2 * TUN - Universal TUN/TAP device driver.
3 * Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
21 * Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
22 * Add TUNSETLINK ioctl to set the link encapsulation
24 * Mark Smith <markzzzsmith@yahoo.com.au>
25 * Use eth_random_addr() for tap MAC address.
27 * Harald Roelle <harald.roelle@ifi.lmu.de> 2004/04/20
28 * Fixes in packet dropping, queue length setting and queue wakeup.
29 * Increased default tx queue length.
33 * Daniel Podlejski <underley@underley.eu.org>
34 * Modifications for 2.3.99-pre5 kernel.
37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
39 #define DRV_NAME "tun"
40 #define DRV_VERSION "1.6"
41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
42 #define DRV_COPYRIGHT "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
44 #include <linux/module.h>
45 #include <linux/errno.h>
46 #include <linux/kernel.h>
47 #include <linux/major.h>
48 #include <linux/slab.h>
49 #include <linux/poll.h>
50 #include <linux/fcntl.h>
51 #include <linux/init.h>
52 #include <linux/skbuff.h>
53 #include <linux/netdevice.h>
54 #include <linux/etherdevice.h>
55 #include <linux/miscdevice.h>
56 #include <linux/ethtool.h>
57 #include <linux/rtnetlink.h>
58 #include <linux/compat.h>
60 #include <linux/if_arp.h>
61 #include <linux/if_ether.h>
62 #include <linux/if_tun.h>
63 #include <linux/crc32.h>
64 #include <linux/nsproxy.h>
65 #include <linux/virtio_net.h>
66 #include <linux/rcupdate.h>
67 #include <net/net_namespace.h>
68 #include <net/netns/generic.h>
69 #include <net/rtnetlink.h>
72 #include <asm/uaccess.h>
74 /* Uncomment to enable debugging */
75 /* #define TUN_DEBUG 1 */
80 #define tun_debug(level, tun, fmt, args...) \
83 netdev_printk(level, tun->dev, fmt, ##args); \
85 #define DBG1(level, fmt, args...) \
88 printk(level fmt, ##args); \
91 #define tun_debug(level, tun, fmt, args...) \
94 netdev_printk(level, tun->dev, fmt, ##args); \
96 #define DBG1(level, fmt, args...) \
99 printk(level fmt, ##args); \
103 #define GOODCOPY_LEN 128
105 #define FLT_EXACT_COUNT 8
107 unsigned int count
; /* Number of addrs. Zero means disabled */
108 u32 mask
[2]; /* Mask of the hashed addrs */
109 unsigned char addr
[FLT_EXACT_COUNT
][ETH_ALEN
];
112 /* 1024 is probably a high enough limit: modern hypervisors seem to support on
113 * the order of 100-200 CPUs so this leaves us some breathing space if we want
114 * to match a queue per guest CPU.
116 #define MAX_TAP_QUEUES 1024
118 /* A tun_file connects an open character device to a tuntap netdevice. It
119 * also contains all socket related strctures (except sock_fprog and tap_filter)
120 * to serve as one transmit queue for tuntap device. The sock_fprog and
121 * tap_filter were kept in tun_struct since they were used for filtering for the
122 * netdevice not for a specific queue (at least I didn't see the reqirement for
126 * The tun_file and tun_struct are loosely coupled, the pointer from on to the
127 * other can only be read while rcu_read_lock or rtnl_lock is held.
131 struct socket socket
;
133 struct tun_struct __rcu
*tun
;
135 struct fasync_struct
*fasync
;
136 /* only used for fasnyc */
141 /* Since the socket were moved to tun_file, to preserve the behavior of persist
142 * device, socket fileter, sndbuf and vnet header size were restore when the
143 * file were attached to a persist device.
146 struct tun_file __rcu
*tfiles
[MAX_TAP_QUEUES
];
147 unsigned int numqueues
;
152 struct net_device
*dev
;
153 netdev_features_t set_features
;
154 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
155 NETIF_F_TSO6|NETIF_F_UFO)
159 struct tap_filter txflt
;
160 struct sock_fprog fprog
;
161 /* protected by rtnl lock */
162 bool filter_attached
;
168 /* We try to identify a flow through its rxhash first. The reason that
169 * we do not check rxq no. is becuase some cards(e.g 82599), chooses
170 * the rxq based on the txq where the last packet of the flow comes. As
171 * the userspace application move between processors, we may get a
172 * different rxq no. here. If we could not get rxhash, then we would
173 * hope the rxq no. may help here.
175 static u16
tun_select_queue(struct net_device
*dev
, struct sk_buff
*skb
)
177 struct tun_struct
*tun
= netdev_priv(dev
);
182 numqueues
= tun
->numqueues
;
184 txq
= skb_get_rxhash(skb
);
186 /* use multiply and shift instead of expensive divide */
187 txq
= ((u64
)txq
* numqueues
) >> 32;
188 } else if (likely(skb_rx_queue_recorded(skb
))) {
189 txq
= skb_get_rx_queue(skb
);
190 while (unlikely(txq
>= numqueues
))
198 static void tun_set_real_num_queues(struct tun_struct
*tun
)
200 netif_set_real_num_tx_queues(tun
->dev
, tun
->numqueues
);
201 netif_set_real_num_rx_queues(tun
->dev
, tun
->numqueues
);
204 static void __tun_detach(struct tun_file
*tfile
, bool clean
)
206 struct tun_file
*ntfile
;
207 struct tun_struct
*tun
;
208 struct net_device
*dev
;
210 tun
= rcu_dereference_protected(tfile
->tun
,
211 lockdep_rtnl_is_held());
213 u16 index
= tfile
->queue_index
;
214 BUG_ON(index
>= tun
->numqueues
);
217 rcu_assign_pointer(tun
->tfiles
[index
],
218 tun
->tfiles
[tun
->numqueues
- 1]);
219 rcu_assign_pointer(tfile
->tun
, NULL
);
220 ntfile
= rcu_dereference_protected(tun
->tfiles
[index
],
221 lockdep_rtnl_is_held());
222 ntfile
->queue_index
= index
;
225 sock_put(&tfile
->sk
);
228 /* Drop read queue */
229 skb_queue_purge(&tfile
->sk
.sk_receive_queue
);
230 tun_set_real_num_queues(tun
);
232 if (tun
->numqueues
== 0 && !(tun
->flags
& TUN_PERSIST
))
233 if (dev
->reg_state
== NETREG_REGISTERED
)
234 unregister_netdevice(dev
);
238 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED
,
239 &tfile
->socket
.flags
));
240 sk_release_kernel(&tfile
->sk
);
244 static void tun_detach(struct tun_file
