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, 0);
 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/*
 545 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
 546 * be shared with the tun/tap driver.
 547 */
 548static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb,
 549                                     struct virtio_net_hdr *vnet_hdr)
 550{
 551        unsigned short gso_type = 0;
 552        if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
 553                switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
 554                case VIRTIO_NET_HDR_GSO_TCPV4:
 555                        gso_type = SKB_GSO_TCPV4;
 556                        break;
 557                case VIRTIO_NET_HDR_GSO_TCPV6:
 558                        gso_type = SKB_GSO_TCPV6;
 559                        break;
 560                case VIRTIO_NET_HDR_GSO_UDP:
 561                        gso_type = SKB_GSO_UDP;
 562                        break;
 563                default:
 564                        return -EINVAL;
 565                }
 566
 567                if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
 568                        gso_type |= SKB_GSO_TCP_ECN;
 569
 570                if (vnet_hdr->gso_size == 0)
 571                        return -EINVAL;
 572        }
 573
 574        if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
 575                if (!skb_partial_csum_set(skb, vnet_hdr->csum_start,
 576                                          vnet_hdr->csum_offset))
 577                        return -EINVAL;
 578        }
 579
 580        if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
 581                skb_shinfo(skb)->gso_size = vnet_hdr->gso_size;
 582                skb_shinfo(skb)->gso_type = gso_type;
 583
 584                /* Header must be checked, and gso_segs computed. */
 585                skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
 586                skb_shinfo(skb)->gso_segs = 0;
 587        }
 588        return 0;
 589}
 590
 591static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb,
 592                                   struct virtio_net_hdr *vnet_hdr)
 593{
 594        memset(vnet_hdr, 0, sizeof(*vnet_hdr));
 595
 596        if (skb_is_gso(skb)) {
 597                struct skb_shared_info *sinfo = skb_shinfo(skb);
 598
 599                /* This is a hint as to how much should be linear. */
 600                vnet_hdr->hdr_len = skb_headlen(skb);
 601                vnet_hdr->gso_size = sinfo->gso_size;
 602                if (sinfo->gso_type & SKB_GSO_TCPV4)
 603                        vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
 604                else if (sinfo->gso_type & SKB_GSO_TCPV6)
 605                        vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
 606                else if (sinfo->gso_type & SKB_GSO_UDP)
 607                        vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
 608                else
 609                        BUG();
 610                if (sinfo->gso_type & SKB_GSO_TCP_ECN)
 611                        vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
 612        } else
 613                vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
 614
 615        if (skb->ip_summed == CHECKSUM_PARTIAL) {
 616                vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 617                vnet_hdr->csum_start = skb_checksum_start_offset(skb);
 618                vnet_hdr->csum_offset = skb->csum_offset;
 619        } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
 620                vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID;
 621        } /* else everything is zero */
 622
 623        return 0;
 624}
 625
 626/* Get packet from user space buffer */
 627static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m,
 628                                const struct iovec *iv, unsigned long total_len,
 629                                size_t count, int noblock)
 630{
 631        struct sk_buff *skb;
 632        struct macvlan_dev *vlan;
 633        unsigned long len = total_len;
 634        int err;
 635        struct virtio_net_hdr vnet_hdr = { 0 };
 636        int vnet_hdr_len = 0;
 637        int copylen = 0;
 638        bool zerocopy = false;
 639        size_t linear;
 640
