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 */
  35struct macvtap_queue {
  36        struct sock sk;
  37        struct socket sock;
  38        struct socket_wq wq;
  39        int vnet_hdr_sz;
  40        struct macvlan_dev __rcu *vlan;
  41        struct file *file;
  42        unsigned int flags;
  43        u16 queue_index;
  44        bool enabled;
  45        struct list_head next;
  46};
  47
  48static struct proto macvtap_proto = {
  49        .name = "macvtap",
  50        .owner = THIS_MODULE,
  51        .obj_size = sizeof (struct macvtap_queue),
  52};
  53
  54/*
  55 * Variables for dealing with macvtaps device numbers.
  56 */
  57static dev_t macvtap_major;
  58#define MACVTAP_NUM_DEVS (1U << MINORBITS)
  59static DEFINE_MUTEX(minor_lock);
  60static DEFINE_IDR(minor_idr);
  61
  62#define GOODCOPY_LEN 128
  63static struct class *macvtap_class;
  64static struct cdev macvtap_cdev;
  65
  66static const struct proto_ops macvtap_socket_ops;
  67
  68#define TUN_OFFLOADS (NETIF_F_HW_CSUM | NETIF_F_TSO_ECN | NETIF_F_TSO | \
  69                      NETIF_F_TSO6 | NETIF_F_UFO)
  70#define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO)
  71#define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG)
  72
  73/*
  74 * RCU usage:
  75 * The macvtap_queue and the macvlan_dev are loosely coupled, the
  76 * pointers from one to the other can only be read while rcu_read_lock
  77 * or rtnl is held.
  78 *
  79 * Both the file and the macvlan_dev hold a reference on the macvtap_queue
  80 * through sock_hold(&q->sk). When the macvlan_dev goes away first,
  81 * q->vlan becomes inaccessible. When the files gets closed,
  82 * macvtap_get_queue() fails.
  83 *
  84 * There may still be references to the struct sock inside of the
  85 * queue from outbound SKBs, but these never reference back to the
  86 * file or the dev. The data structure is freed through __sk_free
  87 * when both our references and any pending SKBs are gone.
  88 */
  89
  90static int macvtap_enable_queue(struct net_device *dev, struct file *file,
  91                                struct macvtap_queue *q)
  92{
  93        struct macvlan_dev *vlan = netdev_priv(dev);
  94        int err = -EINVAL;
  95
  96        ASSERT_RTNL();
  97
  98        if (q->enabled)
  99                goto out;
 100
 101        err = 0;
 102        rcu_assign_pointer(vlan->taps[vlan->numvtaps], q);
 103        q->queue_index = vlan->numvtaps;
 104        q->enabled = true;
 105
 106        vlan->numvtaps++;
 107out:
 108        return err;
 109}
 110
 111static int macvtap_set_queue(struct net_device *dev, struct file *file,
 112                             struct macvtap_queue *q)
 113{
 114        struct macvlan_dev *vlan = netdev_priv(dev);
 115        int err = -EBUSY;
 116
 117        rtnl_lock();
 118        if (vlan->numqueues == MAX_MACVTAP_QUEUES)
 119                goto out;
 120
 121        err = 0;
 122        rcu_assign_pointer(q->vlan, vlan);
 123        rcu_assign_pointer(vlan->taps[vlan->numvtaps], q);
 124        sock_hold(&q->sk);
 125
 126        q->file = file;
 127        q->queue_index = vlan->numvtaps;
 128        q->enabled = true;
 129        file->private_data = q;
 130        list_add_tail(&q->next, &vlan->queue_list);
 131
 132        vlan->numvtaps++;
 133        vlan->numqueues++;
 134
 135out:
 136        rtnl_unlock();
 137        return err;
 138}
 139
 140static int macvtap_disable_queue(struct macvtap_queue *q)
 141{
 142        struct macvlan_dev *vlan;
 143        struct macvtap_queue *nq;
 144
 145        ASSERT_RTNL();
 146        if (!q->enabled)
 147                return -EINVAL;
 148
 149        vlan = rtnl_dereference(q->vlan);
 150
 151        if (vlan) {
 152                int index = q->queue_index;
 153                BUG_ON(index >= vlan->numvtaps);
 154                nq = rtnl_dereference(vlan->taps[vlan->numvtaps - 1]);
 155                nq->queue_index = index;
 156
 157                rcu_assign_pointer(vlan->taps[index], nq);
 158                RCU_INIT_POINTER(vlan->taps[vlan->numvtaps - 1], NULL);
 159                q->enabled = false;
 160
 161                vlan->numvtaps--;
 162        }
 163
 164        return 0;
 165}
 166
 167/*
 168 * The file owning the queue got closed, give up both
 169 * the reference that the files holds as well as the
 170 * one from the macvlan_dev if that still exists.
 171 *
 172 * Using the spinlock makes sure that we don't get
 173 * to the queue again after destroying it.
 174 */
 175static void macvtap_put_queue(struct macvtap_queue *q)
 176{
 177        struct macvlan_dev *vlan;
 178
 179        rtnl_lock();
 180        vlan = rtnl_dereference(q->vlan);
 181
 182        if (vlan) {
 183                if (q->enabled)
 184                        BUG_ON(macvtap_disable_queue(q));
 185
 186                vlan->numqueues--;
 187                RCU_INIT_POINTER(q->vlan, NULL);
 188                sock_put(&q->sk);
 189                list_del_init(&q->next);
 190        }
 191
 192        rtnl_unlock();
 193
 194        synchronize_rcu();
 195        sock_put(&q->sk);
 196}
 197
 198/*
 199 * Select a queue based on the rxq of the device on which this packet
 200 * arrived. If the incoming device is not mq, calculate a flow hash
 201 * to select a queue. If all fails, find the first available queue.
