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>
18 #include <net/net_namespace.h>
19 #include <net/rtnetlink.h>
21 #include <linux/virtio_net.h>
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.
29 * macvtap_proto is used to allocate queues through the sock allocation
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.
37 struct macvtap_queue
{
42 struct macvlan_dev __rcu
*vlan
;
47 static struct proto macvtap_proto
= {
50 .obj_size
= sizeof (struct macvtap_queue
),
54 * Variables for dealing with macvtaps device numbers.
56 static dev_t macvtap_major
;
57 #define MACVTAP_NUM_DEVS (1U << MINORBITS)
58 static DEFINE_MUTEX(minor_lock
);
59 static DEFINE_IDR(minor_idr
);
61 #define GOODCOPY_LEN 128
62 static struct class *macvtap_class
;
63 static struct cdev macvtap_cdev
;
65 static const struct proto_ops macvtap_socket_ops
;
69 * The macvtap_queue and the macvlan_dev are loosely coupled, the
70 * pointers from one to the other can only be read while rcu_read_lock
71 * or macvtap_lock is held.
73 * Both the file and the macvlan_dev hold a reference on the macvtap_queue
74 * through sock_hold(&q->sk). When the macvlan_dev goes away first,
75 * q->vlan becomes inaccessible. When the files gets closed,
76 * macvtap_get_queue() fails.
78 * There may still be references to the struct sock inside of the
79 * queue from outbound SKBs, but these never reference back to the
80 * file or the dev. The data structure is freed through __sk_free
81 * when both our references and any pending SKBs are gone.
83 static DEFINE_SPINLOCK(macvtap_lock
);
86 * get_slot: return a [unused/occupied] slot in vlan->taps[]:
87 * - if 'q' is NULL, return the first empty slot;
88 * - otherwise, return the slot this pointer occupies.
90 static int get_slot(struct macvlan_dev
*vlan
, struct macvtap_queue
*q
)
94 for (i
= 0; i
< MAX_MACVTAP_QUEUES
; i
++) {
95 if (rcu_dereference(vlan
->taps
[i
]) == q
)
99 /* Should never happen */
103 static int macvtap_set_queue(struct net_device
*dev
, struct file
*file
,
104 struct macvtap_queue
*q
)
106 struct macvlan_dev
*vlan
= netdev_priv(dev
);
110 spin_lock(&macvtap_lock
);
111 if (vlan
->numvtaps
== MAX_MACVTAP_QUEUES
)
115 index
= get_slot(vlan
, NULL
);
116 rcu_assign_pointer(q
->vlan
, vlan
);
117 rcu_assign_pointer(vlan
->taps
[index
], q
);
121 file
->private_data
= q
;
126 spin_unlock(&macvtap_lock
);
131 * The file owning the queue got closed, give up both
132 * the reference that the files holds as well as the
133 * one from the macvlan_dev if that still exists.
135 * Using the spinlock makes sure that we don't get
136 * to the queue again after destroying it.
138 static void macvtap_put_queue(struct macvtap_queue
*q
)
140 struct macvlan_dev
*vlan
;
142 spin_lock(&macvtap_lock
);
143 vlan
= rcu_dereference_protected(q
->vlan
,
144 lockdep_is_held(&macvtap_lock
));
146 int index
= get_slot(vlan
, q
);
148 RCU_INIT_POINTER(vlan
->taps
[index
], NULL
);
149 RCU_INIT_POINTER(q
->vlan
, NULL
);
154 spin_unlock(&macvtap_lock
);
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
167 static struct macvtap_queue
*macvtap_get_queue(struct net_device
*dev
,
170 struct macvlan_dev
*vlan
= netdev_priv(dev
);
171 struct macvtap_queue
*tap
= NULL
;
172 int numvtaps
= vlan
->numvtaps
;
178 /* Check if we can use flow to select a queue */
179 rxq
= skb_get_rxhash(skb
);
181 tap
= rcu_dereference(vlan
->taps
[rxq
% numvtaps
]);
186 if (likely(skb_rx_queue_recorded(skb
))) {
187 rxq
= skb_get_rx_queue(skb
);
189 while (unlikely(rxq
>= numvtaps
))
192 tap
= rcu_dereference(vlan
->taps
[rxq
]);
197 /* Everything failed - find first available queue */
198 for (rxq
= 0; rxq
< MAX_MACVTAP_QUEUES
; rxq
++) {
199 tap
= rcu_dereference(vlan
->taps
[rxq
]);
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.