 641        if (q->flags & IFF_VNET_HDR) {
 642                vnet_hdr_len = q->vnet_hdr_sz;
 643
 644                err = -EINVAL;
 645                if (len < vnet_hdr_len)
 646                        goto err;
 647                len -= vnet_hdr_len;
 648
 649                err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0,
 650                                           sizeof(vnet_hdr));
 651                if (err < 0)
 652                        goto err;
 653                if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
 654                     vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
 655                                                        vnet_hdr.hdr_len)
 656                        vnet_hdr.hdr_len = vnet_hdr.csum_start +
 657                                                vnet_hdr.csum_offset + 2;
 658                err = -EINVAL;
 659                if (vnet_hdr.hdr_len > len)
 660                        goto err;
 661        }
 662
 663        err = -EINVAL;
 664        if (unlikely(len < ETH_HLEN))
 665                goto err;
 666
 667        err = -EMSGSIZE;
 668        if (unlikely(count > UIO_MAXIOV))
 669                goto err;
 670
 671        if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) {
 672                copylen = vnet_hdr.hdr_len ? vnet_hdr.hdr_len : GOODCOPY_LEN;
 673                linear = copylen;
 674                if (iov_pages(iv, vnet_hdr_len + copylen, count)
 675                    <= MAX_SKB_FRAGS)
 676                        zerocopy = true;
 677        }
 678
 679        if (!zerocopy) {
 680                copylen = len;
 681                linear = vnet_hdr.hdr_len;
 682        }
 683
 684        skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, copylen,
 685                                linear, noblock, &err);
 686        if (!skb)
 687                goto err;
 688
 689        if (zerocopy)
 690                err = zerocopy_sg_from_iovec(skb, iv, vnet_hdr_len, count);
 691        else {
 692                err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len,
 693                                                   len);
 694                if (!err && m && m->msg_control) {
 695                        struct ubuf_info *uarg = m->msg_control;
 696                        uarg->callback(uarg, false);
 697                }
 698        }
 699
 700        if (err)
 701                goto err_kfree;
 702
 703        skb_set_network_header(skb, ETH_HLEN);
 704        skb_reset_mac_header(skb);
 705        skb->protocol = eth_hdr(skb)->h_proto;
 706
 707        if (vnet_hdr_len) {
 708                err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr);
 709                if (err)
 710                        goto err_kfree;
 711        }
 712
 713        skb_probe_transport_header(skb, ETH_HLEN);
 714
 715        rcu_read_lock();
 716        vlan = rcu_dereference(q->vlan);
 717        /* copy skb_ubuf_info for callback when skb has no error */
 718        if (zerocopy) {
 719                skb_shinfo(skb)->destructor_arg = m->msg_control;
 720                skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
 721                skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
 722        }
 723        if (vlan) {
 724                local_bh_disable();
 725                macvlan_start_xmit(skb, vlan->dev);
 726                local_bh_enable();
 727        } else {
 728                kfree_skb(skb);
 729        }
 730        rcu_read_unlock();
 731
 732        return total_len;
 733
 734err_kfree:
 735        kfree_skb(skb);
 736
 737err:
 738        rcu_read_lock();
 739        vlan = rcu_dereference(q->vlan);
 740        if (vlan)
 741                vlan->dev->stats.tx_dropped++;
 742        rcu_read_unlock();
 743
 744        return err;
 745}
 746
 747static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv,
 748                                 unsigned long count, loff_t pos)
 749{
 750        struct file *file = iocb->ki_filp;
 751        ssize_t result = -ENOLINK;
 752        struct macvtap_queue *q = file->private_data;
 753
 754        result = macvtap_get_user(q, NULL, iv, iov_length(iv, count), count,
 755                                  file->f_flags & O_NONBLOCK);
 756        return result;
 757}
 758
 759/* Put packet to the user space buffer */
