Loading...
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");
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
25/*
26 * A macvtap queue is the central object of this driver, it connects
27 * an open character device to a macvlan interface. There can be
28 * multiple queues on one interface, which map back to queues
29 * implemented in hardware on the underlying device.
30 *
31 * macvtap_proto is used to allocate queues through the sock allocation
32 * mechanism.
33 *
34 */
35struct macvtap_queue {
36 struct sock sk;
37 struct socket sock;
38 struct socket_wq wq;
39 int vnet_hdr_sz;
40 struct macvlan_dev __rcu *vlan;
41 struct file *file;
42 unsigned int flags;
43 u16 queue_index;
44 bool enabled;
45 struct list_head next;
46};
47
48#define MACVTAP_FEATURES (IFF_VNET_HDR | IFF_MULTI_QUEUE)
49
50#define MACVTAP_VNET_LE 0x80000000
51#define MACVTAP_VNET_BE 0x40000000
52
53#ifdef CONFIG_TUN_VNET_CROSS_LE
54static inline bool macvtap_legacy_is_little_endian(struct macvtap_queue *q)
55{
56 return q->flags & MACVTAP_VNET_BE ? false :
57 virtio_legacy_is_little_endian();
58}
59
60static long macvtap_get_vnet_be(struct macvtap_queue *q, int __user *sp)
61{
62 int s = !!(q->flags & MACVTAP_VNET_BE);
63
64 if (put_user(s, sp))
65 return -EFAULT;
66
67 return 0;
68}
69
70static long macvtap_set_vnet_be(struct macvtap_queue *q, int __user *sp)
71{
72 int s;
73
74 if (get_user(s, sp))
75 return -EFAULT;
76
77 if (s)
78 q->flags |= MACVTAP_VNET_BE;
79 else
80 q->flags &= ~MACVTAP_VNET_BE;
81
82 return 0;
83}
84#else
85static inline bool macvtap_legacy_is_little_endian(struct macvtap_queue *q)
86{
87 return virtio_legacy_is_little_endian();
88}
89
90static long macvtap_get_vnet_be(struct macvtap_queue *q, int __user *argp)
91{
92 return -EINVAL;
93}
94
95static long macvtap_set_vnet_be(struct macvtap_queue *q, int __user *argp)
96{
97 return -EINVAL;
98}
99#endif /* CONFIG_TUN_VNET_CROSS_LE */
100
101static inline bool macvtap_is_little_endian(struct macvtap_queue *q)
102{
103 return q->flags & MACVTAP_VNET_LE ||
104 macvtap_legacy_is_little_endian(q);
105}
106
107static inline u16 macvtap16_to_cpu(struct macvtap_queue *q, __virtio16 val)
108{
109 return __virtio16_to_cpu(macvtap_is_little_endian(q), val);
110}
111
112static inline __virtio16 cpu_to_macvtap16(struct macvtap_queue *q, u16 val)
113{
114 return __cpu_to_virtio16(macvtap_is_little_endian(q), val);
115}
116
117static struct proto macvtap_proto = {
118 .name = "macvtap",
119 .owner = THIS_MODULE,
120 .obj_size = sizeof (struct macvtap_queue),
121};
122
123/*
124 * Variables for dealing with macvtaps device numbers.
125 */
126static dev_t macvtap_major;
127#define MACVTAP_NUM_DEVS (1U << MINORBITS)
128static DEFINE_MUTEX(minor_lock);
129static DEFINE_IDR(minor_idr);
130
131#define GOODCOPY_LEN 128
132static struct class *macvtap_class;
133static struct cdev macvtap_cdev;
134
135static const struct proto_ops macvtap_socket_ops;
136
137#define TUN_OFFLOADS (NETIF_F_HW_CSUM | NETIF_F_TSO_ECN | NETIF_F_TSO | \
138 NETIF_F_TSO6 | NETIF_F_UFO)
139#define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO)
140#define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG | NETIF_F_FRAGLIST)
141
142static struct macvlan_dev *macvtap_get_vlan_rcu(const struct net_device *dev)
143{
144 return rcu_dereference(dev->rx_handler_data);
145}
146
147/*
148 * RCU usage:
149 * The macvtap_queue and the macvlan_dev are loosely coupled, the
150 * pointers from one to the other can only be read while rcu_read_lock
151 * or rtnl is held.
152 *
153 * Both the file and the macvlan_dev hold a reference on the macvtap_queue
154 * through sock_hold(&q->sk). When the macvlan_dev goes away first,
155 * q->vlan becomes inaccessible. When the files gets closed,
156 * macvtap_get_queue() fails.
157 *
158 * There may still be references to the struct sock inside of the
159 * queue from outbound SKBs, but these never reference back to the
160 * file or the dev. The data structure is freed through __sk_free
161 * when both our references and any pending SKBs are gone.
