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
35 struct macvtap_queue
{
40 struct macvlan_dev __rcu
*vlan
;
45 struct list_head next
;
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
;
68 #define TUN_OFFLOADS (NETIF_F_HW_CSUM | NETIF_F_TSO_ECN | NETIF_F_TSO | \
69 NETIF_F_TSO6 | NETIF_F_UFO)
70 #define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO)
73 * The macvtap_queue and the macvlan_dev are loosely coupled, the
74 * pointers from one to the other can only be read while rcu_read_lock
77 * Both the file and the macvlan_dev hold a reference on the macvtap_queue
78 * through sock_hold(&q->sk). When the macvlan_dev goes away first,
79 * q->vlan becomes inaccessible. When the files gets closed,
80 * macvtap_get_queue() fails.
82 * There may still be references to the struct sock inside of the
83 * queue from outbound SKBs, but these never reference back to the
84 * file or the dev. The data structure is freed through __sk_free
85 * when both our references and any pending SKBs are gone.
88 static int macvtap_enable_queue(struct net_device
*dev
, struct file
*file
,
89 struct macvtap_queue
*q
)
91 struct macvlan_dev
*vlan
= netdev_priv(dev
);
100 rcu_assign_pointer(vlan
->taps
[vlan
->numvtaps
], q
);
101 q
->queue_index
= vlan
->numvtaps
;
109 static int macvtap_set_queue(struct net_device
*dev
, struct file
*file
,
110 struct macvtap_queue
*q
)
112 struct macvlan_dev
*vlan
= netdev_priv(dev
);
116 if (vlan
->numqueues
== MAX_MACVTAP_QUEUES
)
120 rcu_assign_pointer(q
->vlan
, vlan
);
121 rcu_assign_pointer(vlan
->taps
[vlan
->numvtaps
], q
);
125 q
->queue_index
= vlan
->numvtaps
;
127 file
->private_data
= q
;
128 list_add_tail(&q
->next
, &vlan
->queue_list
);
138 static int macvtap_disable_queue(struct macvtap_queue
*q
)
140 struct macvlan_dev
*vlan
;
141 struct macvtap_queue
*nq
;
147 vlan
= rtnl_dereference(q
->vlan
);
150 int index
= q
->queue_index
;
151 BUG_ON(index
>= vlan
->numvtaps
);
152 nq
= rtnl_dereference(vlan
->taps
[vlan
->numvtaps
- 1]);
153 nq
->queue_index
= index
;
155 rcu_assign_pointer(vlan
->taps
[index
], nq
);
156 RCU_INIT_POINTER(vlan
->taps
[vlan
->numvtaps
- 1], NULL
);
166 * The file owning the queue got closed, give up both
167 * the reference that the files holds as well as the
168 * one from the macvlan_dev if that still exists.
170 * Using the spinlock makes sure that we don't get
171 * to the queue again after destroying it.
173 static void macvtap_put_queue(struct macvtap_queue
*q
)
175 struct macvlan_dev
*vlan
;
178 vlan
= rtnl_dereference(q
->vlan
);
182 BUG_ON(macvtap_disable_queue(q
));
185 RCU_INIT_POINTER(q
->vlan
, NULL
);
187 list_del_init(&q
->next
);
197 * Select a queue based on the rxq of the device on which this packet
198 * arrived. If the incoming device is not mq, calculate a flow hash
199 * to select a queue. If all fails, find the first available queue.
200 * Cache vlan->numvtaps since it can become zero during the execution
203 static struct macvtap_queue
*macvtap_get_queue(struct net_device
*dev
,
206 struct macvlan_dev
*vlan
= netdev_priv(dev
);
207 struct macvtap_queue
*tap
= NULL
;
208 /* Access to taps array is protected by rcu, but access to numvtaps
209 * isn't. Below we use it to lookup a queue, but treat it as a hint
210 * and validate that the result isn't NULL - in case we are
211 * racing against queue removal.
213 int numvtaps
= ACCESS_ONCE(vlan
->numvtaps
);
219 /* Check if we can use flow to select a queue */
220 rxq
= skb_get_rxhash(skb
);
222 tap
= rcu_dereference(vlan
->taps
[rxq
% numvtaps
]);
226 if (likely(skb_rx_queue_recorded(skb
))) {
227 rxq
= skb_get_rx_queue(skb
);
229 while (unlikely(rxq
>= numvtaps
))
232 tap
= rcu_dereference(vlan
->taps
[rxq
]);
236 tap
= rcu_dereference(vlan
->taps
[0]);
242 * The net_device is going away, give up the reference
243 * that it holds on all queues and safely set the pointer
244 * from the queues to NULL.
