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>
18 #include <linux/uio.h>
20 #include <net/net_namespace.h>
21 #include <net/rtnetlink.h>
23 #include <linux/virtio_net.h>
24 #include <linux/skb_array.h>
27 * A macvtap queue is the central object of this driver, it connects
28 * an open character device to a macvlan interface. There can be
29 * multiple queues on one interface, which map back to queues
30 * implemented in hardware on the underlying device.
32 * macvtap_proto is used to allocate queues through the sock allocation
36 struct macvtap_queue
{
41 struct macvlan_dev __rcu
*vlan
;
46 struct list_head next
;
47 struct skb_array skb_array
;
50 #define MACVTAP_FEATURES (IFF_VNET_HDR | IFF_MULTI_QUEUE)
52 #define MACVTAP_VNET_LE 0x80000000
53 #define MACVTAP_VNET_BE 0x40000000
55 #ifdef CONFIG_TUN_VNET_CROSS_LE
56 static inline bool macvtap_legacy_is_little_endian(struct macvtap_queue
*q
)
58 return q
->flags
& MACVTAP_VNET_BE
? false :
59 virtio_legacy_is_little_endian();
62 static long macvtap_get_vnet_be(struct macvtap_queue
*q
, int __user
*sp
)
64 int s
= !!(q
->flags
& MACVTAP_VNET_BE
);
72 static long macvtap_set_vnet_be(struct macvtap_queue
*q
, int __user
*sp
)
80 q
->flags
|= MACVTAP_VNET_BE
;
82 q
->flags
&= ~MACVTAP_VNET_BE
;
87 static inline bool macvtap_legacy_is_little_endian(struct macvtap_queue
*q
)
89 return virtio_legacy_is_little_endian();
92 static long macvtap_get_vnet_be(struct macvtap_queue
*q
, int __user
*argp
)
97 static long macvtap_set_vnet_be(struct macvtap_queue
*q
, int __user
*argp
)
101 #endif /* CONFIG_TUN_VNET_CROSS_LE */
103 static inline bool macvtap_is_little_endian(struct macvtap_queue
*q
)
105 return q
->flags
& MACVTAP_VNET_LE
||
106 macvtap_legacy_is_little_endian(q
);
109 static inline u16
macvtap16_to_cpu(struct macvtap_queue
*q
, __virtio16 val
)
111 return __virtio16_to_cpu(macvtap_is_little_endian(q
), val
);
114 static inline __virtio16
cpu_to_macvtap16(struct macvtap_queue
*q
, u16 val
)
116 return __cpu_to_virtio16(macvtap_is_little_endian(q
), val
);
119 static struct proto macvtap_proto
= {
121 .owner
= THIS_MODULE
,
122 .obj_size
= sizeof (struct macvtap_queue
),
126 * Variables for dealing with macvtaps device numbers.
128 static dev_t macvtap_major
;
129 #define MACVTAP_NUM_DEVS (1U << MINORBITS)
130 static DEFINE_MUTEX(minor_lock
);
131 static DEFINE_IDR(minor_idr
);
133 #define GOODCOPY_LEN 128
134 static const void *macvtap_net_namespace(struct device
*d
)
136 struct net_device
*dev
= to_net_dev(d
->parent
);
140 static struct class macvtap_class
= {
142 .owner
= THIS_MODULE
,
143 .ns_type
= &net_ns_type_operations
,
144 .namespace = macvtap_net_namespace
,
146 static struct cdev macvtap_cdev
;
148 static const struct proto_ops macvtap_socket_ops
;
150 #define TUN_OFFLOADS (NETIF_F_HW_CSUM | NETIF_F_TSO_ECN | NETIF_F_TSO | \
151 NETIF_F_TSO6 | NETIF_F_UFO)
152 #define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO)
153 #define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG | NETIF_F_FRAGLIST)
155 static struct macvlan_dev
*macvtap_get_vlan_rcu(const struct net_device
*dev
)
157 return rcu_dereference(dev
->rx_handler_data
);
162 * The macvtap_queue and the macvlan_dev are loosely coupled, the
163 * pointers from one to the other can only be read while rcu_read_lock
166 * Both the file and the macvlan_dev hold a reference on the macvtap_queue
167 * through sock_hold(&q->sk). When the macvlan_dev goes away first,
168 * q->vlan becomes inaccessible. When the files gets closed,
169 * macvtap_get_queue() fails.
171 * There may still be references to the struct sock inside of the
172 * queue from outbound SKBs, but these never reference back to the
173 * file or the dev. The data structure is freed through __sk_free
174 * when both our references and any pending SKBs are gone.