 202 * Cache vlan->numvtaps since it can become zero during the execution
 203 * of this function.
 204 */
 205static struct macvtap_queue *macvtap_get_queue(struct net_device *dev,
 206                                               struct sk_buff *skb)
 207{
 208        struct macvlan_dev *vlan = netdev_priv(dev);
 209        struct macvtap_queue *tap = NULL;
 210        /* Access to taps array is protected by rcu, but access to numvtaps
 211         * isn't. Below we use it to lookup a queue, but treat it as a hint
 212         * and validate that the result isn't NULL - in case we are
 213         * racing against queue removal.
 214         */
 215        int numvtaps = ACCESS_ONCE(vlan->numvtaps);
 216        __u32 rxq;
 217
 218        if (!numvtaps)
 219                goto out;
 220
 221        /* Check if we can use flow to select a queue */
 222        rxq = skb_get_rxhash(skb);
 223        if (rxq) {
 224                tap = rcu_dereference(vlan->taps[rxq % numvtaps]);
 225                goto out;
 226        }
 227
 228        if (likely(skb_rx_queue_recorded(skb))) {
 229                rxq = skb_get_rx_queue(skb);
 230
 231                while (unlikely(rxq >= numvtaps))
 232                        rxq -= numvtaps;
 233
 234                tap = rcu_dereference(vlan->taps[rxq]);
 235                goto out;
 236        }
 237
 238        tap = rcu_dereference(vlan->taps[0]);
 239out:
 240        return tap;
 241}
 242
 243/*
 244 * The net_device is going away, give up the reference
 245 * that it holds on all queues and safely set the pointer
 246 * from the queues to NULL.
 247 */
 248static void macvtap_del_queues(struct net_device *dev)
 249{
 250        struct macvlan_dev *vlan = netdev_priv(dev);
 251        struct macvtap_queue *q, *tmp, *qlist[MAX_MACVTAP_QUEUES];
 252        int i, j = 0;
 253
 254        ASSERT_RTNL();
 255        list_for_each_entry_safe(q, tmp, &vlan->queue_list, next) {
 256                list_del_init(&q->next);
 257                qlist[j++] = q;
 258                RCU_INIT_POINTER(q->vlan, NULL);
 259                if (q->enabled)
 260                        vlan->numvtaps--;
 261                vlan->numqueues--;
 262        }
 263        for (i = 0; i < vlan->numvtaps; i++)
 264                RCU_INIT_POINTER(vlan->taps[i], NULL);
 265        BUG_ON(vlan->numvtaps);
 266        BUG_ON(vlan->numqueues);
 267        /* guarantee that any future macvtap_set_queue will fail */
 268        vlan->numvtaps = MAX_MACVTAP_QUEUES;
 269
 270        for (--j; j >= 0; j--)
 271                sock_put(&qlist[j]->sk);
 272}
 273
 274/*
 275 * Forward happens for data that gets sent from one macvlan
 276 * endpoint to another one in bridge mode. We just take
 277 * the skb and put it into the receive queue.
 278 */
 279static int macvtap_forward(struct net_device *dev, struct sk_buff *skb)
 280{
 281        struct macvlan_dev *vlan = netdev_priv(dev);
 282        struct macvtap_queue *q = macvtap_get_queue(dev, skb);
 283        netdev_features_t features = TAP_FEATURES;
 284
 285        if (!q)
 286                goto drop;
 287
 288        if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len)
 289                goto drop;
 290
 291        skb->dev = dev;
 292        /* Apply the forward feature mask so that we perform segmentation
 293         * according to users wishes.  This only works if VNET_HDR is
 294         * enabled.
 295         */
 296        if (q->flags & IFF_VNET_HDR)
 297                features |= vlan->tap_features;
 298        if (netif_needs_gso(skb, features)) {
 299                struct sk_buff *segs = __skb_gso_segment(skb, features, false);
 300
 301                if (IS_ERR(segs))
 302                        goto drop;
 303
 304                if (!segs) {
 305                        skb_queue_tail(&q->sk.sk_receive_queue, skb);
 306                        goto wake_up;
 307                }
 308
 309                kfree_skb(skb);
 310                while (segs) {
 311                        struct sk_buff *nskb = segs->next;
 312
 313                        segs->next = NULL;
 314                        skb_queue_tail(&q->sk.sk_receive_queue, segs);
 315                        segs = nskb;
 316                }
 317        } else {
 318                skb_queue_tail(&q->sk.sk_receive_queue, skb);
 319        }
 320
 321wake_up:
 322        wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND);
 323        return NET_RX_SUCCESS;
 324
 325drop:
 326        kfree_skb(skb);
 327        return NET_RX_DROP;
 328}
 329
 330/*
 331 * Receive is for data from the external interface (lowerdev),
 332 * in case of macvtap, we can treat that the same way as
 333 * forward, which macvlan cannot.
