xref: /openbmc/linux/drivers/net/macvtap.c (revision 93cd6fa6806cb3455e8231578840afb031606352)
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 #include <linux/uio.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 #include <linux/skb_array.h>
25 
26 /*
27  * A macvtap queue is the central object of this driver, it connects
28  * an open character device to a macvlan interface. There can be
29  * multiple queues on one interface, which map back to queues
30  * implemented in hardware on the underlying device.
31  *
32  * macvtap_proto is used to allocate queues through the sock allocation
33  * mechanism.
34  *
35  */
36 struct macvtap_queue {
37 	struct sock sk;
38 	struct socket sock;
39 	struct socket_wq wq;
40 	int vnet_hdr_sz;
41 	struct macvlan_dev __rcu *vlan;
42 	struct file *file;
43 	unsigned int flags;
44 	u16 queue_index;
45 	bool enabled;
46 	struct list_head next;
47 	struct skb_array skb_array;
48 };
49 
50 #define MACVTAP_FEATURES (IFF_VNET_HDR | IFF_MULTI_QUEUE)
51 
52 #define MACVTAP_VNET_LE 0x80000000
53 #define MACVTAP_VNET_BE 0x40000000
54 
55 #ifdef CONFIG_TUN_VNET_CROSS_LE
56 static inline bool macvtap_legacy_is_little_endian(struct macvtap_queue *q)
57 {
58 	return q->flags & MACVTAP_VNET_BE ? false :
59 		virtio_legacy_is_little_endian();
60 }
61 
62 static long macvtap_get_vnet_be(struct macvtap_queue *q, int __user *sp)
63 {
64 	int s = !!(q->flags & MACVTAP_VNET_BE);
65 
66 	if (put_user(s, sp))
67 		return -EFAULT;
68 
69 	return 0;
70 }
71 
72 static long macvtap_set_vnet_be(struct macvtap_queue *q, int __user *sp)
73 {
74 	int s;
75 
76 	if (get_user(s, sp))
77 		return -EFAULT;
78 
79 	if (s)
80 		q->flags |= MACVTAP_VNET_BE;
81 	else
82 		q->flags &= ~MACVTAP_VNET_BE;
83 
84 	return 0;
85 }
86 #else
87 static inline bool macvtap_legacy_is_little_endian(struct macvtap_queue *q)
88 {
89 	return virtio_legacy_is_little_endian();
90 }
91 
92 static long macvtap_get_vnet_be(struct macvtap_queue *q, int __user *argp)
93 {
94 	return -EINVAL;
95 }
96 
97 static long macvtap_set_vnet_be(struct macvtap_queue *q, int __user *argp)
98 {
99 	return -EINVAL;
100 }
101 #endif /* CONFIG_TUN_VNET_CROSS_LE */
102 
103 static inline bool macvtap_is_little_endian(struct macvtap_queue *q)
104 {
105 	return q->flags & MACVTAP_VNET_LE ||
106 		macvtap_legacy_is_little_endian(q);
107 }
108 
109 static inline u16 macvtap16_to_cpu(struct macvtap_queue *q, __virtio16 val)
110 {
111 	return __virtio16_to_cpu(macvtap_is_little_endian(q), val);
112 }
113 
114 static inline __virtio16 cpu_to_macvtap16(struct macvtap_queue *q, u16 val)
115 {
116 	return __cpu_to_virtio16(macvtap_is_little_endian(q), val);
117 }
118 
119 static struct proto macvtap_proto = {
120 	.name = "macvtap",
121 	.owner = THIS_MODULE,
122 	.obj_size = sizeof (struct macvtap_queue),
123 };
124 
125 /*
126  * Variables for dealing with macvtaps device numbers.
127  */
128 static dev_t macvtap_major;
129 #define MACVTAP_NUM_DEVS (1U << MINORBITS)
130 static DEFINE_MUTEX(minor_lock);
131 static DEFINE_IDR(minor_idr);
132 
133 #define GOODCOPY_LEN 128
134 static const void *macvtap_net_namespace(struct device *d)
135 {
136 	struct net_device *dev = to_net_dev(d->parent);
137 	return dev_net(dev);
138 }
139 
140 static struct class macvtap_class = {
141 	.name = "macvtap",
142 	.owner = THIS_MODULE,
143 	.ns_type = &net_ns_type_operations,
144 	.namespace = macvtap_net_namespace,
145 };
146 static struct cdev macvtap_cdev;
147 
148 static const struct proto_ops macvtap_socket_ops;
149 
150 #define TUN_OFFLOADS (NETIF_F_HW_CSUM | NETIF_F_TSO_ECN | NETIF_F_TSO | \
151 		      NETIF_F_TSO6 | NETIF_F_UFO)
152 #define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO)
153 #define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG | NETIF_F_FRAGLIST)
154 
155 static struct macvlan_dev *macvtap_get_vlan_rcu(const struct net_device *dev)
156 {
157 	return rcu_dereference(dev->rx_handler_data);
158 }
159 
160 /*
161  * RCU usage:
162  * The macvtap_queue and the macvlan_dev are loosely coupled, the
163  * pointers from one to the other can only be read while rcu_read_lock
164  * or rtnl is held.
165  *
166  * Both the file and the macvlan_dev hold a reference on the macvtap_queue
167  * through sock_hold(&q->sk). When the macvlan_dev goes away first,
168  * q->vlan becomes inaccessible. When the files gets closed,
169  * macvtap_get_queue() fails.
170  *
171  * There may still be references to the struct sock inside of the
172  * queue from outbound SKBs, but these never reference back to the
173  * file or the dev. The data structure is freed through __sk_free
174  * when both our references and any pending SKBs are gone.
