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)
71 #define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG)
75 * The macvtap_queue and the macvlan_dev are loosely coupled, the
76 * pointers from one to the other can only be read while rcu_read_lock
79 * Both the file and the macvlan_dev hold a reference on the macvtap_queue
80 * through sock_hold(&q->sk). When the macvlan_dev goes away first,
81 * q->vlan becomes inaccessible. When the files gets closed,
82 * macvtap_get_queue() fails.
84 * There may still be references to the struct sock inside of the
85 * queue from outbound SKBs, but these never reference back to the
86 * file or the dev. The data structure is freed through __sk_free
87 * when both our references and any pending SKBs are gone.
90 static int macvtap_enable_queue(struct net_device
*dev
, struct file
*file
,
91 struct macvtap_queue
*q
)
93 struct macvlan_dev
*vlan
= netdev_priv(dev
);
102 rcu_assign_pointer(vlan
->taps
[vlan
->numvtaps
], q
);
103 q
->queue_index
= vlan
->numvtaps
;
111 static int macvtap_set_queue(struct net_device
*dev
, struct file
*file
,
112 struct macvtap_queue
*q
)
114 struct macvlan_dev
*vlan
= netdev_priv(dev
);
118 if (vlan
->numqueues
== MAX_MACVTAP_QUEUES
)
122 rcu_assign_pointer(q
->vlan
, vlan
);
123 rcu_assign_pointer(vlan
->taps
[vlan
->numvtaps
], q
);
127 q
->queue_index
= vlan
->numvtaps
;
129 file
->private_data
= q
;
130 list_add_tail(&q
->next
, &vlan
->queue_list
);
140 static int macvtap_disable_queue(struct macvtap_queue
*q
)
142 struct macvlan_dev
*vlan
;
143 struct macvtap_queue
*nq
;
149 vlan
= rtnl_dereference(q
->vlan
);
152 int index
= q
->queue_index
;
153 BUG_ON(index
>= vlan
->numvtaps
);
154 nq
= rtnl_dereference(vlan
->taps
[vlan
->numvtaps
- 1]);
155 nq
->queue_index
= index
;
157 rcu_assign_pointer(vlan
->taps
[index
], nq
);
158 RCU_INIT_POINTER(vlan
->taps
[vlan
->numvtaps
- 1], NULL
);
168 * The file owning the queue got closed, give up both
169 * the reference that the files holds as well as the
170 * one from the macvlan_dev if that still exists.
172 * Using the spinlock makes sure that we don't get
173 * to the queue again after destroying it.
175 static void macvtap_put_queue(struct macvtap_queue
*q
)
177 struct macvlan_dev
*vlan
;
180 vlan
= rtnl_dereference(q
->vlan
);
184 BUG_ON(macvtap_disable_queue(q
));
187 RCU_INIT_POINTER(q
->vlan
, NULL
);
189 list_del_init(&q
->next
);
199 * Select a queue based on the rxq of the device on which this packet
200 * arrived. If the incoming device is not mq, calculate a flow hash
201 * to select a queue. If all fails, find the first available queue.
202 * Cache vlan->numvtaps since it can become zero during the execution
205 static struct macvtap_queue
*macvtap_get_queue(struct net_device
*dev
,
208 struct macvlan_dev
*vlan
= netdev_priv(dev
);
209 struct macvtap_queue
*tap
= NULL
;
210 /* Access to taps array is protected by rcu, but access to numvtaps
211 * isn't. Below we use it to lookup a queue, but treat it as a hint
212 * and validate that the result isn't NULL - in case we are
213 * racing against queue removal.
215 int numvtaps
= ACCESS_ONCE(vlan
->numvtaps
);
221 /* Check if we can use flow to select a queue */
222 rxq
= skb_get_rxhash(skb
);
224 tap
= rcu_dereference(vlan
->taps
[rxq
% numvtaps
]);
228 if (likely(skb_rx_queue_recorded(skb
))) {
229 rxq
= skb_get_rx_queue(skb
);
231 while (unlikely(rxq
>= numvtaps
))
234 tap
= rcu_dereference(vlan
->taps
[rxq
]);
238 tap
= rcu_dereference(vlan
->taps
[0]);
244 * The net_device is going away, give up the reference
245 * that it holds on all queues and safely set the pointer
246 * from the queues to NULL.