177 static int macvtap_enable_queue(struct net_device
*dev
, struct file
*file
,
178 struct macvtap_queue
*q
)
180 struct macvlan_dev
*vlan
= netdev_priv(dev
);
189 rcu_assign_pointer(vlan
->taps
[vlan
->numvtaps
], q
);
190 q
->queue_index
= vlan
->numvtaps
;
199 static int macvtap_set_queue(struct net_device
*dev
, struct file
*file
,
200 struct macvtap_queue
*q
)
202 struct macvlan_dev
*vlan
= netdev_priv(dev
);
204 if (vlan
->numqueues
== MAX_MACVTAP_QUEUES
)
207 rcu_assign_pointer(q
->vlan
, vlan
);
208 rcu_assign_pointer(vlan
->taps
[vlan
->numvtaps
], q
);
212 q
->queue_index
= vlan
->numvtaps
;
214 file
->private_data
= q
;
215 list_add_tail(&q
->next
, &vlan
->queue_list
);
223 static int macvtap_disable_queue(struct macvtap_queue
*q
)
225 struct macvlan_dev
*vlan
;
226 struct macvtap_queue
*nq
;
232 vlan
= rtnl_dereference(q
->vlan
);
235 int index
= q
->queue_index
;
236 BUG_ON(index
>= vlan
->numvtaps
);
237 nq
= rtnl_dereference(vlan
->taps
[vlan
->numvtaps
- 1]);
238 nq
->queue_index
= index
;
240 rcu_assign_pointer(vlan
->taps
[index
], nq
);
241 RCU_INIT_POINTER(vlan
->taps
[vlan
->numvtaps
- 1], NULL
);
251 * The file owning the queue got closed, give up both
252 * the reference that the files holds as well as the
253 * one from the macvlan_dev if that still exists.
255 * Using the spinlock makes sure that we don't get
256 * to the queue again after destroying it.
258 static void macvtap_put_queue(struct macvtap_queue
*q
)
260 struct macvlan_dev
*vlan
;
263 vlan
= rtnl_dereference(q
->vlan
);
267 BUG_ON(macvtap_disable_queue(q
));
270 RCU_INIT_POINTER(q
->vlan
, NULL
);
272 list_del_init(&q
->next
);
278 skb_array_cleanup(&q
->skb_array
);
283 * Select a queue based on the rxq of the device on which this packet
284 * arrived. If the incoming device is not mq, calculate a flow hash
285 * to select a queue. If all fails, find the first available queue.
286 * Cache vlan->numvtaps since it can become zero during the execution
289 static struct macvtap_queue
*macvtap_get_queue(struct net_device
*dev
,
292 struct macvlan_dev
*vlan
= netdev_priv(dev
);
293 struct macvtap_queue
*tap
= NULL
;
294 /* Access to taps array is protected by rcu, but access to numvtaps
295 * isn't. Below we use it to lookup a queue, but treat it as a hint
296 * and validate that the result isn't NULL - in case we are
297 * racing against queue removal.
299 int numvtaps
= ACCESS_ONCE(vlan
->numvtaps
);
308 /* Check if we can use flow to select a queue */
309 rxq
= skb_get_hash(skb
);
311 tap
= rcu_dereference(vlan
->taps
[rxq
% numvtaps
]);
315 if (likely(skb_rx_queue_recorded(skb
))) {
316 rxq
= skb_get_rx_queue(skb
);
318 while (unlikely(rxq
>= numvtaps
))
321 tap
= rcu_dereference(vlan
->taps
[rxq
]);
326 tap
= rcu_dereference(vlan
->taps
[0]);
332 * The net_device is going away, give up the reference
333 * that it holds on all queues and safely set the pointer
334 * from the queues to NULL.
336 static void macvtap_del_queues(struct net_device
*dev
)
338 struct macvlan_dev
*vlan
= netdev_priv(dev
);
339 struct macvtap_queue
*q
, *tmp
;
342 list_for_each_entry_safe(q
, tmp
, &vlan
->queue_list
, next
) {
343 list_del_init(&q
->next
);
344 RCU_INIT_POINTER(q
->vlan
, NULL
);
350 BUG_ON(vlan
->numvtaps
);
351 BUG_ON(vlan
->numqueues
);
352 /* guarantee that any future macvtap_set_queue will fail */
353 vlan
->numvtaps
= MAX_MACVTAP_QUEUES
;
356 static rx_handler_result_t
macvtap_handle_frame(struct sk_buff
**pskb
)
358 struct sk_buff
*skb
= *pskb
;
359 struct net_device
*dev
= skb
->dev
;
360 struct macvlan_dev
*vlan
;
361 struct macvtap_queue
*q
;
362 netdev_features_t features
= TAP_FEATURES
;
364 vlan
= macvtap_get_vlan_rcu(dev
);
366 return RX_HANDLER_PASS
;
368 q
= macvtap_get_queue(dev
, skb
);
370 return RX_HANDLER_PASS
;
372 if (__skb_array_full(&q
->skb_array
))
375 skb_push(skb
, ETH_HLEN
);
377 /* Apply the forward feature mask so that we perform segmentation
378 * according to users wishes. This only works if VNET_HDR is
381 if (q
->flags
& IFF_VNET_HDR
)
382 features
|= vlan
->tap_features
;
383 if (netif_needs_gso(skb
, features
)) {
384 struct sk_buff
*segs
= __skb_gso_segment(skb
, features
, false);
390 if (skb_array_produce(&q
->skb_array
, skb
))
397 struct sk_buff
*nskb
= segs
->next
;
400 if (skb_array_produce(&q
->skb_array
, segs
)) {
402 kfree_skb_list(nskb
);
408 /* If we receive a partial checksum and the tap side
409 * doesn't support checksum offload, compute the checksum.