162 */
163
164static int macvtap_enable_queue(struct net_device *dev, struct file *file,
165 struct macvtap_queue *q)
166{
167 struct macvlan_dev *vlan = netdev_priv(dev);
168 int err = -EINVAL;
169
170 ASSERT_RTNL();
171
172 if (q->enabled)
173 goto out;
174
175 err = 0;
176 rcu_assign_pointer(vlan->taps[vlan->numvtaps], q);
177 q->queue_index = vlan->numvtaps;
178 q->enabled = true;
179
180 vlan->numvtaps++;
181out:
182 return err;
183}
184
185/* Requires RTNL */
186static int macvtap_set_queue(struct net_device *dev, struct file *file,
187 struct macvtap_queue *q)
188{
189 struct macvlan_dev *vlan = netdev_priv(dev);
190
191 if (vlan->numqueues == MAX_MACVTAP_QUEUES)
192 return -EBUSY;
193
194 rcu_assign_pointer(q->vlan, vlan);
195 rcu_assign_pointer(vlan->taps[vlan->numvtaps], q);
196 sock_hold(&q->sk);
197
198 q->file = file;
199 q->queue_index = vlan->numvtaps;
200 q->enabled = true;
201 file->private_data = q;
202 list_add_tail(&q->next, &vlan->queue_list);
203
204 vlan->numvtaps++;
205 vlan->numqueues++;
206
207 return 0;
208}
209
210static int macvtap_disable_queue(struct macvtap_queue *q)
211{
212 struct macvlan_dev *vlan;
213 struct macvtap_queue *nq;
214
215 ASSERT_RTNL();
216 if (!q->enabled)
217 return -EINVAL;
218
219 vlan = rtnl_dereference(q->vlan);
220
221 if (vlan) {
222 int index = q->queue_index;
223 BUG_ON(index >= vlan->numvtaps);
224 nq = rtnl_dereference(vlan->taps[vlan->numvtaps - 1]);
225 nq->queue_index = index;
226
227 rcu_assign_pointer(vlan->taps[index], nq);
228 RCU_INIT_POINTER(vlan->taps[vlan->numvtaps - 1], NULL);
229 q->enabled = false;
230
231 vlan->numvtaps--;
232 }
233
234 return 0;
235}
236
237/*
238 * The file owning the queue got closed, give up both
239 * the reference that the files holds as well as the
240 * one from the macvlan_dev if that still exists.
241 *
242 * Using the spinlock makes sure that we don't get
243 * to the queue again after destroying it.
244 */
245static void macvtap_put_queue(struct macvtap_queue *q)
246{
247 struct macvlan_dev *vlan;
248
249 rtnl_lock();
250 vlan = rtnl_dereference(q->vlan);
251
252 if (vlan) {
253 if (q->enabled)
254 BUG_ON(macvtap_disable_queue(q));
255
256 vlan->numqueues--;
257 RCU_INIT_POINTER(q->vlan, NULL);
258 sock_put(&q->sk);
259 list_del_init(&q->next);
260 }
261
262 rtnl_unlock();
263
264 synchronize_rcu();
265 sock_put(&q->sk);
266}
267
268/*
269 * Select a queue based on the rxq of the device on which this packet
270 * arrived. If the incoming device is not mq, calculate a flow hash
271 * to select a queue. If all fails, find the first available queue.
272 * Cache vlan->numvtaps since it can become zero during the execution
273 * of this function.
274 */
275static struct macvtap_queue *macvtap_get_queue(struct net_device *dev,
276 struct sk_buff *skb)
277{
278 struct macvlan_dev *vlan = netdev_priv(dev);
279 struct macvtap_queue *tap = NULL;
280 /* Access to taps array is protected by rcu, but access to numvtaps
281 * isn't. Below we use it to lookup a queue, but treat it as a hint
282 * and validate that the result isn't NULL - in case we are
283 * racing against queue removal.
284 */
285 int numvtaps = ACCESS_ONCE(vlan->numvtaps);
286 __u32 rxq;
287
288 if (!numvtaps)
289 goto out;
290
291 /* Check if we can use flow to select a queue */
292 rxq = skb_get_hash(skb);
293 if (rxq) {
294 tap = rcu_dereference(vlan->taps[rxq % numvtaps]);
295 goto out;
296 }
297
298 if (likely(skb_rx_queue_recorded(skb))) {
299 rxq = skb_get_rx_queue(skb);
300
301 while (unlikely(rxq >= numvtaps))
302 rxq -= numvtaps;
303
304 tap = rcu_dereference(vlan->taps[rxq]);
305 goto out;
306 }
307
308 tap = rcu_dereference(vlan->taps[0]);
309out:
310 return tap;
311}
312
313/*
314 * The net_device is going away, give up the reference
315 * that it holds on all queues and safely set the pointer
316 * from the queues to NULL.
317 */
318static void macvtap_del_queues(struct net_device *dev)
319{
320 struct macvlan_dev *vlan = netdev_priv(dev);
321 struct macvtap_queue *q, *tmp;
322
323 ASSERT_RTNL();
324 list_for_each_entry_safe(q, tmp, &vlan->queue_list, next) {
325 list_del_init(&q->next);
326 RCU_INIT_POINTER(q->vlan, NULL);
327 if (q->enabled)
328 vlan->numvtaps--;
329 vlan->numqueues--;
330 sock_put(&q->sk);
331 }
332 BUG_ON(vlan->numvtaps);
333 BUG_ON(vlan->numqueues);
334 /* guarantee that any future macvtap_set_queue will fail */
335 vlan->numvtaps = MAX_MACVTAP_QUEUES;
336}
337
338static rx_handler_result_t macvtap_handle_frame(struct sk_buff **pskb)
339{
340 struct sk_buff *skb = *pskb;
341 struct net_device *dev = skb->dev;
342 struct macvlan_dev *vlan;
343 struct macvtap_queue *q;
344 netdev_features_t features = TAP_FEATURES;
345
346 vlan = macvtap_get_vlan_rcu(dev);
347 if (!vlan)
348 return RX_HANDLER_PASS;
349
350 q = macvtap_get_queue(dev, skb);
351 if (!q)
352 return RX_HANDLER_PASS;
353
354 if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len)
355 goto drop;
356
357 skb_push(skb, ETH_HLEN);
358
359 /* Apply the forward feature mask so that we perform segmentation
360 * according to users wishes. This only works if VNET_HDR is
361 * enabled.