248 static void macvtap_del_queues(struct net_device
*dev
)
250 struct macvlan_dev
*vlan
= netdev_priv(dev
);
251 struct macvtap_queue
*q
, *tmp
, *qlist
[MAX_MACVTAP_QUEUES
];
255 list_for_each_entry_safe(q
, tmp
, &vlan
->queue_list
, next
) {
256 list_del_init(&q
->next
);
258 RCU_INIT_POINTER(q
->vlan
, NULL
);
263 for (i
= 0; i
< vlan
->numvtaps
; i
++)
264 RCU_INIT_POINTER(vlan
->taps
[i
], NULL
);
265 BUG_ON(vlan
->numvtaps
);
266 BUG_ON(vlan
->numqueues
);
267 /* guarantee that any future macvtap_set_queue will fail */
268 vlan
->numvtaps
= MAX_MACVTAP_QUEUES
;
270 for (--j
; j
>= 0; j
--)
271 sock_put(&qlist
[j
]->sk
);
275 * Forward happens for data that gets sent from one macvlan
276 * endpoint to another one in bridge mode. We just take
277 * the skb and put it into the receive queue.
279 static int macvtap_forward(struct net_device
*dev
, struct sk_buff
*skb
)
281 struct macvlan_dev
*vlan
= netdev_priv(dev
);
282 struct macvtap_queue
*q
= macvtap_get_queue(dev
, skb
);
283 netdev_features_t features
= TAP_FEATURES
;
288 if (skb_queue_len(&q
->sk
.sk_receive_queue
) >= dev
->tx_queue_len
)
292 /* Apply the forward feature mask so that we perform segmentation
293 * according to users wishes. This only works if VNET_HDR is
296 if (q
->flags
& IFF_VNET_HDR
)
297 features
|= vlan
->tap_features
;
298 if (netif_needs_gso(skb
, features
)) {
299 struct sk_buff
*segs
= __skb_gso_segment(skb
, features
, false);
305 skb_queue_tail(&q
->sk
.sk_receive_queue
, skb
);
311 struct sk_buff
*nskb
= segs
->next
;
314 skb_queue_tail(&q
->sk
.sk_receive_queue
, segs
);
318 skb_queue_tail(&q
->sk
.sk_receive_queue
, skb
);
322 wake_up_interruptible_poll(sk_sleep(&q
->sk
), POLLIN
| POLLRDNORM
| POLLRDBAND
);
323 return NET_RX_SUCCESS
;
331 * Receive is for data from the external interface (lowerdev),
332 * in case of macvtap, we can treat that the same way as
333 * forward, which macvlan cannot.
335 static int macvtap_receive(struct sk_buff
*skb
)
337 skb_push(skb
, ETH_HLEN
);
338 return macvtap_forward(skb
->dev
, skb
);
341 static int macvtap_get_minor(struct macvlan_dev
*vlan
)
343 int retval
= -ENOMEM
;
345 mutex_lock(&minor_lock
);
346 retval
= idr_alloc(&minor_idr
, vlan
, 1, MACVTAP_NUM_DEVS
, GFP_KERNEL
);
348 vlan
->minor
= retval
;
349 } else if (retval
== -ENOSPC
) {
350 printk(KERN_ERR
"too many macvtap devices\n");
353 mutex_unlock(&minor_lock
);
354 return retval
< 0 ? retval
: 0;
357 static void macvtap_free_minor(struct macvlan_dev
*vlan
)
359 mutex_lock(&minor_lock
);
361 idr_remove(&minor_idr
, vlan
->minor
);
364 mutex_unlock(&minor_lock
);
367 static struct net_device
*dev_get_by_macvtap_minor(int minor
)
369 struct net_device
*dev
= NULL
;
370 struct macvlan_dev
*vlan
;
372 mutex_lock(&minor_lock
);
373 vlan
= idr_find(&minor_idr
, minor
);
378 mutex_unlock(&minor_lock
);
382 static int macvtap_newlink(struct net
*src_net
,
383 struct net_device
*dev
,
385 struct nlattr
*data
[])
387 struct macvlan_dev
*vlan
= netdev_priv(dev
);
388 INIT_LIST_HEAD(&vlan
->queue_list
);
390 /* Since macvlan supports all offloads by default, make
391 * tap support all offloads also.
393 vlan
->tap_features
= TUN_OFFLOADS
;
395 /* Don't put anything that may fail after macvlan_common_newlink
396 * because we can't undo what it does.
