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 #define TUN_FLOW_EXPIRE (3 * HZ)
120 /* A tun_file connects an open character device to a tuntap netdevice. It
121 * also contains all socket related strctures (except sock_fprog and tap_filter)
122 * to serve as one transmit queue for tuntap device. The sock_fprog and
123 * tap_filter were kept in tun_struct since they were used for filtering for the
124 * netdevice not for a specific queue (at least I didn't see the requirement for
128 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
129 * other can only be read while rcu_read_lock or rtnl_lock is held.
133 struct socket socket
;
135 struct tun_struct __rcu
*tun
;
137 struct fasync_struct
*fasync
;
138 /* only used for fasnyc */
143 struct tun_flow_entry
{
144 struct hlist_node hash_link
;
146 struct tun_struct
*tun
;
150 unsigned long updated
;
153 #define TUN_NUM_FLOW_ENTRIES 1024
155 /* Since the socket were moved to tun_file, to preserve the behavior of persist
156 * device, socket filter, sndbuf and vnet header size were restore when the
157 * file were attached to a persist device.
160 struct tun_file __rcu
*tfiles
[MAX_TAP_QUEUES
];
161 unsigned int numqueues
;
166 struct net_device
*dev
;
167 netdev_features_t set_features
;
168 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
169 NETIF_F_TSO6|NETIF_F_UFO)
173 struct tap_filter txflt
;
174 struct sock_fprog fprog
;
175 /* protected by rtnl lock */
176 bool filter_attached
;
181 struct kmem_cache
*flow_cache
;
182 struct hlist_head flows
[TUN_NUM_FLOW_ENTRIES
];
183 struct timer_list flow_gc_timer
;
184 unsigned long ageing_time
;
187 static inline u32
tun_hashfn(u32 rxhash
)
189 return rxhash
& 0x3ff;
192 static struct tun_flow_entry
*tun_flow_find(struct hlist_head
*head
, u32 rxhash
)
194 struct tun_flow_entry
*e
;
195 struct hlist_node
*n
;
197 hlist_for_each_entry_rcu(e
, n
, head
, hash_link
) {
198 if (e
->rxhash
== rxhash
)
204 static struct tun_flow_entry
*tun_flow_create(struct tun_struct
*tun
,
205 struct hlist_head
*head
,
206 u32 rxhash
, u16 queue_index
)
208 struct tun_flow_entry
*e
= kmem_cache_alloc(tun
->flow_cache
,
211 tun_debug(KERN_INFO
, tun
, "create flow: hash %u index %u\n",
212 rxhash
, queue_index
);
213 e
->updated
= jiffies
;
215 e
->queue_index
= queue_index
;
217 hlist_add_head_rcu(&e
->hash_link
, head
);
222 static void tun_flow_free(struct rcu_head
*head
)
224 struct tun_flow_entry
*e
225 = container_of(head
, struct tun_flow_entry
, rcu
);
226 kmem_cache_free(e
->tun
->flow_cache
, e
);
229 static void tun_flow_delete(struct tun_struct
*tun
, struct tun_flow_entry
*e
)
231 tun_debug(KERN_INFO
, tun
, "delete flow: hash %u index %u\n",
232 e
->rxhash
, e
->queue_index
);
233 hlist_del_rcu(&e
->hash_link
);
234 call_rcu(&e
->rcu
, tun_flow_free
);
237 static void tun_flow_flush(struct tun_struct
*tun
)
241 spin_lock_bh(&tun
->lock
);
242 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
243 struct tun_flow_entry
*e
;
244 struct hlist_node
*h
, *n
;
246 hlist_for_each_entry_safe(e
, h
, n
, &tun
->flows
[i
], hash_link
)
247 tun_flow_delete(tun
, e
);
249 spin_unlock_bh(&tun
->lock
);
252 static void tun_flow_delete_by_queue(struct tun_struct
*tun
, u16 queue_index
)
256 spin_lock_bh(&tun
->lock
);
257 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
258 struct tun_flow_entry
*e
;
259 struct hlist_node
*h
, *n
;
261 hlist_for_each_entry_safe(e
, h
, n
, &tun
->flows
[i
], hash_link
) {
262 if (e
->queue_index
== queue_index
)
263 tun_flow_delete(tun
, e
);
266 spin_unlock_bh(&tun
->lock
);
269 static void tun_flow_cleanup(unsigned long data
)
271 struct tun_struct
*tun
= (struct tun_struct
*)data
;
272 unsigned long delay
= tun
->ageing_time
;
273 unsigned long next_timer
= jiffies
+ delay
;
274 unsigned long count
= 0;
277 tun_debug(KERN_INFO
, tun
, "tun_flow_cleanup\n");
279 spin_lock_bh(&tun
->lock
);
280 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
281 struct tun_flow_entry
*e
;
282 struct hlist_node
*h
, *n
;
284 hlist_for_each_entry_safe(e
, h
, n
, &tun
->flows
[i
], hash_link
) {
285 unsigned long this_timer
;
287 this_timer
= e
->updated
+ delay
;
288 if (time_before_eq(this_timer
, jiffies
))
289 tun_flow_delete(tun
, e
);
290 else if (time_before(this_timer
, next_timer
))
291 next_timer
= this_timer
;
296 mod_timer(&tun
->flow_gc_timer
, round_jiffies_up(next_timer
));
297 spin_unlock_bh(&tun
->lock
);
300 static void tun_flow_update(struct tun_struct
*tun
, struct sk_buff
*skb
,
303 struct hlist_head
*head
;
304 struct tun_flow_entry
*e
;
305 unsigned long delay
= tun
->ageing_time
;
306 u32 rxhash
= skb_get_rxhash(skb
);
311 head
= &tun
->flows
[tun_hashfn(rxhash
)];
315 if (tun
->numqueues
== 1)
318 e
= tun_flow_find(head
, rxhash
);
320 /* TODO: keep queueing to old queue until it's empty? */
321 e
->queue_index
= queue_index
;
322 e
->updated
= jiffies
;
324 spin_lock_bh(&tun
->lock
);
325 if (!tun_flow_find(head
, rxhash
))
326 tun_flow_create(tun
, head
, rxhash
, queue_index
);
328 if (!timer_pending(&tun
->flow_gc_timer
))
329 mod_timer(&tun
->flow_gc_timer
,
330 round_jiffies_up(jiffies
+ delay
));
331 spin_unlock_bh(&tun
->lock
);
338 /* We try to identify a flow through its rxhash first. The reason that
339 * we do not check rxq no. is becuase some cards(e.g 82599), chooses
340 * the rxq based on the txq where the last packet of the flow comes. As
341 * the userspace application move between processors, we may get a
342 * different rxq no. here. If we could not get rxhash, then we would
343 * hope the rxq no. may help here.