 334 */
 335static int macvtap_receive(struct sk_buff *skb)
 336{
 337        skb_push(skb, ETH_HLEN);
 338        return macvtap_forward(skb->dev, skb);
 339}
 340
 341static int macvtap_get_minor(struct macvlan_dev *vlan)
 342{
 343        int retval = -ENOMEM;
 344
 345        mutex_lock(&minor_lock);
 346        retval = idr_alloc(&minor_idr, vlan, 1, MACVTAP_NUM_DEVS, GFP_KERNEL);
 347        if (retval >= 0) {
 348                vlan->minor = retval;
 349        } else if (retval == -ENOSPC) {
 350                printk(KERN_ERR "too many macvtap devices\n");
 351                retval = -EINVAL;
 352        }
 353        mutex_unlock(&minor_lock);
 354        return retval < 0 ? retval : 0;
 355}
 356
 357static void macvtap_free_minor(struct macvlan_dev *vlan)
 358{
 359        mutex_lock(&minor_lock);
 360        if (vlan->minor) {
 361                idr_remove(&minor_idr, vlan->minor);
 362                vlan->minor = 0;
 363        }
 364        mutex_unlock(&minor_lock);
 365}
 366
 367static struct net_device *dev_get_by_macvtap_minor(int minor)
 368{
 369        struct net_device *dev = NULL;
 370        struct macvlan_dev *vlan;
 371
 372        mutex_lock(&minor_lock);
 373        vlan = idr_find(&minor_idr, minor);
 374        if (vlan) {
 375                dev = vlan->dev;
 376                dev_hold(dev);
 377        }
 378        mutex_unlock(&minor_lock);
 379        return dev;
 380}
 381
 382static int macvtap_newlink(struct net *src_net,
 383                           struct net_device *dev,
 384                           struct nlattr *tb[],
 385                           struct nlattr *data[])
 386{
 387        struct macvlan_dev *vlan = netdev_priv(dev);
 388        INIT_LIST_HEAD(&vlan->queue_list);
 389
 390        /* Since macvlan supports all offloads by default, make
 391         * tap support all offloads also.
 392         */
 393        vlan->tap_features = TUN_OFFLOADS;
 394
 395        /* Don't put anything that may fail after macvlan_common_newlink
 396         * because we can't undo what it does.
 397         */
 398        return macvlan_common_newlink(src_net, dev, tb, data,
 399                                      macvtap_receive, macvtap_forward);
 400}
 401
 402static void macvtap_dellink(struct net_device *dev,
 403                            struct list_head *head)
 404{
 405        macvtap_del_queues(dev);
 406        macvlan_dellink(dev, head);
 407}
 408
 409static void macvtap_setup(struct net_device *dev)
 410{
 411        macvlan_common_setup(dev);
 412        dev->tx_queue_len = TUN_READQ_SIZE;
 413}
 414
 415static struct rtnl_link_ops macvtap_link_ops __read_mostly = {
 416        .kind           = "macvtap",
 417        .setup          = macvtap_setup,
 418        .newlink        = macvtap_newlink,
 419        .dellink        = macvtap_dellink,
 420};
 421
 422
 423static void macvtap_sock_write_space(struct sock *sk)
 424{
 425        wait_queue_head_t *wqueue;
 426
 427        if (!sock_writeable(sk) ||
 428            !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
 429                return;
 430
 431        wqueue = sk_sleep(sk);
 432        if (wqueue && waitqueue_active(wqueue))
 433                wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND);
 434}
 435
 436static void macvtap_sock_destruct(struct sock *sk)
 437{
 438        skb_queue_purge(&sk->sk_receive_queue);
 439}
 440
 441static int macvtap_open(struct inode *inode, struct file *file)
 442{
 443        struct net *net = current->nsproxy->net_ns;
 444        struct net_device *dev = dev_get_by_macvtap_minor(iminor(inode));
 445        struct macvtap_queue *q;
 446        int err;
 447
 448        err = -ENODEV;
 449        if (!dev)
 450                goto out;
 451
 452        err = -ENOMEM;
 453        q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
 454                                             &macvtap_proto);
 455        if (!q)
 456                goto out;
 457
 458        RCU_INIT_POINTER(q->sock.wq, &q->wq);
 459        init_waitqueue_head(&q->wq.wait);
 460        q->sock.type = SOCK_RAW;
 461        q->sock.state = SS_CONNECTED;
 462        q->sock.file = file;
 463        q->sock.ops = &macvtap_socket_ops;
 464        sock_init_data(&q->sock, &q->sk);
 465        q->sk.sk_write_space = macvtap_sock_write_space;
 466        q->sk.sk_destruct = macvtap_sock_destruct;
 467        q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
 468        q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
 469
 470        /*
 471         * so far only KVM virtio_net uses macvtap, enable zero copy between
 472         * guest kernel and host kernel when lower device supports zerocopy
 473         *
 474         * The macvlan supports zerocopy iff the lower device supports zero
 475         * copy so we don't have to look at the lower device directly.