398 return macvlan_common_newlink(src_net
, dev
, tb
, data
,
399 macvtap_receive
, macvtap_forward
);
402 static void macvtap_dellink(struct net_device
*dev
,
403 struct list_head
*head
)
405 macvtap_del_queues(dev
);
406 macvlan_dellink(dev
, head
);
409 static void macvtap_setup(struct net_device
*dev
)
411 macvlan_common_setup(dev
);
412 dev
->tx_queue_len
= TUN_READQ_SIZE
;
415 static struct rtnl_link_ops macvtap_link_ops __read_mostly
= {
417 .setup
= macvtap_setup
,
418 .newlink
= macvtap_newlink
,
419 .dellink
= macvtap_dellink
,
423 static void macvtap_sock_write_space(struct sock
*sk
)
425 wait_queue_head_t
*wqueue
;
427 if (!sock_writeable(sk
) ||
428 !test_and_clear_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
))
431 wqueue
= sk_sleep(sk
);
432 if (wqueue
&& waitqueue_active(wqueue
))
433 wake_up_interruptible_poll(wqueue
, POLLOUT
| POLLWRNORM
| POLLWRBAND
);
436 static void macvtap_sock_destruct(struct sock
*sk
)
438 skb_queue_purge(&sk
->sk_receive_queue
);
441 static int macvtap_open(struct inode
*inode
, struct file
*file
)
443 struct net
*net
= current
->nsproxy
->net_ns
;
444 struct net_device
*dev
= dev_get_by_macvtap_minor(iminor(inode
));
445 struct macvtap_queue
*q
;
453 q
= (struct macvtap_queue
*)sk_alloc(net
, AF_UNSPEC
, GFP_KERNEL
,
458 RCU_INIT_POINTER(q
->sock
.wq
, &q
->wq
);
459 init_waitqueue_head(&q
->wq
.wait
);
460 q
->sock
.type
= SOCK_RAW
;
461 q
->sock
.state
= SS_CONNECTED
;
463 q
->sock
.ops
= &macvtap_socket_ops
;
464 sock_init_data(&q
->sock
, &q
->sk
);
465 q
->sk
.sk_write_space
= macvtap_sock_write_space
;
466 q
->sk
.sk_destruct
= macvtap_sock_destruct
;
467 q
->flags
= IFF_VNET_HDR
| IFF_NO_PI
| IFF_TAP
;
468 q
->vnet_hdr_sz
= sizeof(struct virtio_net_hdr
);
471 * so far only KVM virtio_net uses macvtap, enable zero copy between
472 * guest kernel and host kernel when lower device supports zerocopy
474 * The macvlan supports zerocopy iff the lower device supports zero
475 * copy so we don't have to look at the lower device directly.
477 if ((dev
->features
& NETIF_F_HIGHDMA
) && (dev
->features
& NETIF_F_SG
))
478 sock_set_flag(&q
->sk
, SOCK_ZEROCOPY
);
480 err
= macvtap_set_queue(dev
, file
, q
);
491 static int macvtap_release(struct inode
*inode
, struct file
*file
)
493 struct macvtap_queue
*q
= file
->private_data
;
494 macvtap_put_queue(q
);
498 static unsigned int macvtap_poll(struct file
*file
, poll_table
* wait
)
500 struct macvtap_queue
*q
= file
->private_data
;
501 unsigned int mask
= POLLERR
;
507 poll_wait(file
, &q
->wq
.wait
, wait
);
509 if (!skb_queue_empty(&q
->sk
.sk_receive_queue
))
510 mask
|= POLLIN
| POLLRDNORM
;
512 if (sock_writeable(&q
->sk
) ||
513 (!test_and_set_bit(SOCK_ASYNC_NOSPACE
, &q
->sock
.flags
) &&
514 sock_writeable(&q
->sk
)))
515 mask
|= POLLOUT
| POLLWRNORM
;
521 static inline struct sk_buff
*macvtap_alloc_skb(struct sock
*sk
, size_t prepad
,
522 size_t len
, size_t linear
,
523 int noblock
, int *err
)
527 /* Under a page? Don't bother with paged skb. */
528 if (prepad
+ len
< PAGE_SIZE
|| !linear
)
531 skb
= sock_alloc_send_pskb(sk
, prepad
+ linear
, len
- linear
, noblock
,
536 skb_reserve(skb
, prepad
);
537 skb_put(skb
, linear
);
538 skb
->data_len
= len
- linear
;
539 skb
->len
+= len
- linear
;
544 /* set skb frags from iovec, this can move to core network code for reuse */
545 static int zerocopy_sg_from_iovec(struct sk_buff
