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/idr.h>
19 #include <net/net_namespace.h>
20 #include <net/rtnetlink.h>
22 #include <linux/virtio_net.h>
25 * A macvtap queue is the central object of this driver, it connects
26 * an open character device to a macvlan interface. There can be
27 * multiple queues on one interface, which map back to queues
28 * implemented in hardware on the underlying device.
30 * macvtap_proto is used to allocate queues through the sock allocation
33 * TODO: multiqueue support is currently not implemented, even though
34 * macvtap is basically prepared for that. We will need to add this
35 * here as well as in virtio-net and qemu to get line rate on 10gbit
36 * adapters from a guest.
38 struct macvtap_queue
{
43 struct macvlan_dev __rcu
*vlan
;
48 static struct proto macvtap_proto
= {
51 .obj_size
= sizeof (struct macvtap_queue
),
55 * Variables for dealing with macvtaps device numbers.
57 static dev_t macvtap_major
;
58 #define MACVTAP_NUM_DEVS (1U << MINORBITS)
59 static DEFINE_MUTEX(minor_lock
);
60 static DEFINE_IDR(minor_idr
);
62 #define GOODCOPY_LEN 128
63 static struct class *macvtap_class
;
64 static struct cdev macvtap_cdev
;
66 static const struct proto_ops macvtap_socket_ops
;
70 * The macvtap_queue and the macvlan_dev are loosely coupled, the
71 * pointers from one to the other can only be read while rcu_read_lock
72 * or macvtap_lock is held.
74 * Both the file and the macvlan_dev hold a reference on the macvtap_queue
75 * through sock_hold(&q->sk). When the macvlan_dev goes away first,
76 * q->vlan becomes inaccessible. When the files gets closed,
77 * macvtap_get_queue() fails.
79 * There may still be references to the struct sock inside of the
80 * queue from outbound SKBs, but these never reference back to the
81 * file or the dev. The data structure is freed through __sk_free
82 * when both our references and any pending SKBs are gone.
84 static DEFINE_SPINLOCK(macvtap_lock
);
87 * get_slot: return a [unused/occupied] slot in vlan->taps[]:
88 * - if 'q' is NULL, return the first empty slot;
89 * - otherwise, return the slot this pointer occupies.
91 static int get_slot(struct macvlan_dev
*vlan
, struct macvtap_queue
*q
)
95 for (i
= 0; i
< MAX_MACVTAP_QUEUES
; i
++) {
96 if (rcu_dereference(vlan
->taps
[i
]) == q
)
100 /* Should never happen */
104 static int macvtap_set_queue(struct net_device
*dev
, struct file
*file
,
105 struct macvtap_queue
*q
)
107 struct macvlan_dev
*vlan
= netdev_priv(dev
);
111 spin_lock(&macvtap_lock
);
112 if (vlan
->numvtaps
== MAX_MACVTAP_QUEUES
)
116 index
= get_slot(vlan
, NULL
);
117 rcu_assign_pointer(q
->vlan
, vlan
);
118 rcu_assign_pointer(vlan
->taps
[index
], q
);
122 file
->private_data
= q
;
127 spin_unlock(&macvtap_lock
);
132 * The file owning the queue got closed, give up both
133 * the reference that the files holds as well as the
134 * one from the macvlan_dev if that still exists.
136 * Using the spinlock makes sure that we don't get
137 * to the queue again after destroying it.
139 static void macvtap_put_queue(struct macvtap_queue
*q
)
141 struct macvlan_dev
*vlan
;
143 spin_lock(&macvtap_lock
);
144 vlan
= rcu_dereference_protected(q
->vlan
,
145 lockdep_is_held(&macvtap_lock
));
147 int index
= get_slot(vlan
, q
);
149 RCU_INIT_POINTER(vlan
->taps
[index
], NULL
);
150 RCU_INIT_POINTER(q
->vlan
, NULL
);
155 spin_unlock(&macvtap_lock
);
162 * Select a queue based on the rxq of the device on which this packet
163 * arrived. If the incoming device is not mq, calculate a flow hash
164 * to select a queue. If all fails, find the first available queue.
