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");
v3.15
   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/wait.h>
  15#include <linux/cdev.h>
  16#include <linux/idr.h>
  17#include <linux/fs.h>
  18
  19#include <net/net_namespace.h>
  20#include <net/rtnetlink.h>
  21#include <net/sock.h>
  22#include <linux/virtio_net.h>
  23
  24/*
  25 * A macvtap queue is the central object of this driver, it connects
  26 * an open character device to a macvlan interface. There can be
  27 * multiple queues on one interface, which map back to queues
  28 * implemented in hardware on the underlying device.
  29 *
  30 * macvtap_proto is used to allocate queues through the sock allocation
  31 * mechanism.
  32 *
 
 
 
 
  33 */
  34struct macvtap_queue {
  35	struct sock sk;
  36	struct socket sock;
  37	struct socket_wq wq;
  38	int vnet_hdr_sz;
  39	struct macvlan_dev __rcu *vlan;
  40	struct file *file;
  41	unsigned int flags;
  42	u16 queue_index;
  43	bool enabled;
  44	struct list_head next;
  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 * Variables for dealing with macvtaps device numbers.
 
 
 
 
 
  55 */
  56static dev_t macvtap_major;
  57#define MACVTAP_NUM_DEVS (1U << MINORBITS)
  58static DEFINE_MUTEX(minor_lock);
  59static DEFINE_IDR(minor_idr);
  60
  61#define GOODCOPY_LEN 128
  62static struct class *macvtap_class;
  63static struct cdev macvtap_cdev;
  64
  65static const struct proto_ops macvtap_socket_ops;
  66
  67#define TUN_OFFLOADS (NETIF_F_HW_CSUM | NETIF_F_TSO_ECN | NETIF_F_TSO | \
  68		      NETIF_F_TSO6 | NETIF_F_UFO)
  69#define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO)
  70#define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG)
  71
  72static struct macvlan_dev *macvtap_get_vlan_rcu(const struct net_device *dev)
  73{
  74	return rcu_dereference(dev->rx_handler_data);
  75}
  76
  77/*
  78 * RCU usage:
  79 * The macvtap_queue and the macvlan_dev are loosely coupled, the
  80 * pointers from one to the other can only be read while rcu_read_lock
  81 * or rtnl is held.
  82 *
  83 * Both the file and the macvlan_dev hold a reference on the macvtap_queue
  84 * through sock_hold(&q->sk). When the macvlan_dev goes away first,
  85 * q->vlan becomes inaccessible. When the files gets closed,
  86 * macvtap_get_queue() fails.
  87 *
  88 * There may still be references to the struct sock inside of the
  89 * queue from outbound SKBs, but these never reference back to the
  90 * file or the dev. The data structure is freed through __sk_free
  91 * when both our references and any pending SKBs are gone.
  92 */
 
  93
  94static int macvtap_enable_queue(struct net_device *dev, struct file *file,
  95				struct macvtap_queue *q)
 
 
 
 
  96{
  97	struct macvlan_dev *vlan = netdev_priv(dev);
  98	int err = -EINVAL;
  99
 100	ASSERT_RTNL();
 101
 102	if (q->enabled)
 103		goto out;
 104
 105	err = 0;
 106	rcu_assign_pointer(vlan->taps[vlan->numvtaps], q);
 107	q->queue_index = vlan->numvtaps;
 108	q->enabled = true;
 109
 110	vlan->numvtaps++;
 111out:
 112	return err;
 113}
 114
 115static int macvtap_set_queue(struct net_device *dev, struct file *file,
 116			     struct macvtap_queue *q)
 117{
 118	struct macvlan_dev *vlan = netdev_priv(dev);
 
 119	int err = -EBUSY;
 120
 121	rtnl_lock();
 122	if (vlan->numqueues == MAX_MACVTAP_QUEUES)
 123		goto out;
 124
 125	err = 0;
 