246 static void macvtap_del_queues(struct net_device
*dev
)
248 struct macvlan_dev
*vlan
= netdev_priv(dev
);
249 struct macvtap_queue
*q
, *tmp
, *qlist
[MAX_MACVTAP_QUEUES
];
253 list_for_each_entry_safe(q
, tmp
, &vlan
->queue_list
, next
) {
254 list_del_init(&q
->next
);
256 RCU_INIT_POINTER(q
->vlan
, NULL
);
261 for (i
= 0; i
< vlan
->numvtaps
; i
++)
262 RCU_INIT_POINTER(vlan
->taps
[i
], NULL
);
263 BUG_ON(vlan
->numvtaps
);
264 BUG_ON(vlan
->numqueues
);
265 /* guarantee that any future macvtap_set_queue will fail */
266 vlan
->numvtaps
= MAX_MACVTAP_QUEUES
;
268 for (--j
; j
>= 0; j
--)
269 sock_put(&qlist
[j
]->sk
);
273 * Forward happens for data that gets sent from one macvlan
274 * endpoint to another one in bridge mode. We just take
275 * the skb and put it into the receive queue.
277 static int macvtap_forward(struct net_device
*dev
, struct sk_buff
*skb
)
279 struct macvlan_dev
*vlan
= netdev_priv(dev
);
280 struct macvtap_queue
*q
= macvtap_get_queue(dev
, skb
);
281 netdev_features_t features
;
285 if (skb_queue_len(&q
->sk
.sk_receive_queue
) >= dev
->tx_queue_len
)
289 /* Apply the forward feature mask so that we perform segmentation
290 * according to users wishes.
292 features
= netif_skb_features(skb
) & vlan
->tap_features
;
293 if (netif_needs_gso(skb
, features
)) {
294 struct sk_buff
*segs
= __skb_gso_segment(skb
, features
, false);
300 skb_queue_tail(&q
->sk
.sk_receive_queue
, skb
);
306 struct sk_buff
*nskb
= segs
->next
;
309 skb_queue_tail(&q
->sk
.sk_receive_queue
, segs
);
313 skb_queue_tail(&q
->sk
.sk_receive_queue
, skb
);
317 wake_up_interruptible_poll(sk_sleep(&q
->sk
), POLLIN
| POLLRDNORM
| POLLRDBAND
);
318 return NET_RX_SUCCESS
;
326 * Receive is for data from the external interface (lowerdev),
327 * in case of macvtap, we can treat that the same way as
328 * forward, which macvlan cannot.
330 static int macvtap_receive(struct sk_buff
*skb
)
332 skb_push(skb
, ETH_HLEN
);
333 return macvtap_forward(skb
->dev
, skb
);
336 static int macvtap_get_minor(struct macvlan_dev
*vlan
)
338 int retval
= -ENOMEM
;
340 mutex_lock(&minor_lock
);
341 retval
= idr_alloc(&minor_idr
, vlan
, 1, MACVTAP_NUM_DEVS
, GFP_KERNEL
);
343 vlan
->minor
= retval
;
344 } else if (retval
== -ENOSPC
) {
345 printk(KERN_ERR
"too many macvtap devices\n");
348 mutex_unlock(&minor_lock
);
349 return retval
< 0 ? retval
: 0;
352 static void macvtap_free_minor(struct macvlan_dev
*vlan
)
354 mutex_lock(&minor_lock
);
356 idr_remove(&minor_idr
, vlan
->minor
);
359 mutex_unlock(&minor_lock
);
362 static struct net_device
*dev_get_by_macvtap_minor(int minor
)
364 struct net_device
*dev
= NULL
;
365 struct macvlan_dev
*vlan
;
367 mutex_lock(&minor_lock
);
368 vlan
= idr_find(&minor_idr
, minor
);
373 mutex_unlock(&minor_lock
);
377 static int macvtap_newlink(struct net
*src_net
,
378 struct net_device
*dev
,
380 struct nlattr
*data
[])
382 struct macvlan_dev
*vlan
= netdev_priv(dev
);
383 INIT_LIST_HEAD(&vlan
->queue_list
);
385 /* Since macvlan supports all offloads by default, make
386 * tap support all offloads also.
388 vlan
->tap_features
= TUN_OFFLOADS
;
390 /* Don't put anything that may fail after macvlan_common_newlink
391 * because we can't undo what it does.