*tfile
, bool clean
)
247 __tun_detach(tfile
, clean
);
251 static void tun_detach_all(struct net_device
*dev
)
253 struct tun_struct
*tun
= netdev_priv(dev
);
254 struct tun_file
*tfile
;
255 int i
, n
= tun
->numqueues
;
257 for (i
= 0; i
< n
; i
++) {
258 tfile
= rcu_dereference_protected(tun
->tfiles
[i
],
259 lockdep_rtnl_is_held());
261 wake_up_all(&tfile
->wq
.wait
);
262 rcu_assign_pointer(tfile
->tun
, NULL
);
265 BUG_ON(tun
->numqueues
!= 0);
268 for (i
= 0; i
< n
; i
++) {
269 tfile
= rcu_dereference_protected(tun
->tfiles
[i
],
270 lockdep_rtnl_is_held());
271 /* Drop read queue */
272 skb_queue_purge(&tfile
->sk
.sk_receive_queue
);
273 sock_put(&tfile
->sk
);
277 static int tun_attach(struct tun_struct
*tun
, struct file
*file
)
279 struct tun_file
*tfile
= file
->private_data
;
283 if (rcu_dereference_protected(tfile
->tun
, lockdep_rtnl_is_held()))
287 if (!(tun
->flags
& TUN_TAP_MQ
) && tun
->numqueues
== 1)
291 if (tun
->numqueues
== MAX_TAP_QUEUES
)
296 /* Re-attach the filter to presist device */
297 if (tun
->filter_attached
== true) {
298 err
= sk_attach_filter(&tun
->fprog
, tfile
->socket
.sk
);
302 tfile
->queue_index
= tun
->numqueues
;
303 rcu_assign_pointer(tfile
->tun
, tun
);
304 rcu_assign_pointer(tun
->tfiles
[tun
->numqueues
], tfile
);
305 sock_hold(&tfile
->sk
);
308 tun_set_real_num_queues(tun
);
310 if (tun
->numqueues
== 1)
311 netif_carrier_on(tun
->dev
);
313 /* device is allowed to go away first, so no need to hold extra
321 static struct tun_struct
*__tun_get(struct tun_file
*tfile
)
323 struct tun_struct
*tun
;
326 tun
= rcu_dereference(tfile
->tun
);
334 static struct tun_struct
*tun_get(struct file
*file
)
336 return __tun_get(file
->private_data
);
339 static void tun_put(struct tun_struct
*tun
)
345 static void addr_hash_set(u32
*mask
, const u8
*addr
)
347 int n
= ether_crc(ETH_ALEN
, addr
) >> 26;
348 mask
[n
>> 5] |= (1 << (n
& 31));
351 static unsigned int addr_hash_test(const u32
*mask
, const u8
*addr
)
353 int n
= ether_crc(ETH_ALEN
, addr
) >> 26;
354 return mask
[n
>> 5] & (1 << (n
& 31));
357 static int update_filter(struct tap_filter
*filter
, void __user
*arg
)
359 struct { u8 u
[ETH_ALEN
]; } *addr
;
360 struct tun_filter uf
;
361 int err
, alen
, n
, nexact
;
363 if (copy_from_user(&uf
, arg
, sizeof(uf
)))
372 alen
= ETH_ALEN
* uf
.count
;
373 addr
= kmalloc(alen
, GFP_KERNEL
);
377 if (copy_from_user(addr
, arg
+ sizeof(uf
), alen
)) {
382 /* The filter is updated without holding any locks. Which is
383 * perfectly safe. We disable it first and in the worst
384 * case we'll accept a few undesired packets. */
388 /* Use first set of addresses as an exact filter */
389 for (n
= 0; n
< uf
.count
&& n
< FLT_EXACT_COUNT
; n
++)
390 memcpy(filter
->addr
[n
], addr
[n
].u
, ETH_ALEN
);
394 /* Remaining multicast addresses are hashed,
395 * unicast will leave the filter disabled. */
396 memset(filter
->mask
, 0, sizeof(filter
->mask
));
397 for (; n
< uf
.count
; n
++) {
398 if (!is_multicast_ether_addr(addr
[n
].u
)) {
399 err
= 0; /* no filter */
402 addr_hash_set(filter
->mask
, addr
[n
].u
);
405 /* For ALLMULTI just set the mask to all ones.
406 * This overrides the mask populated above. */
407 if ((uf
.flags
& TUN_FLT_ALLMULTI
))
408 memset(filter
->mask
, ~0, sizeof(filter
->mask
));
410 /* Now enable the filter */
412 filter
->count
= nexact
;
414 /* Return the number of exact filters */
422 /* Returns: 0 - drop, !=0 - accept */
423 static int run_filter(struct tap_filter
*filter
, const struct sk_buff
*skb
)
425 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
427 struct ethhdr
*eh
= (struct ethhdr
*) skb
->data
;
431 for (i
= 0; i
< filter
->count
; i
++)
432 if (ether_addr_equal(eh
->h_dest
, filter
->addr
[i
]))
435 /* Inexact match (multicast only) */
436 if (is_multicast_ether_addr(eh
->h_dest
))
437 return addr_hash_test(filter
->mask
, eh
->h_dest
);
443 * Checks whether the packet is accepted or not.
444 * Returns: 0 - drop, !=0 - accept
446 static int check_filter(struct tap_filter
*filter
, const struct sk_buff
*skb
)
451 return run_filter(filter
, skb
);
454 /* Network device part of the driver */
456 static const struct ethtool_ops tun_ethtool_ops
;
458 /* Net device detach from fd. */
459 static void tun_net_uninit(struct net_device
*dev
)
464 /* Net device open. */
465 static int tun_net_open(struct net_device
*dev
)
467 netif_tx_start_all_queues(dev
);
471 /* Net device close. */
472 static int tun_net_close(struct net_device
*dev
)
474 netif_tx_stop_all_queues(dev
);
478 /* Net device start xmit */
479 static netdev_tx_t
tun_net_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
481 struct tun_struct
*tun
= netdev_priv(dev
);
482 int txq
= skb
->queue_mapping
;
483 struct tun_file
*tfile
;
486 tfile
= rcu_dereference(tun
->tfiles
[txq
]);
488 /* Drop packet if interface is not attached */
489 if (txq
>= tun
->numqueues
)
492 tun_debug(KERN_INFO
, tun
, "tun_net_xmit %d\n", skb
->len
);
496 /* Drop if the filter does not like it.
497 * This is a noop if the filter is disabled.
498 * Filter can be enabled only for the TAP devices. */
499 if (!check_filter(&tun
->txflt
, skb
))
502 if (tfile
->socket
.sk
->sk_filter
&&
503 sk_filter(tfile
->socket
.sk
, skb
))
506 /* Limit the number of packets queued by divining txq length with the
509 if (skb_queue_len(&tfile
->socket
.sk
->sk_receive_queue
)
510 >= dev
->tx_queue_len
/ tun
->numqueues
){
511 if (!(tun
->flags
& TUN_ONE_QUEUE
)) {
512 /* Normal queueing mode. */
513 /* Packet scheduler handles dropping of further packets. */
514 netif_stop_subqueue(dev
, txq
);
516 /* We won't see all dropped packets individually, so overrun
517 * error is more appropriate. */
518 dev
->stats
.tx_fifo_errors
++;
520 /* Single queue mode.
