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        int id;
 283
 284        mutex_lock(&minor_lock);
 285        if (idr_pre_get(&minor_idr, GFP_KERNEL) == 0)
 286                goto exit;
 287
 288        retval = idr_get_new_above(&minor_idr, vlan, 1, &id);
 289        if (retval < 0) {
 290                if (retval == -EAGAIN)
 291                        retval = -ENOMEM;
 292                goto exit;
 293        }
 294        if (id < MACVTAP_NUM_DEVS) {
 295                vlan->minor = id;
 296        } else {
 297                printk(KERN_ERR "too many macvtap devices\n");
 298                retval = -EINVAL;
 299                idr_remove(&minor_idr, id);
 300        }
 301exit:
 302        mutex_unlock(&minor_lock);
 303        return retval;
 304}
 305
 306static void macvtap_free_minor(struct macvlan_dev *vlan)
 307{
 308        mutex_lock(&minor_lock);
 309        if (vlan->minor) {
 310                idr_remove(&minor_idr, vlan->minor);
 311                vlan->minor = 0;
 312        }
 313        mutex_unlock(&minor_lock);
 314}
 315
 316static struct net_device *dev_get_by_macvtap_minor(int minor)
 317{
 318        struct net_device *dev = NULL;
 319        struct macvlan_dev *vlan;
 320
 321        mutex_lock(&minor_lock);
 322        vlan = idr_find(&minor_idr, minor);
 323        if (vlan) {
 324                dev = vlan->dev;
 325                dev_hold(dev);
 326        }
 327        mutex_unlock(&minor_lock);
 328        return dev;
 329}
 330
 331static int macvtap_newlink(struct net *src_net,
 332                           struct net_device *dev,
 333                           struct nlattr *tb[],
 334                           struct nlattr *data[])
 335{
 336        /* Don't put anything that may fail after macvlan_common_newlink
 337         * because we can't undo what it does.
 338         */
 339        return macvlan_common_newlink(src_net, dev, tb, data,
 340                                      macvtap_receive, macvtap_forward);
 341}
 342
 343static void macvtap_dellink(struct net_device *dev,
 344                            struct list_head *head)
 345{
 346        macvtap_del_queues(dev);
 347        macvlan_dellink(dev, head);
 348}
 349
 350static void macvtap_setup(struct net_device *dev)
 351{
 352        macvlan_common_setup(dev);
 353        dev->tx_queue_len = TUN_READQ_SIZE;
 354}
 355
 356static struct rtnl_link_ops macvtap_link_ops __read_mostly = {
 357        .kind           = "macvtap",
 358        .setup          = macvtap_setup,
 359        .newlink        = macvtap_newlink,
 360        .dellink        = macvtap_dellink,
 361};
 362
 363
 364static void macvtap_sock_write_space(struct sock *sk)
 365{
 366        wait_queue_head_t *wqueue;
 367
 368        if (!sock_writeable(sk) ||
 369            !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
 370                return;
 371
 372        wqueue = sk_sleep(sk);
 373        if (wqueue && waitqueue_active(wqueue))
 374                wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND);
 375}
 376
 377static void macvtap_sock_destruct(struct sock *sk)
 378{
 379        skb_queue_purge(&sk->sk_receive_queue);
 380}
 381
 382static int macvtap_open(struct inode *inode, struct file *file)
 383{
 384        struct net *net = current->nsproxy->net_ns;
 385        struct net_device *dev = dev_get_by_macvtap_minor(iminor(inode));
 386        struct macvtap_queue *q;
 387        int err;
 388
 389        err = -ENODEV;
 390        if (!dev)
 391                goto out;
 392
 393        err = -ENOMEM;
 394        q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
 395                                             &macvtap_proto);
 396        if (!q)
 397                goto out;
 398
 399        q->sock.wq = &q->wq;
 400        init_waitqueue_head(&q->wq.wait);
 401        q->sock.type = SOCK_RAW;
 402        q->sock.state = SS_CONNECTED;
 403        q->sock.file = file;
 404        q->sock.ops = &macvtap_socket_ops;
 405        sock_init_data(&q->sock, &q->sk);
 406        q->sk.sk_write_space = macvtap_sock_write_space;
 407        q->sk.sk_destruct = macvtap_sock_destruct;
 408        q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
 409        q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
 410
 411        /*
 412         * so far only KVM virtio_net uses macvtap, enable zero copy between
 413         * guest kernel and host kernel when lower device supports zerocopy
 414         *
 415         * The macvlan supports zerocopy iff the lower device supports zero
 416         * copy so we don't have to look at the lower device directly.
