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
;
50 static struct proto macvtap_proto
= {
53 .obj_size
= sizeof (struct macvtap_queue
),
57 * Variables for dealing with macvtaps device numbers.
59 static dev_t macvtap_major
;
60 #define MACVTAP_NUM_DEVS (1U << MINORBITS)
61 static DEFINE_MUTEX(minor_lock
);
62 static DEFINE_IDR(minor_idr
);
64 #define GOODCOPY_LEN 128
65 static struct class *macvtap_class
;
66 static struct cdev macvtap_cdev
;
68 static const struct proto_ops macvtap_socket_ops
;
72 * The macvtap_queue and the macvlan_dev are loosely coupled, the
73 * pointers from one to the other can only be read while rcu_read_lock
74 * or macvtap_lock is held.
76 * Both the file and the macvlan_dev hold a reference on the macvtap_queue
77 * through sock_hold(&q->sk). When the macvlan_dev goes away first,
78 * q->vlan becomes inaccessible. When the files gets closed,
79 * macvtap_get_queue() fails.
81 * There may still be references to the struct sock inside of the
82 * queue from outbound SKBs, but these never reference back to the
83 * file or the dev. The data structure is freed through __sk_free
84 * when both our references and any pending SKBs are gone.
86 static DEFINE_SPINLOCK(macvtap_lock
);
88 static int macvtap_set_queue(struct net_device
*dev
, struct file
*file
,
89 struct macvtap_queue
*q
)
91 struct macvlan_dev
*vlan
= netdev_priv(dev
);
94 spin_lock(&macvtap_lock
);
95 if (vlan
->numvtaps
== MAX_MACVTAP_QUEUES
)
99 rcu_assign_pointer(q
->vlan
, vlan
);
100 rcu_assign_pointer(vlan
->taps
[vlan
->numvtaps
], q
);
104 q
->queue_index
= vlan
->numvtaps
;
105 file
->private_data
= q
;
110 spin_unlock(&macvtap_lock
);
115 * The file owning the queue got closed, give up both
116 * the reference that the files holds as well as the
117 * one from the macvlan_dev if that still exists.
119 * Using the spinlock makes sure that we don't get
120 * to the queue again after destroying it.
122 static void macvtap_put_queue(struct macvtap_queue
*q
)
124 struct macvtap_queue
*nq
;
125 struct macvlan_dev
*vlan
;
127 spin_lock(&macvtap_lock
);
128 vlan
= rcu_dereference_protected(q
->vlan
,
129 lockdep_is_held(&macvtap_lock
));
131 int index
= q
->queue_index
;
132 BUG_ON(index
>= vlan
->numvtaps
);
134 nq
= rcu_dereference_protected(vlan
->taps
[vlan
->numvtaps
- 1],
135 lockdep_is_held(&macvtap_lock
));
136 rcu_assign_pointer(vlan
->taps
[index
], nq
);
137 nq
->queue_index
= index
;
139 RCU_INIT_POINTER(vlan
->taps
[vlan
->numvtaps
- 1], NULL
);
140 RCU_INIT_POINTER(q
->vlan
, NULL
);
145 spin_unlock(&macvtap_lock
);
152 * Select a queue based on the rxq of the device on which this packet
153 * arrived. If the incoming device is not mq, calculate a flow hash
154 * to select a queue. If all fails, find the first available queue.
155 * Cache vlan->numvtaps since it can become zero during the execution
158 static struct macvtap_queue
*macvtap_get_queue(struct net_device
*dev
,
161 struct macvlan_dev
*vlan
= netdev_priv(dev
);
162 struct macvtap_queue
*tap
= NULL
;
163 int numvtaps
= ACCESS_ONCE(vlan
->numvtaps
);
169 /* Check if we can use flow to select a queue */
170 rxq
= skb_get_rxhash(skb
);
172 tap
= rcu_dereference(vlan
->taps
[rxq
% numvtaps
]);
176 if (likely(skb_rx_queue_recorded(skb
))) {
177 rxq
= skb_get_rx_queue(skb
);
179 while (unlikely(rxq
>= numvtaps
))
182 tap
= rcu_dereference(vlan
->taps
[rxq
]);
186 tap
= rcu_dereference(vlan
->taps
[0]);
192 * The net_device is going away, give up the reference
193 * that it holds on all queues and safely set the pointer
194 * from the queues to NULL.