175  */
176 
177 static int macvtap_enable_queue(struct net_device *dev, struct file *file,
178 				struct macvtap_queue *q)
179 {
180 	struct macvlan_dev *vlan = netdev_priv(dev);
181 	int err = -EINVAL;
182 
183 	ASSERT_RTNL();
184 
185 	if (q->enabled)
186 		goto out;
187 
188 	err = 0;
189 	rcu_assign_pointer(vlan->taps[vlan->numvtaps], q);
190 	q->queue_index = vlan->numvtaps;
191 	q->enabled = true;
192 
193 	vlan->numvtaps++;
194 out:
195 	return err;
196 }
197 
198 /* Requires RTNL */
199 static int macvtap_set_queue(struct net_device *dev, struct file *file,
200 			     struct macvtap_queue *q)
201 {
202 	struct macvlan_dev *vlan = netdev_priv(dev);
203 
204 	if (vlan->numqueues == MAX_MACVTAP_QUEUES)
205 		return -EBUSY;
206 
207 	rcu_assign_pointer(q->vlan, vlan);
208 	rcu_assign_pointer(vlan->taps[vlan->numvtaps], q);
209 	sock_hold(&q->sk);
210 
211 	q->file = file;
212 	q->queue_index = vlan->numvtaps;
213 	q->enabled = true;
214 	file->private_data = q;
215 	list_add_tail(&q->next, &vlan->queue_list);
216 
217 	vlan->numvtaps++;
218 	vlan->numqueues++;
219 
220 	return 0;
221 }
222 
223 static int macvtap_disable_queue(struct macvtap_queue *q)
224 {
225 	struct macvlan_dev *vlan;
226 	struct macvtap_queue *nq;
227 
228 	ASSERT_RTNL();
229 	if (!q->enabled)
230 		return -EINVAL;
231 
232 	vlan = rtnl_dereference(q->vlan);
233 
234 	if (vlan) {
235 		int index = q->queue_index;
236 		BUG_ON(index >= vlan->numvtaps);
237 		nq = rtnl_dereference(vlan->taps[vlan->numvtaps - 1]);
238 		nq->queue_index = index;
239 
240 		rcu_assign_pointer(vlan->taps[index], nq);
241 		RCU_INIT_POINTER(vlan->taps[vlan->numvtaps - 1], NULL);
242 		q->enabled = false;
243 
244 		vlan->numvtaps--;
245 	}
246 
247 	return 0;
248 }
249 
250 /*
251  * The file owning the queue got closed, give up both
252  * the reference that the files holds as well as the
253  * one from the macvlan_dev if that still exists.
254  *
255  * Using the spinlock makes sure that we don't get
256  * to the queue again after destroying it.
257  */
258 static void macvtap_put_queue(struct macvtap_queue *q)
259 {
260 	struct macvlan_dev *vlan;
261 
262 	rtnl_lock();
263 	vlan = rtnl_dereference(q->vlan);
264 
265 	if (vlan) {
266 		if (q->enabled)
267 			BUG_ON(macvtap_disable_queue(q));
268 
269 		vlan->numqueues--;
270 		RCU_INIT_POINTER(q->vlan, NULL);
271 		sock_put(&q->sk);
272 		list_del_init(&q->next);
273 	}
274 
275 	rtnl_unlock();
276 
277 	synchronize_rcu();
278 	sock_put(&q->sk);
279 }
280 
281 /*
282  * Select a queue based on the rxq of the device on which this packet
283  * arrived. If the incoming device is not mq, calculate a flow hash
284  * to select a queue. If all fails, find the first available queue.
285  * Cache vlan->numvtaps since it can become zero during the execution
286  * of this function.
287  */
288 static struct macvtap_queue *macvtap_get_queue(struct net_device *dev,
289 					       struct sk_buff *skb)
290 {
291 	struct macvlan_dev *vlan = netdev_priv(dev);
292 	struct macvtap_queue *tap = NULL;
293 	/* Access to taps array is protected by rcu, but access to numvtaps
294 	 * isn't. Below we use it to lookup a queue, but treat it as a hint
295 	 * and validate that the result isn't NULL - in case we are
296 	 * racing against queue removal.
297 	 */
298 	int numvtaps = ACCESS_ONCE(vlan->numvtaps);
299 	__u32 rxq;
300 
301 	if (!numvtaps)
302 		goto out;
303 
304 	if (numvtaps == 1)
305 		goto single;
306 
307 	/* Check if we can use flow to select a queue */
308 	rxq = skb_get_hash(skb);
309 	if (rxq) {
310 		tap = rcu_dereference(vlan->taps[rxq % numvtaps]);
311 		goto out;
312 	}
313 
314 	if (likely(skb_rx_queue_recorded(skb))) {
315 		rxq = skb_get_rx_queue(skb);
316 
317 		while (unlikely(rxq >= numvtaps))
318 			rxq -= numvtaps;
319 
320 		tap = rcu_dereference(vlan->taps[rxq]);
321 		goto out;
322 	}
323 
324 single:
325 	tap = rcu_dereference(vlan->taps[0]);
326 out:
327 	return tap;
328 }
329 
330 /*
331  * The net_device is going away, give up the reference
332  * that it holds on all queues and safely set the pointer
333  * from the queues to NULL.
334  */
335 static void macvtap_del_queues(struct net_device *dev)
336 {
337 	struct macvlan_dev *vlan = netdev_priv(dev);
338 	struct macvtap_queue *q, *tmp;
339 
340 	ASSERT_RTNL();
341 	list_for_each_entry_safe(q, tmp, &vlan->queue_list, next) {
342 		list_del_init(&q->next);
343 		RCU_INIT_POINTER(q->vlan, NULL);
344 		if (q->enabled)
345 			vlan->numvtaps--;
346 		vlan->numqueues--;
347 		sock_put(&q->sk);
348 	}
349 	BUG_ON(vlan->numvtaps);
350 	BUG_ON(vlan->numqueues);
351 	/* guarantee that any future macvtap_set_queue will fail */
352 	vlan->numvtaps = MAX_MACVTAP_QUEUES;
353 }
354 
355 static rx_handler_result_t macvtap_handle_frame(struct sk_buff **pskb)
356 {
357 	struct sk_buff *skb = *pskb;
358 	struct net_device *dev = skb->dev;
359 	struct macvlan_dev *vlan;
360 	struct macvtap_queue *q;
361 	netdev_features_t features = TAP_FEATURES;
362 
363 	vlan = macvtap_get_vlan_rcu(dev);
364 	if (!vlan)
365 		return RX_HANDLER_PASS;
366 
367 	q = macvtap_get_queue(dev, skb);
368 	if (!q)
369 		return RX_HANDLER_PASS;
370 
371 	if (__skb_array_full(&q->skb_array))
372 		goto drop;
373 
374 	skb_push(skb, ETH_HLEN);
375 
376 	/* Apply the forward feature mask so that we perform segmentation
377 	 * according to users wishes.  This only works if VNET_HDR is
378 	 * enabled.
