xref: /openbmc/linux/drivers/net/macvtap.c (revision 4800cd83)
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  */
37 struct macvtap_queue {
38 	struct sock sk;
39 	struct socket sock;
40 	struct socket_wq wq;
41 	int vnet_hdr_sz;
42 	struct macvlan_dev *vlan;
43 	struct file *file;
44 	unsigned int flags;
45 };
46 
47 static 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  */
61 static dev_t macvtap_major;
62 #define MACVTAP_NUM_DEVS 65536
63 static struct class *macvtap_class;
64 static struct cdev macvtap_cdev;
65 
66 static const struct proto_ops macvtap_socket_ops;
67 
68 /*
69  * RCU usage:
70  * The macvtap_queue and the macvlan_dev are loosely coupled, the
71  * pointers from one to the other can only be read while rcu_read_lock
72  * or macvtap_lock is held.
73  *
74  * Both the file and the macvlan_dev hold a reference on the macvtap_queue
75  * through sock_hold(&q->sk). When the macvlan_dev goes away first,
76  * q->vlan becomes inaccessible. When the files gets closed,
77  * macvtap_get_queue() fails.
78  *
79  * There may still be references to the struct sock inside of the
80  * queue from outbound SKBs, but these never reference back to the
81  * file or the dev. The data structure is freed through __sk_free
82  * when both our references and any pending SKBs are gone.
83  */
84 static DEFINE_SPINLOCK(macvtap_lock);
85 
86 /*
87  * get_slot: return a [unused/occupied] slot in vlan->taps[]:
88  *	- if 'q' is NULL, return the first empty slot;
89  *	- otherwise, return the slot this pointer occupies.
90  */
91 static int get_slot(struct macvlan_dev *vlan, struct macvtap_queue *q)
92 {
93 	int i;
94 
95 	for (i = 0; i < MAX_MACVTAP_QUEUES; i++) {
96 		if (rcu_dereference(vlan->taps[i]) == q)
97 			return i;
98 	}
99 
100 	/* Should never happen */
101 	BUG_ON(1);
102 }
103 
104 static int macvtap_set_queue(struct net_device *dev, struct file *file,
105 				struct macvtap_queue *q)
106 {
107 	struct macvlan_dev *vlan = netdev_priv(dev);
108 	int index;
109 	int err = -EBUSY;
110 
111 	spin_lock(&macvtap_lock);
112 	if (vlan->numvtaps == MAX_MACVTAP_QUEUES)
113 		goto out;
114 
115 	err = 0;
116 	index = get_slot(vlan, NULL);
117 	rcu_assign_pointer(q->vlan, vlan);
118 	rcu_assign_pointer(vlan->taps[index], q);
119 	sock_hold(&q->sk);
120 
121 	q->file = file;
122 	file->private_data = q;
123 
124 	vlan->numvtaps++;
125 
126 out:
127 	spin_unlock(&macvtap_lock);
128 	return err;
129 }
130 
131 /*
132  * The file owning the queue got closed, give up both
133  * the reference that the files holds as well as the
134  * one from the macvlan_dev if that still exists.
135  *
136  * Using the spinlock makes sure that we don't get
137  * to the queue again after destroying it.
138  */
139 static void macvtap_put_queue(struct macvtap_queue *q)
140 {
141 	struct macvlan_dev *vlan;
142 
143 	spin_lock(&macvtap_lock);
144 	vlan = rcu_dereference(q->vlan);
145 	if (vlan) {
146 		int index = get_slot(vlan, q);
147 
148 		rcu_assign_pointer(vlan->taps[index], NULL);
149 		rcu_assign_pointer(q->vlan, NULL);
150 		sock_put(&q->sk);
151 		--vlan->numvtaps;
152 	}
153 
154 	spin_unlock(&macvtap_lock);
155 
156 	synchronize_rcu();
157 	sock_put(&q->sk);
158 }
159 
160 /*
161  * Select a queue based on the rxq of the device on which this packet
162  * arrived. If the incoming device is not mq, calculate a flow hash
163  * to select a queue. If all fails, find the first available queue.
