Linux Audio

Check our new training course

Loading...
v3.1
   1#include <linux/etherdevice.h>
   2#include <linux/if_macvlan.h>
 
   3#include <linux/interrupt.h>
   4#include <linux/nsproxy.h>
   5#include <linux/compat.h>
   6#include <linux/if_tun.h>
   7#include <linux/module.h>
   8#include <linux/skbuff.h>
   9#include <linux/cache.h>
  10#include <linux/sched.h>
  11#include <linux/types.h>
  12#include <linux/slab.h>
  13#include <linux/init.h>
  14#include <linux/wait.h>
  15#include <linux/cdev.h>
 
  16#include <linux/fs.h>
  17
  18#include <net/net_namespace.h>
  19#include <net/rtnetlink.h>
  20#include <net/sock.h>
  21#include <linux/virtio_net.h>
  22
  23/*
  24 * A macvtap queue is the central object of this driver, it connects
  25 * an open character device to a macvlan interface. There can be
  26 * multiple queues on one interface, which map back to queues
  27 * implemented in hardware on the underlying device.
  28 *
  29 * macvtap_proto is used to allocate queues through the sock allocation
  30 * mechanism.
  31 *
  32 * TODO: multiqueue support is currently not implemented, even though
  33 * macvtap is basically prepared for that. We will need to add this
  34 * here as well as in virtio-net and qemu to get line rate on 10gbit
  35 * adapters from a guest.
  36 */
  37struct macvtap_queue {
  38	struct sock sk;
  39	struct socket sock;
  40	struct socket_wq wq;
  41	int vnet_hdr_sz;
  42	struct macvlan_dev __rcu *vlan;
  43	struct file *file;
  44	unsigned int flags;
  45};
  46
  47static struct proto macvtap_proto = {
  48	.name = "macvtap",
  49	.owner = THIS_MODULE,
  50	.obj_size = sizeof (struct macvtap_queue),
  51};
  52
  53/*
  54 * Minor number matches netdev->ifindex, so need a potentially
  55 * large value. This also makes it possible to split the
  56 * tap functionality out again in the future by offering it
  57 * from other drivers besides macvtap. As long as every device
  58 * only has one tap, the interface numbers assure that the
  59 * device nodes are unique.
  60 */
  61static dev_t macvtap_major;
  62#define MACVTAP_NUM_DEVS 65536
 
 
 
  63#define GOODCOPY_LEN 128
  64static struct class *macvtap_class;
  65static struct cdev macvtap_cdev;
  66
  67static const struct proto_ops macvtap_socket_ops;
  68
  69/*
  70 * RCU usage:
  71 * The macvtap_queue and the macvlan_dev are loosely coupled, the
  72 * pointers from one to the other can only be read while rcu_read_lock
  73 * or macvtap_lock is held.
  74 *
  75 * Both the file and the macvlan_dev hold a reference on the macvtap_queue
  76 * through sock_hold(&q->sk). When the macvlan_dev goes away first,
  77 * q->vlan becomes inaccessible. When the files gets closed,
  78 * macvtap_get_queue() fails.
  79 *
  80 * There may still be references to the struct sock inside of the
  81 * queue from outbound SKBs, but these never reference back to the
  82 * file or the dev. The data structure is freed through __sk_free
  83 * when both our references and any pending SKBs are gone.
  84 */
  85static DEFINE_SPINLOCK(macvtap_lock);
  86
  87/*
  88 * get_slot: return a [unused/occupied] slot in vlan->taps[]:
  89 *	- if 'q' is NULL, return the first empty slot;
  90 *	- otherwise, return the slot this pointer occupies.
  91 */
  92static int get_slot(struct macvlan_dev *vlan, struct macvtap_queue *q)
  93{
  94	int i;
  95
  96	for (i = 0; i < MAX_MACVTAP_QUEUES; i++) {
  97		if (rcu_dereference(vlan->taps[i]) == q)
  98			return i;
  99	}
 100
 101	/* Should never happen */
 102	BUG_ON(1);
 103}
 104
 105static int macvtap_set_queue(struct net_device *dev, struct file *file,
 106				struct macvtap_queue *q)
 107{
 108	struct macvlan_dev *vlan = netdev_priv(dev);
 109	int index;
 110	int err = -EBUSY;
 111
 112	spin_lock(&macvtap_lock);
 113	if (vlan->numvtaps == MAX_MACVTAP_QUEUES)
 114		goto out;
 115
 116	err = 0;
 117	index = get_slot(vlan, NULL);
 118	rcu_assign_pointer(q->vlan, vlan);
 119	rcu_assign_pointer(vlan->taps[index], q);
 120	sock_hold(&q->sk);
 121
 122	q->file = file;
 123	file->private_data = q;
 124
 125	vlan->numvtaps++;
 126
 127out:
 128	spin_unlock(&macvtap_lock);
 129	return err;
 130}
 131
 132/*
 133 * The file owning the queue got closed, give up both
 134 * the reference that the files holds as well as the
 135 * one from the macvlan_dev if that still exists.
 136 *
 137 * Using the spinlock makes sure that we don't get
 138 * to the queue again after destroying it.
 139 */
 140static void macvtap_put_queue(struct macvtap_queue *q)
 141{
 142	struct macvlan_dev *vlan;
 143
 144	spin_lock(&macvtap_lock);
 145	vlan = rcu_dereference_protected(q->vlan,
 146					 lockdep_is_held(&macvtap_lock));
 147	if (vlan) {
 148		int index = get_slot(vlan, q);
 149
 150		rcu_assign_pointer(vlan->taps[index], NULL);
 151		rcu_assign_pointer(q->vlan, NULL);
 152		sock_put(&q->sk);
 153		--vlan->numvtaps;
 154	}
 155
 156	spin_unlock(&macvtap_lock);
 157
 158	synchronize_rcu();
 159	sock_put(&q->sk);
 160}
 161
 162/*
 163 * Select a queue based on the rxq of the device on which this packet
 164 * arrived. If the incoming device is not mq, calculate a flow hash
 165 * to select a queue. If all fails, find the first available queue.
 166 * Cache vlan->numvtaps since it can become zero during the execution
 167 * of this function.
 168 */
 169static struct macvtap_queue *macvtap_get_queue(struct net_device *dev,
 170					       struct sk_buff *skb)
 171{
 172	struct macvlan_dev *vlan = netdev_priv(dev);
 173	struct macvtap_queue *tap = NULL;
 174	int numvtaps = vlan->numvtaps;
 175	__u32 rxq;
 176
 177	if (!numvtaps)
 178		goto out;
 179
 
 
 
 
 
 
 
 
 180	if (likely(skb_rx_queue_recorded(skb))) {
 181		rxq = skb_get_rx_queue(skb);
 182
 183		while (unlikely(rxq >= numvtaps))
 184			rxq -= numvtaps;
 185
 186		tap = rcu_dereference(vlan->taps[rxq]);
 187		if (tap)
 188			goto out;
 189	}
 190
 191	/* Check if we can use flow to select a queue */
 192	rxq = skb_get_rxhash(skb);
 193	if (rxq) {
 194		tap = rcu_dereference(vlan->taps[rxq % numvtaps]);
 195		if (tap)
 196			goto out;
 197	}
 198
 199	/* Everything failed - find first available queue */
 200	for (rxq = 0; rxq < MAX_MACVTAP_QUEUES; rxq++) {
 201		tap = rcu_dereference(vlan->taps[rxq]);
 202		if (tap)
 203			break;
 204	}
 205
 206out:
 207	return tap;
 208}
 209
 210/*
 211 * The net_device is going away, give up the reference
 212 * that it holds on all queues and safely set the pointer
 213 * from the queues to NULL.
 214 */
 215static void macvtap_del_queues(struct net_device *dev)
 216{
 217	struct macvlan_dev *vlan = netdev_priv(dev);
 218	struct macvtap_queue *q, *qlist[MAX_MACVTAP_QUEUES];
 219	int i, j = 0;
 220
 221	/* macvtap_put_queue can free some slots, so go through all slots */
 222	spin_lock(&macvtap_lock);
 223	for (i = 0; i < MAX_MACVTAP_QUEUES && vlan->numvtaps; i++) {
 224		q = rcu_dereference_protected(vlan->taps[i],
 225					      lockdep_is_held(&macvtap_lock));
 226		if (q) {
 227			qlist[j++] = q;
 228			rcu_assign_pointer(vlan->taps[i], NULL);
 229			rcu_assign_pointer(q->vlan, NULL);
 230			vlan->numvtaps--;
 231		}
 232	}
 233	BUG_ON(vlan->numvtaps != 0);
 
