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>
21 #include <net/net_namespace.h>
22 #include <net/rtnetlink.h>
24 #include <linux/virtio_net.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
;
49 static struct proto macvtap_proto
= {
52 .obj_size
= sizeof (struct macvtap_queue
),
56 * Variables for dealing with macvtaps device numbers.
58 static dev_t macvtap_major
;
59 #define MACVTAP_NUM_DEVS (1U << MINORBITS)
60 static DEFINE_MUTEX(minor_lock
);
61 static DEFINE_IDR(minor_idr
);
63 #define GOODCOPY_LEN 128
64 static struct class *macvtap_class
;
65 static struct cdev macvtap_cdev
;
67 static const struct proto_ops macvtap_socket_ops
;
69 #define TUN_OFFLOADS (NETIF_F_HW_CSUM | NETIF_F_TSO_ECN | NETIF_F_TSO | \
71 #define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO)
72 #define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG)
74 static struct macvlan_dev
*macvtap_get_vlan_rcu(const struct net_device
*dev
)
76 return rcu_dereference(dev
->rx_handler_data
);
81 * The macvtap_queue and the macvlan_dev are loosely coupled, the
82 * pointers from one to the other can only be read while rcu_read_lock
85 * Both the file and the macvlan_dev hold a reference on the macvtap_queue
86 * through sock_hold(&q->sk). When the macvlan_dev goes away first,
87 * q->vlan becomes inaccessible. When the files gets closed,
88 * macvtap_get_queue() fails.
90 * There may still be references to the struct sock inside of the
91 * queue from outbound SKBs, but these never reference back to the
92 * file or the dev. The data structure is freed through __sk_free
93 * when both our references and any pending SKBs are gone.
96 static int macvtap_enable_queue(struct net_device
*dev
, struct file
*file
,
97 struct macvtap_queue
*q
)
99 struct macvlan_dev
*vlan
= netdev_priv(dev
);
108 rcu_assign_pointer(vlan
->taps
[vlan
->numvtaps
], q
);
109 q
->queue_index
= vlan
->numvtaps
;
118 static int macvtap_set_queue(struct net_device
*dev
, struct file
*file
,
119 struct macvtap_queue
*q
)
121 struct macvlan_dev
*vlan
= netdev_priv(dev
);
123 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
);
142 static int macvtap_disable_queue(struct macvtap_queue
*q
)
144 struct macvlan_dev
*vlan
;
145 struct macvtap_queue
*nq
;
151 vlan
= rtnl_dereference(q
->vlan
);
154 int index
= q
->queue_index
;
155 BUG_ON(index
>= vlan
->numvtaps
);
156 nq
= rtnl_dereference(vlan
->taps
[vlan
->numvtaps
- 1]);
157 nq
->queue_index
= index
;
159 rcu_assign_pointer(vlan
->taps
[index
], nq
);
160 RCU_INIT_POINTER(vlan
->taps
[vlan
->numvtaps
- 1], NULL
);
170 * The file owning the queue got closed, give up both
171 * the reference that the files holds as well as the
172 * one from the macvlan_dev if that still exists.
174 * Using the spinlock makes sure that we don't get
175 * to the queue again after destroying it.
177 static void macvtap_put_queue(struct macvtap_queue
*q
)
179 struct macvlan_dev
*vlan
;
182 vlan
= rtnl_dereference(q
->vlan
);
186 BUG_ON(macvtap_disable_queue(q
));
189 RCU_INIT_POINTER(q
->vlan
, NULL
);
191 list_del_init(&q
->next
);
201 * Select a queue based on the rxq of the device on which this packet
202 * arrived. If the incoming device is not mq, calculate a flow hash
203 * to select a queue. If all fails, find the first available queue.
204 * Cache vlan->numvtaps since it can become zero during the execution
207 static struct macvtap_queue
*macvtap_get_queue(struct net_device
*dev
,
210 struct macvlan_dev
*vlan
= netdev_priv(dev
);
211 struct macvtap_queue
*tap
= NULL
;
212 /* Access to taps array is protected by rcu, but access to numvtaps
213 * isn't. Below we use it to lookup a queue, but treat it as a hint
214 * and validate that the result isn't NULL - in case we are
215 * racing against queue removal.