410 * Note: it doesn't matter which checksum feature to
411 * check, we either support them all or none.
413 if (skb
->ip_summed
== CHECKSUM_PARTIAL
&&
414 !(features
& NETIF_F_CSUM_MASK
) &&
415 skb_checksum_help(skb
))
417 if (skb_array_produce(&q
->skb_array
, skb
))
422 wake_up_interruptible_poll(sk_sleep(&q
->sk
), POLLIN
| POLLRDNORM
| POLLRDBAND
);
423 return RX_HANDLER_CONSUMED
;
426 /* Count errors/drops only here, thus don't care about args. */
427 macvlan_count_rx(vlan
, 0, 0, 0);
429 return RX_HANDLER_CONSUMED
;
432 static int macvtap_get_minor(struct macvlan_dev
*vlan
)
434 int retval
= -ENOMEM
;
436 mutex_lock(&minor_lock
);
437 retval
= idr_alloc(&minor_idr
, vlan
, 1, MACVTAP_NUM_DEVS
, GFP_KERNEL
);
439 vlan
->minor
= retval
;
440 } else if (retval
== -ENOSPC
) {
441 printk(KERN_ERR
"too many macvtap devices\n");
444 mutex_unlock(&minor_lock
);
445 return retval
< 0 ? retval
: 0;
448 static void macvtap_free_minor(struct macvlan_dev
*vlan
)
450 mutex_lock(&minor_lock
);
452 idr_remove(&minor_idr
, vlan
->minor
);
455 mutex_unlock(&minor_lock
);
458 static struct net_device
*dev_get_by_macvtap_minor(int minor
)
460 struct net_device
*dev
= NULL
;
461 struct macvlan_dev
*vlan
;
463 mutex_lock(&minor_lock
);
464 vlan
= idr_find(&minor_idr
, minor
);
469 mutex_unlock(&minor_lock
);
473 static int macvtap_newlink(struct net
*src_net
,
474 struct net_device
*dev
,
476 struct nlattr
*data
[])
478 struct macvlan_dev
*vlan
= netdev_priv(dev
);
481 INIT_LIST_HEAD(&vlan
->queue_list
);
483 /* Since macvlan supports all offloads by default, make
484 * tap support all offloads also.
486 vlan
->tap_features
= TUN_OFFLOADS
;
488 err
= netdev_rx_handler_register(dev
, macvtap_handle_frame
, vlan
);
492 /* Don't put anything that may fail after macvlan_common_newlink
493 * because we can't undo what it does.
495 return macvlan_common_newlink(src_net
, dev
, tb
, data
);
498 static void macvtap_dellink(struct net_device
*dev
,
499 struct list_head
*head
)
501 netdev_rx_handler_unregister(dev
);
502 macvtap_del_queues(dev
);
503 macvlan_dellink(dev
, head
);
506 static void macvtap_setup(struct net_device
*dev
)
508 macvlan_common_setup(dev
);
509 dev
->tx_queue_len
= TUN_READQ_SIZE
;
512 static struct rtnl_link_ops macvtap_link_ops __read_mostly
= {
514 .setup
= macvtap_setup
,
515 .newlink
= macvtap_newlink
,
516 .dellink
= macvtap_dellink
,
520 static void macvtap_sock_write_space(struct sock
*sk
)
522 wait_queue_head_t
*wqueue
;
524 if (!sock_writeable(sk
) ||
525 !test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
))
528 wqueue
= sk_sleep(sk
);
529 if (wqueue
&& waitqueue_active(wqueue
))
530 wake_up_interruptible_poll(wqueue
, POLLOUT
| POLLWRNORM
| POLLWRBAND
);
533 static void macvtap_sock_destruct(struct sock
*sk
)
535 struct macvtap_queue
*q
= container_of(sk
, struct macvtap_queue
, sk
);
538 while ((skb
= skb_array_consume(&q
->skb_array
)) != NULL
)
542 static int macvtap_open(struct inode
*inode
, struct file
*file
)
544 struct net
*net
= current
->nsproxy
->net_ns
;
545 struct net_device
*dev
;
546 struct macvtap_queue
*q
;
550 dev
= dev_get_by_macvtap_minor(iminor(inode
));
555 q
= (struct macvtap_queue
*)sk_alloc(net
, AF_UNSPEC
, GFP_KERNEL
,
560 RCU_INIT_POINTER(q
->sock
.wq
, &q
->wq
);
561 init_waitqueue_head(&q
->wq
.wait
);
562 q
->sock
.type
= SOCK_RAW
;
563 q
->sock
.state
= SS_CONNECTED
;
565 q
->sock
.ops
= &macvtap_socket_ops
;
566 sock_init_data(&q
->sock
, &q
->sk
);
567 q
->sk
.sk_write_space
= macvtap_sock_write_space
;
568 q
->sk
.sk_destruct
= macvtap_sock_destruct
;
569 q
->flags
= IFF_VNET_HDR
| IFF_NO_PI
| IFF_TAP
;
570 q
->vnet_hdr_sz
= sizeof(struct virtio_net_hdr
);
573 * so far only KVM virtio_net uses macvtap, enable zero copy between
574 * guest kernel and host kernel when lower device supports zerocopy
576 * The macvlan supports zerocopy iff the lower device supports zero
577 * copy so we don't have to look at the lower device directly.