196 static void macvtap_del_queues(struct net_device
*dev
)
198 struct macvlan_dev
*vlan
= netdev_priv(dev
);
199 struct macvtap_queue
*q
, *qlist
[MAX_MACVTAP_QUEUES
];
202 spin_lock(&macvtap_lock
);
203 for (i
= 0; i
< vlan
->numvtaps
; i
++) {
204 q
= rcu_dereference_protected(vlan
->taps
[i
],
205 lockdep_is_held(&macvtap_lock
));
208 RCU_INIT_POINTER(vlan
->taps
[i
], NULL
);
209 RCU_INIT_POINTER(q
->vlan
, NULL
);
211 /* guarantee that any future macvtap_set_queue will fail */
212 vlan
->numvtaps
= MAX_MACVTAP_QUEUES
;
213 spin_unlock(&macvtap_lock
);
217 for (--j
; j
>= 0; j
--)
218 sock_put(&qlist
[j
]->sk
);
222 * Forward happens for data that gets sent from one macvlan
223 * endpoint to another one in bridge mode. We just take
224 * the skb and put it into the receive queue.
226 static int macvtap_forward(struct net_device
*dev
, struct sk_buff
*skb
)
228 struct macvtap_queue
*q
= macvtap_get_queue(dev
, skb
);
232 if (skb_queue_len(&q
->sk
.sk_receive_queue
) >= dev
->tx_queue_len
)
235 skb_queue_tail(&q
->sk
.sk_receive_queue
, skb
);
236 wake_up_interruptible_poll(sk_sleep(&q
->sk
), POLLIN
| POLLRDNORM
| POLLRDBAND
);
237 return NET_RX_SUCCESS
;
245 * Receive is for data from the external interface (lowerdev),
246 * in case of macvtap, we can treat that the same way as
247 * forward, which macvlan cannot.
249 static int macvtap_receive(struct sk_buff
*skb
)
251 skb_push(skb
, ETH_HLEN
);
252 return macvtap_forward(skb
->dev
, skb
);
255 static int macvtap_get_minor(struct macvlan_dev
*vlan
)
257 int retval
= -ENOMEM
;
259 mutex_lock(&minor_lock
);
260 retval
= idr_alloc(&minor_idr
, vlan
, 1, MACVTAP_NUM_DEVS
, GFP_KERNEL
);
262 vlan
->minor
= retval
;
263 } else if (retval
== -ENOSPC
) {
264 printk(KERN_ERR
"too many macvtap devices\n");
267 mutex_unlock(&minor_lock
);
268 return retval
< 0 ? retval
: 0;
271 static void macvtap_free_minor(struct macvlan_dev
*vlan
)
273 mutex_lock(&minor_lock
);
275 idr_remove(&minor_idr
, vlan
->minor
);
278 mutex_unlock(&minor_lock
);
281 static struct net_device
*dev_get_by_macvtap_minor(int minor
)
283 struct net_device
*dev
= NULL
;
284 struct macvlan_dev
*vlan
;
286 mutex_lock(&minor_lock
);
287 vlan
= idr_find(&minor_idr
, minor
);
292 mutex_unlock(&minor_lock
);
296 static int macvtap_newlink(struct net
*src_net
,
297 struct net_device
*dev
,
299 struct nlattr
*data
[])
301 /* Don't put anything that may fail after macvlan_common_newlink
302 * because we can't undo what it does.