 
 234	spin_unlock(&macvtap_lock);
 235
 236	synchronize_rcu();
 237
 238	for (--j; j >= 0; j--)
 239		sock_put(&qlist[j]->sk);
 240}
 241
 242/*
 243 * Forward happens for data that gets sent from one macvlan
 244 * endpoint to another one in bridge mode. We just take
 245 * the skb and put it into the receive queue.
 246 */
 247static int macvtap_forward(struct net_device *dev, struct sk_buff *skb)
 248{
 249	struct macvtap_queue *q = macvtap_get_queue(dev, skb);
 250	if (!q)
 251		goto drop;
 252
 253	if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len)
 254		goto drop;
 255
 256	skb_queue_tail(&q->sk.sk_receive_queue, skb);
 257	wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND);
 258	return NET_RX_SUCCESS;
 259
 260drop:
 261	kfree_skb(skb);
 262	return NET_RX_DROP;
 263}
 264
 265/*
 266 * Receive is for data from the external interface (lowerdev),
 267 * in case of macvtap, we can treat that the same way as
 268 * forward, which macvlan cannot.
 269 */
 270static int macvtap_receive(struct sk_buff *skb)
 271{
 272	skb_push(skb, ETH_HLEN);
 273	return macvtap_forward(skb->dev, skb);
 274}
 275
 276static int macvtap_newlink(struct net *src_net,
 277			   struct net_device *dev,
 278			   struct nlattr *tb[],
 279			   struct nlattr *data[])
 280{
 281	struct device *classdev;
 282	dev_t devt;
 283	int err;
 284
 285	err = macvlan_common_newlink(src_net, dev, tb, data,
 286				     macvtap_receive, macvtap_forward);
 287	if (err)
 288		goto out;
 289
 290	devt = MKDEV(MAJOR(macvtap_major), dev->ifindex);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 291
 292	classdev = device_create(macvtap_class, &dev->dev, devt,
 293				 dev, "tap%d", dev->ifindex);
 294	if (IS_ERR(classdev)) {
 295		err = PTR_ERR(classdev);
 296		macvtap_del_queues(dev);
 
 297	}
 
 
 298
 299out:
 300	return err;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 301}
 302
 303static void macvtap_dellink(struct net_device *dev,
 304			    struct list_head *head)
 305{
 306	device_destroy(macvtap_class,
 307		       MKDEV(MAJOR(macvtap_major), dev->ifindex));
 308
 309	macvtap_del_queues(dev);
 310	macvlan_dellink(dev, head);
 311}
 312
 313static void macvtap_setup(struct net_device *dev)
 314{
 315	macvlan_common_setup(dev);
 316	dev->tx_queue_len = TUN_READQ_SIZE;
 317}
 318
 319static struct rtnl_link_ops macvtap_link_ops __read_mostly = {
 320	.kind		= "macvtap",
 321	.setup		= macvtap_setup,
 322	.newlink	= macvtap_newlink,
 323	.dellink	= macvtap_dellink,
 324};
 325
 326
 327static void macvtap_sock_write_space(struct sock *sk)
 328{
 329	wait_queue_head_t *wqueue;
 330
 331	if (!sock_writeable(sk) ||
 332	    !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
 333		return;
 334
 335	wqueue = sk_sleep(sk);
 336	if (wqueue && waitqueue_active(wqueue))
 337		wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND);
 338}
 339
 
 
 
 
 
 340static int macvtap_open(struct inode *inode, struct file *file)
 341{
 342	struct net *net = current->nsproxy->net_ns;
 343	struct net_device *dev = dev_get_by_index(net, iminor(inode));
 344	struct macvlan_dev *vlan = netdev_priv(dev);
 345	struct macvtap_queue *q;
 346	int err;
 347
 348	err = -ENODEV;
 349	if (!dev)
 350		goto out;
 351
 352	/* check if this is a macvtap device */
 353	err = -EINVAL;
 354	if (dev->rtnl_link_ops != &macvtap_link_ops)
 355		goto out;
 356
 357	err = -ENOMEM;
 358	q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
 359					     &macvtap_proto);
 360	if (!q)
 361		goto out;
 362
 363	q->sock.wq = &q->wq;
 364	init_waitqueue_head(&q->wq.wait);
 365	q->sock.type = SOCK_RAW;
 366	q->sock.state = SS_CONNECTED;
 367	q->sock.file = file;
 368	q->sock.ops = &macvtap_socket_ops;
 369	sock_init_data(&q->sock, &q->sk);
 370	q->sk.sk_write_space = macvtap_sock_write_space;
 
 371	q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
 372	q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
 373
 374	/*
 375	 * so far only KVM virtio_net uses macvtap, enable zero copy between
 376	 * guest kernel and host kernel when lower device supports zerocopy
 
 
 
 377	 */
 378	if (vlan) {
 379		if ((vlan->lowerdev->features & NETIF_F_HIGHDMA) &&
 380		    (vlan->lowerdev->features & NETIF_F_SG))
 381			sock_set_flag(&q->sk, SOCK_ZEROCOPY);
 382	}
 383
 384	err = macvtap_set_queue(dev, file, q);
 385	if (err)
 386		sock_put(&q->sk);
 387
 388out:
 389	if (dev)
 390		dev_put(dev);
 391
 392	return err;
 393}
 394
 395static int macvtap_release(struct inode *inode, struct file *file)
 396{
 397	struct macvtap_queue *q = file->private_data;
 398	macvtap_put_queue(q);
 399	return 0;
 400}
 401
 402static unsigned int macvtap_poll(struct file *file, poll_table * wait)
 403{
 404	struct macvtap_queue *q = file->private_data;
 405	unsigned int mask = POLLERR;
 406
 407	if (!q)
 408		goto out;
 409
 410	mask = 0;
 411	poll_wait(file, &q->wq.wait, wait);
 412
 413	if (!skb_queue_empty(&q->sk.sk_receive_queue))
 414		mask |= POLLIN | POLLRDNORM;
 415
 416	if (sock_writeable(&q->sk) ||
 417	    (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) &&
 418	     sock_writeable(&q->sk)))
 419		mask |= POLLOUT | POLLWRNORM;
 420
 421out:
 422	return mask;
 423}
 424
 425static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad,
 426						size_t len, size_t linear,
 427						int noblock, int *err)
 428{
 429	struct sk_buff *skb;
 430
 431	/* Under a page?  Don't bother with paged skb. */
 432	if (prepad + len < PAGE_SIZE || !linear)
 433		linear = len;
 434
 435	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
 436				   err);
 437	if (!skb)
 438		return NULL;
 439
 440	skb_reserve(skb, prepad);
 441	skb_put(skb, linear);
 442	skb->data_len = len - linear;
 443	skb->len += len - linear;
 444
 445	return skb;
 446}
 447
 448/* set skb frags from iovec, this can move to core network code for reuse */
 449static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
 450				  int offset, size_t count)
 451{
 452	int len = iov_length(from, count) - offset;
 453	int copy = skb_headlen(skb);
 454	int size, offset1 = 0;
 455	int i = 0;
 456	skb_frag_t *f;
 457
 458	/* Skip over from offset */
 459	while (count && (offset >= from->iov_len)) {
 460		offset -= from->iov_len;
 461		++from;
 462		--count;
 463	}
 464
 465	/* copy up to skb headlen */
 466	while (count && (copy > 0)) {
 467		size = min_t(unsigned int, copy, from->iov_len - offset);
 468		if (copy_from_user(skb->data + offset1, from->iov_base + offset,
 469				   size))
 470			return -EFAULT;
 471		if (copy > size) {
 472			++from;
 473			--count;
 474		}
 
 
 475		copy -= size;
 476		offset1 += size;
 477		offset = 0;
 478	}
 479
 480	if (len == offset1)
 481		return 0;
 482
 483	while (count--) {
 484		struct page *page[MAX_SKB_FRAGS];
 485		int num_pages;
 486		unsigned long base;
 
 487
 488		len = from->iov_len - offset1;
 489		if (!len) {
 490			offset1 = 0;
 491			++from;
 492			continue;
 493		}
 494		base = (unsigned long)from->iov_base + offset1;
 495		size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
 
 
 496		num_pages = get_user_pages_fast(base, size, 0, &page[i]);
 497		if ((num_pages != size) ||
 498		    (num_pages > MAX_SKB_FRAGS - skb_shinfo(skb)->nr_frags))
 499			/* put_page is in skb free */
 500			return -EFAULT;
 