 126	rcu_assign_pointer(q->vlan, vlan);
 127	rcu_assign_pointer(vlan->taps[vlan->numvtaps], q);
 128	sock_hold(&q->sk);
 129
 130	q->file = file;
 131	q->queue_index = vlan->numvtaps;
 132	q->enabled = true;
 133	file->private_data = q;
 134	list_add_tail(&q->next, &vlan->queue_list);
 135
 136	vlan->numvtaps++;
 137	vlan->numqueues++;
 138
 139out:
 140	rtnl_unlock();
 141	return err;
 142}
 143
 144static int macvtap_disable_queue(struct macvtap_queue *q)
 145{
 146	struct macvlan_dev *vlan;
 147	struct macvtap_queue *nq;
 148
 149	ASSERT_RTNL();
 150	if (!q->enabled)
 151		return -EINVAL;
 152
 153	vlan = rtnl_dereference(q->vlan);
 154
 155	if (vlan) {
 156		int index = q->queue_index;
 157		BUG_ON(index >= vlan->numvtaps);
 158		nq = rtnl_dereference(vlan->taps[vlan->numvtaps - 1]);
 159		nq->queue_index = index;
 160
 161		rcu_assign_pointer(vlan->taps[index], nq);
 162		RCU_INIT_POINTER(vlan->taps[vlan->numvtaps - 1], NULL);
 163		q->enabled = false;
 164
 165		vlan->numvtaps--;
 166	}
 167
 168	return 0;
 169}
 170
 171/*
 172 * The file owning the queue got closed, give up both
 173 * the reference that the files holds as well as the
 174 * one from the macvlan_dev if that still exists.
 175 *
 176 * Using the spinlock makes sure that we don't get
 177 * to the queue again after destroying it.
 178 */
 179static void macvtap_put_queue(struct macvtap_queue *q)
 180{
 181	struct macvlan_dev *vlan;
 182
 183	rtnl_lock();
 184	vlan = rtnl_dereference(q->vlan);
 185
 186	if (vlan) {
 187		if (q->enabled)
 188			BUG_ON(macvtap_disable_queue(q));
 189
 190		vlan->numqueues--;
 191		RCU_INIT_POINTER(q->vlan, NULL);
 192		sock_put(&q->sk);
 193		list_del_init(&q->next);
 194	}
 195
 196	rtnl_unlock();
 197
 198	synchronize_rcu();
 199	sock_put(&q->sk);
 200}
 201
 202/*
 203 * Select a queue based on the rxq of the device on which this packet
 204 * arrived. If the incoming device is not mq, calculate a flow hash
 205 * to select a queue. If all fails, find the first available queue.
 206 * Cache vlan->numvtaps since it can become zero during the execution
 207 * of this function.
 208 */
 209static struct macvtap_queue *macvtap_get_queue(struct net_device *dev,
 210					       struct sk_buff *skb)
 211{
 212	struct macvlan_dev *vlan = netdev_priv(dev);
 213	struct macvtap_queue *tap = NULL;
 214	/* Access to taps array is protected by rcu, but access to numvtaps
 215	 * isn't. Below we use it to lookup a queue, but treat it as a hint
 216	 * and validate that the result isn't NULL - in case we are
 217	 * racing against queue removal.
 218	 */
 219	int numvtaps = ACCESS_ONCE(vlan->numvtaps);
 220	__u32 rxq;
 221
 222	if (!numvtaps)
 223		goto out;
 224
 225	/* Check if we can use flow to select a queue */
 226	rxq = skb_get_hash(skb);
 227	if (rxq) {
 228		tap = rcu_dereference(vlan->taps[rxq % numvtaps]);
 229		goto out;
 230	}
 231
 232	if (likely(skb_rx_queue_recorded(skb))) {
 233		rxq = skb_get_rx_queue(skb);
 234
 235		while (unlikely(rxq >= numvtaps))
 236			rxq -= numvtaps;
 237
 238		tap = rcu_dereference(vlan->taps[rxq]);
 239		goto out;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 240	}
 241
 242	tap = rcu_dereference(vlan->taps[0]);
 243out:
 244	return tap;
 245}
 246
 247/*
 248 * The net_device is going away, give up the reference
 249 * that it holds on all queues and safely set the pointer
 250 * from the queues to NULL.
 251 */
 252static void macvtap_del_queues(struct net_device *dev)
 253{
 254	struct macvlan_dev *vlan = netdev_priv(dev);
 255	struct macvtap_queue *q, *tmp, *qlist[MAX_MACVTAP_QUEUES];
 256	int i, j = 0;
 257
 258	ASSERT_RTNL();
 259	list_for_each_entry_safe(q, tmp, &vlan->queue_list, next) {
 260		list_del_init(&q->next);
 261		qlist[j++] = q;
 262		RCU_INIT_POINTER(q->vlan, NULL);
 263		if (q->enabled)
 
 
 
 264			vlan->numvtaps--;
 265		vlan->numqueues--;
 266	}
 267	for (i = 0; i < vlan->numvtaps; i++)
 268		RCU_INIT_POINTER(vlan->taps[i], NULL);
 269	BUG_ON(vlan->numvtaps);
 270	BUG_ON(vlan->numqueues);
 271	/* guarantee that any future macvtap_set_queue will fail */
 272	vlan->numvtaps = MAX_MACVTAP_QUEUES;
 273
 274	for (--j; j >= 0; j--)
 275		sock_put(&qlist[j]->sk);
 276}
 277
 278static rx_handler_result_t macvtap_handle_frame(struct sk_buff **pskb)
 
 
 
 
 