*skb
, const struct iovec
*from
,
546 int offset
, size_t count
)
548 int len
= iov_length(from
, count
) - offset
;
549 int copy
= skb_headlen(skb
);
550 int size
, offset1
= 0;
553 /* Skip over from offset */
554 while (count
&& (offset
>= from
->iov_len
)) {
555 offset
-= from
->iov_len
;
560 /* copy up to skb headlen */
561 while (count
&& (copy
> 0)) {
562 size
= min_t(unsigned int, copy
, from
->iov_len
- offset
);
563 if (copy_from_user(skb
->data
+ offset1
, from
->iov_base
+ offset
,
580 struct page
*page
[MAX_SKB_FRAGS
];
583 unsigned long truesize
;
585 len
= from
->iov_len
- offset
;
591 base
= (unsigned long)from
->iov_base
+ offset
;
592 size
= ((base
& ~PAGE_MASK
) + len
+ ~PAGE_MASK
) >> PAGE_SHIFT
;
593 if (i
+ size
> MAX_SKB_FRAGS
)
595 num_pages
= get_user_pages_fast(base
, size
, 0, &page
[i
]);
596 if (num_pages
!= size
) {
599 for (j
= 0; j
< num_pages
; j
++)
600 put_page(page
[i
+ j
]);
603 truesize
= size
* PAGE_SIZE
;
604 skb
->data_len
+= len
;
606 skb
->truesize
+= truesize
;
607 atomic_add(truesize
, &skb
->sk
->sk_wmem_alloc
);
609 int off
= base
& ~PAGE_MASK
;
610 int size
= min_t(int, len
, PAGE_SIZE
- off
);
611 __skb_fill_page_desc(skb
, i
, page
[i
], off
, size
);
612 skb_shinfo(skb
)->nr_frags
++;
613 /* increase sk_wmem_alloc */
625 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
626 * be shared with the tun/tap driver.
628 static int macvtap_skb_from_vnet_hdr(struct sk_buff
*skb
,
629 struct virtio_net_hdr
*vnet_hdr
)
631 unsigned short gso_type
= 0;
632 if (vnet_hdr
->gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
633 switch (vnet_hdr
->gso_type
& ~VIRTIO_NET_HDR_GSO_ECN
) {
634 case VIRTIO_NET_HDR_GSO_TCPV4
:
635 gso_type
= SKB_GSO_TCPV4
;
637 case VIRTIO_NET_HDR_GSO_TCPV6
:
638 gso_type
= SKB_GSO_TCPV6
;
640 case VIRTIO_NET_HDR_GSO_UDP
:
641 gso_type
= SKB_GSO_UDP
;
647 if (vnet_hdr
->gso_type
& VIRTIO_NET_HDR_GSO_ECN
)
648 gso_type
|= SKB_GSO_TCP_ECN
;
650 if (vnet_hdr
->gso_size
== 0)
654 if (vnet_hdr
->flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) {
655 if (!skb_partial_csum_set(skb
, vnet_hdr
->csum_start
,
656 vnet_hdr
->csum_offset
))
660 if (vnet_hdr
->gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
661 skb_shinfo(skb
)->gso_size
= vnet_hdr
->gso_size
;
662 skb_shinfo(skb
)->gso_type
= gso_type
;
664 /* Header must be checked, and gso_segs computed. */
665 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
666 skb_shinfo(skb
)->gso_segs
= 0;
671 static int macvtap_skb_to_vnet_hdr(const struct sk_buff
*skb
,
672 struct virtio_net_hdr
*vnet_hdr
)
674 memset(vnet_hdr
, 0, sizeof(*vnet_hdr
));
676 if (skb_is_gso(skb
)) {
677 struct skb_shared_info
*sinfo
= skb_shinfo(skb
);
679 /* This is a hint as to how much should be linear. */
680 vnet_hdr
->hdr_len
= skb_headlen(skb
);
681 vnet_hdr
->gso_size
= sinfo
->gso_size
;
682 if (sinfo
->gso_type
& SKB_GSO_TCPV4
)
683 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_TCPV4
;
684 else if (sinfo
->gso_type
& SKB_GSO_TCPV6
)
685 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_TCPV6
;
686 else if (sinfo
->gso_type
& SKB_GSO_UDP
)
687 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_UDP
;
690 if (sinfo
->gso_type
& SKB_GSO_TCP_ECN
)
691 vnet_hdr
->gso_type
|= VIRTIO_NET_HDR_GSO_ECN
;
693 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
695 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