 
 501		skb->data_len += len;
 502		skb->len += len;
 503		skb->truesize += len;
 504		atomic_add(len, &skb->sk->sk_wmem_alloc);
 505		while (len) {
 506			f = &skb_shinfo(skb)->frags[i];
 507			f->page = page[i];
 508			f->page_offset = base & ~PAGE_MASK;
 509			f->size = min_t(int, len, PAGE_SIZE - f->page_offset);
 510			skb_shinfo(skb)->nr_frags++;
 511			/* increase sk_wmem_alloc */
 512			base += f->size;
 513			len -= f->size;
 514			i++;
 515		}
 516		offset1 = 0;
 517		++from;
 518	}
 519	return 0;
 520}
 521
 522/*
 523 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
 524 * be shared with the tun/tap driver.
 525 */
 526static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb,
 527				     struct virtio_net_hdr *vnet_hdr)
 528{
 529	unsigned short gso_type = 0;
 530	if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
 531		switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
 532		case VIRTIO_NET_HDR_GSO_TCPV4:
 533			gso_type = SKB_GSO_TCPV4;
 534			break;
 535		case VIRTIO_NET_HDR_GSO_TCPV6:
 536			gso_type = SKB_GSO_TCPV6;
 537			break;
 538		case VIRTIO_NET_HDR_GSO_UDP:
 539			gso_type = SKB_GSO_UDP;
 540			break;
 541		default:
 542			return -EINVAL;
 543		}
 544
 545		if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
 546			gso_type |= SKB_GSO_TCP_ECN;
 547
 548		if (vnet_hdr->gso_size == 0)
 549			return -EINVAL;
 550	}
 551
 552	if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
 553		if (!skb_partial_csum_set(skb, vnet_hdr->csum_start,
 554					  vnet_hdr->csum_offset))
 555			return -EINVAL;
 556	}
 557
 558	if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
 559		skb_shinfo(skb)->gso_size = vnet_hdr->gso_size;
 560		skb_shinfo(skb)->gso_type = gso_type;
 561
 562		/* Header must be checked, and gso_segs computed. */
 563		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
 564		skb_shinfo(skb)->gso_segs = 0;
 565	}
 566	return 0;
 567}
 568
 569static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb,
 570				   struct virtio_net_hdr *vnet_hdr)
 571{
 572	memset(vnet_hdr, 0, sizeof(*vnet_hdr));
 573
 574	if (skb_is_gso(skb)) {
 575		struct skb_shared_info *sinfo = skb_shinfo(skb);
 576
 577		/* This is a hint as to how much should be linear. */
 578		vnet_hdr->hdr_len = skb_headlen(skb);
 579		vnet_hdr->gso_size = sinfo->gso_size;
 580		if (sinfo->gso_type & SKB_GSO_TCPV4)
 581			vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
 582		else if (sinfo->gso_type & SKB_GSO_TCPV6)
 583			vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
 584		else if (sinfo->gso_type & SKB_GSO_UDP)
 585			vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
 586		else
 587			BUG();
 588		if (sinfo->gso_type & SKB_GSO_TCP_ECN)
 589			vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
 590	} else
 591		vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
 592
 593	if (skb->ip_summed == CHECKSUM_PARTIAL) {
 594		vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 595		vnet_hdr->csum_start = skb_checksum_start_offset(skb);
 596		vnet_hdr->csum_offset = skb->csum_offset;
 597	} else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
 598		vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID;
 599	} /* else everything is zero */
 600
 601	return 0;
 602}
 603
 604
 605/* Get packet from user space buffer */
 606static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m,
 607				const struct iovec *iv, unsigned long total_len,
 608				size_t count, int noblock)
 609{
 610	struct sk_buff *skb;
 611	struct macvlan_dev *vlan;
 612	unsigned long len = total_len;
 613	int err;
 614	struct virtio_net_hdr vnet_hdr = { 0 };
 615	int vnet_hdr_len = 0;
 616	int copylen;
 617	bool zerocopy = false;
 618
 619	if (q->flags & IFF_VNET_HDR) {
 620		vnet_hdr_len = q->vnet_hdr_sz;
 621
 622		err = -EINVAL;
 623		if (len < vnet_hdr_len)
 624			goto err;
 625		len -= vnet_hdr_len;
 626
 627		err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0,
 628					   sizeof(vnet_hdr));
 629		if (err < 0)
 630			goto err;
 631		if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
 632		     vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
 633							vnet_hdr.hdr_len)
 634			vnet_hdr.hdr_len = vnet_hdr.csum_start +
 635						vnet_hdr.csum_offset + 2;
 636		err = -EINVAL;
 637		if (vnet_hdr.hdr_len > len)
 638			goto err;
 639	}
 640
 641	err = -EINVAL;
 642	if (unlikely(len < ETH_HLEN))
 643		goto err;
 644
 
 
 
 
 645	if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY))
 646		zerocopy = true;
 647
 648	if (zerocopy) {
 
 
 
 
 
 
 
 
 
 
 
 649		/* There are 256 bytes to be copied in skb, so there is enough
 650		 * room for skb expand head in case it is used.
 651		 * The rest buffer is mapped from userspace.
 652		 */
 653		copylen = vnet_hdr.hdr_len;
 
 654		if (!copylen)
 655			copylen = GOODCOPY_LEN;
 656	} else
 657		copylen = len;
 658
 659	skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, copylen,
 660				vnet_hdr.hdr_len, noblock, &err);
 661	if (!skb)
 662		goto err;
 663
 664	if (zerocopy) {
 665		err = zerocopy_sg_from_iovec(skb, iv, vnet_hdr_len, count);
 666		skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
 667	} else
 668		err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len,
 669						   len);
 670	if (err)
 671		goto err_kfree;
 672
 673	skb_set_network_header(skb, ETH_HLEN);
 674	skb_reset_mac_header(skb);
 675	skb->protocol = eth_hdr(skb)->h_proto;
 676
 677	if (vnet_hdr_len) {
 678		err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr);
 679		if (err)
 680			goto err_kfree;
 681	}
 682
 683	rcu_read_lock_bh();
 684	vlan = rcu_dereference_bh(q->vlan);
 685	/* copy skb_ubuf_info for callback when skb has no error */
 686	if (zerocopy)
 687		skb_shinfo(skb)->destructor_arg = m->msg_control;
 
 
 688	if (vlan)
 689		macvlan_start_xmit(skb, vlan->dev);
 690	else
 691		kfree_skb(skb);
 692	rcu_read_unlock_bh();
 693
 694	return total_len;
 695
 696err_kfree:
 697	kfree_skb(skb);
 698
 699err:
 700	rcu_read_lock_bh();
 701	vlan = rcu_dereference_bh(q->vlan);
 702	if (vlan)
 703		vlan->dev->stats.tx_dropped++;
 704	rcu_read_unlock_bh();
 705
 706	return err;
 707}
 708
 709static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv,
 710				 unsigned long count, loff_t pos)
 711{
 712	struct file *file = iocb->ki_filp;
 713	ssize_t result = -ENOLINK;
 714	struct macvtap_queue *q = file->private_data;
 715
 716	result = macvtap_get_user(q, NULL, iv, iov_length(iv, count), count,
 717				  file->f_flags & O_NONBLOCK);
 718	return result;
 719}
 720
 721/* Put packet to the user space buffer */
 722static ssize_t macvtap_put_user(struct macvtap_queue *q,
 723				const struct sk_buff *skb,
 724				const struct iovec *iv, int len)
 725{
 726	struct macvlan_dev *vlan;
 727	int ret;
 728	int vnet_hdr_len = 0;
 
 
 729
 730	if (q->flags & IFF_VNET_HDR) {
 731		struct virtio_net_hdr vnet_hdr;
 732		vnet_hdr_len = q->vnet_hdr_sz;
 733		if ((len -= vnet_hdr_len) < 0)
 734			return -EINVAL;
 735
 736		ret = macvtap_skb_to_vnet_hdr(skb, &vnet_hdr);
 737		if (ret)
 738			return ret;
 739
 740		if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, sizeof(vnet_hdr)))
 741			return -EFAULT;
 742	}
 
 743
 744	len = min_t(int, skb->len, len);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 745
 746	ret = skb_copy_datagram_const_iovec(skb, 0, iv, vnet_hdr_len, len);
 