379 	 */
380 	if (q->flags & IFF_VNET_HDR)
381 		features |= vlan->tap_features;
382 	if (netif_needs_gso(skb, features)) {
383 		struct sk_buff *segs = __skb_gso_segment(skb, features, false);
384 
385 		if (IS_ERR(segs))
386 			goto drop;
387 
388 		if (!segs) {
389 			if (skb_array_produce(&q->skb_array, skb))
390 				goto drop;
391 			goto wake_up;
392 		}
393 
394 		consume_skb(skb);
395 		while (segs) {
396 			struct sk_buff *nskb = segs->next;
397 
398 			segs->next = NULL;
399 			if (skb_array_produce(&q->skb_array, segs)) {
400 				kfree_skb(segs);
401 				kfree_skb_list(nskb);
402 				break;
403 			}
404 			segs = nskb;
405 		}
406 	} else {
407 		/* If we receive a partial checksum and the tap side
408 		 * doesn't support checksum offload, compute the checksum.
409 		 * Note: it doesn't matter which checksum feature to
410 		 *        check, we either support them all or none.
411 		 */
412 		if (skb->ip_summed == CHECKSUM_PARTIAL &&
413 		    !(features & NETIF_F_CSUM_MASK) &&
414 		    skb_checksum_help(skb))
415 			goto drop;
416 		if (skb_array_produce(&q->skb_array, skb))
417 			goto drop;
418 	}
419 
420 wake_up:
421 	wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND);
422 	return RX_HANDLER_CONSUMED;
423 
424 drop:
425 	/* Count errors/drops only here, thus don't care about args. */
426 	macvlan_count_rx(vlan, 0, 0, 0);
427 	kfree_skb(skb);
428 	return RX_HANDLER_CONSUMED;
429 }
430 
431 static int macvtap_get_minor(struct macvlan_dev *vlan)
432 {
433 	int retval = -ENOMEM;
434 
435 	mutex_lock(&minor_lock);
436 	retval = idr_alloc(&minor_idr, vlan, 1, MACVTAP_NUM_DEVS, GFP_KERNEL);
437 	if (retval >= 0) {
438 		vlan->minor = retval;
439 	} else if (retval == -ENOSPC) {
440 		printk(KERN_ERR "too many macvtap devices\n");
441 		retval = -EINVAL;
442 	}
443 	mutex_unlock(&minor_lock);
444 	return retval < 0 ? retval : 0;
445 }
446 
447 static void macvtap_free_minor(struct macvlan_dev *vlan)
448 {
449 	mutex_lock(&minor_lock);
450 	if (vlan->minor) {
451 		idr_remove(&minor_idr, vlan->minor);
452 		vlan->minor = 0;
453 	}
454 	mutex_unlock(&minor_lock);
455 }
456 
457 static struct net_device *dev_get_by_macvtap_minor(int minor)
458 {
459 	struct net_device *dev = NULL;
460 	struct macvlan_dev *vlan;
461 
462 	mutex_lock(&minor_lock);
463 	vlan = idr_find(&minor_idr, minor);
464 	if (vlan) {
465 		dev = vlan->dev;
466 		dev_hold(dev);
467 	}
468 	mutex_unlock(&minor_lock);
469 	return dev;
470 }
471 
472 static int macvtap_newlink(struct net *src_net,
473 			   struct net_device *dev,
474 			   struct nlattr *tb[],
475 			   struct nlattr *data[])
476 {
477 	struct macvlan_dev *vlan = netdev_priv(dev);
478 	int err;
479 
480 	INIT_LIST_HEAD(&vlan->queue_list);
481 
482 	/* Since macvlan supports all offloads by default, make
483 	 * tap support all offloads also.
484 	 */
485 	vlan->tap_features = TUN_OFFLOADS;
486 
487 	err = netdev_rx_handler_register(dev, macvtap_handle_frame, vlan);
488 	if (err)
489 		return err;
490 
491 	/* Don't put anything that may fail after macvlan_common_newlink
492 	 * because we can't undo what it does.
