1 #include <linux/etherdevice.h>
2 #include <linux/if_macvlan.h>
3 #include <linux/if_vlan.h>
4 #include <linux/interrupt.h>
5 #include <linux/nsproxy.h>
6 #include <linux/compat.h>
7 #include <linux/if_tun.h>
8 #include <linux/module.h>
9 #include <linux/skbuff.h>
10 #include <linux/cache.h>
11 #include <linux/sched.h>
12 #include <linux/types.h>
13 #include <linux/slab.h>
14 #include <linux/init.h>
15 #include <linux/wait.h>
16 #include <linux/cdev.h>
17 #include <linux/idr.h>
20 #include <net/net_namespace.h>
21 #include <net/rtnetlink.h>
23 #include <linux/virtio_net.h>
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.
31 * macvtap_proto is used to allocate queues through the sock allocation
34 * TODO: multiqueue support is currently not implemented, even though
35 * macvtap is basically prepared for that. We will need to add this
36 * here as well as in virtio-net and qemu to get line rate on 10gbit
37 * adapters from a guest.
39 struct macvtap_queue
{
44 struct macvlan_dev __rcu
*vlan
;
49 struct list_head next
;
52 static struct proto macvtap_proto
= {
55 .obj_size
= sizeof (struct macvtap_queue
),
59 * Variables for dealing with macvtaps device numbers.
61 static dev_t macvtap_major
;
62 #define MACVTAP_NUM_DEVS (1U << MINORBITS)
63 static DEFINE_MUTEX(minor_lock
);
64 static DEFINE_IDR(minor_idr
);
66 #define GOODCOPY_LEN 128
67 static struct class *macvtap_class
;
68 static struct cdev macvtap_cdev
;
70 static const struct proto_ops macvtap_socket_ops
;
74 * The macvtap_queue and the macvlan_dev are loosely coupled, the
75 * pointers from one to the other can only be read while rcu_read_lock
76 * or macvtap_lock is held.
78 * Both the file and the macvlan_dev hold a reference on the macvtap_queue
79 * through sock_hold(&q->sk). When the macvlan_dev goes away first,
80 * q->vlan becomes inaccessible. When the files gets closed,
81 * macvtap_get_queue() fails.
83 * There may still be references to the struct sock inside of the
84 * queue from outbound SKBs, but these never reference back to the
85 * file or the dev. The data structure is freed through __sk_free
86 * when both our references and any pending SKBs are gone.
88 static DEFINE_SPINLOCK(macvtap_lock
);
90 static int macvtap_enable_queue(struct net_device
*dev
, struct file
*file
,
91 struct macvtap_queue
*q
)
93 struct macvlan_dev
*vlan
= netdev_priv(dev
);
96 spin_lock(&macvtap_lock
);
102 rcu_assign_pointer(vlan
->taps
[vlan
->numvtaps
], q
);
103 q
->queue_index
= vlan
->numvtaps
;
108 spin_unlock(&macvtap_lock
);
112 static int macvtap_set_queue(struct net_device
*dev
, struct file
*file
,
113 struct macvtap_queue
*q
)
115 struct macvlan_dev
*vlan
= netdev_priv(dev
);
118 spin_lock(&macvtap_lock
);
119 if (vlan
->numqueues
== MAX_MACVTAP_QUEUES
)
123 rcu_assign_pointer(q
->vlan
, vlan
);
124 rcu_assign_pointer(vlan
->taps
[vlan
->numvtaps
], q
);
128 q
->queue_index
= vlan
->numvtaps
;
130 file
->private_data
= q
;
131 list_add_tail(&q
->next
, &vlan
->queue_list
);
137 spin_unlock(&macvtap_lock
);
141 static int __macvtap_disable_queue(struct macvtap_queue
*q
)
143 struct macvlan_dev
*vlan
;
144 struct macvtap_queue
*nq
;
146 vlan
= rcu_dereference_protected(q
->vlan
,
147 lockdep_is_held(&macvtap_lock
));
153 int index
= q
->queue_index
;
154 BUG_ON(index
>= vlan
->numvtaps
);
155 nq
= rcu_dereference_protected(vlan
->taps
[vlan
->numvtaps
- 1],
156 lockdep_is_held(&macvtap_lock
));
157 nq
->queue_index
= index
;
159 rcu_assign_pointer(vlan
->taps
[index
], nq
);
160 RCU_INIT_POINTER(vlan
->taps
[vlan
->numvtaps
- 1], NULL
);
169 static int macvtap_disable_queue(struct macvtap_queue
*q
)
173 spin_lock(&macvtap_lock
);
174 err
= __macvtap_disable_queue(q
);
175 spin_unlock(&macvtap_lock
);
181 * The file owning the queue got closed, give up both
182 * the reference that the files holds as well as the
183 * one from the macvlan_dev if that still exists.