393 return macvlan_common_newlink(src_net
, dev
, tb
, data
,
394 macvtap_receive
, macvtap_forward
);
397 static void macvtap_dellink(struct net_device
*dev
,
398 struct list_head
*head
)
400 macvtap_del_queues(dev
);
401 macvlan_dellink(dev
, head
);
404 static void macvtap_setup(struct net_device
*dev
)
406 macvlan_common_setup(dev
);
407 dev
->tx_queue_len
= TUN_READQ_SIZE
;
410 static struct rtnl_link_ops macvtap_link_ops __read_mostly
= {
412 .setup
= macvtap_setup
,
413 .newlink
= macvtap_newlink
,
414 .dellink
= macvtap_dellink
,
418 static void macvtap_sock_write_space(struct sock
*sk
)
420 wait_queue_head_t
*wqueue
;
422 if (!sock_writeable(sk
) ||
423 !test_and_clear_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
))
426 wqueue
= sk_sleep(sk
);
427 if (wqueue
&& waitqueue_active(wqueue
))
428 wake_up_interruptible_poll(wqueue
, POLLOUT
| POLLWRNORM
| POLLWRBAND
);
431 static void macvtap_sock_destruct(struct sock
*sk
)
433 skb_queue_purge(&sk
->sk_receive_queue
);
436 static int macvtap_open(struct inode
*inode
, struct file
*file
)
438 struct net
*net
= current
->nsproxy
->net_ns
;
439 struct net_device
*dev
= dev_get_by_macvtap_minor(iminor(inode
));
440 struct macvtap_queue
*q
;
448 q
= (struct macvtap_queue
*)sk_alloc(net
, AF_UNSPEC
, GFP_KERNEL
,
453 RCU_INIT_POINTER(q
->sock
.wq
, &q
->wq
);
454 init_waitqueue_head(&q
->wq
.wait
);
455 q
->sock
.type
= SOCK_RAW
;
456 q
->sock
.state
= SS_CONNECTED
;
458 q
->sock
.ops
= &macvtap_socket_ops
;
459 sock_init_data(&q
->sock
, &q
->sk
);
460 q
->sk
.sk_write_space
= macvtap_sock_write_space
;
461 q
->sk
.sk_destruct
= macvtap_sock_destruct
;
462 q
->flags
= IFF_VNET_HDR
| IFF_NO_PI
| IFF_TAP
;
463 q
->vnet_hdr_sz
= sizeof(struct virtio_net_hdr
);
466 * so far only KVM virtio_net uses macvtap, enable zero copy between
467 * guest kernel and host kernel when lower device supports zerocopy
469 * The macvlan supports zerocopy iff the lower device supports zero
470 * copy so we don't have to look at the lower device directly.
472 if ((dev
->features
& NETIF_F_HIGHDMA
) && (dev
->features
& NETIF_F_SG
))
473 sock_set_flag(&q
->sk
, SOCK_ZEROCOPY
);
475 err
= macvtap_set_queue(dev
, file
, q
);
486 static int macvtap_release(struct inode
*inode
, struct file
*file
)
488 struct macvtap_queue
*q
= file
->private_data
;
489 macvtap_put_queue(q
);
493 static unsigned int macvtap_poll(struct file
*file
, poll_table
* wait
)
495 struct macvtap_queue
*q
= file
->private_data
;
496 unsigned int mask
= POLLERR
;
502 poll_wait(file
, &q
->wq
.wait
, wait
);
504 if (!skb_queue_empty(&q
->sk
.sk_receive_queue
))
505 mask
|= POLLIN
| POLLRDNORM
;
507 if (sock_writeable(&q
->sk
) ||
508 (!test_and_set_bit(SOCK_ASYNC_NOSPACE
, &q
->sock
.flags
) &&
509 sock_writeable(&q
->sk
)))
510 mask
|= POLLOUT
| POLLWRNORM
;
516 static inline struct sk_buff
*macvtap_alloc_skb(struct sock
*sk
, size_t prepad
,
517 size_t len
, size_t linear
,
518 int noblock
, int *err
)
522 /* Under a page? Don't bother with paged skb. */
523 if (prepad
+ len
< PAGE_SIZE
|| !linear
)
526 skb
= sock_alloc_send_pskb(sk
, prepad
+ linear
, len
- linear
, noblock
,
531 skb_reserve(skb
, prepad
);
532 skb_put(skb
, linear
);
533 skb
->data_len
= len
- linear
;
534 skb
->len
+= len
- linear
;
539 /* set skb frags from iovec, this can move to core network code for reuse */
