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
*vlan
;
47 static struct proto macvtap_proto
= {
50 .obj_size
= sizeof (struct macvtap_queue
),
54 * Minor number matches netdev->ifindex, so need a potentially
55 * large value. This also makes it possible to split the
56 * tap functionality out again in the future by offering it
57 * from other drivers besides macvtap. As long as every device
58 * only has one tap, the interface numbers assure that the
59 * device nodes are unique.
61 static unsigned int macvtap_major
;
62 #define MACVTAP_NUM_DEVS 65536
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 * Choose the next free queue, for now there is only one
89 static int macvtap_set_queue(struct net_device
*dev
, struct file
*file
,
90 struct macvtap_queue
*q
)
92 struct macvlan_dev
*vlan
= netdev_priv(dev
);
95 spin_lock(&macvtap_lock
);
96 if (rcu_dereference(vlan
->tap
))
100 rcu_assign_pointer(q
->vlan
, vlan
);
101 rcu_assign_pointer(vlan
->tap
, q
);
105 file
->private_data
= q
;
108 spin_unlock(&macvtap_lock
);
113 * The file owning the queue got closed, give up both
114 * the reference that the files holds as well as the
115 * one from the macvlan_dev if that still exists.
117 * Using the spinlock makes sure that we don't get
118 * to the queue again after destroying it.
120 static void macvtap_put_queue(struct macvtap_queue
*q
)
122 struct macvlan_dev
*vlan
;
124 spin_lock(&macvtap_lock
);
125 vlan
= rcu_dereference(q
->vlan
);
127 rcu_assign_pointer(vlan
->tap
, NULL
);
128 rcu_assign_pointer(q
->vlan
, NULL
);
132 spin_unlock(&macvtap_lock
);
139 * Since we only support one queue, just dereference the pointer.
141 static struct macvtap_queue
*macvtap_get_queue(struct net_device
*dev
,
144 struct macvlan_dev
*vlan
= netdev_priv(dev
);
146 return rcu_dereference(vlan
->tap
);
150 * The net_device is going away, give up the reference
151 * that it holds on the queue (all the queues one day)
152 * and safely set the pointer from the queues to NULL.
154 static void macvtap_del_queues(struct net_device
*dev
)
156 struct macvlan_dev
*vlan
= netdev_priv(dev
);
157 struct macvtap_queue
*q
;
159 spin_lock(&macvtap_lock
);
160 q
= rcu_dereference(vlan
->tap
);
162 spin_unlock(&macvtap_lock
);
166 rcu_assign_pointer(vlan
->tap
, NULL
);
167 rcu_assign_pointer(q
->vlan
, NULL
);
168 spin_unlock(&macvtap_lock
);
175 * Forward happens for data that gets sent from one macvlan
176 * endpoint to another one in bridge mode. We just take
177 * the skb and put it into the receive queue.
179 static int macvtap_forward(struct net_device
*dev
, struct sk_buff
*skb
)
181 struct macvtap_queue
*q
= macvtap_get_queue(dev
, skb
);
185 skb_queue_tail(&q
->sk
.sk_receive_queue
, skb
);
186 wake_up_interruptible_poll(sk_sleep(&q
->sk
), POLLIN
| POLLRDNORM
| POLLRDBAND
);
191 * Receive is for data from the external interface (lowerdev),
192 * in case of macvtap, we can treat that the same way as
193 * forward, which macvlan cannot.