304 return macvlan_common_newlink(src_net
, dev
, tb
, data
,
305 macvtap_receive
, macvtap_forward
);
308 static void macvtap_dellink(struct net_device
*dev
,
309 struct list_head
*head
)
311 macvtap_del_queues(dev
);
312 macvlan_dellink(dev
, head
);
315 static void macvtap_setup(struct net_device
*dev
)
317 macvlan_common_setup(dev
);
318 dev
->tx_queue_len
= TUN_READQ_SIZE
;
321 static struct rtnl_link_ops macvtap_link_ops __read_mostly
= {
323 .setup
= macvtap_setup
,
324 .newlink
= macvtap_newlink
,
325 .dellink
= macvtap_dellink
,
329 static void macvtap_sock_write_space(struct sock
*sk
)
331 wait_queue_head_t
*wqueue
;
333 if (!sock_writeable(sk
) ||
334 !test_and_clear_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
))
337 wqueue
= sk_sleep(sk
);
338 if (wqueue
&& waitqueue_active(wqueue
))
339 wake_up_interruptible_poll(wqueue
, POLLOUT
| POLLWRNORM
| POLLWRBAND
);
342 static void macvtap_sock_destruct(struct sock
*sk
)
344 skb_queue_purge(&sk
->sk_receive_queue
);
347 static int macvtap_open(struct inode
*inode
, struct file
*file
)
349 struct net
*net
= current
->nsproxy
->net_ns
;
350 struct net_device
*dev
= dev_get_by_macvtap_minor(iminor(inode
));
351 struct macvtap_queue
*q
;
359 q
= (struct macvtap_queue
*)sk_alloc(net
, AF_UNSPEC
, GFP_KERNEL
,
365 init_waitqueue_head(&q
->wq
.wait
);
366 q
->sock
.type
= SOCK_RAW
;
367 q
->sock
.state
= SS_CONNECTED
;
369 q
->sock
.ops
= &macvtap_socket_ops
;
370 sock_init_data(&q
->sock
, &q
->sk
);
371 q
->sk
.sk_write_space
= macvtap_sock_write_space
;
372 q
->sk
.sk_destruct
= macvtap_sock_destruct
;
373 q
->flags
= IFF_VNET_HDR
| IFF_NO_PI
| IFF_TAP
;
374 q
->vnet_hdr_sz
= sizeof(struct virtio_net_hdr
);
377 * so far only KVM virtio_net uses macvtap, enable zero copy between
378 * guest kernel and host kernel when lower device supports zerocopy
380 * The macvlan supports zerocopy iff the lower device supports zero
381 * copy so we don't have to look at the lower device directly.
383 if ((dev
->features
& NETIF_F_HIGHDMA
) && (dev
->features
& NETIF_F_SG
))
384 sock_set_flag(&q
->sk
, SOCK_ZEROCOPY
);
386 err
= macvtap_set_queue(dev
, file
, q
);
397 static int macvtap_release(struct inode
*inode
, struct file
*file
)
399 struct macvtap_queue
*q
= file
->private_data
;
400 macvtap_put_queue(q
);
404 static unsigned int macvtap_poll(struct file
*file
, poll_table
* wait
)
406 struct macvtap_queue
*q
= file
->private_data
;
407 unsigned int mask
= POLLERR
;
413 poll_wait(file
, &q
->wq
.wait
, wait
);
415 if (!skb_queue_empty(&q
->sk
.sk_receive_queue
))
416 mask
|= POLLIN
| POLLRDNORM
;
418 if (sock_writeable(&q
->sk
) ||
419 (!test_and_set_bit(SOCK_ASYNC_NOSPACE
, &q
->sock
.flags
) &&
420 sock_writeable(&q
->sk
)))
421 mask
|= POLLOUT
| POLLWRNORM
;
427 static inline struct sk_buff
*macvtap_alloc_skb(struct sock