165 * Cache vlan->numvtaps since it can become zero during the execution
168 static struct macvtap_queue
*macvtap_get_queue(struct net_device
*dev
,
171 struct macvlan_dev
*vlan
= netdev_priv(dev
);
172 struct macvtap_queue
*tap
= NULL
;
173 int numvtaps
= vlan
->numvtaps
;
179 /* Check if we can use flow to select a queue */
180 rxq
= skb_get_rxhash(skb
);
182 tap
= rcu_dereference(vlan
->taps
[rxq
% numvtaps
]);
187 if (likely(skb_rx_queue_recorded(skb
))) {
188 rxq
= skb_get_rx_queue(skb
);
190 while (unlikely(rxq
>= numvtaps
))
193 tap
= rcu_dereference(vlan
->taps
[rxq
]);
198 /* Everything failed - find first available queue */
199 for (rxq
= 0; rxq
< MAX_MACVTAP_QUEUES
; rxq
++) {
200 tap
= rcu_dereference(vlan
->taps
[rxq
]);
210 * The net_device is going away, give up the reference
211 * that it holds on all queues and safely set the pointer
212 * from the queues to NULL.
214 static void macvtap_del_queues(struct net_device
*dev
)
216 struct macvlan_dev
*vlan
= netdev_priv(dev
);
217 struct macvtap_queue
*q
, *qlist
[MAX_MACVTAP_QUEUES
];
220 /* macvtap_put_queue can free some slots, so go through all slots */
221 spin_lock(&macvtap_lock
);
222 for (i
= 0; i
< MAX_MACVTAP_QUEUES
&& vlan
->numvtaps
; i
++) {
223 q
= rcu_dereference_protected(vlan
->taps
[i
],
224 lockdep_is_held(&macvtap_lock
));
227 RCU_INIT_POINTER(vlan
->taps
[i
], NULL
);
228 RCU_INIT_POINTER(q
->vlan
, NULL
);
232 BUG_ON(vlan
->numvtaps
!= 0);
233 /* guarantee that any future macvtap_set_queue will fail */
234 vlan
->numvtaps
= MAX_MACVTAP_QUEUES
;
235 spin_unlock(&macvtap_lock
);
239 for (--j
; j
>= 0; j
--)
240 sock_put(&qlist
[j
]->sk
);
244 * Forward happens for data that gets sent from one macvlan
245 * endpoint to another one in bridge mode. We just take
246 * the skb and put it into the receive queue.
248 static int macvtap_forward(struct net_device
*dev
, struct sk_buff
*skb
)
250 struct macvtap_queue
*q
= macvtap_get_queue(dev
, skb
);
254 if (skb_queue_len(&q
->sk
.sk_receive_queue
) >= dev
->tx_queue_len
)
257 skb_queue_tail(&q
->sk
.sk_receive_queue
, skb
);
258 wake_up_interruptible_poll(sk_sleep(&q
->sk
), POLLIN
| POLLRDNORM
| POLLRDBAND
);
259 return NET_RX_SUCCESS
;
267 * Receive is for data from the external interface (lowerdev),
268 * in case of macvtap, we can treat that the same way as
269 * forward, which macvlan cannot.