 760static ssize_t macvtap_put_user(struct macvtap_queue *q,
 761                                const struct sk_buff *skb,
 762                                const struct iovec *iv, int len)
 763{
 764        struct macvlan_dev *vlan;
 765        int ret;
 766        int vnet_hdr_len = 0;
 767        int vlan_offset = 0;
 768        int copied;
 769
 770        if (q->flags & IFF_VNET_HDR) {
 771                struct virtio_net_hdr vnet_hdr;
 772                vnet_hdr_len = q->vnet_hdr_sz;
 773                if ((len -= vnet_hdr_len) < 0)
 774                        return -EINVAL;
 775
 776                ret = macvtap_skb_to_vnet_hdr(skb, &vnet_hdr);
 777                if (ret)
 778                        return ret;
 779
 780                if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, sizeof(vnet_hdr)))
 781                        return -EFAULT;
 782        }
 783        copied = vnet_hdr_len;
 784
 785        if (!vlan_tx_tag_present(skb))
 786                len = min_t(int, skb->len, len);
 787        else {
 788                int copy;
 789                struct {
 790                        __be16 h_vlan_proto;
 791                        __be16 h_vlan_TCI;
 792                } veth;
 793                veth.h_vlan_proto = skb->vlan_proto;
 794                veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb));
 795
 796                vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
 797                len = min_t(int, skb->len + VLAN_HLEN, len);
 798
 799                copy = min_t(int, vlan_offset, len);
 800                ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy);
 801                len -= copy;
 802                copied += copy;
 803                if (ret || !len)
 804                        goto done;
 805
 806                copy = min_t(int, sizeof(veth), len);
 807                ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy);
 808                len -= copy;
 809                copied += copy;
 810                if (ret || !len)
 811                        goto done;
 812        }
 813
 814        ret = skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len);
 815        copied += len;
 816
 817done:
 818        rcu_read_lock();
 819        vlan = rcu_dereference(q->vlan);
 820        if (vlan) {
 821                preempt_disable();
 822                macvlan_count_rx(vlan, copied - vnet_hdr_len, ret == 0, 0);
 823                preempt_enable();
 824        }
 825        rcu_read_unlock();
 826
 827        return ret ? ret : copied;
 828}
 829
 830static ssize_t macvtap_do_read(struct macvtap_queue *q, struct kiocb *iocb,
 831                               const struct iovec *iv, unsigned long len,
 832                               int noblock)
 833{
 834        DEFINE_WAIT(wait);
 835        struct sk_buff *skb;
 836        ssize_t ret = 0;
 837
 838        while (len) {
 839                if (!noblock)
 840                        prepare_to_wait(sk_sleep(&q->sk), &wait,
 841                                        TASK_INTERRUPTIBLE);
 842
 843                /* Read frames from the queue */
 844                skb = skb_dequeue(&q->sk.sk_receive_queue);
 845                if (!skb) {
 846                        if (noblock) {
 847                                ret = -EAGAIN;
 848                                break;
 849                        }
 850                        if (signal_pending(current)) {
 851                                ret = -ERESTARTSYS;
 852                                break;
 853                        }
 854                        /* Nothing to read, let's sleep */
 855                        schedule();
 856                        continue;
 857                }
 858                ret = macvtap_put_user(q, skb, iv, len);
 859                kfree_skb(skb);
 860                break;
 861        }
 862
 863        if (!noblock)
 864                finish_wait(sk_sleep(&q->sk), &wait);
 865        return ret;
 866}
 867
 868static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv,
 869                                unsigned long count, loff_t pos)
 870{
 871        struct file *file = iocb->ki_filp;
 872        struct macvtap_queue *q = file->private_data;
 873        ssize_t len, ret = 0;
 874
 875        len = iov_length(iv, count);