345 static u16
tun_select_queue(struct net_device
*dev
, struct sk_buff
*skb
)
347 struct tun_struct
*tun
= netdev_priv(dev
);
348 struct tun_flow_entry
*e
;
353 numqueues
= tun
->numqueues
;
355 txq
= skb_get_rxhash(skb
);
357 e
= tun_flow_find(&tun
->flows
[tun_hashfn(txq
)], txq
);
359 txq
= e
->queue_index
;
361 /* use multiply and shift instead of expensive divide */
362 txq
= ((u64
)txq
* numqueues
) >> 32;
363 } else if (likely(skb_rx_queue_recorded(skb
))) {
364 txq
= skb_get_rx_queue(skb
);
365 while (unlikely(txq
>= numqueues
))
373 static inline bool tun_not_capable(struct tun_struct
*tun
)
375 const struct cred
*cred
= current_cred();
376 struct net
*net
= dev_net(tun
->dev
);
378 return ((uid_valid(tun
->owner
) && !uid_eq(cred
->euid
, tun
->owner
)) ||
379 (gid_valid(tun
->group
) && !in_egroup_p(tun
->group
))) &&
380 !ns_capable(net
->user_ns
, CAP_NET_ADMIN
);
383 static void tun_set_real_num_queues(struct tun_struct
*tun
)
385 netif_set_real_num_tx_queues(tun
->dev
, tun
->numqueues
);
386 netif_set_real_num_rx_queues(tun
->dev
, tun
->numqueues
);
389 static void __tun_detach(struct tun_file
*tfile
, bool clean
)
391 struct tun_file
*ntfile
;
392 struct tun_struct
*tun
;
393 struct net_device
*dev
;
395 tun
= rcu_dereference_protected(tfile
->tun
,
396 lockdep_rtnl_is_held());
398 u16 index
= tfile
->queue_index
;
399 BUG_ON(index
>= tun
->numqueues
);
402 rcu_assign_pointer(tun
->tfiles
[index
],
403 tun
->tfiles
[tun
->numqueues
- 1]);
404 rcu_assign_pointer(tfile
->tun
, NULL
);
405 ntfile
= rcu_dereference_protected(tun
->tfiles
[index
],
406 lockdep_rtnl_is_held());
407 ntfile
->queue_index
= index
;
410 sock_put(&tfile
->sk
);
413 tun_flow_delete_by_queue(tun
, tun
->numqueues
+ 1);
414 /* Drop read queue */
415 skb_queue_purge(&tfile
->sk
.sk_receive_queue
);
416 tun_set_real_num_queues(tun
);
418 if (tun
->numqueues
== 0 && !(tun
->flags
& TUN_PERSIST
))
419 if (dev
->reg_state
== NETREG_REGISTERED
)
420 unregister_netdevice(dev
);
424 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED
,
425 &tfile
->socket
.flags
));
426 sk_release_kernel(&tfile
->sk
);
430 static void tun_detach(struct tun_file
*tfile
, bool clean
)
433 __tun_detach(tfile
, clean
);
437 static void tun_detach_all(struct net_device
*dev
)
439 struct tun_struct
*tun
= netdev_priv(dev
);
440 struct tun_file
*tfile
;
441 int i
, n
= tun
->numqueues
;
443 for (i
= 0; i
< n
; i
++) {
444 tfile
= rcu_dereference_protected(tun
->tfiles
[i
],
445 lockdep_rtnl_is_held());
447 wake_up_all(&tfile
->wq
.wait
);
448 rcu_assign_pointer(tfile
->tun
, NULL
);
451 BUG_ON(tun
->numqueues
!= 0);
454 for (i
= 0; i
< n
; i
++) {
455 tfile
= rcu_dereference_protected(tun
->tfiles
[i
],
456 lockdep_rtnl_is_held());
457 /* Drop read queue */
458 skb_queue_purge(&tfile
->sk
.sk_receive_queue
);
459 sock_put(&tfile
->sk
);
463 static int tun_attach(struct tun_struct
*tun
, struct file
*file
)
465 struct tun_file
*tfile
= file
->private_data
;
469 if (rcu_dereference_protected(tfile
->tun
, lockdep_rtnl_is_held()))
473 if (!(tun
->flags
& TUN_TAP_MQ
) && tun
->numqueues
== 1)
477 if (tun
->numqueues
== MAX_TAP_QUEUES
)
482 /* Re-attach the filter to presist device */
483 if (tun
->filter_attached
== true) {
484 err
= sk_attach_filter(&tun
->fprog
, tfile
->socket
.sk
);
488 tfile
->queue_index
= tun
->numqueues
;
489 rcu_assign_pointer(tfile
->tun
, tun
);
490 rcu_assign_pointer(tun
->tfiles
[tun
->numqueues
], tfile
);
491 sock_hold(&tfile
->sk
);
494 tun_set_real_num_queues(tun
);
496 if (tun
->numqueues
== 1)
497 netif_carrier_on(tun
->dev
);
