1 #include <linux/etherdevice.h>
2 #include <linux/if_tap.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/signal.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>
26 #define TAP_IFFEATURES (IFF_VNET_HDR | IFF_MULTI_QUEUE)
28 #define TAP_VNET_LE 0x80000000
29 #define TAP_VNET_BE 0x40000000
31 #ifdef CONFIG_TUN_VNET_CROSS_LE
32 static inline bool tap_legacy_is_little_endian(struct tap_queue
*q
)
34 return q
->flags
& TAP_VNET_BE
? false :
35 virtio_legacy_is_little_endian();
38 static long tap_get_vnet_be(struct tap_queue
*q
, int __user
*sp
)
40 int s
= !!(q
->flags
& TAP_VNET_BE
);
48 static long tap_set_vnet_be(struct tap_queue
*q
, int __user
*sp
)
56 q
->flags
|= TAP_VNET_BE
;
58 q
->flags
&= ~TAP_VNET_BE
;
63 static inline bool tap_legacy_is_little_endian(struct tap_queue
*q
)
65 return virtio_legacy_is_little_endian();
68 static long tap_get_vnet_be(struct tap_queue
*q
, int __user
*argp
)
73 static long tap_set_vnet_be(struct tap_queue
*q
, int __user
*argp
)
77 #endif /* CONFIG_TUN_VNET_CROSS_LE */
79 static inline bool tap_is_little_endian(struct tap_queue
*q
)
81 return q
->flags
& TAP_VNET_LE
||
82 tap_legacy_is_little_endian(q
);
85 static inline u16
tap16_to_cpu(struct tap_queue
*q
, __virtio16 val
)
87 return __virtio16_to_cpu(tap_is_little_endian(q
), val
);
90 static inline __virtio16
cpu_to_tap16(struct tap_queue
*q
, u16 val
)
92 return __cpu_to_virtio16(tap_is_little_endian(q
), val
);
95 static struct proto tap_proto
= {
98 .obj_size
= sizeof(struct tap_queue
),
101 #define TAP_NUM_DEVS (1U << MINORBITS)
103 static LIST_HEAD(major_list
);
108 struct idr minor_idr
;
109 spinlock_t minor_lock
;
110 const char *device_name
;
111 struct list_head next
;
114 #define GOODCOPY_LEN 128
116 static const struct proto_ops tap_socket_ops
;
118 #define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO)
119 #define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG | NETIF_F_FRAGLIST)
121 static struct tap_dev
*tap_dev_get_rcu(const struct net_device
*dev
)
123 return rcu_dereference(dev
->rx_handler_data
);
128 * The tap_queue and the macvlan_dev are loosely coupled, the
129 * pointers from one to the other can only be read while rcu_read_lock
132 * Both the file and the macvlan_dev hold a reference on the tap_queue
133 * through sock_hold(&q->sk). When the macvlan_dev goes away first,
134 * q->vlan becomes inaccessible. When the files gets closed,
135 * tap_get_queue() fails.
137 * There may still be references to the struct sock inside of the
138 * queue from outbound SKBs, but these never reference back to the
139 * file or the dev. The data structure is freed through __sk_free
140 * when both our references and any pending SKBs are gone.
143 static int tap_enable_queue(struct tap_dev
*tap
, struct file
*file
,
154 rcu_assign_pointer(tap
->taps
[tap
->numvtaps
], q
);
155 q
->queue_index
= tap
->numvtaps
;
164 static int tap_set_queue(struct tap_dev
*tap
, struct file
*file
,
167 if (tap
->numqueues
== MAX_TAP_QUEUES
)
170 rcu_assign_pointer(q
->tap
, tap
);
171 rcu_assign_pointer(tap
->taps
[tap
->numvtaps
], q
);
175 q
->queue_index
= tap
->numvtaps
;
177 file
->private_data
= q
;
178 list_add_tail(&q
->next
, &tap
->queue_list
);
186 static int tap_disable_queue(struct tap_queue
*q
)
189 struct tap_queue
*nq
;
195 tap
= rtnl_dereference(q
->tap
);
198 int index
= q
->queue_index
;
199 BUG_ON(index
>= tap
->numvtaps
);
200 nq
= rtnl_dereference(tap
->taps
[tap
->numvtaps
- 1]);
201 nq
->queue_index
= index
;
203 rcu_assign_pointer(tap
->taps
[index
], nq
);
204 RCU_INIT_POINTER(tap
->taps
[tap
->numvtaps
- 1], NULL
);
214 * The file owning the queue got closed, give up both
215 * the reference that the files holds as well as the
216 * one from the macvlan_dev if that still exists.