362 */
363 if (q->flags & IFF_VNET_HDR)
364 features |= vlan->tap_features;
365 if (netif_needs_gso(skb, features)) {
366 struct sk_buff *segs = __skb_gso_segment(skb, features, false);
367
368 if (IS_ERR(segs))
369 goto drop;
370
371 if (!segs) {
372 skb_queue_tail(&q->sk.sk_receive_queue, skb);
373 goto wake_up;
374 }
375
376 consume_skb(skb);
377 while (segs) {
378 struct sk_buff *nskb = segs->next;
379
380 segs->next = NULL;
381 skb_queue_tail(&q->sk.sk_receive_queue, segs);
382 segs = nskb;
383 }
384 } else {
385 /* If we receive a partial checksum and the tap side
386 * doesn't support checksum offload, compute the checksum.
387 * Note: it doesn't matter which checksum feature to
388 * check, we either support them all or none.
389 */
390 if (skb->ip_summed == CHECKSUM_PARTIAL &&
391 !(features & NETIF_F_CSUM_MASK) &&
392 skb_checksum_help(skb))
393 goto drop;
394 skb_queue_tail(&q->sk.sk_receive_queue, skb);
395 }
396
397wake_up:
398 wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND);
399 return RX_HANDLER_CONSUMED;
400
401drop:
402 /* Count errors/drops only here, thus don't care about args. */
403 macvlan_count_rx(vlan, 0, 0, 0);
404 kfree_skb(skb);
405 return RX_HANDLER_CONSUMED;
406}
407
408static int macvtap_get_minor(struct macvlan_dev *vlan)
409{
410 int retval = -ENOMEM;
411
412 mutex_lock(&minor_lock);
413 retval = idr_alloc(&minor_idr, vlan, 1, MACVTAP_NUM_DEVS, GFP_KERNEL);
414 if (retval >= 0) {
415 vlan->minor = retval;
416 } else if (retval == -ENOSPC) {
417 printk(KERN_ERR "too many macvtap devices\n");
418 retval = -EINVAL;
419 }
420 mutex_unlock(&minor_lock);
421 return retval < 0 ? retval : 0;
422}
423
424static void macvtap_free_minor(struct macvlan_dev *vlan)
425{
426 mutex_lock(&minor_lock);
427 if (vlan->minor) {
428 idr_remove(&minor_idr, vlan->minor);
429 vlan->minor = 0;
430 }
431 mutex_unlock(&minor_lock);
432}
433
434static struct net_device *dev_get_by_macvtap_minor(int minor)
435{
436 struct net_device *dev = NULL;
437 struct macvlan_dev *vlan;
438
439 mutex_lock(&minor_lock);
440 vlan = idr_find(&minor_idr, minor);
441 if (vlan) {
442 dev = vlan->dev;
443 dev_hold(dev);
444 }
445 mutex_unlock(&minor_lock);
446 return dev;
447}
448
449static int macvtap_newlink(struct net *src_net,
450 struct net_device *dev,
451 struct nlattr *tb[],
452 struct nlattr *data[])
453{
454 struct macvlan_dev *vlan = netdev_priv(dev);
455 int err;
456
457 INIT_LIST_HEAD(&vlan->queue_list);
458
459 /* Since macvlan supports all offloads by default, make
460 * tap support all offloads also.
461 */
462 vlan->tap_features = TUN_OFFLOADS;
463
464 err = netdev_rx_handler_register(dev, macvtap_handle_frame, vlan);
465 if (err)
466 return err;
467
468 /* Don't put anything that may fail after macvlan_common_newlink
469 * because we can't undo what it does.
470 */
471 return macvlan_common_newlink(src_net, dev, tb, data);
472}
473
474static void macvtap_dellink(struct net_device *dev,
475 struct list_head *head)
476{
477 netdev_rx_handler_unregister(dev);
478 macvtap_del_queues(dev);
479 macvlan_dellink(dev, head);
480}
481
482static void macvtap_setup(struct net_device *dev)
483{
484 macvlan_common_setup(dev);
485 dev->tx_queue_len = TUN_READQ_SIZE;
486}
487
488static struct rtnl_link_ops macvtap_link_ops __read_mostly = {
489 .kind = "macvtap",
490 .setup = macvtap_setup,
491 .newlink = macvtap_newlink,
492 .dellink = macvtap_dellink,
493};
494
495
496static void macvtap_sock_write_space(struct sock *sk)
497{
498 wait_queue_head_t *wqueue;
499
500 if (!sock_writeable(sk) ||
501 !test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
502 return;
503
504 wqueue = sk_sleep(sk);
505 if (wqueue && waitqueue_active(wqueue))
506 wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND);
507}
508
509static void macvtap_sock_destruct(struct sock *sk)
510{
511 skb_queue_purge(&sk->sk_receive_queue);
512}
513
514static int macvtap_open(struct inode *inode, struct file *file)
515{
516 struct net *net = current->nsproxy->net_ns;
517 struct net_device *dev;
518 struct macvtap_queue *q;
519 int err = -ENODEV;
520
521 rtnl_lock();
522 dev = dev_get_by_macvtap_minor(iminor(inode));
523 if (!dev)
524 goto out;
525
526 err = -ENOMEM;
527 q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
528 &macvtap_proto, 0);
529 if (!q)
530 goto out;
531
532 RCU_INIT_POINTER(q->sock.wq, &q->wq);
533 init_waitqueue_head(&q->wq.wait);
534 q->sock.type = SOCK_RAW;
535 q->sock.state = SS_CONNECTED;
536 q->sock.file = file;
537 q->sock.ops = &macvtap_socket_ops;
538 sock_init_data(&q->sock, &q->sk);
539 q->sk.sk_write_space = macvtap_sock_write_space;
540 q->sk.sk_destruct = macvtap_sock_destruct;
541 q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
542 q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
543
544 /*
545 * so far only KVM virtio_net uses macvtap, enable zero copy between
546 * guest kernel and host kernel when lower device supports zerocopy
547 *
548 * The macvlan supports zerocopy iff the lower device supports zero
549 * copy so we don't have to look at the lower device directly.