499 /* device is allowed to go away first, so no need to hold extra
507 static struct tun_struct
*__tun_get(struct tun_file
*tfile
)
509 struct tun_struct
*tun
;
512 tun
= rcu_dereference(tfile
->tun
);
520 static struct tun_struct
*tun_get(struct file
*file
)
522 return __tun_get(file
->private_data
);
525 static void tun_put(struct tun_struct
*tun
)
531 static void addr_hash_set(u32
*mask
, const u8
*addr
)
533 int n
= ether_crc(ETH_ALEN
, addr
) >> 26;
534 mask
[n
>> 5] |= (1 << (n
& 31));
537 static unsigned int addr_hash_test(const u32
*mask
, const u8
*addr
)
539 int n
= ether_crc(ETH_ALEN
, addr
) >> 26;
540 return mask
[n
>> 5] & (1 << (n
& 31));
543 static int update_filter(struct tap_filter
*filter
, void __user
*arg
)
545 struct { u8 u
[ETH_ALEN
]; } *addr
;
546 struct tun_filter uf
;
547 int err
, alen
, n
, nexact
;
549 if (copy_from_user(&uf
, arg
, sizeof(uf
)))
558 alen
= ETH_ALEN
* uf
.count
;
559 addr
= kmalloc(alen
, GFP_KERNEL
);
563 if (copy_from_user(addr
, arg
+ sizeof(uf
), alen
)) {
568 /* The filter is updated without holding any locks. Which is
569 * perfectly safe. We disable it first and in the worst
570 * case we'll accept a few undesired packets. */
574 /* Use first set of addresses as an exact filter */
575 for (n
= 0; n
< uf
.count
&& n
< FLT_EXACT_COUNT
; n
++)
576 memcpy(filter
->addr
[n
], addr
[n
].u
, ETH_ALEN
);
580 /* Remaining multicast addresses are hashed,
581 * unicast will leave the filter disabled. */
582 memset(filter
->mask
, 0, sizeof(filter
->mask
));
583 for (; n
< uf
.count
; n
++) {
584 if (!is_multicast_ether_addr(addr
[n
].u
)) {
585 err
= 0; /* no filter */
588 addr_hash_set(filter
->mask
, addr
[n
].u
);
591 /* For ALLMULTI just set the mask to all ones.
592 * This overrides the mask populated above. */
593 if ((uf
.flags
& TUN_FLT_ALLMULTI
))
594 memset(filter
->mask
, ~0, sizeof(filter
->mask
));
596 /* Now enable the filter */
598 filter
->count
= nexact
;
600 /* Return the number of exact filters */
608 /* Returns: 0 - drop, !=0 - accept */
609 static int run_filter(struct tap_filter
*filter
, const struct sk_buff
*skb
)
611 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
613 struct ethhdr
*eh
= (struct ethhdr
*) skb
->data
;
617 for (i
= 0; i
< filter
->count
; i
++)
618 if (ether_addr_equal(eh
->h_dest
, filter
->addr
[i
]))
621 /* Inexact match (multicast only) */
622 if (is_multicast_ether_addr(eh
->h_dest
))
623 return addr_hash_test(filter
->mask
, eh
->h_dest
);
629 * Checks whether the packet is accepted or not.
630 * Returns: 0 - drop, !=0 - accept
632 static int check_filter(struct tap_filter
*filter
, const struct sk_buff
*skb
)
637 return run_filter(filter
, skb
);
640 /* Network device part of the driver */
642 static const struct ethtool_ops tun_ethtool_ops
;
644 /* Net device detach from fd. */
645 static void tun_net_uninit(struct net_device
*dev
)
650 /* Net device open. */
651 static int tun_net_open(struct net_device
*dev
)
653 netif_tx_start_all_queues(dev
);
657 /* Net device close. */
658 static int tun_net_close(struct net_device
*dev
)
660 netif_tx_stop_all_queues(dev
);
664 /* Net device start xmit */
665 static netdev_tx_t
tun_net_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
667 struct tun_struct
*tun
= netdev_priv(dev
);
668 int txq
= skb
->queue_mapping
;
669 struct tun_file
*tfile
;
672 tfile
= rcu_dereference(tun
->tfiles
[txq
]);
674 /* Drop packet if interface is not attached */
675 if (txq
>= tun
->numqueues
)
678 tun_debug(KERN_INFO
, tun
, "tun_net_xmit %d\n", skb
->len
);
682 /* Drop if the filter does not like it.
683 * This is a noop if the filter is disabled.
