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/wait.h>
15 #include <linux/cdev.h>
16 #include <linux/idr.h>
19 #include <net/net_namespace.h>
20 #include <net/rtnetlink.h>
22 #include <linux/virtio_net.h>
25 * A macvtap queue is the central object of this driver, it connects
26 * an open character device to a macvlan interface. There can be
27 * multiple queues on one interface, which map back to queues
28 * implemented in hardware on the underlying device.
30 * macvtap_proto is used to allocate queues through the sock allocation
34 struct macvtap_queue
{
39 struct macvlan_dev __rcu
*vlan
;
44 struct list_head next
;
47 static struct proto macvtap_proto
= {
50 .obj_size
= sizeof (struct macvtap_queue
),
54 * Variables for dealing with macvtaps device numbers.
56 static dev_t macvtap_major
;
57 #define MACVTAP_NUM_DEVS (1U << MINORBITS)
58 static DEFINE_MUTEX(minor_lock
);
59 static DEFINE_IDR(minor_idr
);
61 #define GOODCOPY_LEN 128
62 static struct class *macvtap_class
;
63 static struct cdev macvtap_cdev
;
65 static const struct proto_ops macvtap_socket_ops
;
67 #define TUN_OFFLOADS (NETIF_F_HW_CSUM | NETIF_F_TSO_ECN | NETIF_F_TSO | \
68 NETIF_F_TSO6 | NETIF_F_UFO)
69 #define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO)
70 #define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG)
72 static struct macvlan_dev
*macvtap_get_vlan_rcu(const struct net_device
*dev
)
74 return rcu_dereference(dev
->rx_handler_data
);
79 * The macvtap_queue and the macvlan_dev are loosely coupled, the
80 * pointers from one to the other can only be read while rcu_read_lock
83 * Both the file and the macvlan_dev hold a reference on the macvtap_queue
84 * through sock_hold(&q->sk). When the macvlan_dev goes away first,
85 * q->vlan becomes inaccessible. When the files gets closed,
86 * macvtap_get_queue() fails.
88 * There may still be references to the struct sock inside of the
89 * queue from outbound SKBs, but these never reference back to the
90 * file or the dev. The data structure is freed through __sk_free
91 * when both our references and any pending SKBs are gone.
94 static int macvtap_enable_queue(struct net_device
*dev
, struct file
*file
,
95 struct macvtap_queue
*q
)
97 struct macvlan_dev
*vlan
= netdev_priv(dev
);
106 rcu_assign_pointer(vlan
->taps
[vlan
->numvtaps
], q
);
107 q
->queue_index
= vlan
->numvtaps
;
115 static int macvtap_set_queue(struct net_device
*dev
, struct file
*file
,
116 struct macvtap_queue
*q
)
118 struct macvlan_dev
*vlan
= netdev_priv(dev
);
122 if (vlan
->numqueues
== MAX_MACVTAP_QUEUES
)
126 rcu_assign_pointer(q
->vlan
, vlan
);
127 rcu_assign_pointer(vlan
->taps
[vlan
->numvtaps
], q
);
131 q
->queue_index
= vlan
->numvtaps
;
133 file
->private_data
= q
;
134 list_add_tail(&q
->next
, &vlan
->queue_list
);
144 static int macvtap_disable_queue(struct macvtap_queue
*q
)
146 struct macvlan_dev
*vlan
;
147 struct macvtap_queue
*nq
;
153 vlan
= rtnl_dereference(q
->vlan
);
156 int index
= q
->queue_index
;
157 BUG_ON(index
>= vlan
->numvtaps
);
158 nq
= rtnl_dereference(vlan
->taps
[vlan
->numvtaps
- 1]);
159 nq
->queue_index
= index
;
161 rcu_assign_pointer(vlan
->taps
[index
], nq
);
162 RCU_INIT_POINTER(vlan
->taps
[vlan
->numvtaps
- 1], NULL
);
172 * The file owning the queue got closed, give up both
173 * the reference that the files holds as well as the
174 * one from the macvlan_dev if that still exists.
176 * Using the spinlock makes sure that we don't get
177 * to the queue again after destroying it.
179 static void macvtap_put_queue(struct macvtap_queue
*q
)
181 struct macvlan_dev
*vlan
;
184 vlan
= rtnl_dereference(q
->vlan
);
188 BUG_ON(macvtap_disable_queue(q
));
191 RCU_INIT_POINTER(q
->vlan
, NULL
);
193 list_del_init(&q
->next
);
203 * Select a queue based on the rxq of the device on which this packet
204 * arrived. If the incoming device is not mq, calculate a flow hash
205 * to select a queue. If all fails, find the first available queue.