 279{
 280	struct sk_buff *skb = *pskb;
 281	struct net_device *dev = skb->dev;
 282	struct macvlan_dev *vlan;
 283	struct macvtap_queue *q;
 284	netdev_features_t features = TAP_FEATURES;
 285
 286	vlan = macvtap_get_vlan_rcu(dev);
 287	if (!vlan)
 288		return RX_HANDLER_PASS;
 289
 290	q = macvtap_get_queue(dev, skb);
 291	if (!q)
 292		return RX_HANDLER_PASS;
 293
 294	if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len)
 295		goto drop;
 296
 297	skb_push(skb, ETH_HLEN);
 298
 299	/* Apply the forward feature mask so that we perform segmentation
 300	 * according to users wishes.  This only works if VNET_HDR is
 301	 * enabled.
 302	 */
 303	if (q->flags & IFF_VNET_HDR)
 304		features |= vlan->tap_features;
 305	if (netif_needs_gso(skb, features)) {
 306		struct sk_buff *segs = __skb_gso_segment(skb, features, false);
 307
 308		if (IS_ERR(segs))
 309			goto drop;
 310
 311		if (!segs) {
 312			skb_queue_tail(&q->sk.sk_receive_queue, skb);
 313			goto wake_up;
 314		}
 315
 316		kfree_skb(skb);
 317		while (segs) {
 318			struct sk_buff *nskb = segs->next;
 319
 320			segs->next = NULL;
 321			skb_queue_tail(&q->sk.sk_receive_queue, segs);
 322			segs = nskb;
 323		}
 324	} else {
 325		/* If we receive a partial checksum and the tap side
 326		 * doesn't support checksum offload, compute the checksum.
 327		 * Note: it doesn't matter which checksum feature to
 328		 *        check, we either support them all or none.
 329		 */
 330		if (skb->ip_summed == CHECKSUM_PARTIAL &&
 331		    !(features & NETIF_F_ALL_CSUM) &&
 332		    skb_checksum_help(skb))
 333			goto drop;
 334		skb_queue_tail(&q->sk.sk_receive_queue, skb);
 335	}
 336
 337wake_up:
 338	wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND);
 339	return RX_HANDLER_CONSUMED;
 340
 341drop:
 342	/* Count errors/drops only here, thus don't care about args. */
 343	macvlan_count_rx(vlan, 0, 0, 0);
 344	kfree_skb(skb);
 345	return RX_HANDLER_CONSUMED;
 346}
 347
 348static int macvtap_get_minor(struct macvlan_dev *vlan)
 
 
 
 
 
 349{
 350	int retval = -ENOMEM;
 351
 352	mutex_lock(&minor_lock);
 353	retval = idr_alloc(&minor_idr, vlan, 1, MACVTAP_NUM_DEVS, GFP_KERNEL);
 354	if (retval >= 0) {
 355		vlan->minor = retval;
 356	} else if (retval == -ENOSPC) {
 357		printk(KERN_ERR "too many macvtap devices\n");
 358		retval = -EINVAL;
 359	}
 360	mutex_unlock(&minor_lock);
 361	return retval < 0 ? retval : 0;
 362}
 363
 364static void macvtap_free_minor(struct macvlan_dev *vlan)
 365{
 366	mutex_lock(&minor_lock);
 367	if (vlan->minor) {
 368		idr_remove(&minor_idr, vlan->minor);
 369		vlan->minor = 0;
 370	}
 371	mutex_unlock(&minor_lock);
 372}
 373
 374static struct net_device *dev_get_by_macvtap_minor(int minor)
 375{
 376	struct net_device *dev = NULL;
 377	struct macvlan_dev *vlan;
 378
 379	mutex_lock(&minor_lock);
 380	vlan = idr_find(&minor_idr, minor);
 381	if (vlan) {
 382		dev = vlan->dev;
 383		dev_hold(dev);
 384	}
 385	mutex_unlock(&minor_lock);
 386	return dev;
 387}
 388
 389static int macvtap_newlink(struct net *src_net,
 390			   struct net_device *dev,
 391			   struct nlattr *tb[],
 392			   struct nlattr *data[])
 393{
 394	struct macvlan_dev *vlan = netdev_priv(dev);
 
 395	int err;
 396
 397	INIT_LIST_HEAD(&vlan->queue_list);
 
 
 
 398
 399	/* Since macvlan supports all offloads by default, make
 400	 * tap support all offloads also.
 401	 */
 402	vlan->tap_features = TUN_OFFLOADS;
 403
 404	err = netdev_rx_handler_register(dev, macvtap_handle_frame, vlan);
 405	if (err)
 406		return err;
 
 
 
 407
 408	/* Don't put anything that may fail after macvlan_common_newlink
 409	 * because we can't undo what it does.
 410	 */
 411	return macvlan_common_newlink(src_net, dev, tb, data);
 412}
 413
 414static void macvtap_dellink(struct net_device *dev,
 415			    struct list_head *head)
 416{
 417	netdev_rx_handler_unregister(dev);
 
 
 418	macvtap_del_queues(dev);
 419	macvlan_dellink(dev, head);
 420}
 421
 422static void macvtap_setup(struct net_device *dev)
 423{
 424	macvlan_common_setup(dev);
 425	dev->tx_queue_len = TUN_READQ_SIZE;
 426}
 427
 428static struct rtnl_link_ops macvtap_link_ops __read_mostly = {
 429	.kind		= "macvtap",
 430	.setup		= macvtap_setup,
 431	.newlink	= macvtap_newlink,
 432	.dellink	= macvtap_dellink,
 433};
 434
 435
 436static void macvtap_sock_write_space(struct sock *sk)
 437{
 438	wait_queue_head_t *wqueue;
 439
 440	if (!sock_writeable(sk) ||
 441	    !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
 442		return;
 443
 444	wqueue = sk_sleep(sk);
 445	if (wqueue && waitqueue_active(wqueue))
 446		wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND);
 447}
 448
 449static void macvtap_sock_destruct(struct sock *sk)
 450{
 451	skb_queue_purge(&sk->sk_receive_queue);
 452}
 453
 454static int macvtap_open(struct inode *inode, struct file *file)
 455{
 456	struct net *net = current->nsproxy->net_ns;
 457	struct net_device *dev = dev_get_by_macvtap_minor(iminor(inode));
 