579 if ((dev
->features
& NETIF_F_HIGHDMA
) && (dev
->features
& NETIF_F_SG
))
580 sock_set_flag(&q
->sk
, SOCK_ZEROCOPY
);
583 if (skb_array_init(&q
->skb_array
, dev
->tx_queue_len
, GFP_KERNEL
))
586 err
= macvtap_set_queue(dev
, file
, q
);
596 skb_array_cleanup(&q
->skb_array
);
607 static int macvtap_release(struct inode
*inode
, struct file
*file
)
609 struct macvtap_queue
*q
= file
->private_data
;
610 macvtap_put_queue(q
);
614 static unsigned int macvtap_poll(struct file
*file
, poll_table
* wait
)
616 struct macvtap_queue
*q
= file
->private_data
;
617 unsigned int mask
= POLLERR
;
623 poll_wait(file
, &q
->wq
.wait
, wait
);
625 if (!skb_array_empty(&q
->skb_array
))
626 mask
|= POLLIN
| POLLRDNORM
;
628 if (sock_writeable(&q
->sk
) ||
629 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE
, &q
->sock
.flags
) &&
630 sock_writeable(&q
->sk
)))
631 mask
|= POLLOUT
| POLLWRNORM
;
637 static inline struct sk_buff
*macvtap_alloc_skb(struct sock
*sk
, size_t prepad
,
638 size_t len
, size_t linear
,
639 int noblock
, int *err
)
643 /* Under a page? Don't bother with paged skb. */
644 if (prepad
+ len
< PAGE_SIZE
|| !linear
)
647 skb
= sock_alloc_send_pskb(sk
, prepad
+ linear
, len
- linear
, noblock
,
652 skb_reserve(skb
, prepad
);
653 skb_put(skb
, linear
);
654 skb
->data_len
= len
- linear
;
655 skb
->len
+= len
- linear
;
660 /* Neighbour code has some assumptions on HH_DATA_MOD alignment */
661 #define MACVTAP_RESERVE HH_DATA_OFF(ETH_HLEN)
663 /* Get packet from user space buffer */
664 static ssize_t
macvtap_get_user(struct macvtap_queue
*q
, struct msghdr
*m
,
665 struct iov_iter
*from
, int noblock
)
667 int good_linear
= SKB_MAX_HEAD(MACVTAP_RESERVE
);
669 struct macvlan_dev
*vlan
;
670 unsigned long total_len
= iov_iter_count(from
);
671 unsigned long len
= total_len
;
673 struct virtio_net_hdr vnet_hdr
= { 0 };
674 int vnet_hdr_len
= 0;
677 bool zerocopy
= false;
681 if (q
->flags
& IFF_VNET_HDR
) {
682 vnet_hdr_len
= q
->vnet_hdr_sz
;
685 if (len
< vnet_hdr_len
)
690 n
= copy_from_iter(&vnet_hdr
, sizeof(vnet_hdr
), from
);
691 if (n
!= sizeof(vnet_hdr
))
693 iov_iter_advance(from
, vnet_hdr_len
- sizeof(vnet_hdr
));
694 if ((vnet_hdr
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) &&
695 macvtap16_to_cpu(q
, vnet_hdr
.csum_start
) +
696 macvtap16_to_cpu(q
, vnet_hdr
.csum_offset
) + 2 >
697 macvtap16_to_cpu(q
, vnet_hdr
.hdr_len
))
698 vnet_hdr
.hdr_len
= cpu_to_macvtap16(q
,
699 macvtap16_to_cpu(q
, vnet_hdr
.csum_start
) +
700 macvtap16_to_cpu(q
, vnet_hdr
.csum_offset
) + 2);
702 if (macvtap16_to_cpu(q
, vnet_hdr
.hdr_len
) > len
)
707 if (unlikely(len
< ETH_HLEN
))
710 if (m
&& m
->msg_control
&& sock_flag(&q
->sk
, SOCK_ZEROCOPY
)) {
713 copylen
= vnet_hdr
.hdr_len
?