684 * Filter can be enabled only for the TAP devices. */
685 if (!check_filter(&tun
->txflt
, skb
))
688 if (tfile
->socket
.sk
->sk_filter
&&
689 sk_filter(tfile
->socket
.sk
, skb
))
692 /* Limit the number of packets queued by dividing txq length with the
695 if (skb_queue_len(&tfile
->socket
.sk
->sk_receive_queue
)
696 >= dev
->tx_queue_len
/ tun
->numqueues
){
697 if (!(tun
->flags
& TUN_ONE_QUEUE
)) {
698 /* Normal queueing mode. */
699 /* Packet scheduler handles dropping of further packets. */
700 netif_stop_subqueue(dev
, txq
);
702 /* We won't see all dropped packets individually, so overrun
703 * error is more appropriate. */
704 dev
->stats
.tx_fifo_errors
++;
706 /* Single queue mode.
707 * Driver handles dropping of all packets itself. */
712 /* Orphan the skb - required as we might hang on to it
713 * for indefinite time. */
714 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
719 skb_queue_tail(&tfile
->socket
.sk
->sk_receive_queue
, skb
);
721 /* Notify and wake up reader process */
722 if (tfile
->flags
& TUN_FASYNC
)
723 kill_fasync(&tfile
->fasync
, SIGIO
, POLL_IN
);
724 wake_up_interruptible_poll(&tfile
->wq
.wait
, POLLIN
|
725 POLLRDNORM
| POLLRDBAND
);
731 dev
->stats
.tx_dropped
++;
738 static void tun_net_mclist(struct net_device
*dev
)
741 * This callback is supposed to deal with mc filter in
742 * _rx_ path and has nothing to do with the _tx_ path.
743 * In rx path we always accept everything userspace gives us.
748 #define MAX_MTU 65535
751 tun_net_change_mtu(struct net_device
*dev
, int new_mtu
)
753 if (new_mtu
< MIN_MTU
|| new_mtu
+ dev
->hard_header_len
> MAX_MTU
)
759 static netdev_features_t
tun_net_fix_features(struct net_device
*dev
,
760 netdev_features_t features
)
762 struct tun_struct
*tun
= netdev_priv(dev
);
764 return (features
& tun
->set_features
) | (features
& ~TUN_USER_FEATURES
);
766 #ifdef CONFIG_NET_POLL_CONTROLLER
767 static void tun_poll_controller(struct net_device
*dev
)
770 * Tun only receives frames when:
771 * 1) the char device endpoint gets data from user space
772 * 2) the tun socket gets a sendmsg call from user space
773 * Since both of those are syncronous operations, we are guaranteed
774 * never to have pending data when we poll for it
775 * so theres nothing to do here but return.
776 * We need this though so netpoll recognizes us as an interface that
777 * supports polling, which enables bridge devices in virt setups to
778 * still use netconsole
783 static const struct net_device_ops tun_netdev_ops
= {
784 .ndo_uninit
= tun_net_uninit
,
785 .ndo_open
= tun_net_open
,
786 .ndo_stop
= tun_net_close
,
787 .ndo_start_xmit
= tun_net_xmit
,
788 .ndo_change_mtu
= tun_net_change_mtu
,
789 .ndo_fix_features
= tun_net_fix_features
,
790 .ndo_select_queue
= tun_select_queue
,
791 #ifdef CONFIG_NET_POLL_CONTROLLER
792 .ndo_poll_controller
= tun_poll_controller
,
796 static const struct net_device_ops tap_netdev_ops
= {
797 .ndo_uninit
= tun_net_uninit
,
798 .ndo_open
= tun_net_open
,
799 .ndo_stop
= tun_net_close
,
800 .ndo_start_xmit
= tun_net_xmit
,
801 .ndo_change_mtu
= tun_net_change_mtu
,
802 .ndo_fix_features
= tun_net_fix_features
,
803 .ndo_set_rx_mode
= tun_net_mclist
,
804 .ndo_set_mac_address
= eth_mac_addr
,
805 .ndo_validate_addr
= eth_validate_addr
,
806 .ndo_select_queue
= tun_select_queue
,
807 #ifdef CONFIG_NET_POLL_CONTROLLER
808 .ndo_poll_controller
= tun_poll_controller
,
812 static int tun_flow_init(struct tun_struct
*tun
)
816 tun
->flow_cache
= kmem_cache_create("tun_flow_cache",
817 sizeof(struct tun_flow_entry
), 0, 0,
819 if (!tun
->flow_cache
)
822 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++)
823 INIT_HLIST_HEAD(&tun
->flows
[i
]);
825 tun
->ageing_time
= TUN_FLOW_EXPIRE
;
826 setup_timer(&tun
->flow_gc_timer
, tun_flow_cleanup
, (unsigned long)tun
);
827 mod_timer(&tun
->flow_gc_timer
,
828 round_jiffies_up(jiffies
+ tun
->ageing_time
));
833 static void tun_flow_uninit(struct tun_struct
*tun
)
835 del_timer_sync(&tun
->flow_gc_timer
);
838 /* Wait for completion of call_rcu()'s */
840 kmem_cache_destroy(tun
->flow_cache
);
843 /* Initialize net device. */
844 static void tun_net_init(struct net_device
*dev
)
846 struct tun_struct
*tun
= netdev_priv(dev
);
848 switch (tun
->flags
& TUN_TYPE_MASK
) {
850 dev
->netdev_ops
= &tun_netdev_ops
;
852 /* Point-to-Point TUN Device */
853 dev
->hard_header_len
= 0;
857 /* Zero header length */
858 dev
->type
= ARPHRD_NONE
;
859 dev
->flags
= IFF_POINTOPOINT
| IFF_NOARP
| IFF_MULTICAST
;
860 dev
->tx_queue_len
= TUN_READQ_SIZE
; /* We prefer our own queue length */
864 dev
->netdev_ops
= &tap_netdev_ops
;
865 /* Ethernet TAP Device */
867 dev
->priv_flags
&= ~IFF_TX_SKB_SHARING
;
869 eth_hw_addr_random(dev
);
871 dev
->tx_queue_len
= TUN_READQ_SIZE
; /* We prefer our own queue length */
876 /* Character device part */
879 static unsigned int tun_chr_poll(struct file
*file
, poll_table
*wait
)
881 struct tun_file
*tfile
= file
->private_data
;
882 struct tun_struct
*tun
= __tun_get(tfile
);
884 unsigned int mask
= 0;
889 sk
= tfile
->socket
.sk
;
891 tun_debug(KERN_INFO
, tun
, "tun_chr_poll\n");
893 poll_wait(file
, &tfile
->wq
.wait
, wait
);
895 if (!skb_queue_empty(&sk
->sk_receive_queue
))
896 mask
|= POLLIN
| POLLRDNORM
;
898 if (sock_writeable(sk
) ||
899 (!test_and_set_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
) &&
901 mask
|= POLLOUT
| POLLWRNORM
;
903 if (tun
->dev
->reg_state
!= NETREG_REGISTERED
)
910 /* prepad is the amount to reserve at front. len is length after that.