185 * Using the spinlock makes sure that we don't get
186 * to the queue again after destroying it.
188 static void macvtap_put_queue(struct macvtap_queue
*q
)
190 struct macvlan_dev
*vlan
;
192 spin_lock(&macvtap_lock
);
193 vlan
= rcu_dereference_protected(q
->vlan
,
194 lockdep_is_held(&macvtap_lock
));
197 BUG_ON(__macvtap_disable_queue(q
));
200 RCU_INIT_POINTER(q
->vlan
, NULL
);
202 list_del_init(&q
->next
);
205 spin_unlock(&macvtap_lock
);
212 * Select a queue based on the rxq of the device on which this packet
213 * arrived. If the incoming device is not mq, calculate a flow hash
214 * to select a queue. If all fails, find the first available queue.
215 * Cache vlan->numvtaps since it can become zero during the execution
218 static struct macvtap_queue
*macvtap_get_queue(struct net_device
*dev
,
221 struct macvlan_dev
*vlan
= netdev_priv(dev
);
222 struct macvtap_queue
*tap
= NULL
;
223 /* Access to taps array is protected by rcu, but access to numvtaps
224 * isn't. Below we use it to lookup a queue, but treat it as a hint
225 * and validate that the result isn't NULL - in case we are
226 * racing against queue removal.
228 int numvtaps
= ACCESS_ONCE(vlan
->numvtaps
);
234 /* Check if we can use flow to select a queue */
235 rxq
= skb_get_rxhash(skb
);
237 tap
= rcu_dereference(vlan
->taps
[rxq
% numvtaps
]);
241 if (likely(skb_rx_queue_recorded(skb
))) {
242 rxq
= skb_get_rx_queue(skb
);
244 while (unlikely(rxq
>= numvtaps
))
247 tap
= rcu_dereference(vlan
->taps
[rxq
]);
251 tap
= rcu_dereference(vlan
->taps
[0]);
257 * The net_device is going away, give up the reference
258 * that it holds on all queues and safely set the pointer
259 * from the queues to NULL.
261 static void macvtap_del_queues(struct net_device
*dev
)
263 struct macvlan_dev
*vlan
= netdev_priv(dev
);
264 struct macvtap_queue
*q
, *tmp
, *qlist
[MAX_MACVTAP_QUEUES
];
267 spin_lock(&macvtap_lock
);
268 list_for_each_entry_safe(q
, tmp
, &vlan
->queue_list
, next
) {
269 list_del_init(&q
->next
);
271 RCU_INIT_POINTER(q
->vlan
, NULL
);
276 for (i
= 0; i
< vlan
->numvtaps
; i
++)
277 RCU_INIT_POINTER(vlan
->taps
[i
], NULL
);
278 BUG_ON(vlan
->numvtaps
);
279 BUG_ON(vlan
->numqueues
);
280 /* guarantee that any future macvtap_set_queue will fail */
281 vlan
->numvtaps
= MAX_MACVTAP_QUEUES
;
282 spin_unlock(&macvtap_lock
);
286 for (--j
; j
>= 0; j
--)
287 sock_put(&qlist
[j
]->sk
);
291 * Forward happens for data that gets sent from one macvlan
292 * endpoint to another one in bridge mode. We just take
293 * the skb and put it into the receive queue.
295 static int macvtap_forward(struct net_device
*dev
, struct sk_buff
*skb
)
297 struct macvtap_queue
*q
= macvtap_get_queue(dev
, skb
);
301 if (skb_queue_len(&q
->sk
.sk_receive_queue
) >= dev
->tx_queue_len
)
304 skb_queue_tail(&q
->sk
.sk_receive_queue
, skb
);
305 wake_up_interruptible_poll(sk_sleep(&q
->sk
), POLLIN
| POLLRDNORM
| POLLRDBAND
);
306 return NET_RX_SUCCESS
;
314 * Receive is for data from the external interface (lowerdev),
315 * in case of macvtap, we can treat that the same way as
316 * forward, which macvlan cannot.