 876        if (len < 0) {
 877                ret = -EINVAL;
 878                goto out;
 879        }
 880
 881        ret = macvtap_do_read(q, iocb, iv, len, file->f_flags & O_NONBLOCK);
 882        ret = min_t(ssize_t, ret, len); /* XXX copied from tun.c. Why? */
 883out:
 884        return ret;
 885}
 886
 887static struct macvlan_dev *macvtap_get_vlan(struct macvtap_queue *q)
 888{
 889        struct macvlan_dev *vlan;
 890
 891        ASSERT_RTNL();
 892        vlan = rtnl_dereference(q->vlan);
 893        if (vlan)
 894                dev_hold(vlan->dev);
 895
 896        return vlan;
 897}
 898
 899static void macvtap_put_vlan(struct macvlan_dev *vlan)
 900{
 901        dev_put(vlan->dev);
 902}
 903
 904static int macvtap_ioctl_set_queue(struct file *file, unsigned int flags)
 905{
 906        struct macvtap_queue *q = file->private_data;
 907        struct macvlan_dev *vlan;
 908        int ret;
 909
 910        vlan = macvtap_get_vlan(q);
 911        if (!vlan)
 912                return -EINVAL;
 913
 914        if (flags & IFF_ATTACH_QUEUE)
 915                ret = macvtap_enable_queue(vlan->dev, file, q);
 916        else if (flags & IFF_DETACH_QUEUE)
 917                ret = macvtap_disable_queue(q);
 918        else
 919                ret = -EINVAL;
 920
 921        macvtap_put_vlan(vlan);
 922        return ret;
 923}
 924
 925static int set_offload(struct macvtap_queue *q, unsigned long arg)
 926{
 927        struct macvlan_dev *vlan;
 928        netdev_features_t features;
 929        netdev_features_t feature_mask = 0;
 930
 931        vlan = rtnl_dereference(q->vlan);
 932        if (!vlan)
 933                return -ENOLINK;
 934
 935        features = vlan->dev->features;
 936
 937        if (arg & TUN_F_CSUM) {
 938                feature_mask = NETIF_F_HW_CSUM;
 939
 940                if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) {
 941                        if (arg & TUN_F_TSO_ECN)
 942                                feature_mask |= NETIF_F_TSO_ECN;
 943                        if (arg & TUN_F_TSO4)
 944                                feature_mask |= NETIF_F_TSO;
 945                        if (arg & TUN_F_TSO6)
 946                                feature_mask |= NETIF_F_TSO6;
 947                }
 948
 949                if (arg & TUN_F_UFO)
 950                        feature_mask |= NETIF_F_UFO;
 951        }
 952
 953        /* tun/tap driver inverts the usage for TSO offloads, where
 954         * setting the TSO bit means that the userspace wants to
 955         * accept TSO frames and turning it off means that user space
 956         * does not support TSO.
 957         * For macvtap, we have to invert it to mean the same thing.
 958         * When user space turns off TSO, we turn off GSO/LRO so that
 959         * user-space will not receive TSO frames.
 960         */
 961        if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_UFO))
 962                features |= RX_OFFLOADS;
 963        else
 964                features &= ~RX_OFFLOADS;
 965
 966        /* tap_features are the same as features on tun/tap and
 967         * reflect user expectations.
 968         */
 969        vlan->tap_features = feature_mask;
 970        vlan->set_features = features;
 971        netdev_update_features(vlan->dev);
 972
 973        return 0;
 974}
 975
 976/*
 977 * provide compatibility with generic tun/tap interface
 978 */
 979static long macvtap_ioctl(struct file *file, unsigned int cmd,
 980                          unsigned long arg)
 981{
 982        struct macvtap_queue *q = file->private_data;
 983        struct macvlan_dev *vlan;
 984        void __user *argp = (void __user *)arg;
 985        struct ifreq __user *ifr = argp;
 986        unsigned int __user *up = argp;
 987        unsigned int u;
 988        int __user *sp = argp;
 989        int s;
 990        int ret;
 991
 992        switch (cmd) {
 993        case TUNSETIFF:
 994                /* ignore the name, just look at flags */
 995                if (get_user(u, &ifr->ifr_flags))
 996                        return -EFAULT;
 997
 998                ret = 0;
 999                if ((u & ~(IFF_VNET_HDR | IFF_MULTI_QUEUE)) !=
1000                    (IFF_NO_PI | IFF_TAP))
1001                        ret = -EINVAL;
1002                else