218 * Using the spinlock makes sure that we don't get
219 * to the queue again after destroying it.
221 static void tap_put_queue(struct tap_queue
*q
)
226 tap
= rtnl_dereference(q
->tap
);
230 BUG_ON(tap_disable_queue(q
));
233 RCU_INIT_POINTER(q
->tap
, NULL
);
235 list_del_init(&q
->next
);
245 * Select a queue based on the rxq of the device on which this packet
246 * arrived. If the incoming device is not mq, calculate a flow hash
247 * to select a queue. If all fails, find the first available queue.
248 * Cache vlan->numvtaps since it can become zero during the execution
251 static struct tap_queue
*tap_get_queue(struct tap_dev
*tap
,
254 struct tap_queue
*queue
= NULL
;
255 /* Access to taps array is protected by rcu, but access to numvtaps
256 * isn't. Below we use it to lookup a queue, but treat it as a hint
257 * and validate that the result isn't NULL - in case we are
258 * racing against queue removal.
260 int numvtaps
= READ_ONCE(tap
->numvtaps
);
269 /* Check if we can use flow to select a queue */
270 rxq
= skb_get_hash(skb
);
272 queue
= rcu_dereference(tap
->taps
[rxq
% numvtaps
]);
276 if (likely(skb_rx_queue_recorded(skb
))) {
277 rxq
= skb_get_rx_queue(skb
);
279 while (unlikely(rxq
>= numvtaps
))
282 queue
= rcu_dereference(tap
->taps
[rxq
]);
287 queue
= rcu_dereference(tap
->taps
[0]);
293 * The net_device is going away, give up the reference
294 * that it holds on all queues and safely set the pointer
295 * from the queues to NULL.
297 void tap_del_queues(struct tap_dev
*tap
)
299 struct tap_queue
*q
, *tmp
;
302 list_for_each_entry_safe(q
, tmp
, &tap
->queue_list
, next
) {
303 list_del_init(&q
->next
);
304 RCU_INIT_POINTER(q
->tap
, NULL
);
310 BUG_ON(tap
->numvtaps
);
311 BUG_ON(tap
->numqueues
);
312 /* guarantee that any future tap_set_queue will fail */
313 tap
->numvtaps
= MAX_TAP_QUEUES
;
315 EXPORT_SYMBOL_GPL(tap_del_queues
);
317 rx_handler_result_t
tap_handle_frame(struct sk_buff
**pskb
)
319 struct sk_buff
*skb
= *pskb
;
320 struct net_device
*dev
= skb
->dev
;
323 netdev_features_t features
= TAP_FEATURES
;
325 tap
= tap_dev_get_rcu(dev
);
327 return RX_HANDLER_PASS
;
329 q
= tap_get_queue(tap
, skb
);
331 return RX_HANDLER_PASS
;
333 if (__skb_array_full(&q
->skb_array
))
336 skb_push(skb
, ETH_HLEN
);
338 /* Apply the forward feature mask so that we perform segmentation
339 * according to users wishes. This only works if VNET_HDR is
342 if (q
->flags
& IFF_VNET_HDR
)
343 features
|= tap
->tap_features
;
344 if (netif_needs_gso(skb
, features
)) {
345 struct sk_buff
*segs
= __skb_gso_segment(skb
, features
, false);
351 if (skb_array_produce(&q
->skb_array
, skb
))
358 struct sk_buff
*nskb
= segs
->next
;
361 if (skb_array_produce(&q
->skb_array
, segs
)) {
363 kfree_skb_list(nskb
);
369 /* If we receive a partial checksum and the tap side
370 * doesn't support checksum offload, compute the checksum.
371 * Note: it doesn't matter which checksum feature to
372 * check, we either support them all or none.