318 static int macvtap_receive(struct sk_buff
*skb
)
320 skb_push(skb
, ETH_HLEN
);
321 return macvtap_forward(skb
->dev
, skb
);
324 static int macvtap_get_minor(struct macvlan_dev
*vlan
)
326 int retval
= -ENOMEM
;
328 mutex_lock(&minor_lock
);
329 retval
= idr_alloc(&minor_idr
, vlan
, 1, MACVTAP_NUM_DEVS
, GFP_KERNEL
);
331 vlan
->minor
= retval
;
332 } else if (retval
== -ENOSPC
) {
333 printk(KERN_ERR
"too many macvtap devices\n");
336 mutex_unlock(&minor_lock
);
337 return retval
< 0 ? retval
: 0;
340 static void macvtap_free_minor(struct macvlan_dev
*vlan
)
342 mutex_lock(&minor_lock
);
344 idr_remove(&minor_idr
, vlan
->minor
);
347 mutex_unlock(&minor_lock
);
350 static struct net_device
*dev_get_by_macvtap_minor(int minor
)
352 struct net_device
*dev
= NULL
;
353 struct macvlan_dev
*vlan
;
355 mutex_lock(&minor_lock
);
356 vlan
= idr_find(&minor_idr
, minor
);
361 mutex_unlock(&minor_lock
);
365 static int macvtap_newlink(struct net
*src_net
,
366 struct net_device
*dev
,
368 struct nlattr
*data
[])
370 struct macvlan_dev
*vlan
= netdev_priv(dev
);
371 INIT_LIST_HEAD(&vlan
->queue_list
);
373 /* Don't put anything that may fail after macvlan_common_newlink
374 * because we can't undo what it does.
376 return macvlan_common_newlink(src_net
, dev
, tb
, data
,
377 macvtap_receive
, macvtap_forward
);
380 static void macvtap_dellink(struct net_device
*dev
,
381 struct list_head
*head
)
383 macvtap_del_queues(dev
);
384 macvlan_dellink(dev
, head
);
387 static void macvtap_setup(struct net_device
*dev
)
389 macvlan_common_setup(dev
);
390 dev
->tx_queue_len
= TUN_READQ_SIZE
;
393 static struct rtnl_link_ops macvtap_link_ops __read_mostly
= {
395 .setup
= macvtap_setup
,
396 .newlink
= macvtap_newlink
,
397 .dellink
= macvtap_dellink
,
401 static void macvtap_sock_write_space(struct sock
*sk
)
403 wait_queue_head_t
*wqueue
;
405 if (!sock_writeable(sk
) ||
406 !test_and_clear_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
))
409 wqueue
= sk_sleep(sk
);
410 if (wqueue
&& waitqueue_active(wqueue
))
411 wake_up_interruptible_poll(wqueue
, POLLOUT
| POLLWRNORM
| POLLWRBAND
);
414 static void macvtap_sock_destruct(struct sock
*sk
)
416 skb_queue_purge(&sk
->sk_receive_queue
);
419 static int macvtap_open(struct inode
*inode
, struct file
*file
)
421 struct net
*net
= current
->nsproxy
->net_ns
;
422 struct net_device
*dev
= dev_get_by_macvtap_minor(iminor(inode
));
423 struct macvtap_queue
*q
;
431 q
= (struct macvtap_queue
*)sk_alloc(net
, AF_UNSPEC
, GFP_KERNEL
,
437 init_waitqueue_head(&q
->wq
.wait
);
438 q
->sock
.type
= SOCK_RAW
;
439 q
->sock
.state
= SS_CONNECTED
;
441 q
->sock
.ops
= &macvtap_socket_ops
;
442 sock_init_data(&q
->sock
, &q
->sk
);
443 q
->sk
.sk_write_space
= macvtap_sock_write_space
;
444 q
->sk
.sk_destruct
= macvtap_sock_destruct
;
445 q
->flags
= IFF_VNET_HDR
| IFF_NO_PI
| IFF_TAP
;
446 q
->vnet_hdr_sz
= sizeof(struct virtio_net_hdr
);
449 * so far only KVM virtio_net uses macvtap, enable zero copy between
450 * guest kernel and host kernel when lower device supports zerocopy
452 * The macvlan supports zerocopy iff the lower device supports zero
453 * copy so we don't have to look at the lower device directly.