550 */
551 if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG))
552 sock_set_flag(&q->sk, SOCK_ZEROCOPY);
553
554 err = macvtap_set_queue(dev, file, q);
555 if (err)
556 sock_put(&q->sk);
557
558out:
559 if (dev)
560 dev_put(dev);
561
562 rtnl_unlock();
563 return err;
564}
565
566static int macvtap_release(struct inode *inode, struct file *file)
567{
568 struct macvtap_queue *q = file->private_data;
569 macvtap_put_queue(q);
570 return 0;
571}
572
573static unsigned int macvtap_poll(struct file *file, poll_table * wait)
574{
575 struct macvtap_queue *q = file->private_data;
576 unsigned int mask = POLLERR;
577
578 if (!q)
579 goto out;
580
581 mask = 0;
582 poll_wait(file, &q->wq.wait, wait);
583
584 if (!skb_queue_empty(&q->sk.sk_receive_queue))
585 mask |= POLLIN | POLLRDNORM;
586
587 if (sock_writeable(&q->sk) ||
588 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &q->sock.flags) &&
589 sock_writeable(&q->sk)))
590 mask |= POLLOUT | POLLWRNORM;
591
592out:
593 return mask;
594}
595
596static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad,
597 size_t len, size_t linear,
598 int noblock, int *err)
599{
600 struct sk_buff *skb;
601
602 /* Under a page? Don't bother with paged skb. */
603 if (prepad + len < PAGE_SIZE || !linear)
604 linear = len;
605
606 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
607 err, 0);
608 if (!skb)
609 return NULL;
610
611 skb_reserve(skb, prepad);
612 skb_put(skb, linear);
613 skb->data_len = len - linear;
614 skb->len += len - linear;
615
616 return skb;
617}
618
619/*
620 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
621 * be shared with the tun/tap driver.
622 */
623static int macvtap_skb_from_vnet_hdr(struct macvtap_queue *q,
624 struct sk_buff *skb,
625 struct virtio_net_hdr *vnet_hdr)
626{
627 unsigned short gso_type = 0;
628 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
629 switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
630 case VIRTIO_NET_HDR_GSO_TCPV4:
631 gso_type = SKB_GSO_TCPV4;
632 break;
633 case VIRTIO_NET_HDR_GSO_TCPV6:
634 gso_type = SKB_GSO_TCPV6;
635 break;
636 case VIRTIO_NET_HDR_GSO_UDP:
637 gso_type = SKB_GSO_UDP;
638 break;
639 default:
640 return -EINVAL;
641 }
642
643 if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
644 gso_type |= SKB_GSO_TCP_ECN;
645
646 if (vnet_hdr->gso_size == 0)
647 return -EINVAL;
648 }
649
650 if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
651 if (!skb_partial_csum_set(skb, macvtap16_to_cpu(q, vnet_hdr->csum_start),
652 macvtap16_to_cpu(q, vnet_hdr->csum_offset)))
653 return -EINVAL;
654 }
655
656 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
657 skb_shinfo(skb)->gso_size = macvtap16_to_cpu(q, vnet_hdr->gso_size);
658 skb_shinfo(skb)->gso_type = gso_type;
659
660 /* Header must be checked, and gso_segs computed. */
661 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
662 skb_shinfo(skb)->gso_segs = 0;
663 }
664 return 0;
665}
666
667static void macvtap_skb_to_vnet_hdr(struct macvtap_queue *q,
668 const struct sk_buff *skb,
669 struct virtio_net_hdr *vnet_hdr)
670{
671 memset(vnet_hdr, 0, sizeof(*vnet_hdr));
672
673 if (skb_is_gso(skb)) {
674 struct skb_shared_info *sinfo = skb_shinfo(skb);
675
676 /* This is a hint as to how much should be linear. */
677 vnet_hdr->hdr_len = cpu_to_macvtap16(q, skb_headlen(skb));
678 vnet_hdr->gso_size = cpu_to_macvtap16(q, sinfo->gso_size);
679 if (sinfo->gso_type & SKB_GSO_TCPV4)
680 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
681 else if (sinfo->gso_type & SKB_GSO_TCPV6)
682 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
683 else if (sinfo->gso_type & SKB_GSO_UDP)
684 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
685 else
686 BUG();
687 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
688 vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
689 } else
690 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
691
692 if (skb->ip_summed == CHECKSUM_PARTIAL) {
693 vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
694 if (skb_vlan_tag_present(skb))
695 vnet_hdr->csum_start = cpu_to_macvtap16(q,
696 skb_checksum_start_offset(skb) + VLAN_HLEN);
697 else
698 vnet_hdr->csum_start = cpu_to_macvtap16(q,
699 skb_checksum_start_offset(skb));
700 vnet_hdr->csum_offset = cpu_to_macvtap16(q, skb->csum_offset);
701 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