213 static void macvtap_del_queues(struct net_device
*dev
)
215 struct macvlan_dev
*vlan
= netdev_priv(dev
);
216 struct macvtap_queue
*q
, *qlist
[MAX_MACVTAP_QUEUES
];
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_protected(vlan
->taps
[i
],
223 lockdep_is_held(&macvtap_lock
));
226 RCU_INIT_POINTER(vlan
->taps
[i
], NULL
);
227 RCU_INIT_POINTER(q
->vlan
, NULL
);
231 BUG_ON(vlan
->numvtaps
!= 0);
232 /* guarantee that any future macvtap_set_queue will fail */
233 vlan
->numvtaps
= MAX_MACVTAP_QUEUES
;
234 spin_unlock(&macvtap_lock
);
238 for (--j
; j
>= 0; j
--)
239 sock_put(&qlist
[j
]->sk
);
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.
247 static int macvtap_forward(struct net_device
*dev
, struct sk_buff
*skb
)
249 struct macvtap_queue
*q
= macvtap_get_queue(dev
, skb
);
253 if (skb_queue_len(&q
->sk
.sk_receive_queue
) >= dev
->tx_queue_len
)
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
;
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.
270 static int macvtap_receive(struct sk_buff
*skb
)
272 skb_push(skb
, ETH_HLEN
);
273 return macvtap_forward(skb
->dev
, skb
);
276 static int macvtap_get_minor(struct macvlan_dev
*vlan
)
278 int retval
= -ENOMEM
;
281 mutex_lock(&minor_lock
);
282 if (idr_pre_get(&minor_idr
, GFP_KERNEL
) == 0)
285 retval
= idr_get_new_above(&minor_idr
, vlan
, 1, &id
);
287 if (retval
== -EAGAIN
)
291 if (id
< MACVTAP_NUM_DEVS
) {
294 printk(KERN_ERR
"too many macvtap devices\n");
296 idr_remove(&minor_idr
, id
);
299 mutex_unlock(&minor_lock
);
303 static void macvtap_free_minor(struct macvlan_dev
*vlan
)
305 mutex_lock(&minor_lock
);
307 idr_remove(&minor_idr
, vlan
->minor
);
310 mutex_unlock(&minor_lock
);
313 static struct net_device
*dev_get_by_macvtap_minor(int minor
)
315 struct net_device
*dev
= NULL
;
316 struct macvlan_dev
*vlan
;
318 mutex_lock(&minor_lock
);
319 vlan
= idr_find(&minor_idr
, minor
);
324 mutex_unlock(&minor_lock
);
328 static int macvtap_newlink(struct net
*src_net
,
329 struct net_device
*dev
,
331 struct nlattr
*data
[])
333 /* Don't put anything that may fail after macvlan_common_newlink
334 * because we can't undo what it does.