164  * Cache vlan->numvtaps since it can become zero during the execution
165  * of this function.
166  */
167 static struct macvtap_queue *macvtap_get_queue(struct net_device *dev,
168 					       struct sk_buff *skb)
169 {
170 	struct macvlan_dev *vlan = netdev_priv(dev);
171 	struct macvtap_queue *tap = NULL;
172 	int numvtaps = vlan->numvtaps;
173 	__u32 rxq;
174 
175 	if (!numvtaps)
176 		goto out;
177 
178 	if (likely(skb_rx_queue_recorded(skb))) {
179 		rxq = skb_get_rx_queue(skb);
180 
181 		while (unlikely(rxq >= numvtaps))
182 			rxq -= numvtaps;
183 
184 		tap = rcu_dereference(vlan->taps[rxq]);
185 		if (tap)
186 			goto out;
187 	}
188 
189 	/* Check if we can use flow to select a queue */
190 	rxq = skb_get_rxhash(skb);
191 	if (rxq) {
192 		tap = rcu_dereference(vlan->taps[rxq % numvtaps]);
193 		if (tap)
194 			goto out;
195 	}
196 
197 	/* Everything failed - find first available queue */
198 	for (rxq = 0; rxq < MAX_MACVTAP_QUEUES; rxq++) {
199 		tap = rcu_dereference(vlan->taps[rxq]);
200 		if (tap)
201 			break;
202 	}
203 
204 out:
205 	return tap;
206 }
207 
208 /*
209  * The net_device is going away, give up the reference
210  * that it holds on all queues and safely set the pointer
211  * from the queues to NULL.
212  */
213 static void macvtap_del_queues(struct net_device *dev)
214 {
215 	struct macvlan_dev *vlan = netdev_priv(dev);
216 	struct macvtap_queue *q, *qlist[MAX_MACVTAP_QUEUES];
217 	int i, j = 0;
218 
219 	/* macvtap_put_queue can free some slots, so go through all slots */
220 	spin_lock(&macvtap_lock);
221 	for (i = 0; i < MAX_MACVTAP_QUEUES && vlan->numvtaps; i++) {
222 		q = rcu_dereference(vlan->taps[i]);
223 		if (q) {
224 			qlist[j++] = q;
225 			rcu_assign_pointer(vlan->taps[i], NULL);
226 			rcu_assign_pointer(q->vlan, NULL);
227 			vlan->numvtaps--;
228 		}
229 	}
230 	BUG_ON(vlan->numvtaps != 0);
231 	spin_unlock(&macvtap_lock);
232 
233 	synchronize_rcu();
234 
235 	for (--j; j >= 0; j--)
236 		sock_put(&qlist[j]->sk);
237 }
238 
239 /*
240  * Forward happens for data that gets sent from one macvlan
241  * endpoint to another one in bridge mode. We just take
242  * the skb and put it into the receive queue.
243  */
244 static int macvtap_forward(struct net_device *dev, struct sk_buff *skb)
245 {
246 	struct macvtap_queue *q = macvtap_get_queue(dev, skb);
247 	if (!q)
248 		goto drop;
249 
250 	if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len)
251 		goto drop;
252 
253 	skb_queue_tail(&q->sk.sk_receive_queue, skb);
254 	wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND);
255 	return NET_RX_SUCCESS;
256 
257 drop:
258 	kfree_skb(skb);
259 	return NET_RX_DROP;
260 }
261 
262 /*
263  * Receive is for data from the external interface (lowerdev),
264  * in case of macvtap, we can treat that the same way as
265  * forward, which macvlan cannot.