911 * linear is a hint as to how much to copy (usually headers). */
912 static struct sk_buff
*tun_alloc_skb(struct tun_file
*tfile
,
913 size_t prepad
, size_t len
,
914 size_t linear
, int noblock
)
916 struct sock
*sk
= tfile
->socket
.sk
;
920 /* Under a page? Don't bother with paged skb. */
921 if (prepad
+ len
< PAGE_SIZE
|| !linear
)
924 skb
= sock_alloc_send_pskb(sk
, prepad
+ linear
, len
- linear
, noblock
,
929 skb_reserve(skb
, prepad
);
930 skb_put(skb
, linear
);
931 skb
->data_len
= len
- linear
;
932 skb
->len
+= len
- linear
;
937 /* set skb frags from iovec, this can move to core network code for reuse */
938 static int zerocopy_sg_from_iovec(struct sk_buff
*skb
, const struct iovec
*from
,
939 int offset
, size_t count
)
941 int len
= iov_length(from
, count
) - offset
;
942 int copy
= skb_headlen(skb
);
943 int size
, offset1
= 0;
946 /* Skip over from offset */
947 while (count
&& (offset
>= from
->iov_len
)) {
948 offset
-= from
->iov_len
;
953 /* copy up to skb headlen */
954 while (count
&& (copy
> 0)) {
955 size
= min_t(unsigned int, copy
, from
->iov_len
- offset
);
956 if (copy_from_user(skb
->data
+ offset1
, from
->iov_base
+ offset
,
973 struct page
*page
[MAX_SKB_FRAGS
];
976 unsigned long truesize
;
978 len
= from
->iov_len
- offset
;
984 base
= (unsigned long)from
->iov_base
+ offset
;
985 size
= ((base
& ~PAGE_MASK
) + len
+ ~PAGE_MASK
) >> PAGE_SHIFT
;
986 if (i
+ size
> MAX_SKB_FRAGS
)
988 num_pages
= get_user_pages_fast(base
, size
, 0, &page
[i
]);
989 if (num_pages
!= size
) {
990 for (i
= 0; i
< num_pages
; i
++)
994 truesize
= size
* PAGE_SIZE
;
995 skb
->data_len
+= len
;
997 skb
->truesize
+= truesize
;
998 atomic_add(truesize
, &skb
->sk
->sk_wmem_alloc
);
1000 int off
= base
& ~PAGE_MASK
;
1001 int size
= min_t(int, len
, PAGE_SIZE
- off
);
1002 __skb_fill_page_desc(skb
, i
, page
[i
], off
, size
);
1003 skb_shinfo(skb
)->nr_frags
++;
1004 /* increase sk_wmem_alloc */
1015 /* Get packet from user space buffer */
1016 static ssize_t
tun_get_user(struct tun_struct
*tun
, struct tun_file
*tfile
,
1017 void *msg_control
, const struct iovec
*iv
,
1018 size_t total_len
, size_t count
, int noblock
)
1020 struct tun_pi pi
= { 0, cpu_to_be16(ETH_P_IP
) };
1021 struct sk_buff
*skb
;
1022 size_t len
= total_len
, align
= NET_SKB_PAD
;
1023 struct virtio_net_hdr gso
= { 0 };
1026 bool zerocopy
= false;
1029 if (!(tun
->flags
& TUN_NO_PI
)) {
1030 if ((len
-= sizeof(pi
)) > total_len
)
1033 if (memcpy_fromiovecend((void *)&pi
, iv
, 0, sizeof(pi
)))
1035 offset
+= sizeof(pi
);
1038 if (tun
->flags
& TUN_VNET_HDR
) {
1039 if ((len
-= tun
->vnet_hdr_sz
) > total_len
)
1042 if (memcpy_fromiovecend((void *)&gso
, iv
, offset
, sizeof(gso
)))
1045 if ((gso
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) &&
1046 gso
.csum_start
+ gso
.csum_offset
+ 2 > gso
.hdr_len
)
1047 gso
.hdr_len
= gso
.csum_start
+ gso
.csum_offset
+ 2;
1049 if (gso
.hdr_len
> len
)
1051 offset
+= tun
->vnet_hdr_sz
;
1054 if ((tun
->flags
& TUN_TYPE_MASK
) == TUN_TAP_DEV
) {
1055 align
+= NET_IP_ALIGN
;
1056 if (unlikely(len
< ETH_HLEN
||
1057 (gso
.hdr_len
&& gso
.hdr_len
< ETH_HLEN
)))
1065 /* Userspace may produce vectors with count greater than
1066 * MAX_SKB_FRAGS, so we need to linearize parts of the skb
1067 * to let the rest of data to be fit in the frags.