696 vnet_hdr
->flags
= VIRTIO_NET_HDR_F_NEEDS_CSUM
;
697 vnet_hdr
->csum_start
= skb_checksum_start_offset(skb
);
698 vnet_hdr
->csum_offset
= skb
->csum_offset
;
699 } else if (skb
->ip_summed
== CHECKSUM_UNNECESSARY
) {
700 vnet_hdr
->flags
= VIRTIO_NET_HDR_F_DATA_VALID
;
701 } /* else everything is zero */
706 static unsigned long iov_pages(const struct iovec
*iv
, int offset
,
707 unsigned long nr_segs
)
709 unsigned long seg
, base
;
710 int pages
= 0, len
, size
;
712 while (nr_segs
&& (offset
>= iv
->iov_len
)) {
713 offset
-= iv
->iov_len
;
718 for (seg
= 0; seg
< nr_segs
; seg
++) {
719 base
= (unsigned long)iv
[seg
].iov_base
+ offset
;
720 len
= iv
[seg
].iov_len
- offset
;
721 size
= ((base
& ~PAGE_MASK
) + len
+ ~PAGE_MASK
) >> PAGE_SHIFT
;
729 /* Get packet from user space buffer */
730 static ssize_t
macvtap_get_user(struct macvtap_queue
*q
, struct msghdr
*m
,
731 const struct iovec
*iv
, unsigned long total_len
,
732 size_t count
, int noblock
)
735 struct macvlan_dev
*vlan
;
736 unsigned long len
= total_len
;
738 struct virtio_net_hdr vnet_hdr
= { 0 };
739 int vnet_hdr_len
= 0;
741 bool zerocopy
= false;
744 if (q
->flags
& IFF_VNET_HDR
) {
745 vnet_hdr_len
= q
->vnet_hdr_sz
;
748 if (len
< vnet_hdr_len
)
752 err
= memcpy_fromiovecend((void *)&vnet_hdr
, iv
, 0,
756 if ((vnet_hdr
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) &&
757 vnet_hdr
.csum_start
+ vnet_hdr
.csum_offset
+ 2 >
759 vnet_hdr
.hdr_len
= vnet_hdr
.csum_start
+
760 vnet_hdr
.csum_offset
+ 2;
762 if (vnet_hdr
.hdr_len
> len
)
767 if (unlikely(len
< ETH_HLEN
))
771 if (unlikely(count
> UIO_MAXIOV
))
774 if (m
&& m
->msg_control
&& sock_flag(&q
->sk
, SOCK_ZEROCOPY
)) {
775 copylen
= vnet_hdr
.hdr_len
? vnet_hdr
.hdr_len
: GOODCOPY_LEN
;
777 if (iov_pages(iv
, vnet_hdr_len
+ copylen
, count
)
784 linear
= vnet_hdr
.hdr_len
;
787 skb
= macvtap_alloc_skb(&q
->sk
, NET_IP_ALIGN
, copylen
,
788 linear
, noblock
, &err
);
793 err
= zerocopy_sg_from_iovec(skb
, iv
, vnet_hdr_len
, count
);
795 err
= skb_copy_datagram_from_iovec(skb
, 0, iv
, vnet_hdr_len
,
797 if (!err
&& m
&& m
->msg_control
) {
798 struct ubuf_info
*uarg
= m
->msg_control
;
799 uarg
->callback(uarg
, false);
806 skb_set_network_header(skb
, ETH_HLEN
);
807 skb_reset_mac_header(skb
);
808 skb
->protocol
= eth_hdr(skb
)->h_proto
;
811 err
= macvtap_skb_from_vnet_hdr(skb
, &vnet_hdr
);
816 skb_probe_transport_header(skb
, ETH_HLEN
);
819 vlan
= rcu_dereference(q
->vlan
);
820 /* copy skb_ubuf_info for callback when skb has no error */
822 skb_shinfo(skb
)->destructor_arg
= m
->msg_control
;
823 skb_shinfo(skb
)->tx_flags
|= SKBTX_DEV_ZEROCOPY
;
824 skb_shinfo(skb
)->tx_flags
|= SKBTX_SHARED_FRAG
;
828 macvlan_start_xmit(skb
, vlan
->dev
);
842 vlan
= rcu_dereference(q
->vlan
);
844 vlan
->dev
->stats
.tx_dropped
++;
850 static ssize_t
macvtap_aio_write(struct kiocb
*iocb
, const struct iovec
*iv
,
851 unsigned long count
, loff_t pos
)
853 struct file
*file
= iocb
->ki_filp
;
854 ssize_t result
= -ENOLINK
;
855 struct macvtap_queue
*q
= file
->private_data
;
857 result
= macvtap_get_user(q
, NULL
, iv
, iov_length(iv
, count
), count
,
858 file
->f_flags
& O_NONBLOCK
);
862 /* Put packet to the user space buffer */
863 static ssize_t
macvtap_put_user(struct macvtap_queue
*q
,
864 const struct sk_buff
*skb
,
865 const struct iovec
*iv
, int len
)
867 struct macvlan_dev