271 static int macvtap_receive(struct sk_buff
*skb
)
273 skb_push(skb
, ETH_HLEN
);
274 return macvtap_forward(skb
->dev
, skb
);
277 static int macvtap_get_minor(struct macvlan_dev
*vlan
)
279 int retval
= -ENOMEM
;
282 mutex_lock(&minor_lock
);
283 if (idr_pre_get(&minor_idr
, GFP_KERNEL
) == 0)
286 retval
= idr_get_new_above(&minor_idr
, vlan
, 1, &id
);
288 if (retval
== -EAGAIN
)
292 if (id
< MACVTAP_NUM_DEVS
) {
295 printk(KERN_ERR
"too many macvtap devices\n");
297 idr_remove(&minor_idr
, id
);
300 mutex_unlock(&minor_lock
);
304 static void macvtap_free_minor(struct macvlan_dev
*vlan
)
306 mutex_lock(&minor_lock
);
308 idr_remove(&minor_idr
, vlan
->minor
);
311 mutex_unlock(&minor_lock
);
314 static struct net_device
*dev_get_by_macvtap_minor(int minor
)
316 struct net_device
*dev
= NULL
;
317 struct macvlan_dev
*vlan
;
319 mutex_lock(&minor_lock
);
320 vlan
= idr_find(&minor_idr
, minor
);
325 mutex_unlock(&minor_lock
);
329 static int macvtap_newlink(struct net
*src_net
,
330 struct net_device
*dev
,
332 struct nlattr
*data
[])
334 /* Don't put anything that may fail after macvlan_common_newlink
335 * because we can't undo what it does.
337 return macvlan_common_newlink(src_net
, dev
, tb
, data
,
338 macvtap_receive
, macvtap_forward
);
341 static void macvtap_dellink(struct net_device
*dev
,
342 struct list_head
*head
)
344 macvtap_del_queues(dev
);
345 macvlan_dellink(dev
, head
);
348 static void macvtap_setup(struct net_device
*dev
)
350 macvlan_common_setup(dev
);
351 dev
->tx_queue_len
= TUN_READQ_SIZE
;
354 static struct rtnl_link_ops macvtap_link_ops __read_mostly
= {
356 .setup
= macvtap_setup
,
357 .newlink
= macvtap_newlink
,
358 .dellink
= macvtap_dellink
,
362 static void macvtap_sock_write_space(struct sock
*sk
)
364 wait_queue_head_t
*wqueue
;
366 if (!sock_writeable(sk
) ||
367 !test_and_clear_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
))
370 wqueue
= sk_sleep(sk
);
371 if (wqueue
&& waitqueue_active(wqueue
))
372 wake_up_interruptible_poll(wqueue
, POLLOUT
| POLLWRNORM
| POLLWRBAND
);
375 static void macvtap_sock_destruct(struct sock
*sk
)
377 skb_queue_purge(&sk
->sk_receive_queue
);
380 static int macvtap_open(struct inode
*inode
, struct file
*file
)
382 struct net
*net
= current
->nsproxy
->net_ns
;
383 struct net_device
*dev
= dev_get_by_macvtap_minor(iminor(inode
));
384 struct macvtap_queue
*q
;
392 q
= (struct macvtap_queue
*)sk_alloc(net
, AF_UNSPEC
, GFP_KERNEL
,
398 init_waitqueue_head(&q
->wq
.wait
);
399 q
->sock
.type
= SOCK_RAW
;
400 q
->sock
.state
= SS_CONNECTED
;
402 q
->sock
.ops
= &macvtap_socket_ops
;
403 sock_init_data(&q
->sock
, &q
->sk
);
404 q
->sk
.sk_write_space
= macvtap_sock_write_space
;
405 q
->sk
.sk_destruct
= macvtap_sock_destruct
;
406 q
->flags
= IFF_VNET_HDR
| IFF_NO_PI
| IFF_TAP
;
407 q
->vnet_hdr_sz
= sizeof(struct virtio_net_hdr
);
410 * so far only KVM virtio_net uses macvtap, enable zero copy between
411 * guest kernel and host kernel when lower device supports zerocopy
413 * The macvlan supports zerocopy iff the lower device supports zero
414 * copy so we don't have to look at the lower device directly.