195 static int macvtap_receive(struct sk_buff
*skb
)
197 skb_push(skb
, ETH_HLEN
);
198 return macvtap_forward(skb
->dev
, skb
);
201 static int macvtap_newlink(struct net
*src_net
,
202 struct net_device
*dev
,
204 struct nlattr
*data
[])
206 struct device
*classdev
;
210 err
= macvlan_common_newlink(src_net
, dev
, tb
, data
,
211 macvtap_receive
, macvtap_forward
);
215 devt
= MKDEV(MAJOR(macvtap_major
), dev
->ifindex
);
217 classdev
= device_create(macvtap_class
, &dev
->dev
, devt
,
218 dev
, "tap%d", dev
->ifindex
);
219 if (IS_ERR(classdev
)) {
220 err
= PTR_ERR(classdev
);
221 macvtap_del_queues(dev
);
228 static void macvtap_dellink(struct net_device
*dev
,
229 struct list_head
*head
)
231 device_destroy(macvtap_class
,
232 MKDEV(MAJOR(macvtap_major
), dev
->ifindex
));
234 macvtap_del_queues(dev
);
235 macvlan_dellink(dev
, head
);
238 static struct rtnl_link_ops macvtap_link_ops __read_mostly
= {
240 .newlink
= macvtap_newlink
,
241 .dellink
= macvtap_dellink
,
245 static void macvtap_sock_write_space(struct sock
*sk
)
247 wait_queue_head_t
*wqueue
;
249 if (!sock_writeable(sk
) ||
250 !test_and_clear_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
))
253 wqueue
= sk_sleep(sk
);
254 if (wqueue
&& waitqueue_active(wqueue
))
255 wake_up_interruptible_poll(wqueue
, POLLOUT
| POLLWRNORM
| POLLWRBAND
);
258 static int macvtap_open(struct inode
*inode
, struct file
*file
)
260 struct net
*net
= current
->nsproxy
->net_ns
;
261 struct net_device
*dev
= dev_get_by_index(net
, iminor(inode
));
262 struct macvtap_queue
*q
;
269 /* check if this is a macvtap device */
271 if (dev
->rtnl_link_ops
!= &macvtap_link_ops
)
275 q
= (struct macvtap_queue
*)sk_alloc(net
, AF_UNSPEC
, GFP_KERNEL
,
281 init_waitqueue_head(&q
->wq
.wait
);
282 q
->sock
.type
= SOCK_RAW
;
283 q
->sock
.state
= SS_CONNECTED
;
285 q
->sock
.ops
= &macvtap_socket_ops
;
286 sock_init_data(&q
->sock
, &q
->sk
);
287 q
->sk
.sk_write_space
= macvtap_sock_write_space
;
288 q
->flags
= IFF_VNET_HDR
| IFF_NO_PI
| IFF_TAP
;
289 q
->vnet_hdr_sz
= sizeof(struct virtio_net_hdr
);
291 err
= macvtap_set_queue(dev
, file
, q
);
302 static int macvtap_release(struct inode
*inode
, struct file
*file
)
304 struct macvtap_queue
*q
= file
->private_data
;
305 macvtap_put_queue(q
);
309 static unsigned int macvtap_poll(struct file
*file
, poll_table
* wait
)
311 struct macvtap_queue
*q
= file
->private_data
;
312 unsigned int mask
= POLLERR
;
318 poll_wait(file
, &q
->wq
.wait
, wait
);
320 if (!skb_queue_empty(&q
->sk
.sk_receive_queue
))
321 mask
|= POLLIN
| POLLRDNORM
;
323 if (sock_writeable(&q
->sk
) ||
324 (!test_and_set_bit(SOCK_ASYNC_NOSPACE
, &q
->sock
.flags
) &&
325 sock_writeable(&q
->sk
)))
326 mask
|= POLLOUT
| POLLWRNORM
;
332 static inline struct sk_buff
*macvtap_alloc_skb(struct sock
*sk
, size_t prepad
,
333 size_t len
, size_t linear
,
334 int noblock
, int *err
)
338 /* Under a page? Don't bother with paged skb. */
339 if (prepad
+ len
< PAGE_SIZE
|| !linear
)
342 skb
= sock_alloc_send_pskb(sk
, prepad
+ linear
, len
- linear
, noblock
,
347 skb_reserve(skb
, prepad
);
348 skb_put(skb
, linear
);
349 skb
->data_len
= len
- linear
;
350 skb
->len
+= len
- linear
;
356 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
357 * be shared with the tun/tap driver.