 417         */
 418        if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG))
 419                sock_set_flag(&q->sk, SOCK_ZEROCOPY);
 420
 421        err = macvtap_set_queue(dev, file, q);
 422        if (err)
 423                sock_put(&q->sk);
 424
 425out:
 426        if (dev)
 427                dev_put(dev);
 428
 429        return err;
 430}
 431
 432static int macvtap_release(struct inode *inode, struct file *file)
 433{
 434        struct macvtap_queue *q = file->private_data;
 435        macvtap_put_queue(q);
 436        return 0;
 437}
 438
 439static unsigned int macvtap_poll(struct file *file, poll_table * wait)
 440{
 441        struct macvtap_queue *q = file->private_data;
 442        unsigned int mask = POLLERR;
 443
 444        if (!q)
 445                goto out;
 446
 447        mask = 0;
 448        poll_wait(file, &q->wq.wait, wait);
 449
 450        if (!skb_queue_empty(&q->sk.sk_receive_queue))
 451                mask |= POLLIN | POLLRDNORM;
 452
 453        if (sock_writeable(&q->sk) ||
 454            (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) &&
 455             sock_writeable(&q->sk)))
 456                mask |= POLLOUT | POLLWRNORM;
 457
 458out:
 459        return mask;
 460}
 461
 462static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad,
 463                                                size_t len, size_t linear,
 464                                                int noblock, int *err)
 465{
 466        struct sk_buff *skb;
 467
 468        /* Under a page?  Don't bother with paged skb. */
 469        if (prepad + len < PAGE_SIZE || !linear)
 470                linear = len;
 471
 472        skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
 473                                   err);
 474        if (!skb)
 475                return NULL;
 476
 477        skb_reserve(skb, prepad);
 478        skb_put(skb, linear);
 479        skb->data_len = len - linear;
 480        skb->len += len - linear;
 481
 482        return skb;
 483}
 484
 485/* set skb frags from iovec, this can move to core network code for reuse */
 486static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
 487                                  int offset, size_t count)
 488{
 489        int len = iov_length(from, count) - offset;
 490        int copy = skb_headlen(skb);
 491        int size, offset1 = 0;
 492        int i = 0;
 493
 494        /* Skip over from offset */
 495        while (count && (offset >= from->iov_len)) {
 496                offset -= from->iov_len;
 497                ++from;
 498                --count;
 499        }
 500
 501        /* copy up to skb headlen */
 502        while (count && (copy > 0)) {
 503                size = min_t(unsigned int, copy, from->iov_len - offset);
 504                if (copy_from_user(skb->data + offset1, from->iov_base + offset,
 505                                   size))
 506                        return -EFAULT;
 507                if (copy > size) {
 508                        ++from;
 509                        --count;
 510                        offset = 0;
 511                } else
 512                        offset += size;
 513                copy -= size;
 514                offset1 += size;
 515        }
 516
 517        if (len == offset1)
 518                return 0;
 519
 520        while (count--) {
 521                struct page *page[MAX_SKB_FRAGS];
 522                int num_pages;
 523                unsigned long base;
 524                unsigned long truesize;
 525
 526                len = from->iov_len - offset;
 527                if (!len) {
 528                        offset = 0;
 529                        ++from;
 530                        continue;
 531                }
 532                base = (unsigned long)from->iov_base + offset;
 533                size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
 534                if (i + size > MAX_SKB_FRAGS)
 535                        return -EMSGSIZE;
 536                num_pages = get_user_pages_fast(base, size, 0, &page[i]);
 537                if (num_pages != size) {
 538                        for (i = 0; i < num_pages; i++)
 539                                put_page(page[i]);
 540                        return -EFAULT;
 541                }
 542                truesize = size * PAGE_SIZE;
 543                skb->data_len += len;
 544                skb->len += len;
 545                skb->truesize += truesize;
 546                atomic_add(truesize, &skb->sk->sk_wmem_alloc);
 547                while (len) {
 548                        int off = base & ~PAGE_MASK;
 549                        int size = min_t(int, len, PAGE_SIZE - off);
 550                        __skb_fill_page_desc(skb, i, page[i], off, size);
 551                        skb_shinfo(skb)->nr_frags++;
 552                        /* increase sk_wmem_alloc */
 553                        base += size;
 554                        len -= size;
 555                        i++;
 556                }
 557                offset = 0;
 558                ++from;
 559        }
 560        return 0;
 561}
 562
 563/*
 564 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
 565 * be shared with the tun/tap driver.