 458	struct macvtap_queue *q;
 459	int err;
 460
 461	err = -ENODEV;
 462	if (!dev)
 463		goto out;
 464
 
 
 
 
 
 465	err = -ENOMEM;
 466	q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
 467					     &macvtap_proto);
 468	if (!q)
 469		goto out;
 470
 471	RCU_INIT_POINTER(q->sock.wq, &q->wq);
 472	init_waitqueue_head(&q->wq.wait);
 473	q->sock.type = SOCK_RAW;
 474	q->sock.state = SS_CONNECTED;
 475	q->sock.file = file;
 476	q->sock.ops = &macvtap_socket_ops;
 477	sock_init_data(&q->sock, &q->sk);
 478	q->sk.sk_write_space = macvtap_sock_write_space;
 479	q->sk.sk_destruct = macvtap_sock_destruct;
 480	q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
 481	q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
 482
 483	/*
 484	 * so far only KVM virtio_net uses macvtap, enable zero copy between
 485	 * guest kernel and host kernel when lower device supports zerocopy
 486	 *
 487	 * The macvlan supports zerocopy iff the lower device supports zero
 488	 * copy so we don't have to look at the lower device directly.
 489	 */
 490	if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG))
 491		sock_set_flag(&q->sk, SOCK_ZEROCOPY);
 
 
 
 492
 493	err = macvtap_set_queue(dev, file, q);
 494	if (err)
 495		sock_put(&q->sk);
 496
 497out:
 498	if (dev)
 499		dev_put(dev);
 500
 501	return err;
 502}
 503
 504static int macvtap_release(struct inode *inode, struct file *file)
 505{
 506	struct macvtap_queue *q = file->private_data;
 507	macvtap_put_queue(q);
 508	return 0;
 509}
 510
 511static unsigned int macvtap_poll(struct file *file, poll_table * wait)
 512{
 513	struct macvtap_queue *q = file->private_data;
 514	unsigned int mask = POLLERR;
 515
 516	if (!q)
 517		goto out;
 518
 519	mask = 0;
 520	poll_wait(file, &q->wq.wait, wait);
 521
 522	if (!skb_queue_empty(&q->sk.sk_receive_queue))
 523		mask |= POLLIN | POLLRDNORM;
 524
 525	if (sock_writeable(&q->sk) ||
 526	    (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) &&
 527	     sock_writeable(&q->sk)))
 528		mask |= POLLOUT | POLLWRNORM;
 529
 530out:
 531	return mask;
 532}
 533
 534static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad,
 535						size_t len, size_t linear,
 536						int noblock, int *err)
 537{
 538	struct sk_buff *skb;
 539
 540	/* Under a page?  Don't bother with paged skb. */
 541	if (prepad + len < PAGE_SIZE || !linear)
 542		linear = len;
 543
 544	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
 545				   err, 0);
 546	if (!skb)
 547		return NULL;
 548
 549	skb_reserve(skb, prepad);
 550	skb_put(skb, linear);
 551	skb->data_len = len - linear;
 552	skb->len += len - linear;
 553
 554	return skb;
 555}
 556
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 557/*
 558 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
 559 * be shared with the tun/tap driver.
 560 */
 561static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb,
 562				     struct virtio_net_hdr *vnet_hdr)
 563{
 564	unsigned short gso_type = 0;
 565	if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
 566		switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
 567		case VIRTIO_NET_HDR_GSO_TCPV4:
 568			gso_type = SKB_GSO_TCPV4;
 569			break;
 570		case VIRTIO_NET_HDR_GSO_TCPV6:
 571			gso_type = SKB_GSO_TCPV6;
 572			break;
 573		case VIRTIO_NET_HDR_GSO_UDP:
 574			gso_type = SKB_GSO_UDP;
 575			break;
 576		default:
 577			return -EINVAL;
 578		}
 579
 580		if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
 581			gso_type |= SKB_GSO_TCP_ECN;
 582
 583		if (vnet_hdr->gso_size == 0)
 584			return -EINVAL;
 585	}
 586
 587	if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
 588		if (!skb_partial_csum_set(skb, vnet_hdr->csum_start,
 589					  vnet_hdr->csum_offset))
 590			return -EINVAL;
 591	}
 592
 593	if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
 594		skb_shinfo(skb)->gso_size = vnet_hdr->gso_size;
 595		skb_shinfo(skb)->gso_type = gso_type;
 596
 597		/* Header must be checked, and gso_segs computed. */
 598		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
 599		skb_shinfo(skb)->gso_segs = 0;
 600	}
 601	return 0;
 602}
 603
 604static void macvtap_skb_to_vnet_hdr(const struct sk_buff *skb,
 605				   struct virtio_net_hdr *vnet_hdr)
 606{
 607	memset(vnet_hdr, 0, sizeof(*vnet_hdr));
 608
 609	if (skb_is_gso(skb)) {
 610		struct skb_shared_info *sinfo = skb_shinfo(skb);
 611
 612		/* This is a hint as to how much should be linear. */
 613		vnet_hdr->hdr_len = skb_headlen(skb);
 614		vnet_hdr->gso_size = sinfo->gso_size;
 615		if (sinfo->gso_type & SKB_GSO_TCPV4)
 616			vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
 617		else if (sinfo->gso_type & SKB_GSO_TCPV6)
 618			vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
 619		else if (sinfo->gso_type & SKB_GSO_UDP)
 620			vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
 621		else
 622			BUG();
 623		if (sinfo->gso_type & SKB_GSO_TCP_ECN)
 624			vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
 625	} else
 626		vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
 627
 628	if (skb->ip_summed == CHECKSUM_PARTIAL) {
 629		vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 630		vnet_hdr->csum_start = skb_checksum_start_offset(skb);
 631		vnet_hdr->csum_offset = skb->csum_offset;
 632	} else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
 633		vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID;
 634	} /* else everything is zero */
 