359 static int macvtap_skb_from_vnet_hdr(struct sk_buff
*skb
,
360 struct virtio_net_hdr
*vnet_hdr
)
362 unsigned short gso_type
= 0;
363 if (vnet_hdr
->gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
364 switch (vnet_hdr
->gso_type
& ~VIRTIO_NET_HDR_GSO_ECN
) {
365 case VIRTIO_NET_HDR_GSO_TCPV4
:
366 gso_type
= SKB_GSO_TCPV4
;
368 case VIRTIO_NET_HDR_GSO_TCPV6
:
369 gso_type
= SKB_GSO_TCPV6
;
371 case VIRTIO_NET_HDR_GSO_UDP
:
372 gso_type
= SKB_GSO_UDP
;
378 if (vnet_hdr
->gso_type
& VIRTIO_NET_HDR_GSO_ECN
)
379 gso_type
|= SKB_GSO_TCP_ECN
;
381 if (vnet_hdr
->gso_size
== 0)
385 if (vnet_hdr
->flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) {
386 if (!skb_partial_csum_set(skb
, vnet_hdr
->csum_start
,
387 vnet_hdr
->csum_offset
))
391 if (vnet_hdr
->gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
392 skb_shinfo(skb
)->gso_size
= vnet_hdr
->gso_size
;
393 skb_shinfo(skb
)->gso_type
= gso_type
;
395 /* Header must be checked, and gso_segs computed. */
396 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
397 skb_shinfo(skb
)->gso_segs
= 0;
402 static int macvtap_skb_to_vnet_hdr(const struct sk_buff
*skb
,
403 struct virtio_net_hdr
*vnet_hdr
)
405 memset(vnet_hdr
, 0, sizeof(*vnet_hdr
));
407 if (skb_is_gso(skb
)) {
408 struct skb_shared_info
*sinfo
= skb_shinfo(skb
);
410 /* This is a hint as to how much should be linear. */
411 vnet_hdr
->hdr_len
= skb_headlen(skb
);
412 vnet_hdr
->gso_size
= sinfo
->gso_size
;
413 if (sinfo
->gso_type
& SKB_GSO_TCPV4
)
414 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_TCPV4
;
415 else if (sinfo
->gso_type
& SKB_GSO_TCPV6
)
416 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_TCPV6
;
417 else if (sinfo
->gso_type
& SKB_GSO_UDP
)
418 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_UDP
;
421 if (sinfo
->gso_type
& SKB_GSO_TCP_ECN
)
422 vnet_hdr
->gso_type
|= VIRTIO_NET_HDR_GSO_ECN
;
424 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
426 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
427 vnet_hdr
->flags
= VIRTIO_NET_HDR_F_NEEDS_CSUM
;
428 vnet_hdr
->csum_start
= skb
->csum_start
-
430 vnet_hdr
->csum_offset
= skb
->csum_offset
;
431 } /* else everything is zero */
437 /* Get packet from user space buffer */
438 static ssize_t
macvtap_get_user(struct macvtap_queue
*q
,
439 const struct iovec
*iv
, size_t count
,
443 struct macvlan_dev
*vlan
;
446 struct virtio_net_hdr vnet_hdr
= { 0 };
447 int vnet_hdr_len
= 0;
449 if (q
->flags
& IFF_VNET_HDR
) {
450 vnet_hdr_len
= q
->vnet_hdr_sz
;
453 if ((len
-= vnet_hdr_len
) < 0)
456 err
= memcpy_fromiovecend((void *)&vnet_hdr
, iv
, 0,
460 if ((vnet_hdr
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) &&
461 vnet_hdr
.csum_start
+ vnet_hdr
.csum_offset
+ 2 >
463 vnet_hdr
.hdr_len
= vnet_hdr
.csum_start
+
464 vnet_hdr
.csum_offset
+ 2;
466 if (vnet_hdr
.hdr_len
> len
)
471 if (unlikely(len
< ETH_HLEN
))
474 skb
= macvtap_alloc_skb(&q
->sk
, NET_IP_ALIGN
, len
, vnet_hdr
.hdr_len
,
479 err
= skb_copy_datagram_from_iovec(skb
, 0, iv
, vnet_hdr_len
, len
);
483 skb_set_network_header(skb
, ETH_HLEN
);
484 skb_reset_mac_header(skb
);
485 skb
->protocol
= eth_hdr(skb
)->h_proto
;
488 err
= macvtap_skb_from_vnet_hdr(skb