702 vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID;
703 } /* else everything is zero */
704}
705
706/* Neighbour code has some assumptions on HH_DATA_MOD alignment */
707#define MACVTAP_RESERVE HH_DATA_OFF(ETH_HLEN)
708
709/* Get packet from user space buffer */
710static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m,
711 struct iov_iter *from, int noblock)
712{
713 int good_linear = SKB_MAX_HEAD(MACVTAP_RESERVE);
714 struct sk_buff *skb;
715 struct macvlan_dev *vlan;
716 unsigned long total_len = iov_iter_count(from);
717 unsigned long len = total_len;
718 int err;
719 struct virtio_net_hdr vnet_hdr = { 0 };
720 int vnet_hdr_len = 0;
721 int copylen = 0;
722 int depth;
723 bool zerocopy = false;
724 size_t linear;
725 ssize_t n;
726
727 if (q->flags & IFF_VNET_HDR) {
728 vnet_hdr_len = q->vnet_hdr_sz;
729
730 err = -EINVAL;
731 if (len < vnet_hdr_len)
732 goto err;
733 len -= vnet_hdr_len;
734
735 err = -EFAULT;
736 n = copy_from_iter(&vnet_hdr, sizeof(vnet_hdr), from);
737 if (n != sizeof(vnet_hdr))
738 goto err;
739 iov_iter_advance(from, vnet_hdr_len - sizeof(vnet_hdr));
740 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
741 macvtap16_to_cpu(q, vnet_hdr.csum_start) +
742 macvtap16_to_cpu(q, vnet_hdr.csum_offset) + 2 >
743 macvtap16_to_cpu(q, vnet_hdr.hdr_len))
744 vnet_hdr.hdr_len = cpu_to_macvtap16(q,
745 macvtap16_to_cpu(q, vnet_hdr.csum_start) +
746 macvtap16_to_cpu(q, vnet_hdr.csum_offset) + 2);
747 err = -EINVAL;
748 if (macvtap16_to_cpu(q, vnet_hdr.hdr_len) > len)
749 goto err;
750 }
751
752 err = -EINVAL;
753 if (unlikely(len < ETH_HLEN))
754 goto err;
755
756 if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) {
757 struct iov_iter i;
758
759 copylen = vnet_hdr.hdr_len ?
760 macvtap16_to_cpu(q, vnet_hdr.hdr_len) : GOODCOPY_LEN;
761 if (copylen > good_linear)
762 copylen = good_linear;
763 else if (copylen < ETH_HLEN)
764 copylen = ETH_HLEN;
765 linear = copylen;
766 i = *from;
767 iov_iter_advance(&i, copylen);
768 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
769 zerocopy = true;
770 }
771
772 if (!zerocopy) {
773 copylen = len;
774 linear = macvtap16_to_cpu(q, vnet_hdr.hdr_len);
775 if (linear > good_linear)
776 linear = good_linear;
777 else if (linear < ETH_HLEN)
778 linear = ETH_HLEN;
779 }
780
781 skb = macvtap_alloc_skb(&q->sk, MACVTAP_RESERVE, copylen,
782 linear, noblock, &err);
783 if (!skb)
784 goto err;
785
786 if (zerocopy)
787 err = zerocopy_sg_from_iter(skb, from);
788 else {
789 err = skb_copy_datagram_from_iter(skb, 0, from, len);
790 if (!err && m && m->msg_control) {
791 struct ubuf_info *uarg = m->msg_control;
792 uarg->callback(uarg, false);
793 }
794 }
795
796 if (err)
797 goto err_kfree;
798
799 skb_set_network_header(skb, ETH_HLEN);
800 skb_reset_mac_header(skb);
801 skb->protocol = eth_hdr(skb)->h_proto;
802
803 if (vnet_hdr_len) {
804 err = macvtap_skb_from_vnet_hdr(q, skb, &vnet_hdr);
805 if (err)
806 goto err_kfree;
807 }
808
809 skb_probe_transport_header(skb, ETH_HLEN);
810
811 /* Move network header to the right position for VLAN tagged packets */
812 if ((skb->protocol == htons(ETH_P_8021Q) ||
813 skb->protocol == htons(ETH_P_8021AD)) &&
814 __vlan_get_protocol(skb, skb->protocol, &depth) != 0)
815 skb_set_network_header(skb, depth);
816
817 rcu_read_lock();
818 vlan = rcu_dereference(q->vlan);
819 /* copy skb_ubuf_info for callback when skb has no error */
820 if (zerocopy) {
821 skb_shinfo(skb)->destructor_arg = m->msg_control;
822 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
823 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
824 }
825 if (vlan) {
826 skb->dev = vlan->dev;
827 dev_queue_xmit(skb);
828 } else {
829 kfree_skb(skb);
830 }
831 rcu_read_unlock();
832
833 return total_len;
834
835err_kfree:
836 kfree_skb(skb);
837
838err:
839 rcu_read_lock();
840 vlan = rcu_dereference(q->vlan);
841 if (vlan)
842 this_cpu_inc(vlan->pcpu_stats->tx_dropped);
843 rcu_read_unlock();
844
845 return err;
846}
847
848static ssize_t macvtap_write_iter(struct kiocb *iocb, struct iov_iter *from)
849{
850 struct file *file = iocb->ki_filp;
851 struct macvtap_queue *q = file->private_data;
852
853 return macvtap_get_user(q, NULL, from, file->f_flags & O_NONBLOCK);
854}
855
856/* Put packet to the user space buffer */
857static ssize_t macvtap_put_user(struct macvtap_queue *q,
858 const struct sk_buff *skb,
859 struct iov_iter *iter)
860{
861 int ret;