493 	 */
494 	err = macvlan_common_newlink(src_net, dev, tb, data);
495 	if (err) {
496 		netdev_rx_handler_unregister(dev);
497 		return err;
498 	}
499 
500 	return 0;
501 }
502 
503 static void macvtap_dellink(struct net_device *dev,
504 			    struct list_head *head)
505 {
506 	netdev_rx_handler_unregister(dev);
507 	macvtap_del_queues(dev);
508 	macvlan_dellink(dev, head);
509 }
510 
511 static void macvtap_setup(struct net_device *dev)
512 {
513 	macvlan_common_setup(dev);
514 	dev->tx_queue_len = TUN_READQ_SIZE;
515 }
516 
517 static struct rtnl_link_ops macvtap_link_ops __read_mostly = {
518 	.kind		= "macvtap",
519 	.setup		= macvtap_setup,
520 	.newlink	= macvtap_newlink,
521 	.dellink	= macvtap_dellink,
522 };
523 
524 
525 static void macvtap_sock_write_space(struct sock *sk)
526 {
527 	wait_queue_head_t *wqueue;
528 
529 	if (!sock_writeable(sk) ||
530 	    !test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
531 		return;
532 
533 	wqueue = sk_sleep(sk);
534 	if (wqueue && waitqueue_active(wqueue))
535 		wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND);
536 }
537 
538 static void macvtap_sock_destruct(struct sock *sk)
539 {
540 	struct macvtap_queue *q = container_of(sk, struct macvtap_queue, sk);
541 
542 	skb_array_cleanup(&q->skb_array);
543 }
544 
545 static int macvtap_open(struct inode *inode, struct file *file)
546 {
547 	struct net *net = current->nsproxy->net_ns;
548 	struct net_device *dev;
549 	struct macvtap_queue *q;
550 	int err = -ENODEV;
551 
552 	rtnl_lock();
553 	dev = dev_get_by_macvtap_minor(iminor(inode));
554 	if (!dev)
555 		goto err;
556 
557 	err = -ENOMEM;
558 	q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
559 					     &macvtap_proto, 0);
560 	if (!q)
561 		goto err;
562 
563 	RCU_INIT_POINTER(q->sock.wq, &q->wq);
564 	init_waitqueue_head(&q->wq.wait);
565 	q->sock.type = SOCK_RAW;
566 	q->sock.state = SS_CONNECTED;
567 	q->sock.file = file;
568 	q->sock.ops = &macvtap_socket_ops;
569 	sock_init_data(&q->sock, &q->sk);
570 	q->sk.sk_write_space = macvtap_sock_write_space;
571 	q->sk.sk_destruct = macvtap_sock_destruct;
572 	q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
573 	q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
574 
575 	/*
576 	 * so far only KVM virtio_net uses macvtap, enable zero copy between
577 	 * guest kernel and host kernel when lower device supports zerocopy
578 	 *
579 	 * The macvlan supports zerocopy iff the lower device supports zero
580 	 * copy so we don't have to look at the lower device directly.
581 	 */
582 	if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG))
583 		sock_set_flag(&q->sk, SOCK_ZEROCOPY);
584 
585 	err = -ENOMEM;
586 	if (skb_array_init(&q->skb_array, dev->tx_queue_len, GFP_KERNEL))
587 		goto err_array;
588 
589 	err = macvtap_set_queue(dev, file, q);
590 	if (err)
591 		goto err_queue;
592 
593 	dev_put(dev);
594 
595 	rtnl_unlock();
596 	return err;
597 
598 err_queue:
599 	skb_array_cleanup(&q->skb_array);
600 err_array:
601 	sock_put(&q->sk);
602 err:
603 	if (dev)
604 		dev_put(dev);
605 
606 	rtnl_unlock();
607 	return err;
608 }
609 
610 static int macvtap_release(struct inode *inode, struct file *file)
611 {
612 	struct macvtap_queue *q = file->private_data;
613 	macvtap_put_queue(q);
614 	return 0;
615 }
616 
617 static unsigned int macvtap_poll(struct file *file, poll_table * wait)
618 {
619 	struct macvtap_queue *q = file->private_data;
620 	unsigned int mask = POLLERR;
621 
622 	if (!q)
623 		goto out;
624 
625 	mask = 0;
626 	poll_wait(file, &q->wq.wait, wait);
627 
628 	if (!skb_array_empty(&q->skb_array))
629 		mask |= POLLIN | POLLRDNORM;
630 
631 	if (sock_writeable(&q->sk) ||
632 	    (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &q->sock.flags) &&
633 	     sock_writeable(&q->sk)))
634 		mask |= POLLOUT | POLLWRNORM;
635 
636 out:
637 	return mask;
638 }
639 
640 static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad,
641 						size_t len, size_t linear,
642 						int noblock, int *err)
643 {
644 	struct sk_buff *skb;
645 
646 	/* Under a page?  Don't bother with paged skb. */
647 	if (prepad + len < PAGE_SIZE || !linear)
648 		linear = len;
649 
650 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
651 				   err, 0);
652 	if (!skb)
653 		return NULL;
654 
655 	skb_reserve(skb, prepad);
656 	skb_put(skb, linear);
657 	skb->data_len = len - linear;
658 	skb->len += len - linear;
659 
660 	return skb;
661 }
662 
663 /* Neighbour code has some assumptions on HH_DATA_MOD alignment */
664 #define MACVTAP_RESERVE HH_DATA_OFF(ETH_HLEN)
665 
666 /* Get packet from user space buffer */
667 static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m,
668 				struct iov_iter *from, int noblock)
669 {
670 	int good_linear = SKB_MAX_HEAD(MACVTAP_RESERVE);
671 	struct sk_buff *skb;
672 	struct macvlan_dev *vlan;
673 	unsigned long total_len = iov_iter_count(from);
674 	unsigned long len = total_len;
675 	int err;
676 	struct virtio_net_hdr vnet_hdr = { 0 };
677 	int vnet_hdr_len = 0;
678 	int copylen = 0;
679 	int depth;
680 	bool zerocopy = false;
681 	size_t linear;
682 	ssize_t n;
683 
684 	if (q->flags & IFF_VNET_HDR) {
685 		vnet_hdr_len = q->vnet_hdr_sz;
686 
687 		err = -EINVAL;
688 		if (len < vnet_hdr_len)
689 			goto err;
690 		len -= vnet_hdr_len;
691 
692 		err = -EFAULT;
693 		n = copy_from_iter(&vnet_hdr, sizeof(vnet_hdr), from);
694 		if (n != sizeof(vnet_hdr))
695 			goto err;
696 		iov_iter_advance(from, vnet_hdr_len - sizeof(vnet_hdr));
697 		if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
698 		     macvtap16_to_cpu(q, vnet_hdr.csum_start) +
699 		     macvtap16_to_cpu(q, vnet_hdr.csum_offset) + 2 >
700 			     macvtap16_to_cpu(q, vnet_hdr.hdr_len))
701 			vnet_hdr.hdr_len = cpu_to_macvtap16(q,
702 				 macvtap16_to_cpu(q, vnet_hdr.csum_start) +
703 				 macvtap16_to_cpu(q, vnet_hdr.csum_offset) + 2);
704 		err = -EINVAL;
705 		if (macvtap16_to_cpu(q, vnet_hdr.hdr_len) > len)
706 			goto err;
707 	}
708 
709 	err = -EINVAL;
710 	if (unlikely(len < ETH_HLEN))
711 		goto err;
712 
713 	if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) {
714 		struct iov_iter i;
715 
716 		copylen = vnet_hdr.hdr_len ?