*vlan
;
869 int vnet_hdr_len
= 0;
873 if (q
->flags
& IFF_VNET_HDR
) {
874 struct virtio_net_hdr vnet_hdr
;
875 vnet_hdr_len
= q
->vnet_hdr_sz
;
876 if ((len
-= vnet_hdr_len
) < 0)
879 ret
= macvtap_skb_to_vnet_hdr(skb
, &vnet_hdr
);
883 if (memcpy_toiovecend(iv
, (void *)&vnet_hdr
, 0, sizeof(vnet_hdr
)))
886 copied
= vnet_hdr_len
;
888 if (!vlan_tx_tag_present(skb
))
889 len
= min_t(int, skb
->len
, len
);
896 veth
.h_vlan_proto
= skb
->vlan_proto
;
897 veth
.h_vlan_TCI
= htons(vlan_tx_tag_get(skb
));
899 vlan_offset
= offsetof(struct vlan_ethhdr
, h_vlan_proto
);
900 len
= min_t(int, skb
->len
+ VLAN_HLEN
, len
);
902 copy
= min_t(int, vlan_offset
, len
);
903 ret
= skb_copy_datagram_const_iovec(skb
, 0, iv
, copied
, copy
);
909 copy
= min_t(int, sizeof(veth
), len
);
910 ret
= memcpy_toiovecend(iv
, (void *)&veth
, copied
, copy
);
917 ret
= skb_copy_datagram_const_iovec(skb
, vlan_offset
, iv
, copied
, len
);
922 vlan
= rcu_dereference(q
->vlan
);
925 macvlan_count_rx(vlan
, copied
- vnet_hdr_len
, ret
== 0, 0);
930 return ret
? ret
: copied
;
933 static ssize_t
macvtap_do_read(struct macvtap_queue
*q
, struct kiocb
*iocb
,
934 const struct iovec
*iv
, unsigned long len
,
943 prepare_to_wait(sk_sleep(&q
->sk
), &wait
,
946 /* Read frames from the queue */
947 skb
= skb_dequeue(&q
->sk
.sk_receive_queue
);
953 if (signal_pending(current
)) {
957 /* Nothing to read, let's sleep */
961 ret
= macvtap_put_user(q
, skb
, iv
, len
);
967 finish_wait(sk_sleep(&q
->sk
), &wait
);
971 static ssize_t
macvtap_aio_read(struct kiocb
*iocb
, const struct iovec
*iv
,
972 unsigned long count
, loff_t pos
)
974 struct file
*file
= iocb
->ki_filp
;
975 struct macvtap_queue
*q
= file
->private_data
;
976 ssize_t len
, ret
= 0;
978 len
= iov_length(iv
, count
);
984 ret
= macvtap_do_read(q
, iocb
, iv
, len
, file
->f_flags
& O_NONBLOCK
);
985 ret
= min_t(ssize_t
, ret
, len
); /* XXX copied from tun.c. Why? */
990 static struct macvlan_dev
*macvtap_get_vlan(struct macvtap_queue
*q
)
992 struct macvlan_dev
*vlan
;
995 vlan
= rtnl_dereference(q
->vlan
);
1002 static void macvtap_put_vlan(struct macvlan_dev
*vlan
)
1007 static int macvtap_ioctl_set_queue(struct file
*file
, unsigned int flags
)
1009 struct macvtap_queue
*q
= file
->private_data
;
1010 struct macvlan_dev
*vlan
;
1013 vlan
= macvtap_get_vlan(q
);
1017 if (flags
& IFF_ATTACH_QUEUE
)
1018 ret
= macvtap_enable_queue(vlan
->dev
, file
, q
);
1019 else if (flags
& IFF_DETACH_QUEUE
)
1020 ret
= macvtap_disable_queue(q
);
1024 macvtap_put_vlan(vlan
);
1028 static int set_offload(struct macvtap_queue
*q
, unsigned long arg
)
1030 struct macvlan_dev
*vlan
;
1031 netdev_features_t features
;
1032 netdev_features_t feature_mask
= 0;
1034 vlan
= rtnl_dereference(q
->vlan
);
1038 features
= vlan
->dev
->features
;
1040 if (arg
& TUN_F_CSUM
) {
1041 feature_mask
= NETIF_F_HW_CSUM
;
1043 if (arg
& (TUN_F_TSO4
| TUN_F_TSO6
)) {
1044 if (arg
& TUN_F_TSO_ECN
)
1045 feature_mask
|= NETIF_F_TSO_ECN
;
1046 if (arg
& TUN_F_TSO4
)
1047 feature_mask
|= NETIF_F_TSO
;
1048 if (arg
& TUN_F_TSO6
)
1049 feature_mask
|= NETIF_F_TSO6
;
1052 if (arg
& TUN_F_UFO
)
1053 feature_mask
|= NETIF_F_UFO
;
1056 /* tun/tap driver inverts the usage for TSO offloads, where
1057 * setting the TSO bit means that the userspace wants to
1058 * accept TSO frames and turning it off means that user space
1059 * does not support TSO.