*sk
, size_t prepad
,
428 size_t len
, size_t linear
,
429 int noblock
, int *err
)
433 /* Under a page? Don't bother with paged skb. */
434 if (prepad
+ len
< PAGE_SIZE
|| !linear
)
437 skb
= sock_alloc_send_pskb(sk
, prepad
+ linear
, len
- linear
, noblock
,
442 skb_reserve(skb
, prepad
);
443 skb_put(skb
, linear
);
444 skb
->data_len
= len
- linear
;
445 skb
->len
+= len
- linear
;
450 /* set skb frags from iovec, this can move to core network code for reuse */
451 static int zerocopy_sg_from_iovec(struct sk_buff
*skb
, const struct iovec
*from
,
452 int offset
, size_t count
)
454 int len
= iov_length(from
, count
) - offset
;
455 int copy
= skb_headlen(skb
);
456 int size
, offset1
= 0;
459 /* Skip over from offset */
460 while (count
&& (offset
>= from
->iov_len
)) {
461 offset
-= from
->iov_len
;
466 /* copy up to skb headlen */
467 while (count
&& (copy
> 0)) {
468 size
= min_t(unsigned int, copy
, from
->iov_len
- offset
);
469 if (copy_from_user(skb
->data
+ offset1
, from
->iov_base
+ offset
,
486 struct page
*page
[MAX_SKB_FRAGS
];
489 unsigned long truesize
;
491 len
= from
->iov_len
- offset
;
497 base
= (unsigned long)from
->iov_base
+ offset
;
498 size
= ((base
& ~PAGE_MASK
) + len
+ ~PAGE_MASK
) >> PAGE_SHIFT
;
499 if (i
+ size
> MAX_SKB_FRAGS
)
501 num_pages
= get_user_pages_fast(base
, size
, 0, &page
[i
]);
502 if (num_pages
!= size
) {
503 for (i
= 0; i
< num_pages
; i
++)
507 truesize
= size
* PAGE_SIZE
;
508 skb
->data_len
+= len
;
510 skb
->truesize
+= truesize
;
511 atomic_add(truesize
, &skb
->sk
->sk_wmem_alloc
);
513 int off
= base
& ~PAGE_MASK
;
514 int size
= min_t(int, len
, PAGE_SIZE
- off
);
515 __skb_fill_page_desc(skb
, i
, page
[i
], off
, size
);
516 skb_shinfo(skb
)->nr_frags
++;
517 /* increase sk_wmem_alloc */
529 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
530 * be shared with the tun/tap driver.
532 static int macvtap_skb_from_vnet_hdr(struct sk_buff
*skb
,
533 struct virtio_net_hdr
*vnet_hdr
)
535 unsigned short gso_type
= 0;
536 if (vnet_hdr
->gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
537 switch (vnet_hdr
->gso_type
& ~VIRTIO_NET_HDR_GSO_ECN
) {
538 case VIRTIO_NET_HDR_GSO_TCPV4
:
539 gso_type
= SKB_GSO_TCPV4
;
541 case VIRTIO_NET_HDR_GSO_TCPV6
:
542 gso_type
= SKB_GSO_TCPV6
;
544 case VIRTIO_NET_HDR_GSO_UDP
:
545 gso_type
= SKB_GSO_UDP
;
551 if (vnet_hdr
->gso_type
& VIRTIO_NET_HDR_GSO_ECN
)
552 gso_type
|= SKB_GSO_TCP_ECN
;
554 if (vnet_hdr
->gso_size
== 0)
558 if (vnet_hdr
->flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) {
559 if (!skb_partial_csum_set(skb
, vnet_hdr
->csum_start
,
560 vnet_hdr
->csum_offset
))
564 if (vnet_hdr
->gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
565 skb_shinfo(skb
)->gso_size
= vnet_hdr
->gso_size
;
566 skb_shinfo(skb
)->gso_type