 
 635}
 636
 
 637/* Get packet from user space buffer */
 638static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m,
 639				const struct iovec *iv, unsigned long total_len,
 640				size_t count, int noblock)
 641{
 642	int good_linear = SKB_MAX_HEAD(NET_IP_ALIGN);
 643	struct sk_buff *skb;
 644	struct macvlan_dev *vlan;
 645	unsigned long len = total_len;
 646	int err;
 647	struct virtio_net_hdr vnet_hdr = { 0 };
 648	int vnet_hdr_len = 0;
 649	int copylen = 0;
 650	bool zerocopy = false;
 651	size_t linear;
 652
 653	if (q->flags & IFF_VNET_HDR) {
 654		vnet_hdr_len = q->vnet_hdr_sz;
 655
 656		err = -EINVAL;
 657		if (len < vnet_hdr_len)
 658			goto err;
 659		len -= vnet_hdr_len;
 660
 661		err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0,
 662					   sizeof(vnet_hdr));
 663		if (err < 0)
 664			goto err;
 665		if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
 666		     vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
 667							vnet_hdr.hdr_len)
 668			vnet_hdr.hdr_len = vnet_hdr.csum_start +
 669						vnet_hdr.csum_offset + 2;
 670		err = -EINVAL;
 671		if (vnet_hdr.hdr_len > len)
 672			goto err;
 673	}
 674
 675	err = -EINVAL;
 676	if (unlikely(len < ETH_HLEN))
 677		goto err;
 678
 679	err = -EMSGSIZE;
 680	if (unlikely(count > UIO_MAXIOV))
 681		goto err;
 682
 683	if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) {
 684		copylen = vnet_hdr.hdr_len ? vnet_hdr.hdr_len : GOODCOPY_LEN;
 685		if (copylen > good_linear)
 686			copylen = good_linear;
 687		linear = copylen;
 688		if (iov_pages(iv, vnet_hdr_len + copylen, count)
 689		    <= MAX_SKB_FRAGS)
 690			zerocopy = true;
 691	}
 692
 693	if (!zerocopy) {
 
 
 
 
 
 
 
 
 694		copylen = len;
 695		if (vnet_hdr.hdr_len > good_linear)
 696			linear = good_linear;
 697		else
 698			linear = vnet_hdr.hdr_len;
 699	}
 700
 701	skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, copylen,
 702				linear, noblock, &err);
 703	if (!skb)
 704		goto err;
 705
 706	if (zerocopy)
 707		err = zerocopy_sg_from_iovec(skb, iv, vnet_hdr_len, count);
 708	else {
 
 709		err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len,
 710						   len);
 711		if (!err && m && m->msg_control) {
 712			struct ubuf_info *uarg = m->msg_control;
 713			uarg->callback(uarg, false);
 714		}
 715	}
 716
 717	if (err)
 718		goto err_kfree;
 719
 720	skb_set_network_header(skb, ETH_HLEN);
 721	skb_reset_mac_header(skb);
 722	skb->protocol = eth_hdr(skb)->h_proto;
 723
 724	if (vnet_hdr_len) {
 725		err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr);
 726		if (err)
 727			goto err_kfree;
 728	}
 729
 730	skb_probe_transport_header(skb, ETH_HLEN);
 731
 732	rcu_read_lock();
 733	vlan = rcu_dereference(q->vlan);
 734	/* copy skb_ubuf_info for callback when skb has no error */
 735	if (zerocopy) {
 736		skb_shinfo(skb)->destructor_arg = m->msg_control;
 737		skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
 738		skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
 739	}
 740	if (vlan) {
 741		skb->dev = vlan->dev;
 742		dev_queue_xmit(skb);
 743	} else {
 744		kfree_skb(skb);
 745	}
 746	rcu_read_unlock();
 747
 748	return total_len;
 749
 750err_kfree:
 751	kfree_skb(skb);
 752
 753err:
 754	rcu_read_lock();
 755	vlan = rcu_dereference(q->vlan);
 756	if (vlan)
 757		this_cpu_inc(vlan->pcpu_stats->tx_dropped);
 758	rcu_read_unlock();
 759
 760	return err;
 761}
 762
 763static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv,
 764				 unsigned long count, loff_t pos)
 765{
 766	struct file *file = iocb->ki_filp;
 767	ssize_t result = -ENOLINK;
 768	struct macvtap_queue *q = file->private_data;
 769
 770	result = macvtap_get_user(q, NULL, iv, iov_length(iv, count), count,
 771				  file->f_flags & O_NONBLOCK);
 772	return result;
 773}
 774
 775/* Put packet to the user space buffer */
 776static ssize_t macvtap_put_user(struct macvtap_queue *q,
 777				const struct sk_buff *skb,
 778				const struct iovec *iv, int len)
 779{
 