 747
 
 748	rcu_read_lock_bh();
 749	vlan = rcu_dereference_bh(q->vlan);
 750	if (vlan)
 751		macvlan_count_rx(vlan, len, ret == 0, 0);
 752	rcu_read_unlock_bh();
 753
 754	return ret ? ret : (len + vnet_hdr_len);
 755}
 756
 757static ssize_t macvtap_do_read(struct macvtap_queue *q, struct kiocb *iocb,
 758			       const struct iovec *iv, unsigned long len,
 759			       int noblock)
 760{
 761	DECLARE_WAITQUEUE(wait, current);
 762	struct sk_buff *skb;
 763	ssize_t ret = 0;
 764
 765	add_wait_queue(sk_sleep(&q->sk), &wait);
 766	while (len) {
 767		current->state = TASK_INTERRUPTIBLE;
 768
 769		/* Read frames from the queue */
 770		skb = skb_dequeue(&q->sk.sk_receive_queue);
 771		if (!skb) {
 772			if (noblock) {
 773				ret = -EAGAIN;
 774				break;
 775			}
 776			if (signal_pending(current)) {
 777				ret = -ERESTARTSYS;
 778				break;
 779			}
 780			/* Nothing to read, let's sleep */
 781			schedule();
 782			continue;
 783		}
 784		ret = macvtap_put_user(q, skb, iv, len);
 785		kfree_skb(skb);
 786		break;
 787	}
 788
 789	current->state = TASK_RUNNING;
 790	remove_wait_queue(sk_sleep(&q->sk), &wait);
 791	return ret;
 792}
 793
 794static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv,
 795				unsigned long count, loff_t pos)
 796{
 797	struct file *file = iocb->ki_filp;
 798	struct macvtap_queue *q = file->private_data;
 799	ssize_t len, ret = 0;
 800
 801	len = iov_length(iv, count);
 802	if (len < 0) {
 803		ret = -EINVAL;
 804		goto out;
 805	}
 806
 807	ret = macvtap_do_read(q, iocb, iv, len, file->f_flags & O_NONBLOCK);
 808	ret = min_t(ssize_t, ret, len); /* XXX copied from tun.c. Why? */
 809out:
 810	return ret;
 811}
 812
 813/*
 814 * provide compatibility with generic tun/tap interface
 815 */
 816static long macvtap_ioctl(struct file *file, unsigned int cmd,
 817			  unsigned long arg)
 818{
 819	struct macvtap_queue *q = file->private_data;
 820	struct macvlan_dev *vlan;
 821	void __user *argp = (void __user *)arg;
 822	struct ifreq __user *ifr = argp;
 823	unsigned int __user *up = argp;
 824	unsigned int u;
 825	int __user *sp = argp;
 826	int s;
 827	int ret;
 828
 829	switch (cmd) {
 830	case TUNSETIFF:
 831		/* ignore the name, just look at flags */
 832		if (get_user(u, &ifr->ifr_flags))
 833			return -EFAULT;
 834
 835		ret = 0;
 836		if ((u & ~IFF_VNET_HDR) != (IFF_NO_PI | IFF_TAP))
 837			ret = -EINVAL;
 838		else
 839			q->flags = u;
 840
 841		return ret;
 842
 843	case TUNGETIFF:
 844		rcu_read_lock_bh();
 845		vlan = rcu_dereference_bh(q->vlan);
 846		if (vlan)
 847			dev_hold(vlan->dev);
 848		rcu_read_unlock_bh();
 849
 850		if (!vlan)
 851			return -ENOLINK;
 852
 853		ret = 0;
 854		if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) ||
 855		    put_user(q->flags, &ifr->ifr_flags))
 856			ret = -EFAULT;
 857		dev_put(vlan->dev);
 858		return ret;
 859
 860	case TUNGETFEATURES:
 861		if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR, up))
 862			return -EFAULT;
 863		return 0;
 864
 865	case TUNSETSNDBUF:
 866		if (get_user(u, up))
 867			return -EFAULT;
 868
 869		q->sk.sk_sndbuf = u;
 870		return 0;
 871
 872	case TUNGETVNETHDRSZ:
 873		s = q->vnet_hdr_sz;
 874		if (put_user(s, sp))
 875			return -EFAULT;
 876		return 0;
 877
 878	case TUNSETVNETHDRSZ:
 879		if (get_user(s, sp))
 880			return -EFAULT;
 881		if (s < (int)sizeof(struct virtio_net_hdr))
 882			return -EINVAL;
 883
 884		q->vnet_hdr_sz = s;
 885		return 0;
 886
 887	case TUNSETOFFLOAD:
 888		/* let the user check for future flags */
 889		if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
 890			    TUN_F_TSO_ECN | TUN_F_UFO))
 891			return -EINVAL;
 892
 893		/* TODO: only accept frames with the features that
 894			 got enabled for forwarded frames */
 895		if (!(q->flags & IFF_VNET_HDR))
 896			return  -EINVAL;
 897		return 0;
 898
 899	default:
 900		return -EINVAL;
 901	}
 902}
 903
 904#ifdef CONFIG_COMPAT
 905static long macvtap_compat_ioctl(struct file *file, unsigned int cmd,
 906				 unsigned long arg)
 907{
 908	return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
 909}
 910#endif
 911
 912static const struct file_operations macvtap_fops = {
 913	.owner		= THIS_MODULE,
 914	.open		= macvtap_open,
 915	.release	= macvtap_release,
 916	.aio_read	= macvtap_aio_read,
 917	.aio_write	= macvtap_aio_write,
 918	.poll		= macvtap_poll,
 919	.llseek		= no_llseek,
 920	.unlocked_ioctl	= macvtap_ioctl,
 921#ifdef CONFIG_COMPAT
 922	.compat_ioctl	= macvtap_compat_ioctl,
 923#endif
 924};
 925
 926static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock,
 927			   struct msghdr *m, size_t total_len)
 928{
 929	struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
 930	return macvtap_get_user(q, m, m->msg_iov, total_len, m->msg_iovlen,
 931			    m->msg_flags & MSG_DONTWAIT);
 932}
 933
 934static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock,
 935			   struct msghdr *m, size_t total_len,
 936			   int flags)
 937{
 938	struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
 939	int ret;
 940	if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
 941		return -EINVAL;
 942	ret = macvtap_do_read(q, iocb, m->msg_iov, total_len,
 943			  flags & MSG_DONTWAIT);
 944	if (ret > total_len) {
 945		m->msg_flags |= MSG_TRUNC;
 946		ret = flags & MSG_TRUNC ? ret : total_len;
 947	}
 948	return ret;
 949}
 950
 951/* Ops structure to mimic raw sockets with tun */
 952static const struct proto_ops macvtap_socket_ops = {
 953	.sendmsg = macvtap_sendmsg,
 954	.recvmsg = macvtap_recvmsg,
 955};
 956
 957/* Get an underlying socket object from tun file.  Returns error unless file is
 958 * attached to a device.  The returned object works like a packet socket, it
 959 * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
 960 * holding a reference to the file for as long as the socket is in use. */
 961struct socket *macvtap_get_socket(struct file *file)
 962{
 963	struct macvtap_queue *q;
 964	if (file->f_op != &macvtap_fops)
 965		return ERR_PTR(-EINVAL);
 966	q = file->private_data;
 967	if (!q)
 968		return ERR_PTR(-EBADFD);
 969	return &q->sock;
 970}
 971EXPORT_SYMBOL_GPL(macvtap_get_socket);
 972
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 973static int macvtap_init(void)
 974{
 975	int err;
 976
 977	err = alloc_chrdev_region(&macvtap_major, 0,
 978				MACVTAP_NUM_DEVS, "macvtap");
 979	if (err)
 980		goto out1;
 981
 982	cdev_init(&macvtap_cdev, &macvtap_fops);
 983	err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS);
 984	if (err)
 985		goto out2;
 986
 987	macvtap_class = class_create(THIS_MODULE, "macvtap");
 988	if (IS_ERR(macvtap_class)) {
 989		err = PTR_ERR(macvtap_class);
 990		goto out3;
 991	}
 992
 993	err = macvlan_link_register(&macvtap_link_ops);
 994	if (err)
 995		goto out4;
 996
 
 
 
 
 997	return 0;
 998
 
 
 999out4:
1000	class_unregister(macvtap_class);
1001out3:
1002	cdev_del(&macvtap_cdev);
1003out2:
1004	unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1005out1:
1006	return err;
1007}
1008module_init(macvtap_init);
1009
1010static void macvtap_exit(void)
1011{
1012	rtnl_link_unregister(&macvtap_link_ops);
 
1013	class_unregister(macvtap_class);
1014	cdev_del(&macvtap_cdev);
1015	unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1016}
1017module_exit(macvtap_exit);
1018
1019MODULE_ALIAS_RTNL_LINK("macvtap");
1020MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1021MODULE_LICENSE("GPL");
v3.5.6
   1#include <linux/etherdevice.h>
   2#include <linux/if_macvlan.h>
   3#include <linux/if_vlan.h>
   4#include <linux/interrupt.h>
   5#include <linux/nsproxy.h>
   6#include <linux/compat.h>
   7#include <linux/if_tun.h>
   8#include <linux/module.h>
   9#include <linux/skbuff.h>
  10#include <linux/cache.h>
  11#include <linux/sched.h>
  12#include <linux/types.h>
  13#include <linux/slab.h>
  14#include <linux/init.h>
  15#include <linux/wait.h>
  16#include <linux/cdev.h>
  17#include <linux/idr.h>
  18#include <linux/fs.h>
  19
  20#include <net/net_namespace.h>
  21#include <net/rtnetlink.h>
  22#include <net/sock.h>
  23#include <linux/virtio_net.h>
  24
  25/*
  26 * A macvtap queue is the central object of this driver, it connects
  27 * an open character device to a macvlan interface. There can be
  28 * multiple queues on one interface, which map back to queues
  29 * implemented in hardware on the underlying device.
  30 *
  31 * macvtap_proto is used to allocate queues through the sock allocation
  32 * mechanism.
  33 *
  34 * TODO: multiqueue support is currently not implemented, even though
  35 * macvtap is basically prepared for that. We will need to add this
  36 * here as well as in virtio-net and qemu to get line rate on 10gbit
  37 * adapters from a guest.
  38 */
  39struct macvtap_queue {
  40	struct sock sk;
  41	struct socket sock;
  42	struct socket_wq wq;
  43	int vnet_hdr_sz;
  44	struct macvlan_dev __rcu *vlan;
  45	struct file *file;
  46	unsigned int flags;
  47};
  48
  49static struct proto macvtap_proto = {
  50	.name = "macvtap",
  51	.owner = THIS_MODULE,
  52	.obj_size = sizeof (struct macvtap_queue),
  53};
  54
  55/*
  56 * Variables for dealing with macvtaps device numbers.
 