266  */
267 static int macvtap_receive(struct sk_buff *skb)
268 {
269 	skb_push(skb, ETH_HLEN);
270 	return macvtap_forward(skb->dev, skb);
271 }
272 
273 static int macvtap_newlink(struct net *src_net,
274 			   struct net_device *dev,
275 			   struct nlattr *tb[],
276 			   struct nlattr *data[])
277 {
278 	struct device *classdev;
279 	dev_t devt;
280 	int err;
281 
282 	err = macvlan_common_newlink(src_net, dev, tb, data,
283 				     macvtap_receive, macvtap_forward);
284 	if (err)
285 		goto out;
286 
287 	devt = MKDEV(MAJOR(macvtap_major), dev->ifindex);
288 
289 	classdev = device_create(macvtap_class, &dev->dev, devt,
290 				 dev, "tap%d", dev->ifindex);
291 	if (IS_ERR(classdev)) {
292 		err = PTR_ERR(classdev);
293 		macvtap_del_queues(dev);
294 	}
295 
296 out:
297 	return err;
298 }
299 
300 static void macvtap_dellink(struct net_device *dev,
301 			    struct list_head *head)
302 {
303 	device_destroy(macvtap_class,
304 		       MKDEV(MAJOR(macvtap_major), dev->ifindex));
305 
306 	macvtap_del_queues(dev);
307 	macvlan_dellink(dev, head);
308 }
309 
310 static void macvtap_setup(struct net_device *dev)
311 {
312 	macvlan_common_setup(dev);
313 	dev->tx_queue_len = TUN_READQ_SIZE;
314 }
315 
316 static struct rtnl_link_ops macvtap_link_ops __read_mostly = {
317 	.kind		= "macvtap",
318 	.setup		= macvtap_setup,
319 	.newlink	= macvtap_newlink,
320 	.dellink	= macvtap_dellink,
321 };
322 
323 
324 static void macvtap_sock_write_space(struct sock *sk)
325 {
326 	wait_queue_head_t *wqueue;
327 
328 	if (!sock_writeable(sk) ||
329 	    !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
330 		return;
331 
332 	wqueue = sk_sleep(sk);
333 	if (wqueue && waitqueue_active(wqueue))
334 		wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND);
335 }
336 
337 static int macvtap_open(struct inode *inode, struct file *file)
338 {
339 	struct net *net = current->nsproxy->net_ns;
340 	struct net_device *dev = dev_get_by_index(net, iminor(inode));
341 	struct macvtap_queue *q;
342 	int err;
343 
344 	err = -ENODEV;
345 	if (!dev)
346 		goto out;
347 
348 	/* check if this is a macvtap device */
349 	err = -EINVAL;
350 	if (dev->rtnl_link_ops != &macvtap_link_ops)
351 		goto out;
352 
353 	err = -ENOMEM;
354 	q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
355 					     &macvtap_proto);
356 	if (!q)
357 		goto out;
358 
359 	q->sock.wq = &q->wq;
360 	init_waitqueue_head(&q->wq.wait);
361 	q->sock.type = SOCK_RAW;
362 	q->sock.state = SS_CONNECTED;
363 	q->sock.file = file;
364 	q->sock.ops = &macvtap_socket_ops;
365 	sock_init_data(&q->sock, &q->sk);
366 	q->sk.sk_write_space = macvtap_sock_write_space;
367 	q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
368 	q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
369 
370 	err = macvtap_set_queue(dev, file, q);
371 	if (err)
372 		sock_put(&q->sk);
373 
374 out:
375 	if (dev)
376 		dev_put(dev);
377 
378 	return err;
379 }
380 
381 static int macvtap_release(struct inode *inode, struct file *file)
382 {
383 	struct macvtap_queue *q = file->private_data;
384 	macvtap_put_queue(q);
385 	return 0;
386 }
387 
388 static unsigned int macvtap_poll(struct file *file, poll_table * wait)
389 {
390 	struct macvtap_queue *q = file->private_data;
391 	unsigned int mask = POLLERR;
392 
393 	if (!q)
394 		goto out;
395 
396 	mask = 0;
397 	poll_wait(file, &q->wq.wait, wait);
398 
399 	if (!skb_queue_empty(&q->sk.sk_receive_queue))
400 		mask |= POLLIN | POLLRDNORM;
401 
402 	if (sock_writeable(&q->sk) ||
403 	    (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) &&
404 	     sock_writeable(&q->sk)))
405 		mask |= POLLOUT | POLLWRNORM;
406 
407 out:
408 	return mask;
409 }
410 
411 static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad,
412 						size_t len, size_t linear,
413 						int noblock, int *err)
414 {
415 	struct sk_buff *skb;
416 
417 	/* Under a page?  Don't bother with paged skb. */
418 	if (prepad + len < PAGE_SIZE || !linear)
419 		linear = len;
420 
421 	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
422 				   err);
423 	if (!skb)
424 		return NULL;
425 
426 	skb_reserve(skb, prepad);
427 	skb_put(skb, linear);
428 	skb->data_len = len - linear;
429 	skb->len += len - linear;
430 
431 	return skb;
432 }
433 
434 /*
435  * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
436  * be shared with the tun/tap driver.