416 if ((dev
->features
& NETIF_F_HIGHDMA
) && (dev
->features
& NETIF_F_SG
))
417 sock_set_flag(&q
->sk
, SOCK_ZEROCOPY
);
419 err
= macvtap_set_queue(dev
, file
, q
);
430 static int macvtap_release(struct inode
*inode
, struct file
*file
)
432 struct macvtap_queue
*q
= file
->private_data
;
433 macvtap_put_queue(q
);
437 static unsigned int macvtap_poll(struct file
*file
, poll_table
* wait
)
439 struct macvtap_queue
*q
= file
->private_data
;
440 unsigned int mask
= POLLERR
;
446 poll_wait(file
, &q
->wq
.wait
, wait
);
448 if (!skb_queue_empty(&q
->sk
.sk_receive_queue
))
449 mask
|= POLLIN
| POLLRDNORM
;
451 if (sock_writeable(&q
->sk
) ||
452 (!test_and_set_bit(SOCK_ASYNC_NOSPACE
, &q
->sock
.flags
) &&
453 sock_writeable(&q
->sk
)))
454 mask
|= POLLOUT
| POLLWRNORM
;
460 static inline struct sk_buff
*macvtap_alloc_skb(struct sock
*sk
, size_t prepad
,
461 size_t len
, size_t linear
,
462 int noblock
, int *err
)
466 /* Under a page? Don't bother with paged skb. */
467 if (prepad
+ len
< PAGE_SIZE
|| !linear
)
470 skb
= sock_alloc_send_pskb(sk
, prepad
+ linear
, len
- linear
, noblock
,
475 skb_reserve(skb
, prepad
);
476 skb_put(skb
, linear
);
477 skb
->data_len
= len
- linear
;
478 skb
->len
+= len
- linear
;
483 /* set skb frags from iovec, this can move to core network code for reuse */
484 static int zerocopy_sg_from_iovec(struct sk_buff
*skb
, const struct iovec
*from
,
485 int offset
, size_t count
)
487 int len
= iov_length(from
, count
) - offset
;
488 int copy
= skb_headlen(skb
);
489 int size
, offset1
= 0;
492 /* Skip over from offset */
493 while (count
&& (offset
>= from
->iov_len
)) {
494 offset
-= from
->iov_len
;
499 /* copy up to skb headlen */
500 while (count
&& (copy
> 0)) {
501 size
= min_t(unsigned int, copy
, from
->iov_len
- offset
);
502 if (copy_from_user(skb
->data
+ offset1
, from
->iov_base
+ offset
,
518 struct page
*page
[MAX_SKB_FRAGS
];
522 len
= from
->iov_len
- offset1
;
528 base
= (unsigned long)from
->iov_base
+ offset1
;
529 size
= ((base
& ~PAGE_MASK
) + len
+ ~PAGE_MASK
) >> PAGE_SHIFT
;
530 num_pages
= get_user_pages_fast(base
, size
, 0, &page
[i
]);
531 if ((num_pages
!= size
) ||
532 (num_pages
> MAX_SKB_FRAGS
- skb_shinfo(skb
)->nr_frags
))
533 /* put_page is in skb free */
535 skb
->data_len
+= len
;
537 skb
->truesize
+= len
;
538 atomic_add(len
, &skb
->sk
->sk_wmem_alloc
);
540 int off
= base
& ~PAGE_MASK
;
541 int size
= min_t(int, len
, PAGE_SIZE
- off
);
542 __skb_fill_page_desc(skb
, i
, page
[i
], off
, size
);
543 skb_shinfo(skb
)->nr_frags
++;
544 /* increase sk_wmem_alloc */
556 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
557 * be shared with the tun/tap driver.