217 int numvtaps
= ACCESS_ONCE(vlan
->numvtaps
);
223 /* Check if we can use flow to select a queue */
224 rxq
= skb_get_hash(skb
);
226 tap
= rcu_dereference(vlan
->taps
[rxq
% numvtaps
]);
230 if (likely(skb_rx_queue_recorded(skb
))) {
231 rxq
= skb_get_rx_queue(skb
);
233 while (unlikely(rxq
>= numvtaps
))
236 tap
= rcu_dereference(vlan
->taps
[rxq
]);
240 tap
= rcu_dereference(vlan
->taps
[0]);
246 * The net_device is going away, give up the reference
247 * that it holds on all queues and safely set the pointer
248 * from the queues to NULL.
250 static void macvtap_del_queues(struct net_device
*dev
)
252 struct macvlan_dev
*vlan
= netdev_priv(dev
);
253 struct macvtap_queue
*q
, *tmp
, *qlist
[MAX_MACVTAP_QUEUES
];
257 list_for_each_entry_safe(q
, tmp
, &vlan
->queue_list
, next
) {
258 list_del_init(&q
->next
);
260 RCU_INIT_POINTER(q
->vlan
, NULL
);
265 for (i
= 0; i
< vlan
->numvtaps
; i
++)
266 RCU_INIT_POINTER(vlan
->taps
[i
], NULL
);
267 BUG_ON(vlan
->numvtaps
);
268 BUG_ON(vlan
->numqueues
);
269 /* guarantee that any future macvtap_set_queue will fail */
270 vlan
->numvtaps
= MAX_MACVTAP_QUEUES
;
272 for (--j
; j
>= 0; j
--)
273 sock_put(&qlist
[j
]->sk
);
276 static rx_handler_result_t
macvtap_handle_frame(struct sk_buff
**pskb
)
278 struct sk_buff
*skb
= *pskb
;
279 struct net_device
*dev
= skb
->dev
;
280 struct macvlan_dev
*vlan
;
281 struct macvtap_queue
*q
;
282 netdev_features_t features
= TAP_FEATURES
;
284 vlan
= macvtap_get_vlan_rcu(dev
);
286 return RX_HANDLER_PASS
;
288 q
= macvtap_get_queue(dev
, skb
);
290 return RX_HANDLER_PASS
;
292 if (skb_queue_len(&q
->sk
.sk_receive_queue
) >= dev
->tx_queue_len
)
295 skb_push(skb
, ETH_HLEN
);
297 /* Apply the forward feature mask so that we perform segmentation
298 * according to users wishes. This only works if VNET_HDR is
301 if (q
->flags
& IFF_VNET_HDR
)
302 features
|= vlan
->tap_features
;
303 if (netif_needs_gso(dev
, skb
, features
)) {
304 struct sk_buff
*segs
= __skb_gso_segment(skb
, features
, false);
310 skb_queue_tail(&q
->sk
.sk_receive_queue
, skb
);
316 struct sk_buff
*nskb
= segs
->next
;
319 skb_queue_tail(&q
->sk
.sk_receive_queue
, segs
);
323 /* If we receive a partial checksum and the tap side
324 * doesn't support checksum offload, compute the checksum.
325 * Note: it doesn't matter which checksum feature to
326 * check, we either support them all or none.