 566 */
 567static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb,
 568                                     struct virtio_net_hdr *vnet_hdr)
 569{
 570        unsigned short gso_type = 0;
 571        if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
 572                switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
 573                case VIRTIO_NET_HDR_GSO_TCPV4:
 574                        gso_type = SKB_GSO_TCPV4;
 575                        break;
 576                case VIRTIO_NET_HDR_GSO_TCPV6:
 577                        gso_type = SKB_GSO_TCPV6;
 578                        break;
 579                case VIRTIO_NET_HDR_GSO_UDP:
 580                        gso_type = SKB_GSO_UDP;
 581                        break;
 582                default:
 583                        return -EINVAL;
 584                }
 585
 586                if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
 587                        gso_type |= SKB_GSO_TCP_ECN;
 588
 589                if (vnet_hdr->gso_size == 0)
 590                        return -EINVAL;
 591        }
 592
 593        if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
 594                if (!skb_partial_csum_set(skb, vnet_hdr->csum_start,
 595                                          vnet_hdr->csum_offset))
 596                        return -EINVAL;
 597        }
 598
 599        if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
 600                skb_shinfo(skb)->gso_size = vnet_hdr->gso_size;
 601                skb_shinfo(skb)->gso_type = gso_type;
 602
 603                /* Header must be checked, and gso_segs computed. */
 604                skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
 605                skb_shinfo(skb)->gso_segs = 0;
 606        }
 607        return 0;
 608}
 609
 610static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb,
 611                                   struct virtio_net_hdr *vnet_hdr)
 612{
 613        memset(vnet_hdr, 0, sizeof(*vnet_hdr));
 614
 615        if (skb_is_gso(skb)) {
 616                struct skb_shared_info *sinfo = skb_shinfo(skb);
 617
 618                /* This is a hint as to how much should be linear. */
 619                vnet_hdr->hdr_len = skb_headlen(skb);
 620                vnet_hdr->gso_size = sinfo->gso_size;
 621                if (sinfo->gso_type & SKB_GSO_TCPV4)
 622                        vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
 623                else if (sinfo->gso_type & SKB_GSO_TCPV6)
 624                        vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
 625                else if (sinfo->gso_type & SKB_GSO_UDP)
 626                        vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
 627                else
 628                        BUG();
 629                if (sinfo->gso_type & SKB_GSO_TCP_ECN)
 630                        vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
 631        } else
 632                vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
 633
 634        if (skb->ip_summed == CHECKSUM_PARTIAL) {
 635                vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 636                vnet_hdr->csum_start = skb_checksum_start_offset(skb);
 637                vnet_hdr->csum_offset = skb->csum_offset;
 638        } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
 639                vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID;
 640        } /* else everything is zero */
 641
 642        return 0;
 643}
 644
 645
 646/* Get packet from user space buffer */
 647static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m,
 648                                const struct iovec *iv, unsigned long total_len,
 649                                size_t count, int noblock)
 650{
 651        struct sk_buff *skb;
 652        struct macvlan_dev *vlan;
 653        unsigned long len = total_len;
 654        int err;
 655        struct virtio_net_hdr vnet_hdr = { 0 };
 656        int vnet_hdr_len = 0;
 657        int copylen = 0;
 658        bool zerocopy = false;
 659
 660        if (q->flags & IFF_VNET_HDR) {
 661                vnet_hdr_len = q->vnet_hdr_sz;
 662
 663                err = -EINVAL;
 664                if (len < vnet_hdr_len)
 665                        goto err;
 666                len -= vnet_hdr_len;
 667
 668                err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0,
 669                                           sizeof(vnet_hdr));
 670                if (err < 0)
 671                        goto err;
 672                if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
 673                     vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
 674                                                        vnet_hdr.hdr_len)
 675                        vnet_hdr.hdr_len = vnet_hdr.csum_start +
 676                                                vnet_hdr.csum_offset + 2;
 677                err = -EINVAL;
 678                if (vnet_hdr.hdr_len > len)
 679                        goto err;
 680        }
 681
 682        err = -EINVAL;
 683        if (unlikely(len < ETH_HLEN))
 684                goto err;
 685
 686        err = -EMSGSIZE;
 687        if (unlikely(count > UIO_MAXIOV))
 688                goto err;
 689
 690        if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY))
 691                zerocopy = true;
 692
 693        if (zerocopy) {
 694                /* Userspace may produce vectors with count greater than
 695                 * MAX_SKB_FRAGS, so we need to linearize parts of the skb
 696                 * to let the rest of data to be fit in the frags.