714 macvtap16_to_cpu(q
, vnet_hdr
.hdr_len
) : GOODCOPY_LEN
;
715 if (copylen
> good_linear
)
716 copylen
= good_linear
;
717 else if (copylen
< ETH_HLEN
)
721 iov_iter_advance(&i
, copylen
);
722 if (iov_iter_npages(&i
, INT_MAX
) <= MAX_SKB_FRAGS
)
728 linear
= macvtap16_to_cpu(q
, vnet_hdr
.hdr_len
);
729 if (linear
> good_linear
)
730 linear
= good_linear
;
731 else if (linear
< ETH_HLEN
)
735 skb
= macvtap_alloc_skb(&q
->sk
, MACVTAP_RESERVE
, copylen
,
736 linear
, noblock
, &err
);
741 err
= zerocopy_sg_from_iter(skb
, from
);
743 err
= skb_copy_datagram_from_iter(skb
, 0, from
, len
);
744 if (!err
&& m
&& m
->msg_control
) {
745 struct ubuf_info
*uarg
= m
->msg_control
;
746 uarg
->callback(uarg
, false);
753 skb_set_network_header(skb
, ETH_HLEN
);
754 skb_reset_mac_header(skb
);
755 skb
->protocol
= eth_hdr(skb
)->h_proto
;
758 err
= virtio_net_hdr_to_skb(skb
, &vnet_hdr
,
759 macvtap_is_little_endian(q
));
764 skb_probe_transport_header(skb
, ETH_HLEN
);
766 /* Move network header to the right position for VLAN tagged packets */
767 if ((skb
->protocol
== htons(ETH_P_8021Q
) ||
768 skb
->protocol
== htons(ETH_P_8021AD
)) &&
769 __vlan_get_protocol(skb
, skb
->protocol
, &depth
) != 0)
770 skb_set_network_header(skb
, depth
);
773 vlan
= rcu_dereference(q
->vlan
);
774 /* copy skb_ubuf_info for callback when skb has no error */
776 skb_shinfo(skb
)->destructor_arg
= m
->msg_control
;
777 skb_shinfo(skb
)->tx_flags
|= SKBTX_DEV_ZEROCOPY
;
778 skb_shinfo(skb
)->tx_flags
|= SKBTX_SHARED_FRAG
;
781 skb
->dev
= vlan
->dev
;
795 vlan
= rcu_dereference(q
->vlan
);
797 this_cpu_inc(vlan
->pcpu_stats
->tx_dropped
);
803 static ssize_t
macvtap_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
805 struct file
*file
= iocb
->ki_filp
;
806 struct macvtap_queue
*q
= file
->private_data
;
808 return macvtap_get_user(q
, NULL
, from
, file
->f_flags
& O_NONBLOCK
);
811 /* Put packet to the user space buffer */
812 static ssize_t
macvtap_put_user(struct macvtap_queue
*q
,
813 const struct sk_buff
*skb
,
814 struct iov_iter
*iter
)
817 int vnet_hdr_len
= 0;
821 if (q
->flags
& IFF_VNET_HDR
) {
822 struct virtio_net_hdr vnet_hdr
;
823 vnet_hdr_len
= q
->vnet_hdr_sz
;
824 if (iov_iter_count(iter
) < vnet_hdr_len
)
827 ret
= virtio_net_hdr_from_skb(skb
, &vnet_hdr
,
828 macvtap_is_little_endian(q
));
832 if (copy_to_iter(&vnet_hdr
, sizeof(vnet_hdr
), iter
) !=
836 iov_iter_advance(iter
, vnet_hdr_len
- sizeof(vnet_hdr
));
838 total
= vnet_hdr_len
;
841 if (skb_vlan_tag_present(skb
)) {
846 veth
.h_vlan_proto
= skb
->vlan_proto
;
847 veth
.h_vlan_TCI
= htons(skb_vlan_tag_get(skb
));
849 vlan_offset
= offsetof(struct vlan_ethhdr
, h_vlan_proto
);
852 ret
= skb_copy_datagram_iter(skb
, 0, iter
, vlan_offset
);
853 if (ret
|| !iov_iter_count(iter