455 if ((dev
->features
& NETIF_F_HIGHDMA
) && (dev
->features
& NETIF_F_SG
))
456 sock_set_flag(&q
->sk
, SOCK_ZEROCOPY
);
458 err
= macvtap_set_queue(dev
, file
, q
);
469 static int macvtap_release(struct inode
*inode
, struct file
*file
)
471 struct macvtap_queue
*q
= file
->private_data
;
472 macvtap_put_queue(q
);
476 static unsigned int macvtap_poll(struct file
*file
, poll_table
* wait
)
478 struct macvtap_queue
*q
= file
->private_data
;
479 unsigned int mask
= POLLERR
;
485 poll_wait(file
, &q
->wq
.wait
, wait
);
487 if (!skb_queue_empty(&q
->sk
.sk_receive_queue
))
488 mask
|= POLLIN
| POLLRDNORM
;
490 if (sock_writeable(&q
->sk
) ||
491 (!test_and_set_bit(SOCK_ASYNC_NOSPACE
, &q
->sock
.flags
) &&
492 sock_writeable(&q
->sk
)))
493 mask
|= POLLOUT
| POLLWRNORM
;
499 static inline struct sk_buff
*macvtap_alloc_skb(struct sock
*sk
, size_t prepad
,
500 size_t len
, size_t linear
,
501 int noblock
, int *err
)
505 /* Under a page? Don't bother with paged skb. */
506 if (prepad
+ len
< PAGE_SIZE
|| !linear
)
509 skb
= sock_alloc_send_pskb(sk
, prepad
+ linear
, len
- linear
, noblock
,
514 skb_reserve(skb
, prepad
);
515 skb_put(skb
, linear
);
516 skb
->data_len
= len
- linear
;
517 skb
->len
+= len
- linear
;
522 /* set skb frags from iovec, this can move to core network code for reuse */
523 static int zerocopy_sg_from_iovec(struct sk_buff
*skb
, const struct iovec
*from
,
524 int offset
, size_t count
)
526 int len
= iov_length(from
, count
) - offset
;
527 int copy
= skb_headlen(skb
);
528 int size
, offset1
= 0;
531 /* Skip over from offset */
532 while (count
&& (offset
>= from
->iov_len
)) {
533 offset
-= from
->iov_len
;
538 /* copy up to skb headlen */
539 while (count
&& (copy
> 0)) {
540 size
= min_t(unsigned int, copy
, from
->iov_len
- offset
);
541 if (copy_from_user(skb
->data
+ offset1
, from
->iov_base
+ offset
,
558 struct page
*page
[MAX_SKB_FRAGS
];
561 unsigned long truesize
;
563 len
= from
->iov_len
- offset
;
569 base
= (unsigned long)from
->iov_base
+ offset
;
570 size
= ((base
& ~PAGE_MASK
) + len
+ ~PAGE_MASK
) >> PAGE_SHIFT
;
571 if (i
+ size
> MAX_SKB_FRAGS
)
573 num_pages
= get_user_pages_fast(base
, size
, 0, &page
[i
]);
574 if (num_pages
!= size
) {
575 for (i
= 0; i
< num_pages
; i
++)
579 truesize
= size
* PAGE_SIZE
;
580 skb
->data_len
+= len
;
582 skb
->truesize
+= truesize
;
583 atomic_add(truesize
, &skb
->sk
->sk_wmem_alloc
);
585 int off
= base
& ~PAGE_MASK
;
586 int size
= min_t(int, len
, PAGE_SIZE
- off
);
587 __skb_fill_page_desc(skb
, i
, page
[i
], off
, size
);
588 skb_shinfo(skb
)->nr_frags
++;
589 /* increase sk_wmem_alloc */
601 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
602 * be shared with the tun/tap driver.