 476         */
 477        if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG))
 478                sock_set_flag(&q->sk, SOCK_ZEROCOPY);
 479
 480        err = macvtap_set_queue(dev, file, q);
 481        if (err)
 482                sock_put(&q->sk);
 483
 484out:
 485        if (dev)
 486                dev_put(dev);
 487
 488        return err;
 489}
 490
 491static int macvtap_release(struct inode *inode, struct file *file)
 492{
 493        struct macvtap_queue *q = file->private_data;
 494        macvtap_put_queue(q);
 495        return 0;
 496}
 497
 498static unsigned int macvtap_poll(struct file *file, poll_table * wait)
 499{
 500        struct macvtap_queue *q = file->private_data;
 501        unsigned int mask = POLLERR;
 502
 503        if (!q)
 504                goto out;
 505
 506        mask = 0;
 507        poll_wait(file, &q->wq.wait, wait);
 508
 509        if (!skb_queue_empty(&q->sk.sk_receive_queue))
 510                mask |= POLLIN | POLLRDNORM;
 511
 512        if (sock_writeable(&q->sk) ||
 513            (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) &&
 514             sock_writeable(&q->sk)))
 515                mask |= POLLOUT | POLLWRNORM;
 516
 517out:
 518        return mask;
 519}
 520
 521static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad,
 522                                                size_t len, size_t linear,
 523                                                int noblock, int *err)
 524{
 525        struct sk_buff *skb;
 526
 527        /* Under a page?  Don't bother with paged skb. */
 528        if (prepad + len < PAGE_SIZE || !linear)
 529                linear = len;
 530
 531        skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
 532                                   err);
 533        if (!skb)
 534                return NULL;
 535
 536        skb_reserve(skb, prepad);
 537        skb_put(skb, linear);
 538        skb->data_len = len - linear;
 539        skb->len += len - linear;
 540
 541        return skb;
 542}
 543
 544/* set skb frags from iovec, this can move to core network code for reuse */
 545static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
 546                                  int offset, size_t count)
 547{
 548        int len = iov_length(from, count) - offset;
 549        int copy = skb_headlen(skb);
 550        int size, offset1 = 0;
 551        int i = 0;
 552
 553        /* Skip over from offset */
 554        while (count && (offset >= from->iov_len)) {
 555                offset -= from->iov_len;
 556                ++from;
 557                --count;
 558        }
 559
 560        /* copy up to skb headlen */
 561        while (count && (copy > 0)) {
 562                size = min_t(unsigned int, copy, from->iov_len - offset);
 563                if (copy_from_user(skb->data + offset1, from->iov_base + offset,
 564                                   size))
 565                        return -EFAULT;
 566                if (copy > size) {
 567                        ++from;
 568                        --count;
 569                        offset = 0;
 570                } else
 571                        offset += size;
 572                copy -= size;
 573                offset1 += size;
 574        }
 575
 576        if (len == offset1)
 577                return 0;
 578
 579        while (count--) {
 580                struct page *page[MAX_SKB_FRAGS];
 581                int num_pages;
 582                unsigned long base;
 583                unsigned long truesize;
 584
 585                len = from->iov_len - offset;
 586                if (!len) {
 587                        offset = 0;
 588                        ++from;
 589                        continue;
 590                }
 591                base = (unsigned long)from->iov_base + offset;
 592                size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
 593                if (i + size > MAX_SKB_FRAGS)
 594                        return -EMSGSIZE;
 595                num_pages = get_user_pages_fast(base, size, 0, &page[i]);
 596                if (num_pages != size) {
 597                        int j;
 598
 599                        for (j = 0; j < num_pages; j++)
 600                                put_page(page[i + j]);
 601                        return -EFAULT;
 602                }
 603                truesize = size * PAGE_SIZE;
 604                skb->data_len += len;
 605                skb->len += len;
 606                skb->truesize += truesize;
 607                atomic_add(truesize, &skb->sk->sk_wmem_alloc);
 608                while (len) {
 609                        int off = base & ~PAGE_MASK;
 610                        int size = min_t(int, len, PAGE_SIZE - off);
 611                        __skb_fill_page_desc(skb, i, page[i], off, size);
 612                        skb_shinfo(skb)->nr_frags++;
 613                        /* increase sk_wmem_alloc */
 614                        base += size;
 615                        len -= size;
 616                        i++;
 617                }
 618                offset = 0;
 619                ++from;
 620        }
 621        return 0;
 622}
 623
 624/*
 625 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
 626 * be shared with the tun/tap driver.
 627 */
 628static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb,
 629                                     struct virtio_net_hdr *vnet_hdr)
 630{
 631        unsigned short gso_type = 0;
 632        if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
 633                switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
 634                case VIRTIO_NET_HDR_GSO_TCPV4:
 635                        gso_type = SKB_GSO_TCPV4;
 636                        break;
 637                case VIRTIO_NET_HDR_GSO_TCPV6:
 638                        gso_type = SKB_GSO_TCPV6;
 639                        break;
 640                case VIRTIO_NET_HDR_GSO_UDP:
 641                        gso_type = SKB_GSO_UDP;
 642                        break;
 643                default:
 644                        return -EINVAL;
 645                }
 646
 647                if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
 648                        gso_type |= SKB_GSO_TCP_ECN;
 649
 650                if (vnet_hdr->gso_size == 0)
 651                        return -EINVAL;
 652        }
 653
 654        if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
 655                if (!skb_partial_csum_set(skb, vnet_hdr->csum_start,
 656                                          vnet_hdr->csum_offset))
 657                        return -EINVAL;
 658        }
 659
 660        if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
 661                skb_shinfo(skb)->gso_size = vnet_hdr->gso_size;
 662                skb_shinfo(skb)->gso_type = gso_type;
 663
 664                /* Header must be checked, and gso_segs computed. */
 665                skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
 666                skb_shinfo(skb)->gso_segs = 0;
 667        }
 668        return 0;
 669}
 670
 671static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb,