521 * Driver handles dropping of all packets itself. */
526 /* Orphan the skb - required as we might hang on to it
527 * for indefinite time. */
528 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
533 skb_queue_tail(&tfile
->socket
.sk
->sk_receive_queue
, skb
);
535 /* Notify and wake up reader process */
536 if (tfile
->flags
& TUN_FASYNC
)
537 kill_fasync(&tfile
->fasync
, SIGIO
, POLL_IN
);
538 wake_up_interruptible_poll(&tfile
->wq
.wait
, POLLIN
|
539 POLLRDNORM
| POLLRDBAND
);
545 dev
->stats
.tx_dropped
++;
551 static void tun_net_mclist(struct net_device
*dev
)
554 * This callback is supposed to deal with mc filter in
555 * _rx_ path and has nothing to do with the _tx_ path.
556 * In rx path we always accept everything userspace gives us.
561 #define MAX_MTU 65535
564 tun_net_change_mtu(struct net_device
*dev
, int new_mtu
)
566 if (new_mtu
< MIN_MTU
|| new_mtu
+ dev
->hard_header_len
> MAX_MTU
)
572 static netdev_features_t
tun_net_fix_features(struct net_device
*dev
,
573 netdev_features_t features
)
575 struct tun_struct
*tun
= netdev_priv(dev
);
577 return (features
& tun
->set_features
) | (features
& ~TUN_USER_FEATURES
);
579 #ifdef CONFIG_NET_POLL_CONTROLLER
580 static void tun_poll_controller(struct net_device
*dev
)
583 * Tun only receives frames when:
584 * 1) the char device endpoint gets data from user space
585 * 2) the tun socket gets a sendmsg call from user space
586 * Since both of those are syncronous operations, we are guaranteed
587 * never to have pending data when we poll for it
588 * so theres nothing to do here but return.
589 * We need this though so netpoll recognizes us as an interface that
590 * supports polling, which enables bridge devices in virt setups to
591 * still use netconsole
596 static const struct net_device_ops tun_netdev_ops
= {
597 .ndo_uninit
= tun_net_uninit
,
598 .ndo_open
= tun_net_open
,
599 .ndo_stop
= tun_net_close
,
600 .ndo_start_xmit
= tun_net_xmit
,
601 .ndo_change_mtu
= tun_net_change_mtu
,
602 .ndo_fix_features
= tun_net_fix_features
,
603 .ndo_select_queue
= tun_select_queue
,
604 #ifdef CONFIG_NET_POLL_CONTROLLER
605 .ndo_poll_controller
= tun_poll_controller
,
609 static const struct net_device_ops tap_netdev_ops
= {
610 .ndo_uninit
= tun_net_uninit
,
611 .ndo_open
= tun_net_open
,
612 .ndo_stop
= tun_net_close
,
613 .ndo_start_xmit
= tun_net_xmit
,
614 .ndo_change_mtu
= tun_net_change_mtu
,
615 .ndo_fix_features
= tun_net_fix_features
,
616 .ndo_set_rx_mode
= tun_net_mclist
,
617 .ndo_set_mac_address
= eth_mac_addr
,
618 .ndo_validate_addr
= eth_validate_addr
,
619 .ndo_select_queue
= tun_select_queue
,
620 #ifdef CONFIG_NET_POLL_CONTROLLER
621 .ndo_poll_controller
= tun_poll_controller
,
625 /* Initialize net device. */
626 static void tun_net_init(struct net_device
*dev
)
628 struct tun_struct
*tun
= netdev_priv(dev
);
630 switch (tun
->flags
& TUN_TYPE_MASK
) {
632 dev
->netdev_ops
= &tun_netdev_ops
;
634 /* Point-to-Point TUN Device */
635 dev
->hard_header_len
= 0;
639 /* Zero header length */
640 dev
->type
= ARPHRD_NONE
;
641 dev
->flags
= IFF_POINTOPOINT
| IFF_NOARP
| IFF_MULTICAST
;
642 dev
->tx_queue_len
= TUN_READQ_SIZE
; /* We prefer our own queue length */
646 dev
->netdev_ops
= &tap_netdev_ops
;
647 /* Ethernet TAP Device */
649 dev
->priv_flags
&= ~IFF_TX_SKB_SHARING
;
651 eth_hw_addr_random(dev
);
653 dev
->tx_queue_len
= TUN_READQ_SIZE
; /* We prefer our own queue length */
658 /* Character device part */
661 static unsigned int tun_chr_poll(struct file
*file
, poll_table
*wait
)
663 struct tun_file
*tfile
= file
->private_data
;
664 struct tun_struct
*tun
= __tun_get(tfile
);
666 unsigned int mask
= 0;
671 sk
= tfile
->socket
.sk
;
673 tun_debug(KERN_INFO
, tun
, "tun_chr_poll\n");
675 poll_wait(file
, &tfile
->wq
.wait
, wait
);
677 if (!skb_queue_empty(&sk
->sk_receive_queue
))
678 mask
|= POLLIN
| POLLRDNORM
;
680 if (sock_writeable(sk
) ||
681 (!test_and_set_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
) &&
683 mask
|= POLLOUT
| POLLWRNORM
;
685 if (tun
->dev
->reg_state
!= NETREG_REGISTERED
)
692 /* prepad is the amount to reserve at front. len is length after that.