862 int vnet_hdr_len = 0;
863 int vlan_offset = 0;
864 int total;
865
866 if (q->flags & IFF_VNET_HDR) {
867 struct virtio_net_hdr vnet_hdr;
868 vnet_hdr_len = q->vnet_hdr_sz;
869 if (iov_iter_count(iter) < vnet_hdr_len)
870 return -EINVAL;
871
872 macvtap_skb_to_vnet_hdr(q, skb, &vnet_hdr);
873
874 if (copy_to_iter(&vnet_hdr, sizeof(vnet_hdr), iter) !=
875 sizeof(vnet_hdr))
876 return -EFAULT;
877
878 iov_iter_advance(iter, vnet_hdr_len - sizeof(vnet_hdr));
879 }
880 total = vnet_hdr_len;
881 total += skb->len;
882
883 if (skb_vlan_tag_present(skb)) {
884 struct {
885 __be16 h_vlan_proto;
886 __be16 h_vlan_TCI;
887 } veth;
888 veth.h_vlan_proto = skb->vlan_proto;
889 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
890
891 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
892 total += VLAN_HLEN;
893
894 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
895 if (ret || !iov_iter_count(iter))
896 goto done;
897
898 ret = copy_to_iter(&veth, sizeof(veth), iter);
899 if (ret != sizeof(veth) || !iov_iter_count(iter))
900 goto done;
901 }
902
903 ret = skb_copy_datagram_iter(skb, vlan_offset, iter,
904 skb->len - vlan_offset);
905
906done:
907 return ret ? ret : total;
908}
909
910static ssize_t macvtap_do_read(struct macvtap_queue *q,
911 struct iov_iter *to,
912 int noblock)
913{
914 DEFINE_WAIT(wait);
915 struct sk_buff *skb;
916 ssize_t ret = 0;
917
918 if (!iov_iter_count(to))
919 return 0;
920
921 while (1) {
922 if (!noblock)
923 prepare_to_wait(sk_sleep(&q->sk), &wait,
924 TASK_INTERRUPTIBLE);
925
926 /* Read frames from the queue */
927 skb = skb_dequeue(&q->sk.sk_receive_queue);
928 if (skb)
929 break;
930 if (noblock) {
931 ret = -EAGAIN;
932 break;
933 }
934 if (signal_pending(current)) {
935 ret = -ERESTARTSYS;
936 break;
937 }
938 /* Nothing to read, let's sleep */
939 schedule();
940 }
941 if (!noblock)
942 finish_wait(sk_sleep(&q->sk), &wait);
943
944 if (skb) {
945 ret = macvtap_put_user(q, skb, to);
946 if (unlikely(ret < 0))
947 kfree_skb(skb);
948 else
949 consume_skb(skb);
950 }
951 return ret;
952}
953
954static ssize_t macvtap_read_iter(struct kiocb *iocb, struct iov_iter *to)
955{
956 struct file *file = iocb->ki_filp;
957 struct macvtap_queue *q = file->private_data;
958 ssize_t len = iov_iter_count(to), ret;
959
960 ret = macvtap_do_read(q, to, file->f_flags & O_NONBLOCK);
961 ret = min_t(ssize_t, ret, len);
962 if (ret > 0)
963 iocb->ki_pos = ret;
964 return ret;
965}
966
967static struct macvlan_dev *macvtap_get_vlan(struct macvtap_queue *q)
968{
969 struct macvlan_dev *vlan;
970
971 ASSERT_RTNL();
972 vlan = rtnl_dereference(q->vlan);
973 if (vlan)
974 dev_hold(vlan->dev);
975
976 return vlan;
977}
978
979static void macvtap_put_vlan(struct macvlan_dev *vlan)
980{
981 dev_put(vlan->dev);
982}
983
984static int macvtap_ioctl_set_queue(struct file *file, unsigned int flags)
985{
986 struct macvtap_queue *q = file->private_data;
987 struct macvlan_dev *vlan;
988 int ret;
989
990 vlan = macvtap_get_vlan(q);
991 if (!vlan)
992 return -EINVAL;
993
994 if (flags & IFF_ATTACH_QUEUE)
995 ret = macvtap_enable_queue(vlan->dev, file, q);
996 else if (flags & IFF_DETACH_QUEUE)
997 ret = macvtap_disable_queue(q);
998 else
999 ret = -EINVAL;
1000
1001 macvtap_put_vlan(vlan);
1002 return ret;
1003}
1004
1005static int set_offload(struct macvtap_queue *q, unsigned long arg)
1006{
1007 struct macvlan_dev *vlan;
1008 netdev_features_t features;
1009 netdev_features_t feature_mask = 0;
1010
1011 vlan = rtnl_dereference(q->vlan);
1012 if (!vlan)
1013 return -ENOLINK;
1014
1015 features = vlan->dev->features;
1016
1017 if (arg & TUN_F_CSUM) {
1018 feature_mask = NETIF_F_HW_CSUM;
1019
1020 if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) {
1021 if (arg & TUN_F_TSO_ECN)
1022 feature_mask |= NETIF_F_TSO_ECN;
1023 if (arg & TUN_F_TSO4)
1024 feature_mask |= NETIF_F_TSO;
1025 if (arg & TUN_F_TSO6)
1026 feature_mask |= NETIF_F_TSO6;
1027 }
1028
1029 if (arg & TUN_F_UFO)
1030 feature_mask |= NETIF_F_UFO;
1031 }
1032
1033 /* tun/tap driver inverts the usage for TSO offloads, where
1034 * setting the TSO bit means that the userspace wants to
1035 * accept TSO frames and turning it off means that user space
1036 * does not support TSO.
1037 * For macvtap, we have to invert it to mean the same thing.
1038 * When user space turns off TSO, we turn off GSO/LRO so that
1039 * user-space will not receive TSO frames.