 672                                   struct virtio_net_hdr *vnet_hdr)
 673{
 674        memset(vnet_hdr, 0, sizeof(*vnet_hdr));
 675
 676        if (skb_is_gso(skb)) {
 677                struct skb_shared_info *sinfo = skb_shinfo(skb);
 678
 679                /* This is a hint as to how much should be linear. */
 680                vnet_hdr->hdr_len = skb_headlen(skb);
 681                vnet_hdr->gso_size = sinfo->gso_size;
 682                if (sinfo->gso_type & SKB_GSO_TCPV4)
 683                        vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
 684                else if (sinfo->gso_type & SKB_GSO_TCPV6)
 685                        vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
 686                else if (sinfo->gso_type & SKB_GSO_UDP)
 687                        vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
 688                else
 689                        BUG();
 690                if (sinfo->gso_type & SKB_GSO_TCP_ECN)
 691                        vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
 692        } else
 693                vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
 694
 695        if (skb->ip_summed == CHECKSUM_PARTIAL) {
 696                vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 697                vnet_hdr->csum_start = skb_checksum_start_offset(skb);
 698                vnet_hdr->csum_offset = skb->csum_offset;
 699        } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
 700                vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID;
 701        } /* else everything is zero */
 702
 703        return 0;
 704}
 705
 706static unsigned long iov_pages(const struct iovec *iv, int offset,
 707                               unsigned long nr_segs)
 708{
 709        unsigned long seg, base;
 710        int pages = 0, len, size;
 711
 712        while (nr_segs && (offset >= iv->iov_len)) {
 713                offset -= iv->iov_len;
 714                ++iv;
 715                --nr_segs;
 716        }
 717
 718        for (seg = 0; seg < nr_segs; seg++) {
 719                base = (unsigned long)iv[seg].iov_base + offset;
 720                len = iv[seg].iov_len - offset;
 721                size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
 722                pages += size;
 723                offset = 0;
 724        }
 725
 726        return pages;
 727}
 728
 729/* Get packet from user space buffer */
 730static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m,
 731                                const struct iovec *iv, unsigned long total_len,
 732                                size_t count, int noblock)
 733{
 734        struct sk_buff *skb;
 735        struct macvlan_dev *vlan;
 736        unsigned long len = total_len;
 737        int err;
 738        struct virtio_net_hdr vnet_hdr = { 0 };
 739        int vnet_hdr_len = 0;
 740        int copylen = 0;
 741        bool zerocopy = false;
 742        size_t linear;
 743
 744        if (q->flags & IFF_VNET_HDR) {
 745                vnet_hdr_len = q->vnet_hdr_sz;
 746
 747                err = -EINVAL;
 748                if (len < vnet_hdr_len)
 749                        goto err;
 750                len -= vnet_hdr_len;
 751
 752                err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0,
 753                                           sizeof(vnet_hdr));
 754                if (err < 0)
 755                        goto err;
 756                if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
 757                     vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
 758                                                        vnet_hdr.hdr_len)
 759                        vnet_hdr.hdr_len = vnet_hdr.csum_start +
 760                                                vnet_hdr.csum_offset + 2;
 761                err = -EINVAL;
 762                if (vnet_hdr.hdr_len > len)
 763                        goto err;
 764        }
 765
 766        err = -EINVAL;
 767        if (unlikely(len < ETH_HLEN))
 768                goto err;
 769
 770        err = -EMSGSIZE;
 771        if (unlikely(count > UIO_MAXIOV))
 772                goto err;
 773
 774        if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) {
 775                copylen = vnet_hdr.hdr_len ? vnet_hdr.hdr_len : GOODCOPY_LEN;
 776                linear = copylen;
 777                if (iov_pages(iv, vnet_hdr_len + copylen, count)
 778                    <= MAX_SKB_FRAGS)
 779                        zerocopy = true;
 780        }
 781
 782        if (!zerocopy) {
 783                copylen = len;
 784                linear = vnet_hdr.hdr_len;
 785        }
 786
 787        skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, copylen,
 788                                linear, noblock, &err);
 789        if (!skb)
 790                goto err;
 791
 792        if (zerocopy)
 793                err = zerocopy_sg_from_iovec(skb, iv, vnet_hdr_len, count);
 794        else {
 795                err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len,
 796                                                   len);
 797                if (!err && m && m->msg_control) {
 798                        struct ubuf_info *uarg = m->msg_control;
 799                        uarg->callback(uarg, false);
 800                }
 801        }
 802
 803        if (err)
 804                goto err_kfree;
 805
 806        skb_set_network_header(skb, ETH_HLEN);
 807        skb_reset_mac_header(skb);
 808        skb->protocol = eth_hdr(skb)->h_proto;
 809
 810        if (vnet_hdr_len) {
 811                err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr);
 812                if (err)
 813                        goto err_kfree;
 814        }
 815
 816        skb_probe_transport_header(skb, ETH_HLEN);
 817
 818        rcu_read_lock();
 819        vlan = rcu_dereference(q->vlan);
 820        /* copy skb_ubuf_info for callback when skb has no error */
 821        if (zerocopy) {
 822                skb_shinfo(skb)->destructor_arg = m->msg_control;
 823                skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
 824                skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
 825        }
 826        if (vlan) {
 827                local_bh_disable();
 828                macvlan_start_xmit(skb, vlan->dev);
 829                local_bh_enable();
 830        } else {
 831                kfree_skb(skb);
 832        }
 833        rcu_read_unlock();
 834
 835        return total_len;
 836
 837err_kfree:
 838        kfree_skb(skb);
 839
 840err:
 841        rcu_read_lock();
 842        vlan = rcu_dereference(q->vlan);
 843        if (vlan)
 844                vlan->dev->stats.tx_dropped++;
 845        rcu_read_unlock();
 846
 847        return err;
 848}
 849
 850static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv,
 851                                 unsigned long count, loff_t pos)
 852{
 853        struct file *file = iocb->ki_filp;
 854        ssize_t result = -ENOLINK;
 855        struct macvtap_queue *q = file->private_data;
 856