559 static int macvtap_skb_from_vnet_hdr(struct sk_buff
*skb
,
560 struct virtio_net_hdr
*vnet_hdr
)
562 unsigned short gso_type
= 0;
563 if (vnet_hdr
->gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
564 switch (vnet_hdr
->gso_type
& ~VIRTIO_NET_HDR_GSO_ECN
) {
565 case VIRTIO_NET_HDR_GSO_TCPV4
:
566 gso_type
= SKB_GSO_TCPV4
;
568 case VIRTIO_NET_HDR_GSO_TCPV6
:
569 gso_type
= SKB_GSO_TCPV6
;
571 case VIRTIO_NET_HDR_GSO_UDP
:
572 gso_type
= SKB_GSO_UDP
;
578 if (vnet_hdr
->gso_type
& VIRTIO_NET_HDR_GSO_ECN
)
579 gso_type
|= SKB_GSO_TCP_ECN
;
581 if (vnet_hdr
->gso_size
== 0)
585 if (vnet_hdr
->flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) {
586 if (!skb_partial_csum_set(skb
, vnet_hdr
->csum_start
,
587 vnet_hdr
->csum_offset
))
591 if (vnet_hdr
->gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
592 skb_shinfo(skb
)->gso_size
= vnet_hdr
->gso_size
;
593 skb_shinfo(skb
)->gso_type
= gso_type
;
595 /* Header must be checked, and gso_segs computed. */
596 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
597 skb_shinfo(skb
)->gso_segs
= 0;
602 static int macvtap_skb_to_vnet_hdr(const struct sk_buff
*skb
,
603 struct virtio_net_hdr
*vnet_hdr
)
605 memset(vnet_hdr
, 0, sizeof(*vnet_hdr
));
607 if (skb_is_gso(skb
)) {
608 struct skb_shared_info
*sinfo
= skb_shinfo(skb
);
610 /* This is a hint as to how much should be linear. */
611 vnet_hdr
->hdr_len
= skb_headlen(skb
);
612 vnet_hdr
->gso_size
= sinfo
->gso_size
;
613 if (sinfo
->gso_type
& SKB_GSO_TCPV4
)
614 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_TCPV4
;
615 else if (sinfo
->gso_type
& SKB_GSO_TCPV6
)
616 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_TCPV6
;
617 else if (sinfo
->gso_type
& SKB_GSO_UDP
)
618 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_UDP
;
621 if (sinfo
->gso_type
& SKB_GSO_TCP_ECN
)
622 vnet_hdr
->gso_type
|= VIRTIO_NET_HDR_GSO_ECN
;
624 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
626 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
627 vnet_hdr
->flags
= VIRTIO_NET_HDR_F_NEEDS_CSUM
;
628 vnet_hdr
->csum_start
= skb_checksum_start_offset(skb
);
629 vnet_hdr
->csum_offset
= skb
->csum_offset
;
630 } else if (skb
->ip_summed
== CHECKSUM_UNNECESSARY
) {
631 vnet_hdr
->flags
= VIRTIO_NET_HDR_F_DATA_VALID
;
632 } /* else everything is zero */
638 /* Get packet from user space buffer */
639 static ssize_t
macvtap_get_user(struct macvtap_queue
*q
, struct msghdr
*m
,
640 const struct iovec
*iv
, unsigned long total_len
,
641 size_t count
, int noblock
)
644 struct macvlan_dev
*vlan
;
645 unsigned long len
= total_len
;
647 struct virtio_net_hdr vnet_hdr
= { 0 };
648 int vnet_hdr_len
= 0;
650 bool zerocopy
= false;
652 if (q
->flags
& IFF_VNET_HDR
) {
653 vnet_hdr_len
= q
->vnet_hdr_sz
;
656 if (len
< vnet_hdr_len
)
660 err
= memcpy_fromiovecend((void *)&vnet_hdr
, iv
, 0,
664 if ((vnet_hdr
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) &&
665 vnet_hdr
.csum_start
+ vnet_hdr
.csum_offset
+ 2 >
667 vnet_hdr
.hdr_len
= vnet_hdr
.csum_start
+
668 vnet_hdr
.csum_offset
+ 2;
670 if (vnet_hdr
.hdr_len
> len
)
675 if (unlikely(len
< ETH_HLEN
))
678 if (m
&& m
->msg_control
&& sock_flag(&q
->sk
, SOCK_ZEROCOPY
))
682 /* There are 256 bytes to be copied in skb, so there is enough
683 * room for skb expand head in case it is used.