206 * Cache vlan->numvtaps since it can become zero during the execution
209 static struct macvtap_queue
*macvtap_get_queue(struct net_device
*dev
,
212 struct macvlan_dev
*vlan
= netdev_priv(dev
);
213 struct macvtap_queue
*tap
= NULL
;
214 /* Access to taps array is protected by rcu, but access to numvtaps
215 * isn't. Below we use it to lookup a queue, but treat it as a hint
216 * and validate that the result isn't NULL - in case we are
217 * racing against queue removal.
219 int numvtaps
= ACCESS_ONCE(vlan
->numvtaps
);
225 /* Check if we can use flow to select a queue */
226 rxq
= skb_get_hash(skb
);
228 tap
= rcu_dereference(vlan
->taps
[rxq
% numvtaps
]);
232 if (likely(skb_rx_queue_recorded(skb
))) {
233 rxq
= skb_get_rx_queue(skb
);
235 while (unlikely(rxq
>= numvtaps
))
238 tap
= rcu_dereference(vlan
->taps
[rxq
]);
242 tap
= rcu_dereference(vlan
->taps
[0]);
248 * The net_device is going away, give up the reference
249 * that it holds on all queues and safely set the pointer
250 * from the queues to NULL.
252 static void macvtap_del_queues(struct net_device
*dev
)
254 struct macvlan_dev
*vlan
= netdev_priv(dev
);
255 struct macvtap_queue
*q
, *tmp
, *qlist
[MAX_MACVTAP_QUEUES
];
259 list_for_each_entry_safe(q
, tmp
, &vlan
->queue_list
, next
) {
260 list_del_init(&q
->next
);
262 RCU_INIT_POINTER(q
->vlan
, NULL
);
267 for (i
= 0; i
< vlan
->numvtaps
; i
++)
268 RCU_INIT_POINTER(vlan
->taps
[i
], NULL
);
269 BUG_ON(vlan
->numvtaps
);
270 BUG_ON(vlan
->numqueues
);
271 /* guarantee that any future macvtap_set_queue will fail */
272 vlan
->numvtaps
= MAX_MACVTAP_QUEUES
;
274 for (--j
; j
>= 0; j
--)
275 sock_put(&qlist
[j
]->sk
);
278 static rx_handler_result_t
macvtap_handle_frame(struct sk_buff
**pskb
)
280 struct sk_buff
*skb
= *pskb
;
281 struct net_device
*dev
= skb
->dev
;
282 struct macvlan_dev
*vlan
;
283 struct macvtap_queue
*q
;
284 netdev_features_t features
= TAP_FEATURES
;
286 vlan
= macvtap_get_vlan_rcu(dev
);
288 return RX_HANDLER_PASS
;
290 q
= macvtap_get_queue(dev
, skb
);
292 return RX_HANDLER_PASS
;
294 if (skb_queue_len(&q
->sk
.sk_receive_queue
) >= dev
->tx_queue_len
)
297 skb_push(skb
, ETH_HLEN
);
299 /* Apply the forward feature mask so that we perform segmentation
300 * according to users wishes. This only works if VNET_HDR is
303 if (q
->flags
& IFF_VNET_HDR
)
304 features
|= vlan
->tap_features
;
305 if (netif_needs_gso(skb
, features
)) {
306 struct sk_buff
*segs
= __skb_gso_segment(skb
, features
, false);
312 skb_queue_tail(&q
->sk
.sk_receive_queue
, skb
);
318 struct sk_buff
*nskb
= segs
->next
;
321 skb_queue_tail(&q
->sk
.sk_receive_queue
, segs
);
325 /* If we receive a partial checksum and the tap side
326 * doesn't support checksum offload, compute the checksum.
327 * Note: it doesn't matter which checksum feature to
328 * check, we either support them all or none.