717 			macvtap16_to_cpu(q, vnet_hdr.hdr_len) : GOODCOPY_LEN;
718 		if (copylen > good_linear)
719 			copylen = good_linear;
720 		else if (copylen < ETH_HLEN)
721 			copylen = ETH_HLEN;
722 		linear = copylen;
723 		i = *from;
724 		iov_iter_advance(&i, copylen);
725 		if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
726 			zerocopy = true;
727 	}
728 
729 	if (!zerocopy) {
730 		copylen = len;
731 		linear = macvtap16_to_cpu(q, vnet_hdr.hdr_len);
732 		if (linear > good_linear)
733 			linear = good_linear;
734 		else if (linear < ETH_HLEN)
735 			linear = ETH_HLEN;
736 	}
737 
738 	skb = macvtap_alloc_skb(&q->sk, MACVTAP_RESERVE, copylen,
739 				linear, noblock, &err);
740 	if (!skb)
741 		goto err;
742 
743 	if (zerocopy)
744 		err = zerocopy_sg_from_iter(skb, from);
745 	else
746 		err = skb_copy_datagram_from_iter(skb, 0, from, len);
747 
748 	if (err)
749 		goto err_kfree;
750 
751 	skb_set_network_header(skb, ETH_HLEN);
752 	skb_reset_mac_header(skb);
753 	skb->protocol = eth_hdr(skb)->h_proto;
754 
755 	if (vnet_hdr_len) {
756 		err = virtio_net_hdr_to_skb(skb, &vnet_hdr,
757 					    macvtap_is_little_endian(q));
758 		if (err)
759 			goto err_kfree;
760 	}
761 
762 	skb_probe_transport_header(skb, ETH_HLEN);
763 
764 	/* Move network header to the right position for VLAN tagged packets */
765 	if ((skb->protocol == htons(ETH_P_8021Q) ||
766 	     skb->protocol == htons(ETH_P_8021AD)) &&
767 	    __vlan_get_protocol(skb, skb->protocol, &depth) != 0)
768 		skb_set_network_header(skb, depth);
769 
770 	rcu_read_lock();
771 	vlan = rcu_dereference(q->vlan);
772 	/* copy skb_ubuf_info for callback when skb has no error */
773 	if (zerocopy) {
774 		skb_shinfo(skb)->destructor_arg = m->msg_control;
775 		skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
776 		skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
777 	} else if (m && m->msg_control) {
778 		struct ubuf_info *uarg = m->msg_control;
779 		uarg->callback(uarg, false);
780 	}
781 
782 	if (vlan) {
783 		skb->dev = vlan->dev;
784 		dev_queue_xmit(skb);
785 	} else {
786 		kfree_skb(skb);
787 	}
788 	rcu_read_unlock();
789 
790 	return total_len;
791 
792 err_kfree:
793 	kfree_skb(skb);
794 
795 err:
796 	rcu_read_lock();
797 	vlan = rcu_dereference(q->vlan);
798 	if (vlan)
799 		this_cpu_inc(vlan->pcpu_stats->tx_dropped);
800 	rcu_read_unlock();
801 
802 	return err;
803 }
804 
805 static ssize_t macvtap_write_iter(struct kiocb *iocb, struct iov_iter *from)
806 {
807 	struct file *file = iocb->ki_filp;
808 	struct macvtap_queue *q = file->private_data;
809 
810 	return macvtap_get_user(q, NULL, from, file->f_flags & O_NONBLOCK);
811 }
812 
813 /* Put packet to the user space buffer */
814 static ssize_t macvtap_put_user(struct macvtap_queue *q,
815 				const struct sk_buff *skb,
816 				struct iov_iter *iter)
817 {
818 	int ret;
819 	int vnet_hdr_len = 0;
820 	int vlan_offset = 0;
821 	int total;
822 
823 	if (q->flags & IFF_VNET_HDR) {
824 		struct virtio_net_hdr vnet_hdr;
825 		vnet_hdr_len = q->vnet_hdr_sz;
826 		if (iov_iter_count(iter) < vnet_hdr_len)
827 			return -EINVAL;
828 
829 		ret = virtio_net_hdr_from_skb(skb, &vnet_hdr,
830 					      macvtap_is_little_endian(q));
831 		if (ret)
832 			BUG();
833 
834 		if (copy_to_iter(&vnet_hdr, sizeof(vnet_hdr), iter) !=
835 		    sizeof(vnet_hdr))
836 			return -EFAULT;
837 
838 		iov_iter_advance(iter, vnet_hdr_len - sizeof(vnet_hdr));
839 	}
840 	total = vnet_hdr_len;
841 	total += skb->len;
842 
843 	if (skb_vlan_tag_present(skb)) {
844 		struct {
845 			__be16 h_vlan_proto;
846 			__be16 h_vlan_TCI;
847 		} veth;
848 		veth.h_vlan_proto = skb->vlan_proto;
849 		veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
850 
851 		vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
852 		total += VLAN_HLEN;
853 
854 		ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
855 		if (ret || !iov_iter_count(iter))
856 			goto done;