684 * The rest buffer is mapped from userspace.
686 copylen
= vnet_hdr
.hdr_len
;
688 copylen
= GOODCOPY_LEN
;
692 skb
= macvtap_alloc_skb(&q
->sk
, NET_IP_ALIGN
, copylen
,
693 vnet_hdr
.hdr_len
, noblock
, &err
);
698 err
= zerocopy_sg_from_iovec(skb
, iv
, vnet_hdr_len
, count
);
699 skb_shinfo(skb
)->tx_flags
|= SKBTX_DEV_ZEROCOPY
;
701 err
= skb_copy_datagram_from_iovec(skb
, 0, iv
, vnet_hdr_len
,
706 skb_set_network_header(skb
, ETH_HLEN
);
707 skb_reset_mac_header(skb
);
708 skb
->protocol
= eth_hdr(skb
)->h_proto
;
711 err
= macvtap_skb_from_vnet_hdr(skb
, &vnet_hdr
);
717 vlan
= rcu_dereference_bh(q
->vlan
);
718 /* copy skb_ubuf_info for callback when skb has no error */
720 skb_shinfo(skb
)->destructor_arg
= m
->msg_control
;
722 macvlan_start_xmit(skb
, vlan
->dev
);
725 rcu_read_unlock_bh();
734 vlan
= rcu_dereference_bh(q
->vlan
);
736 vlan
->dev
->stats
.tx_dropped
++;
737 rcu_read_unlock_bh();
742 static ssize_t
macvtap_aio_write(struct kiocb
*iocb
, const struct iovec
*iv
,
743 unsigned long count
, loff_t pos
)
745 struct file
*file
= iocb
->ki_filp
;
746 ssize_t result
= -ENOLINK
;
747 struct macvtap_queue
*q
= file
->private_data
;
749 result
= macvtap_get_user(q
, NULL
, iv
, iov_length(iv
, count
), count
,
750 file
->f_flags
& O_NONBLOCK
);
754 /* Put packet to the user space buffer */
755 static ssize_t
macvtap_put_user(struct macvtap_queue
*q
,
756 const struct sk_buff
*skb
,
757 const struct iovec
*iv
, int len
)
759 struct macvlan_dev
*vlan
;
761 int vnet_hdr_len
= 0;
763 if (q
->flags
& IFF_VNET_HDR
) {
764 struct virtio_net_hdr vnet_hdr
;
765 vnet_hdr_len
= q
->vnet_hdr_sz
;
766 if ((len
-= vnet_hdr_len
) < 0)
769 ret
= macvtap_skb_to_vnet_hdr(skb
, &vnet_hdr
);
773 if (memcpy_toiovecend(iv
, (void *)&vnet_hdr
, 0, sizeof(vnet_hdr
)))
777 len
= min_t(int, skb
->len
, len
);
779 ret
= skb_copy_datagram_const_iovec(skb
, 0, iv
, vnet_hdr_len
, len
);
782 vlan
= rcu_dereference_bh(q
->vlan
);
784 macvlan_count_rx(vlan
, len
, ret
== 0, 0);
785 rcu_read_unlock_bh();
787 return ret
? ret
: (len
+ vnet_hdr_len
);
790 static ssize_t
macvtap_do_read(struct macvtap_queue
*q
, struct kiocb
*iocb
,
791 const struct iovec
*iv
, unsigned long len
,
794 DECLARE_WAITQUEUE(wait
, current
);
798 add_wait_queue(sk_sleep(&q
->sk
), &wait
);
800 current
->state
= TASK_INTERRUPTIBLE
;
802 /* Read frames from the queue */
803 skb
= skb_dequeue(&q
->sk
.sk_receive_queue
);
809 if (signal_pending(current
)) {
813 /* Nothing to read, let's sleep */
817 ret
= macvtap_put_user(q
, skb
, iv
, len
);
822 current
->state
= TASK_RUNNING
;
823 remove_wait_queue(sk_sleep(&q
->sk
), &wait
);
827 static ssize_t
macvtap_aio_read(struct kiocb
*iocb
, const struct iovec
*iv
,
828 unsigned long count
, loff_t pos
)
830 struct file
*file
= iocb
->ki_filp