374 if (skb
->ip_summed
== CHECKSUM_PARTIAL
&&
375 !(features
& NETIF_F_CSUM_MASK
) &&
376 skb_checksum_help(skb
))
378 if (skb_array_produce(&q
->skb_array
, skb
))
383 wake_up_interruptible_poll(sk_sleep(&q
->sk
), POLLIN
| POLLRDNORM
| POLLRDBAND
);
384 return RX_HANDLER_CONSUMED
;
387 /* Count errors/drops only here, thus don't care about args. */
388 if (tap
->count_rx_dropped
)
389 tap
->count_rx_dropped(tap
);
391 return RX_HANDLER_CONSUMED
;
393 EXPORT_SYMBOL_GPL(tap_handle_frame
);
395 static struct major_info
*tap_get_major(int major
)
397 struct major_info
*tap_major
;
399 list_for_each_entry_rcu(tap_major
, &major_list
, next
) {
400 if (tap_major
->major
== major
)
407 int tap_get_minor(dev_t major
, struct tap_dev
*tap
)
409 int retval
= -ENOMEM
;
410 struct major_info
*tap_major
;
413 tap_major
= tap_get_major(MAJOR(major
));
419 spin_lock(&tap_major
->minor_lock
);
420 retval
= idr_alloc(&tap_major
->minor_idr
, tap
, 1, TAP_NUM_DEVS
, GFP_ATOMIC
);
423 } else if (retval
== -ENOSPC
) {
424 netdev_err(tap
->dev
, "Too many tap devices\n");
427 spin_unlock(&tap_major
->minor_lock
);
431 return retval
< 0 ? retval
: 0;
433 EXPORT_SYMBOL_GPL(tap_get_minor
);
435 void tap_free_minor(dev_t major
, struct tap_dev
*tap
)
437 struct major_info
*tap_major
;
440 tap_major
= tap_get_major(MAJOR(major
));
445 spin_lock(&tap_major
->minor_lock
);
447 idr_remove(&tap_major
->minor_idr
, tap
->minor
);
450 spin_unlock(&tap_major
->minor_lock
);
455 EXPORT_SYMBOL_GPL(tap_free_minor
);
457 static struct tap_dev
*dev_get_by_tap_file(int major
, int minor
)
459 struct net_device
*dev
= NULL
;
461 struct major_info
*tap_major
;
464 tap_major
= tap_get_major(major
);
470 spin_lock(&tap_major
->minor_lock
);
471 tap
= idr_find(&tap_major
->minor_idr
, minor
);
476 spin_unlock(&tap_major
->minor_lock
);
483 static void tap_sock_write_space(struct sock
*sk
)
485 wait_queue_head_t
*wqueue
;
487 if (!sock_writeable(sk
) ||
488 !test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
))
491 wqueue
= sk_sleep(sk
);
492 if (wqueue
&& waitqueue_active(wqueue
))
493 wake_up_interruptible_poll(wqueue
, POLLOUT
| POLLWRNORM
| POLLWRBAND
);
496 static void tap_sock_destruct(struct sock
*sk
)
498 struct tap_queue
*q
= container_of(sk
, struct tap_queue
, sk
);
500 skb_array_cleanup(&q
->skb_array
);
503 static int tap_open(struct inode
*inode
, struct file
*file
)
505 struct net
*net
= current
->nsproxy
->net_ns
;
511 tap
= dev_get_by_tap_file(imajor(inode
), iminor(inode
));
516 q
= (struct tap_queue
*)sk_alloc(net
, AF_UNSPEC
, GFP_KERNEL
,
521 RCU_INIT_POINTER(q
->sock
.wq
, &q
->wq
);
522 init_waitqueue_head(&q
->wq
.wait
);
523 q
->sock
.type
= SOCK_RAW
;
524 q
->sock
.state
= SS_CONNECTED
;
526 q
->sock
.ops
= &tap_socket_ops
;
527 sock_init_data(&q
->sock
, &q
->sk
);
528 q
->sk
.sk_write_space
= tap_sock_write_space
;
529 q
->sk
.sk_destruct
= tap_sock_destruct
;
530 q
->flags
= IFF_VNET_HDR
| IFF_NO_PI
| IFF_TAP
;
531 q
->vnet_hdr_sz
= sizeof(struct virtio_net_hdr
);
534 * so far only KVM virtio_net uses tap, enable zero copy between
535 * guest kernel and host kernel when lower device supports zerocopy
537 * The macvlan supports zerocopy iff the lower device supports zero
538 * copy so we don't have to look at the lower device directly.