857 
858 		ret = copy_to_iter(&veth, sizeof(veth), iter);
859 		if (ret != sizeof(veth) || !iov_iter_count(iter))
860 			goto done;
861 	}
862 
863 	ret = skb_copy_datagram_iter(skb, vlan_offset, iter,
864 				     skb->len - vlan_offset);
865 
866 done:
867 	return ret ? ret : total;
868 }
869 
870 static ssize_t macvtap_do_read(struct macvtap_queue *q,
871 			       struct iov_iter *to,
872 			       int noblock)
873 {
874 	DEFINE_WAIT(wait);
875 	struct sk_buff *skb;
876 	ssize_t ret = 0;
877 
878 	if (!iov_iter_count(to))
879 		return 0;
880 
881 	while (1) {
882 		if (!noblock)
883 			prepare_to_wait(sk_sleep(&q->sk), &wait,
884 					TASK_INTERRUPTIBLE);
885 
886 		/* Read frames from the queue */
887 		skb = skb_array_consume(&q->skb_array);
888 		if (skb)
889 			break;
890 		if (noblock) {
891 			ret = -EAGAIN;
892 			break;
893 		}
894 		if (signal_pending(current)) {
895 			ret = -ERESTARTSYS;
896 			break;
897 		}
898 		/* Nothing to read, let's sleep */
899 		schedule();
900 	}
901 	if (!noblock)
902 		finish_wait(sk_sleep(&q->sk), &wait);
903 
904 	if (skb) {
905 		ret = macvtap_put_user(q, skb, to);
906 		if (unlikely(ret < 0))
907 			kfree_skb(skb);
908 		else
909 			consume_skb(skb);
910 	}
911 	return ret;
912 }
913 
914 static ssize_t macvtap_read_iter(struct kiocb *iocb, struct iov_iter *to)
915 {
916 	struct file *file = iocb->ki_filp;
917 	struct macvtap_queue *q = file->private_data;
918 	ssize_t len = iov_iter_count(to), ret;
919 
920 	ret = macvtap_do_read(q, to, file->f_flags & O_NONBLOCK);
921 	ret = min_t(ssize_t, ret, len);
922 	if (ret > 0)
923 		iocb->ki_pos = ret;
924 	return ret;
925 }
926 
927 static struct macvlan_dev *macvtap_get_vlan(struct macvtap_queue *q)
928 {
929 	struct macvlan_dev *vlan;
930 
931 	ASSERT_RTNL();
932 	vlan = rtnl_dereference(q->vlan);
933 	if (vlan)
934 		dev_hold(vlan->dev);
935 
936 	return vlan;
937 }
938 
939 static void macvtap_put_vlan(struct macvlan_dev *vlan)
940 {
941 	dev_put(vlan->dev);
942 }
943 
944 static int macvtap_ioctl_set_queue(struct file *file, unsigned int flags)
945 {
946 	struct macvtap_queue *q = file->private_data;
947 	struct macvlan_dev *vlan;
948 	int ret;
949 
950 	vlan = macvtap_get_vlan(q);
951 	if (!vlan)
952 		return -EINVAL;
953 
954 	if (flags & IFF_ATTACH_QUEUE)
955 		ret = macvtap_enable_queue(vlan->dev, file, q);
956 	else if (flags & IFF_DETACH_QUEUE)
957 		ret = macvtap_disable_queue(q);
958 	else
959 		ret = -EINVAL;
960 
961 	macvtap_put_vlan(vlan);
962 	return ret;
963 }
964 
965 static int set_offload(struct macvtap_queue *q, unsigned long arg)
966 {
967 	struct macvlan_dev *vlan;
968 	netdev_features_t features;
969 	netdev_features_t feature_mask = 0;
970 
971 	vlan = rtnl_dereference(q->vlan);
972 	if (!vlan)
973 		return -ENOLINK;
974 
975 	features = vlan->dev->features;
976 
977 	if (arg & TUN_F_CSUM) {
978 		feature_mask = NETIF_F_HW_CSUM;
979 
980 		if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) {
981 			if (arg & TUN_F_TSO_ECN)
982 				feature_mask |= NETIF_F_TSO_ECN;
983 			if (arg & TUN_F_TSO4)
984 				feature_mask |= NETIF_F_TSO;
985 			if (arg & TUN_F_TSO6)
986 				feature_mask |= NETIF_F_TSO6;
987 		}
988 
989 		if (arg & TUN_F_UFO)
990 			feature_mask |= NETIF_F_UFO;
991 	}
992 
993 	/* tun/tap driver inverts the usage for TSO offloads, where
994 	 * setting the TSO bit means that the userspace wants to
995 	 * accept TSO frames and turning it off means that user space
996 	 * does not support TSO.
997 	 * For macvtap, we have to invert it to mean the same thing.
998 	 * When user space turns off TSO, we turn off GSO/LRO so that
999 	 * user-space will not receive TSO frames.
1000 	 */
1001 	if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_UFO))
1002 		features |= RX_OFFLOADS;
1003 	else
1004 		features &= ~RX_OFFLOADS;
1005 
1006 	/* tap_features are the same as features on tun/tap and
1007 	 * reflect user expectations.