1003                        q->flags = u;
1004
1005                return ret;
1006
1007        case TUNGETIFF:
1008                rtnl_lock();
1009                vlan = macvtap_get_vlan(q);
1010                if (!vlan) {
1011                        rtnl_unlock();
1012                        return -ENOLINK;
1013                }
1014
1015                ret = 0;
1016                if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) ||
1017                    put_user(q->flags, &ifr->ifr_flags))
1018                        ret = -EFAULT;
1019                macvtap_put_vlan(vlan);
1020                rtnl_unlock();
1021                return ret;
1022
1023        case TUNSETQUEUE:
1024                if (get_user(u, &ifr->ifr_flags))
1025                        return -EFAULT;
1026                rtnl_lock();
1027                ret = macvtap_ioctl_set_queue(file, u);
1028                rtnl_unlock();
1029                return ret;
1030
1031        case TUNGETFEATURES:
1032                if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR |
1033                             IFF_MULTI_QUEUE, up))
1034                        return -EFAULT;
1035                return 0;
1036
1037        case TUNSETSNDBUF:
1038                if (get_user(u, up))
1039                        return -EFAULT;
1040
1041                q->sk.sk_sndbuf = u;
1042                return 0;
1043
1044        case TUNGETVNETHDRSZ:
1045                s = q->vnet_hdr_sz;
1046                if (put_user(s, sp))
1047                        return -EFAULT;
1048                return 0;
1049
1050        case TUNSETVNETHDRSZ:
1051                if (get_user(s, sp))
1052                        return -EFAULT;
1053                if (s < (int)sizeof(struct virtio_net_hdr))
1054                        return -EINVAL;
1055
1056                q->vnet_hdr_sz = s;
1057                return 0;
1058
1059        case TUNSETOFFLOAD:
1060                /* let the user check for future flags */
1061                if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
1062                            TUN_F_TSO_ECN | TUN_F_UFO))
1063                        return -EINVAL;
1064
1065                rtnl_lock();
1066                ret = set_offload(q, arg);
1067                rtnl_unlock();
1068                return ret;
1069
1070        default:
1071                return -EINVAL;
1072        }
1073}
1074
1075#ifdef CONFIG_COMPAT
1076static long macvtap_compat_ioctl(struct file *file, unsigned int cmd,
1077                                 unsigned long arg)
1078{
1079        return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
1080}
1081#endif
1082
1083static const struct file_operations macvtap_fops = {
1084        .owner          = THIS_MODULE,
1085        .open           = macvtap_open,
1086        .release        = macvtap_release,
1087        .aio_read       = macvtap_aio_read,
1088        .aio_write      = macvtap_aio_write,
1089        .poll           = macvtap_poll,
1090        .llseek         = no_llseek,
1091        .unlocked_ioctl = macvtap_ioctl,
1092#ifdef CONFIG_COMPAT
1093        .compat_ioctl   = macvtap_compat_ioctl,
1094#endif
1095};
1096
1097static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock,
1098                           struct msghdr *m, size_t total_len)
1099{
1100        struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1101        return macvtap_get_user(q, m, m->msg_iov, total_len, m->msg_iovlen,
1102                            m->msg_flags & MSG_DONTWAIT);
1103}
1104
1105static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock,
1106                           struct msghdr *m, size_t total_len,
1107                           int flags)
1108{
1109        struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1110        int ret;
1111        if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1112                return -EINVAL;
1113        ret = macvtap_do_read(q, iocb, m->msg_iov, total_len,
1114                          flags & MSG_DONTWAIT);
1115        if (ret > total_len) {
1116                m->msg_flags |= MSG_TRUNC;
1117                ret = flags & MSG_TRUNC ? ret : total_len;
1118        }
1119        return ret;
1120}
1121
1122/* Ops structure to mimic raw sockets with tun */
1123static const struct proto_ops macvtap_socket_ops = {