540 if ((tap
->dev
->features
& NETIF_F_HIGHDMA
) && (tap
->dev
->features
& NETIF_F_SG
))
541 sock_set_flag(&q
->sk
, SOCK_ZEROCOPY
);
544 if (skb_array_init(&q
->skb_array
, tap
->dev
->tx_queue_len
, GFP_KERNEL
))
547 err
= tap_set_queue(tap
, file
, q
);
557 skb_array_cleanup(&q
->skb_array
);
568 static int tap_release(struct inode
*inode
, struct file
*file
)
570 struct tap_queue
*q
= file
->private_data
;
575 static unsigned int tap_poll(struct file
*file
, poll_table
*wait
)
577 struct tap_queue
*q
= file
->private_data
;
578 unsigned int mask
= POLLERR
;
584 poll_wait(file
, &q
->wq
.wait
, wait
);
586 if (!skb_array_empty(&q
->skb_array
))
587 mask
|= POLLIN
| POLLRDNORM
;
589 if (sock_writeable(&q
->sk
) ||
590 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE
, &q
->sock
.flags
) &&
591 sock_writeable(&q
->sk
)))
592 mask
|= POLLOUT
| POLLWRNORM
;
598 static inline struct sk_buff
*tap_alloc_skb(struct sock
*sk
, size_t prepad
,
599 size_t len
, size_t linear
,
600 int noblock
, int *err
)
604 /* Under a page? Don't bother with paged skb. */
605 if (prepad
+ len
< PAGE_SIZE
|| !linear
)
608 skb
= sock_alloc_send_pskb(sk
, prepad
+ linear
, len
- linear
, noblock
,
613 skb_reserve(skb
, prepad
);
614 skb_put(skb
, linear
);
615 skb
->data_len
= len
- linear
;
616 skb
->len
+= len
- linear
;
621 /* Neighbour code has some assumptions on HH_DATA_MOD alignment */
622 #define TAP_RESERVE HH_DATA_OFF(ETH_HLEN)
624 /* Get packet from user space buffer */
625 static ssize_t
tap_get_user(struct tap_queue
*q
, struct msghdr
*m
,
626 struct iov_iter
*from
, int noblock
)
628 int good_linear
= SKB_MAX_HEAD(TAP_RESERVE
);
631 unsigned long total_len
= iov_iter_count(from
);
632 unsigned long len
= total_len
;
634 struct virtio_net_hdr vnet_hdr
= { 0 };
635 int vnet_hdr_len
= 0;
638 bool zerocopy
= false;
641 if (q
->flags
& IFF_VNET_HDR
) {
642 vnet_hdr_len
= READ_ONCE(q
->vnet_hdr_sz
);
645 if (len
< vnet_hdr_len
)
650 if (!copy_from_iter_full(&vnet_hdr
, sizeof(vnet_hdr
), from
))
652 iov_iter_advance(from
, vnet_hdr_len
- sizeof(vnet_hdr
));
653 if ((vnet_hdr
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) &&
654 tap16_to_cpu(q
, vnet_hdr
.csum_start
) +
655 tap16_to_cpu(q
, vnet_hdr
.csum_offset
) + 2 >
656 tap16_to_cpu(q
, vnet_hdr
.hdr_len
))
657 vnet_hdr
.hdr_len
= cpu_to_tap16(q
,
658 tap16_to_cpu(q
, vnet_hdr
.csum_start
) +
659 tap16_to_cpu(q
, vnet_hdr
.csum_offset
) + 2);
661 if (tap16_to_cpu(q
, vnet_hdr
.hdr_len
) > len
)
666 if (unlikely(len
< ETH_HLEN
))
669 if (m
&& m
->msg_control
&& sock_flag(&q
->sk
, SOCK_ZEROCOPY
)) {
672 copylen
= vnet_hdr
.hdr_len
?