437  */
438 static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb,
439 				     struct virtio_net_hdr *vnet_hdr)
440 {
441 	unsigned short gso_type = 0;
442 	if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
443 		switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
444 		case VIRTIO_NET_HDR_GSO_TCPV4:
445 			gso_type = SKB_GSO_TCPV4;
446 			break;
447 		case VIRTIO_NET_HDR_GSO_TCPV6:
448 			gso_type = SKB_GSO_TCPV6;
449 			break;
450 		case VIRTIO_NET_HDR_GSO_UDP:
451 			gso_type = SKB_GSO_UDP;
452 			break;
453 		default:
454 			return -EINVAL;
455 		}
456 
457 		if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
458 			gso_type |= SKB_GSO_TCP_ECN;
459 
460 		if (vnet_hdr->gso_size == 0)
461 			return -EINVAL;
462 	}
463 
464 	if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
465 		if (!skb_partial_csum_set(skb, vnet_hdr->csum_start,
466 					  vnet_hdr->csum_offset))
467 			return -EINVAL;
468 	}
469 
470 	if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
471 		skb_shinfo(skb)->gso_size = vnet_hdr->gso_size;
472 		skb_shinfo(skb)->gso_type = gso_type;
473 
474 		/* Header must be checked, and gso_segs computed. */
475 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
476 		skb_shinfo(skb)->gso_segs = 0;
477 	}
478 	return 0;
479 }
480 
481 static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb,
482 				   struct virtio_net_hdr *vnet_hdr)
483 {
484 	memset(vnet_hdr, 0, sizeof(*vnet_hdr));
485 
486 	if (skb_is_gso(skb)) {
487 		struct skb_shared_info *sinfo = skb_shinfo(skb);
488 
489 		/* This is a hint as to how much should be linear. */
490 		vnet_hdr->hdr_len = skb_headlen(skb);
491 		vnet_hdr->gso_size = sinfo->gso_size;
492 		if (sinfo->gso_type & SKB_GSO_TCPV4)
493 			vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
494 		else if (sinfo->gso_type & SKB_GSO_TCPV6)
495 			vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
496 		else if (sinfo->gso_type & SKB_GSO_UDP)
497 			vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
498 		else
499 			BUG();
500 		if (sinfo->gso_type & SKB_GSO_TCP_ECN)
501 			vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
502 	} else
503 		vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
504 
505 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
506 		vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
507 		vnet_hdr->csum_start = skb_checksum_start_offset(skb);
508 		vnet_hdr->csum_offset = skb->csum_offset;
509 	} /* else everything is zero */
510 
511 	return 0;
512 }
513 
514 
515 /* Get packet from user space buffer */
516 static ssize_t macvtap_get_user(struct macvtap_queue *q,
517 				const struct iovec *iv, size_t count,
518 				int noblock)
519 {
520 	struct sk_buff *skb;
521 	struct macvlan_dev *vlan;
522 	size_t len = count;
523 	int err;
524 	struct virtio_net_hdr vnet_hdr = { 0 };
525 	int vnet_hdr_len = 0;
526 
527 	if (q->flags & IFF_VNET_HDR) {
528 		vnet_hdr_len = q->vnet_hdr_sz;
529 
530 		err = -EINVAL;
531 		if (len < vnet_hdr_len)
532 			goto err;
533 		len -= vnet_hdr_len;
534 
535 		err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0,
536 					   sizeof(vnet_hdr));
537 		if (err < 0)
538 			goto err;
539 		if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