 697                 */
 698                if (count > MAX_SKB_FRAGS) {
 699                        copylen = iov_length(iv, count - MAX_SKB_FRAGS);
 700                        if (copylen < vnet_hdr_len)
 701                                copylen = 0;
 702                        else
 703                                copylen -= vnet_hdr_len;
 704                }
 705                /* There are 256 bytes to be copied in skb, so there is enough
 706                 * room for skb expand head in case it is used.
 707                 * The rest buffer is mapped from userspace.
 708                 */
 709                if (copylen < vnet_hdr.hdr_len)
 710                        copylen = vnet_hdr.hdr_len;
 711                if (!copylen)
 712                        copylen = GOODCOPY_LEN;
 713        } else
 714                copylen = len;
 715
 716        skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, copylen,
 717                                vnet_hdr.hdr_len, noblock, &err);
 718        if (!skb)
 719                goto err;
 720
 721        if (zerocopy)
 722                err = zerocopy_sg_from_iovec(skb, iv, vnet_hdr_len, count);
 723        else
 724                err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len,
 725                                                   len);
 726        if (err)
 727                goto err_kfree;
 728
 729        skb_set_network_header(skb, ETH_HLEN);
 730        skb_reset_mac_header(skb);
 731        skb->protocol = eth_hdr(skb)->h_proto;
 732
 733        if (vnet_hdr_len) {
 734                err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr);
 735                if (err)
 736                        goto err_kfree;
 737        }
 738
 739        rcu_read_lock_bh();
 740        vlan = rcu_dereference_bh(q->vlan);
 741        /* copy skb_ubuf_info for callback when skb has no error */
 742        if (zerocopy) {
 743                skb_shinfo(skb)->destructor_arg = m->msg_control;
 744                skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
 745        }
 746        if (vlan)
 747                macvlan_start_xmit(skb, vlan->dev);
 748        else
 749                kfree_skb(skb);
 750        rcu_read_unlock_bh();
 751
 752        return total_len;
 753
 754err_kfree:
 755        kfree_skb(skb);
 756
 757err:
 758        rcu_read_lock_bh();
 759        vlan = rcu_dereference_bh(q->vlan);
 760        if (vlan)
 761                vlan->dev->stats.tx_dropped++;
 762        rcu_read_unlock_bh();
 763
 764        return err;
 765}
 766
 767static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv,
 768                                 unsigned long count, loff_t pos)
 769{
 770        struct file *file = iocb->ki_filp;
 771        ssize_t result = -ENOLINK;
 772        struct macvtap_queue *q = file->private_data;
 773
 774        result = macvtap_get_user(q, NULL, iv, iov_length(iv, count), count,
 775                                  file->f_flags & O_NONBLOCK);
 776        return result;
 777}
 778
 779/* Put packet to the user space buffer */
 780static ssize_t macvtap_put_user(struct macvtap_queue *q,
 781                                const struct sk_buff *skb,
 782                                const struct iovec *iv, int len)
 783{
 784        struct macvlan_dev *vlan;
 785        int ret;
 786        int vnet_hdr_len = 0;
 787        int vlan_offset = 0;
 788        int copied;
 789
 790        if (q->flags & IFF_VNET_HDR) {
 791                struct virtio_net_hdr vnet_hdr;
 792                vnet_hdr_len = q->vnet_hdr_sz;
 793                if ((len -= vnet_hdr_len) < 0)
 794                        return -EINVAL;
 795
 796                ret = macvtap_skb_to_vnet_hdr(skb, &vnet_hdr);
 797                if (ret)
 798                        return ret;
 799
 800                if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, sizeof(vnet_hdr)))
 801                        return -EFAULT;
 802        }
 803        copied = vnet_hdr_len;
 804
 805        if (!vlan_tx_tag_present(skb))
 806                len = min_t(int, skb->len, len);
 807        else {
 808                int copy;
 809                struct {
 810                        __be16 h_vlan_proto;