1008 	 */
1009 	vlan->tap_features = feature_mask;
1010 	vlan->set_features = features;
1011 	netdev_update_features(vlan->dev);
1012 
1013 	return 0;
1014 }
1015 
1016 /*
1017  * provide compatibility with generic tun/tap interface
1018  */
1019 static long macvtap_ioctl(struct file *file, unsigned int cmd,
1020 			  unsigned long arg)
1021 {
1022 	struct macvtap_queue *q = file->private_data;
1023 	struct macvlan_dev *vlan;
1024 	void __user *argp = (void __user *)arg;
1025 	struct ifreq __user *ifr = argp;
1026 	unsigned int __user *up = argp;
1027 	unsigned short u;
1028 	int __user *sp = argp;
1029 	struct sockaddr sa;
1030 	int s;
1031 	int ret;
1032 
1033 	switch (cmd) {
1034 	case TUNSETIFF:
1035 		/* ignore the name, just look at flags */
1036 		if (get_user(u, &ifr->ifr_flags))
1037 			return -EFAULT;
1038 
1039 		ret = 0;
1040 		if ((u & ~MACVTAP_FEATURES) != (IFF_NO_PI | IFF_TAP))
1041 			ret = -EINVAL;
1042 		else
1043 			q->flags = (q->flags & ~MACVTAP_FEATURES) | u;
1044 
1045 		return ret;
1046 
1047 	case TUNGETIFF:
1048 		rtnl_lock();
1049 		vlan = macvtap_get_vlan(q);
1050 		if (!vlan) {
1051 			rtnl_unlock();
1052 			return -ENOLINK;
1053 		}
1054 
1055 		ret = 0;
1056 		u = q->flags;
1057 		if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) ||
1058 		    put_user(u, &ifr->ifr_flags))
1059 			ret = -EFAULT;
1060 		macvtap_put_vlan(vlan);
1061 		rtnl_unlock();
1062 		return ret;
1063 
1064 	case TUNSETQUEUE:
1065 		if (get_user(u, &ifr->ifr_flags))
1066 			return -EFAULT;
1067 		rtnl_lock();
1068 		ret = macvtap_ioctl_set_queue(file, u);
1069 		rtnl_unlock();
1070 		return ret;
1071 
1072 	case TUNGETFEATURES:
1073 		if (put_user(IFF_TAP | IFF_NO_PI | MACVTAP_FEATURES, up))
1074 			return -EFAULT;
1075 		return 0;
1076 
1077 	case TUNSETSNDBUF:
1078 		if (get_user(s, sp))
1079 			return -EFAULT;
1080 
1081 		q->sk.sk_sndbuf = s;
1082 		return 0;
1083 
1084 	case TUNGETVNETHDRSZ:
1085 		s = q->vnet_hdr_sz;
1086 		if (put_user(s, sp))
1087 			return -EFAULT;
1088 		return 0;
1089 
1090 	case TUNSETVNETHDRSZ:
1091 		if (get_user(s, sp))
1092 			return -EFAULT;
1093 		if (s < (int)sizeof(struct virtio_net_hdr))
1094 			return -EINVAL;
1095 
1096 		q->vnet_hdr_sz = s;
1097 		return 0;
1098 
1099 	case TUNGETVNETLE:
1100 		s = !!(q->flags & MACVTAP_VNET_LE);
1101 		if (put_user(s, sp))
1102 			return -EFAULT;
1103 		return 0;
1104 
1105 	case TUNSETVNETLE:
1106 		if (get_user(s, sp))
1107 			return -EFAULT;
1108 		if (s)
1109 			q->flags |= MACVTAP_VNET_LE;
1110 		else
1111 			q->flags &= ~MACVTAP_VNET_LE;
1112 		return 0;
1113 
1114 	case TUNGETVNETBE:
1115 		return macvtap_get_vnet_be(q, sp);
1116 
1117 	case TUNSETVNETBE:
1118 		return macvtap_set_vnet_be(q, sp);
1119 
1120 	case TUNSETOFFLOAD:
1121 		/* let the user check for future flags */
1122 		if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
1123 			    TUN_F_TSO_ECN | TUN_F_UFO))
1124 			return -EINVAL;
1125 
1126 		rtnl_lock();
1127 		ret = set_offload(q, arg);
1128 		rtnl_unlock();
1129 		return ret;
1130 
1131 	case SIOCGIFHWADDR:
1132 		rtnl_lock();
1133 		vlan = macvtap_get_vlan(q);
1134 		if (!vlan) {
1135 			rtnl_unlock();
1136 			return -ENOLINK;
1137 		}
1138 		ret = 0;
1139 		u = vlan->dev->type;
1140 		if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) ||
1141 		    copy_to_user(&ifr->ifr_hwaddr.sa_data, vlan->dev->dev_addr, ETH_ALEN) ||
1142 		    put_user(u, &ifr->ifr_hwaddr.sa_family))
1143 			ret = -EFAULT;
1144 		macvtap_put_vlan(vlan);
1145 		rtnl_unlock();
1146 		return ret;
1147 
1148 	case SIOCSIFHWADDR:
1149 		if (copy_from_user(&sa, &ifr->ifr_hwaddr, sizeof(sa)))
1150 			return -EFAULT;
1151 		rtnl_lock();
1152 		vlan = macvtap_get_vlan(q);
1153 		if (!vlan) {
1154 			rtnl_unlock();
1155 			return -ENOLINK;
1156 		}
1157 		ret = dev_set_mac_address(vlan->dev, &sa);
1158 		macvtap_put_vlan(vlan);
1159 		rtnl_unlock();
1160 		return ret;
1161 
1162 	default:
1163 		return -EINVAL;
1164 	}
1165 }
1166 
1167 #ifdef CONFIG_COMPAT
1168 static long macvtap_compat_ioctl(struct file *file, unsigned int cmd,
1169 				 unsigned long arg)
1170 {
1171 	return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
1172 }
1173 #endif
1174 
1175 static const struct file_operations macvtap_fops = {
1176 	.owner		= THIS_MODULE,
1177 	.open		= macvtap_open,
1178 	.release	= macvtap_release,
1179 	.read_iter	= macvtap_read_iter,
1180 	.write_iter	= macvtap_write_iter,
1181 	.poll		= macvtap_poll,
1182 	.llseek		= no_llseek,
1183 	.unlocked_ioctl	= macvtap_ioctl,
1184 #ifdef CONFIG_COMPAT
1185 	.compat_ioctl	= macvtap_compat_ioctl,
1186 #endif
1187 };
1188 
1189 static int macvtap_sendmsg(struct socket *sock, struct msghdr *m,
1190 			   size_t total_len)
1191 {
1192 	struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1193 	return macvtap_get_user(q, m, &m->msg_iter, m->msg_flags & MSG_DONTWAIT);
1194 }
1195 
1196 static int macvtap_recvmsg(struct socket *sock, struct msghdr *m,
1197 			   size_t total_len, int flags)
1198 {
1199 	struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1200 	int ret;
1201 	if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1202 		return -EINVAL;