328 if (skb
->ip_summed
== CHECKSUM_PARTIAL
&&
329 !(features
& NETIF_F_ALL_CSUM
) &&
330 skb_checksum_help(skb
))
332 skb_queue_tail(&q
->sk
.sk_receive_queue
, skb
);
336 wake_up_interruptible_poll(sk_sleep(&q
->sk
), POLLIN
| POLLRDNORM
| POLLRDBAND
);
337 return RX_HANDLER_CONSUMED
;
340 /* Count errors/drops only here, thus don't care about args. */
341 macvlan_count_rx(vlan
, 0, 0, 0);
343 return RX_HANDLER_CONSUMED
;
346 static int macvtap_get_minor(struct macvlan_dev
*vlan
)
348 int retval
= -ENOMEM
;
350 mutex_lock(&minor_lock
);
351 retval
= idr_alloc(&minor_idr
, vlan
, 1, MACVTAP_NUM_DEVS
, GFP_KERNEL
);
353 vlan
->minor
= retval
;
354 } else if (retval
== -ENOSPC
) {
355 printk(KERN_ERR
"too many macvtap devices\n");
358 mutex_unlock(&minor_lock
);
359 return retval
< 0 ? retval
: 0;
362 static void macvtap_free_minor(struct macvlan_dev
*vlan
)
364 mutex_lock(&minor_lock
);
366 idr_remove(&minor_idr
, vlan
->minor
);
369 mutex_unlock(&minor_lock
);
372 static struct net_device
*dev_get_by_macvtap_minor(int minor
)
374 struct net_device
*dev
= NULL
;
375 struct macvlan_dev
*vlan
;
377 mutex_lock(&minor_lock
);
378 vlan
= idr_find(&minor_idr
, minor
);
383 mutex_unlock(&minor_lock
);
387 static int macvtap_newlink(struct net
*src_net
,
388 struct net_device
*dev
,
390 struct nlattr
*data
[])
392 struct macvlan_dev
*vlan
= netdev_priv(dev
);
395 INIT_LIST_HEAD(&vlan
->queue_list
);
397 /* Since macvlan supports all offloads by default, make
398 * tap support all offloads also.
400 vlan
->tap_features
= TUN_OFFLOADS
;
402 err
= netdev_rx_handler_register(dev
, macvtap_handle_frame
, vlan
);
406 /* Don't put anything that may fail after macvlan_common_newlink
407 * because we can't undo what it does.
409 return macvlan_common_newlink(src_net
, dev
, tb
, data
);
412 static void macvtap_dellink(struct net_device
*dev
,
413 struct list_head
*head
)
415 netdev_rx_handler_unregister(dev
);
416 macvtap_del_queues(dev
);
417 macvlan_dellink(dev
, head
);
420 static void macvtap_setup(struct net_device
*dev
)
422 macvlan_common_setup(dev
);
423 dev
->tx_queue_len
= TUN_READQ_SIZE
;
426 static struct rtnl_link_ops macvtap_link_ops __read_mostly
= {
428 .setup
= macvtap_setup
,
429 .newlink
= macvtap_newlink
,
430 .dellink
= macvtap_dellink
,
434 static void macvtap_sock_write_space(struct sock
*sk
)
436 wait_queue_head_t
*wqueue
;
438 if (!sock_writeable(sk
) ||
439 !test_and_clear_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
))
442 wqueue
= sk_sleep(sk
);
443 if (wqueue
&& waitqueue_active(wqueue
))
444 wake_up_interruptible_poll(wqueue
, POLLOUT
| POLLWRNORM
| POLLWRBAND
);
447 static void macvtap_sock_destruct(struct sock
*sk
)
449 skb_queue_purge(&sk
->sk_receive_queue
);
452 static int macvtap_open(struct inode
*inode
, struct file
*file
)
454 struct net
*net
= current
->nsproxy
->net_ns
;
455 struct net_device
*dev
;
456 struct macvtap_queue
*q
;
460 dev
= dev_get_by_macvtap_minor(iminor(inode
));
465 q
= (struct macvtap_queue
*)sk_alloc(net
, AF_UNSPEC
, GFP_KERNEL
,
470 RCU_INIT_POINTER(q
->sock
.wq
, &q
->wq
);
471 init_waitqueue_head(&q
->wq
.wait
);
472 q
->sock
.type
= SOCK_RAW
;
473 q
->sock
.state
= SS_CONNECTED
;
475 q
->sock
.ops
= &macvtap_socket_ops
;
476 sock_init_data(&q
->sock
, &q
->sk
);
477 q
->sk
.sk_write_space
= macvtap_sock_write_space
;
478 q
->sk
.sk_destruct
= macvtap_sock_destruct
;
479 q
->flags
= IFF_VNET_HDR
| IFF_NO_PI
| IFF_TAP
;
480 q
->vnet_hdr_sz
= sizeof(struct virtio_net_hdr
);
483 * so far only KVM virtio_net uses macvtap, enable zero copy between
484 * guest kernel and host kernel when lower device supports zerocopy
486 * The macvlan supports zerocopy iff the lower device supports zero
487 * copy so we don't have to look at the lower device directly.