673 tap16_to_cpu(q
, vnet_hdr
.hdr_len
) : GOODCOPY_LEN
;
674 if (copylen
> good_linear
)
675 copylen
= good_linear
;
676 else if (copylen
< ETH_HLEN
)
680 iov_iter_advance(&i
, copylen
);
681 if (iov_iter_npages(&i
, INT_MAX
) <= MAX_SKB_FRAGS
)
687 linear
= tap16_to_cpu(q
, vnet_hdr
.hdr_len
);
688 if (linear
> good_linear
)
689 linear
= good_linear
;
690 else if (linear
< ETH_HLEN
)
694 skb
= tap_alloc_skb(&q
->sk
, TAP_RESERVE
, copylen
,
695 linear
, noblock
, &err
);
700 err
= zerocopy_sg_from_iter(skb
, from
);
702 err
= skb_copy_datagram_from_iter(skb
, 0, from
, len
);
707 skb_set_network_header(skb
, ETH_HLEN
);
708 skb_reset_mac_header(skb
);
709 skb
->protocol
= eth_hdr(skb
)->h_proto
;
712 err
= virtio_net_hdr_to_skb(skb
, &vnet_hdr
,
713 tap_is_little_endian(q
));
718 skb_probe_transport_header(skb
, ETH_HLEN
);
720 /* Move network header to the right position for VLAN tagged packets */
721 if ((skb
->protocol
== htons(ETH_P_8021Q
) ||
722 skb
->protocol
== htons(ETH_P_8021AD
)) &&
723 __vlan_get_protocol(skb
, skb
->protocol
, &depth
) != 0)
724 skb_set_network_header(skb
, depth
);
727 tap
= rcu_dereference(q
->tap
);
728 /* copy skb_ubuf_info for callback when skb has no error */
730 skb_shinfo(skb
)->destructor_arg
= m
->msg_control
;
731 skb_shinfo(skb
)->tx_flags
|= SKBTX_DEV_ZEROCOPY
;
732 skb_shinfo(skb
)->tx_flags
|= SKBTX_SHARED_FRAG
;
733 } else if (m
&& m
->msg_control
) {
734 struct ubuf_info
*uarg
= m
->msg_control
;
735 uarg
->callback(uarg
, false);
753 tap
= rcu_dereference(q
->tap
);
754 if (tap
&& tap
->count_tx_dropped
)
755 tap
->count_tx_dropped(tap
);
761 static ssize_t
tap_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
763 struct file
*file
= iocb
->ki_filp
;
764 struct tap_queue
*q
= file
->private_data
;
766 return tap_get_user(q
, NULL
, from
, file
->f_flags
& O_NONBLOCK
);
769 /* Put packet to the user space buffer */
770 static ssize_t
tap_put_user(struct tap_queue
*q
,
771 const struct sk_buff
*skb
,
772 struct iov_iter
*iter
)
775 int vnet_hdr_len
= 0;
779 if (q
->flags
& IFF_VNET_HDR
) {
780 struct virtio_net_hdr vnet_hdr
;
781 vnet_hdr_len
= READ_ONCE(q
->vnet_hdr_sz
);
782 if (iov_iter_count(iter
) < vnet_hdr_len
)
785 if (virtio_net_hdr_from_skb(skb
, &vnet_hdr
,
786 tap_is_little_endian(q
), true))
789 if (copy_to_iter(&vnet_hdr
, sizeof(vnet_hdr
), iter
) !=
793 iov_iter_advance(iter
, vnet_hdr_len
- sizeof(vnet_hdr
));
795 total
= vnet_hdr_len
;
798 if (skb_vlan_tag_present(skb
)) {
803 veth
.h_vlan_proto
= skb
->vlan_proto
;
804 veth
.h_vlan_TCI
= htons(skb_vlan_tag_get(skb
));
806 vlan_offset
= offsetof(struct vlan_ethhdr
, h_vlan_proto
);
809 ret
= skb_copy_datagram_iter(skb