330 if (skb
->ip_summed
== CHECKSUM_PARTIAL
&&
331 !(features
& NETIF_F_ALL_CSUM
) &&
332 skb_checksum_help(skb
))
334 skb_queue_tail(&q
->sk
.sk_receive_queue
, skb
);
338 wake_up_interruptible_poll(sk_sleep(&q
->sk
), POLLIN
| POLLRDNORM
| POLLRDBAND
);
339 return RX_HANDLER_CONSUMED
;
342 /* Count errors/drops only here, thus don't care about args. */
343 macvlan_count_rx(vlan
, 0, 0, 0);
345 return RX_HANDLER_CONSUMED
;
348 static int macvtap_get_minor(struct macvlan_dev
*vlan
)
350 int retval
= -ENOMEM
;
352 mutex_lock(&minor_lock
);
353 retval
= idr_alloc(&minor_idr
, vlan
, 1, MACVTAP_NUM_DEVS
, GFP_KERNEL
);
355 vlan
->minor
= retval
;
356 } else if (retval
== -ENOSPC
) {
357 printk(KERN_ERR
"too many macvtap devices\n");
360 mutex_unlock(&minor_lock
);
361 return retval
< 0 ? retval
: 0;
364 static void macvtap_free_minor(struct macvlan_dev
*vlan
)
366 mutex_lock(&minor_lock
);
368 idr_remove(&minor_idr
, vlan
->minor
);
371 mutex_unlock(&minor_lock
);
374 static struct net_device
*dev_get_by_macvtap_minor(int minor
)
376 struct net_device
*dev
= NULL
;
377 struct macvlan_dev
*vlan
;
379 mutex_lock(&minor_lock
);
380 vlan
= idr_find(&minor_idr
, minor
);
385 mutex_unlock(&minor_lock
);
389 static int macvtap_newlink(struct net
*src_net
,
390 struct net_device
*dev
,
392 struct nlattr
*data
[])
394 struct macvlan_dev
*vlan
= netdev_priv(dev
);
397 INIT_LIST_HEAD(&vlan
->queue_list
);
399 /* Since macvlan supports all offloads by default, make
400 * tap support all offloads also.
402 vlan
->tap_features
= TUN_OFFLOADS
;
404 err
= netdev_rx_handler_register(dev
, macvtap_handle_frame
, vlan
);
408 /* Don't put anything that may fail after macvlan_common_newlink
409 * because we can't undo what it does.
411 return macvlan_common_newlink(src_net
, dev
, tb
, data
);
414 static void macvtap_dellink(struct net_device
*dev
,
415 struct list_head
*head
)
417 netdev_rx_handler_unregister(dev
);
418 macvtap_del_queues(dev
);
419 macvlan_dellink(dev
, head
);
422 static void macvtap_setup(struct net_device
*dev
)
424 macvlan_common_setup(dev
);
425 dev
->tx_queue_len
= TUN_READQ_SIZE
;
428 static struct rtnl_link_ops macvtap_link_ops __read_mostly
= {
430 .setup
= macvtap_setup
,
431 .newlink
= macvtap_newlink
,
432 .dellink
= macvtap_dellink
,
436 static void macvtap_sock_write_space(struct sock
*sk
)
438 wait_queue_head_t
*wqueue
;
440 if (!sock_writeable(sk
) ||
441 !test_and_clear_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
))
444 wqueue
= sk_sleep(sk
);
445 if (wqueue
&& waitqueue_active(wqueue
))
446 wake_up_interruptible_poll(wqueue
, POLLOUT
| POLLWRNORM
| POLLWRBAND
);
449 static void macvtap_sock_destruct(struct sock
*sk
)
451 skb_queue_purge(&sk
->sk_receive_queue
);
454 static int macvtap_open(struct inode
*inode
, struct file
*file
)
456 struct net
*net
= current
->nsproxy
->net_ns
;
457 struct net_device
*dev
= dev_get_by_macvtap_minor(iminor(inode
));
458 struct macvtap_queue
*q
;
466 q
= (struct macvtap_queue
*)sk_alloc(net
, AF_UNSPEC
, GFP_KERNEL
,
471 RCU_INIT_POINTER(q
->sock
.wq
, &q
->wq
);
472 init_waitqueue_head(&q
->wq
.wait
);
473 q
->sock
.type
= SOCK_RAW
;
474 q
->sock
.state
= SS_CONNECTED
;
476 q
->sock
.ops
= &macvtap_socket_ops
;
477 sock_init_data(&q
->sock
, &q
->sk
);
478 q
->sk
.sk_write_space
= macvtap_sock_write_space
;
479 q
->sk
.sk_destruct
= macvtap_sock_destruct
;
480 q
->flags
= IFF_VNET_HDR
| IFF_NO_PI
| IFF_TAP
;
481 q
->vnet_hdr_sz
= sizeof(struct virtio_net_hdr
);
484 * so far only KVM virtio_net uses macvtap, enable zero copy between
485 * guest kernel and host kernel when lower device supports zerocopy
487 * The macvlan supports zerocopy iff the lower device supports zero
488 * copy so we don't have to look at the lower device directly.