540 		     vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
541 							vnet_hdr.hdr_len)
542 			vnet_hdr.hdr_len = vnet_hdr.csum_start +
543 						vnet_hdr.csum_offset + 2;
544 		err = -EINVAL;
545 		if (vnet_hdr.hdr_len > len)
546 			goto err;
547 	}
548 
549 	err = -EINVAL;
550 	if (unlikely(len < ETH_HLEN))
551 		goto err;
552 
553 	skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, len, vnet_hdr.hdr_len,
554 				noblock, &err);
555 	if (!skb)
556 		goto err;
557 
558 	err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len, len);
559 	if (err)
560 		goto err_kfree;
561 
562 	skb_set_network_header(skb, ETH_HLEN);
563 	skb_reset_mac_header(skb);
564 	skb->protocol = eth_hdr(skb)->h_proto;
565 
566 	if (vnet_hdr_len) {
567 		err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr);
568 		if (err)
569 			goto err_kfree;
570 	}
571 
572 	rcu_read_lock_bh();
573 	vlan = rcu_dereference(q->vlan);
574 	if (vlan)
575 		macvlan_start_xmit(skb, vlan->dev);
576 	else
577 		kfree_skb(skb);
578 	rcu_read_unlock_bh();
579 
580 	return count;
581 
582 err_kfree:
583 	kfree_skb(skb);
584 
585 err:
586 	rcu_read_lock_bh();
587 	vlan = rcu_dereference(q->vlan);
588 	if (vlan)
589 		vlan->dev->stats.tx_dropped++;
590 	rcu_read_unlock_bh();
591 
592 	return err;
593 }
594 
595 static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv,
596 				 unsigned long count, loff_t pos)
597 {
598 	struct file *file = iocb->ki_filp;
599 	ssize_t result = -ENOLINK;
600 	struct macvtap_queue *q = file->private_data;
601 
602 	result = macvtap_get_user(q, iv, iov_length(iv, count),
603 			      file->f_flags & O_NONBLOCK);
604 	return result;
605 }
606 
607 /* Put packet to the user space buffer */
608 static ssize_t macvtap_put_user(struct macvtap_queue *q,
609 				const struct sk_buff *skb,
610 				const struct iovec *iv, int len)
611 {
612 	struct macvlan_dev *vlan;
613 	int ret;
614 	int vnet_hdr_len = 0;
615 
616 	if (q->flags & IFF_VNET_HDR) {
617 		struct virtio_net_hdr vnet_hdr;
618 		vnet_hdr_len = q->vnet_hdr_sz;
619 		if ((len -= vnet_hdr_len) < 0)
620 			return -EINVAL;
621 
622 		ret = macvtap_skb_to_vnet_hdr(skb, &vnet_hdr);
623 		if (ret)
624 			return ret;
625 
626 		if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, sizeof(vnet_hdr)))
627 			return -EFAULT;
628 	}
629 
630 	len = min_t(int, skb->len, len);
631 
632 	ret = skb_copy_datagram_const_iovec(skb, 0, iv, vnet_hdr_len, len);
633 
634 	rcu_read_lock_bh();
635 	vlan = rcu_dereference(q->vlan);
636 	if (vlan)
637 		macvlan_count_rx(vlan, len, ret == 0, 0);
638 	rcu_read_unlock_bh();
639 
640 	return ret ? ret : (len + vnet_hdr_len);
641 }
642 
643 static ssize_t macvtap_do_read(struct macvtap_queue *q, struct kiocb *iocb,
644 			       const struct iovec *iv, unsigned long len,
645 			       int noblock)
646 {
647 	DECLARE_WAITQUEUE(wait, current);
648 	struct sk_buff *skb;
649 	ssize_t ret = 0;
650 
651 	add_wait_queue(sk_sleep(&q->sk), &wait);
652 	while (len) {
653 		current->state = TASK_INTERRUPTIBLE;
654 
655 		/* Read frames from the queue */
656 		skb = skb_dequeue(&q->sk.sk_receive_queue);
657 		if (!skb) {
658 			if (noblock) {
659 				ret = -EAGAIN;
660 				break;
661 			}
662 			if (signal_pending(current)) {
663 				ret = -ERESTARTSYS;