, 0, iter
, vlan_offset
);
810 if (ret
|| !iov_iter_count(iter
))
813 ret
= copy_to_iter(&veth
, sizeof(veth
), iter
);
814 if (ret
!= sizeof(veth
) || !iov_iter_count(iter
))
818 ret
= skb_copy_datagram_iter(skb
, vlan_offset
, iter
,
819 skb
->len
- vlan_offset
);
822 return ret
? ret
: total
;
825 static ssize_t
tap_do_read(struct tap_queue
*q
,
827 int noblock
, struct sk_buff
*skb
)
832 if (!iov_iter_count(to
))
840 prepare_to_wait(sk_sleep(&q
->sk
), &wait
,
843 /* Read frames from the queue */
844 skb
= skb_array_consume(&q
->skb_array
);
851 if (signal_pending(current
)) {
855 /* Nothing to read, let's sleep */
859 finish_wait(sk_sleep(&q
->sk
), &wait
);
863 ret
= tap_put_user(q
, skb
, to
);
864 if (unlikely(ret
< 0))
872 static ssize_t
tap_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
874 struct file
*file
= iocb
->ki_filp
;
875 struct tap_queue
*q
= file
->private_data
;
876 ssize_t len
= iov_iter_count(to
), ret
;
878 ret
= tap_do_read(q
, to
, file
->f_flags
& O_NONBLOCK
, NULL
);
879 ret
= min_t(ssize_t
, ret
, len
);
885 static struct tap_dev
*tap_get_tap_dev(struct tap_queue
*q
)
890 tap
= rtnl_dereference(q
->tap
);
897 static void tap_put_tap_dev(struct tap_dev
*tap
)
902 static int tap_ioctl_set_queue(struct file
*file
, unsigned int flags
)
904 struct tap_queue
*q
= file
->private_data
;
908 tap
= tap_get_tap_dev(q
);
912 if (flags
& IFF_ATTACH_QUEUE
)
913 ret
= tap_enable_queue(tap
, file
, q
);
914 else if (flags
& IFF_DETACH_QUEUE
)
915 ret
= tap_disable_queue(q
);
919 tap_put_tap_dev(tap
);
923 static int set_offload(struct tap_queue
*q
, unsigned long arg
)
926 netdev_features_t features
;
927 netdev_features_t feature_mask
= 0;
929 tap
= rtnl_dereference(q
->tap
);
933 features
= tap
->dev
->features
;
935 if (arg
& TUN_F_CSUM
) {
936 feature_mask
= NETIF_F_HW_CSUM
;
938 if (arg
& (TUN_F_TSO4
| TUN_F_TSO6
)) {
939 if (arg
& TUN_F_TSO_ECN
)
940 feature_mask
|= NETIF_F_TSO_ECN
;
941 if (arg
& TUN_F_TSO4
)
942 feature_mask
|= NETIF_F_TSO
;
943 if (arg
& TUN_F_TSO6
)
944 feature_mask
|= NETIF_F_TSO6
;
948 /* tun/tap driver inverts the usage for TSO offloads, where
949 * setting the TSO bit means that the userspace wants to
950 * accept TSO frames and turning it off means that user space
951 * does not support TSO.
952 * For tap, we have to invert it to mean the same thing.
953 * When user space turns off TSO, we turn off GSO/LRO so that
954 * user-space will not receive TSO frames.
956 if (feature_mask
& (NETIF_F_TSO
| NETIF_F_TSO6
))
957 features
|= RX_OFFLOADS
;
959 features
&= ~RX_OFFLOADS
;
961 /* tap_features are the same as features on tun/tap and
962 * reflect user expectations.