;
831 struct macvtap_queue
*q
= file
->private_data
;
832 ssize_t len
, ret
= 0;
834 len
= iov_length(iv
, count
);
840 ret
= macvtap_do_read(q
, iocb
, iv
, len
, file
->f_flags
& O_NONBLOCK
);
841 ret
= min_t(ssize_t
, ret
, len
); /* XXX copied from tun.c. Why? */
847 * provide compatibility with generic tun/tap interface
849 static long macvtap_ioctl(struct file
*file
, unsigned int cmd
,
852 struct macvtap_queue
*q
= file
->private_data
;
853 struct macvlan_dev
*vlan
;
854 void __user
*argp
= (void __user
*)arg
;
855 struct ifreq __user
*ifr
= argp
;
856 unsigned int __user
*up
= argp
;
858 int __user
*sp
= argp
;
864 /* ignore the name, just look at flags */
865 if (get_user(u
, &ifr
->ifr_flags
))
869 if ((u
& ~IFF_VNET_HDR
) != (IFF_NO_PI
| IFF_TAP
))
878 vlan
= rcu_dereference_bh(q
->vlan
);
881 rcu_read_unlock_bh();
887 if (copy_to_user(&ifr
->ifr_name
, vlan
->dev
->name
, IFNAMSIZ
) ||
888 put_user(q
->flags
, &ifr
->ifr_flags
))
894 if (put_user(IFF_TAP
| IFF_NO_PI
| IFF_VNET_HDR
, up
))
905 case TUNGETVNETHDRSZ
:
911 case TUNSETVNETHDRSZ
:
914 if (s
< (int)sizeof(struct virtio_net_hdr
))
921 /* let the user check for future flags */
922 if (arg
& ~(TUN_F_CSUM
| TUN_F_TSO4
| TUN_F_TSO6
|
923 TUN_F_TSO_ECN
| TUN_F_UFO
))
926 /* TODO: only accept frames with the features that
927 got enabled for forwarded frames */
928 if (!(q
->flags
& IFF_VNET_HDR
))
938 static long macvtap_compat_ioctl(struct file
*file
, unsigned int cmd
,
941 return macvtap_ioctl(file
, cmd
, (unsigned long)compat_ptr(arg
));
945 static const struct file_operations macvtap_fops
= {
946 .owner
= THIS_MODULE
,
947 .open
= macvtap_open
,
948 .release
= macvtap_release
,
949 .aio_read
= macvtap_aio_read
,
950 .aio_write
= macvtap_aio_write
,
951 .poll
= macvtap_poll
,
953 .unlocked_ioctl
= macvtap_ioctl
,
955 .compat_ioctl
= macvtap_compat_ioctl
,
959 static int macvtap_sendmsg(struct kiocb
*iocb
, struct socket
*sock
,
960 struct msghdr
*m
, size_t total_len
)
962 struct macvtap_queue
*q
= container_of(sock
, struct macvtap_queue
, sock
);
963 return macvtap_get_user(q
, m
, m
->msg_iov
, total_len
, m
->msg_iovlen
,
964 m
->msg_flags
& MSG_DONTWAIT
);
967 static int macvtap_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
968 struct msghdr
*m
, size_t total_len
,
971 struct macvtap_queue
*q
= container_of(sock
, struct macvtap_queue
, sock
);
973 if (flags
& ~(MSG_DONTWAIT
|MSG_TRUNC
))
975 ret
= macvtap_do_read(q
, iocb
, m
->msg_iov
, total_len
,
976 flags
& MSG_DONTWAIT
);
977 if (ret
> total_len
) {
978 m
->msg_flags
|= MSG_TRUNC
;
979 ret
= flags
& MSG_TRUNC
? ret
: total_len
;
984 /* Ops structure to mimic raw sockets with tun */
985 static const struct proto_ops macvtap_socket_ops
= {
986 .sendmsg
= macvtap_sendmsg
,
987 .recvmsg
= macvtap_recvmsg
,