336 return macvlan_common_newlink(src_net
, dev
, tb
, data
,
337 macvtap_receive
, macvtap_forward
);
340 static void macvtap_dellink(struct net_device
*dev
,
341 struct list_head
*head
)
343 macvtap_del_queues(dev
);
344 macvlan_dellink(dev
, head
);
347 static void macvtap_setup(struct net_device
*dev
)
349 macvlan_common_setup(dev
);
350 dev
->tx_queue_len
= TUN_READQ_SIZE
;
353 static struct rtnl_link_ops macvtap_link_ops __read_mostly
= {
355 .setup
= macvtap_setup
,
356 .newlink
= macvtap_newlink
,
357 .dellink
= macvtap_dellink
,
361 static void macvtap_sock_write_space(struct sock
*sk
)
363 wait_queue_head_t
*wqueue
;
365 if (!sock_writeable(sk
) ||
366 !test_and_clear_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
))
369 wqueue
= sk_sleep(sk
);
370 if (wqueue
&& waitqueue_active(wqueue
))
371 wake_up_interruptible_poll(wqueue
, POLLOUT
| POLLWRNORM
| POLLWRBAND
);
374 static void macvtap_sock_destruct(struct sock
*sk
)
376 skb_queue_purge(&sk
->sk_receive_queue
);
379 static int macvtap_open(struct inode
*inode
, struct file
*file
)
381 struct net
*net
= current
->nsproxy
->net_ns
;
382 struct net_device
*dev
= dev_get_by_macvtap_minor(iminor(inode
));
383 struct macvtap_queue
*q
;
391 q
= (struct macvtap_queue
*)sk_alloc(net
, AF_UNSPEC
, GFP_KERNEL
,
397 init_waitqueue_head(&q
->wq
.wait
);
398 q
->sock
.type
= SOCK_RAW
;
399 q
->sock
.state
= SS_CONNECTED
;
401 q
->sock
.ops
= &macvtap_socket_ops
;
402 sock_init_data(&q
->sock
, &q
->sk
);
403 q
->sk
.sk_write_space
= macvtap_sock_write_space
;
404 q
->sk
.sk_destruct
= macvtap_sock_destruct
;
405 q
->flags
= IFF_VNET_HDR
| IFF_NO_PI
| IFF_TAP
;
406 q
->vnet_hdr_sz
= sizeof(struct virtio_net_hdr
);
409 * so far only KVM virtio_net uses macvtap, enable zero copy between
410 * guest kernel and host kernel when lower device supports zerocopy
412 * The macvlan supports zerocopy iff the lower device supports zero
413 * copy so we don't have to look at the lower device directly.
415 if ((dev
->features
& NETIF_F_HIGHDMA
) && (dev
->features
& NETIF_F_SG
))
416 sock_set_flag(&q
->sk
, SOCK_ZEROCOPY
);
418 err
= macvtap_set_queue(dev
, file
, q
);
429 static int macvtap_release(struct inode
*inode
, struct file
*file
)
431 struct macvtap_queue
*q
= file
->private_data
;
432 macvtap_put_queue(q
);
436 static unsigned int macvtap_poll(struct file
*file
, poll_table
* wait
)
438 struct macvtap_queue
*q
= file
->private_data
;
439 unsigned int mask
= POLLERR
;
445 poll_wait(file
, &q
->wq
.wait
, wait
);
447 if (!skb_queue_empty(&q
->sk
.sk_receive_queue
))
448 mask
|= POLLIN
| POLLRDNORM
;
450 if (sock_writeable(&q
->sk
) ||
451 (!test_and_set_bit(SOCK_ASYNC_NOSPACE
, &q
->sock
.flags
) &&
452 sock_writeable(&q
->sk
)))
453 mask
|= POLLOUT
| POLLWRNORM
;
459 static inline struct sk_buff
*macvtap_alloc_skb(struct sock
*sk
, size_t prepad
,
460 size_t len
, size_t linear
,
461 int noblock
, int *err
)
465 /* Under a page? Don't bother with paged skb. */
466 if (prepad
+ len
< PAGE_SIZE
|| !linear
)
469 skb
= sock_alloc_send_pskb(sk
, prepad
+ linear
, len
- linear
, noblock
,
474 skb_reserve(skb
, prepad
);
475 skb_put(skb
, linear
);
476 skb
->data_len
= len
- linear
;
477 skb
->len
+= len
- linear
;
482 /* set skb frags from iovec, this can move to core network code for reuse */
483 static int zerocopy_sg_from_iovec(struct sk_buff
*skb
, const struct iovec
*from
,
484 int offset
, size_t count
)
486 int len
= iov_length(from
, count
) - offset
;
487 int copy
= skb_headlen(skb
);
488 int size
, offset1
= 0;
491 /* Skip over from offset */
492 while (count
&& (offset
>= from
->iov_len
)) {
493 offset
-= from
->iov_len
;
498 /* copy up to skb headlen */
499 while (count
&& (copy
> 0)) {
500 size
= min_t(unsigned int, copy
, from
->iov_len
- offset
);
501 if (copy_from_user(skb
->data
+ offset1
, from
->iov_base
+ offset
,
517 struct page
*page
[MAX_SKB_FRAGS
];
521 len
= from
->iov_len
- offset1
;
527 base
= (unsigned long)from
->iov_base
+ offset1
;
528 size
= ((base
& ~PAGE_MASK
) + len
+ ~PAGE_MASK
) >> PAGE_SHIFT
;
529 num_pages
= get_user_pages_fast(base
, size
, 0, &page
[i
]);
530 if ((num_pages
!= size
) ||
531 (num_pages
> MAX_SKB_FRAGS
- skb_shinfo(skb
)->nr_frags
))
532 /* put_page is in skb free */
534 skb
->data_len
+= len
;
536 skb
->truesize
+= len
;
537 atomic_add(len
, &skb
->sk
->sk_wmem_alloc
);
539 int off
= base
& ~PAGE_MASK
;
540 int size
= min_t(int, len
, PAGE_SIZE
- off
);
541 __skb_fill_page_desc(skb
, i
, page
[i
], off
, size
);
542 skb_shinfo(skb
)->nr_frags
++;
543 /* increase sk_wmem_alloc */
555 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
556 * be shared with the tun/tap driver.