604 static int macvtap_skb_from_vnet_hdr(struct sk_buff
*skb
,
605 struct virtio_net_hdr
*vnet_hdr
)
607 unsigned short gso_type
= 0;
608 if (vnet_hdr
->gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
609 switch (vnet_hdr
->gso_type
& ~VIRTIO_NET_HDR_GSO_ECN
) {
610 case VIRTIO_NET_HDR_GSO_TCPV4
:
611 gso_type
= SKB_GSO_TCPV4
;
613 case VIRTIO_NET_HDR_GSO_TCPV6
:
614 gso_type
= SKB_GSO_TCPV6
;
616 case VIRTIO_NET_HDR_GSO_UDP
:
617 gso_type
= SKB_GSO_UDP
;
623 if (vnet_hdr
->gso_type
& VIRTIO_NET_HDR_GSO_ECN
)
624 gso_type
|= SKB_GSO_TCP_ECN
;
626 if (vnet_hdr
->gso_size
== 0)
630 if (vnet_hdr
->flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) {
631 if (!skb_partial_csum_set(skb
, vnet_hdr
->csum_start
,
632 vnet_hdr
->csum_offset
))
636 if (vnet_hdr
->gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
637 skb_shinfo(skb
)->gso_size
= vnet_hdr
->gso_size
;
638 skb_shinfo(skb
)->gso_type
= gso_type
;
640 /* Header must be checked, and gso_segs computed. */
641 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
642 skb_shinfo(skb
)->gso_segs
= 0;
647 static int macvtap_skb_to_vnet_hdr(const struct sk_buff
*skb
,
648 struct virtio_net_hdr
*vnet_hdr
)
650 memset(vnet_hdr
, 0, sizeof(*vnet_hdr
));
652 if (skb_is_gso(skb
)) {
653 struct skb_shared_info
*sinfo
= skb_shinfo(skb
);
655 /* This is a hint as to how much should be linear. */
656 vnet_hdr
->hdr_len
= skb_headlen(skb
);
657 vnet_hdr
->gso_size
= sinfo
->gso_size
;
658 if (sinfo
->gso_type
& SKB_GSO_TCPV4
)
659 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_TCPV4
;
660 else if (sinfo
->gso_type
& SKB_GSO_TCPV6
)
661 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_TCPV6
;
662 else if (sinfo
->gso_type
& SKB_GSO_UDP
)
663 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_UDP
;
666 if (sinfo
->gso_type
& SKB_GSO_TCP_ECN
)
667 vnet_hdr
->gso_type
|= VIRTIO_NET_HDR_GSO_ECN
;
669 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
671 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
672 vnet_hdr
->flags
= VIRTIO_NET_HDR_F_NEEDS_CSUM
;
673 vnet_hdr
->csum_start
= skb_checksum_start_offset(skb
);
674 vnet_hdr
->csum_offset
= skb
->csum_offset
;
675 } else if (skb
->ip_summed
== CHECKSUM_UNNECESSARY
) {
676 vnet_hdr
->flags
= VIRTIO_NET_HDR_F_DATA_VALID
;
677 } /* else everything is zero */
683 /* Get packet from user space buffer */
684 static ssize_t
macvtap_get_user(struct macvtap_queue
*q
, struct msghdr
*m
,
685 const struct iovec
*iv
, unsigned long total_len
,
686 size_t count
, int noblock
)
689 struct macvlan_dev
*vlan
;
690 unsigned long len
= total_len
;
692 struct virtio_net_hdr vnet_hdr
= { 0 };
693 int vnet_hdr_len
= 0;
695 bool zerocopy
= false;
697 if (q
->flags
& IFF_VNET_HDR
) {
698 vnet_hdr_len
= q
->vnet_hdr_sz
;
701 if (len
< vnet_hdr_len
)
705 err
= memcpy_fromiovecend((void *)&vnet_hdr
, iv
, 0,
709 if ((vnet_hdr
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) &&
710 vnet_hdr
.csum_start
+ vnet_hdr
.csum_offset
+ 2 >
712 vnet_hdr
.hdr_len
= vnet_hdr
.csum_start
+
713 vnet_hdr
.csum_offset
+ 2;
715 if (vnet_hdr
.hdr_len
> len
)
720 if (unlikely(len
< ETH_HLEN
))
724 if (unlikely(count
> UIO_MAXIOV
))
727 if (m
&& m
->msg_control
&& sock_flag(&q
->sk
, SOCK_ZEROCOPY
))
731 /* Userspace may produce vectors with count greater than
732 * MAX_SKB_FRAGS, so we need to linearize parts of the skb
733 * to let the rest of data to be fit in the frags.
735 if (count
> MAX_SKB_FRAGS
) {
736 copylen
= iov_length(iv
, count
- MAX_SKB_FRAGS
);
737 if (copylen
< vnet_hdr_len
)
740 copylen
-= vnet_hdr_len
;
742 /* There are 256 bytes to be copied in skb, so there is enough
743 * room for skb expand head in case it is used.
744 * The rest buffer is mapped from userspace.