489 if ((dev
->features
& NETIF_F_HIGHDMA
) && (dev
->features
& NETIF_F_SG
))
490 sock_set_flag(&q
->sk
, SOCK_ZEROCOPY
);
492 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 pr_warn_once("macvtap: %s: using disabled UFO feature; please fix this program\n",
576 gso_type
= SKB_GSO_UDP
;
577 if (skb
->protocol
== htons(ETH_P_IPV6
))
578 ipv6_proxy_select_ident(skb
);
584 if (vnet_hdr
->gso_type
& VIRTIO_NET_HDR_GSO_ECN
)
585 gso_type
|= SKB_GSO_TCP_ECN
;
587 if (vnet_hdr
->gso_size
== 0)
591 if (vnet_hdr
->flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) {
592 if (!skb_partial_csum_set(skb
, vnet_hdr
->csum_start
,
593 vnet_hdr
->csum_offset
))
597 if (vnet_hdr
->gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
598 skb_shinfo(skb
)->gso_size
= vnet_hdr
->gso_size
;
599 skb_shinfo(skb
)->gso_type
= gso_type
;
601 /* Header must be checked, and gso_segs computed. */
602 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
603 skb_shinfo(skb
)->gso_segs
= 0;
608 static void macvtap_skb_to_vnet_hdr(const struct sk_buff
*skb
,
609 struct virtio_net_hdr
*vnet_hdr
)
611 memset(vnet_hdr
, 0, sizeof(*vnet_hdr
));
613 if (skb_is_gso(skb
)) {
614 struct skb_shared_info
*sinfo
= skb_shinfo(skb
);
616 /* This is a hint as to how much should be linear. */
617 vnet_hdr
->hdr_len
= skb_headlen(skb
);
618 vnet_hdr
->gso_size
= sinfo
->gso_size
;
619 if (sinfo
->gso_type
& SKB_GSO_TCPV4
)
620 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_TCPV4
;
621 else if (sinfo
->gso_type
& SKB_GSO_TCPV6
)
622 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_TCPV6
;
625 if (sinfo
->gso_type
& SKB_GSO_TCP_ECN
)
626 vnet_hdr
->gso_type
|= VIRTIO_NET_HDR_GSO_ECN
;
628 vnet_hdr
->gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
630 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
631 vnet_hdr
->flags
= VIRTIO_NET_HDR_F_NEEDS_CSUM
;
632 vnet_hdr
->csum_start
= skb_checksum_start_offset(skb
);
633 if (vlan_tx_tag_present(skb
))
634 vnet_hdr
->csum_start
+= VLAN_HLEN
;
635 vnet_hdr
->csum_offset
= skb
->csum_offset
;
636 } else if (skb
->ip_summed
== CHECKSUM_UNNECESSARY
) {
637 vnet_hdr
->flags
= VIRTIO_NET_HDR_F_DATA_VALID
;
638 } /* else everything is zero */
641 /* Get packet from user space buffer */
642 static ssize_t
macvtap_get_user(struct macvtap_queue
*q
, struct msghdr
*m
,
643 struct iov_iter
*from
, int noblock
)
645 int good_linear
= SKB_MAX_HEAD(NET_IP_ALIGN
);
647 struct macvlan_dev
*vlan
;
648 unsigned long total_len
= iov_iter_count(from
);
649 unsigned long len
= total_len
;
651 struct virtio_net_hdr vnet_hdr
= { 0 };
652 int vnet_hdr_len
= 0;
654 bool zerocopy
= false;
658 if (q
->flags
& IFF_VNET_HDR