1060 * For macvtap, we have to invert it to mean the same thing.
1061 * When user space turns off TSO, we turn off GSO/LRO so that
1062 * user-space will not receive TSO frames.
1064 if (feature_mask
& (NETIF_F_TSO
| NETIF_F_TSO6
| NETIF_F_UFO
))
1065 features
|= RX_OFFLOADS
;
1067 features
&= ~RX_OFFLOADS
;
1069 /* tap_features are the same as features on tun/tap and
1070 * reflect user expectations.
1072 vlan
->tap_features
= feature_mask
;
1073 vlan
->set_features
= features
;
1074 netdev_update_features(vlan
->dev
);
1080 * provide compatibility with generic tun/tap interface
1082 static long macvtap_ioctl(struct file
*file
, unsigned int cmd
,
1085 struct macvtap_queue
*q
= file
->private_data
;
1086 struct macvlan_dev
*vlan
;
1087 void __user
*argp
= (void __user
*)arg
;
1088 struct ifreq __user
*ifr
= argp
;
1089 unsigned int __user
*up
= argp
;
1091 int __user
*sp
= argp
;
1097 /* ignore the name, just look at flags */
1098 if (get_user(u
, &ifr
->ifr_flags
))
1102 if ((u
& ~(IFF_VNET_HDR
| IFF_MULTI_QUEUE
)) !=
1103 (IFF_NO_PI
| IFF_TAP
))
1112 vlan
= macvtap_get_vlan(q
);
1119 if (copy_to_user(&ifr
->ifr_name
, vlan
->dev
->name
, IFNAMSIZ
) ||
1120 put_user(q
->flags
, &ifr
->ifr_flags
))
1122 macvtap_put_vlan(vlan
);
1127 if (get_user(u
, &ifr
->ifr_flags
))
1130 ret
= macvtap_ioctl_set_queue(file
, u
);
1134 case TUNGETFEATURES
:
1135 if (put_user(IFF_TAP
| IFF_NO_PI
| IFF_VNET_HDR
|
1136 IFF_MULTI_QUEUE
, up
))
1141 if (get_user(u
, up
))
1144 q
->sk
.sk_sndbuf
= u
;
1147 case TUNGETVNETHDRSZ
:
1149 if (put_user(s
, sp
))
1153 case TUNSETVNETHDRSZ
:
1154 if (get_user(s
, sp
))
1156 if (s
< (int)sizeof(struct virtio_net_hdr
))
1163 /* let the user check for future flags */
1164 if (arg
& ~(TUN_F_CSUM
| TUN_F_TSO4
| TUN_F_TSO6
|
1165 TUN_F_TSO_ECN
| TUN_F_UFO
))
1169 ret
= set_offload(q
, arg
);
1178 #ifdef CONFIG_COMPAT
1179 static long macvtap_compat_ioctl(struct file
*file
, unsigned int cmd
,
1182 return macvtap_ioctl(file
, cmd
, (unsigned long)compat_ptr(arg
));
1186 static const struct file_operations macvtap_fops
= {
1187 .owner
= THIS_MODULE
,
1188 .open
= macvtap_open
,
1189 .release
= macvtap_release
,
1190 .aio_read
= macvtap_aio_read
,
1191 .aio_write
= macvtap_aio_write
,
1192 .poll
= macvtap_poll
,
1193 .llseek
= no_llseek
,
1194 .unlocked_ioctl
= macvtap_ioctl
,
1195 #ifdef CONFIG_COMPAT
1196 .compat_ioctl
= macvtap_compat_ioctl
,
1200 static int macvtap_sendmsg(struct kiocb
*iocb
, struct socket
*sock
,
1201 struct msghdr
*m
, size_t total_len
)
1203 struct macvtap_queue
*q
= container_of(sock
, struct macvtap_queue
, sock
);
1204 return macvtap_get_user(q
, m
, m
->msg_iov
, total_len
, m
->msg_iovlen
,
1205 m
->msg_flags
& MSG_DONTWAIT
);
1208 static int macvtap_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
1209 struct msghdr
*m
, size_t total_len
,
1212 struct macvtap_queue
*q
= container_of(sock
, struct macvtap_queue
, sock
);
1214 if (flags
& ~(MSG_DONTWAIT
|MSG_TRUNC
))
1216 ret
= macvtap_do_read(q
, iocb
, m
->msg_iov
, total_len
,
1217 flags
& MSG_DONTWAIT
);
1218 if (ret
> total_len
) {
1219 m