 780	int ret;
 781	int vnet_hdr_len = 0;
 782	int vlan_offset = 0;
 783	int copied, total;
 784
 785	if (q->flags & IFF_VNET_HDR) {
 786		struct virtio_net_hdr vnet_hdr;
 787		vnet_hdr_len = q->vnet_hdr_sz;
 788		if ((len -= vnet_hdr_len) < 0)
 789			return -EINVAL;
 790
 791		macvtap_skb_to_vnet_hdr(skb, &vnet_hdr);
 
 
 792
 793		if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, sizeof(vnet_hdr)))
 794			return -EFAULT;
 795	}
 796	total = copied = vnet_hdr_len;
 797	total += skb->len;
 798
 799	if (!vlan_tx_tag_present(skb))
 800		len = min_t(int, skb->len, len);
 801	else {
 802		int copy;
 803		struct {
 804			__be16 h_vlan_proto;
 805			__be16 h_vlan_TCI;
 806		} veth;
 807		veth.h_vlan_proto = skb->vlan_proto;
 808		veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb));
 809
 810		vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
 811		len = min_t(int, skb->len + VLAN_HLEN, len);
 812		total += VLAN_HLEN;
 813
 814		copy = min_t(int, vlan_offset, len);
 815		ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy);
 816		len -= copy;
 817		copied += copy;
 818		if (ret || !len)
 819			goto done;
 820
 821		copy = min_t(int, sizeof(veth), len);
 822		ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy);
 823		len -= copy;
 824		copied += copy;
 825		if (ret || !len)
 826			goto done;
 827	}
 828
 829	ret = skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len);
 
 
 
 
 830
 831done:
 832	return ret ? ret : total;
 833}
 834
 835static ssize_t macvtap_do_read(struct macvtap_queue *q,
 836			       const struct iovec *iv, unsigned long len,
 837			       int noblock)
 838{
 839	DEFINE_WAIT(wait);
 840	struct sk_buff *skb;
 841	ssize_t ret = 0;
 842
 
 843	while (len) {
 844		if (!noblock)
 845			prepare_to_wait(sk_sleep(&q->sk), &wait,
 846					TASK_INTERRUPTIBLE);
 847
 848		/* Read frames from the queue */
 849		skb = skb_dequeue(&q->sk.sk_receive_queue);
 850		if (!skb) {
 851			if (noblock) {
 852				ret = -EAGAIN;
 853				break;
 854			}
 855			if (signal_pending(current)) {
 856				ret = -ERESTARTSYS;
 857				break;
 858			}
 859			/* Nothing to read, let's sleep */
 860			schedule();
 861			continue;
 862		}
 863		ret = macvtap_put_user(q, skb, iv, len);
 864		kfree_skb(skb);
 865		break;
 866	}
 867
 868	if (!noblock)
 869		finish_wait(sk_sleep(&q->sk), &wait);
 870	return ret;
 871}
 872
 873static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv,
 874				unsigned long count, loff_t pos)
 875{
 876	struct file *file = iocb->ki_filp;
 877	struct macvtap_queue *q = file->private_data;
 878	ssize_t len, ret = 0;
 879
 880	len = iov_length(iv, count);
 881	if (len < 0) {
 882		ret = -EINVAL;
 883		goto out;
 884	}
 885
 886	ret = macvtap_do_read(q, iv, len, file->f_flags & O_NONBLOCK);
 887	ret = min_t(ssize_t, ret, len);
 888	if (ret > 0)
 889		iocb->ki_pos = ret;
 890out:
 891	return ret;
 892}
 893
 894static struct macvlan_dev *macvtap_get_vlan(struct macvtap_queue *q)
 895{
 896	struct macvlan_dev *vlan;
 897
 898	ASSERT_RTNL();
 899	vlan = rtnl_dereference(q->vlan);
 900	if (vlan)
 901		dev_hold(vlan->dev);
 902
 903	return vlan;
 904}
 905
 906static void macvtap_put_vlan(struct macvlan_dev *vlan)
 907{
 908	dev_put(vlan->dev);
 909}
 910
 911static int macvtap_ioctl_set_queue(struct file *file, unsigned int flags)
 912{
 913	struct macvtap_queue *q = file->private_data;
 914	struct macvlan_dev *vlan;
 915	int ret;
 916
 917	vlan = macvtap_get_vlan(q);
 918	if (!vlan)
 919		return -EINVAL;
 920
 921	if (flags & IFF_ATTACH_QUEUE)
 922		ret = macvtap_enable_queue(vlan->dev, file, q);
 923	else if (flags & IFF_DETACH_QUEUE)
 924		ret = macvtap_disable_queue(q);
 925	else
 926		ret = -EINVAL;
 927
 928	macvtap_put_vlan(vlan);
 929	return ret;
 930}
 931
 932static int set_offload(struct macvtap_queue *q, unsigned long arg)
 933{
 934	struct macvlan_dev *vlan;
 935	netdev_features_t features;
 936	netdev_features_t feature_mask = 0;
 937
 938	vlan = rtnl_dereference(q->vlan);
 939	if (!vlan)
 940		return -ENOLINK;
 941
 942	features = vlan->dev->features;
 943
 944	if (arg & TUN_F_CSUM) {
 945		feature_mask = NETIF_F_HW_CSUM;
 946
 947		if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) {
 948			if (arg & TUN_F_TSO_ECN)
 949				feature_mask |= NETIF_F_TSO_ECN;
 950			if (arg & TUN_F_TSO4)
 951				feature_mask |= NETIF_F_TSO;
 952			if (arg & TUN_F_TSO6)
 953				feature_mask |= NETIF_F_TSO6;
 954		}
 955
 956		if (arg & TUN_F_UFO)
 957			feature_mask |= NETIF_F_UFO;
 958	}
 959
 960	/* tun/tap driver inverts the usage for TSO offloads, where
 961	 * setting the TSO bit means that the userspace wants to
 962	 * accept TSO frames and turning it off means that user space
 963	 * does not support TSO.
 964	 * For macvtap, we have to invert it to mean the same thing.
 965	 * When user space turns off TSO, we turn off GSO/LRO so that
 966	 * user-space will not receive TSO frames.
 967	 */
 968	if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_UFO))
 969		features |= RX_OFFLOADS;
 970	else
 971		features &= ~RX_OFFLOADS;
 972
 973	/* tap_features are the same as features on tun/tap and
 974	 * reflect user expectations.
 975	 */
 976	vlan->tap_features = feature_mask;
 977	vlan->set_features = features;
 978	netdev_update_features(vlan->dev);
 979
 980	return 0;
 981}
 982
 983/*
 984 * provide compatibility with generic tun/tap interface
 985 */
 986static long macvtap_ioctl(struct file *file, unsigned int cmd,
 987			  unsigned long arg)
 988{
 989	struct macvtap_queue *q = file->private_data;
 990	struct macvlan_dev *vlan;
 991	void __user *argp = (void __user *)arg;
 992	struct ifreq __user *ifr = argp;
 993	unsigned int __user *up = argp;
 994	unsigned int u;
 995	int __user *sp = argp;
 996	int s;
 997	int ret;
 998
 999	switch (cmd) {
1000	case TUNSETIFF:
1001		/* ignore the name, just look at flags */
1002		if (get_user(u, &ifr->ifr_flags))
1003			return -EFAULT;
1004
1005		ret = 0;
1006		if ((u & ~(IFF_VNET_HDR | IFF_MULTI_QUEUE)) !=
1007		    (IFF_NO_PI | IFF_TAP))
1008			ret = -EINVAL;
1009		else
1010			q->flags = u;
1011
1012		return ret;
1013
1014	case TUNGETIFF:
1015		rtnl_lock();
1016		vlan = macvtap_get_vlan(q);
1017		if (!vlan) {
1018			rtnl_unlock();
 