))
856 ret
= copy_to_iter(&veth
, sizeof(veth
), iter
);
857 if (ret
!= sizeof(veth
) || !iov_iter_count(iter
))
861 ret
= skb_copy_datagram_iter(skb
, vlan_offset
, iter
,
862 skb
->len
- vlan_offset
);
865 return ret
? ret
: total
;
868 static ssize_t
macvtap_do_read(struct macvtap_queue
*q
,
876 if (!iov_iter_count(to
))
881 prepare_to_wait(sk_sleep(&q
->sk
), &wait
,
884 /* Read frames from the queue */
885 skb
= skb_array_consume(&q
->skb_array
);
892 if (signal_pending(current
)) {
896 /* Nothing to read, let's sleep */
900 finish_wait(sk_sleep(&q
->sk
), &wait
);
903 ret
= macvtap_put_user(q
, skb
, to
);
904 if (unlikely(ret
< 0))
912 static ssize_t
macvtap_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
914 struct file
*file
= iocb
->ki_filp
;
915 struct macvtap_queue
*q
= file
->private_data
;
916 ssize_t len
= iov_iter_count(to
), ret
;
918 ret
= macvtap_do_read(q
, to
, file
->f_flags
& O_NONBLOCK
);
919 ret
= min_t(ssize_t
, ret
, len
);
925 static struct macvlan_dev
*macvtap_get_vlan(struct macvtap_queue
*q
)
927 struct macvlan_dev
*vlan
;
930 vlan
= rtnl_dereference(q
->vlan
);
937 static void macvtap_put_vlan(struct macvlan_dev
*vlan
)
942 static int macvtap_ioctl_set_queue(struct file
*file
, unsigned int flags
)
944 struct macvtap_queue
*q
= file
->private_data
;
945 struct macvlan_dev
*vlan
;
948 vlan
= macvtap_get_vlan(q
);
952 if (flags
& IFF_ATTACH_QUEUE
)
953 ret
= macvtap_enable_queue(vlan
->dev
, file
, q
);
954 else if (flags
& IFF_DETACH_QUEUE
)
955 ret
= macvtap_disable_queue(q
);
959 macvtap_put_vlan(vlan
);
963 static int set_offload(struct macvtap_queue
*q
, unsigned long arg
)
965 struct macvlan_dev
*vlan
;
966 netdev_features_t features
;
967 netdev_features_t feature_mask
= 0;
969 vlan
= rtnl_dereference(q
->vlan
);
973 features
= vlan
->dev
->features
;
975 if (arg
& TUN_F_CSUM
) {
976 feature_mask
= NETIF_F_HW_CSUM
;
978 if (arg
& (TUN_F_TSO4
| TUN_F_TSO6
)) {
979 if (arg
& TUN_F_TSO_ECN
)
980 feature_mask
|= NETIF_F_TSO_ECN
;
981 if (arg
& TUN_F_TSO4
)
982 feature_mask
|= NETIF_F_TSO
;
983 if (arg
& TUN_F_TSO6
)
984 feature_mask
|= NETIF_F_TSO6
;
988 feature_mask
|= NETIF_F_UFO
;
991 /* tun/tap driver inverts the usage for TSO offloads, where
992 * setting the TSO bit means that the userspace wants to
993 * accept TSO frames and turning it off means that user space
994 * does not support TSO.
995 * For macvtap, we have to invert it to mean the same thing.
996 * When user space turns off TSO, we turn off GSO/LRO so that
997 * user-space will not receive TSO frames.
999 if (feature_mask
& (NETIF_F_TSO
| NETIF_F_TSO6
| NETIF_F_UFO
))
1000 features
|= RX_OFFLOADS
;
1002 features
&= ~RX_OFFLOADS
;
1004 /* tap_features are the same as features on tun/tap and
1005 * reflect user expectations.