) {
659 vnet_hdr_len
= q
->vnet_hdr_sz
;
662 if (len
< vnet_hdr_len
)
667 n
= copy_from_iter(&vnet_hdr
, sizeof(vnet_hdr
), from
);
668 if (n
!= sizeof(vnet_hdr
))
670 iov_iter_advance(from
, vnet_hdr_len
- sizeof(vnet_hdr
));
671 if ((vnet_hdr
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) &&
672 vnet_hdr
.csum_start
+ vnet_hdr
.csum_offset
+ 2 >
674 vnet_hdr
.hdr_len
= vnet_hdr
.csum_start
+
675 vnet_hdr
.csum_offset
+ 2;
677 if (vnet_hdr
.hdr_len
> len
)
682 if (unlikely(len
< ETH_HLEN
))
685 if (m
&& m
->msg_control
&& sock_flag(&q
->sk
, SOCK_ZEROCOPY
)) {
688 copylen
= vnet_hdr
.hdr_len
? vnet_hdr
.hdr_len
: GOODCOPY_LEN
;
689 if (copylen
> good_linear
)
690 copylen
= good_linear
;
693 iov_iter_advance(&i
, copylen
);
694 if (iov_iter_npages(&i
, INT_MAX
) <= MAX_SKB_FRAGS
)
700 if (vnet_hdr
.hdr_len
> good_linear
)
701 linear
= good_linear
;
703 linear
= vnet_hdr
.hdr_len
;
706 skb
= macvtap_alloc_skb(&q
->sk
, NET_IP_ALIGN
, copylen
,
707 linear
, noblock
, &err
);
712 err
= zerocopy_sg_from_iter(skb
, from
);
714 err
= skb_copy_datagram_from_iter(skb
, 0, from
, len
);
715 if (!err
&& m
&& m
->msg_control
) {
716 struct ubuf_info
*uarg
= m
->msg_control
;
717 uarg
->callback(uarg
, false);
724 skb_set_network_header(skb
, ETH_HLEN
);
725 skb_reset_mac_header(skb
);
726 skb
->protocol
= eth_hdr(skb
)->h_proto
;
729 err
= macvtap_skb_from_vnet_hdr(skb
, &vnet_hdr
);
734 skb_probe_transport_header(skb
, ETH_HLEN
);
737 vlan
= rcu_dereference(q
->vlan
);
738 /* copy skb_ubuf_info for callback when skb has no error */
740 skb_shinfo(skb
)->destructor_arg
= m
->msg_control
;
741 skb_shinfo(skb
)->tx_flags
|= SKBTX_DEV_ZEROCOPY
;
742 skb_shinfo(skb
)->tx_flags
|= SKBTX_SHARED_FRAG
;
745 skb
->dev
= vlan
->dev
;
759 vlan
= rcu_dereference(q
->vlan
);
761 this_cpu_inc(vlan
->pcpu_stats
->tx_dropped
);
767 static ssize_t
macvtap_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
769 struct file
*file
= iocb
->ki_filp
;
770 struct macvtap_queue
*q
= file
->private_data
;
772 return macvtap_get_user(q
, NULL
, from
, 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 struct iov_iter
*iter
)
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 (iov_iter_count(iter
) < vnet_hdr_len
)
791 macvtap_skb_to_vnet_hdr(skb
, &vnet_hdr
);
793 if (copy_to_iter(&vnet_hdr
, sizeof(vnet_hdr
), iter
) !=
797 iov_iter_advance(iter
, vnet_hdr_len
- sizeof(vnet_hdr
));
799 total
= vnet_hdr_len
;
802 if (vlan_tx_tag_present(skb
)) {
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
);
813 ret
= skb_copy_datagram_iter(skb
, 0, iter
, vlan_offset
);
814 if (ret
|| !iov_iter_count(iter
))
817 ret
= copy_to_iter(&veth