558 static int macvtap_skb_from_vnet_hdr(struct sk_buff
*skb
,
559 struct virtio_net_hdr
*vnet_hdr
)
561 unsigned short gso_type
= 0;
562 if (vnet_hdr
->gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
563 switch (vnet_hdr
->gso_type
& ~VIRTIO_NET_HDR_GSO_ECN
) {
564 case VIRTIO_NET_HDR_GSO_TCPV4
:
565 gso_type
= SKB_GSO_TCPV4
;
567 case VIRTIO_NET_HDR_GSO_TCPV6
:
568 gso_type
= SKB_GSO_TCPV6
;
570 case VIRTIO_NET_HDR_GSO_UDP
:
571 gso_type
= SKB_GSO_UDP
;
577 if (vnet_hdr
->gso_type
& VIRTIO_NET_HDR_GSO_ECN
)
578 gso_type
|= SKB_GSO_TCP_ECN
;
580 if (vnet_hdr
->gso_size
== 0)
584 if (vnet_hdr
->flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) {
585 if (!skb_partial_csum_set(skb
, vnet_hdr
->csum_start
,
586 vnet_hdr
->csum_offset
))
590 if (vnet_hdr
->gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
591 skb_shinfo(skb
)->gso_size
= vnet_hdr
->gso_size
;
592 skb_shinfo(skb
)->gso_type
= gso_type
;
594 /* Header must be checked, and gso_segs computed. */
595 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
596 skb_shinfo(skb
)->gso_segs
= 0;
601 static int macvtap_skb_to_vnet_hdr(const struct sk_buff
*skb
,
602 struct virtio_net_hdr
*vnet_hdr
)
604 memset(vnet_hdr
, 0, sizeof(*vnet_hdr
));
606 if (skb_is_gso(skb
)) {
607 struct skb_shared_info
*sinfo
= skb_shinfo(skb
);
609 /* This is a hint as to how much should be linear. */
610 vnet_hdr
->hdr_len
= skb_headlen(skb
);
611 vnet_hdr
->gso_size
= sinfo
->gso_size
;
612 if (sinfo
->gso_type
& SKB_GSO_TCPV4
)
613 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_TCPV4
;
614 else if (sinfo
->gso_type
& SKB_GSO_TCPV6
)
615 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_TCPV6
;
616 else if (sinfo
->gso_type
& SKB_GSO_UDP
)
617 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_UDP
;
620 if (sinfo
->gso_type
& SKB_GSO_TCP_ECN
)
621 vnet_hdr
->gso_type
|= VIRTIO_NET_HDR_GSO_ECN
;
623 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
625 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
626 vnet_hdr
->flags
= VIRTIO_NET_HDR_F_NEEDS_CSUM
;
627 vnet_hdr
->csum_start
= skb_checksum_start_offset(skb
);
628 vnet_hdr
->csum_offset
= skb
->csum_offset
;
629 } else if (skb
->ip_summed
== CHECKSUM_UNNECESSARY
) {
630 vnet_hdr
->flags
= VIRTIO_NET_HDR_F_DATA_VALID
;
631 } /* else everything is zero */
637 /* Get packet from user space buffer */
638 static ssize_t
macvtap_get_user(struct macvtap_queue
*q
, struct msghdr
*m
,
639 const struct iovec
*iv
, unsigned long total_len
,
640 size_t count
, int noblock
)
643 struct macvlan_dev
*vlan
;
644 unsigned long len
= total_len
;
646 struct virtio_net_hdr vnet_hdr
= { 0 };
647 int vnet_hdr_len
= 0;
649 bool zerocopy
= false;
651 if (q
->flags
& IFF_VNET_HDR
) {
652 vnet_hdr_len
= q
->vnet_hdr_sz
;
655 if (len
< vnet_hdr_len
)
659 err
= memcpy_fromiovecend((void *)&vnet_hdr
, iv
, 0,
663 if ((vnet_hdr
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) &&
664 vnet_hdr
.csum_start
+ vnet_hdr
.csum_offset
+ 2 >
666 vnet_hdr
.hdr_len
= vnet_hdr
.csum_start
+
667 vnet_hdr
.csum_offset
+ 2;
669 if (vnet_hdr
.hdr_len
> len
)
674 if (unlikely(len
< ETH_HLEN
))
677 if (m
&& m
->msg_control
&& sock_flag(&q
->sk
, SOCK_ZEROCOPY
))
681 /* There are 256 bytes to be copied in skb, so there is enough
682 * room for skb expand head in case it is used.