 
 
 
 
  57 */
  58static dev_t macvtap_major;
  59#define MACVTAP_NUM_DEVS (1U << MINORBITS)
  60static DEFINE_MUTEX(minor_lock);
  61static DEFINE_IDR(minor_idr);
  62
  63#define GOODCOPY_LEN 128
  64static struct class *macvtap_class;
  65static struct cdev macvtap_cdev;
  66
  67static const struct proto_ops macvtap_socket_ops;
  68
  69/*
  70 * RCU usage:
  71 * The macvtap_queue and the macvlan_dev are loosely coupled, the
  72 * pointers from one to the other can only be read while rcu_read_lock
  73 * or macvtap_lock is held.
  74 *
  75 * Both the file and the macvlan_dev hold a reference on the macvtap_queue
  76 * through sock_hold(&q->sk). When the macvlan_dev goes away first,
  77 * q->vlan becomes inaccessible. When the files gets closed,
  78 * macvtap_get_queue() fails.
  79 *
  80 * There may still be references to the struct sock inside of the
  81 * queue from outbound SKBs, but these never reference back to the
  82 * file or the dev. The data structure is freed through __sk_free
  83 * when both our references and any pending SKBs are gone.
  84 */
  85static DEFINE_SPINLOCK(macvtap_lock);
  86
  87/*
  88 * get_slot: return a [unused/occupied] slot in vlan->taps[]:
  89 *	- if 'q' is NULL, return the first empty slot;
  90 *	- otherwise, return the slot this pointer occupies.
  91 */
  92static int get_slot(struct macvlan_dev *vlan, struct macvtap_queue *q)
  93{
  94	int i;
  95
  96	for (i = 0; i < MAX_MACVTAP_QUEUES; i++) {
  97		if (rcu_dereference(vlan->taps[i]) == q)
  98			return i;
  99	}
 100
 101	/* Should never happen */
 102	BUG_ON(1);
 103}
 104
 105static int macvtap_set_queue(struct net_device *dev, struct file *file,
 106				struct macvtap_queue *q)
 107{
 108	struct macvlan_dev *vlan = netdev_priv(dev);
 109	int index;
 110	int err = -EBUSY;
 111
 112	spin_lock(&macvtap_lock);
 113	if (vlan->numvtaps == MAX_MACVTAP_QUEUES)
 114		goto out;
 115
 116	err = 0;
 117	index = get_slot(vlan, NULL);
 118	rcu_assign_pointer(q->vlan, vlan);
 119	rcu_assign_pointer(vlan->taps[index], q);
 120	sock_hold(&q->sk);
 121
 122	q->file = file;
 123	file->private_data = q;
 124
 125	vlan->numvtaps++;
 126
 127out:
 128	spin_unlock(&macvtap_lock);
 129	return err;
 130}
 131
 132/*
 133 * The file owning the queue got closed, give up both
 134 * the reference that the files holds as well as the
 135 * one from the macvlan_dev if that still exists.
 136 *
 137 * Using the spinlock makes sure that we don't get
 138 * to the queue again after destroying it.
 139 */
 140static void macvtap_put_queue(struct macvtap_queue *q)
 141{
 142	struct macvlan_dev *vlan;
 143
 144	spin_lock(&macvtap_lock);
 145	vlan = rcu_dereference_protected(q->vlan,
 146					 lockdep_is_held(&macvtap_lock));
 147	if (vlan) {
 148		int index = get_slot(vlan, q);
 149
 150		RCU_INIT_POINTER(vlan->taps[index], NULL);
 151		RCU_INIT_POINTER(q->vlan, NULL);
 152		sock_put(&q->sk);
 153		--vlan->numvtaps;
 154	}
 155
 156	spin_unlock(&macvtap_lock);
 157
 158	synchronize_rcu();
 159	sock_put(&q->sk);
 160}
 161
 162/*
 163 * Select a queue based on the rxq of the device on which this packet
 164 * arrived. If the incoming device is not mq, calculate a flow hash
 165 * to select a queue. If all fails, find the first available queue.
 166 * Cache vlan->numvtaps since it can become zero during the execution
 167 * of this function.
 168 */
 169static struct macvtap_queue *macvtap_get_queue(struct net_device *dev,
 170					       struct sk_buff *skb)
 171{
 172	struct macvlan_dev *vlan = netdev_priv(dev);
 173	struct macvtap_queue *tap = NULL;
 174	int numvtaps = vlan->numvtaps;
 175	__u32 rxq;
 176
 177	if (!numvtaps)
 178		goto out;
 179
 180	/* Check if we can use flow to select a queue */
 181	rxq = skb_get_rxhash(skb);
 182	if (rxq) {
 183		tap = rcu_dereference(vlan->taps[rxq % numvtaps]);
 184		if (tap)
 185			goto out;
 186	}
 187
 188	if (likely(skb_rx_queue_recorded(skb))) {
 189		rxq = skb_get_rx_queue(skb);
 190
 191		while (unlikely(rxq >= numvtaps))
 192			rxq -= numvtaps;
 193
 194		tap = rcu_dereference(vlan->taps[rxq]);
 195		if (tap)
 196			goto out;
 197	}
 198
 
 
 
 
 
 
 
 
 199	/* Everything failed - find first available queue */
 200	for (rxq = 0; rxq < MAX_MACVTAP_QUEUES; rxq++) {
 201		tap = rcu_dereference(vlan->taps[rxq]);
 202		if (tap)
 203			break;
 204	}
 205
 206out:
 207	return tap;
 208}
 209
 210/*
 211 * The net_device is going away, give up the reference
 212 * that it holds on all queues and safely set the pointer
 213 * from the queues to NULL.
 214 */
 215static void macvtap_del_queues(struct net_device *dev)
 216{
 217	struct macvlan_dev *vlan = netdev_priv(dev);
 218	struct macvtap_queue *q, *qlist[MAX_MACVTAP_QUEUES];
 219	int i, j = 0;
 220
 221	/* macvtap_put_queue can free some slots, so go through all slots */
 222	spin_lock(&macvtap_lock);
 223	for (i = 0; i < MAX_MACVTAP_QUEUES && vlan->numvtaps; i++) {
 224		q = rcu_dereference_protected(vlan->taps[i],
 225					      lockdep_is_held(&macvtap_lock));
 226		if (q) {
 227			qlist[j++] = q;
 228			RCU_INIT_POINTER(vlan->taps[i], NULL);
 229			RCU_INIT_POINTER(q->vlan, NULL);
 230			vlan->numvtaps--;
 231		}
 232	}
 233	BUG_ON(vlan->numvtaps != 0);
 234	/* guarantee that any future macvtap_set_queue will fail */
 235	vlan->numvtaps = MAX_MACVTAP_QUEUES;
 236	spin_unlock(&macvtap_lock);
 237
 238	synchronize_rcu();
 239
 240	for (--j; j >= 0; j--)
 241		sock_put(&qlist[j]->sk);
 242}
 243
 244/*
 245 * Forward happens for data that gets sent from one macvlan
 246 * endpoint to another one in bridge mode. We just take
 247 * the skb and put it into the receive queue.
 248 */
 249static int macvtap_forward(struct net_device *dev, struct sk_buff *skb)
 250{
 251	struct macvtap_queue *q = macvtap_get_queue(dev, skb);
 252	if (!q)
 253		goto drop;
 254
 255	if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len)
 256		goto drop;
 257
 258	skb_queue_tail(&q->sk.sk_receive_queue, skb);
 259	wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND);
 260	return NET_RX_SUCCESS;
 261
 262drop:
 263	kfree_skb(skb);
 264	return NET_RX_DROP;
 265}
 266
 267/*
 268 * Receive is for data from the external interface (lowerdev),
 269 * in case of macvtap, we can treat that the same way as
 270 * forward, which macvlan cannot.
 271 */
 272static int macvtap_receive(struct sk_buff *skb)
 273{
 274	skb_push(skb, ETH_HLEN);
 275	return macvtap_forward(skb->dev, skb);
 276}
 277
 278static int macvtap_get_minor(struct macvlan_dev *vlan)
 
 
 
 279{
 280	int retval = -ENOMEM;
 281	int id;
 
 282
 283	mutex_lock(&minor_lock);
 284	if (idr_pre_get(&minor_idr, GFP_KERNEL) == 0)
 285		goto exit;
 
 286
 287	retval = idr_get_new_above(&minor_idr, vlan, 1, &id);
 288	if (retval < 0) {
 289		if (retval == -EAGAIN)
 290			retval = -ENOMEM;
 291		goto exit;
 292	}
 293	if (id < MACVTAP_NUM_DEVS) {
 294		vlan->minor = id;
 295	} else {
 296		printk(KERN_ERR "too many macvtap devices\n");
 297		retval = -EINVAL;
 298		idr_remove(&minor_idr, id);
 299	}
 300exit:
 301	mutex_unlock(&minor_lock);
 302	return retval;
 303}
 304
 305static void macvtap_free_minor(struct macvlan_dev *vlan)
 306{
 307	mutex_lock(&minor_lock);
 308	if (vlan->minor) {
 309		idr_remove(&minor_idr, vlan->minor);
 310		vlan->minor = 0;
 311	}
 312	mutex_unlock(&minor_lock);
 313}
 314
 315static struct net_device *dev_get_by_macvtap_minor(int minor)
 316{
 317	struct net_device *dev = NULL;
 318	struct macvlan_dev *vlan;
 319
 320	mutex_lock(&minor_lock);
 321	vlan = idr_find(&minor_idr, minor);
 322	if (vlan) {
 323		dev = vlan->dev;
 324		dev_hold(dev);
 325	}
 326	mutex_unlock(&minor_lock);
 327	return dev;
 328}
 329
 330static int macvtap_newlink(struct net *src_net,
 331			   struct net_device *dev,
 332			   struct nlattr *tb[],
 333			   struct nlattr *data[])
 334{
 335	/* Don't put anything that may fail after macvlan_common_newlink
 336	 * because we can't undo what it does.
 337	 */
 338	return macvlan_common_newlink(src_net, dev, tb, data,
 339				      macvtap_receive, macvtap_forward);
 340}
 341
 342static void macvtap_dellink(struct net_device *dev,
 343			    struct list_head *head)
 344{
 