 811                        __be16 h_vlan_TCI;
 812                } veth;
 813                veth.h_vlan_proto = htons(ETH_P_8021Q);
 814                veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb));
 815
 816                vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
 817                len = min_t(int, skb->len + VLAN_HLEN, len);
 818
 819                copy = min_t(int, vlan_offset, len);
 820                ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy);
 821                len -= copy;
 822                copied += copy;
 823                if (ret || !len)
 824                        goto done;
 825
 826                copy = min_t(int, sizeof(veth), len);
 827                ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy);
 828                len -= copy;
 829                copied += copy;
 830                if (ret || !len)
 831                        goto done;
 832        }
 833
 834        ret = skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len);
 835        copied += len;
 836
 837done:
 838        rcu_read_lock_bh();
 839        vlan = rcu_dereference_bh(q->vlan);
 840        if (vlan)
 841                macvlan_count_rx(vlan, copied - vnet_hdr_len, ret == 0, 0);
 842        rcu_read_unlock_bh();
 843
 844        return ret ? ret : copied;
 845}
 846
 847static ssize_t macvtap_do_read(struct macvtap_queue *q, struct kiocb *iocb,
 848                               const struct iovec *iv, unsigned long len,
 849                               int noblock)
 850{
 851        DEFINE_WAIT(wait);
 852        struct sk_buff *skb;
 853        ssize_t ret = 0;
 854
 855        while (len) {
 856                prepare_to_wait(sk_sleep(&q->sk), &wait, TASK_INTERRUPTIBLE);
 857
 858                /* Read frames from the queue */
 859                skb = skb_dequeue(&q->sk.sk_receive_queue);
 860                if (!skb) {
 861                        if (noblock) {
 862                                ret = -EAGAIN;
 863                                break;
 864                        }
 865                        if (signal_pending(current)) {
 866                                ret = -ERESTARTSYS;
 867                                break;
 868                        }
 869                        /* Nothing to read, let's sleep */
 870                        schedule();
 871                        continue;
 872                }
 873                ret = macvtap_put_user(q, skb, iv, len);
 874                kfree_skb(skb);
 875                break;
 876        }
 877
 878        finish_wait(sk_sleep(&q->sk), &wait);
 879        return ret;
 880}
 881
 882static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv,
 883                                unsigned long count, loff_t pos)
 884{
 885        struct file *file = iocb->ki_filp;
 886        struct macvtap_queue *q = file->private_data;
 887        ssize_t len, ret = 0;
 888
 889        len = iov_length(iv, count);
 890        if (len < 0) {
 891                ret = -EINVAL;
 892                goto out;
 893        }
 894
 895        ret = macvtap_do_read(q, iocb, iv, len, file->f_flags & O_NONBLOCK);
 896        ret = min_t(ssize_t, ret, len); /* XXX copied from tun.c. Why? */
 897out:
 898        return ret;
 899}
 900
 901/*
 902 * provide compatibility with generic tun/tap interface
 903 */
 904static long macvtap_ioctl(struct file *file, unsigned int cmd,
 905                          unsigned long arg)
 906{
 907        struct macvtap_queue *q = file->private_data;
 908        struct macvlan_dev *vlan;
 909        void __user *argp = (void __user *)arg;
 910        struct ifreq __user *ifr = argp;
 911        unsigned int __user *up = argp;
 912        unsigned int u;
 913        int __user *sp = argp;
 914        int s;
 915        int ret;
 916
 917        switch (cmd) {
 918        case TUNSETIFF:
 919                /* ignore the name, just look at flags */
 920                if (get_user(u, &ifr->ifr_flags))
 921                        return -EFAULT;
 922
 923                ret = 0;
 924                if ((u & ~IFF_VNET_HDR) != (IFF_NO_PI | IFF_TAP))
 925                        ret = -EINVAL;
 926                else