540 static int zerocopy_sg_from_iovec(struct sk_buff
*skb
, const struct iovec
*from
,
541 int offset
, size_t count
)
543 int len
= iov_length(from
, count
) - offset
;
544 int copy
= skb_headlen(skb
);
545 int size
, offset1
= 0;
548 /* Skip over from offset */
549 while (count
&& (offset
>= from
->iov_len
)) {
550 offset
-= from
->iov_len
;
555 /* copy up to skb headlen */
556 while (count
&& (copy
> 0)) {
557 size
= min_t(unsigned int, copy
, from
->iov_len
- offset
);
558 if (copy_from_user(skb
->data
+ offset1
, from
->iov_base
+ offset
,
575 struct page
*page
[MAX_SKB_FRAGS
];
578 unsigned long truesize
;
580 len
= from
->iov_len
- offset
;
586 base
= (unsigned long)from
->iov_base
+ offset
;
587 size
= ((base
& ~PAGE_MASK
) + len
+ ~PAGE_MASK
) >> PAGE_SHIFT
;
588 if (i
+ size
> MAX_SKB_FRAGS
)
590 num_pages
= get_user_pages_fast(base
, size
, 0, &page
[i
]);
591 if (num_pages
!= size
) {
594 for (j
= 0; j
< num_pages
; j
++)
595 put_page(page
[i
+ j
]);
598 truesize
= size
* PAGE_SIZE
;
599 skb
->data_len
+= len
;
601 skb
->truesize
+= truesize
;
602 atomic_add(truesize
, &skb
->sk
->sk_wmem_alloc
);
604 int off
= base
& ~PAGE_MASK
;
605 int size
= min_t(int, len
, PAGE_SIZE
- off
);
606 __skb_fill_page_desc(skb
, i
, page
[i
], off
, size
);
607 skb_shinfo(skb
)->nr_frags
++;
608 /* increase sk_wmem_alloc */
620 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
621 * be shared with the tun/tap driver.
623 static int macvtap_skb_from_vnet_hdr(struct sk_buff
*skb
,
624 struct virtio_net_hdr
*vnet_hdr
)
626 unsigned short gso_type
= 0;
627 if (vnet_hdr
->gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
628 switch (vnet_hdr
->gso_type
& ~VIRTIO_NET_HDR_GSO_ECN
) {
629 case VIRTIO_NET_HDR_GSO_TCPV4
:
630 gso_type
= SKB_GSO_TCPV4
;
632 case VIRTIO_NET_HDR_GSO_TCPV6
:
633 gso_type
= SKB_GSO_TCPV6
;
635 case VIRTIO_NET_HDR_GSO_UDP
:
636 gso_type
= SKB_GSO_UDP
;
642 if (vnet_hdr
->gso_type
& VIRTIO_NET_HDR_GSO_ECN
)
643 gso_type
|= SKB_GSO_TCP_ECN
;
645 if (vnet_hdr
->gso_size
== 0)
649 if (vnet_hdr
->flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) {
650 if (!skb_partial_csum_set(skb
, vnet_hdr
->csum_start
,
651 vnet_hdr
->csum_offset
))
655 if (vnet_hdr
->gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
656 skb_shinfo(skb
)->gso_size
= vnet_hdr
->gso_size
;
657 skb_shinfo(skb
)->gso_type
= gso_type
;
659 /* Header must be checked, and gso_segs computed. */
660 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
661 skb_shinfo(skb
)->gso_segs
= 0;
666 static int macvtap_skb_to_vnet_hdr(const struct sk_buff
*skb
,
667 struct virtio_net_hdr
*vnet_hdr
)
669 memset(vnet_hdr
, 0, sizeof(*vnet_hdr
));
671 if (skb_is_gso(skb
)) {
672 struct skb_shared_info
*sinfo
= skb_shinfo(skb
);
674 /* This is a hint as to how much should be linear. */
675 vnet_hdr
->hdr_len
= skb_headlen(skb
);
676 vnet_hdr
->gso_size
= sinfo
->gso_size
;
677 if (sinfo
->gso_type
& SKB_GSO_TCPV4
)
678 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_TCPV4
;
679 else if (sinfo
->gso_type
& SKB_GSO_TCPV6
)
680 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_TCPV6
;
681 else if (sinfo
->gso_type
& SKB_GSO_UDP
)
682 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_UDP
;
685 if (sinfo
->gso_type
& SKB_GSO_TCP_ECN
)
686 vnet_hdr
->gso_type
|= VIRTIO_NET_HDR_GSO_ECN
;
688 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