 
 
 345	macvtap_del_queues(dev);
 346	macvlan_dellink(dev, head);
 347}
 348
 349static void macvtap_setup(struct net_device *dev)
 350{
 351	macvlan_common_setup(dev);
 352	dev->tx_queue_len = TUN_READQ_SIZE;
 353}
 354
 355static struct rtnl_link_ops macvtap_link_ops __read_mostly = {
 356	.kind		= "macvtap",
 357	.setup		= macvtap_setup,
 358	.newlink	= macvtap_newlink,
 359	.dellink	= macvtap_dellink,
 360};
 361
 362
 363static void macvtap_sock_write_space(struct sock *sk)
 364{
 365	wait_queue_head_t *wqueue;
 366
 367	if (!sock_writeable(sk) ||
 368	    !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
 369		return;
 370
 371	wqueue = sk_sleep(sk);
 372	if (wqueue && waitqueue_active(wqueue))
 373		wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND);
 374}
 375
 376static void macvtap_sock_destruct(struct sock *sk)
 377{
 378	skb_queue_purge(&sk->sk_receive_queue);
 379}
 380
 381static int macvtap_open(struct inode *inode, struct file *file)
 382{
 383	struct net *net = current->nsproxy->net_ns;
 384	struct net_device *dev = dev_get_by_macvtap_minor(iminor(inode));
 
 385	struct macvtap_queue *q;
 386	int err;
 387
 388	err = -ENODEV;
 389	if (!dev)
 390		goto out;
 391
 
 
 
 
 
 392	err = -ENOMEM;
 393	q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
 394					     &macvtap_proto);
 395	if (!q)
 396		goto out;
 397
 398	q->sock.wq = &q->wq;
 399	init_waitqueue_head(&q->wq.wait);
 400	q->sock.type = SOCK_RAW;
 401	q->sock.state = SS_CONNECTED;
 402	q->sock.file = file;
 403	q->sock.ops = &macvtap_socket_ops;
 404	sock_init_data(&q->sock, &q->sk);
 405	q->sk.sk_write_space = macvtap_sock_write_space;
 406	q->sk.sk_destruct = macvtap_sock_destruct;
 407	q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
 408	q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
 409
 410	/*
 411	 * so far only KVM virtio_net uses macvtap, enable zero copy between
 412	 * guest kernel and host kernel when lower device supports zerocopy
 413	 *
 414	 * The macvlan supports zerocopy iff the lower device supports zero
 415	 * copy so we don't have to look at the lower device directly.
 416	 */
 417	if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG))
 418		sock_set_flag(&q->sk, SOCK_ZEROCOPY);
 
 
 
 419
 420	err = macvtap_set_queue(dev, file, q);
 421	if (err)
 422		sock_put(&q->sk);
 423
 424out:
 425	if (dev)
 426		dev_put(dev);
 427
 428	return err;
 429}
 430
 431static int macvtap_release(struct inode *inode, struct file *file)
 432{
 433	struct macvtap_queue *q = file->private_data;
 434	macvtap_put_queue(q);
 435	return 0;
 436}
 437
 438static unsigned int macvtap_poll(struct file *file, poll_table * wait)
 439{
 440	struct macvtap_queue *q = file->private_data;
 441	unsigned int mask = POLLERR;
 442
 443	if (!q)
 444		goto out;
 445
 446	mask = 0;
 447	poll_wait(file, &q->wq.wait, wait);
 448
 449	if (!skb_queue_empty(&q->sk.sk_receive_queue))
 450		mask |= POLLIN | POLLRDNORM;
 451
 452	if (sock_writeable(&q->sk) ||
 453	    (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) &&
 454	     sock_writeable(&q->sk)))
 455		mask |= POLLOUT | POLLWRNORM;
 456
 457out:
 458	return mask;
 459}
 460
 461static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad,
 462						size_t len, size_t linear,
 463						int noblock, int *err)
 464{
 465	struct sk_buff *skb;
 466
 467	/* Under a page?  Don't bother with paged skb. */
 468	if (prepad + len < PAGE_SIZE || !linear)
 469		linear = len;
 470
 471	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
 472				   err);
 473	if (!skb)
 474		return NULL;
 475
 476	skb_reserve(skb, prepad);
 477	skb_put(skb, linear);
 478	skb->data_len = len - linear;
 479	skb->len += len - linear;
 480
 481	return skb;
 482}
 483
 484/* set skb frags from iovec, this can move to core network code for reuse */
 485static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
 486				  int offset, size_t count)
 487{
 488	int len = iov_length(from, count) - offset;
 489	int copy = skb_headlen(skb);
 490	int size, offset1 = 0;
 491	int i = 0;
 
 492
 493	/* Skip over from offset */
 494	while (count && (offset >= from->iov_len)) {
 495		offset -= from->iov_len;
 496		++from;
 497		--count;
 498	}
 499
 500	/* copy up to skb headlen */
 501	while (count && (copy > 0)) {
 502		size = min_t(unsigned int, copy, from->iov_len - offset);
 503		if (copy_from_user(skb->data + offset1, from->iov_base + offset,
 504				   size))
 505			return -EFAULT;
 506		if (copy > size) {
 507			++from;
 508			--count;
 509			offset = 0;
 510		} else
 511			offset += size;
 512		copy -= size;
 513		offset1 += size;
 
 514	}
 515
 516	if (len == offset1)
 517		return 0;
 518
 519	while (count--) {
 520		struct page *page[MAX_SKB_FRAGS];
 521		int num_pages;
 522		unsigned long base;
 523		unsigned long truesize;
 524
 525		len = from->iov_len - offset;
 526		if (!len) {
 527			offset = 0;
 528			++from;
 529			continue;
 530		}
 531		base = (unsigned long)from->iov_base + offset;
 532		size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
 533		if (i + size > MAX_SKB_FRAGS)
 534			return -EMSGSIZE;
 535		num_pages = get_user_pages_fast(base, size, 0, &page[i]);
 536		if (num_pages != size) {
 537			for (i = 0; i < num_pages; i++)
 538				put_page(page[i]);
 539			return -EFAULT;
 540		}
 541		truesize = size * PAGE_SIZE;
 542		skb->data_len += len;
 543		skb->len += len;
 544		skb->truesize += truesize;
 545		atomic_add(truesize, &skb->sk->sk_wmem_alloc);
 546		while (len) {
 547			int off = base & ~PAGE_MASK;
 548			int size = min_t(int, len, PAGE_SIZE - off);
 549			__skb_fill_page_desc(skb, i, page[i], off, size);
 