490 if ((dev
->features
& NETIF_F_HIGHDMA
) && (dev
->features
& NETIF_F_SG
))
491 sock_set_flag(&q
->sk
, SOCK_ZEROCOPY
);
493 err
= macvtap_set_queue(dev
, file
, q
);
504 static int macvtap_release(struct inode
*inode
, struct file
*file
)
506 struct macvtap_queue
*q
= file
->private_data
;
507 macvtap_put_queue(q
);
511 static unsigned int macvtap_poll(struct file
*file
, poll_table
* wait
)
513 struct macvtap_queue
*q
= file
->private_data
;
514 unsigned int mask
= POLLERR
;
520 poll_wait(file
, &q
->wq
.wait
, wait
);
522 if (!skb_queue_empty(&q
->sk
.sk_receive_queue
))
523 mask
|= POLLIN
| POLLRDNORM
;
525 if (sock_writeable(&q
->sk
) ||
526 (!test_and_set_bit(SOCK_ASYNC_NOSPACE
, &q
->sock
.flags
) &&
527 sock_writeable(&q
->sk
)))
528 mask
|= POLLOUT
| POLLWRNORM
;
534 static inline struct sk_buff
*macvtap_alloc_skb(struct sock
*sk
, size_t prepad
,
535 size_t len
, size_t linear
,
536 int noblock
, int *err
)
540 /* Under a page? Don't bother with paged skb. */
541 if (prepad
+ len
< PAGE_SIZE
|| !linear
)
544 skb
= sock_alloc_send_pskb(sk
, prepad
+ linear
, len
- linear
, noblock
,
549 skb_reserve(skb
, prepad
);
550 skb_put(skb
, linear
);
551 skb
->data_len
= len
- linear
;
552 skb
->len
+= len
- linear
;
558 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
559 * be shared with the tun/tap driver.
561 static int macvtap_skb_from_vnet_hdr(struct sk_buff
*skb
,
562 struct virtio_net_hdr
*vnet_hdr
)
564 unsigned short gso_type
= 0;
565 if (vnet_hdr
->gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
566 switch (vnet_hdr
->gso_type
& ~VIRTIO_NET_HDR_GSO_ECN
) {
567 case VIRTIO_NET_HDR_GSO_TCPV4
:
568 gso_type
= SKB_GSO_TCPV4
;
570 case VIRTIO_NET_HDR_GSO_TCPV6
:
571 gso_type
= SKB_GSO_TCPV6
;
573 case VIRTIO_NET_HDR_GSO_UDP
:
574 gso_type
= SKB_GSO_UDP
;
580 if (vnet_hdr
->gso_type
& VIRTIO_NET_HDR_GSO_ECN
)
581 gso_type
|= SKB_GSO_TCP_ECN
;
583 if (vnet_hdr
->gso_size
== 0)
587 if (vnet_hdr
->flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) {
588 if (!skb_partial_csum_set(skb
, vnet_hdr
->csum_start
,
589 vnet_hdr
->csum_offset
))
593 if (vnet_hdr
->gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
594 skb_shinfo(skb
)->gso_size
= vnet_hdr
->gso_size
;
595 skb_shinfo(skb
)->gso_type
= gso_type
;
597 /* Header must be checked, and gso_segs computed. */
598 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
599 skb_shinfo(skb
)->gso_segs
= 0;
604 static void macvtap_skb_to_vnet_hdr(const struct sk_buff
*skb
,
605 struct virtio_net_hdr
*vnet_hdr
)
607 memset(vnet_hdr
, 0, sizeof(*vnet_hdr
));
609 if (skb_is_gso(skb
)) {
610 struct skb_shared_info
*sinfo
= skb_shinfo(skb
);
612 /* This is a hint as to how much should be linear. */
613 vnet_hdr
->hdr_len
= skb_headlen(skb
);
614 vnet_hdr
->gso_size
= sinfo
->gso_size
;
615 if (sinfo
->gso_type
& SKB_GSO_TCPV4
)
616 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_TCPV4
;
617 else if (sinfo
->gso_type
& SKB_GSO_TCPV6
)
618 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_TCPV6
;
619 else if (sinfo
->gso_type
& SKB_GSO_UDP
)
620 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_UDP
;
623 if (sinfo
->gso_type
& SKB_GSO_TCP_ECN
)
624 vnet_hdr