693 * linear is a hint as to how much to copy (usually headers). */
694 static struct sk_buff
*tun_alloc_skb(struct tun_file
*tfile
,
695 size_t prepad
, size_t len
,
696 size_t linear
, int noblock
)
698 struct sock
*sk
= tfile
->socket
.sk
;
702 /* Under a page? Don't bother with paged skb. */
703 if (prepad
+ len
< PAGE_SIZE
|| !linear
)
706 skb
= sock_alloc_send_pskb(sk
, prepad
+ linear
, len
- linear
, noblock
,
711 skb_reserve(skb
, prepad
);
712 skb_put(skb
, linear
);
713 skb
->data_len
= len
- linear
;
714 skb
->len
+= len
- linear
;
719 /* set skb frags from iovec, this can move to core network code for reuse */
720 static int zerocopy_sg_from_iovec(struct sk_buff
*skb
, const struct iovec
*from
,
721 int offset
, size_t count
)
723 int len
= iov_length(from
, count
) - offset
;
724 int copy
= skb_headlen(skb
);
725 int size
, offset1
= 0;
728 /* Skip over from offset */
729 while (count
&& (offset
>= from
->iov_len
)) {
730 offset
-= from
->iov_len
;
735 /* copy up to skb headlen */
736 while (count
&& (copy
> 0)) {
737 size
= min_t(unsigned int, copy
, from
->iov_len
- offset
);
738 if (copy_from_user(skb
->data
+ offset1
, from
->iov_base
+ offset
,
755 struct page
*page
[MAX_SKB_FRAGS
];
758 unsigned long truesize
;
760 len
= from
->iov_len
- offset
;
766 base
= (unsigned long)from
->iov_base
+ offset
;
767 size
= ((base
& ~PAGE_MASK
) + len
+ ~PAGE_MASK
) >> PAGE_SHIFT
;
768 if (i
+ size
> MAX_SKB_FRAGS
)
770 num_pages
= get_user_pages_fast(base
, size
, 0, &page
[i
]);
771 if (num_pages
!= size
) {
772 for (i
= 0; i
< num_pages
; i
++)
776 truesize
= size
* PAGE_SIZE
;
777 skb
->data_len
+= len
;
779 skb
->truesize
+= truesize
;
780 atomic_add(truesize
, &skb
->sk
->sk_wmem_alloc
);
782 int off
= base
& ~PAGE_MASK
;
783 int size
= min_t(int, len
, PAGE_SIZE
- off
);
784 __skb_fill_page_desc(skb
, i
, page
[i
], off
, size
);
785 skb_shinfo(skb
)->nr_frags
++;
786 /* increase sk_wmem_alloc */
797 /* Get packet from user space buffer */
798 static ssize_t
tun_get_user(struct tun_struct
*tun
, struct tun_file
*tfile
,
799 void *msg_control
, const struct iovec
*iv
,
800 size_t total_len
, size_t count
, int noblock
)
802 struct tun_pi pi
= { 0, cpu_to_be16(ETH_P_IP
) };
804 size_t len
= total_len
, align
= NET_SKB_PAD
;
805 struct virtio_net_hdr gso
= { 0 };
808 bool zerocopy
= false;
811 if (!(tun
->flags
& TUN_NO_PI
)) {
812 if ((len
-= sizeof(pi
)) > total_len
)
815 if (memcpy_fromiovecend((void *)&pi
, iv
, 0, sizeof(pi
)))
817 offset
+= sizeof(pi
);
820 if (tun
->flags
& TUN_VNET_HDR
) {
821 if ((len
-= tun
->vnet_hdr_sz
) > total_len
)
824 if (memcpy_fromiovecend((void *)&gso
, iv
, offset
, sizeof(gso
)))
827 if ((gso
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) &&
828 gso
.csum_start
+ gso
.csum_offset
+ 2 > gso
.hdr_len
)
829 gso
.hdr_len
= gso
.csum_start
+ gso
.csum_offset
+ 2;
831 if (gso
.hdr_len
> len
)
833 offset
+= tun
->vnet_hdr_sz
;
836 if ((tun
->flags
& TUN_TYPE_MASK
) == TUN_TAP_DEV
) {
837 align
+= NET_IP_ALIGN
;
838 if (unlikely(len
< ETH_HLEN
||
839 (gso
.hdr_len
&& gso
.hdr_len
< ETH_HLEN
)))
847 /* Userspace may produce vectors with count greater than
848 * MAX_SKB_FRAGS, so we need to linearize parts of the skb
849 * to let the rest of data to be fit in the frags.
851 if (count
> MAX_SKB_FRAGS
) {
852 copylen
= iov_length(iv
, count
- MAX_SKB_FRAGS
);
853 if (copylen
< offset
)
859 /* There are 256 bytes to be copied in skb, so there is enough
860 * room for skb expand head in case it is used.
861 * The rest of the buffer is mapped from userspace.
863 if (copylen
< gso
.hdr_len
)
864 copylen
= gso
.hdr_len
;
866 copylen
= GOODCOPY_LEN
;
870 skb
= tun_alloc_skb(tfile
, align
, copylen
, gso
.hdr_len
, noblock
);
872 if (PTR_ERR(skb
) != -EAGAIN
)
873 tun
->dev
->stats
.rx_dropped
++;
878 err
= zerocopy_sg_from_iovec(skb
, iv
, offset
, count
);
880 err
= skb_copy_datagram_from_iovec(skb
, 0, iv
, offset
, len
);
883 tun
->dev
->stats
.rx_dropped
++;
888 if (gso
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) {
889 if (!skb_partial_csum_set(skb
, gso
.csum_start
,
891 tun
->dev
->stats
.rx_frame_errors
++;
897 switch (tun
->flags
& TUN_TYPE_MASK
) {
899 if (tun
->flags
& TUN_NO_PI
) {
900 switch (skb
->data
[0] & 0xf0) {
902 pi
.proto
= htons(ETH_P_IP
);
905 pi
.proto
= htons(ETH_P_IPV6
);
908 tun
->dev
->stats
.rx_dropped
++;
914 skb_reset_mac_header(skb
);
915 skb
->protocol
= pi
.proto
;
919 skb
->protocol
= eth_type_trans(skb
, tun
->dev
);
923 if (gso
.gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
925 switch (gso
.gso_type
& ~VIRTIO_NET_HDR_GSO_ECN
) {
926 case VIRTIO_NET_HDR_GSO_TCPV4
:
927 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV4
;
929 case VIRTIO_NET_HDR_GSO_TCPV6
:
930 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV6
;
932 case VIRTIO_NET_HDR_GSO_UDP
:
933 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
936 tun
->dev
->stats
.rx_frame_errors
++;
941 if (gso
.gso_type
& VIRTIO_NET_HDR_GSO_ECN
)
942 skb_shinfo(skb
)->gso_type
|= SKB_GSO_TCP_ECN
;
944 skb_shinfo(skb
)->gso_size
= gso
.gso_size
;
945 if (skb_shinfo(skb
)->gso_size
== 0) {
946 tun
->dev
->stats
.rx_frame_errors
++;
951 /* Header must be checked, and gso_segs computed. */
952 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
953 skb_shinfo(skb
)->gso_segs
= 0;
956 /* copy skb_ubuf_info for callback when skb has no error */
958 skb_shinfo(skb
)->destructor_arg
= msg_control
;
959 skb_shinfo(skb
)->tx_flags
|= SKBTX_DEV_ZEROCOPY
;
964 tun
->dev
->stats
.rx_packets
++;
965 tun
->dev
->stats
.rx_bytes
+= len
;
970 static ssize_t
tun_chr_aio_write(struct kiocb
*iocb
, const struct iovec
*iv
,
971 unsigned long count
, loff_t pos
)
973 struct file
*file
= iocb
->ki_filp
;
974 struct tun_struct
*tun
= tun_get(file
);
975 struct tun_file
*tfile
= file
->private_data
;
981 tun_debug(KERN_INFO
, tun
, "tun_chr_write %ld\n", count
);
983 result
= tun_get_user(tun
, tfile
, NULL
, iv
, iov_length(iv
, count
),
984 count
, file
->f_flags
& O_NONBLOCK
);
990 /* Put packet to the user space buffer */
991 static ssize_t
tun_put_user(struct tun_struct
*tun
,
992 struct tun_file
*tfile
,
994 const struct iovec
*iv
, int len
)
996 struct tun_pi pi
= { 0, skb