1007 vlan
->tap_features
= feature_mask
;
1008 vlan
->set_features
= features
;
1009 netdev_update_features(vlan
->dev
);
1015 * provide compatibility with generic tun/tap interface
1017 static long macvtap_ioctl(struct file
*file
, unsigned int cmd
,
1020 struct macvtap_queue
*q
= file
->private_data
;
1021 struct macvlan_dev
*vlan
;
1022 void __user
*argp
= (void __user
*)arg
;
1023 struct ifreq __user
*ifr
= argp
;
1024 unsigned int __user
*up
= argp
;
1026 int __user
*sp
= argp
;
1033 /* ignore the name, just look at flags */
1034 if (get_user(u
, &ifr
->ifr_flags
))
1038 if ((u
& ~MACVTAP_FEATURES
) != (IFF_NO_PI
| IFF_TAP
))
1041 q
->flags
= (q
->flags
& ~MACVTAP_FEATURES
) | u
;
1047 vlan
= macvtap_get_vlan(q
);
1055 if (copy_to_user(&ifr
->ifr_name
, vlan
->dev
->name
, IFNAMSIZ
) ||
1056 put_user(u
, &ifr
->ifr_flags
))
1058 macvtap_put_vlan(vlan
);
1063 if (get_user(u
, &ifr
->ifr_flags
))
1066 ret
= macvtap_ioctl_set_queue(file
, u
);
1070 case TUNGETFEATURES
:
1071 if (put_user(IFF_TAP
| IFF_NO_PI
| MACVTAP_FEATURES
, up
))
1076 if (get_user(s
, sp
))
1079 q
->sk
.sk_sndbuf
= s
;
1082 case TUNGETVNETHDRSZ
:
1084 if (put_user(s
, sp
))
1088 case TUNSETVNETHDRSZ
:
1089 if (get_user(s
, sp
))
1091 if (s
< (int)sizeof(struct virtio_net_hdr
))
1098 s
= !!(q
->flags
& MACVTAP_VNET_LE
);
1099 if (put_user(s
, sp
))
1104 if (get_user(s
, sp
))
1107 q
->flags
|= MACVTAP_VNET_LE
;
1109 q
->flags
&= ~MACVTAP_VNET_LE
;
1113 return macvtap_get_vnet_be(q
, sp
);
1116 return macvtap_set_vnet_be(q
, sp
);
1119 /* let the user check for future flags */
1120 if (arg
& ~(TUN_F_CSUM
| TUN_F_TSO4
| TUN_F_TSO6
|
1121 TUN_F_TSO_ECN
| TUN_F_UFO
))
1125 ret
= set_offload(q
, arg
);
1131 vlan
= macvtap_get_vlan(q
);
1137 u
= vlan
->dev
->type
;
1138 if (copy_to_user(&ifr
->ifr_name
, vlan
->dev
->name
, IFNAMSIZ
) ||
1139 copy_to_user(&ifr
->ifr_hwaddr
.sa_data
, vlan
->dev
->dev_addr
, ETH_ALEN
) ||
1140 put_user(u
, &ifr
->ifr_hwaddr
.sa_family
))
1142 macvtap_put_vlan(vlan
);
1147 if (copy_from_user(&sa
, &ifr
->ifr_hwaddr
, sizeof(sa
)))
1150 vlan
= macvtap_get_vlan(q
);
1155 ret
= dev_set_mac_address(vlan
->dev
, &sa
);
1156 macvtap_put_vlan(vlan
);
1165 #ifdef CONFIG_COMPAT
1166 static long macvtap_compat_ioctl(struct file
*file
, unsigned int cmd
,
1169 return macvtap_ioctl(file
, cmd
, (unsigned long)compat_ptr(arg
));
1173 static const struct file_operations macvtap_fops
= {
1174 .owner
= THIS_MODULE
,
1175 .open
= macvtap_open
,
1176 .release
= macvtap_release
,
1177 .read_iter
= macvtap_read_iter
,
1178 .write_iter
= macvtap_write_iter
,
1179 .poll
= macvtap_poll
,
1180 .llseek
= no_llseek
,
1181 .unlocked_ioctl
= macvtap_ioctl
,
1182 #ifdef CONFIG_COMPAT
1183 .compat_ioctl
= macvtap_compat_ioctl
,
1187 static int macvtap_sendmsg(struct socket
*sock
, struct msghdr
*m
,
1190 struct macvtap_queue
*q
= container_of(sock
, struct macvtap_queue
, sock
);
1191 return macvtap_get_user(q
, m
, &m
->msg_iter
, m
->msg_flags
& MSG_DONTWAIT
);
1194 static int macvtap_recvmsg(struct socket
*sock
, struct msghdr
*m
,
1195 size_t total_len
, int flags
)
1197 struct macvtap_queue
*q
= container_of(sock
, struct macvtap_queue
, sock
);
1199 if (flags
& ~(MSG_DONTWAIT
|MSG_TRUNC
))
1201 ret
= macvtap_do_read(q
, &m
->msg_iter
, flags
& MSG_DONTWAIT
);
1202 if (ret
> total_len
) {
1203 m
->msg_flags
|= MSG_TRUNC
;
1204 ret
= flags
& MSG_TRUNC
? ret
: total_len
;
1209 static int macvtap_peek_len(struct socket
*sock
)
1211 struct macvtap_queue
*q
= container_of(sock
, struct macvtap_queue
,
1213 return skb_array_peek_len(&q
->skb_array
);
1216 /* Ops structure to mimic raw sockets with tun */
1217 static const struct proto_ops macvtap_socket_ops
= {
1218 .sendmsg