 857        result = macvtap_get_user(q, NULL, iv, iov_length(iv, count), count,
 858                                  file->f_flags & O_NONBLOCK);
 859        return result;
 860}
 861
 862/* Put packet to the user space buffer */
 863static ssize_t macvtap_put_user(struct macvtap_queue *q,
 864                                const struct sk_buff *skb,
 865                                const struct iovec *iv, int len)
 866{
 867        struct macvlan_dev *vlan;
 868        int ret;
 869        int vnet_hdr_len = 0;
 870        int vlan_offset = 0;
 871        int copied;
 872
 873        if (q->flags & IFF_VNET_HDR) {
 874                struct virtio_net_hdr vnet_hdr;
 875                vnet_hdr_len = q->vnet_hdr_sz;
 876                if ((len -= vnet_hdr_len) < 0)
 877                        return -EINVAL;
 878
 879                ret = macvtap_skb_to_vnet_hdr(skb, &vnet_hdr);
 880                if (ret)
 881                        return ret;
 882
 883                if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, sizeof(vnet_hdr)))
 884                        return -EFAULT;
 885        }
 886        copied = vnet_hdr_len;
 887
 888        if (!vlan_tx_tag_present(skb))
 889                len = min_t(int, skb->len, len);
 890        else {
 891                int copy;
 892                struct {
 893                        __be16 h_vlan_proto;
 894                        __be16 h_vlan_TCI;
 895                } veth;
 896                veth.h_vlan_proto = skb->vlan_proto;
 897                veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb));
 898
 899                vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
 900                len = min_t(int, skb->len + VLAN_HLEN, len);
 901
 902                copy = min_t(int, vlan_offset, len);
 903                ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy);
 904                len -= copy;
 905                copied += copy;
 906                if (ret || !len)
 907                        goto done;
 908
 909                copy = min_t(int, sizeof(veth), len);
 910                ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy);
 911                len -= copy;
 912                copied += copy;
 913                if (ret || !len)
 914                        goto done;
 915        }
 916
 917        ret = skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len);
 918        copied += len;
 919
 920done:
 921        rcu_read_lock();
 922        vlan = rcu_dereference(q->vlan);
 923        if (vlan) {
 924                preempt_disable();
 925                macvlan_count_rx(vlan, copied - vnet_hdr_len, ret == 0, 0);
 926                preempt_enable();
 927        }
 928        rcu_read_unlock();
 929
 930        return ret ? ret : copied;
 931}
 932
 933static ssize_t macvtap_do_read(struct macvtap_queue *q, struct kiocb *iocb,
 934                               const struct iovec *iv, unsigned long len,
 935                               int noblock)
 936{
 937        DEFINE_WAIT(wait);
 938        struct sk_buff *skb;
 939        ssize_t ret = 0;
 940
 941        while (len) {
 942                if (!noblock)
 943                        prepare_to_wait(sk_sleep(&q->sk), &wait,
 944                                        TASK_INTERRUPTIBLE);
 945
 946                /* Read frames from the queue */
 947                skb = skb_dequeue(&q->sk.sk_receive_queue);
 948                if (!skb) {
 949                        if (noblock) {
 950                                ret = -EAGAIN;
 951                                break;
 952                        }
 953                        if (signal_pending(current)) {
 954                                ret = -ERESTARTSYS;
 955                                break;
 956                        }
 957                        /* Nothing to read, let's sleep */
 958                        schedule();
 959                        continue;
 960                }
 961                ret = macvtap_put_user(q, skb, iv, len);
 962                kfree_skb(skb);
 963                break;
 964        }
 965
 966        if (!noblock)
 967                finish_wait(sk_sleep(&q->sk), &wait);
 968        return ret;
 969}
 970
 971static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv,
 972                                unsigned long count, loff_t pos)
 973{
 974        struct file *file = iocb->ki_filp;
 975        struct macvtap_queue *q = file->private_data;
 976        ssize_t len, ret = 0;
 977
 978        len = iov_length(iv, count);
 979        if (len < 0) {
 980                ret = -EINVAL;
 981                goto out;
 982        }
 983
 984        ret = macvtap_do_read(q, iocb, iv, len, file->f_flags & O_NONBLOCK);
 985        ret = min_t(ssize_t, ret, len); /* XXX copied from tun.c. Why? */
 986out:
 987        return ret;
 988}
 989
 990static struct macvlan_dev *macvtap_get_vlan(struct macvtap_queue *q)
 991{
 992        struct macvlan_dev *vlan;
 993
 994        ASSERT_RTNL();
 995        vlan = rtnl_dereference(q->vlan);
 996        if (vlan)
 997                dev_hold(vlan->dev);
 998
 999        return vlan;
1000}
1001
1002static void macvtap_put_vlan(struct macvlan_dev *vlan)
1003{
1004        dev_put(vlan->dev);
1005}
1006
1007static int macvtap_ioctl_set_queue(struct file *file, unsigned int flags)
1008{
1009        struct macvtap_queue *q = file->private_data;
1010        struct macvlan_dev *vlan;
1011        int ret;
1012
1013        vlan = macvtap_get_vlan(q);
1014        if (!vlan)
1015                return -EINVAL;
1016
1017        if (flags & IFF_ATTACH_QUEUE)
1018                ret = macvtap_enable_queue(vlan->dev, file, q);
1019        else if (flags & IFF_DETACH_QUEUE)
1020                ret = macvtap_disable_queue(q);
1021        else
1022                ret = -EINVAL;
1023
1024        macvtap_put_vlan(vlan);
1025        return ret;
1026}
1027
1028static int set_offload(struct macvtap_queue *q, unsigned long arg)
1029{
1030        struct macvlan_dev *vlan;
1031        netdev_features_t features;
1032        netdev_features_t feature_mask = 0;
1033
1034        vlan = rtnl_dereference(q->vlan);
1035        if (!vlan)
1036                return -ENOLINK;
1037
1038        features = vlan->dev->features;
1039
1040        if (arg & TUN_F_CSUM) {
1041                feature_mask = NETIF_F_HW_CSUM;
1042
1043                if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) {
1044                        if (arg & TUN_F_TSO_ECN)
1045                                feature_mask |= NETIF_F_TSO_ECN;
1046                        if (arg & TUN_F_TSO4)
1047                                feature_mask |= NETIF_F_TSO;
1048                        if (arg & TUN_F_TSO6)
1049                                feature_mask |= NETIF_F_TSO6;
1050                }
1051
1052                if (arg & TUN_F_UFO)
1053                        feature_mask |= NETIF_F_UFO;
1054        }
1055
1056        /* tun/tap driver inverts the usage for TSO offloads, where
1057         * setting the TSO bit means that the userspace wants to
1058         * accept TSO frames and turning it off means that user space
1059         * does not support TSO.