1124        .sendmsg = macvtap_sendmsg,
1125        .recvmsg = macvtap_recvmsg,
1126};
1127
1128/* Get an underlying socket object from tun file.  Returns error unless file is
1129 * attached to a device.  The returned object works like a packet socket, it
1130 * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
1131 * holding a reference to the file for as long as the socket is in use. */
1132struct socket *macvtap_get_socket(struct file *file)
1133{
1134        struct macvtap_queue *q;
1135        if (file->f_op != &macvtap_fops)
1136                return ERR_PTR(-EINVAL);
1137        q = file->private_data;
1138        if (!q)
1139                return ERR_PTR(-EBADFD);
1140        return &q->sock;
1141}
1142EXPORT_SYMBOL_GPL(macvtap_get_socket);
1143
1144static int macvtap_device_event(struct notifier_block *unused,
1145                                unsigned long event, void *ptr)
1146{
1147        struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1148        struct macvlan_dev *vlan;
1149        struct device *classdev;
1150        dev_t devt;
1151        int err;
1152
1153        if (dev->rtnl_link_ops != &macvtap_link_ops)
1154                return NOTIFY_DONE;
1155
1156        vlan = netdev_priv(dev);
1157
1158        switch (event) {
1159        case NETDEV_REGISTER:
1160                /* Create the device node here after the network device has
1161                 * been registered but before register_netdevice has
1162                 * finished running.
1163                 */
1164                err = macvtap_get_minor(vlan);
1165                if (err)
1166                        return notifier_from_errno(err);
1167
1168                devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
1169                classdev = device_create(macvtap_class, &dev->dev, devt,
1170                                         dev, "tap%d", dev->ifindex);
1171                if (IS_ERR(classdev)) {
1172                        macvtap_free_minor(vlan);
1173                        return notifier_from_errno(PTR_ERR(classdev));
1174                }
1175                break;
1176        case NETDEV_UNREGISTER:
1177                devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
1178                device_destroy(macvtap_class, devt);
1179                macvtap_free_minor(vlan);
1180                break;
1181        }
1182
1183        return NOTIFY_DONE;
1184}
1185
1186static struct notifier_block macvtap_notifier_block __read_mostly = {
1187        .notifier_call  = macvtap_device_event,
1188};
1189
1190static int macvtap_init(void)
1191{
1192        int err;
1193
1194        err = alloc_chrdev_region(&macvtap_major, 0,
1195                                MACVTAP_NUM_DEVS, "macvtap");
1196        if (err)
1197                goto out1;
1198
1199        cdev_init(&macvtap_cdev, &macvtap_fops);
1200        err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS);
1201        if (err)
1202                goto out2;
1203
1204        macvtap_class = class_create(THIS_MODULE, "macvtap");
1205        if (IS_ERR(macvtap_class)) {
1206                err = PTR_ERR(macvtap_class);
1207                goto out3;
1208        }
1209
1210        err = register_netdevice_notifier(&macvtap_notifier_block);
1211        if (err)
1212                goto out4;
1213
1214        err = macvlan_link_register(&macvtap_link_ops);
1215        if (err)
1216                goto out5;
1217
1218        return 0;
1219
1220out5:
1221        unregister_netdevice_notifier(&macvtap_notifier_block);
1222out4:
1223        class_unregister(macvtap_class);
1224out3:
1225        cdev_del(&macvtap_cdev);
1226out2:
1227        unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1228out1:
1229        return err;
1230}
1231module_init(macvtap_init);
1232
1233static void macvtap_exit(void)
1234{
1235        rtnl_link_unregister(&macvtap_link_ops);
1236        unregister_netdevice_notifier(&macvtap_notifier_block);
1237        class_unregister(macvtap_class);
1238        cdev_del(&macvtap_cdev);
1239        unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1240}
1241module_exit(macvtap_exit);
1242
1243MODULE_ALIAS_RTNL_LINK("macvtap");
1244MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1245MODULE_LICENSE("GPL");
1246
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.