1040 */
1041 if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_UFO))
1042 features |= RX_OFFLOADS;
1043 else
1044 features &= ~RX_OFFLOADS;
1045
1046 /* tap_features are the same as features on tun/tap and
1047 * reflect user expectations.
1048 */
1049 vlan->tap_features = feature_mask;
1050 vlan->set_features = features;
1051 netdev_update_features(vlan->dev);
1052
1053 return 0;
1054}
1055
1056/*
1057 * provide compatibility with generic tun/tap interface
1058 */
1059static long macvtap_ioctl(struct file *file, unsigned int cmd,
1060 unsigned long arg)
1061{
1062 struct macvtap_queue *q = file->private_data;
1063 struct macvlan_dev *vlan;
1064 void __user *argp = (void __user *)arg;
1065 struct ifreq __user *ifr = argp;
1066 unsigned int __user *up = argp;
1067 unsigned short u;
1068 int __user *sp = argp;
1069 struct sockaddr sa;
1070 int s;
1071 int ret;
1072
1073 switch (cmd) {
1074 case TUNSETIFF:
1075 /* ignore the name, just look at flags */
1076 if (get_user(u, &ifr->ifr_flags))
1077 return -EFAULT;
1078
1079 ret = 0;
1080 if ((u & ~MACVTAP_FEATURES) != (IFF_NO_PI | IFF_TAP))
1081 ret = -EINVAL;
1082 else
1083 q->flags = (q->flags & ~MACVTAP_FEATURES) | u;
1084
1085 return ret;
1086
1087 case TUNGETIFF:
1088 rtnl_lock();
1089 vlan = macvtap_get_vlan(q);
1090 if (!vlan) {
1091 rtnl_unlock();
1092 return -ENOLINK;
1093 }
1094
1095 ret = 0;
1096 u = q->flags;
1097 if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) ||
1098 put_user(u, &ifr->ifr_flags))
1099 ret = -EFAULT;
1100 macvtap_put_vlan(vlan);
1101 rtnl_unlock();
1102 return ret;
1103
1104 case TUNSETQUEUE:
1105 if (get_user(u, &ifr->ifr_flags))
1106 return -EFAULT;
1107 rtnl_lock();
1108 ret = macvtap_ioctl_set_queue(file, u);
1109 rtnl_unlock();
1110 return ret;
1111
1112 case TUNGETFEATURES:
1113 if (put_user(IFF_TAP | IFF_NO_PI | MACVTAP_FEATURES, up))
1114 return -EFAULT;
1115 return 0;
1116
1117 case TUNSETSNDBUF:
1118 if (get_user(s, sp))
1119 return -EFAULT;
1120
1121 q->sk.sk_sndbuf = s;
1122 return 0;
1123
1124 case TUNGETVNETHDRSZ:
1125 s = q->vnet_hdr_sz;
1126 if (put_user(s, sp))
1127 return -EFAULT;
1128 return 0;
1129
1130 case TUNSETVNETHDRSZ:
1131 if (get_user(s, sp))
1132 return -EFAULT;
1133 if (s < (int)sizeof(struct virtio_net_hdr))
1134 return -EINVAL;
1135
1136 q->vnet_hdr_sz = s;
1137 return 0;
1138
1139 case TUNGETVNETLE:
1140 s = !!(q->flags & MACVTAP_VNET_LE);
1141 if (put_user(s, sp))
1142 return -EFAULT;
1143 return 0;
1144
1145 case TUNSETVNETLE:
1146 if (get_user(s, sp))
1147 return -EFAULT;
1148 if (s)
1149 q->flags |= MACVTAP_VNET_LE;
1150 else
1151 q->flags &= ~MACVTAP_VNET_LE;
1152 return 0;
1153
1154 case TUNGETVNETBE:
1155 return macvtap_get_vnet_be(q, sp);
1156
1157 case TUNSETVNETBE:
1158 return macvtap_set_vnet_be(q, sp);
1159
1160 case TUNSETOFFLOAD:
1161 /* let the user check for future flags */
1162 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
1163 TUN_F_TSO_ECN | TUN_F_UFO))
1164 return -EINVAL;
1165
1166 rtnl_lock();
1167 ret = set_offload(q, arg);
1168 rtnl_unlock();
1169 return ret;
1170
1171 case SIOCGIFHWADDR:
1172 rtnl_lock();
1173 vlan = macvtap_get_vlan(q);
1174 if (!vlan) {
1175 rtnl_unlock();
1176 return -ENOLINK;
1177 }
1178 ret = 0;
1179 u = vlan->dev->type;
1180 if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) ||
1181 copy_to_user(&ifr->ifr_hwaddr.sa_data, vlan->dev->dev_addr, ETH_ALEN) ||
1182 put_user(u, &ifr->ifr_hwaddr.sa_family))
1183 ret = -EFAULT;
1184 macvtap_put_vlan(vlan);
1185 rtnl_unlock();
1186 return ret;
1187
1188 case SIOCSIFHWADDR:
1189 if (copy_from_user(&sa, &ifr->ifr_hwaddr, sizeof(sa)))
1190 return -EFAULT;
1191 rtnl_lock();
1192 vlan = macvtap_get_vlan(q);
1193 if (!vlan) {
1194 rtnl_unlock();
1195 return -ENOLINK;
1196 }
1197 ret = dev_set_mac_address(vlan->dev, &sa);
1198 macvtap_put_vlan(vlan);
1199 rtnl_unlock();
1200 return ret;
1201
1202 default:
1203 return -EINVAL;
1204 }
1205}
1206
1207#ifdef CONFIG_COMPAT
1208static long macvtap_compat_ioctl(struct file *file, unsigned int cmd,
1209 unsigned long arg)
1210{
1211 return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
1212}
1213#endif
1214
1215static const struct file_operations macvtap_fops = {
1216 .owner = THIS_MODULE,
1217 .open = macvtap_open,
1218 .release = macvtap_release,
1219 .read_iter = macvtap_read_iter,
1220 .write_iter = macvtap_write_iter,
1221 .poll = macvtap_poll,
1222 .llseek = no_llseek,