1069 if (count
> MAX_SKB_FRAGS
) {
1070 copylen
= iov_length(iv
, count
- MAX_SKB_FRAGS
);
1071 if (copylen
< offset
)
1077 /* There are 256 bytes to be copied in skb, so there is enough
1078 * room for skb expand head in case it is used.
1079 * The rest of the buffer is mapped from userspace.
1081 if (copylen
< gso
.hdr_len
)
1082 copylen
= gso
.hdr_len
;
1084 copylen
= GOODCOPY_LEN
;
1088 skb
= tun_alloc_skb(tfile
, align
, copylen
, gso
.hdr_len
, noblock
);
1090 if (PTR_ERR(skb
) != -EAGAIN
)
1091 tun
->dev
->stats
.rx_dropped
++;
1092 return PTR_ERR(skb
);
1096 err
= zerocopy_sg_from_iovec(skb
, iv
, offset
, count
);
1098 err
= skb_copy_datagram_from_iovec(skb
, 0, iv
, offset
, len
);
1101 tun
->dev
->stats
.rx_dropped
++;
1106 if (gso
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) {
1107 if (!skb_partial_csum_set(skb
, gso
.csum_start
,
1109 tun
->dev
->stats
.rx_frame_errors
++;
1115 switch (tun
->flags
& TUN_TYPE_MASK
) {
1117 if (tun
->flags
& TUN_NO_PI
) {
1118 switch (skb
->data
[0] & 0xf0) {
1120 pi
.proto
= htons(ETH_P_IP
);
1123 pi
.proto
= htons(ETH_P_IPV6
);
1126 tun
->dev
->stats
.rx_dropped
++;
1132 skb_reset_mac_header(skb
);
1133 skb
->protocol
= pi
.proto
;
1134 skb
->dev
= tun
->dev
;
1137 skb
->protocol
= eth_type_trans(skb
, tun
->dev
);
1141 if (gso
.gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
1143 switch (gso
.gso_type
& ~VIRTIO_NET_HDR_GSO_ECN
) {
1144 case VIRTIO_NET_HDR_GSO_TCPV4
:
1145 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV4
;
1147 case VIRTIO_NET_HDR_GSO_TCPV6
:
1148 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV6
;
1150 case VIRTIO_NET_HDR_GSO_UDP
:
1151 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
1154 tun
->dev
->stats
.rx_frame_errors
++;
1159 if (gso
.gso_type
& VIRTIO_NET_HDR_GSO_ECN
)
1160 skb_shinfo(skb
)->gso_type
|= SKB_GSO_TCP_ECN
;
1162 skb_shinfo(skb
)->gso_size
= gso
.gso_size
;
1163 if (skb_shinfo(skb
)->gso_size
== 0) {
1164 tun
->dev
->stats
.rx_frame_errors
++;
1169 /* Header must be checked, and gso_segs computed. */
1170 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
1171 skb_shinfo(skb
)->gso_segs
= 0;
1174 /* copy skb_ubuf_info for callback when skb has no error */
1176 skb_shinfo(skb
)->destructor_arg
= msg_control
;
1177 skb_shinfo(skb
)->tx_flags
|= SKBTX_DEV_ZEROCOPY
;
1182 tun
->dev
->stats
.rx_packets
++;
1183 tun
->dev
->stats
.rx_bytes
+= len
;
1185 tun_flow_update(tun
, skb
, tfile
->queue_index
);
1189 static ssize_t
tun_chr_aio_write(struct kiocb
*iocb
, const struct iovec
*iv
,
1190 unsigned long count
, loff_t pos
)
1192 struct file
*file
= iocb
->ki_filp
;
1193 struct tun_struct
*tun
= tun_get(file
);
1194 struct tun_file
*tfile
= file
->private_data
;
1200 tun_debug(KERN_INFO
, tun
, "tun_chr_write %ld\n", count
);
1202 result
= tun_get_user(tun
, tfile
, NULL
, iv
, iov_length(iv
, count
),
1203 count
, file
->f_flags
& O_NONBLOCK
);
1209 /* Put packet to the user space buffer */
1210 static ssize_t
tun_put_user(struct tun_struct
*tun
,
1211 struct tun_file
*tfile
,
1212 struct sk_buff
*skb
,
1213 const struct iovec
*iv
, int len
)
1215 struct tun_pi pi
= { 0, skb
->protocol
};
1218 if (!(tun
->flags
& TUN_NO_PI
)) {
1219 if ((len
-= sizeof(pi
)) < 0)
1222 if (len
< skb
->len
) {
1223 /* Packet will be striped */
1224 pi
.flags
|= TUN_PKT_STRIP
;
1227 if (memcpy_toiovecend(iv
, (void *) &pi
, 0, sizeof(pi
)))
1229 total
+= sizeof(pi
);
1232 if (tun
->flags