->protocol
};
999 if (!(tun
->flags
& TUN_NO_PI
)) {
1000 if ((len
-= sizeof(pi
)) < 0)
1003 if (len
< skb
->len
) {
1004 /* Packet will be striped */
1005 pi
.flags
|= TUN_PKT_STRIP
;
1008 if (memcpy_toiovecend(iv
, (void *) &pi
, 0, sizeof(pi
)))
1010 total
+= sizeof(pi
);
1013 if (tun
->flags
& TUN_VNET_HDR
) {
1014 struct virtio_net_hdr gso
= { 0 }; /* no info leak */
1015 if ((len
-= tun
->vnet_hdr_sz
) < 0)
1018 if (skb_is_gso(skb
)) {
1019 struct skb_shared_info
*sinfo
= skb_shinfo(skb
);
1021 /* This is a hint as to how much should be linear. */
1022 gso
.hdr_len
= skb_headlen(skb
);
1023 gso
.gso_size
= sinfo
->gso_size
;
1024 if (sinfo
->gso_type
& SKB_GSO_TCPV4
)
1025 gso
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV4
;
1026 else if (sinfo
->gso_type
& SKB_GSO_TCPV6
)
1027 gso
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV6
;
1028 else if (sinfo
->gso_type
& SKB_GSO_UDP
)
1029 gso
.gso_type
= VIRTIO_NET_HDR_GSO_UDP
;
1031 pr_err("unexpected GSO type: "
1032 "0x%x, gso_size %d, hdr_len %d\n",
1033 sinfo
->gso_type
, gso
.gso_size
,
1035 print_hex_dump(KERN_ERR
, "tun: ",
1038 min((int)gso
.hdr_len
, 64), true);
1042 if (sinfo
->gso_type
& SKB_GSO_TCP_ECN
)
1043 gso
.gso_type
|= VIRTIO_NET_HDR_GSO_ECN
;
1045 gso
.gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
1047 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1048 gso
.flags
= VIRTIO_NET_HDR_F_NEEDS_CSUM
;
1049 gso
.csum_start
= skb_checksum_start_offset(skb
);
1050 gso
.csum_offset
= skb
->csum_offset
;
1051 } else if (skb
->ip_summed
== CHECKSUM_UNNECESSARY
) {
1052 gso
.flags
= VIRTIO_NET_HDR_F_DATA_VALID
;
1053 } /* else everything is zero */
1055 if (unlikely(memcpy_toiovecend(iv
, (void *)&gso
, total
,
1058 total
+= tun
->vnet_hdr_sz
;
1061 len
= min_t(int, skb
->len
, len
);
1063 skb_copy_datagram_const_iovec(skb
, 0, iv
, total
, len
);
1066 tun
->dev
->stats
.tx_packets
++;
1067 tun
->dev
->stats
.tx_bytes
+= len
;
1072 static ssize_t
tun_do_read(struct tun_struct
*tun
, struct tun_file
*tfile
,
1073 struct kiocb
*iocb
, const struct iovec
*iv
,
1074 ssize_t len
, int noblock
)
1076 DECLARE_WAITQUEUE(wait
, current
);
1077 struct sk_buff
*skb
;
1080 tun_debug(KERN_INFO
, tun
, "tun_chr_read\n");
1082 if (unlikely(!noblock
))
1083 add_wait_queue(&tfile
->wq
.wait
, &wait
);
1085 current
->state
= TASK_INTERRUPTIBLE
;
1087 /* Read frames from the queue */
1088 if (!(skb
= skb_dequeue(&tfile
->socket
.sk
->sk_receive_queue
))) {
1093 if (signal_pending(current
)) {
1097 if (tun
->dev
->reg_state
!= NETREG_REGISTERED
) {
1102 /* Nothing to read, let's sleep */
1106 netif_wake_subqueue(tun
->dev
, tfile
->queue_index
);
1108 ret
= tun_put_user(tun
, tfile
, skb
, iv
, len
);
1113 current
->state
= TASK_RUNNING
;
1114 if (unlikely(!noblock
))
1115 remove_wait_queue(&tfile
->wq
.wait
, &wait
);
1120 static ssize_t
tun_chr_aio_read(struct kiocb
*iocb
, const struct iovec
*iv
,
1121 unsigned long count
, loff_t pos
)
1123 struct file
*file
= iocb
->ki_filp
;
1124 struct tun_file
*tfile
= file
->private_data
;
1125 struct tun_struct
*tun
= __tun_get(tfile
);
1130 len
= iov_length(iv
, count
);
1136 ret
= tun_do_read(tun
, tfile
, iocb
, iv
, len
,
1137 file
->f_flags
& O_NONBLOCK
);
1138 ret
= min_t(ssize_t
, ret
, len
);
1144 static void tun_setup(struct net_device
*dev
)
1146 struct tun_struct
*tun
= netdev_priv(dev
);
1148 tun
->owner
= INVALID_UID
;
1149 tun
->group
= INVALID_GID
;
1151 dev
->ethtool_ops
= &tun_ethtool_ops
;
1152 dev
->destructor
= free_netdev
;
1155 /* Trivial set of netlink ops to allow deleting tun or tap
1156 * device with netlink.
1158 static int tun_validate(struct nlattr
*tb
[], struct nlattr
*data
[])
1163 static struct rtnl_link_ops tun_link_ops __read_mostly
= {
1165 .priv_size
= sizeof(struct tun_struct
),
1167 .validate
= tun_validate
,
1170 static void tun_sock_write_space(struct sock
*sk
)
1172 struct tun_file
*tfile
;
1173 wait_queue_head_t
*wqueue
;
1175 if (!sock_writeable(sk
))
1178 if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
))
1181 wqueue
= sk_sleep(sk
);
1182 if (wqueue
&& waitqueue_active(wqueue
))
1183 wake_up_interruptible_sync_poll(wqueue
, POLLOUT
|
1184 POLLWRNORM
| POLLWRBAND
);
1186 tfile
= container_of(sk
, struct tun_file
, sk
);
1187 kill_fasync(&tfile
->fasync
, SIGIO
, POLL_OUT
);
1190 static int tun_sendmsg(struct kiocb
*iocb
, struct socket
*sock
,
1191 struct msghdr
*m
, size_t total_len
)
1194 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
1195 struct tun_struct
*tun
= __tun_get(tfile
);
1199 ret
= tun_get_user(tun
, tfile
, m
->msg_control
, m
->msg_iov
, total_len
,
1200 m
->msg_iovlen
, m
->msg_flags
& MSG_DONTWAIT
);
1206 static int tun_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
1207 struct msghdr
*m
, size_t total_len
,
1210 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
1211 struct tun_struct
*tun
= __tun_get(tfile
);
1217 if (flags
& ~(MSG_DONTWAIT
|MSG_TRUNC
))
1219 ret
= tun_do_read(tun
, tfile
, iocb
, m
->msg_iov
, total_len
,
1220 flags
& MSG_DONTWAIT
);
1221 if (ret
> total_len
) {
1222 m
->msg_flags
|= MSG_TRUNC
;
1223 ret
= flags
& MSG_TRUNC
? ret
: total_len
;
1229 static int tun_release(struct socket
*sock
)
1236 /* Ops structure to mimic raw sockets with tun */
1237 static const struct proto_ops tun_socket_ops
= {
1238 .sendmsg
= tun_sendmsg
,
1239 .recvmsg
= tun_recvmsg
,
1240 .release
= tun_release
,
1243 static struct proto tun_proto
= {
1245 .owner
= THIS_MODULE
,
1246 .obj_size
= sizeof(struct tun_file
),
1249 static int tun_flags(struct tun_struct
*tun
)
1253 if (tun
->flags
& TUN_TUN_DEV
)
1258 if (tun
->flags
& TUN_NO_PI
)
1261 if (tun
->flags
& TUN_ONE_QUEUE
)
1262 flags
|= IFF_ONE_QUEUE
;
1264 if (tun
->flags
& TUN_VNET_HDR
)
1265 flags
|= IFF_VNET_HDR
;
1267 if (tun
->flags
& TUN_TAP_MQ
)
1268 flags
|= IFF_MULTI_QUEUE
;
1273 static ssize_t
tun_show_flags(struct device
*dev
, struct device_attribute
*attr
,
1276 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
1277 return sprintf(buf
, "0x%x\n", tun_flags(tun
));
1280 static ssize_t
tun_show_owner(struct device
*dev
, struct device_attribute
*attr
,
1283 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
1284 return uid_valid(tun
->owner
)?