->msg_flags
|= MSG_TRUNC
;
1220 ret
= flags
& MSG_TRUNC
? ret
: total_len
;
1225 /* Ops structure to mimic raw sockets with tun */
1226 static const struct proto_ops macvtap_socket_ops
= {
1227 .sendmsg
= macvtap_sendmsg
,
1228 .recvmsg
= macvtap_recvmsg
,
1231 /* Get an underlying socket object from tun file. Returns error unless file is
1232 * attached to a device. The returned object works like a packet socket, it
1233 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
1234 * holding a reference to the file for as long as the socket is in use. */
1235 struct socket
*macvtap_get_socket(struct file
*file
)
1237 struct macvtap_queue
*q
;
1238 if (file
->f_op
!= &macvtap_fops
)
1239 return ERR_PTR(-EINVAL
);
1240 q
= file
->private_data
;
1242 return ERR_PTR(-EBADFD
);
1245 EXPORT_SYMBOL_GPL(macvtap_get_socket
);
1247 static int macvtap_device_event(struct notifier_block
*unused
,
1248 unsigned long event
, void *ptr
)
1250 struct net_device
*dev
= netdev_notifier_info_to_dev(ptr
);
1251 struct macvlan_dev
*vlan
;
1252 struct device
*classdev
;
1256 if (dev
->rtnl_link_ops
!= &macvtap_link_ops
)
1259 vlan
= netdev_priv(dev
);
1262 case NETDEV_REGISTER
:
1263 /* Create the device node here after the network device has
1264 * been registered but before register_netdevice has
1267 err
= macvtap_get_minor(vlan
);
1269 return notifier_from_errno(err
);
1271 devt
= MKDEV(MAJOR(macvtap_major
), vlan
->minor
);
1272 classdev
= device_create(macvtap_class
, &dev
->dev
, devt
,
1273 dev
, "tap%d", dev
->ifindex
);
1274 if (IS_ERR(classdev
)) {
1275 macvtap_free_minor(vlan
);
1276 return notifier_from_errno(PTR_ERR(classdev
));
1279 case NETDEV_UNREGISTER
:
1280 devt
= MKDEV(MAJOR(macvtap_major
), vlan
->minor
);
1281 device_destroy(macvtap_class
, devt
);
1282 macvtap_free_minor(vlan
);
1289 static struct notifier_block macvtap_notifier_block __read_mostly
= {
1290 .notifier_call
= macvtap_device_event
,
1293 static int macvtap_init(void)
1297 err
= alloc_chrdev_region(&macvtap_major
, 0,
1298 MACVTAP_NUM_DEVS
, "macvtap");
1302 cdev_init(&macvtap_cdev
, &macvtap_fops
);
1303 err
= cdev_add(&macvtap_cdev
, macvtap_major
, MACVTAP_NUM_DEVS
);
1307 macvtap_class
= class_create(THIS_MODULE
, "macvtap");
1308 if (IS_ERR(macvtap_class
)) {
1309 err
= PTR_ERR(macvtap_class
);
1313 err
= register_netdevice_notifier(&macvtap_notifier_block
);
1317 err
= macvlan_link_register(&macvtap_link_ops
);
1324 unregister_netdevice_notifier(&macvtap_notifier_block
);
1326 class_unregister(macvtap_class
);
1328 cdev_del(&macvtap_cdev
);
1330 unregister_chrdev_region(macvtap_major
, MACVTAP_NUM_DEVS
);
1334 module_init(macvtap_init
);
1336 static void macvtap_exit(void)
1338 rtnl_link_unregister(&macvtap_link_ops
);
1339 unregister_netdevice_notifier(&macvtap_notifier_block
);
1340 class_unregister(macvtap_class
);
1341 cdev_del(&macvtap_cdev
);
1342 unregister_chrdev_region(macvtap_major
, MACVTAP_NUM_DEVS
);
1344 module_exit(macvtap_exit
);
1346 MODULE_ALIAS_RTNL_LINK("macvtap");
1347 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1348 MODULE_LICENSE("GPL");