->gso_type
|= VIRTIO_NET_HDR_GSO_ECN
;
626 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
628 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
629 vnet_hdr
->flags
= VIRTIO_NET_HDR_F_NEEDS_CSUM
;
630 vnet_hdr
->csum_start
= skb_checksum_start_offset(skb
);
631 vnet_hdr
->csum_offset
= skb
->csum_offset
;
632 } else if (skb
->ip_summed
== CHECKSUM_UNNECESSARY
) {
633 vnet_hdr
->flags
= VIRTIO_NET_HDR_F_DATA_VALID
;
634 } /* else everything is zero */
637 /* Get packet from user space buffer */
638 static ssize_t
macvtap_get_user(struct macvtap_queue
*q
, struct msghdr
*m
,
639 const struct iovec
*iv
, unsigned long total_len
,
640 size_t count
, int noblock
)
642 int good_linear
= SKB_MAX_HEAD(NET_IP_ALIGN
);
644 struct macvlan_dev
*vlan
;
645 unsigned long len
= total_len
;
647 struct virtio_net_hdr vnet_hdr
= { 0 };
648 int vnet_hdr_len
= 0;
650 bool zerocopy
= false;
653 if (q
->flags
& IFF_VNET_HDR
) {
654 vnet_hdr_len
= q
->vnet_hdr_sz
;
657 if (len
< vnet_hdr_len
)
661 err
= memcpy_fromiovecend((void *)&vnet_hdr
, iv
, 0,
665 if ((vnet_hdr
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) &&
666 vnet_hdr
.csum_start
+ vnet_hdr
.csum_offset
+ 2 >
668 vnet_hdr
.hdr_len
= vnet_hdr
.csum_start
+
669 vnet_hdr
.csum_offset
+ 2;
671 if (vnet_hdr
.hdr_len
> len
)
676 if (unlikely(len
< ETH_HLEN
))
680 if (unlikely(count
> UIO_MAXIOV
))
683 if (m
&& m
->msg_control
&& sock_flag(&q
->sk
, SOCK_ZEROCOPY
)) {
684 copylen
= vnet_hdr
.hdr_len
? vnet_hdr
.hdr_len
: GOODCOPY_LEN
;
685 if (copylen
> good_linear
)
686 copylen
= good_linear
;
688 if (iov_pages(iv
, vnet_hdr_len
+ copylen
, count
)
695 if (vnet_hdr
.hdr_len
> good_linear
)
696 linear
= good_linear
;
698 linear
= vnet_hdr
.hdr_len
;
701 skb
= macvtap_alloc_skb(&q
->sk
, NET_IP_ALIGN
, copylen
,
702 linear
, noblock
, &err
);
707 err
= zerocopy_sg_from_iovec(skb
, iv
, vnet_hdr_len
, count
);
709 err
= skb_copy_datagram_from_iovec(skb
, 0, iv
, vnet_hdr_len
,
711 if (!err
&& m
&& m
->msg_control
) {
712 struct ubuf_info
*uarg
= m
->msg_control
;
713 uarg
->callback(uarg
, false);
720 skb_set_network_header(skb
, ETH_HLEN
);
721 skb_reset_mac_header(skb
);
722 skb
->protocol
= eth_hdr(skb
)->h_proto
;
725 err
= macvtap_skb_from_vnet_hdr(skb
, &vnet_hdr
);
730 skb_probe_transport_header(skb
, ETH_HLEN
);
733 vlan
= rcu_dereference(q
->vlan
);
734 /* copy skb_ubuf_info for callback when skb has no error */
736 skb_shinfo(skb
)->destructor_arg
= m
->msg_control
;
737 skb_shinfo(skb
)->tx_flags
|= SKBTX_DEV_ZEROCOPY
;
738 skb_shinfo(skb
)->tx_flags
|= SKBTX_SHARED_FRAG
;
741 skb
->dev
= vlan
->dev
;
755 vlan
= rcu_dereference(q
->vlan
);
757 this_cpu_inc(vlan
->pcpu_stats
->tx_dropped
);
763 static ssize_t
macvtap_aio_write(struct kiocb
*iocb
, const struct iovec
*iv
,
764 unsigned long count
, loff_t pos
)
766 struct file
*file
= iocb
->ki_filp
;
767 ssize_t result
= -ENOLINK
;
768 struct macvtap_queue
*q
= file
->private_data
;
770 result
= macvtap_get_user(q
, NULL
, iv
, iov_length(iv
, count
), count
,
771 file
->f_flags
& O_NONBLOCK
);
775 /* Put packet to the user space buffer */
776 static ssize_t
macvtap_put_user(struct macvtap_queue
*q
,
777 const struct sk_buff
*skb
,