& TUN_VNET_HDR
) {
1233 struct virtio_net_hdr gso
= { 0 }; /* no info leak */
1234 if ((len
-= tun
->vnet_hdr_sz
) < 0)
1237 if (skb_is_gso(skb
)) {
1238 struct skb_shared_info
*sinfo
= skb_shinfo(skb
);
1240 /* This is a hint as to how much should be linear. */
1241 gso
.hdr_len
= skb_headlen(skb
);
1242 gso
.gso_size
= sinfo
->gso_size
;
1243 if (sinfo
->gso_type
& SKB_GSO_TCPV4
)
1244 gso
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV4
;
1245 else if (sinfo
->gso_type
& SKB_GSO_TCPV6
)
1246 gso
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV6
;
1247 else if (sinfo
->gso_type
& SKB_GSO_UDP
)
1248 gso
.gso_type
= VIRTIO_NET_HDR_GSO_UDP
;
1250 pr_err("unexpected GSO type: "
1251 "0x%x, gso_size %d, hdr_len %d\n",
1252 sinfo
->gso_type
, gso
.gso_size
,
1254 print_hex_dump(KERN_ERR
, "tun: ",
1257 min((int)gso
.hdr_len
, 64), true);
1261 if (sinfo
->gso_type
& SKB_GSO_TCP_ECN
)
1262 gso
.gso_type
|= VIRTIO_NET_HDR_GSO_ECN
;
1264 gso
.gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
1266 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1267 gso
.flags
= VIRTIO_NET_HDR_F_NEEDS_CSUM
;
1268 gso
.csum_start
= skb_checksum_start_offset(skb
);
1269 gso
.csum_offset
= skb
->csum_offset
;
1270 } else if (skb
->ip_summed
== CHECKSUM_UNNECESSARY
) {
1271 gso
.flags
= VIRTIO_NET_HDR_F_DATA_VALID
;
1272 } /* else everything is zero */
1274 if (unlikely(memcpy_toiovecend(iv
, (void *)&gso
, total
,
1277 total
+= tun
->vnet_hdr_sz
;
1280 len
= min_t(int, skb
->len
, len
);
1282 skb_copy_datagram_const_iovec(skb
, 0, iv
, total
, len
);
1285 tun
->dev
->stats
.tx_packets
++;
1286 tun
->dev
->stats
.tx_bytes
+= len
;
1291 static ssize_t
tun_do_read(struct tun_struct
*tun
, struct tun_file
*tfile
,
1292 struct kiocb
*iocb
, const struct iovec
*iv
,
1293 ssize_t len
, int noblock
)
1295 DECLARE_WAITQUEUE(wait
, current
);
1296 struct sk_buff
*skb
;
1299 tun_debug(KERN_INFO
, tun
, "tun_do_read\n");
1301 if (unlikely(!noblock
))
1302 add_wait_queue(&tfile
->wq
.wait
, &wait
);
1304 current
->state
= TASK_INTERRUPTIBLE
;
1306 /* Read frames from the queue */
1307 if (!(skb
= skb_dequeue(&tfile
->socket
.sk
->sk_receive_queue
))) {
1312 if (signal_pending(current
)) {
1316 if (tun
->dev
->reg_state
!= NETREG_REGISTERED
) {
1321 /* Nothing to read, let's sleep */
1325 netif_wake_subqueue(tun
->dev
, tfile
->queue_index
);
1327 ret
= tun_put_user(tun
, tfile
, skb
, iv
, len
);
1332 current
->state
= TASK_RUNNING
;
1333 if (unlikely(!noblock
))
1334 remove_wait_queue(&tfile
->wq
.wait
, &wait
);
1339 static ssize_t
tun_chr_aio_read(struct kiocb
*iocb
, const struct iovec
*iv
,
1340 unsigned long count
, loff_t pos
)
1342 struct file
*file
= iocb
->ki_filp
;
1343 struct tun_file
*tfile
= file
->private_data
;
1344 struct tun_struct
*tun
= __tun_get(tfile
);
1349 len
= iov_length(iv
, count
);
1355 ret
= tun_do_read(tun
, tfile
, iocb
, iv
, len
,
1356 file
->f_flags
& O_NONBLOCK
);
1357 ret
= min_t(ssize_t
, ret
, len
);
1363 static void tun_free_netdev(struct net_device
*dev
)
1365 struct tun_struct
*tun
= netdev_priv(dev
);
1367 tun_flow_uninit(tun
);
1371 static void tun_setup(struct net_device
*dev
)
1373 struct tun_struct
*tun
= netdev_priv(dev
);
1375 tun
->owner
= INVALID_UID
;
1376 tun
->group
= INVALID_GID
;
1378 dev
->ethtool_ops
= &tun_ethtool_ops
;
1379 dev
->destructor
= tun_free_netdev
;
1382 /* Trivial set of netlink ops to allow deleting tun or tap
1383 * device with netlink.