664 				break;
665 			}
666 			/* Nothing to read, let's sleep */
667 			schedule();
668 			continue;
669 		}
670 		ret = macvtap_put_user(q, skb, iv, len);
671 		kfree_skb(skb);
672 		break;
673 	}
674 
675 	current->state = TASK_RUNNING;
676 	remove_wait_queue(sk_sleep(&q->sk), &wait);
677 	return ret;
678 }
679 
680 static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv,
681 				unsigned long count, loff_t pos)
682 {
683 	struct file *file = iocb->ki_filp;
684 	struct macvtap_queue *q = file->private_data;
685 	ssize_t len, ret = 0;
686 
687 	len = iov_length(iv, count);
688 	if (len < 0) {
689 		ret = -EINVAL;
690 		goto out;
691 	}
692 
693 	ret = macvtap_do_read(q, iocb, iv, len, file->f_flags & O_NONBLOCK);
694 	ret = min_t(ssize_t, ret, len); /* XXX copied from tun.c. Why? */
695 out:
696 	return ret;
697 }
698 
699 /*
700  * provide compatibility with generic tun/tap interface
701  */
702 static long macvtap_ioctl(struct file *file, unsigned int cmd,
703 			  unsigned long arg)
704 {
705 	struct macvtap_queue *q = file->private_data;
706 	struct macvlan_dev *vlan;
707 	void __user *argp = (void __user *)arg;
708 	struct ifreq __user *ifr = argp;
709 	unsigned int __user *up = argp;
710 	unsigned int u;
711 	int __user *sp = argp;
712 	int s;
713 	int ret;
714 
715 	switch (cmd) {
716 	case TUNSETIFF:
717 		/* ignore the name, just look at flags */
718 		if (get_user(u, &ifr->ifr_flags))
719 			return -EFAULT;
720 
721 		ret = 0;
722 		if ((u & ~IFF_VNET_HDR) != (IFF_NO_PI | IFF_TAP))
723 			ret = -EINVAL;
724 		else
725 			q->flags = u;
726 
727 		return ret;
728 
729 	case TUNGETIFF:
730 		rcu_read_lock_bh();
731 		vlan = rcu_dereference(q->vlan);
732 		if (vlan)
733 			dev_hold(vlan->dev);
734 		rcu_read_unlock_bh();
735 
736 		if (!vlan)
737 			return -ENOLINK;
738 
739 		ret = 0;
740 		if (copy_to_user(&ifr->ifr_name, q->vlan->dev->name, IFNAMSIZ) ||
741 		    put_user(q->flags, &ifr->ifr_flags))
742 			ret = -EFAULT;
743 		dev_put(vlan->dev);
744 		return ret;
745 
746 	case TUNGETFEATURES:
747 		if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR, up))
748 			return -EFAULT;
749 		return 0;
750 
751 	case TUNSETSNDBUF:
752 		if (get_user(u, up))
753 			return -EFAULT;
754 
755 		q->sk.sk_sndbuf = u;
756 		return 0;
757 
758 	case TUNGETVNETHDRSZ:
759 		s = q->vnet_hdr_sz;
760 		if (put_user(s, sp))
761 			return -EFAULT;
762 		return 0;
763 
764 	case TUNSETVNETHDRSZ:
765 		if (get_user(s, sp))
766 			return -EFAULT;
767 		if (s < (int)sizeof(struct virtio_net_hdr))
768 			return -EINVAL;
769 
770 		q->vnet_hdr_sz = s;
771 		return 0;
772 
773 	case TUNSETOFFLOAD:
774 		/* let the user check for future flags */
775 		if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
776 			    TUN_F_TSO_ECN | TUN_F_UFO))
777 			return -EINVAL;
778 
779 		/* TODO: only accept frames with the features that
780 			 got enabled for forwarded frames */
781 		if (!(q->flags & IFF_VNET_HDR))
782 			return  -EINVAL;
783 		return 0;
784 
785 	default:
786 		return -EINVAL;
787 	}
788 }
789 
790 #ifdef CONFIG_COMPAT
791 static long macvtap_compat_ioctl(struct file *file, unsigned int cmd,
792 				 unsigned long arg)
793 {
794 	return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