= gso_type
;
568 /* Header must be checked, and gso_segs computed. */
569 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
570 skb_shinfo(skb
)->gso_segs
= 0;
575 static int macvtap_skb_to_vnet_hdr(const struct sk_buff
*skb
,
576 struct virtio_net_hdr
*vnet_hdr
)
578 memset(vnet_hdr
, 0, sizeof(*vnet_hdr
));
580 if (skb_is_gso(skb
)) {
581 struct skb_shared_info
*sinfo
= skb_shinfo(skb
);
583 /* This is a hint as to how much should be linear. */
584 vnet_hdr
->hdr_len
= skb_headlen(skb
);
585 vnet_hdr
->gso_size
= sinfo
->gso_size
;
586 if (sinfo
->gso_type
& SKB_GSO_TCPV4
)
587 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_TCPV4
;
588 else if (sinfo
->gso_type
& SKB_GSO_TCPV6
)
589 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_TCPV6
;
590 else if (sinfo
->gso_type
& SKB_GSO_UDP
)
591 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_UDP
;
594 if (sinfo
->gso_type
& SKB_GSO_TCP_ECN
)
595 vnet_hdr
->gso_type
|= VIRTIO_NET_HDR_GSO_ECN
;
597 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
599 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
600 vnet_hdr
->flags
= VIRTIO_NET_HDR_F_NEEDS_CSUM
;
601 vnet_hdr
->csum_start
= skb_checksum_start_offset(skb
);
602 vnet_hdr
->csum_offset
= skb
->csum_offset
;
603 } else if (skb
->ip_summed
== CHECKSUM_UNNECESSARY
) {
604 vnet_hdr
->flags
= VIRTIO_NET_HDR_F_DATA_VALID
;
605 } /* else everything is zero */
611 /* Get packet from user space buffer */
612 static ssize_t
macvtap_get_user(struct macvtap_queue
*q
, struct msghdr
*m
,
613 const struct iovec
*iv
, unsigned long total_len
,
614 size_t count
, int noblock
)
617 struct macvlan_dev
*vlan
;
618 unsigned long len
= total_len
;
620 struct virtio_net_hdr vnet_hdr
= { 0 };
621 int vnet_hdr_len
= 0;
623 bool zerocopy
= false;
625 if (q
->flags
& IFF_VNET_HDR
) {
626 vnet_hdr_len
= q
->vnet_hdr_sz
;
629 if (len
< vnet_hdr_len
)
633 err
= memcpy_fromiovecend((void *)&vnet_hdr
, iv
, 0,
637 if ((vnet_hdr
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) &&
638 vnet_hdr
.csum_start
+ vnet_hdr
.csum_offset
+ 2 >
640 vnet_hdr
.hdr_len
= vnet_hdr
.csum_start
+
641 vnet_hdr
.csum_offset
+ 2;
643 if (vnet_hdr
.hdr_len
> len
)
648 if (unlikely(len
< ETH_HLEN
))
652 if (unlikely(count
> UIO_MAXIOV
))
655 if (m
&& m
->msg_control
&& sock_flag(&q
->sk
, SOCK_ZEROCOPY
))
659 /* Userspace may produce vectors with count greater than
660 * MAX_SKB_FRAGS, so we need to linearize parts of the skb
661 * to let the rest of data to be fit in the frags.
663 if (count
> MAX_SKB_FRAGS
) {
664 copylen
= iov_length(iv
, count
- MAX_SKB_FRAGS
);
665 if (copylen
< vnet_hdr_len
)
668 copylen
-= vnet_hdr_len
;
670 /* There are 256 bytes to be copied in skb, so there is enough
671 * room for skb expand head in case it is used.
672 * The rest buffer is mapped from userspace.