= macvtap_sendmsg
,
1219 .recvmsg
= macvtap_recvmsg
,
1220 .peek_len
= macvtap_peek_len
,
1223 /* Get an underlying socket object from tun file. Returns error unless file is
1224 * attached to a device. The returned object works like a packet socket, it
1225 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
1226 * holding a reference to the file for as long as the socket is in use. */
1227 struct socket
*macvtap_get_socket(struct file
*file
)
1229 struct macvtap_queue
*q
;
1230 if (file
->f_op
!= &macvtap_fops
)
1231 return ERR_PTR(-EINVAL
);
1232 q
= file
->private_data
;
1234 return ERR_PTR(-EBADFD
);
1237 EXPORT_SYMBOL_GPL(macvtap_get_socket
);
1239 static int macvtap_queue_resize(struct macvlan_dev
*vlan
)
1241 struct net_device
*dev
= vlan
->dev
;
1242 struct macvtap_queue
*q
;
1243 struct skb_array
**arrays
;
1244 int n
= vlan
->numqueues
;
1247 arrays
= kmalloc(sizeof *arrays
* n
, GFP_KERNEL
);
1251 list_for_each_entry(q
, &vlan
->queue_list
, next
)
1252 arrays
[i
++] = &q
->skb_array
;
1254 ret
= skb_array_resize_multiple(arrays
, n
,
1255 dev
->tx_queue_len
, GFP_KERNEL
);
1261 static int macvtap_device_event(struct notifier_block
*unused
,
1262 unsigned long event
, void *ptr
)
1264 struct net_device
*dev
= netdev_notifier_info_to_dev(ptr
);
1265 struct macvlan_dev
*vlan
;
1266 struct device
*classdev
;
1269 char tap_name
[IFNAMSIZ
];
1271 if (dev
->rtnl_link_ops
!= &macvtap_link_ops
)
1274 snprintf(tap_name
, IFNAMSIZ
, "tap%d", dev
->ifindex
);
1275 vlan
= netdev_priv(dev
);
1278 case NETDEV_REGISTER
:
1279 /* Create the device node here after the network device has
1280 * been registered but before register_netdevice has
1283 err
= macvtap_get_minor(vlan
);
1285 return notifier_from_errno(err
);
1287 devt
= MKDEV(MAJOR(macvtap_major
), vlan
->minor
);
1288 classdev
= device_create(&macvtap_class
, &dev
->dev
, devt
,
1290 if (IS_ERR(classdev
)) {
1291 macvtap_free_minor(vlan
);
1292 return notifier_from_errno(PTR_ERR(classdev
));
1294 err
= sysfs_create_link(&dev
->dev
.kobj
, &classdev
->kobj
,
1297 return notifier_from_errno(err
);
1299 case NETDEV_UNREGISTER
:
1300 /* vlan->minor == 0 if NETDEV_REGISTER above failed */
1301 if (vlan
->minor
== 0)
1303 sysfs_remove_link(&dev
->dev
.kobj
, tap_name
);
1304 devt
= MKDEV(MAJOR(macvtap_major
), vlan
->minor
);
1305 device_destroy(&macvtap_class
, devt
);
1306 macvtap_free_minor(vlan
);
1308 case NETDEV_CHANGE_TX_QUEUE_LEN
:
1309 if (macvtap_queue_resize(vlan
))
1317 static struct notifier_block macvtap_notifier_block __read_mostly
= {
1318 .notifier_call
= macvtap_device_event
,
1321 static int macvtap_init(void)
1325 err
= alloc_chrdev_region(&macvtap_major
, 0,
1326 MACVTAP_NUM_DEVS
, "macvtap");
1330 cdev_init(&macvtap_cdev
, &macvtap_fops
);
1331 err
= cdev_add(&macvtap_cdev
, macvtap_major
, MACVTAP_NUM_DEVS
);
1335 err
= class_register(&macvtap_class
);
1339 err
= register_netdevice_notifier(&macvtap_notifier_block
);
1343 err
= macvlan_link_register(&macvtap_link_ops
);
1350 unregister_netdevice_notifier(&macvtap_notifier_block
);
1352 class_unregister(&macvtap_class
);
1354 cdev_del(&macvtap_cdev
);
1356 unregister_chrdev_region(macvtap_major
, MACVTAP_NUM_DEVS
);
1360 module_init(macvtap_init
);
1362 static void macvtap_exit(void)
1364 rtnl_link_unregister(&macvtap_link_ops
);
1365 unregister_netdevice_notifier(&macvtap_notifier_block
);
1366 class_unregister(&macvtap_class
);
1367 cdev_del(&macvtap_cdev
);
1368 unregister_chrdev_region(macvtap_major
, MACVTAP_NUM_DEVS
);
1369 idr_destroy(&minor_idr
);
1371 module_exit(macvtap_exit
);
1373 MODULE_ALIAS_RTNL_LINK("macvtap");
1374 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1375 MODULE_LICENSE("GPL");