690 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
691 vnet_hdr
->flags
= VIRTIO_NET_HDR_F_NEEDS_CSUM
;
692 vnet_hdr
->csum_start
= skb_checksum_start_offset(skb
);
693 vnet_hdr
->csum_offset
= skb
->csum_offset
;
694 } else if (skb
->ip_summed
== CHECKSUM_UNNECESSARY
) {
695 vnet_hdr
->flags
= VIRTIO_NET_HDR_F_DATA_VALID
;
696 } /* else everything is zero */
702 /* Get packet from user space buffer */
703 static ssize_t
macvtap_get_user(struct macvtap_queue
*q
, struct msghdr
*m
,
704 const struct iovec
*iv
, unsigned long total_len
,
705 size_t count
, int noblock
)
708 struct macvlan_dev
*vlan
;
709 unsigned long len
= total_len
;
711 struct virtio_net_hdr vnet_hdr
= { 0 };
712 int vnet_hdr_len
= 0;
714 bool zerocopy
= false;
716 if (q
->flags
& IFF_VNET_HDR
) {
717 vnet_hdr_len
= q
->vnet_hdr_sz
;
720 if (len
< vnet_hdr_len
)
724 err
= memcpy_fromiovecend((void *)&vnet_hdr
, iv
, 0,
728 if ((vnet_hdr
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) &&
729 vnet_hdr
.csum_start
+ vnet_hdr
.csum_offset
+ 2 >
731 vnet_hdr
.hdr_len
= vnet_hdr
.csum_start
+
732 vnet_hdr
.csum_offset
+ 2;
734 if (vnet_hdr
.hdr_len
> len
)
739 if (unlikely(len
< ETH_HLEN
))
743 if (unlikely(count
> UIO_MAXIOV
))
746 if (m
&& m
->msg_control
&& sock_flag(&q
->sk
, SOCK_ZEROCOPY
))
750 /* Userspace may produce vectors with count greater than
751 * MAX_SKB_FRAGS, so we need to linearize parts of the skb
752 * to let the rest of data to be fit in the frags.
754 if (count
> MAX_SKB_FRAGS
) {
755 copylen
= iov_length(iv
, count
- MAX_SKB_FRAGS
);
756 if (copylen
< vnet_hdr_len
)
759 copylen
-= vnet_hdr_len
;
761 /* There are 256 bytes to be copied in skb, so there is enough
762 * room for skb expand head in case it is used.
763 * The rest buffer is mapped from userspace.
765 if (copylen
< vnet_hdr
.hdr_len
)
766 copylen
= vnet_hdr
.hdr_len
;
768 copylen
= GOODCOPY_LEN
;
772 skb
= macvtap_alloc_skb(&q
->sk
, NET_IP_ALIGN
, copylen
,
773 vnet_hdr
.hdr_len
, noblock
, &err
);
778 err
= zerocopy_sg_from_iovec(skb
, iv
, vnet_hdr_len
, count
);
780 err
= skb_copy_datagram_from_iovec(skb
, 0, iv
, vnet_hdr_len
,
785 skb_set_network_header(skb
, ETH_HLEN
);
786 skb_reset_mac_header(skb
);
787 skb
->protocol
= eth_hdr(skb
)->h_proto
;
790 err
= macvtap_skb_from_vnet_hdr(skb
, &vnet_hdr
);
795 skb_probe_transport_header(skb
, ETH_HLEN
);
798 vlan
= rcu_dereference(q
->vlan
);
799 /* copy skb_ubuf_info for callback when skb has no error */
801 skb_shinfo(skb
)->destructor_arg
= m
->msg_control
;
802 skb_shinfo(skb
)->tx_flags
|= SKBTX_DEV_ZEROCOPY
;
803 skb_shinfo(skb
)->tx_flags
|= SKBTX_SHARED_FRAG
;
806 macvlan_start_xmit(skb
, vlan
->dev
);
818 vlan
= rcu_dereference(q
->vlan
);
820 vlan
->dev
->stats
.tx_dropped
++;
826 static ssize_t
macvtap_aio_write(struct kiocb
*iocb
, const struct iovec
*iv
,
827 unsigned long count
, loff_t pos
)
829 struct file
*file
= iocb
->ki_filp
;
830 ssize_t result
= -ENOLINK
;
831 struct macvtap_queue
*q
= file
->private_data
;
833 result
= macvtap_get_user(q
, NULL
, iv
, iov_length(iv
, count
), count
,
834 file
->f_flags
& O_NONBLOCK
);
838 /* Put packet to the user space buffer */
839 static ssize_t
macvtap_put_user(struct macvtap_queue
*q
,
840 const struct sk_buff
*skb
,
841 const struct iovec
*iv
, int len
)
843 struct macvlan_dev
*vlan
;
845 int vnet_hdr_len
