recvmsg,
1037};
1038
1039/* Get an underlying socket object from tun file.  Returns error unless file is
1040 * attached to a device.  The returned object works like a packet socket, it
1041 * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
1042 * holding a reference to the file for as long as the socket is in use. */
1043struct socket *macvtap_get_socket(struct file *file)
1044{
1045        struct macvtap_queue *q;
1046        if (file->f_op != &macvtap_fops)
1047                return ERR_PTR(-EINVAL);
1048        q = file->private_data;
1049        if (!q)
1050                return ERR_PTR(-EBADFD);
1051        return &q->sock;
1052}
1053EXPORT_SYMBOL_GPL(macvtap_get_socket);
1054
1055static int macvtap_device_event(struct notifier_block *unused,
1056                                unsigned long event, void *ptr)
1057{
1058        struct net_device *dev = ptr;
1059        struct macvlan_dev *vlan;
1060        struct device *classdev;
1061        dev_t devt;
1062        int err;
1063
1064        if (dev->rtnl_link_ops != &macvtap_link_ops)
1065                return NOTIFY_DONE;
1066
1067        vlan = netdev_priv(dev);
1068
1069        switch (event) {
1070        case NETDEV_REGISTER:
1071                /* Create the device node here after the network device has
1072                 * been registered but before register_netdevice has
1073                 * finished running.
1074                 */
1075                err = macvtap_get_minor(vlan);
1076                if (err)
1077                        return notifier_from_errno(err);
1078
1079                devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
1080                classdev = device_create(macvtap_class, &dev->dev, devt,
1081                                         dev, "tap%d", dev->ifindex);
1082                if (IS_ERR(classdev)) {
1083                        macvtap_free_minor(vlan);
1084                        return notifier_from_errno(PTR_ERR(classdev));
1085                }
1086                break;
1087        case NETDEV_UNREGISTER:
1088                devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
1089                device_destroy(macvtap_class, devt);
1090                macvtap_free_minor(vlan);
1091                break;
1092        }
1093
1094        return NOTIFY_DONE;
1095}
1096
1097static struct notifier_block macvtap_notifier_block __read_mostly = {
1098        .notifier_call  = macvtap_device_event,
1099};
1100
1101static int macvtap_init(void)
1102{
1103        int err;
1104
1105        err = alloc_chrdev_region(&macvtap_major, 0,
1106                                MACVTAP_NUM_DEVS, "macvtap");
1107        if (err)
1108                goto out1;
1109
1110        cdev_init(&macvtap_cdev, &macvtap_fops);
1111        err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS);
1112        if (err)
1113                goto out2;
1114
1115        macvtap_class = class_create(THIS_MODULE, "macvtap");
1116        if (IS_ERR(macvtap_class)) {
1117                err = PTR_ERR(macvtap_class);
1118                goto out3;
1119        }
1120
1121        err = register_netdevice_notifier(&macvtap_notifier_block);
1122        if (err)
1123                goto out4;
1124
1125        err = macvlan_link_register(&macvtap_link_ops);
1126        if (err)
1127                goto out5;
1128
1129        return 0;
1130
1131out5:
1132        unregister_netdevice_notifier(&macvtap_notifier_block);
1133out4:
1134        class_unregister(macvtap_class);
1135out3:
1136        cdev_del(&macvtap_cdev);
1137out2:
1138        unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1139out1:
1140        return err;
1141}
1142module_init(macvtap_init);
1143
1144static void macvtap_exit(void)
1145{
1146        rtnl_link_unregister(&macvtap_link_ops);
1147        unregister_netdevice_notifier(&macvtap_notifier_block);
1148        class_unregister(macvtap_class);
1149        cdev_del(&macvtap_cdev);
1150        unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1151}
1152module_exit(macvtap_exit);
1153
1154MODULE_ALIAS_RTNL_LINK("macvtap");
1155MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1156MODULE_LICENSE("GPL");
1157
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.