1285 sprintf(buf
, "%u\n",
1286 from_kuid_munged(current_user_ns(), tun
->owner
)):
1287 sprintf(buf
, "-1\n");
1290 static ssize_t
tun_show_group(struct device
*dev
, struct device_attribute
*attr
,
1293 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
1294 return gid_valid(tun
->group
) ?
1295 sprintf(buf
, "%u\n",
1296 from_kgid_munged(current_user_ns(), tun
->group
)):
1297 sprintf(buf
, "-1\n");
1300 static DEVICE_ATTR(tun_flags
, 0444, tun_show_flags
, NULL
);
1301 static DEVICE_ATTR(owner
, 0444, tun_show_owner
, NULL
);
1302 static DEVICE_ATTR(group
, 0444, tun_show_group
, NULL
);
1304 static int tun_set_iff(struct net
*net
, struct file
*file
, struct ifreq
*ifr
)
1306 struct tun_struct
*tun
;
1307 struct tun_file
*tfile
= file
->private_data
;
1308 struct net_device
*dev
;
1311 dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
1313 const struct cred
*cred
= current_cred();
1315 if (ifr
->ifr_flags
& IFF_TUN_EXCL
)
1317 if ((ifr
->ifr_flags
& IFF_TUN
) && dev
->netdev_ops
== &tun_netdev_ops
)
1318 tun
= netdev_priv(dev
);
1319 else if ((ifr
->ifr_flags
& IFF_TAP
) && dev
->netdev_ops
== &tap_netdev_ops
)
1320 tun
= netdev_priv(dev
);
1324 if (((uid_valid(tun
->owner
) && !uid_eq(cred
->euid
, tun
->owner
)) ||
1325 (gid_valid(tun
->group
) && !in_egroup_p(tun
->group
))) &&
1326 !capable(CAP_NET_ADMIN
))
1328 err
= security_tun_dev_attach(tfile
->socket
.sk
);
1332 err
= tun_attach(tun
, file
);
1338 unsigned long flags
= 0;
1340 if (!capable(CAP_NET_ADMIN
))
1342 err
= security_tun_dev_create();
1347 if (ifr
->ifr_flags
& IFF_TUN
) {
1349 flags
|= TUN_TUN_DEV
;
1351 } else if (ifr
->ifr_flags
& IFF_TAP
) {
1353 flags
|= TUN_TAP_DEV
;
1359 name
= ifr
->ifr_name
;
1361 dev
= alloc_netdev_mqs(sizeof(struct tun_struct
), name
,
1363 MAX_TAP_QUEUES
, MAX_TAP_QUEUES
);
1367 dev_net_set(dev
, net
);
1368 dev
->rtnl_link_ops
= &tun_link_ops
;
1370 tun
= netdev_priv(dev
);
1373 tun
->txflt
.count
= 0;
1374 tun
->vnet_hdr_sz
= sizeof(struct virtio_net_hdr
);
1376 tun
->filter_attached
= false;
1377 tun
->sndbuf
= tfile
->socket
.sk
->sk_sndbuf
;
1379 security_tun_dev_post_create(&tfile
->sk
);
1383 dev
->hw_features
= NETIF_F_SG
| NETIF_F_FRAGLIST
|
1385 dev
->features
= dev
->hw_features
;
1387 err
= register_netdevice(tun
->dev
);
1391 if (device_create_file(&tun
->dev
->dev
, &dev_attr_tun_flags
) ||
1392 device_create_file(&tun
->dev
->dev
, &dev_attr_owner
) ||
1393 device_create_file(&tun
->dev
->dev
, &dev_attr_group
))
1394 pr_err("Failed to create tun sysfs files\n");
1396 err
= tun_attach(tun
, file
);
1401 tun_debug(KERN_INFO
, tun
, "tun_set_iff\n");
1403 if (ifr
->ifr_flags
& IFF_NO_PI
)
1404 tun
->flags
|= TUN_NO_PI
;
1406 tun
->flags
&= ~TUN_NO_PI
;
1408 if (ifr
->ifr_flags
& IFF_ONE_QUEUE
)
1409 tun
->flags
|= TUN_ONE_QUEUE
;
1411 tun
->flags
&= ~TUN_ONE_QUEUE
;
1413 if (ifr
->ifr_flags
& IFF_VNET_HDR
)
1414 tun
->flags
|= TUN_VNET_HDR
;
1416 tun
->flags
&= ~TUN_VNET_HDR
;
1418 if (ifr
->ifr_flags
& IFF_MULTI_QUEUE
)
1419 tun
->flags
|= TUN_TAP_MQ
;
1421 tun
->flags
&= ~TUN_TAP_MQ
;
1423 /* Make sure persistent devices do not get stuck in
1426 if (netif_running(tun
->dev
))
1427 netif_tx_wake_all_queues(tun
->dev
);
1429 strcpy(ifr
->ifr_name
, tun
->dev
->name
);
1437 static int tun_get_iff(struct net
*net
, struct tun_struct
*tun
,
1440 tun_debug(KERN_INFO
, tun
, "tun_get_iff\n");
1442 strcpy(ifr
->ifr_name
, tun
->dev
->name
);
1444 ifr
->ifr_flags
= tun_flags(tun
);
1449 /* This is like a cut-down ethtool ops, except done via tun fd so no
1450 * privs required. */
1451 static int set_offload(struct tun_struct
*tun
, unsigned long arg
)
1453 netdev_features_t features
= 0;
1455 if (arg
& TUN_F_CSUM
) {
1456 features
|= NETIF_F_HW_CSUM
;
1459 if (arg
& (TUN_F_TSO4
|TUN_F_TSO6
)) {
1460 if (arg
& TUN_F_TSO_ECN
) {
1461 features
|= NETIF_F_TSO_ECN
;
1462 arg
&= ~TUN_F_TSO_ECN
;
1464 if (arg
& TUN_F_TSO4
)
1465 features
|= NETIF_F_TSO
;
1466 if (arg
& TUN_F_TSO6
)
1467 features
|= NETIF_F_TSO6
;
1468 arg
&= ~(TUN_F_TSO4
|TUN_F_TSO6
);
1471 if (arg
& TUN_F_UFO
) {
1472 features
|= NETIF_F_UFO
;
1477 /* This gives the user a way to test for new features in future by
1478 * trying to set them. */
1482 tun
->set_features
= features
;
1483 netdev_update_features(tun
->dev
);
1488 static void tun_detach_filter(struct tun_struct
*tun
, int n
)
1491 struct tun_file
*tfile
;
1493 for (i
= 0; i
< n
; i
++) {
1494 tfile
= rcu_dereference_protected(tun
->tfiles
[i
],
1495 lockdep_rtnl_is_held());
1496 sk_detach_filter(tfile
->socket
.sk
);
1499 tun
->filter_attached
= false;
1502 static int tun_attach_filter(struct tun_struct
*tun
)
1505 struct tun_file
*tfile
;
1507 for (i
= 0; i
< tun
->numqueues
; i
++) {
1508 tfile
= rcu_dereference_protected(tun
->tfiles
[i
],
1509 lockdep_rtnl_is_held());
1510 ret
= sk_attach_filter(&tun
->fprog
, tfile
->socket
.sk
);
1512 tun_detach_filter(tun
, i
);
1517 tun
->filter_attached
= true;
1521 static void tun_set_sndbuf(struct tun_struct
*tun
)
1523 struct tun_file
*tfile
;
1526 for (i
= 0; i
< tun
->numqueues
; i
++) {
1527 tfile
= rcu_dereference_protected(tun
->tfiles
[i
],
1528 lockdep_rtnl_is_held());
1529 tfile
->socket
.sk
->sk_sndbuf
= tun
->sndbuf
;
1533 static long __tun_chr_ioctl(struct file
*file
, unsigned int cmd
,
1534 unsigned long arg
, int ifreq_len
)
1536 struct tun_file
*tfile
= file
->private_data
;
1537 struct tun_struct
*tun
;
1538 void __user
* argp
= (void __user
*)arg
;
1546 if (cmd
== TUNSETIFF
|| _IOC_TYPE(cmd
) == 0x89) {
1547 if (copy_from_user(&ifr
, argp
, ifreq_len
))
1550 memset(&ifr
, 0, sizeof(ifr
));
1552 if (cmd
== TUNGETFEATURES
) {
1553 /* Currently this just means: "what IFF flags are valid?".