1203 	ret = macvtap_do_read(q, &m->msg_iter, flags & MSG_DONTWAIT);
1204 	if (ret > total_len) {
1205 		m->msg_flags |= MSG_TRUNC;
1206 		ret = flags & MSG_TRUNC ? ret : total_len;
1207 	}
1208 	return ret;
1209 }
1210 
1211 static int macvtap_peek_len(struct socket *sock)
1212 {
1213 	struct macvtap_queue *q = container_of(sock, struct macvtap_queue,
1214 					       sock);
1215 	return skb_array_peek_len(&q->skb_array);
1216 }
1217 
1218 /* Ops structure to mimic raw sockets with tun */
1219 static const struct proto_ops macvtap_socket_ops = {
1220 	.sendmsg = macvtap_sendmsg,
1221 	.recvmsg = macvtap_recvmsg,
1222 	.peek_len = macvtap_peek_len,
1223 };
1224 
1225 /* Get an underlying socket object from tun file.  Returns error unless file is
1226  * attached to a device.  The returned object works like a packet socket, it
1227  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
1228  * holding a reference to the file for as long as the socket is in use. */
1229 struct socket *macvtap_get_socket(struct file *file)
1230 {
1231 	struct macvtap_queue *q;
1232 	if (file->f_op != &macvtap_fops)
1233 		return ERR_PTR(-EINVAL);
1234 	q = file->private_data;
1235 	if (!q)
1236 		return ERR_PTR(-EBADFD);
1237 	return &q->sock;
1238 }
1239 EXPORT_SYMBOL_GPL(macvtap_get_socket);
1240 
1241 static int macvtap_queue_resize(struct macvlan_dev *vlan)
1242 {
1243 	struct net_device *dev = vlan->dev;
1244 	struct macvtap_queue *q;
1245 	struct skb_array **arrays;
1246 	int n = vlan->numqueues;
1247 	int ret, i = 0;
1248 
1249 	arrays = kmalloc(sizeof *arrays * n, GFP_KERNEL);
1250 	if (!arrays)
1251 		return -ENOMEM;
1252 
1253 	list_for_each_entry(q, &vlan->queue_list, next)
1254 		arrays[i++] = &q->skb_array;
1255 
1256 	ret = skb_array_resize_multiple(arrays, n,
1257 					dev->tx_queue_len, GFP_KERNEL);
1258 
1259 	kfree(arrays);
1260 	return ret;
1261 }
1262 
1263 static int macvtap_device_event(struct notifier_block *unused,
1264 				unsigned long event, void *ptr)
1265 {
1266 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1267 	struct macvlan_dev *vlan;
1268 	struct device *classdev;
1269 	dev_t devt;
1270 	int err;
1271 	char tap_name[IFNAMSIZ];
1272 
1273 	if (dev->rtnl_link_ops != &macvtap_link_ops)
1274 		return NOTIFY_DONE;
1275 
1276 	snprintf(tap_name, IFNAMSIZ, "tap%d", dev->ifindex);
1277 	vlan = netdev_priv(dev);
1278 
1279 	switch (event) {
1280 	case NETDEV_REGISTER:
1281 		/* Create the device node here after the network device has
1282 		 * been registered but before register_netdevice has
1283 		 * finished running.
1284 		 */
1285 		err = macvtap_get_minor(vlan);
1286 		if (err)
1287 			return notifier_from_errno(err);
1288 
1289 		devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
1290 		classdev = device_create(&macvtap_class, &dev->dev, devt,
1291 					 dev, tap_name);
1292 		if (IS_ERR(classdev)) {
1293 			macvtap_free_minor(vlan);
1294 			return notifier_from_errno(PTR_ERR(classdev));
1295 		}
1296 		err = sysfs_create_link(&dev->dev.kobj, &classdev->kobj,
1297 					tap_name);
1298 		if (err)
1299 			return notifier_from_errno(err);
1300 		break;
1301 	case NETDEV_UNREGISTER:
1302 		/* vlan->minor == 0 if NETDEV_REGISTER above failed */
1303 		if (vlan->minor == 0)
1304 			break;
1305 		sysfs_remove_link(&dev->dev.kobj, tap_name);
1306 		devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
1307 		device_destroy(&macvtap_class, devt);
1308 		macvtap_free_minor(vlan);
1309 		break;
1310 	case NETDEV_CHANGE_TX_QUEUE_LEN:
1311 		if (macvtap_queue_resize(vlan))
1312 			return NOTIFY_BAD;
1313 		break;
1314 	}
1315 
1316 	return NOTIFY_DONE;
1317 }
1318 
1319 static struct notifier_block macvtap_notifier_block __read_mostly = {
1320 	.notifier_call	= macvtap_device_event,
1321 };
1322 
1323 static int macvtap_init(void)
1324 {
1325 	int err;
1326 
1327 	err = alloc_chrdev_region(&macvtap_major, 0,
1328 				MACVTAP_NUM_DEVS, "macvtap");
1329 	if (err)
1330 		goto out1;
1331 
1332 	cdev_init(&macvtap_cdev, &macvtap_fops);
1333 	err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS);
1334 	if (err)
1335 		goto out2;
1336 
1337 	err = class_register(&macvtap_class);
1338 	if (err)
1339 		goto out3;
1340 
1341 	err = register_netdevice_notifier(&macvtap_notifier_block);
1342 	if (err)
1343 		goto out4;
1344 
1345 	err = macvlan_link_register(&macvtap_link_ops);
1346 	if (err)
1347 		goto out5;
1348 
1349 	return 0;
1350 
1351 out5:
1352 	unregister_netdevice_notifier(&macvtap_notifier_block);
1353 out4:
1354 	class_unregister(&macvtap_class);
1355 out3:
1356 	cdev_del(&macvtap_cdev);
1357 out2:
1358 	unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1359 out1:
1360 	return err;
1361 }
1362 module_init(macvtap_init);
1363 
1364 static void macvtap_exit(void)
1365 {
1366 	rtnl_link_unregister(&macvtap_link_ops);
1367 	unregister_netdevice_notifier(&macvtap_notifier_block);
1368 	class_unregister(&macvtap_class);
1369 	cdev_del(&macvtap_cdev);
1370 	unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1371 	idr_destroy(&minor_idr);
1372 }
1373 module_exit(macvtap_exit);
1374 
1375 MODULE_ALIAS_RTNL_LINK("macvtap");
1376 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1377 MODULE_LICENSE("GPL");
1378