= 0;
849 if (q
->flags
& IFF_VNET_HDR
) {
850 struct virtio_net_hdr vnet_hdr
;
851 vnet_hdr_len
= q
->vnet_hdr_sz
;
852 if ((len
-= vnet_hdr_len
) < 0)
855 ret
= macvtap_skb_to_vnet_hdr(skb
, &vnet_hdr
);
859 if (memcpy_toiovecend(iv
, (void *)&vnet_hdr
, 0, sizeof(vnet_hdr
)))
862 copied
= vnet_hdr_len
;
864 if (!vlan_tx_tag_present(skb
))
865 len
= min_t(int, skb
->len
, len
);
872 veth
.h_vlan_proto
= htons(ETH_P_8021Q
);
873 veth
.h_vlan_TCI
= htons(vlan_tx_tag_get(skb
));
875 vlan_offset
= offsetof(struct vlan_ethhdr
, h_vlan_proto
);
876 len
= min_t(int, skb
->len
+ VLAN_HLEN
, len
);
878 copy
= min_t(int, vlan_offset
, len
);
879 ret
= skb_copy_datagram_const_iovec(skb
, 0, iv
, copied
, copy
);
885 copy
= min_t(int, sizeof(veth
), len
);
886 ret
= memcpy_toiovecend(iv
, (void *)&veth
, copied
, copy
);
893 ret
= skb_copy_datagram_const_iovec(skb
, vlan_offset
, iv
, copied
, len
);
898 vlan
= rcu_dereference(q
->vlan
);
900 macvlan_count_rx(vlan
, copied
- vnet_hdr_len
, ret
== 0, 0);
903 return ret
? ret
: copied
;
906 static ssize_t
macvtap_do_read(struct macvtap_queue
*q
, struct kiocb
*iocb
,
907 const struct iovec
*iv
, unsigned long len
,
916 prepare_to_wait(sk_sleep(&q
->sk
), &wait
,
919 /* Read frames from the queue */
920 skb
= skb_dequeue(&q
->sk
.sk_receive_queue
);
926 if (signal_pending(current
)) {
930 /* Nothing to read, let's sleep */
934 ret
= macvtap_put_user(q
, skb
, iv
, len
);
940 finish_wait(sk_sleep(&q
->sk
), &wait
);
944 static ssize_t
macvtap_aio_read(struct kiocb
*iocb
, const struct iovec
*iv
,
945 unsigned long count
, loff_t pos
)
947 struct file
*file
= iocb
->ki_filp
;
948 struct macvtap_queue
*q
= file
->private_data
;
949 ssize_t len
, ret
= 0;
951 len
= iov_length(iv
, count
);
957 ret
= macvtap_do_read(q
, iocb
, iv
, len
, file
->f_flags
& O_NONBLOCK
);
958 ret
= min_t(ssize_t
, ret
, len
); /* XXX copied from tun.c. Why? */
963 static struct macvlan_dev
*macvtap_get_vlan(struct macvtap_queue
*q
)
965 struct macvlan_dev
*vlan
;
968 vlan
= rtnl_dereference(q
->vlan
);
975 static void macvtap_put_vlan(struct macvlan_dev
*vlan
)
980 static int macvtap_ioctl_set_queue(struct file
*file
, unsigned int flags
)
982 struct macvtap_queue
*q
= file
->private_data
;
983 struct macvlan_dev
*vlan
;
986 vlan
= macvtap_get_vlan(q
);
990 if (flags
& IFF_ATTACH_QUEUE
)
991 ret
= macvtap_enable_queue(vlan
->dev
, file
, q
);
992 else if (flags
& IFF_DETACH_QUEUE
)
993 ret
= macvtap_disable_queue(q
);
997 macvtap_put_vlan(vlan
);
1001 static int set_offload(struct macvtap_queue
*q
, unsigned long arg
)
1003 struct macvlan_dev
*vlan
;
1004 netdev_features_t features
;
1005 netdev_features_t feature_mask
= 0;
1007 vlan
= rtnl_dereference(q
->vlan
);
1011 features
= vlan
->dev
->features
;
1013 if (arg
& TUN_F_CSUM
) {
1014 feature_mask
= NETIF_F_HW_CSUM
;
1016 if (arg
& (TUN_F_TSO4
| TUN_F_TSO6
)) {
1017 if (arg
& TUN_F_TSO_ECN
)
1018 feature_mask
|= NETIF_F_TSO_ECN
;
1019 if (arg
& TUN_F_TSO4
)
1020 feature_mask
|= NETIF_F_TSO
;
1021 if (arg
& TUN_F_TSO6
)
1022 feature_mask
|= NETIF_F_TSO6
;
1025 if (arg
& TUN_F_UFO
)
1026 feature_mask
|= NETIF_F_UFO
;
1029 /* tun/tap driver inverts the usage for TSO offloads, where
1030 * setting the TSO bit means that the userspace wants to
1031 * accept TSO frames and turning it off means that user space
1032 * does not support TSO.