778 const struct iovec
*iv
, int len
)
781 int vnet_hdr_len
= 0;
785 if (q
->flags
& IFF_VNET_HDR
) {
786 struct virtio_net_hdr vnet_hdr
;
787 vnet_hdr_len
= q
->vnet_hdr_sz
;
788 if ((len
-= vnet_hdr_len
) < 0)
791 macvtap_skb_to_vnet_hdr(skb
, &vnet_hdr
);
793 if (memcpy_toiovecend(iv
, (void *)&vnet_hdr
, 0, sizeof(vnet_hdr
)))
796 total
= copied
= vnet_hdr_len
;
799 if (!vlan_tx_tag_present(skb
))
800 len
= min_t(int, skb
->len
, len
);
807 veth
.h_vlan_proto
= skb
->vlan_proto
;
808 veth
.h_vlan_TCI
= htons(vlan_tx_tag_get(skb
));
810 vlan_offset
= offsetof(struct vlan_ethhdr
, h_vlan_proto
);
811 len
= min_t(int, skb
->len
+ VLAN_HLEN
, len
);
814 copy
= min_t(int, vlan_offset
, len
);
815 ret
= skb_copy_datagram_const_iovec(skb
, 0, iv
, copied
, copy
);
821 copy
= min_t(int, sizeof(veth
), len
);
822 ret
= memcpy_toiovecend(iv
, (void *)&veth
, copied
, copy
);
829 ret
= skb_copy_datagram_const_iovec(skb
, vlan_offset
, iv
, copied
, len
);
832 return ret
? ret
: total
;
835 static ssize_t
macvtap_do_read(struct macvtap_queue
*q
,
836 const struct iovec
*iv
, unsigned long len
,
845 prepare_to_wait(sk_sleep(&q
->sk
), &wait
,
848 /* Read frames from the queue */
849 skb
= skb_dequeue(&q
->sk
.sk_receive_queue
);
855 if (signal_pending(current
)) {
859 /* Nothing to read, let's sleep */
863 ret
= macvtap_put_user(q
, skb
, iv
, len
);
869 finish_wait(sk_sleep(&q
->sk
), &wait
);
873 static ssize_t
macvtap_aio_read(struct kiocb
*iocb
, const struct iovec
*iv
,
874 unsigned long count
, loff_t pos
)
876 struct file
*file
= iocb
->ki_filp
;
877 struct macvtap_queue
*q
= file
->private_data
;
878 ssize_t len
, ret
= 0;
880 len
= iov_length(iv
, count
);
886 ret
= macvtap_do_read(q
, iv
, len
, file
->f_flags
& O_NONBLOCK
);
887 ret
= min_t(ssize_t
, ret
, len
);
894 static struct macvlan_dev
*macvtap_get_vlan(struct macvtap_queue
*q
)
896 struct macvlan_dev
*vlan
;
899 vlan
= rtnl_dereference(q
->vlan
);
906 static void macvtap_put_vlan(struct macvlan_dev
*vlan
)
911 static int macvtap_ioctl_set_queue(struct file
*file
, unsigned int flags
)
913 struct macvtap_queue
*q
= file
->private_data
;
914 struct macvlan_dev
*vlan
;
917 vlan
= macvtap_get_vlan(q
);
921 if (flags
& IFF_ATTACH_QUEUE
)
922 ret
= macvtap_enable_queue(vlan
->dev
, file
, q
);
923 else if (flags
& IFF_DETACH_QUEUE
)
924 ret
= macvtap_disable_queue(q
);
928 macvtap_put_vlan(vlan
);
932 static int set_offload(struct macvtap_queue
*q
, unsigned long arg
)
934 struct macvlan_dev
*vlan
;
935 netdev_features_t features
;
936 netdev_features_t feature_mask
= 0;
938 vlan
= rtnl_dereference(q
->vlan
);
942 features
= vlan
->dev
->features
;
944 if (arg
& TUN_F_CSUM
) {
945 feature_mask
= NETIF_F_HW_CSUM
;
947 if (arg
& (TUN_F_TSO4
| TUN_F_TSO6
)) {
948 if (arg
& TUN_F_TSO_ECN
)
949 feature_mask
|= NETIF_F_TSO_ECN
;
950 if (arg
& TUN_F_TSO4
)
951 feature_mask
|= NETIF_F_TSO
;
952 if (arg
& TUN_F_TSO6
)
953 feature_mask
|= NETIF_F_TSO6
;
957 feature_mask
|= NETIF_F_UFO
;
960 /* tun/tap driver inverts the usage for TSO offloads, where
961 * setting the TSO bit means that the userspace wants to
962 * accept TSO frames and turning it off means that user space
963 * does not support TSO.