1554 * This is needed because we never checked for invalid flags on
1556 return put_user(IFF_TUN
| IFF_TAP
| IFF_NO_PI
| IFF_ONE_QUEUE
|
1558 (unsigned int __user
*)argp
);
1564 tun
= __tun_get(tfile
);
1565 if (cmd
== TUNSETIFF
&& !tun
) {
1566 ifr
.ifr_name
[IFNAMSIZ
-1] = '\0';
1568 ret
= tun_set_iff(tfile
->net
, file
, &ifr
);
1573 if (copy_to_user(argp
, &ifr
, ifreq_len
))
1582 tun_debug(KERN_INFO
, tun
, "tun_chr_ioctl cmd %u\n", cmd
);
1587 ret
= tun_get_iff(current
->nsproxy
->net_ns
, tun
, &ifr
);
1591 if (copy_to_user(argp
, &ifr
, ifreq_len
))
1596 /* Disable/Enable checksum */
1598 /* [unimplemented] */
1599 tun_debug(KERN_INFO
, tun
, "ignored: set checksum %s\n",
1600 arg
? "disabled" : "enabled");
1604 /* Disable/Enable persist mode. Keep an extra reference to the
1605 * module to prevent the module being unprobed.
1608 tun
->flags
|= TUN_PERSIST
;
1609 __module_get(THIS_MODULE
);
1611 tun
->flags
&= ~TUN_PERSIST
;
1612 module_put(THIS_MODULE
);
1615 tun_debug(KERN_INFO
, tun
, "persist %s\n",
1616 arg
? "enabled" : "disabled");
1620 /* Set owner of the device */
1621 owner
= make_kuid(current_user_ns(), arg
);
1622 if (!uid_valid(owner
)) {
1627 tun_debug(KERN_INFO
, tun
, "owner set to %u\n",
1628 from_kuid(&init_user_ns
, tun
->owner
));
1632 /* Set group of the device */
1633 group
= make_kgid(current_user_ns(), arg
);
1634 if (!gid_valid(group
)) {
1639 tun_debug(KERN_INFO
, tun
, "group set to %u\n",
1640 from_kgid(&init_user_ns
, tun
->group
));
1644 /* Only allow setting the type when the interface is down */
1645 if (tun
->dev
->flags
& IFF_UP
) {
1646 tun_debug(KERN_INFO
, tun
,
1647 "Linktype set failed because interface is up\n");
1650 tun
->dev
->type
= (int) arg
;
1651 tun_debug(KERN_INFO
, tun
, "linktype set to %d\n",
1663 ret
= set_offload(tun
, arg
);
1666 case TUNSETTXFILTER
:
1667 /* Can be set only for TAPs */
1669 if ((tun
->flags
& TUN_TYPE_MASK
) != TUN_TAP_DEV
)
1671 ret
= update_filter(&tun
->txflt
, (void __user
*)arg
);
1675 /* Get hw address */
1676 memcpy(ifr
.ifr_hwaddr
.sa_data
, tun
->dev
->dev_addr
, ETH_ALEN
);
1677 ifr
.ifr_hwaddr
.sa_family
= tun
->dev
->type
;
1678 if (copy_to_user(argp
, &ifr
, ifreq_len
))
1683 /* Set hw address */
1684 tun_debug(KERN_DEBUG
, tun
, "set hw address: %pM\n",
1685 ifr
.ifr_hwaddr
.sa_data
);
1687 ret
= dev_set_mac_address(tun
->dev
, &ifr
.ifr_hwaddr
);
1691 sndbuf
= tfile
->socket
.sk
->sk_sndbuf
;
1692 if (copy_to_user(argp
, &sndbuf
, sizeof(sndbuf
)))
1697 if (copy_from_user(&sndbuf
, argp
, sizeof(sndbuf
))) {
1702 tun
->sndbuf
= sndbuf
;
1703 tun_set_sndbuf(tun
);
1706 case TUNGETVNETHDRSZ
:
1707 vnet_hdr_sz
= tun
->vnet_hdr_sz
;
1708 if (copy_to_user(argp
, &vnet_hdr_sz
, sizeof(vnet_hdr_sz
)))
1712 case TUNSETVNETHDRSZ
:
1713 if (copy_from_user(&vnet_hdr_sz
, argp
, sizeof(vnet_hdr_sz
))) {
1717 if (vnet_hdr_sz
< (int)sizeof(struct virtio_net_hdr
)) {
1722 tun
->vnet_hdr_sz
= vnet_hdr_sz
;
1725 case TUNATTACHFILTER
:
1726 /* Can be set only for TAPs */
1728 if ((tun
->flags
& TUN_TYPE_MASK
) != TUN_TAP_DEV
)
1731 if (copy_from_user(&tun
->fprog
, argp
, sizeof(tun
->fprog
)))
1734 ret
= tun_attach_filter(tun
);
1737 case TUNDETACHFILTER
:
1738 /* Can be set only for TAPs */
1740 if ((tun
->flags
& TUN_TYPE_MASK
) != TUN_TAP_DEV
)
1743 tun_detach_filter(tun
, tun
->numqueues
);
1758 static long tun_chr_ioctl(struct file
*file
,
1759 unsigned int cmd
, unsigned long arg
)
1761 return __tun_chr_ioctl(file
, cmd
, arg
, sizeof (struct ifreq
));
1764 #ifdef CONFIG_COMPAT
1765 static long tun_chr_compat_ioctl(struct file
*file
,
1766 unsigned int cmd
, unsigned long arg
)
1771 case TUNSETTXFILTER
:
1776 arg
= (unsigned long)compat_ptr(arg
);
1779 arg
= (compat_ulong_t
)arg
;
1784 * compat_ifreq is shorter than ifreq, so we must not access beyond
1785 * the end of that structure. All fields that are used in this
1786 * driver are compatible though, we don't need to convert the
1789 return __tun_chr_ioctl(file
, cmd
, arg
, sizeof(struct compat_ifreq
));
1791 #endif /* CONFIG_COMPAT */
1793 static int tun_chr_fasync(int fd
, struct file
*file
, int on
)
1795 struct tun_file
*tfile
= file
->private_data
;
1798 if ((ret
= fasync_helper(fd
, file
, on
, &tfile
->fasync
)) < 0)
1802 ret
= __f_setown(file