683 * The rest buffer is mapped from userspace.
685 copylen
= vnet_hdr
.hdr_len
;
687 copylen
= GOODCOPY_LEN
;
691 skb
= macvtap_alloc_skb(&q
->sk
, NET_IP_ALIGN
, copylen
,
692 vnet_hdr
.hdr_len
, noblock
, &err
);
697 err
= zerocopy_sg_from_iovec(skb
, iv
, vnet_hdr_len
, count
);
698 skb_shinfo(skb
)->tx_flags
|= SKBTX_DEV_ZEROCOPY
;
700 err
= skb_copy_datagram_from_iovec(skb
, 0, iv
, vnet_hdr_len
,
705 skb_set_network_header(skb
, ETH_HLEN
);
706 skb_reset_mac_header(skb
);
707 skb
->protocol
= eth_hdr(skb
)->h_proto
;
710 err
= macvtap_skb_from_vnet_hdr(skb
, &vnet_hdr
);
716 vlan
= rcu_dereference_bh(q
->vlan
);
717 /* copy skb_ubuf_info for callback when skb has no error */
719 skb_shinfo(skb
)->destructor_arg
= m
->msg_control
;
721 macvlan_start_xmit(skb
, vlan
->dev
);
724 rcu_read_unlock_bh();
733 vlan
= rcu_dereference_bh(q
->vlan
);
735 vlan
->dev
->stats
.tx_dropped
++;
736 rcu_read_unlock_bh();
741 static ssize_t
macvtap_aio_write(struct kiocb
*iocb
, const struct iovec
*iv
,
742 unsigned long count
, loff_t pos
)
744 struct file
*file
= iocb
->ki_filp
;
745 ssize_t result
= -ENOLINK
;
746 struct macvtap_queue
*q
= file
->private_data
;
748 result
= macvtap_get_user(q
, NULL
, iv
, iov_length(iv
, count
), count
,
749 file
->f_flags
& O_NONBLOCK
);
753 /* Put packet to the user space buffer */
754 static ssize_t
macvtap_put_user(struct macvtap_queue
*q
,
755 const struct sk_buff
*skb
,
756 const struct iovec
*iv
, int len
)
758 struct macvlan_dev
*vlan
;
760 int vnet_hdr_len
= 0;
762 if (q
->flags
& IFF_VNET_HDR
) {
763 struct virtio_net_hdr vnet_hdr
;
764 vnet_hdr_len
= q
->vnet_hdr_sz
;
765 if ((len
-= vnet_hdr_len
) < 0)
768 ret
= macvtap_skb_to_vnet_hdr(skb
, &vnet_hdr
);
772 if (memcpy_toiovecend(iv
, (void *)&vnet_hdr
, 0, sizeof(vnet_hdr
)))
776 len
= min_t(int, skb
->len
, len
);
778 ret
= skb_copy_datagram_const_iovec(skb
, 0, iv
, vnet_hdr_len
, len
);
781 vlan
= rcu_dereference_bh(q
->vlan
);
783 macvlan_count_rx(vlan
, len
, ret
== 0, 0);
784 rcu_read_unlock_bh();
786 return ret
? ret
: (len
+ vnet_hdr_len
);
789 static ssize_t
macvtap_do_read(struct macvtap_queue
*q
, struct kiocb
*iocb
,
790 const struct iovec
*iv
, unsigned long len
,
793 DECLARE_WAITQUEUE(wait
, current
);
797 add_wait_queue(sk_sleep(&q
->sk
), &wait
);
799 current
->state
= TASK_INTERRUPTIBLE
;
801 /* Read frames from the queue */
802 skb
= skb_dequeue(&q
->sk
.sk_receive_queue
);
808 if (signal_pending(current
)) {
812 /* Nothing to read, let's sleep */
816 ret
= macvtap_put_user(q
, skb
, iv
, len
);
821 current
->state
= TASK_RUNNING
;
822 remove_wait_queue(sk_sleep(&q
->sk
), &wait
);
826 static ssize_t
macvtap_aio_read(struct kiocb
*iocb
, const struct iovec
*iv
,
827 unsigned long count
, loff_t pos
)
829 struct file
*file
= iocb
->ki_filp