964 * For macvtap, we have to invert it to mean the same thing.
965 * When user space turns off TSO, we turn off GSO/LRO so that
966 * user-space will not receive TSO frames.
968 if (feature_mask
& (NETIF_F_TSO
| NETIF_F_TSO6
| NETIF_F_UFO
))
969 features
|= RX_OFFLOADS
;
971 features
&= ~RX_OFFLOADS
;
973 /* tap_features are the same as features on tun/tap and
974 * reflect user expectations.
976 vlan
->tap_features
= feature_mask
;
977 vlan
->set_features
= features
;
978 netdev_update_features(vlan
->dev
);
984 * provide compatibility with generic tun/tap interface
986 static long macvtap_ioctl(struct file
*file
, unsigned int cmd
,
989 struct macvtap_queue
*q
= file
->private_data
;
990 struct macvlan_dev
*vlan
;
991 void __user
*argp
= (void __user
*)arg
;
992 struct ifreq __user
*ifr
= argp
;
993 unsigned int __user
*up
= argp
;
995 int __user
*sp
= argp
;
1001 /* ignore the name, just look at flags */
1002 if (get_user(u
, &ifr
->ifr_flags
))
1006 if ((u
& ~(IFF_VNET_HDR
| IFF_MULTI_QUEUE
)) !=
1007 (IFF_NO_PI
| IFF_TAP
))
1016 vlan
= macvtap_get_vlan(q
);
1023 if (copy_to_user(&ifr
->ifr_name
, vlan
->dev
->name
, IFNAMSIZ
) ||
1024 put_user(q
->flags
, &ifr
->ifr_flags
))
1026 macvtap_put_vlan(vlan
);
1031 if (get_user(u
, &ifr
->ifr_flags
))
1034 ret
= macvtap_ioctl_set_queue(file
, u
);
1038 case TUNGETFEATURES
:
1039 if (put_user(IFF_TAP
| IFF_NO_PI
| IFF_VNET_HDR
|
1040 IFF_MULTI_QUEUE
, up
))
1045 if (get_user(u
, up
))
1048 q
->sk
.sk_sndbuf
= u
;
1051 case TUNGETVNETHDRSZ
:
1053 if (put_user(s
, sp
))
1057 case TUNSETVNETHDRSZ
:
1058 if (get_user(s
, sp
))
1060 if (s
< (int)sizeof(struct virtio_net_hdr
))
1067 /* let the user check for future flags */
1068 if (arg
& ~(TUN_F_CSUM
| TUN_F_TSO4
| TUN_F_TSO6
|
1069 TUN_F_TSO_ECN
| TUN_F_UFO
))
1073 ret
= set_offload(q
, arg
);
1082 #ifdef CONFIG_COMPAT
1083 static long macvtap_compat_ioctl(struct file
*file
, unsigned int cmd
,
1086 return macvtap_ioctl(file
, cmd
, (unsigned long)compat_ptr(arg
));
1090 static const struct file_operations macvtap_fops
= {
1091 .owner
= THIS_MODULE
,
1092 .open
= macvtap_open
,
1093 .release
= macvtap_release
,
1094 .aio_read
= macvtap_aio_read
,
1095 .aio_write
= macvtap_aio_write
,
1096 .poll
= macvtap_poll
,
1097 .llseek
= no_llseek
,
1098 .unlocked_ioctl
= macvtap_ioctl
,
1099 #ifdef CONFIG_COMPAT
1100 .compat_ioctl
= macvtap_compat_ioctl
,
1104 static int macvtap_sendmsg(struct kiocb
*iocb
, struct socket
*sock
,
1105 struct msghdr
*m
, size_t total_len
)
1107 struct macvtap_queue
*q
= container_of(sock
, struct macvtap_queue
, sock
);
1108 return macvtap_get_user(q
, m
, m
->msg_iov
, total_len
, m
->msg_iovlen
,
1109 m
->msg_flags
& MSG_DONTWAIT
);
1112 static int macvtap_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
1113 struct msghdr
*m
, size_t total_len
,
1116 struct macvtap_queue
*q
= container_of(sock
, struct macvtap_queue
, sock
);
1118 if (flags
& ~(MSG_DONTWAIT
|MSG_TRUNC
))
1120 ret
= macvtap_do_read(q
, m
->msg_iov
, total_len
,
1121 flags
& MSG_DONTWAIT
);
1122 if (ret
> total_len
) {
1123 m
->msg_flags