1033 * For macvtap, we have to invert it to mean the same thing.
1034 * When user space turns off TSO, we turn off GSO/LRO so that
1035 * user-space will not receive TSO frames.
1037 if (feature_mask
& (NETIF_F_TSO
| NETIF_F_TSO6
| NETIF_F_UFO
))
1038 features
|= RX_OFFLOADS
;
1040 features
&= ~RX_OFFLOADS
;
1042 /* tap_features are the same as features on tun/tap and
1043 * reflect user expectations.
1045 vlan
->tap_features
= vlan
->dev
->features
&
1046 (feature_mask
| ~TUN_OFFLOADS
);
1047 vlan
->set_features
= features
;
1048 netdev_update_features(vlan
->dev
);
1054 * provide compatibility with generic tun/tap interface
1056 static long macvtap_ioctl(struct file
*file
, unsigned int cmd
,
1059 struct macvtap_queue
*q
= file
->private_data
;
1060 struct macvlan_dev
*vlan
;
1061 void __user
*argp
= (void __user
*)arg
;
1062 struct ifreq __user
*ifr
= argp
;
1063 unsigned int __user
*up
= argp
;
1065 int __user
*sp
= argp
;
1071 /* ignore the name, just look at flags */
1072 if (get_user(u
, &ifr
->ifr_flags
))
1076 if ((u
& ~(IFF_VNET_HDR
| IFF_MULTI_QUEUE
)) !=
1077 (IFF_NO_PI
| IFF_TAP
))
1086 vlan
= macvtap_get_vlan(q
);
1093 if (copy_to_user(&ifr
->ifr_name
, vlan
->dev
->name
, IFNAMSIZ
) ||
1094 put_user(q
->flags
, &ifr
->ifr_flags
))
1096 macvtap_put_vlan(vlan
);
1101 if (get_user(u
, &ifr
->ifr_flags
))
1104 ret
= macvtap_ioctl_set_queue(file
, u
);
1107 case TUNGETFEATURES
:
1108 if (put_user(IFF_TAP
| IFF_NO_PI
| IFF_VNET_HDR
|
1109 IFF_MULTI_QUEUE
, up
))
1114 if (get_user(u
, up
))
1117 q
->sk
.sk_sndbuf
= u
;
1120 case TUNGETVNETHDRSZ
:
1122 if (put_user(s
, sp
))
1126 case TUNSETVNETHDRSZ
:
1127 if (get_user(s
, sp
))
1129 if (s
< (int)sizeof(struct virtio_net_hdr
))
1136 /* let the user check for future flags */
1137 if (arg
& ~(TUN_F_CSUM
| TUN_F_TSO4
| TUN_F_TSO6
|
1138 TUN_F_TSO_ECN
| TUN_F_UFO
))
1141 /* TODO: only accept frames with the features that
1142 got enabled for forwarded frames */
1143 if (!(q
->flags
& IFF_VNET_HDR
))
1146 ret
= set_offload(q
, arg
);
1155 #ifdef CONFIG_COMPAT
1156 static long macvtap_compat_ioctl(struct file
*file
, unsigned int cmd
,
1159 return macvtap_ioctl(file
, cmd
, (unsigned long)compat_ptr(arg
));
1163 static const struct file_operations macvtap_fops
= {
1164 .owner
= THIS_MODULE
,
1165 .open
= macvtap_open
,
1166 .release
= macvtap_release
,
1167 .aio_read
= macvtap_aio_read
,
1168 .aio_write
= macvtap_aio_write
,
1169 .poll
= macvtap_poll
,
1170 .llseek
= no_llseek
,
1171 .unlocked_ioctl
= macvtap_ioctl
,
1172 #ifdef CONFIG_COMPAT
1173 .compat_ioctl
= macvtap_compat_ioctl
,
1177 static int macvtap_sendmsg(struct kiocb
*iocb
, struct socket
*sock
,
1178 struct msghdr
*m
, size_t total_len
)
1180 struct macvtap_queue
*q
= container_of(sock
, struct macvtap_queue
, sock
);
1181 return macvtap_get_user(q
, m
, m
->msg_iov
, total_len
, m
->msg_iovlen
,
1182 m
->msg_flags
& MSG_DONTWAIT
);
1185 static int macvtap_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
1186 struct msghdr
*m
, size_t total_len
,
1189 struct macvtap_queue
*q
= container_of(sock
, struct macvtap_queue
, sock
);
1191 if (flags
& ~(MSG_DONTWAIT
|MSG_TRUNC
))
1193 ret
= macvtap_do_read(q
, iocb