 927                        q->flags = u;
 928
 929                return ret;
 930
 931        case TUNGETIFF:
 932                rcu_read_lock_bh();
 933                vlan = rcu_dereference_bh(q->vlan);
 934                if (vlan)
 935                        dev_hold(vlan->dev);
 936                rcu_read_unlock_bh();
 937
 938                if (!vlan)
 939                        return -ENOLINK;
 940
 941                ret = 0;
 942                if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) ||
 943                    put_user(q->flags, &ifr->ifr_flags))
 944                        ret = -EFAULT;
 945                dev_put(vlan->dev);
 946                return ret;
 947
 948        case TUNGETFEATURES:
 949                if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR, up))
 950                        return -EFAULT;
 951                return 0;
 952
 953        case TUNSETSNDBUF:
 954                if (get_user(u, up))
 955                        return -EFAULT;
 956
 957                q->sk.sk_sndbuf = u;
 958                return 0;
 959
 960        case TUNGETVNETHDRSZ:
 961                s = q->vnet_hdr_sz;
 962                if (put_user(s, sp))
 963                        return -EFAULT;
 964                return 0;
 965
 966        case TUNSETVNETHDRSZ:
 967                if (get_user(s, sp))
 968                        return -EFAULT;
 969                if (s < (int)sizeof(struct virtio_net_hdr))
 970                        return -EINVAL;
 971
 972                q->vnet_hdr_sz = s;
 973                return 0;
 974
 975        case TUNSETOFFLOAD:
 976                /* let the user check for future flags */
 977                if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
 978                            TUN_F_TSO_ECN | TUN_F_UFO))
 979                        return -EINVAL;
 980
 981                /* TODO: only accept frames with the features that
 982                         got enabled for forwarded frames */
 983                if (!(q->flags & IFF_VNET_HDR))
 984                        return  -EINVAL;
 985                return 0;
 986
 987        default:
 988                return -EINVAL;
 989        }
 990}
 991
 992#ifdef CONFIG_COMPAT
 993static long macvtap_compat_ioctl(struct file *file, unsigned int cmd,
 994                                 unsigned long arg)
 995{
 996        return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
 997}
 998#endif
 999
1000static const struct file_operations macvtap_fops = {
1001        .owner          = THIS_MODULE,
1002        .open           = macvtap_open,
1003        .release        = macvtap_release,
1004        .aio_read       = macvtap_aio_read,
1005        .aio_write      = macvtap_aio_write,
1006        .poll           = macvtap_poll,
1007        .llseek         = no_llseek,
1008        .unlocked_ioctl = macvtap_ioctl,
1009#ifdef CONFIG_COMPAT
1010        .compat_ioctl   = macvtap_compat_ioctl,
1011#endif
1012};
1013
1014static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock,
1015                           struct msghdr *m, size_t total_len)
1016{
1017        struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1018        return macvtap_get_user(q, m, m->msg_iov, total_len, m->msg_iovlen,
1019                            m->msg_flags & MSG_DONTWAIT);
1020}
1021
1022static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock,
1023                           struct msghdr *m, size_t total_len,
1024                           int flags)
1025{
1026        struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1027        int ret;
1028        if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1029                return -EINVAL;
1030        ret = macvtap_do_read(q, iocb, m->msg_iov, total_len,
1031                          flags & MSG_DONTWAIT);
1032        if (ret > total_len) {
1033                m->msg_flags |= MSG_TRUNC;
1034                ret = flags & MSG_TRUNC ? ret : total_len;
1035        }
1036        return ret;
1037}
1038
1039/* Ops structure to mimic raw sockets with tun */
1040static const struct proto_ops macvtap_socket_ops = {
1041        .sendmsg = macvtap_sendmsg,
1042        .recvmsg = macvtap_recvmsg,