|= MSG_TRUNC
;
1124 ret
= flags
& MSG_TRUNC
? ret
: total_len
;
1129 /* Ops structure to mimic raw sockets with tun */
1130 static const struct proto_ops macvtap_socket_ops
= {
1131 .sendmsg
= macvtap_sendmsg
,
1132 .recvmsg
= macvtap_recvmsg
,
1135 /* Get an underlying socket object from tun file. Returns error unless file is
1136 * attached to a device. The returned object works like a packet socket, it
1137 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
1138 * holding a reference to the file for as long as the socket is in use. */
1139 struct socket
*macvtap_get_socket(struct file
*file
)
1141 struct macvtap_queue
*q
;
1142 if (file
->f_op
!= &macvtap_fops
)
1143 return ERR_PTR(-EINVAL
);
1144 q
= file
->private_data
;
1146 return ERR_PTR(-EBADFD
);
1149 EXPORT_SYMBOL_GPL(macvtap_get_socket
);
1151 static int macvtap_device_event(struct notifier_block
*unused
,
1152 unsigned long event
, void *ptr
)
1154 struct net_device
*dev
= netdev_notifier_info_to_dev(ptr
);
1155 struct macvlan_dev
*vlan
;
1156 struct device
*classdev
;
1160 if (dev
->rtnl_link_ops
!= &macvtap_link_ops
)
1163 vlan
= netdev_priv(dev
);
1166 case NETDEV_REGISTER
:
1167 /* Create the device node here after the network device has
1168 * been registered but before register_netdevice has
1171 err
= macvtap_get_minor(vlan
);
1173 return notifier_from_errno(err
);
1175 devt
= MKDEV(MAJOR(macvtap_major
), vlan
->minor
);
1176 classdev
= device_create(macvtap_class
, &dev
->dev
, devt
,
1177 dev
, "tap%d", dev
->ifindex
);
1178 if (IS_ERR(classdev
)) {
1179 macvtap_free_minor(vlan
);
1180 return notifier_from_errno(PTR_ERR(classdev
));
1183 case NETDEV_UNREGISTER
:
1184 devt
= MKDEV(MAJOR(macvtap_major
), vlan
->minor
);
1185 device_destroy(macvtap_class
, devt
);
1186 macvtap_free_minor(vlan
);
1193 static struct notifier_block macvtap_notifier_block __read_mostly
= {
1194 .notifier_call
= macvtap_device_event
,
1197 static int macvtap_init(void)
1201 err
= alloc_chrdev_region(&macvtap_major
, 0,
1202 MACVTAP_NUM_DEVS
, "macvtap");
1206 cdev_init(&macvtap_cdev
, &macvtap_fops
);
1207 err
= cdev_add(&macvtap_cdev
, macvtap_major
, MACVTAP_NUM_DEVS
);
1211 macvtap_class
= class_create(THIS_MODULE
, "macvtap");
1212 if (IS_ERR(macvtap_class
)) {
1213 err
= PTR_ERR(macvtap_class
);
1217 err
= register_netdevice_notifier(&macvtap_notifier_block
);
1221 err
= macvlan_link_register(&macvtap_link_ops
);
1228 unregister_netdevice_notifier(&macvtap_notifier_block
);
1230 class_unregister(macvtap_class
);
1232 cdev_del(&macvtap_cdev
);
1234 unregister_chrdev_region(macvtap_major
, MACVTAP_NUM_DEVS
);
1238 module_init(macvtap_init
);
1240 static void macvtap_exit(void)
1242 rtnl_link_unregister(&macvtap_link_ops
);
1243 unregister_netdevice_notifier(&macvtap_notifier_block
);
1244 class_unregister(macvtap_class
);
1245 cdev_del(&macvtap_cdev
);
1246 unregister_chrdev_region(macvtap_major
, MACVTAP_NUM_DEVS
);
1248 module_exit(macvtap_exit
);
1250 MODULE_ALIAS_RTNL_LINK("macvtap");
1251 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1252 MODULE_LICENSE("GPL");