964 tap
->tap_features
= feature_mask
;
965 if (tap
->update_features
)
966 tap
->update_features(tap
, features
);
972 * provide compatibility with generic tun/tap interface
974 static long tap_ioctl(struct file
*file
, unsigned int cmd
,
977 struct tap_queue
*q
= file
->private_data
;
979 void __user
*argp
= (void __user
*)arg
;
980 struct ifreq __user
*ifr
= argp
;
981 unsigned int __user
*up
= argp
;
983 int __user
*sp
= argp
;
990 /* ignore the name, just look at flags */
991 if (get_user(u
, &ifr
->ifr_flags
))
995 if ((u
& ~TAP_IFFEATURES
) != (IFF_NO_PI
| IFF_TAP
))
998 q
->flags
= (q
->flags
& ~TAP_IFFEATURES
) | u
;
1004 tap
= tap_get_tap_dev(q
);
1012 if (copy_to_user(&ifr
->ifr_name
, tap
->dev
->name
, IFNAMSIZ
) ||
1013 put_user(u
, &ifr
->ifr_flags
))
1015 tap_put_tap_dev(tap
);
1020 if (get_user(u
, &ifr
->ifr_flags
))
1023 ret
= tap_ioctl_set_queue(file
, u
);
1027 case TUNGETFEATURES
:
1028 if (put_user(IFF_TAP
| IFF_NO_PI
| TAP_IFFEATURES
, up
))
1033 if (get_user(s
, sp
))
1036 q
->sk
.sk_sndbuf
= s
;
1039 case TUNGETVNETHDRSZ
:
1041 if (put_user(s
, sp
))
1045 case TUNSETVNETHDRSZ
:
1046 if (get_user(s
, sp
))
1048 if (s
< (int)sizeof(struct virtio_net_hdr
))
1055 s
= !!(q
->flags
& TAP_VNET_LE
);
1056 if (put_user(s
, sp
))
1061 if (get_user(s
, sp
))
1064 q
->flags
|= TAP_VNET_LE
;
1066 q
->flags
&= ~TAP_VNET_LE
;
1070 return tap_get_vnet_be(q
, sp
);
1073 return tap_set_vnet_be(q
, sp
);
1076 /* let the user check for future flags */
1077 if (arg
& ~(TUN_F_CSUM
| TUN_F_TSO4
| TUN_F_TSO6
|
1082 ret
= set_offload(q
, arg
);
1088 tap
= tap_get_tap_dev(q
);
1095 if (copy_to_user(&ifr
->ifr_name
, tap
->dev
->name
, IFNAMSIZ
) ||
1096 copy_to_user(&ifr
->ifr_hwaddr
.sa_data
, tap
->dev
->dev_addr
, ETH_ALEN
) ||
1097 put_user(u
, &ifr
->ifr_hwaddr
.sa_family
))
1099 tap_put_tap_dev(tap
);
1104 if (copy_from_user(&sa
, &ifr
->ifr_hwaddr
, sizeof(sa
)))
1107 tap
= tap_get_tap_dev(q
);
1112 ret
= dev_set_mac_address(tap
->dev
, &sa
);
1113 tap_put_tap_dev(tap
);
1122 #ifdef CONFIG_COMPAT
1123 static long tap_compat_ioctl(struct file
*file
, unsigned int cmd
,
1126 return tap_ioctl(file
, cmd
, (unsigned long)compat_ptr(arg
));
1130 static const struct file_operations tap_fops
= {
1131 .owner
= THIS_MODULE
,
1133 .release
= tap_release
,
1134 .read_iter
= tap_read_iter
,
1135 .write_iter
= tap_write_iter
,
1137 .llseek
= no_llseek
,
1138 .unlocked_ioctl
= tap_ioctl
,
1139 #ifdef CONFIG_COMPAT
1140 .compat_ioctl
= tap_compat_ioctl
,
1144 static int tap_sendmsg(struct socket
*sock
, struct msghdr
*m
,
1147 struct tap_queue
*q
= container_of(sock
, struct tap_queue
, sock
);
1148 return tap_get_user(q
, m
, &m
->msg_iter
, m
->msg_flags
& MSG_DONTWAIT
);
1151 static int tap_recvmsg(struct socket
*sock
, struct msghdr
*m
,
1152 size_t total_len
, int flags
)
1154 struct tap_queue
*q
= container_of(sock
, struct tap_queue
, sock
);
1156 if (flags
& ~(MSG_DONTWAIT
|MSG_TRUNC
))
1158 ret
= tap_do_read(q
, &m
->msg_iter
, flags
& MSG_DONTWAIT
,
1160 if (ret
> total_len
) {
1161 m
->msg_flags
|= MSG_TRUNC
;
1162 ret
= flags
& MSG_TRUNC
? ret