990 /* Get an underlying socket object from tun file. Returns error unless file is
991 * attached to a device. The returned object works like a packet socket, it
992 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
993 * holding a reference to the file for as long as the socket is in use. */
994 struct socket
*macvtap_get_socket(struct file
*file
)
996 struct macvtap_queue
*q
;
997 if (file
->f_op
!= &macvtap_fops
)
998 return ERR_PTR(-EINVAL
);
999 q
= file
->private_data
;
1001 return ERR_PTR(-EBADFD
);
1004 EXPORT_SYMBOL_GPL(macvtap_get_socket
);
1006 static int macvtap_device_event(struct notifier_block
*unused
,
1007 unsigned long event
, void *ptr
)
1009 struct net_device
*dev
= ptr
;
1010 struct macvlan_dev
*vlan
;
1011 struct device
*classdev
;
1015 if (dev
->rtnl_link_ops
!= &macvtap_link_ops
)
1018 vlan
= netdev_priv(dev
);
1021 case NETDEV_REGISTER
:
1022 /* Create the device node here after the network device has
1023 * been registered but before register_netdevice has
1026 err
= macvtap_get_minor(vlan
);
1028 return notifier_from_errno(err
);
1030 devt
= MKDEV(MAJOR(macvtap_major
), vlan
->minor
);
1031 classdev
= device_create(macvtap_class
, &dev
->dev
, devt
,
1032 dev
, "tap%d", dev
->ifindex
);
1033 if (IS_ERR(classdev
)) {
1034 macvtap_free_minor(vlan
);
1035 return notifier_from_errno(PTR_ERR(classdev
));
1038 case NETDEV_UNREGISTER
:
1039 devt
= MKDEV(MAJOR(macvtap_major
), vlan
->minor
);
1040 device_destroy(macvtap_class
, devt
);
1041 macvtap_free_minor(vlan
);
1048 static struct notifier_block macvtap_notifier_block __read_mostly
= {
1049 .notifier_call
= macvtap_device_event
,
1052 static int macvtap_init(void)
1056 err
= alloc_chrdev_region(&macvtap_major
, 0,
1057 MACVTAP_NUM_DEVS
, "macvtap");
1061 cdev_init(&macvtap_cdev
, &macvtap_fops
);
1062 err
= cdev_add(&macvtap_cdev
, macvtap_major
, MACVTAP_NUM_DEVS
);
1066 macvtap_class
= class_create(THIS_MODULE
, "macvtap");
1067 if (IS_ERR(macvtap_class
)) {
1068 err
= PTR_ERR(macvtap_class
);
1072 err
= register_netdevice_notifier(&macvtap_notifier_block
);
1076 err
= macvlan_link_register(&macvtap_link_ops
);
1083 unregister_netdevice_notifier(&macvtap_notifier_block
);
1085 class_unregister(macvtap_class
);
1087 cdev_del(&macvtap_cdev
);
1089 unregister_chrdev_region(macvtap_major
, MACVTAP_NUM_DEVS
);
1093 module_init(macvtap_init
);
1095 static void macvtap_exit(void)
1097 rtnl_link_unregister(&macvtap_link_ops
);
1098 unregister_netdevice_notifier(&macvtap_notifier_block
);
1099 class_unregister(macvtap_class
);
1100 cdev_del(&macvtap_cdev
);
1101 unregister_chrdev_region(macvtap_major
, MACVTAP_NUM_DEVS
);
1103 module_exit(macvtap_exit
);
1105 MODULE_ALIAS_RTNL_LINK("macvtap");
1106 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1107 MODULE_LICENSE("GPL");