, &vnet_hdr
);
494 vlan
= rcu_dereference(q
->vlan
);
496 macvlan_start_xmit(skb
, vlan
->dev
);
499 rcu_read_unlock_bh();
508 vlan
= rcu_dereference(q
->vlan
);
510 netdev_get_tx_queue(vlan
->dev
, 0)->tx_dropped
++;
511 rcu_read_unlock_bh();
516 static ssize_t
macvtap_aio_write(struct kiocb
*iocb
, const struct iovec
*iv
,
517 unsigned long count
, loff_t pos
)
519 struct file
*file
= iocb
->ki_filp
;
520 ssize_t result
= -ENOLINK
;
521 struct macvtap_queue
*q
= file
->private_data
;
523 result
= macvtap_get_user(q
, iv
, iov_length(iv
, count
),
524 file
->f_flags
& O_NONBLOCK
);
528 /* Put packet to the user space buffer */
529 static ssize_t
macvtap_put_user(struct macvtap_queue
*q
,
530 const struct sk_buff
*skb
,
531 const struct iovec
*iv
, int len
)
533 struct macvlan_dev
*vlan
;
535 int vnet_hdr_len
= 0;
537 if (q
->flags
& IFF_VNET_HDR
) {
538 struct virtio_net_hdr vnet_hdr
;
539 vnet_hdr_len
= q
->vnet_hdr_sz
;
540 if ((len
-= vnet_hdr_len
) < 0)
543 ret
= macvtap_skb_to_vnet_hdr(skb
, &vnet_hdr
);
547 if (memcpy_toiovecend(iv
, (void *)&vnet_hdr
, 0, sizeof(vnet_hdr
)))
551 len
= min_t(int, skb
->len
, len
);
553 ret
= skb_copy_datagram_const_iovec(skb
, 0, iv
, vnet_hdr_len
, len
);
556 vlan
= rcu_dereference(q
->vlan
);
558 macvlan_count_rx(vlan
, len
, ret
== 0, 0);
559 rcu_read_unlock_bh();
561 return ret
? ret
: (len
+ vnet_hdr_len
);
564 static ssize_t
macvtap_do_read(struct macvtap_queue
*q
, struct kiocb
*iocb
,
565 const struct iovec
*iv
, unsigned long len
,
568 DECLARE_WAITQUEUE(wait
, current
);
572 add_wait_queue(sk_sleep(&q
->sk
), &wait
);
574 current
->state
= TASK_INTERRUPTIBLE
;
576 /* Read frames from the queue */
577 skb
= skb_dequeue(&q
->sk
.sk_receive_queue
);
583 if (signal_pending(current
)) {
587 /* Nothing to read, let's sleep */
591 ret
= macvtap_put_user(q
, skb
, iv
, len
);
596 current
->state
= TASK_RUNNING
;
597 remove_wait_queue(sk_sleep(&q
->sk
), &wait
);
601 static ssize_t
macvtap_aio_read(struct kiocb
*iocb
, const struct iovec
*iv
,
602 unsigned long count
, loff_t pos
)
604 struct file
*file
= iocb
->ki_filp
;
605 struct macvtap_queue
*q
= file
->private_data
;
606 ssize_t len
, ret
= 0;
608 len
= iov_length(iv
, count
);
614 ret
= macvtap_do_read(q
, iocb
, iv
, len
, file
->f_flags
& O_NONBLOCK
);
615 ret
= min_t(ssize_t
, ret
, len
); /* XXX copied from tun.c. Why? */
621 * provide compatibility with generic tun/tap interface
623 static long macvtap_ioctl(struct file
*file
, unsigned int cmd
,
626 struct macvtap_queue
*q
= file
->private_data
;
627 struct macvlan_dev
*vlan
;
628 void __user
*argp
= (void __user
*)arg
;
629 struct ifreq __user
*ifr
= argp
;
630 unsigned int __user
*up
= argp
;
632 int __user
*sp
= argp
;
638 /* ignore the name, just look at flags */
639 if (get_user(u
, &ifr
->ifr_flags
))
643 if ((u
& ~IFF_VNET_HDR
) != (IFF_NO_PI
| IFF_TAP
))
652 vlan
= rcu_dereference(q
->vlan
);
655 rcu_read_unlock_bh();
661 if (copy_to_user(&ifr
->ifr_name
, q
->vlan
->dev
->name
, IFNAMSIZ
) ||
662 put_user(q
->flags
, &ifr
->ifr_flags
))
668 if (put_user(IFF_TAP
| IFF_NO_PI
| IFF_VNET_HDR
, up
))
679 case TUNGETVNETHDRSZ
:
685 case TUNSETVNETHDRSZ
:
688 if (s