, sizeof(veth
), iter
);
818 if (ret
!= sizeof(veth
) || !iov_iter_count(iter
))
822 ret
= skb_copy_datagram_iter(skb
, vlan_offset
, iter
,
823 skb
->len
- vlan_offset
);
826 return ret
? ret
: total
;
829 static ssize_t
macvtap_do_read(struct macvtap_queue
*q
,
837 if (!iov_iter_count(to
))
842 prepare_to_wait(sk_sleep(&q
->sk
), &wait
,
845 /* Read frames from the queue */
846 skb
= skb_dequeue(&q
->sk
.sk_receive_queue
);
853 if (signal_pending(current
)) {
857 /* Nothing to read, let's sleep */
861 ret
= macvtap_put_user(q
, skb
, to
);
865 finish_wait(sk_sleep(&q
->sk
), &wait
);
869 static ssize_t
macvtap_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
871 struct file
*file
= iocb
->ki_filp
;
872 struct macvtap_queue
*q
= file
->private_data
;
873 ssize_t len
= iov_iter_count(to
), ret
;
875 ret
= macvtap_do_read(q
, to
, file
->f_flags
& O_NONBLOCK
);
876 ret
= min_t(ssize_t
, ret
, len
);
882 static struct macvlan_dev
*macvtap_get_vlan(struct macvtap_queue
*q
)
884 struct macvlan_dev
*vlan
;
887 vlan
= rtnl_dereference(q
->vlan
);
894 static void macvtap_put_vlan(struct macvlan_dev
*vlan
)
899 static int macvtap_ioctl_set_queue(struct file
*file
, unsigned int flags
)
901 struct macvtap_queue
*q
= file
->private_data
;
902 struct macvlan_dev
*vlan
;
905 vlan
= macvtap_get_vlan(q
);
909 if (flags
& IFF_ATTACH_QUEUE
)
910 ret
= macvtap_enable_queue(vlan
->dev
, file
, q
);
911 else if (flags
& IFF_DETACH_QUEUE
)
912 ret
= macvtap_disable_queue(q
);
916 macvtap_put_vlan(vlan
);
920 static int set_offload(struct macvtap_queue
*q
, unsigned long arg
)
922 struct macvlan_dev
*vlan
;
923 netdev_features_t features
;
924 netdev_features_t feature_mask
= 0;
926 vlan
= rtnl_dereference(q
->vlan
);
930 features
= vlan
->dev
->features
;
932 if (arg
& TUN_F_CSUM
) {
933 feature_mask
= NETIF_F_HW_CSUM
;
935 if (arg
& (TUN_F_TSO4
| TUN_F_TSO6
)) {
936 if (arg
& TUN_F_TSO_ECN
)
937 feature_mask
|= NETIF_F_TSO_ECN
;
938 if (arg
& TUN_F_TSO4
)
939 feature_mask
|= NETIF_F_TSO
;
940 if (arg
& TUN_F_TSO6
)
941 feature_mask
|= NETIF_F_TSO6
;
945 /* tun/tap driver inverts the usage for TSO offloads, where
946 * setting the TSO bit means that the userspace wants to
947 * accept TSO frames and turning it off means that user space
948 * does not support TSO.
949 * For macvtap, we have to invert it to mean the same thing.
950 * When user space turns off TSO, we turn off GSO/LRO so that
951 * user-space will not receive TSO frames.
953 if (feature_mask
& (NETIF_F_TSO
| NETIF_F_TSO6
))
954 features
|= RX_OFFLOADS
;
956 features
&= ~RX_OFFLOADS
;
958 /* tap_features are the same as features on tun/tap and
959 * reflect user expectations.