674 if (copylen
< vnet_hdr
.hdr_len
)
675 copylen
= vnet_hdr
.hdr_len
;
677 copylen
= GOODCOPY_LEN
;
681 skb
= macvtap_alloc_skb(&q
->sk
, NET_IP_ALIGN
, copylen
,
682 vnet_hdr
.hdr_len
, noblock
, &err
);
687 err
= zerocopy_sg_from_iovec(skb
, iv
, vnet_hdr_len
, count
);
689 err
= skb_copy_datagram_from_iovec(skb
, 0, iv
, vnet_hdr_len
,
694 skb_set_network_header(skb
, ETH_HLEN
);
695 skb_reset_mac_header(skb
);
696 skb
->protocol
= eth_hdr(skb
)->h_proto
;
699 err
= macvtap_skb_from_vnet_hdr(skb
, &vnet_hdr
);
704 skb_probe_transport_header(skb
, ETH_HLEN
);
707 vlan
= rcu_dereference_bh(q
->vlan
);
708 /* copy skb_ubuf_info for callback when skb has no error */
710 skb_shinfo(skb
)->destructor_arg
= m
->msg_control
;
711 skb_shinfo(skb
)->tx_flags
|= SKBTX_DEV_ZEROCOPY
;
712 skb_shinfo(skb
)->tx_flags
|= SKBTX_SHARED_FRAG
;
715 macvlan_start_xmit(skb
, vlan
->dev
);
718 rcu_read_unlock_bh();
727 vlan
= rcu_dereference_bh(q
->vlan
);
729 vlan
->dev
->stats
.tx_dropped
++;
730 rcu_read_unlock_bh();
735 static ssize_t
macvtap_aio_write(struct kiocb
*iocb
, const struct iovec
*iv
,
736 unsigned long count
, loff_t pos
)
738 struct file
*file
= iocb
->ki_filp
;
739 ssize_t result
= -ENOLINK
;
740 struct macvtap_queue
*q
= file
->private_data
;
742 result
= macvtap_get_user(q
, NULL
, iv
, iov_length(iv
, count
), count
,
743 file
->f_flags
& O_NONBLOCK
);
747 /* Put packet to the user space buffer */
748 static ssize_t
macvtap_put_user(struct macvtap_queue
*q
,
749 const struct sk_buff
*skb
,
750 const struct iovec
*iv
, int len
)
752 struct macvlan_dev
*vlan
;
754 int vnet_hdr_len
= 0;
758 if (q
->flags
& IFF_VNET_HDR
) {
759 struct virtio_net_hdr vnet_hdr
;
760 vnet_hdr_len
= q
->vnet_hdr_sz
;
761 if ((len
-= vnet_hdr_len
) < 0)
764 ret
= macvtap_skb_to_vnet_hdr(skb
, &vnet_hdr
);
768 if (memcpy_toiovecend(iv
, (void *)&vnet_hdr
, 0, sizeof(vnet_hdr
)))
771 copied
= vnet_hdr_len
;
773 if (!vlan_tx_tag_present(skb
))
774 len
= min_t(int, skb
->len
, len
);
781 veth
.h_vlan_proto
= htons(ETH_P_8021Q
);
782 veth
.h_vlan_TCI
= htons(vlan_tx_tag_get(skb
));
784 vlan_offset
= offsetof(struct vlan_ethhdr
, h_vlan_proto
);
785 len
= min_t(int, skb
->len
+ VLAN_HLEN
, len
);
787 copy
= min_t(int, vlan_offset
, len
);
788 ret
= skb_copy_datagram_const_iovec(skb
, 0, iv
, copied
, copy
);
794 copy
= min_t(int, sizeof(veth
), len
);
795 ret
= memcpy_toiovecend(iv
, (void *)&veth
, copied
, copy
);
802 ret
= skb_copy_datagram_const_iovec(skb
, vlan_offset
, iv
, copied
, len
);
807 vlan
= rcu_dereference_bh(q
->vlan
);
809 macvlan_count_rx(vlan
, copied
- vnet_hdr_len
, ret
== 0, 0);
810 rcu_read_unlock_bh();
812 return ret
? ret
: copied
;
815 static ssize_t
macvtap_do_read(struct macvtap_queue
*q
, struct kiocb
*iocb
,
816 const struct iovec
*iv
, unsigned long len
,
825 prepare_to_wait(sk_sleep(&q
->sk
), &wait
,
828 /* Read frames from the queue */
829 skb
= skb_dequeue(&q
->sk
.sk_receive_queue
);
835 if (signal_pending(current
)) {
839 /* Nothing to read, let's sleep */
843 ret
= macvtap_put_user(q