, m
->msg_iov
, total_len
,
1194 flags
& MSG_DONTWAIT
);
1195 if (ret
> total_len
) {
1196 m
->msg_flags
|= MSG_TRUNC
;
1197 ret
= flags
& MSG_TRUNC
? ret
: total_len
;
1202 /* Ops structure to mimic raw sockets with tun */
1203 static const struct proto_ops macvtap_socket_ops
= {
1204 .sendmsg
= macvtap_sendmsg
,
1205 .recvmsg
= macvtap_recvmsg
,
1208 /* Get an underlying socket object from tun file. Returns error unless file is
1209 * attached to a device. The returned object works like a packet socket, it
1210 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
1211 * holding a reference to the file for as long as the socket is in use. */
1212 struct socket
*macvtap_get_socket(struct file
*file
)
1214 struct macvtap_queue
*q
;
1215 if (file
->f_op
!= &macvtap_fops
)
1216 return ERR_PTR(-EINVAL
);
1217 q
= file
->private_data
;
1219 return ERR_PTR(-EBADFD
);
1222 EXPORT_SYMBOL_GPL(macvtap_get_socket
);
1224 static int macvtap_device_event(struct notifier_block
*unused
,
1225 unsigned long event
, void *ptr
)
1227 struct net_device
*dev
= netdev_notifier_info_to_dev(ptr
);
1228 struct macvlan_dev
*vlan
;
1229 struct device
*classdev
;
1233 if (dev
->rtnl_link_ops
!= &macvtap_link_ops
)
1236 vlan
= netdev_priv(dev
);
1239 case NETDEV_REGISTER
:
1240 /* Create the device node here after the network device has
1241 * been registered but before register_netdevice has
1244 err
= macvtap_get_minor(vlan
);
1246 return notifier_from_errno(err
);
1248 devt
= MKDEV(MAJOR(macvtap_major
), vlan
->minor
);
1249 classdev
= device_create(macvtap_class
, &dev
->dev
, devt
,
1250 dev
, "tap%d", dev
->ifindex
);
1251 if (IS_ERR(classdev
)) {
1252 macvtap_free_minor(vlan
);
1253 return notifier_from_errno(PTR_ERR(classdev
));
1256 case NETDEV_UNREGISTER
:
1257 devt
= MKDEV(MAJOR(macvtap_major
), vlan
->minor
);
1258 device_destroy(macvtap_class
, devt
);
1259 macvtap_free_minor(vlan
);
1266 static struct notifier_block macvtap_notifier_block __read_mostly
= {
1267 .notifier_call
= macvtap_device_event
,
1270 static int macvtap_init(void)
1274 err
= alloc_chrdev_region(&macvtap_major
, 0,
1275 MACVTAP_NUM_DEVS
, "macvtap");
1279 cdev_init(&macvtap_cdev
, &macvtap_fops
);
1280 err
= cdev_add(&macvtap_cdev
, macvtap_major
, MACVTAP_NUM_DEVS
);
1284 macvtap_class
= class_create(THIS_MODULE
, "macvtap");
1285 if (IS_ERR(macvtap_class
)) {
1286 err
= PTR_ERR(macvtap_class
);
1290 err
= register_netdevice_notifier(&macvtap_notifier_block
);
1294 err
= macvlan_link_register(&macvtap_link_ops
);
1301 unregister_netdevice_notifier(&macvtap_notifier_block
);
1303 class_unregister(macvtap_class
);
1305 cdev_del(&macvtap_cdev
);
1307 unregister_chrdev_region(macvtap_major
, MACVTAP_NUM_DEVS
);
1311 module_init(macvtap_init
);
1313 static void macvtap_exit(void)
1315 rtnl_link_unregister(&macvtap_link_ops
);
1316 unregister_netdevice_notifier(&macvtap_notifier_block
);
1317 class_unregister(macvtap_class
);
1318 cdev_del(&macvtap_cdev
);
1319 unregister_chrdev_region(macvtap_major
, MACVTAP_NUM_DEVS
);
1321 module_exit(macvtap_exit
);
1323 MODULE_ALIAS_RTNL_LINK("macvtap");
1324 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1325 MODULE_LICENSE("GPL");