 550			skb_shinfo(skb)->nr_frags++;
 551			/* increase sk_wmem_alloc */
 552			base += size;
 553			len -= size;
 554			i++;
 555		}
 556		offset = 0;
 557		++from;
 558	}
 559	return 0;
 560}
 561
 562/*
 563 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
 564 * be shared with the tun/tap driver.
 565 */
 566static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb,
 567				     struct virtio_net_hdr *vnet_hdr)
 568{
 569	unsigned short gso_type = 0;
 570	if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
 571		switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
 572		case VIRTIO_NET_HDR_GSO_TCPV4:
 573			gso_type = SKB_GSO_TCPV4;
 574			break;
 575		case VIRTIO_NET_HDR_GSO_TCPV6:
 576			gso_type = SKB_GSO_TCPV6;
 577			break;
 578		case VIRTIO_NET_HDR_GSO_UDP:
 579			gso_type = SKB_GSO_UDP;
 580			break;
 581		default:
 582			return -EINVAL;
 583		}
 584
 585		if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
 586			gso_type |= SKB_GSO_TCP_ECN;
 587
 588		if (vnet_hdr->gso_size == 0)
 589			return -EINVAL;
 590	}
 591
 592	if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
 593		if (!skb_partial_csum_set(skb, vnet_hdr->csum_start,
 594					  vnet_hdr->csum_offset))
 595			return -EINVAL;
 596	}
 597
 598	if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
 599		skb_shinfo(skb)->gso_size = vnet_hdr->gso_size;
 600		skb_shinfo(skb)->gso_type = gso_type;
 601
 602		/* Header must be checked, and gso_segs computed. */
 603		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
 604		skb_shinfo(skb)->gso_segs = 0;
 605	}
 606	return 0;
 607}
 608
 609static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb,
 610				   struct virtio_net_hdr *vnet_hdr)
 611{
 612	memset(vnet_hdr, 0, sizeof(*vnet_hdr));
 613
 614	if (skb_is_gso(skb)) {
 615		struct skb_shared_info *sinfo = skb_shinfo(skb);
 616
 617		/* This is a hint as to how much should be linear. */
 618		vnet_hdr->hdr_len = skb_headlen(skb);
 619		vnet_hdr->gso_size = sinfo->gso_size;
 620		if (sinfo->gso_type & SKB_GSO_TCPV4)
 621			vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
 622		else if (sinfo->gso_type & SKB_GSO_TCPV6)
 623			vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
 624		else if (sinfo->gso_type & SKB_GSO_UDP)
 625			vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
 626		else
 627			BUG();
 628		if (sinfo->gso_type & SKB_GSO_TCP_ECN)
 629			vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
 630	} else
 631		vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
 632
 633	if (skb->ip_summed == CHECKSUM_PARTIAL) {
 634		vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 635		vnet_hdr->csum_start = skb_checksum_start_offset(skb);
 636		vnet_hdr->csum_offset = skb->csum_offset;
 637	} else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
 638		vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID;
 639	} /* else everything is zero */
 640
 641	return 0;
 642}
 643
 644
 645/* Get packet from user space buffer */
 646static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m,
 647				const struct iovec *iv, unsigned long total_len,
 648				size_t count, int noblock)
 649{
 650	struct sk_buff *skb;
 651	struct macvlan_dev *vlan;
 652	unsigned long len = total_len;
 653	int err;
 654	struct virtio_net_hdr vnet_hdr = { 0 };
 655	int vnet_hdr_len = 0;
 656	int copylen = 0;
 657	bool zerocopy = false;
 658
 659	if (q->flags & IFF_VNET_HDR) {
 660		vnet_hdr_len = q->vnet_hdr_sz;
 661
 662		err = -EINVAL;
 663		if (len < vnet_hdr_len)
 664			goto err;
 665		len -= vnet_hdr_len;
 666
 667		err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0,
 668					   sizeof(vnet_hdr));
 669		if (err < 0)
 670			goto err;
 671		if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
 672		     vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
 673							vnet_hdr.hdr_len)
 674			vnet_hdr.hdr_len = vnet_hdr.csum_start +
 675						vnet_hdr.csum_offset + 2;
 676		err = -EINVAL;
 677		if (vnet_hdr.hdr_len > len)
 678			goto err;
 679	}
 680
 681	err = -EINVAL;
 682	if (unlikely(len < ETH_HLEN))
 683		goto err;
 684
 685	err = -EMSGSIZE;
 686	if (unlikely(count > UIO_MAXIOV))
 687		goto err;
 688
 689	if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY))
 690		zerocopy = true;
 691
 692	if (zerocopy) {
 693		/* Userspace may produce vectors with count greater than
 694		 * MAX_SKB_FRAGS, so we need to linearize parts of the skb
 695		 * to let the rest of data to be fit in the frags.
 696		 */
 697		if (count > MAX_SKB_FRAGS) {
 698			copylen = iov_length(iv, count - MAX_SKB_FRAGS);
 699			if (copylen < vnet_hdr_len)
 700				copylen = 0;
 701			else
 702				copylen -= vnet_hdr_len;
 703		}
 704		/* There are 256 bytes to be copied in skb, so there is enough
 705		 * room for skb expand head in case it is used.
 706		 * The rest buffer is mapped from userspace.
 707		 */
 708		if (copylen < vnet_hdr.hdr_len)
 709			copylen = vnet_hdr.hdr_len;
 710		if (!copylen)
 711			copylen = GOODCOPY_LEN;
 712	} else
 713		copylen = len;
 714
 715	skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, copylen,
 716				vnet_hdr.hdr_len, noblock, &err);
 717	if (!skb)
 718		goto err;
 719
 720	if (zerocopy)
 721		err = zerocopy_sg_from_iovec(skb, iv, vnet_hdr_len, count);
 722	else
 