961 vlan
->tap_features
= feature_mask
;
962 vlan
->set_features
= features
;
963 netdev_update_features(vlan
->dev
);
969 * provide compatibility with generic tun/tap interface
971 static long macvtap_ioctl(struct file
*file
, unsigned int cmd
,
974 struct macvtap_queue
*q
= file
->private_data
;
975 struct macvlan_dev
*vlan
;
976 void __user
*argp
= (void __user
*)arg
;
977 struct ifreq __user
*ifr
= argp
;
978 unsigned int __user
*up
= argp
;
980 int __user
*sp
= argp
;
986 /* ignore the name, just look at flags */
987 if (get_user(u
, &ifr
->ifr_flags
))
991 if ((u
& ~(IFF_VNET_HDR
| IFF_MULTI_QUEUE
)) !=
992 (IFF_NO_PI
| IFF_TAP
))
1001 vlan
= macvtap_get_vlan(q
);
1008 if (copy_to_user(&ifr
->ifr_name
, vlan
->dev
->name
, IFNAMSIZ
) ||
1009 put_user(q
->flags
, &ifr
->ifr_flags
))
1011 macvtap_put_vlan(vlan
);
1016 if (get_user(u
, &ifr
->ifr_flags
))
1019 ret
= macvtap_ioctl_set_queue(file
, u
);
1023 case TUNGETFEATURES
:
1024 if (put_user(IFF_TAP
| IFF_NO_PI
| IFF_VNET_HDR
|
1025 IFF_MULTI_QUEUE
, up
))
1030 if (get_user(u
, up
))
1033 q
->sk
.sk_sndbuf
= u
;
1036 case TUNGETVNETHDRSZ
:
1038 if (put_user(s
, sp
))
1042 case TUNSETVNETHDRSZ
:
1043 if (get_user(s
, sp
))
1045 if (s
< (int)sizeof(struct virtio_net_hdr
))
1052 /* let the user check for future flags */
1053 if (arg
& ~(TUN_F_CSUM
| TUN_F_TSO4
| TUN_F_TSO6
|
1058 ret
= set_offload(q
, arg
);
1067 #ifdef CONFIG_COMPAT
1068 static long macvtap_compat_ioctl(struct file
*file
, unsigned int cmd
,
1071 return macvtap_ioctl(file
, cmd
, (unsigned long)compat_ptr(arg
));
1075 static const struct file_operations macvtap_fops
= {
1076 .owner
= THIS_MODULE
,
1077 .open
= macvtap_open
,
1078 .release
= macvtap_release
,
1079 .read
= new_sync_read
,
1080 .write
= new_sync_write
,
1081 .read_iter
= macvtap_read_iter
,
1082 .write_iter
= macvtap_write_iter
,
1083 .poll
= macvtap_poll
,
1084 .llseek
= no_llseek
,
1085 .unlocked_ioctl
= macvtap_ioctl
,
1086 #ifdef CONFIG_COMPAT
1087 .compat_ioctl
= macvtap_compat_ioctl
,
1091 static int macvtap_sendmsg(struct kiocb
*iocb
, struct socket
*sock
,
1092 struct msghdr
*m
, size_t total_len
)
1094 struct macvtap_queue
*q
= container_of(sock
, struct macvtap_queue
, sock
);
1095 struct iov_iter from
;
1096 iov_iter_init(&from
, WRITE
, m
->msg_iov
, m
->msg_iovlen
, total_len
);
1097 return macvtap_get_user(q
, m
, &from
, m
->msg_flags
& MSG_DONTWAIT
);
1100 static int macvtap_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
1101 struct msghdr
*m
, size_t total_len
,
1104 struct macvtap_queue
*q
= container_of(sock
, struct macvtap_queue
, sock
);
1107 if (flags
& ~(MSG_DONTWAIT
|MSG_TRUNC
))
1109 iov_iter_init(&to
, READ
, m
->msg_iov
, m
->msg_iovlen
, total_len
);
1110 ret
= macvtap_do_read(q
, &to
, flags
& MSG_DONTWAIT
);
1111 if (ret
> total_len
) {
1112 m
->msg_flags
|= MSG_TRUNC
;
1113 ret
= flags
& MSG_TRUNC
? ret
: total_len
;
1118 /* Ops structure to mimic raw sockets with tun */