 
 
1019			return -ENOLINK;
1020		}
1021
1022		ret = 0;
1023		if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) ||
1024		    put_user(q->flags, &ifr->ifr_flags))
1025			ret = -EFAULT;
1026		macvtap_put_vlan(vlan);
1027		rtnl_unlock();
1028		return ret;
1029
1030	case TUNSETQUEUE:
1031		if (get_user(u, &ifr->ifr_flags))
1032			return -EFAULT;
1033		rtnl_lock();
1034		ret = macvtap_ioctl_set_queue(file, u);
1035		rtnl_unlock();
1036		return ret;
1037
1038	case TUNGETFEATURES:
1039		if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR |
1040			     IFF_MULTI_QUEUE, up))
1041			return -EFAULT;
1042		return 0;
1043
1044	case TUNSETSNDBUF:
1045		if (get_user(u, up))
1046			return -EFAULT;
1047
1048		q->sk.sk_sndbuf = u;
1049		return 0;
1050
1051	case TUNGETVNETHDRSZ:
1052		s = q->vnet_hdr_sz;
1053		if (put_user(s, sp))
1054			return -EFAULT;
1055		return 0;
1056
1057	case TUNSETVNETHDRSZ:
1058		if (get_user(s, sp))
1059			return -EFAULT;
1060		if (s < (int)sizeof(struct virtio_net_hdr))
1061			return -EINVAL;
1062
1063		q->vnet_hdr_sz = s;
1064		return 0;
1065
1066	case TUNSETOFFLOAD:
1067		/* let the user check for future flags */
1068		if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
1069			    TUN_F_TSO_ECN | TUN_F_UFO))
1070			return -EINVAL;
1071
1072		rtnl_lock();
1073		ret = set_offload(q, arg);
1074		rtnl_unlock();
1075		return ret;
 