: total_len
;
1167 static int tap_peek_len(struct socket
*sock
)
1169 struct tap_queue
*q
= container_of(sock
, struct tap_queue
,
1171 return skb_array_peek_len(&q
->skb_array
);
1174 /* Ops structure to mimic raw sockets with tun */
1175 static const struct proto_ops tap_socket_ops
= {
1176 .sendmsg
= tap_sendmsg
,
1177 .recvmsg
= tap_recvmsg
,
1178 .peek_len
= tap_peek_len
,
1181 /* Get an underlying socket object from tun file. Returns error unless file is
1182 * attached to a device. The returned object works like a packet socket, it
1183 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
1184 * holding a reference to the file for as long as the socket is in use. */
1185 struct socket
*tap_get_socket(struct file
*file
)
1187 struct tap_queue
*q
;
1188 if (file
->f_op
!= &tap_fops
)
1189 return ERR_PTR(-EINVAL
);
1190 q
= file
->private_data
;
1192 return ERR_PTR(-EBADFD
);
1195 EXPORT_SYMBOL_GPL(tap_get_socket
);
1197 struct skb_array
*tap_get_skb_array(struct file
*file
)
1199 struct tap_queue
*q
;
1201 if (file
->f_op
!= &tap_fops
)
1202 return ERR_PTR(-EINVAL
);
1203 q
= file
->private_data
;
1205 return ERR_PTR(-EBADFD
);
1206 return &q
->skb_array
;
1208 EXPORT_SYMBOL_GPL(tap_get_skb_array
);
1210 int tap_queue_resize(struct tap_dev
*tap
)
1212 struct net_device
*dev
= tap
->dev
;
1213 struct tap_queue
*q
;
1214 struct skb_array
**arrays
;
1215 int n
= tap
->numqueues
;
1218 arrays
= kmalloc_array(n
, sizeof(*arrays
), GFP_KERNEL
);
1222 list_for_each_entry(q
, &tap
->queue_list
, next
)
1223 arrays
[i
++] = &q
->skb_array
;
1225 ret
= skb_array_resize_multiple(arrays
, n
,
1226 dev
->tx_queue_len
, GFP_KERNEL
);
1231 EXPORT_SYMBOL_GPL(tap_queue_resize
);
1233 static int tap_list_add(dev_t major
, const char *device_name
)
1235 struct major_info
*tap_major
;
1237 tap_major
= kzalloc(sizeof(*tap_major
), GFP_ATOMIC
);
1241 tap_major
->major
= MAJOR(major
);
1243 idr_init(&tap_major
->minor_idr
);
1244 spin_lock_init(&tap_major
->minor_lock
);
1246 tap_major
->device_name
= device_name
;
1248 list_add_tail_rcu(&tap_major
->next
, &major_list
);
1252 int tap_create_cdev(struct cdev
*tap_cdev
,
1253 dev_t
*tap_major
, const char *device_name
)
1257 err
= alloc_chrdev_region(tap_major
, 0, TAP_NUM_DEVS
, device_name
);
1261 cdev_init(tap_cdev
, &tap_fops
);
1262 err
= cdev_add(tap_cdev
, *tap_major
, TAP_NUM_DEVS
);
1266 err
= tap_list_add(*tap_major
, device_name
);
1275 unregister_chrdev_region(*tap_major
, TAP_NUM_DEVS
);
1279 EXPORT_SYMBOL_GPL(tap_create_cdev
);
1281 void tap_destroy_cdev(dev_t major
, struct cdev
*tap_cdev
)
1283 struct major_info
*tap_major
, *tmp
;
1286 unregister_chrdev_region(major
, TAP_NUM_DEVS
);
1287 list_for_each_entry_safe(tap_major
, tmp
, &major_list
, next
) {
1288 if (tap_major
->major
== MAJOR(major
)) {
1289 idr_destroy(&tap_major
->minor_idr
);
1290 list_del_rcu(&tap_major
->next
);
1291 kfree_rcu(tap_major
, rcu
);
1295 EXPORT_SYMBOL_GPL(tap_destroy_cdev
);
1297 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1298 MODULE_AUTHOR("Sainath Grandhi <sainath.grandhi@intel.com>");
1299 MODULE_LICENSE("GPL");