1119 static const struct proto_ops macvtap_socket_ops
= {
1120 .sendmsg
= macvtap_sendmsg
,
1121 .recvmsg
= macvtap_recvmsg
,
1124 /* Get an underlying socket object from tun file. Returns error unless file is
1125 * attached to a device. The returned object works like a packet socket, it
1126 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
1127 * holding a reference to the file for as long as the socket is in use. */
1128 struct socket
*macvtap_get_socket(struct file
*file
)
1130 struct macvtap_queue
*q
;
1131 if (file
->f_op
!= &macvtap_fops
)
1132 return ERR_PTR(-EINVAL
);
1133 q
= file
->private_data
;
1135 return ERR_PTR(-EBADFD
);
1138 EXPORT_SYMBOL_GPL(macvtap_get_socket
);
1140 static int macvtap_device_event(struct notifier_block
*unused
,
1141 unsigned long event
, void *ptr
)
1143 struct net_device
*dev
= netdev_notifier_info_to_dev(ptr
);
1144 struct macvlan_dev
*vlan
;
1145 struct device
*classdev
;
1149 if (dev
->rtnl_link_ops
!= &macvtap_link_ops
)
1152 vlan
= netdev_priv(dev
);
1155 case NETDEV_REGISTER
:
1156 /* Create the device node here after the network device has
1157 * been registered but before register_netdevice has
1160 err
= macvtap_get_minor(vlan
);
1162 return notifier_from_errno(err
);
1164 devt
= MKDEV(MAJOR(macvtap_major
), vlan
->minor
);
1165 classdev
= device_create(macvtap_class
, &dev
->dev
, devt
,
1166 dev
, "tap%d", dev
->ifindex
);
1167 if (IS_ERR(classdev
)) {
1168 macvtap_free_minor(vlan
);
1169 return notifier_from_errno(PTR_ERR(classdev
));
1172 case NETDEV_UNREGISTER
:
1173 devt
= MKDEV(MAJOR(macvtap_major
), vlan
->minor
);
1174 device_destroy(macvtap_class
, devt
);
1175 macvtap_free_minor(vlan
);
1182 static struct notifier_block macvtap_notifier_block __read_mostly
= {
1183 .notifier_call
= macvtap_device_event
,
1186 static int macvtap_init(void)
1190 err
= alloc_chrdev_region(&macvtap_major
, 0,
1191 MACVTAP_NUM_DEVS
, "macvtap");
1195 cdev_init(&macvtap_cdev
, &macvtap_fops
);
1196 err
= cdev_add(&macvtap_cdev
, macvtap_major
, MACVTAP_NUM_DEVS
);
1200 macvtap_class
= class_create(THIS_MODULE
, "macvtap");
1201 if (IS_ERR(macvtap_class
)) {
1202 err
= PTR_ERR(macvtap_class
);
1206 err
= register_netdevice_notifier(&macvtap_notifier_block
);
1210 err
= macvlan_link_register(&macvtap_link_ops
);
1217 unregister_netdevice_notifier(&macvtap_notifier_block
);
1219 class_unregister(macvtap_class
);
1221 cdev_del(&macvtap_cdev
);
1223 unregister_chrdev_region(macvtap_major
, MACVTAP_NUM_DEVS
);
1227 module_init(macvtap_init
);
1229 static void macvtap_exit(void)
1231 rtnl_link_unregister(&macvtap_link_ops
);
1232 unregister_netdevice_notifier(&macvtap_notifier_block
);
1233 class_unregister(macvtap_class
);
1234 cdev_del(&macvtap_cdev
);
1235 unregister_chrdev_region(macvtap_major
, MACVTAP_NUM_DEVS
);
1237 module_exit(macvtap_exit
);
1239 MODULE_ALIAS_RTNL_LINK("macvtap");
1240 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1241 MODULE_LICENSE("GPL");