, skb
, iv
, len
);
849 finish_wait(sk_sleep(&q
->sk
), &wait
);
853 static ssize_t
macvtap_aio_read(struct kiocb
*iocb
, const struct iovec
*iv
,
854 unsigned long count
, loff_t pos
)
856 struct file
*file
= iocb
->ki_filp
;
857 struct macvtap_queue
*q
= file
->private_data
;
858 ssize_t len
, ret
= 0;
860 len
= iov_length(iv
, count
);
866 ret
= macvtap_do_read(q
, iocb
, iv
, len
, file
->f_flags
& O_NONBLOCK
);
867 ret
= min_t(ssize_t
, ret
, len
); /* XXX copied from tun.c. Why? */
872 static struct macvlan_dev
*macvtap_get_vlan(struct macvtap_queue
*q
)
874 struct macvlan_dev
*vlan
;
877 vlan
= rcu_dereference_bh(q
->vlan
);
880 rcu_read_unlock_bh();
885 static void macvtap_put_vlan(struct macvlan_dev
*vlan
)
891 * provide compatibility with generic tun/tap interface
893 static long macvtap_ioctl(struct file
*file
, unsigned int cmd
,
896 struct macvtap_queue
*q
= file
->private_data
;
897 struct macvlan_dev
*vlan
;
898 void __user
*argp
= (void __user
*)arg
;
899 struct ifreq __user
*ifr
= argp
;
900 unsigned int __user
*up
= argp
;
902 int __user
*sp
= argp
;
908 /* ignore the name, just look at flags */
909 if (get_user(u
, &ifr
->ifr_flags
))
913 if ((u
& ~IFF_VNET_HDR
) != (IFF_NO_PI
| IFF_TAP
))
921 vlan
= macvtap_get_vlan(q
);
926 if (copy_to_user(&ifr
->ifr_name
, vlan
->dev
->name
, IFNAMSIZ
) ||
927 put_user(q
->flags
, &ifr
->ifr_flags
))
929 macvtap_put_vlan(vlan
);
933 if (put_user(IFF_TAP
| IFF_NO_PI
| IFF_VNET_HDR
, up
))
944 case TUNGETVNETHDRSZ
:
950 case TUNSETVNETHDRSZ
:
953 if (s
< (int)sizeof(struct virtio_net_hdr
))
960 /* let the user check for future flags */
961 if (arg
& ~(TUN_F_CSUM
| TUN_F_TSO4
| TUN_F_TSO6
|
962 TUN_F_TSO_ECN
| TUN_F_UFO
))
965 /* TODO: only accept frames with the features that
966 got enabled for forwarded frames */
967 if (!(q
->flags
& IFF_VNET_HDR
))
977 static long macvtap_compat_ioctl(struct file
*file
, unsigned int cmd
,
980 return macvtap_ioctl(file
, cmd
, (unsigned long)compat_ptr(arg
));
984 static const struct file_operations macvtap_fops
= {
985 .owner
= THIS_MODULE
,
986 .open
= macvtap_open
,
987 .release
= macvtap_release
,
988 .aio_read
= macvtap_aio_read
,
989 .aio_write
= macvtap_aio_write
,
990 .poll
= macvtap_poll
,
992 .unlocked_ioctl
= macvtap_ioctl
,
994 .compat_ioctl
= macvtap_compat_ioctl
,
998 static int macvtap_sendmsg(struct kiocb
*iocb
, struct socket
*sock
,
999 struct msghdr
*m
, size_t total_len
)
1001 struct macvtap_queue
*q
= container_of(sock
, struct macvtap_queue
, sock
);
1002 return macvtap_get_user(q
, m
, m
->msg_iov
, total_len
, m
->msg_iovlen
,
1003 m
->msg_flags
& MSG_DONTWAIT
);
1006 static int macvtap_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
1007 struct msghdr
*m
, size_t total_len
,
1010 struct macvtap_queue
*q
= container_of(sock
, struct macvtap_queue
, sock
);
1012 if (flags
& ~(MSG_DONTWAIT
|MSG_TRUNC
))
1014 ret
= macvtap_do_read(q
, iocb
, m
->msg_iov
, total_len
,
1015 flags
& MSG_DONTWAIT
);
1016 if (ret
> total_len
) {
1017 m
->msg_flags
|= MSG_TRUNC
;
1018 ret
= flags