;
830 struct macvtap_queue
*q
= file
->private_data
;
831 ssize_t len
, ret
= 0;
833 len
= iov_length(iv
, count
);
839 ret
= macvtap_do_read(q
, iocb
, iv
, len
, file
->f_flags
& O_NONBLOCK
);
840 ret
= min_t(ssize_t
, ret
, len
); /* XXX copied from tun.c. Why? */
846 * provide compatibility with generic tun/tap interface
848 static long macvtap_ioctl(struct file
*file
, unsigned int cmd
,
851 struct macvtap_queue
*q
= file
->private_data
;
852 struct macvlan_dev
*vlan
;
853 void __user
*argp
= (void __user
*)arg
;
854 struct ifreq __user
*ifr
= argp
;
855 unsigned int __user
*up
= argp
;
857 int __user
*sp
= argp
;
863 /* ignore the name, just look at flags */
864 if (get_user(u
, &ifr
->ifr_flags
))
868 if ((u
& ~IFF_VNET_HDR
) != (IFF_NO_PI
| IFF_TAP
))
877 vlan
= rcu_dereference_bh(q
->vlan
);
880 rcu_read_unlock_bh();
886 if (copy_to_user(&ifr
->ifr_name
, vlan
->dev
->name
, IFNAMSIZ
) ||
887 put_user(q
->flags
, &ifr
->ifr_flags
))
893 if (put_user(IFF_TAP
| IFF_NO_PI
| IFF_VNET_HDR
, up
))
904 case TUNGETVNETHDRSZ
:
910 case TUNSETVNETHDRSZ
:
913 if (s
< (int)sizeof(struct virtio_net_hdr
))
920 /* let the user check for future flags */
921 if (arg
& ~(TUN_F_CSUM
| TUN_F_TSO4
| TUN_F_TSO6
|
922 TUN_F_TSO_ECN
| TUN_F_UFO
))
925 /* TODO: only accept frames with the features that
926 got enabled for forwarded frames */
927 if (!(q
->flags
& IFF_VNET_HDR
))
937 static long macvtap_compat_ioctl(struct file
*file
, unsigned int cmd
,
940 return macvtap_ioctl(file
, cmd
, (unsigned long)compat_ptr(arg
));
944 static const struct file_operations macvtap_fops
= {
945 .owner
= THIS_MODULE
,
946 .open
= macvtap_open
,
947 .release
= macvtap_release
,
948 .aio_read
= macvtap_aio_read
,
949 .aio_write
= macvtap_aio_write
,
950 .poll
= macvtap_poll
,
952 .unlocked_ioctl
= macvtap_ioctl
,
954 .compat_ioctl
= macvtap_compat_ioctl
,
958 static int macvtap_sendmsg(struct kiocb
*iocb
, struct socket
*sock
,
959 struct msghdr
*m
, size_t total_len
)
961 struct macvtap_queue
*q
= container_of(sock
, struct macvtap_queue
, sock
);
962 return macvtap_get_user(q
, m
, m
->msg_iov
, total_len
, m
->msg_iovlen
,
963 m
->msg_flags
& MSG_DONTWAIT
);
966 static int macvtap_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
967 struct msghdr
*m
, size_t total_len
,
970 struct macvtap_queue
*q
= container_of(sock
, struct macvtap_queue
, sock
);
972 if (flags
& ~(MSG_DONTWAIT
|MSG_TRUNC
))
974 ret
= macvtap_do_read(q
, iocb
, m
->msg_iov
, total_len
,
975 flags
& MSG_DONTWAIT
);
976 if (ret
> total_len
) {
977 m
->msg_flags
|= MSG_TRUNC
;
978 ret
= flags
& MSG_TRUNC
? ret
: total_len
;
983 /* Ops structure to mimic raw sockets with tun */
984 static const struct proto_ops macvtap_socket_ops
= {
985 .sendmsg
= macvtap_sendmsg
,
986 .recvmsg
= macvtap_recvmsg
,