< (int)sizeof(struct virtio_net_hdr
))
695 /* let the user check for future flags */
696 if (arg
& ~(TUN_F_CSUM
| TUN_F_TSO4
| TUN_F_TSO6
|
697 TUN_F_TSO_ECN
| TUN_F_UFO
))
700 /* TODO: only accept frames with the features that
701 got enabled for forwarded frames */
702 if (!(q
->flags
& IFF_VNET_HDR
))
712 static long macvtap_compat_ioctl(struct file
*file
, unsigned int cmd
,
715 return macvtap_ioctl(file
, cmd
, (unsigned long)compat_ptr(arg
));
719 static const struct file_operations macvtap_fops
= {
720 .owner
= THIS_MODULE
,
721 .open
= macvtap_open
,
722 .release
= macvtap_release
,
723 .aio_read
= macvtap_aio_read
,
724 .aio_write
= macvtap_aio_write
,
725 .poll
= macvtap_poll
,
727 .unlocked_ioctl
= macvtap_ioctl
,
729 .compat_ioctl
= macvtap_compat_ioctl
,
733 static int macvtap_sendmsg(struct kiocb
*iocb
, struct socket
*sock
,
734 struct msghdr
*m
, size_t total_len
)
736 struct macvtap_queue
*q
= container_of(sock
, struct macvtap_queue
, sock
);
737 return macvtap_get_user(q
, m
->msg_iov
, total_len
,
738 m
->msg_flags
& MSG_DONTWAIT
);
741 static int macvtap_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
742 struct msghdr
*m
, size_t total_len
,
745 struct macvtap_queue
*q
= container_of(sock
, struct macvtap_queue
, sock
);
747 if (flags
& ~(MSG_DONTWAIT
|MSG_TRUNC
))
749 ret
= macvtap_do_read(q
, iocb
, m
->msg_iov
, total_len
,
750 flags
& MSG_DONTWAIT
);
751 if (ret
> total_len
) {
752 m
->msg_flags
|= MSG_TRUNC
;
753 ret
= flags
& MSG_TRUNC
? ret
: total_len
;
758 /* Ops structure to mimic raw sockets with tun */
759 static const struct proto_ops macvtap_socket_ops
= {
760 .sendmsg
= macvtap_sendmsg
,
761 .recvmsg
= macvtap_recvmsg
,
764 /* Get an underlying socket object from tun file. Returns error unless file is
765 * attached to a device. The returned object works like a packet socket, it
766 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
767 * holding a reference to the file for as long as the socket is in use. */
768 struct socket
*macvtap_get_socket(struct file
*file
)
770 struct macvtap_queue
*q
;
771 if (file
->f_op
!= &macvtap_fops
)
772 return ERR_PTR(-EINVAL
);
773 q
= file
->private_data
;
775 return ERR_PTR(-EBADFD
);
778 EXPORT_SYMBOL_GPL(macvtap_get_socket
);
780 static int macvtap_init(void)
784 err
= alloc_chrdev_region(&macvtap_major
, 0,
785 MACVTAP_NUM_DEVS
, "macvtap");
789 cdev_init(&macvtap_cdev
, &macvtap_fops
);
790 err
= cdev_add(&macvtap_cdev
, macvtap_major
, MACVTAP_NUM_DEVS
);
794 macvtap_class
= class_create(THIS_MODULE
, "macvtap");
795 if (IS_ERR(macvtap_class
)) {
796 err
= PTR_ERR(macvtap_class
);
800 err
= macvlan_link_register(&macvtap_link_ops
);
807 class_unregister(macvtap_class
);
809 cdev_del(&macvtap_cdev
);
811 unregister_chrdev_region(macvtap_major
, MACVTAP_NUM_DEVS
);
815 module_init(macvtap_init
);
817 static void macvtap_exit(void)
819 rtnl_link_unregister(&macvtap_link_ops
);
820 class_unregister(macvtap_class
);
821 cdev_del(&macvtap_cdev
);
822 unregister_chrdev_region(macvtap_major
, MACVTAP_NUM_DEVS
);
824 module_exit(macvtap_exit
);
826 MODULE_ALIAS_RTNL_LINK("macvtap");
827 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
828 MODULE_LICENSE("GPL");