& MSG_TRUNC
? ret
: total_len
;
1023 /* Ops structure to mimic raw sockets with tun */
1024 static const struct proto_ops macvtap_socket_ops
= {
1025 .sendmsg
= macvtap_sendmsg
,
1026 .recvmsg
= macvtap_recvmsg
,
1029 /* Get an underlying socket object from tun file. Returns error unless file is
1030 * attached to a device. The returned object works like a packet socket, it
1031 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
1032 * holding a reference to the file for as long as the socket is in use. */
1033 struct socket
*macvtap_get_socket(struct file
*file
)
1035 struct macvtap_queue
*q
;
1036 if (file
->f_op
!= &macvtap_fops
)
1037 return ERR_PTR(-EINVAL
);
1038 q
= file
->private_data
;
1040 return ERR_PTR(-EBADFD
);
1043 EXPORT_SYMBOL_GPL(macvtap_get_socket
);
1045 static int macvtap_device_event(struct notifier_block
*unused
,
1046 unsigned long event
, void *ptr
)
1048 struct net_device
*dev
= netdev_notifier_info_to_dev(ptr
);
1049 struct macvlan_dev
*vlan
;
1050 struct device
*classdev
;
1054 if (dev
->rtnl_link_ops
!= &macvtap_link_ops
)
1057 vlan
= netdev_priv(dev
);
1060 case NETDEV_REGISTER
:
1061 /* Create the device node here after the network device has
1062 * been registered but before register_netdevice has
1065 err
= macvtap_get_minor(vlan
);
1067 return notifier_from_errno(err
);
1069 devt
= MKDEV(MAJOR(macvtap_major
), vlan
->minor
);
1070 classdev
= device_create(macvtap_class
, &dev
->dev
, devt
,
1071 dev
, "tap%d", dev
->ifindex
);
1072 if (IS_ERR(classdev
)) {
1073 macvtap_free_minor(vlan
);
1074 return notifier_from_errno(PTR_ERR(classdev
));
1077 case NETDEV_UNREGISTER
:
1078 devt
= MKDEV(MAJOR(macvtap_major
), vlan
->minor
);
1079 device_destroy(macvtap_class
, devt
);
1080 macvtap_free_minor(vlan
);
1087 static struct notifier_block macvtap_notifier_block __read_mostly
= {
1088 .notifier_call
= macvtap_device_event
,
1091 static int macvtap_init(void)
1095 err
= alloc_chrdev_region(&macvtap_major
, 0,
1096 MACVTAP_NUM_DEVS
, "macvtap");
1100 cdev_init(&macvtap_cdev
, &macvtap_fops
);
1101 err
= cdev_add(&macvtap_cdev
, macvtap_major
, MACVTAP_NUM_DEVS
);
1105 macvtap_class
= class_create(THIS_MODULE
, "macvtap");
1106 if (IS_ERR(macvtap_class
)) {
1107 err
= PTR_ERR(macvtap_class
);
1111 err
= register_netdevice_notifier(&macvtap_notifier_block
);
1115 err
= macvlan_link_register(&macvtap_link_ops
);
1122 unregister_netdevice_notifier(&macvtap_notifier_block
);
1124 class_unregister(macvtap_class
);
1126 cdev_del(&macvtap_cdev
);
1128 unregister_chrdev_region(macvtap_major
, MACVTAP_NUM_DEVS
);
1132 module_init(macvtap_init
);
1134 static void macvtap_exit(void)
1136 rtnl_link_unregister(&macvtap_link_ops
);
1137 unregister_netdevice_notifier(&macvtap_notifier_block
);
1138 class_unregister(macvtap_class
);
1139 cdev_del(&macvtap_cdev
);
1140 unregister_chrdev_region(macvtap_major
, MACVTAP_NUM_DEVS
);
1142 module_exit(macvtap_exit
);
1144 MODULE_ALIAS_RTNL_LINK("macvtap");
1145 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1146 MODULE_LICENSE("GPL");