746 if (copylen
< vnet_hdr
.hdr_len
)
747 copylen
= vnet_hdr
.hdr_len
;
749 copylen
= GOODCOPY_LEN
;
753 skb
= macvtap_alloc_skb(&q
->sk
, NET_IP_ALIGN
, copylen
,
754 vnet_hdr
.hdr_len
, noblock
, &err
);
759 err
= zerocopy_sg_from_iovec(skb
, iv
, vnet_hdr_len
, count
);
761 err
= skb_copy_datagram_from_iovec(skb
, 0, iv
, vnet_hdr_len
,
766 skb_set_network_header(skb
, ETH_HLEN
);
767 skb_reset_mac_header(skb
);
768 skb
->protocol
= eth_hdr(skb
)->h_proto
;
771 err
= macvtap_skb_from_vnet_hdr(skb
, &vnet_hdr
);
776 skb_probe_transport_header(skb
, ETH_HLEN
);
779 vlan
= rcu_dereference_bh(q
->vlan
);
780 /* copy skb_ubuf_info for callback when skb has no error */
782 skb_shinfo(skb
)->destructor_arg
= m
->msg_control
;
783 skb_shinfo(skb
)->tx_flags
|= SKBTX_DEV_ZEROCOPY
;
784 skb_shinfo(skb
)->tx_flags
|= SKBTX_SHARED_FRAG
;
787 macvlan_start_xmit(skb
, vlan
->dev
);
790 rcu_read_unlock_bh();
799 vlan
= rcu_dereference_bh(q
->vlan
);
801 vlan
->dev
->stats
.tx_dropped
++;
802 rcu_read_unlock_bh();
807 static ssize_t
macvtap_aio_write(struct kiocb
*iocb
, const struct iovec
*iv
,
808 unsigned long count
, loff_t pos
)
810 struct file
*file
= iocb
->ki_filp
;
811 ssize_t result
= -ENOLINK
;
812 struct macvtap_queue
*q
= file
->private_data
;
814 result
= macvtap_get_user(q
, NULL
, iv
, iov_length(iv
, count
), count
,
815 file
->f_flags
& O_NONBLOCK
);
819 /* Put packet to the user space buffer */
820 static ssize_t
macvtap_put_user(struct macvtap_queue
*q
,
821 const struct sk_buff
*skb
,
822 const struct iovec
*iv
, int len
)
824 struct macvlan_dev
*vlan
;
826 int vnet_hdr_len
= 0;
830 if (q
->flags
& IFF_VNET_HDR
) {
831 struct virtio_net_hdr vnet_hdr
;
832 vnet_hdr_len
= q
->vnet_hdr_sz
;
833 if ((len
-= vnet_hdr_len
) < 0)
836 ret
= macvtap_skb_to_vnet_hdr(skb
, &vnet_hdr
);
840 if (memcpy_toiovecend(iv
, (void *)&vnet_hdr
, 0, sizeof(vnet_hdr
)))
843 copied
= vnet_hdr_len
;
845 if (!vlan_tx_tag_present(skb
))
846 len
= min_t(int, skb
->len
, len
);
853 veth
.h_vlan_proto
= htons(ETH_P_8021Q
);
854 veth
.h_vlan_TCI
= htons(vlan_tx_tag_get(skb
));
856 vlan_offset
= offsetof(struct vlan_ethhdr
, h_vlan_proto
);
857 len
= min_t(int, skb
->len
+ VLAN_HLEN
, len
);
859 copy
= min_t(int, vlan_offset
, len
);
860 ret
= skb_copy_datagram_const_iovec(skb
, 0, iv
, copied
, copy
);
866 copy
= min_t(int, sizeof(veth
), len
);
867 ret
= memcpy_toiovecend(iv
, (void *)&veth
, copied
, copy
);
874 ret
= skb_copy_datagram_const_iovec(skb
, vlan_offset
, iv
, copied
, len
);
879 vlan
= rcu_dereference_bh(q
->vlan
);
881 macvlan_count_rx(vlan
, copied
- vnet_hdr_len
, ret
== 0, 0);
882 rcu_read_unlock_bh();
884 return ret
? ret
: copied
;
887 static ssize_t
macvtap_do_read(struct macvtap_queue
*q
, struct kiocb
*iocb
,
888 const struct iovec
*iv
, unsigned long len
,
897 prepare_to_wait(sk_sleep(&q
->sk
), &wait
,
900 /* Read frames from the queue */
901 skb
= skb_dequeue(&q
->sk
.sk_receive_queue
);
907 if (signal_pending(current
)) {
911 /* Nothing to read, let's sleep */
915 ret
= macvtap_put_user(q
, skb
, iv
, len
);
921 finish_wait(sk_sleep(&q
->sk
), &wait
);
925 static ssize_t
macvtap_aio_read(struct kiocb
*iocb
, const struct iovec
*iv
,
926 unsigned long count
, loff_t pos
)
928 struct file
*file
= iocb
->ki_filp
;
929 struct macvtap_queue