1060         * For macvtap, we have to invert it to mean the same thing.
1061         * When user space turns off TSO, we turn off GSO/LRO so that
1062         * user-space will not receive TSO frames.
1063         */
1064        if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_UFO))
1065                features |= RX_OFFLOADS;
1066        else
1067                features &= ~RX_OFFLOADS;
1068
1069        /* tap_features are the same as features on tun/tap and
1070         * reflect user expectations.
1071         */
1072        vlan->tap_features = feature_mask;
1073        vlan->set_features = features;
1074        netdev_update_features(vlan->dev);
1075
1076        return 0;
1077}
1078
1079/*
1080 * provide compatibility with generic tun/tap interface
1081 */
1082static long macvtap_ioctl(struct file *file, unsigned int cmd,
1083                          unsigned long arg)
1084{
1085        struct macvtap_queue *q = file->private_data;
1086        struct macvlan_dev *vlan;
1087        void __user *argp = (void __user *)arg;
1088        struct ifreq __user *ifr = argp;
1089        unsigned int __user *up = argp;
1090        unsigned int u;
1091        int __user *sp = argp;
1092        int s;
1093        int ret;
1094
1095        switch (cmd) {
1096        case TUNSETIFF:
1097                /* ignore the name, just look at flags */
1098                if (get_user(u, &ifr->ifr_flags))
1099                        return -EFAULT;
1100
1101                ret = 0;
1102                if ((u & ~(IFF_VNET_HDR | IFF_MULTI_QUEUE)) !=
1103                    (IFF_NO_PI | IFF_TAP))
1104                        ret = -EINVAL;
1105                else
1106                        q->flags = u;
1107
1108                return ret;
1109
1110        case TUNGETIFF:
1111                rtnl_lock();
1112                vlan = macvtap_get_vlan(q);
1113                if (!vlan) {
1114                        rtnl_unlock();
1115                        return -ENOLINK;
1116                }
1117
1118                ret = 0;
1119                if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) ||
1120                    put_user(q->flags, &ifr->ifr_flags))
1121                        ret = -EFAULT;
1122                macvtap_put_vlan(vlan);
1123                rtnl_unlock();
1124                return ret;
1125
1126        case TUNSETQUEUE:
1127                if (get_user(u, &ifr->ifr_flags))
1128                        return -EFAULT;
1129                rtnl_lock();
1130                ret = macvtap_ioctl_set_queue(file, u);
1131                rtnl_unlock();
1132                return ret;
1133
1134        case TUNGETFEATURES:
1135                if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR |
1136                             IFF_MULTI_QUEUE, up))
1137                        return -EFAULT;
1138                return 0;
1139
1140        case TUNSETSNDBUF:
1141                if (get_user(u, up))
1142                        return -EFAULT;
1143
1144                q->sk.sk_sndbuf = u;
1145                return 0;
1146
1147        case TUNGETVNETHDRSZ:
1148                s = q->vnet_hdr_sz;
1149                if (put_user(s, sp))
1150                        return -EFAULT;
1151                return 0;
1152
1153        case TUNSETVNETHDRSZ:
1154                if (get_user(s, sp))
1155                        return -EFAULT;
1156                if (s < (int)sizeof(struct virtio_net_hdr))
1157                        return -EINVAL;
1158
1159                q->vnet_hdr_sz = s;
1160                return 0;
1161
1162        case TUNSETOFFLOAD:
1163                /* let the user check for future flags */
1164                if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
1165                            TUN_F_TSO_ECN | TUN_F_UFO))
1166                        return -EINVAL;
1167
1168                rtnl_lock();
1169                ret = set_offload(q, arg);
1170                rtnl_unlock();
1171                return ret;
1172
1173        default:
1174                return -EINVAL;
1175        }
1176}
1177
1178#ifdef CONFIG_COMPAT
1179static long macvtap_compat_ioctl(struct file *file, unsigned int cmd,
1180                                 unsigned long arg)
1181{
1182        return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
1183}
1184#endif
1185
1186static const struct file_operations macvtap_fops = {
1187        .owner          = THIS_MODULE,
1188        .open           = macvtap_open,
1189        .release        = macvtap_release,
1190        .aio_read       = macvtap_aio_read,
1191        .aio_write      = macvtap_aio_write,
1192        .poll           = macvtap_poll,
1193        .llseek         = no_llseek,
1194        .unlocked_ioctl = macvtap_ioctl,
1195#ifdef CONFIG_COMPAT
1196        .compat_ioctl   = macvtap_compat_ioctl,
1197#endif
1198};
1199
1200static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock,
1201                           struct msghdr *m, size_t total_len)
1202{
1203        struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1204        return macvtap_get_user(q, m, m->msg_iov, total_len, m->msg_iovlen,