1043};
1044
1045/* Get an underlying socket object from tun file.  Returns error unless file is
1046 * attached to a device.  The returned object works like a packet socket, it
1047 * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
1048 * holding a reference to the file for as long as the socket is in use. */
1049struct socket *macvtap_get_socket(struct file *file)
1050{
1051        struct macvtap_queue *q;
1052        if (file->f_op != &macvtap_fops)
1053                return ERR_PTR(-EINVAL);
1054        q = file->private_data;
1055        if (!q)
1056                return ERR_PTR(-EBADFD);
1057        return &q->sock;
1058}
1059EXPORT_SYMBOL_GPL(macvtap_get_socket);
1060
1061static int macvtap_device_event(struct notifier_block *unused,
1062                                unsigned long event, void *ptr)
1063{
1064        struct net_device *dev = ptr;
1065        struct macvlan_dev *vlan;
1066        struct device *classdev;
1067        dev_t devt;
1068        int err;
1069
1070        if (dev->rtnl_link_ops != &macvtap_link_ops)
1071                return NOTIFY_DONE;
1072
1073        vlan = netdev_priv(dev);
1074
1075        switch (event) {
1076        case NETDEV_REGISTER:
1077                /* Create the device node here after the network device has
1078                 * been registered but before register_netdevice has
1079                 * finished running.
1080                 */
1081                err = macvtap_get_minor(vlan);
1082                if (err)
1083                        return notifier_from_errno(err);
1084
1085                devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
1086                classdev = device_create(macvtap_class, &dev->dev, devt,
1087                                         dev, "tap%d", dev->ifindex);
1088                if (IS_ERR(classdev)) {
1089                        macvtap_free_minor(vlan);
1090                        return notifier_from_errno(PTR_ERR(classdev));
1091                }
1092                break;
1093        case NETDEV_UNREGISTER:
1094                devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
1095                device_destroy(macvtap_class, devt);
1096                macvtap_free_minor(vlan);
1097                break;
1098        }
1099
1100        return NOTIFY_DONE;
1101}
1102
1103static struct notifier_block macvtap_notifier_block __read_mostly = {
1104        .notifier_call  = macvtap_device_event,
1105};
1106
1107static int macvtap_init(void)
1108{
1109        int err;
1110
1111        err = alloc_chrdev_region(&macvtap_major, 0,
1112                                MACVTAP_NUM_DEVS, "macvtap");
1113        if (err)
1114                goto out1;
1115
1116        cdev_init(&macvtap_cdev, &macvtap_fops);
1117        err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS);
1118        if (err)
1119                goto out2;
1120
1121        macvtap_class = class_create(THIS_MODULE, "macvtap");
1122        if (IS_ERR(macvtap_class)) {
1123                err = PTR_ERR(macvtap_class);
1124                goto out3;
1125        }
1126
1127        err = register_netdevice_notifier(&macvtap_notifier_block);
1128        if (err)
1129                goto out4;
1130
1131        err = macvlan_link_register(&macvtap_link_ops);
1132        if (err)
1133                goto out5;
1134
1135        return 0;
1136
1137out5:
1138        unregister_netdevice_notifier(&macvtap_notifier_block);
1139out4:
1140        class_unregister(macvtap_class);
1141out3:
1142        cdev_del(&macvtap_cdev);
1143out2:
1144        unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1145out1:
1146        return err;
1147}
1148module_init(macvtap_init);
1149
1150static void macvtap_exit(void)
1151{
1152        rtnl_link_unregister(&macvtap_link_ops);
1153        unregister_netdevice_notifier(&macvtap_notifier_block);
1154        class_unregister(macvtap_class);
1155        cdev_del(&macvtap_cdev);
1156        unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1157}
1158module_exit(macvtap_exit);
1159
1160MODULE_ALIAS_RTNL_LINK("macvtap");
1161MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1162MODULE_LICENSE("GPL");
1163
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.