795 }
796 #endif
797 
798 static const struct file_operations macvtap_fops = {
799 	.owner		= THIS_MODULE,
800 	.open		= macvtap_open,
801 	.release	= macvtap_release,
802 	.aio_read	= macvtap_aio_read,
803 	.aio_write	= macvtap_aio_write,
804 	.poll		= macvtap_poll,
805 	.llseek		= no_llseek,
806 	.unlocked_ioctl	= macvtap_ioctl,
807 #ifdef CONFIG_COMPAT
808 	.compat_ioctl	= macvtap_compat_ioctl,
809 #endif
810 };
811 
812 static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock,
813 			   struct msghdr *m, size_t total_len)
814 {
815 	struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
816 	return macvtap_get_user(q, m->msg_iov, total_len,
817 			    m->msg_flags & MSG_DONTWAIT);
818 }
819 
820 static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock,
821 			   struct msghdr *m, size_t total_len,
822 			   int flags)
823 {
824 	struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
825 	int ret;
826 	if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
827 		return -EINVAL;
828 	ret = macvtap_do_read(q, iocb, m->msg_iov, total_len,
829 			  flags & MSG_DONTWAIT);
830 	if (ret > total_len) {
831 		m->msg_flags |= MSG_TRUNC;
832 		ret = flags & MSG_TRUNC ? ret : total_len;
833 	}
834 	return ret;
835 }
836 
837 /* Ops structure to mimic raw sockets with tun */
838 static const struct proto_ops macvtap_socket_ops = {
839 	.sendmsg = macvtap_sendmsg,
840 	.recvmsg = macvtap_recvmsg,
841 };
842 
843 /* Get an underlying socket object from tun file.  Returns error unless file is
844  * attached to a device.  The returned object works like a packet socket, it
845  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
846  * holding a reference to the file for as long as the socket is in use. */
847 struct socket *macvtap_get_socket(struct file *file)
848 {
849 	struct macvtap_queue *q;
850 	if (file->f_op != &macvtap_fops)
851 		return ERR_PTR(-EINVAL);
852 	q = file->private_data;
853 	if (!q)
854 		return ERR_PTR(-EBADFD);
855 	return &q->sock;
856 }
857 EXPORT_SYMBOL_GPL(macvtap_get_socket);
858 
859 static int macvtap_init(void)
860 {
861 	int err;
862 
863 	err = alloc_chrdev_region(&macvtap_major, 0,
864 				MACVTAP_NUM_DEVS, "macvtap");
865 	if (err)
866 		goto out1;
867 
868 	cdev_init(&macvtap_cdev, &macvtap_fops);
869 	err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS);
870 	if (err)
871 		goto out2;
872 
873 	macvtap_class = class_create(THIS_MODULE, "macvtap");
874 	if (IS_ERR(macvtap_class)) {
875 		err = PTR_ERR(macvtap_class);
876 		goto out3;
877 	}
878 
879 	err = macvlan_link_register(&macvtap_link_ops);
880 	if (err)
881 		goto out4;
882 
883 	return 0;
884 
885 out4:
886 	class_unregister(macvtap_class);
887 out3:
888 	cdev_del(&macvtap_cdev);
889 out2:
890 	unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
891 out1:
892 	return err;
893 }
894 module_init(macvtap_init);
895 
896 static void macvtap_exit(void)
897 {
898 	rtnl_link_unregister(&macvtap_link_ops);
899 	class_unregister(macvtap_class);
900 	cdev_del(&macvtap_cdev);
901 	unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
902 }
903 module_exit(macvtap_exit);
904 
905 MODULE_ALIAS_RTNL_LINK("macvtap");
906 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
907 MODULE_LICENSE("GPL");
908