*q
= file
->private_data
;
930 ssize_t len
, ret
= 0;
932 len
= iov_length(iv
, count
);
938 ret
= macvtap_do_read(q
, iocb
, iv
, len
, file
->f_flags
& O_NONBLOCK
);
939 ret
= min_t(ssize_t
, ret
, len
); /* XXX copied from tun.c. Why? */
944 static struct macvlan_dev
*macvtap_get_vlan(struct macvtap_queue
*q
)
946 struct macvlan_dev
*vlan
;
949 vlan
= rcu_dereference_bh(q
->vlan
);
952 rcu_read_unlock_bh();
957 static void macvtap_put_vlan(struct macvlan_dev
*vlan
)
962 static int macvtap_ioctl_set_queue(struct file
*file
, unsigned int flags
)
964 struct macvtap_queue
*q
= file
->private_data
;
965 struct macvlan_dev
*vlan
;
968 vlan
= macvtap_get_vlan(q
);
972 if (flags
& IFF_ATTACH_QUEUE
)
973 ret
= macvtap_enable_queue(vlan
->dev
, file
, q
);
974 else if (flags
& IFF_DETACH_QUEUE
)
975 ret
= macvtap_disable_queue(q
);
977 macvtap_put_vlan(vlan
);
982 * provide compatibility with generic tun/tap interface
984 static long macvtap_ioctl(struct file
*file
, unsigned int cmd
,
987 struct macvtap_queue
*q
= file
->private_data
;
988 struct macvlan_dev
*vlan
;
989 void __user
*argp
= (void __user
*)arg
;
990 struct ifreq __user
*ifr
= argp
;
991 unsigned int __user
*up
= argp
;
993 int __user
*sp
= argp
;
999 /* ignore the name, just look at flags */
1000 if (get_user(u
, &ifr
->ifr_flags
))
1004 if ((u
& ~(IFF_VNET_HDR
| IFF_MULTI_QUEUE
)) !=
1005 (IFF_NO_PI
| IFF_TAP
))
1013 vlan
= macvtap_get_vlan(q
);
1018 if (copy_to_user(&ifr
->ifr_name
, vlan
->dev
->name
, IFNAMSIZ
) ||
1019 put_user(q
->flags
, &ifr
->ifr_flags
))
1021 macvtap_put_vlan(vlan
);
1025 if (get_user(u
, &ifr
->ifr_flags
))
1027 return macvtap_ioctl_set_queue(file
, u
);
1029 case TUNGETFEATURES
:
1030 if (put_user(IFF_TAP
| IFF_NO_PI
| IFF_VNET_HDR
, up
))
1035 if (get_user(u
, up
))
1038 q
->sk
.sk_sndbuf
= u
;
1041 case TUNGETVNETHDRSZ
:
1043 if (put_user(s
, sp
))
1047 case TUNSETVNETHDRSZ
:
1048 if (get_user(s
, sp
))
1050 if (s
< (int)sizeof(struct virtio_net_hdr
))
1057 /* let the user check for future flags */
1058 if (arg
& ~(TUN_F_CSUM
| TUN_F_TSO4
| TUN_F_TSO6
|
1059 TUN_F_TSO_ECN
| TUN_F_UFO
))
1062 /* TODO: only accept frames with the features that
1063 got enabled for forwarded frames */
1064 if (!(q
->flags
& IFF_VNET_HDR
))
1073 #ifdef CONFIG_COMPAT
1074 static long macvtap_compat_ioctl(struct file
*file
, unsigned int cmd
,
1077 return macvtap_ioctl(file
, cmd
, (unsigned long)compat_ptr(arg
));
1081 static const struct file_operations macvtap_fops
= {
1082 .owner
= THIS_MODULE
,
1083 .open
= macvtap_open
,
1084 .release
= macvtap_release
,
1085 .aio_read
= macvtap_aio_read
,
1086 .aio_write
= macvtap_aio_write
,
1087 .poll
= macvtap_poll
,
1088 .llseek
= no_llseek
,
1089 .unlocked_ioctl
= macvtap_ioctl
,
1090 #ifdef CONFIG_COMPAT
1091 .compat_ioctl
= macvtap_compat_ioctl
,
1095 static int macvtap_sendmsg(struct kiocb
*iocb
, struct socket
*sock
,
1096 struct msghdr
*m
, size_t total_len
)
1098 struct macvtap_queue
*q
= container_of(sock
, struct macvtap_queue
, sock
);
1099 return macvtap_get_user(q
, m
, m
->msg_iov
, total_len
, m
->msg_iovlen
,
1100 m
->msg_flags
& MSG_DONTWAIT
);
1103 static int macvtap_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
1104 struct msghdr
*m
, size_t total_len
,
1107 struct macvtap_queue
*q
= container_of(sock