989 /* Get an underlying socket object from tun file. Returns error unless file is
990 * attached to a device. The returned object works like a packet socket, it
991 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
992 * holding a reference to the file for as long as the socket is in use. */
993 struct socket
*macvtap_get_socket(struct file
*file
)
995 struct macvtap_queue
*q
;
996 if (file
->f_op
!= &macvtap_fops
)
997 return ERR_PTR(-EINVAL
);
998 q
= file
->private_data
;
1000 return ERR_PTR(-EBADFD
);
1003 EXPORT_SYMBOL_GPL(macvtap_get_socket
);
1005 static int macvtap_device_event(struct notifier_block
*unused
,
1006 unsigned long event
, void *ptr
)
1008 struct net_device
*dev
= ptr
;
1009 struct macvlan_dev
*vlan
;
1010 struct device
*classdev
;
1014 if (dev
->rtnl_link_ops
!= &macvtap_link_ops
)
1017 vlan
= netdev_priv(dev
);
1020 case NETDEV_REGISTER
:
1021 /* Create the device node here after the network device has
1022 * been registered but before register_netdevice has
1025 err
= macvtap_get_minor(vlan
);
1027 return notifier_from_errno(err
);
1029 devt
= MKDEV(MAJOR(macvtap_major
), vlan
->minor
);
1030 classdev
= device_create(macvtap_class
, &dev
->dev
, devt
,
1031 dev
, "tap%d", dev
->ifindex
);
1032 if (IS_ERR(classdev
)) {
1033 macvtap_free_minor(vlan
);
1034 return notifier_from_errno(PTR_ERR(classdev
));
1037 case NETDEV_UNREGISTER
:
1038 devt
= MKDEV(MAJOR(macvtap_major
), vlan
->minor
);
1039 device_destroy(macvtap_class
, devt
);
1040 macvtap_free_minor(vlan
);
1047 static struct notifier_block macvtap_notifier_block __read_mostly
= {
1048 .notifier_call
= macvtap_device_event
,
1051 static int macvtap_init(void)
1055 err
= alloc_chrdev_region(&macvtap_major
, 0,
1056 MACVTAP_NUM_DEVS
, "macvtap");
1060 cdev_init(&macvtap_cdev
, &macvtap_fops
);
1061 err
= cdev_add(&macvtap_cdev
, macvtap_major
, MACVTAP_NUM_DEVS
);
1065 macvtap_class
= class_create(THIS_MODULE
, "macvtap");
1066 if (IS_ERR(macvtap_class
)) {
1067 err
= PTR_ERR(macvtap_class
);
1071 err
= register_netdevice_notifier(&macvtap_notifier_block
);
1075 err
= macvlan_link_register(&macvtap_link_ops
);
1082 unregister_netdevice_notifier(&macvtap_notifier_block
);
1084 class_unregister(macvtap_class
);
1086 cdev_del(&macvtap_cdev
);
1088 unregister_chrdev_region(macvtap_major
, MACVTAP_NUM_DEVS
);
1092 module_init(macvtap_init
);
1094 static void macvtap_exit(void)
1096 rtnl_link_unregister(&macvtap_link_ops
);
1097 unregister_netdevice_notifier(&macvtap_notifier_block
);
1098 class_unregister(macvtap_class
);
1099 cdev_del(&macvtap_cdev
);
1100 unregister_chrdev_region(macvtap_major
, MACVTAP_NUM_DEVS
);
1102 module_exit(macvtap_exit
);
1104 MODULE_ALIAS_RTNL_LINK("macvtap");
1105 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1106 MODULE_LICENSE("GPL");