1223 .unlocked_ioctl = macvtap_ioctl,
1224#ifdef CONFIG_COMPAT
1225 .compat_ioctl = macvtap_compat_ioctl,
1226#endif
1227};
1228
1229static int macvtap_sendmsg(struct socket *sock, struct msghdr *m,
1230 size_t total_len)
1231{
1232 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1233 return macvtap_get_user(q, m, &m->msg_iter, m->msg_flags & MSG_DONTWAIT);
1234}
1235
1236static int macvtap_recvmsg(struct socket *sock, struct msghdr *m,
1237 size_t total_len, int flags)
1238{
1239 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1240 int ret;
1241 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1242 return -EINVAL;
1243 ret = macvtap_do_read(q, &m->msg_iter, flags & MSG_DONTWAIT);
1244 if (ret > total_len) {
1245 m->msg_flags |= MSG_TRUNC;
1246 ret = flags & MSG_TRUNC ? ret : total_len;
1247 }
1248 return ret;
1249}
1250
1251/* Ops structure to mimic raw sockets with tun */
1252static const struct proto_ops macvtap_socket_ops = {
1253 .sendmsg = macvtap_sendmsg,
1254 .recvmsg = macvtap_recvmsg,
1255};
1256
1257/* Get an underlying socket object from tun file. Returns error unless file is
1258 * attached to a device. The returned object works like a packet socket, it
1259 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
1260 * holding a reference to the file for as long as the socket is in use. */
1261struct socket *macvtap_get_socket(struct file *file)
1262{
1263 struct macvtap_queue *q;
1264 if (file->f_op != &macvtap_fops)
1265 return ERR_PTR(-EINVAL);
1266 q = file->private_data;
1267 if (!q)
1268 return ERR_PTR(-EBADFD);
1269 return &q->sock;
1270}
1271EXPORT_SYMBOL_GPL(macvtap_get_socket);
1272
1273static int macvtap_device_event(struct notifier_block *unused,
1274 unsigned long event, void *ptr)
1275{
1276 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1277 struct macvlan_dev *vlan;
1278 struct device *classdev;
1279 dev_t devt;
1280 int err;
1281
1282 if (dev->rtnl_link_ops != &macvtap_link_ops)
1283 return NOTIFY_DONE;
1284
1285 vlan = netdev_priv(dev);
1286
1287 switch (event) {
1288 case NETDEV_REGISTER:
1289 /* Create the device node here after the network device has
1290 * been registered but before register_netdevice has
1291 * finished running.
1292 */
1293 err = macvtap_get_minor(vlan);
1294 if (err)
1295 return notifier_from_errno(err);
1296
1297 devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
1298 classdev = device_create(macvtap_class, &dev->dev, devt,
1299 dev, "tap%d", dev->ifindex);
1300 if (IS_ERR(classdev)) {
1301 macvtap_free_minor(vlan);
1302 return notifier_from_errno(PTR_ERR(classdev));
1303 }
1304 break;
1305 case NETDEV_UNREGISTER:
1306 devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
1307 device_destroy(macvtap_class, devt);
1308 macvtap_free_minor(vlan);
1309 break;
1310 }
1311
1312 return NOTIFY_DONE;
1313}
1314
1315static struct notifier_block macvtap_notifier_block __read_mostly = {
1316 .notifier_call = macvtap_device_event,
1317};
1318
1319static int macvtap_init(void)
1320{
1321 int err;
1322
1323 err = alloc_chrdev_region(&macvtap_major, 0,
1324 MACVTAP_NUM_DEVS, "macvtap");
1325 if (err)
1326 goto out1;
1327
1328 cdev_init(&macvtap_cdev, &macvtap_fops);
1329 err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS);
1330 if (err)
1331 goto out2;
1332
1333 macvtap_class = class_create(THIS_MODULE, "macvtap");
1334 if (IS_ERR(macvtap_class)) {
1335 err = PTR_ERR(macvtap_class);
1336 goto out3;
1337 }
1338
1339 err = register_netdevice_notifier(&macvtap_notifier_block);
1340 if (err)
1341 goto out4;
1342
1343 err = macvlan_link_register(&macvtap_link_ops);
1344 if (err)
1345 goto out5;
1346
1347 return 0;
1348
1349out5:
1350 unregister_netdevice_notifier(&macvtap_notifier_block);
1351out4:
1352 class_unregister(macvtap_class);
1353out3:
1354 cdev_del(&macvtap_cdev);
1355out2:
1356 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1357out1:
1358 return err;
1359}
1360module_init(macvtap_init);
1361
1362static void macvtap_exit(void)
1363{
1364 rtnl_link_unregister(&macvtap_link_ops);
1365 unregister_netdevice_notifier(&macvtap_notifier_block);
1366 class_unregister(macvtap_class);
1367 cdev_del(&macvtap_cdev);
1368 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1369 idr_destroy(&minor_idr);
1370}
1371module_exit(macvtap_exit);
1372
1373MODULE_ALIAS_RTNL_LINK("macvtap");
1374MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1375MODULE_LICENSE("GPL");