, struct macvtap_queue
, sock
);
1109 if (flags
& ~(MSG_DONTWAIT
|MSG_TRUNC
))
1111 ret
= macvtap_do_read(q
, iocb
, m
->msg_iov
, total_len
,
1112 flags
& MSG_DONTWAIT
);
1113 if (ret
> total_len
) {
1114 m
->msg_flags
|= MSG_TRUNC
;
1115 ret
= flags
& MSG_TRUNC
? ret
: total_len
;
1120 /* Ops structure to mimic raw sockets with tun */
1121 static const struct proto_ops macvtap_socket_ops
= {
1122 .sendmsg
= macvtap_sendmsg
,
1123 .recvmsg
= macvtap_recvmsg
,
1126 /* Get an underlying socket object from tun file. Returns error unless file is
1127 * attached to a device. The returned object works like a packet socket, it
1128 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
1129 * holding a reference to the file for as long as the socket is in use. */
1130 struct socket
*macvtap_get_socket(struct file
*file
)
1132 struct macvtap_queue
*q
;
1133 if (file
->f_op
!= &macvtap_fops
)
1134 return ERR_PTR(-EINVAL
);
1135 q
= file
->private_data
;
1137 return ERR_PTR(-EBADFD
);
1140 EXPORT_SYMBOL_GPL(macvtap_get_socket
);
1142 static int macvtap_device_event(struct notifier_block
*unused
,
1143 unsigned long event
, void *ptr
)
1145 struct net_device
*dev
= netdev_notifier_info_to_dev(ptr
);
1146 struct macvlan_dev
*vlan
;
1147 struct device
*classdev
;
1151 if (dev
->rtnl_link_ops
!= &macvtap_link_ops
)
1154 vlan
= netdev_priv(dev
);
1157 case NETDEV_REGISTER
:
1158 /* Create the device node here after the network device has
1159 * been registered but before register_netdevice has
1162 err
= macvtap_get_minor(vlan
);
1164 return notifier_from_errno(err
);
1166 devt
= MKDEV(MAJOR(macvtap_major
), vlan
->minor
);
1167 classdev
= device_create(macvtap_class
, &dev
->dev
, devt
,
1168 dev
, "tap%d", dev
->ifindex
);
1169 if (IS_ERR(classdev
)) {
1170 macvtap_free_minor(vlan
);
1171 return notifier_from_errno(PTR_ERR(classdev
));
1174 case NETDEV_UNREGISTER
:
1175 devt
= MKDEV(MAJOR(macvtap_major
), vlan
->minor
);
1176 device_destroy(macvtap_class
, devt
);
1177 macvtap_free_minor(vlan
);
1184 static struct notifier_block macvtap_notifier_block __read_mostly
= {
1185 .notifier_call
= macvtap_device_event
,
1188 static int macvtap_init(void)
1192 err
= alloc_chrdev_region(&macvtap_major
, 0,
1193 MACVTAP_NUM_DEVS
, "macvtap");
1197 cdev_init(&macvtap_cdev
, &macvtap_fops
);
1198 err
= cdev_add(&macvtap_cdev
, macvtap_major
, MACVTAP_NUM_DEVS
);
1202 macvtap_class
= class_create(THIS_MODULE
, "macvtap");
1203 if (IS_ERR(macvtap_class
)) {
1204 err
= PTR_ERR(macvtap_class
);
1208 err
= register_netdevice_notifier(&macvtap_notifier_block
);
1212 err
= macvlan_link_register(&macvtap_link_ops
);
1219 unregister_netdevice_notifier(&macvtap_notifier_block
);
1221 class_unregister(macvtap_class
);
1223 cdev_del(&macvtap_cdev
);
1225 unregister_chrdev_region(macvtap_major
, MACVTAP_NUM_DEVS
);
1229 module_init(macvtap_init
);
1231 static void macvtap_exit(void)
1233 rtnl_link_unregister(&macvtap_link_ops
);
1234 unregister_netdevice_notifier(&macvtap_notifier_block
);
1235 class_unregister(macvtap_class
);
1236 cdev_del(&macvtap_cdev
);
1237 unregister_chrdev_region(macvtap_major
, MACVTAP_NUM_DEVS
);
1239 module_exit(macvtap_exit
);
1241 MODULE_ALIAS_RTNL_LINK("macvtap");
1242 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1243 MODULE_LICENSE("GPL");