1205                            m->msg_flags & MSG_DONTWAIT);
1206}
1207
1208static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock,
1209                           struct msghdr *m, size_t total_len,
1210                           int flags)
1211{
1212        struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1213        int ret;
1214        if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1215                return -EINVAL;
1216        ret = macvtap_do_read(q, iocb, m->msg_iov, total_len,
1217                          flags & MSG_DONTWAIT);
1218        if (ret > total_len) {
1219                m->msg_flags |= MSG_TRUNC;
1220                ret = flags & MSG_TRUNC ? ret : total_len;
1221        }
1222        return ret;
1223}
1224
1225/* Ops structure to mimic raw sockets with tun */
1226static const struct proto_ops macvtap_socket_ops = {
1227        .sendmsg = macvtap_sendmsg,
1228        .recvmsg = macvtap_recvmsg,
1229};
1230
1231/* Get an underlying socket object from tun file.  Returns error unless file is
1232 * attached to a device.  The returned object works like a packet socket, it
1233 * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
1234 * holding a reference to the file for as long as the socket is in use. */
1235struct socket *macvtap_get_socket(struct file *file)
1236{
1237        struct macvtap_queue *q;
1238        if (file->f_op != &macvtap_fops)
1239                return ERR_PTR(-EINVAL);
1240        q = file->private_data;
1241        if (!q)
1242                return ERR_PTR(-EBADFD);
1243        return &q->sock;
1244}
1245EXPORT_SYMBOL_GPL(macvtap_get_socket);
1246
1247static int macvtap_device_event(struct notifier_block *unused,
1248                                unsigned long event, void *ptr)
1249{
1250        struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1251        struct macvlan_dev *vlan;
1252        struct device *classdev;
1253        dev_t devt;
1254        int err;
1255
1256        if (dev->rtnl_link_ops != &macvtap_link_ops)
1257                return NOTIFY_DONE;
1258
1259        vlan = netdev_priv(dev);
1260
1261        switch (event) {
1262        case NETDEV_REGISTER:
1263                /* Create the device node here after the network device has
1264                 * been registered but before register_netdevice has
1265                 * finished running.
1266                 */
1267                err = macvtap_get_minor(vlan);
1268                if (err)
1269                        return notifier_from_errno(err);
1270
1271                devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
1272                classdev = device_create(macvtap_class, &dev->dev, devt,
1273                                         dev, "tap%d", dev->ifindex);
1274                if (IS_ERR(classdev)) {
1275                        macvtap_free_minor(vlan);
1276                        return notifier_from_errno(PTR_ERR(classdev));
1277                }
1278                break;
1279        case NETDEV_UNREGISTER:
1280                devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
1281                device_destroy(macvtap_class, devt);
1282                macvtap_free_minor(vlan);
1283                break;
1284        }
1285
1286        return NOTIFY_DONE;
1287}
1288
1289static struct notifier_block macvtap_notifier_block __read_mostly = {
1290        .notifier_call  = macvtap_device_event,
1291};
1292
1293static int macvtap_init(void)
1294{
1295        int err;
1296
1297        err = alloc_chrdev_region(&macvtap_major, 0,
1298                                MACVTAP_NUM_DEVS, "macvtap");
1299        if (err)
1300                goto out1;
1301
1302        cdev_init(&macvtap_cdev, &macvtap_fops);
1303        err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS);
1304        if (err)
1305                goto out2;
1306
1307        macvtap_class = class_create(THIS_MODULE, "macvtap");
1308        if (IS_ERR(macvtap_class)) {
1309                err = PTR_ERR(macvtap_class);
1310                goto out3;
1311        }
1312
1313        err = register_netdevice_notifier(&macvtap_notifier_block);
1314        if (err)
1315                goto out4;
1316
1317        err = macvlan_link_register(&macvtap_link_ops);
1318        if (err)
1319                goto out5;
1320
1321        return 0;
1322
1323out5:
1324        unregister_netdevice_notifier(&macvtap_notifier_block);
1325out4:
1326        class_unregister(macvtap_class);
1327out3:
1328        cdev_del(&macvtap_cdev);
1329out2:
1330        unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1331out1:
1332        return err;
1333}
1334module_init(macvtap_init);
1335
1336static void macvtap_exit(void)
1337{
1338        rtnl_link_unregister(&macvtap_link_ops);
1339        unregister_netdevice_notifier(&macvtap_notifier_block);
1340        class_unregister(macvtap_class);
1341        cdev_del(&macvtap_cdev);
1342        unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1343}
1344module_exit(macvtap_exit);
1345
1346MODULE_ALIAS_RTNL_LINK("macvtap");
1347MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1348MODULE_LICENSE("GPL");
1349
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.