1076
1077	default:
1078		return -EINVAL;
1079	}
1080}
1081
1082#ifdef CONFIG_COMPAT
1083static long macvtap_compat_ioctl(struct file *file, unsigned int cmd,
1084				 unsigned long arg)
1085{
1086	return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
1087}
1088#endif
1089
1090static const struct file_operations macvtap_fops = {
1091	.owner		= THIS_MODULE,
1092	.open		= macvtap_open,
1093	.release	= macvtap_release,
1094	.aio_read	= macvtap_aio_read,
1095	.aio_write	= macvtap_aio_write,
1096	.poll		= macvtap_poll,
1097	.llseek		= no_llseek,
1098	.unlocked_ioctl	= macvtap_ioctl,
1099#ifdef CONFIG_COMPAT
1100	.compat_ioctl	= macvtap_compat_ioctl,
1101#endif
1102};
1103
1104static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock,
1105			   struct msghdr *m, size_t total_len)
1106{
1107	struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1108	return macvtap_get_user(q, m, m->msg_iov, total_len, m->msg_iovlen,
1109			    m->msg_flags & MSG_DONTWAIT);
1110}
1111
1112static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock,
1113			   struct msghdr *m, size_t total_len,
1114			   int flags)
1115{
1116	struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1117	int ret;
1118	if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1119		return -EINVAL;
1120	ret = macvtap_do_read(q, m->msg_iov, total_len,
1121			  flags & MSG_DONTWAIT);
1122	if (ret > total_len) {
1123		m->msg_flags |= MSG_TRUNC;
1124		ret = flags & MSG_TRUNC ? ret : total_len;
1125	}
1126	return ret;
1127}
1128
1129/* Ops structure to mimic raw sockets with tun */
1130static const struct proto_ops macvtap_socket_ops = {
1131	.sendmsg = macvtap_sendmsg,
1132	.recvmsg = macvtap_recvmsg,
1133};
1134
1135/* Get an underlying socket object from tun file.  Returns error unless file is
1136 * attached to a device.  The returned object works like a packet socket, it
1137 * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
1138 * holding a reference to the file for as long as the socket is in use. */
1139struct socket *macvtap_get_socket(struct file *file)
1140{
1141	struct macvtap_queue *q;
1142	if (file->f_op != &macvtap_fops)
1143		return ERR_PTR(-EINVAL);
1144	q = file->private_data;
1145	if (!q)
1146		return ERR_PTR(-EBADFD);
1147	return &q->sock;
1148}
1149EXPORT_SYMBOL_GPL(macvtap_get_socket);
1150
1151static int macvtap_device_event(struct notifier_block *unused,
1152				unsigned long event, void *ptr)
1153{
1154	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1155	struct macvlan_dev *vlan;
1156	struct device *classdev;
1157	dev_t devt;
1158	int err;
1159
1160	if (dev->rtnl_link_ops != &macvtap_link_ops)
1161		return NOTIFY_DONE;
1162
1163	vlan = netdev_priv(dev);
1164
1165	switch (event) {
1166	case NETDEV_REGISTER:
1167		/* Create the device node here after the network device has
1168		 * been registered but before register_netdevice has
1169		 * finished running.
1170		 */
1171		err = macvtap_get_minor(vlan);
1172		if (err)
1173			return notifier_from_errno(err);
1174
1175		devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
1176		classdev = device_create(macvtap_class, &dev->dev, devt,
1177					 dev, "tap%d", dev->ifindex);
1178		if (IS_ERR(classdev)) {
1179			macvtap_free_minor(vlan);
1180			return notifier_from_errno(PTR_ERR(classdev));
1181		}
1182		break;
1183	case NETDEV_UNREGISTER:
1184		devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
1185		device_destroy(macvtap_class, devt);
1186		macvtap_free_minor(vlan);
1187		break;
1188	}
1189
1190	return NOTIFY_DONE;
1191}
1192
1193static struct notifier_block macvtap_notifier_block __read_mostly = {
1194	.notifier_call	= macvtap_device_event,
1195};
1196
1197static int macvtap_init(void)
1198{
1199	int err;
1200
1201	err = alloc_chrdev_region(&macvtap_major, 0,
1202				MACVTAP_NUM_DEVS, "macvtap");
1203	if (err)
1204		goto out1;
1205
1206	cdev_init(&macvtap_cdev, &macvtap_fops);
1207	err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS);
1208	if (err)
1209		goto out2;
1210
1211	macvtap_class = class_create(THIS_MODULE, "macvtap");
1212	if (IS_ERR(macvtap_class)) {
1213		err = PTR_ERR(macvtap_class);
1214		goto out3;
1215	}
1216
1217	err = register_netdevice_notifier(&macvtap_notifier_block);
1218	if (err)
1219		goto out4;
1220
1221	err = macvlan_link_register(&macvtap_link_ops);
1222	if (err)
1223		goto out5;
1224
1225	return 0;
1226
1227out5:
1228	unregister_netdevice_notifier(&macvtap_notifier_block);
1229out4:
1230	class_unregister(macvtap_class);
1231out3:
1232	cdev_del(&macvtap_cdev);
1233out2:
1234	unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1235out1:
1236	return err;
1237}
1238module_init(macvtap_init);
1239
1240static void macvtap_exit(void)
1241{
1242	rtnl_link_unregister(&macvtap_link_ops);
1243	unregister_netdevice_notifier(&macvtap_notifier_block);
1244	class_unregister(macvtap_class);
1245	cdev_del(&macvtap_cdev);
1246	unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1247}
1248module_exit(macvtap_exit);
1249
1250MODULE_ALIAS_RTNL_LINK("macvtap");
1251MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1252MODULE_LICENSE("GPL");