 723		err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len,
 724						   len);
 725	if (err)
 726		goto err_kfree;
 727
 728	skb_set_network_header(skb, ETH_HLEN);
 729	skb_reset_mac_header(skb);
 730	skb->protocol = eth_hdr(skb)->h_proto;
 731
 732	if (vnet_hdr_len) {
 733		err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr);
 734		if (err)
 735			goto err_kfree;
 736	}
 737
 738	rcu_read_lock_bh();
 739	vlan = rcu_dereference_bh(q->vlan);
 740	/* copy skb_ubuf_info for callback when skb has no error */
 741	if (zerocopy) {
 742		skb_shinfo(skb)->destructor_arg = m->msg_control;
 743		skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
 744	}
 745	if (vlan)
 746		macvlan_start_xmit(skb, vlan->dev);
 747	else
 748		kfree_skb(skb);
 749	rcu_read_unlock_bh();
 750
 751	return total_len;
 752
 753err_kfree:
 754	kfree_skb(skb);
 755
 756err:
 757	rcu_read_lock_bh();
 758	vlan = rcu_dereference_bh(q->vlan);
 759	if (vlan)
 760		vlan->dev->stats.tx_dropped++;
 761	rcu_read_unlock_bh();
 762
 763	return err;
 764}
 765
 766static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv,
 767				 unsigned long count, loff_t pos)
 768{
 769	struct file *file = iocb->ki_filp;
 770	ssize_t result = -ENOLINK;
 771	struct macvtap_queue *q = file->private_data;
 772
 773	result = macvtap_get_user(q, NULL, iv, iov_length(iv, count), count,
 774				  file->f_flags & O_NONBLOCK);
 775	return result;
 776}
 777
 778/* Put packet to the user space buffer */
 779static ssize_t macvtap_put_user(struct macvtap_queue *q,
 780				const struct sk_buff *skb,
 781				const struct iovec *iv, int len)
 782{
 783	struct macvlan_dev *vlan;
 784	int ret;
 785	int vnet_hdr_len = 0;
 786	int vlan_offset = 0;
 787	int copied;
 788
 789	if (q->flags & IFF_VNET_HDR) {
 790		struct virtio_net_hdr vnet_hdr;
 791		vnet_hdr_len = q->vnet_hdr_sz;
 792		if ((len -= vnet_hdr_len) < 0)
 793			return -EINVAL;
 794
 795		ret = macvtap_skb_to_vnet_hdr(skb, &vnet_hdr);
 796		if (ret)
 797			return ret;
 798
 799		if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, sizeof(vnet_hdr)))
 800			return -EFAULT;
 801	}
 802	copied = vnet_hdr_len;
 803
 804	if (!vlan_tx_tag_present(skb))
 805		len = min_t(int, skb->len, len);
 806	else {
 807		int copy;
 808		struct {
 809			__be16 h_vlan_proto;
 810			__be16 h_vlan_TCI;
 811		} veth;
 812		veth.h_vlan_proto = htons(ETH_P_8021Q);
 813		veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb));
 814
 815		vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
 816		len = min_t(int, skb->len + VLAN_HLEN, len);
 817
 818		copy = min_t(int, vlan_offset, len);
 819		ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy);
 820		len -= copy;
 821		copied += copy;
 822		if (ret || !len)
 823			goto done;
 824
 825		copy = min_t(int, sizeof(veth), len);
 826		ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy);
 827		len -= copy;
 828		copied += copy;
 829		if (ret || !len)
 830			goto done;
 831	}
 832
 833	ret = skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len);
 834	copied += len;
 835
 836done:
 837	rcu_read_lock_bh();
 838	vlan = rcu_dereference_bh(q->vlan);
 839	if (vlan)
 840		macvlan_count_rx(vlan, copied - vnet_hdr_len, ret == 0, 0);
 841	rcu_read_unlock_bh();
 842
 843	return ret ? ret : copied;
 844}
 845
 846static ssize_t macvtap_do_read(struct macvtap_queue *q, struct kiocb *iocb,
 847			       const struct iovec *iv, unsigned long len,
 848			       int noblock)
 849{
 850	DECLARE_WAITQUEUE(wait, current);
 851	struct sk_buff *skb;
 852	ssize_t ret = 0;
 853
 854	add_wait_queue(sk_sleep(&q->sk), &wait);
 855	while (len) {
 856		current->state = TASK_INTERRUPTIBLE;
 857
 858		/* Read frames from the queue */
 859		skb = skb_dequeue(&q->sk.sk_receive_queue);
 860		if (!skb) {
 861			if (noblock) {
 862				ret = -EAGAIN;
 863				break;
 864			}
 865			if (signal_pending(current)) {
 866				ret = -ERESTARTSYS;
 867				break;
 868			}
 869			/* Nothing to read, let's sleep */
 870			schedule();
 871			continue;
 872		}
 873		ret = macvtap_put_user(q, skb, iv, len);
 874		kfree_skb(skb);
 875		break;
 876	}
 877
 878	current->state = TASK_RUNNING;
 879	remove_wait_queue(sk_sleep(&q->sk), &wait);
 880	return ret;
 881}
 882
 883static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv,
 884				unsigned long count, loff_t pos)
 885{
 886	struct file *file = iocb->ki_filp;
 887	struct macvtap_queue *q = file->private_data;
 888	ssize_t len, ret = 0;
 889
 890	len = iov_length(iv, count);
 891	if (len < 0) {
 892		ret = -EINVAL;
 893		goto out;
 894	}
 895
 896	ret = macvtap_do_read(q, iocb, iv, len, file->f_flags & O_NONBLOCK);
 897	ret = min_t(ssize_t, ret, len); /* XXX copied from tun.c. Why? */
 898out:
 899	return ret;
 900}
 901
 902/*
 903 * provide compatibility with generic tun/tap interface
 904 */
 905static long macvtap_ioctl(struct file *file, unsigned int cmd,
 906			  unsigned long arg)
 907{
 908	struct macvtap_queue *q = file->private_data;
 909	struct macvlan_dev *vlan;
 910	void __user *argp = (void __user *)arg;
 911	struct ifreq __user *ifr = argp;
 912	unsigned int __user *up = argp;
 913	unsigned int u;
 914	int __user *sp = argp;
 915	int s;
 916	int ret;
 917
 918	switch (cmd) {
 919	case TUNSETIFF:
 920		/* ignore the name, just look at flags */
 921		if (get_user(u, &ifr->ifr_flags))
 922			return -EFAULT;
 923
 924		ret = 0;
 925		if ((u & ~IFF_VNET_HDR) != (IFF_NO_PI | IFF_TAP))
 926			ret = -EINVAL;
 927		else
 928			q->flags = u;
 929
 930		return ret;
 931
 932	case TUNGETIFF:
 933		rcu_read_lock_bh();
 934		vlan = rcu_dereference_bh(q->vlan);
 935		if (vlan)
 936			dev_hold(vlan->dev);
 937		rcu_read_unlock_bh();
 938
 939		if (!vlan)
 940			return -ENOLINK;
 941
 942		ret = 0;
 943		if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) ||
 944		    put_user(q->flags, &ifr->ifr_flags))
 945			ret = -EFAULT;
 946		dev_put(vlan->dev);
 947		return ret;
 948
 949	case TUNGETFEATURES:
 950		if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR, up))
 951			return -EFAULT;
 952		return 0;
 953
 954	case TUNSETSNDBUF:
 955		if (get_user(u, up))
 956			return -EFAULT;
 957
 958		q->sk.sk_sndbuf = u;
 959		return 0;
 960
 961	case TUNGETVNETHDRSZ:
 962		s = q->vnet_hdr_sz;
 963		if (put_user(s, sp))
 964			return -EFAULT;
 965		return 0;
 966
 967	case TUNSETVNETHDRSZ:
 968		if (get_user(s, sp))
 969			return -EFAULT;
 970		if (s < (int)sizeof(struct virtio_net_hdr))
 971			return -EINVAL;
 972
 973		q->vnet_hdr_sz = s;
 974		return 0;
 975
 976	case TUNSETOFFLOAD:
 977		/* let the user check for future flags */
 978		if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
 979			    TUN_F_TSO_ECN | TUN_F_UFO))
 980			return -EINVAL;
 981
 982		/* TODO: only accept frames with the features that
 983			 got enabled for forwarded frames */
 984		if (!(q->flags & IFF_VNET_HDR))
 985			return  -EINVAL;
 986		return 0;
 987
 988	default:
 989		return -EINVAL;
 990	}
 991}
 992
 993#ifdef CONFIG_COMPAT
 994static long macvtap_compat_ioctl(struct file *file, unsigned int cmd,
 995				 unsigned long arg)
 996{
 997	return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
 998}
 999#endif
1000
1001static const struct file_operations macvtap_fops = {
1002	.owner		= THIS_MODULE,
1003	.open		= macvtap_open,
1004	.release	= macvtap_release,
1005	.aio_read	= macvtap_aio_read,
1006	.aio_write	= macvtap_aio_write,
1007	.poll		= macvtap_poll,
1008	.llseek		= no_llseek,
1009	.unlocked_ioctl	= macvtap_ioctl,
1010#ifdef CONFIG_COMPAT
1011	.compat_ioctl	= macvtap_compat_ioctl,
1012#endif
1013};
1014
1015static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock,
1016			   struct msghdr *m, size_t total_len)
1017{
1018	struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1019	return macvtap_get_user(q, m, m->msg_iov, total_len, m->msg_iovlen,
1020			    m->msg_flags & MSG_DONTWAIT);
1021}
1022
1023static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock,
1024			   struct msghdr *m, size_t total_len,
1025			   int flags)
1026{
1027	struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1028	int ret;
1029	if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1030		return -EINVAL;
1031	ret = macvtap_do_read(q, iocb, m->msg_iov, total_len,
1032			  flags & MSG_DONTWAIT);
1033	if (ret > total_len) {
1034		m->msg_flags |= MSG_TRUNC;
1035		ret = flags & MSG_TRUNC ? ret : total_len;
1036	}
1037	return ret;
1038}
1039
1040/* Ops structure to mimic raw sockets with tun */
1041static const struct proto_ops macvtap_socket_ops = {
1042	.sendmsg = macvtap_sendmsg,
1043	.recvmsg = macvtap_recvmsg,
1044};
1045
1046/* Get an underlying socket object from tun file.  Returns error unless file is
1047 * attached to a device.  The returned object works like a packet socket, it
1048 * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
1049 * holding a reference to the file for as long as the socket is in use. */
1050struct socket *macvtap_get_socket(struct file *file)
1051{
1052	struct macvtap_queue *q;
1053	if (file->f_op != &macvtap_fops)
1054		return ERR_PTR(-EINVAL);
1055	q = file->private_data;
1056	if (!q)
1057		return ERR_PTR(-EBADFD);
1058	return &q->sock;
1059}
1060EXPORT_SYMBOL_GPL(macvtap_get_socket);
1061
1062static int macvtap_device_event(struct notifier_block *unused,
1063				unsigned long event, void *ptr)
1064{
1065	struct net_device *dev = ptr;
1066	struct macvlan_dev *vlan;
1067	struct device *classdev;
1068	dev_t devt;
1069	int err;
1070
1071	if (dev->rtnl_link_ops != &macvtap_link_ops)
1072		return NOTIFY_DONE;
1073
1074	vlan = netdev_priv(dev);
1075
1076	switch (event) {
1077	case NETDEV_REGISTER:
1078		/* Create the device node here after the network device has
1079		 * been registered but before register_netdevice has
1080		 * finished running.
1081		 */
1082		err = macvtap_get_minor(vlan);
1083		if (err)
1084			return notifier_from_errno(err);
1085
1086		devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
1087		classdev = device_create(macvtap_class, &dev->dev, devt,
1088					 dev, "tap%d", dev->ifindex);
1089		if (IS_ERR(classdev)) {
1090			macvtap_free_minor(vlan);
1091			return notifier_from_errno(PTR_ERR(classdev));
1092		}
1093		break;
1094	case NETDEV_UNREGISTER:
1095		devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
1096		device_destroy(macvtap_class, devt);
1097		macvtap_free_minor(vlan);
1098		break;
1099	}
1100
1101	return NOTIFY_DONE;
1102}
1103
1104static struct notifier_block macvtap_notifier_block __read_mostly = {
1105	.notifier_call	= macvtap_device_event,
1106};
1107
1108static int macvtap_init(void)
1109{
1110	int err;
1111
1112	err = alloc_chrdev_region(&macvtap_major, 0,
1113				MACVTAP_NUM_DEVS, "macvtap");
1114	if (err)
1115		goto out1;
1116
1117	cdev_init(&macvtap_cdev, &macvtap_fops);
1118	err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS);
1119	if (err)
1120		goto out2;
1121
1122	macvtap_class = class_create(THIS_MODULE, "macvtap");
1123	if (IS_ERR(macvtap_class)) {
1124		err = PTR_ERR(macvtap_class);
1125		goto out3;
1126	}
1127
1128	err = register_netdevice_notifier(&macvtap_notifier_block);
1129	if (err)
1130		goto out4;
1131
1132	err = macvlan_link_register(&macvtap_link_ops);
1133	if (err)
1134		goto out5;
1135
1136	return 0;
1137
1138out5:
1139	unregister_netdevice_notifier(&macvtap_notifier_block);
1140out4:
1141	class_unregister(macvtap_class);
1142out3:
1143	cdev_del(&macvtap_cdev);
1144out2:
1145	unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1146out1:
1147	return err;
1148}
1149module_init(macvtap_init);
1150
1151static void macvtap_exit(void)
1152{
1153	rtnl_link_unregister(&macvtap_link_ops);
1154	unregister_netdevice_notifier(&macvtap_notifier_block);
1155	class_unregister(macvtap_class);
1156	cdev_del(&macvtap_cdev);
1157	unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1158}
1159module_exit(macvtap_exit);
1160
1161MODULE_ALIAS_RTNL_LINK("macvtap");
1162MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1163MODULE_LICENSE("GPL");