, task_pid(current
), PIDTYPE_PID
, 0);
1805 tfile
->flags
|= TUN_FASYNC
;
1807 tfile
->flags
&= ~TUN_FASYNC
;
1813 static int tun_chr_open(struct inode
*inode
, struct file
* file
)
1815 struct tun_file
*tfile
;
1817 DBG1(KERN_INFO
, "tunX: tun_chr_open\n");
1819 tfile
= (struct tun_file
*)sk_alloc(&init_net
, AF_UNSPEC
, GFP_KERNEL
,
1823 rcu_assign_pointer(tfile
->tun
, NULL
);
1824 tfile
->net
= get_net(current
->nsproxy
->net_ns
);
1827 rcu_assign_pointer(tfile
->socket
.wq
, &tfile
->wq
);
1828 init_waitqueue_head(&tfile
->wq
.wait
);
1830 tfile
->socket
.file
= file
;
1831 tfile
->socket
.ops
= &tun_socket_ops
;
1833 sock_init_data(&tfile
->socket
, &tfile
->sk
);
1834 sk_change_net(&tfile
->sk
, tfile
->net
);
1836 tfile
->sk
.sk_write_space
= tun_sock_write_space
;
1837 tfile
->sk
.sk_sndbuf
= INT_MAX
;
1839 file
->private_data
= tfile
;
1840 set_bit(SOCK_EXTERNALLY_ALLOCATED
, &tfile
->socket
.flags
);
1845 static int tun_chr_close(struct inode
*inode
, struct file
*file
)
1847 struct tun_file
*tfile
= file
->private_data
;
1848 struct net
*net
= tfile
->net
;
1850 tun_detach(tfile
, true);
1856 static const struct file_operations tun_fops
= {
1857 .owner
= THIS_MODULE
,
1858 .llseek
= no_llseek
,
1859 .read
= do_sync_read
,
1860 .aio_read
= tun_chr_aio_read
,
1861 .write
= do_sync_write
,
1862 .aio_write
= tun_chr_aio_write
,
1863 .poll
= tun_chr_poll
,
1864 .unlocked_ioctl
= tun_chr_ioctl
,
1865 #ifdef CONFIG_COMPAT
1866 .compat_ioctl
= tun_chr_compat_ioctl
,
1868 .open
= tun_chr_open
,
1869 .release
= tun_chr_close
,
1870 .fasync
= tun_chr_fasync
1873 static struct miscdevice tun_miscdev
= {
1876 .nodename
= "net/tun",
1880 /* ethtool interface */
1882 static int tun_get_settings(struct net_device
*dev
, struct ethtool_cmd
*cmd
)
1885 cmd
->advertising
= 0;
1886 ethtool_cmd_speed_set(cmd
, SPEED_10
);
1887 cmd
->duplex
= DUPLEX_FULL
;
1888 cmd
->port
= PORT_TP
;
1889 cmd
->phy_address
= 0;
1890 cmd
->transceiver
= XCVR_INTERNAL
;
1891 cmd
->autoneg
= AUTONEG_DISABLE
;
1897 static void tun_get_drvinfo(struct net_device
*dev
, struct ethtool_drvinfo
*info
)
1899 struct tun_struct
*tun
= netdev_priv(dev
);
1901 strlcpy(info
->driver
, DRV_NAME
, sizeof(info
->driver
));
1902 strlcpy(info
->version
, DRV_VERSION
, sizeof(info
->version
));
1904 switch (tun
->flags
& TUN_TYPE_MASK
) {
1906 strlcpy(info
->bus_info
, "tun", sizeof(info
->bus_info
));
1909 strlcpy(info
->bus_info
, "tap", sizeof(info
->bus_info
));
1914 static u32
tun_get_msglevel(struct net_device
*dev
)
1917 struct tun_struct
*tun
= netdev_priv(dev
);
1924 static void tun_set_msglevel(struct net_device
*dev
, u32 value
)
1927 struct tun_struct
*tun
= netdev_priv(dev
);
1932 static const struct ethtool_ops tun_ethtool_ops
= {
1933 .get_settings
= tun_get_settings
,
1934 .get_drvinfo
= tun_get_drvinfo
,
1935 .get_msglevel
= tun_get_msglevel
,
1936 .set_msglevel
= tun_set_msglevel
,
1937 .get_link
= ethtool_op_get_link
,
1941 static int __init
tun_init(void)
1945 pr_info("%s, %s\n", DRV_DESCRIPTION
, DRV_VERSION
);
1946 pr_info("%s\n", DRV_COPYRIGHT
);
1948 ret
= rtnl_link_register(&tun_link_ops
);
1950 pr_err("Can't register link_ops\n");
1954 ret
= misc_register(&tun_miscdev
);
1956 pr_err("Can't register misc device %d\n", TUN_MINOR
);
1961 rtnl_link_unregister(&tun_link_ops
);
1966 static void tun_cleanup(void)
1968 misc_deregister(&tun_miscdev
);
1969 rtnl_link_unregister(&tun_link_ops
);
1972 /* Get an underlying socket object from tun file. Returns error unless file is
1973 * attached to a device. The returned object works like a packet socket, it
1974 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
1975 * holding a reference to the file for as long as the socket is in use. */
1976 struct socket
*tun_get_socket(struct file
*file
)
1978 struct tun_file
*tfile
;
1979 if (file
->f_op
!= &tun_fops
)
1980 return ERR_PTR(-EINVAL
);
1981 tfile
= file
->private_data
;
1983 return ERR_PTR(-EBADFD
);
1984 return &tfile
->socket
;
1986 EXPORT_SYMBOL_GPL(tun_get_socket
);
1988 module_init(tun_init
);
1989 module_exit(tun_cleanup
);
1990 MODULE_DESCRIPTION(DRV_DESCRIPTION
);
1991 MODULE_AUTHOR(DRV_COPYRIGHT
);
1992 MODULE_LICENSE("GPL");
1993 MODULE_ALIAS_MISCDEV(TUN_MINOR
);
1994 MODULE_ALIAS("devname:net/tun");