1385 static int tun_validate(struct nlattr
*tb
[], struct nlattr
*data
[])
1390 static struct rtnl_link_ops tun_link_ops __read_mostly
= {
1392 .priv_size
= sizeof(struct tun_struct
),
1394 .validate
= tun_validate
,
1397 static void tun_sock_write_space(struct sock
*sk
)
1399 struct tun_file
*tfile
;
1400 wait_queue_head_t
*wqueue
;
1402 if (!sock_writeable(sk
))
1405 if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
))
1408 wqueue
= sk_sleep(sk
);
1409 if (wqueue
&& waitqueue_active(wqueue
))
1410 wake_up_interruptible_sync_poll(wqueue
, POLLOUT
|
1411 POLLWRNORM
| POLLWRBAND
);
1413 tfile
= container_of(sk
, struct tun_file
, sk
);
1414 kill_fasync(&tfile
->fasync
, SIGIO
, POLL_OUT
);
1417 static int tun_sendmsg(struct kiocb
*iocb
, struct socket
*sock
,
1418 struct msghdr
*m
, size_t total_len
)
1421 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
1422 struct tun_struct
*tun
= __tun_get(tfile
);
1426 ret
= tun_get_user(tun
, tfile
, m
->msg_control
, m
->msg_iov
, total_len
,
1427 m
->msg_iovlen
, m
->msg_flags
& MSG_DONTWAIT
);
1433 static int tun_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
1434 struct msghdr
*m
, size_t total_len
,
1437 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
1438 struct tun_struct
*tun
= __tun_get(tfile
);
1444 if (flags
& ~(MSG_DONTWAIT
|MSG_TRUNC
))
1446 ret
= tun_do_read(tun
, tfile
, iocb
, m
->msg_iov
, total_len
,
1447 flags
& MSG_DONTWAIT
);
1448 if (ret
> total_len
) {
1449 m
->msg_flags
|= MSG_TRUNC
;
1450 ret
= flags
& MSG_TRUNC
? ret
: total_len
;
1456 static int tun_release(struct socket
*sock
)
1463 /* Ops structure to mimic raw sockets with tun */
1464 static const struct proto_ops tun_socket_ops
= {
1465 .sendmsg
= tun_sendmsg
,
1466 .recvmsg
= tun_recvmsg
,
1467 .release
= tun_release
,
1470 static struct proto tun_proto
= {
1472 .owner
= THIS_MODULE
,
1473 .obj_size
= sizeof(struct tun_file
),
1476 static int tun_flags(struct tun_struct
*tun
)
1480 if (tun
->flags
& TUN_TUN_DEV
)
1485 if (tun
->flags
& TUN_NO_PI
)
1488 if (tun
->flags
& TUN_ONE_QUEUE
)
1489 flags
|= IFF_ONE_QUEUE
;
1491 if (tun
->flags
& TUN_VNET_HDR
)
1492 flags
|= IFF_VNET_HDR
;
1494 if (tun
->flags
& TUN_TAP_MQ
)
1495 flags
|= IFF_MULTI_QUEUE
;
1500 static ssize_t
tun_show_flags(struct device
*dev
, struct device_attribute
*attr
,
1503 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
1504 return sprintf(buf
, "0x%x\n", tun_flags(tun
));
1507 static ssize_t
tun_show_owner(struct device
*dev
, struct device_attribute
*attr
,
1510 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
1511 return uid_valid(tun
->owner
)?
1512 sprintf(buf
, "%u\n",
1513 from_kuid_munged(current_user_ns(), tun
->owner
)):
1514 sprintf(buf
, "-1\n");
1517 static ssize_t
tun_show_group(struct device
*dev
, struct device_attribute
*attr
,
1520 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
1521 return gid_valid(tun
->group
) ?
1522 sprintf(buf
, "%u\n",
1523 from_kgid_munged(current_user_ns(), tun
->group
)):
1524 sprintf(buf
, "-1\n");
1527 static DEVICE_ATTR(tun_flags
, 0444, tun_show_flags
, NULL
);
1528 static DEVICE_ATTR(owner
, 0444, tun_show_owner
, NULL
);
1529 static DEVICE_ATTR(group
, 0444, tun_show_group
, NULL
);
1531 static int tun_set_iff(struct net
*net
, struct file
*file
, struct ifreq
*ifr
)
1533 struct tun_struct
*tun
;
1534 struct tun_file
*tfile
= file
->private_data
;
1535 struct net_device
*dev
;
1538 dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
1540 if (ifr
->ifr_flags
& IFF_TUN_EXCL
)
1542 if ((ifr
->ifr_flags
& IFF_TUN
) && dev
->netdev_ops
== &tun_netdev_ops
)
1543 tun
= netdev_priv(dev
);
1544 else if ((ifr
->ifr_flags
& IFF_TAP
) && dev
->netdev_ops
== &tap_netdev_ops
)
1545 tun
= netdev_priv(dev
);
1549 if (tun_not_capable(tun
))
1551 err
= security_tun_dev_attach(tfile
->socket
.sk
);
1555 err
= tun_attach(tun
, file
);
1561 unsigned long flags
= 0;
1563 if (!ns_capable(net
->user_ns
, CAP_NET_ADMIN
))
1565 err
= security_tun_dev_create();
1570 if (ifr
->ifr_flags
& IFF_TUN
) {
1572 flags
|= TUN_TUN_DEV
;
1574 } else if (ifr
->ifr_flags
& IFF_TAP
) {
1576 flags
|= TUN_TAP_DEV
;
1582 name
= ifr
->ifr_name
;
1584 dev
= alloc_netdev_mqs(sizeof(struct tun_struct
), name
,
1586 MAX_TAP_QUEUES
, MAX_TAP_QUEUES
);
1590 dev_net_set(dev
, net
);
1591 dev
->rtnl_link_ops
= &tun_link_ops
;
1593 tun
= netdev_priv(dev
);
1596 tun
->txflt
.count
= 0;
1597 tun
->vnet_hdr_sz
= sizeof(struct virtio_net_hdr
);
1599 tun
->filter_attached
= false;
1600 tun
->sndbuf
= tfile
->socket
.sk
->sk_sndbuf
;
1602 spin_lock_init(&tun
->lock
);
1604 security_tun_dev_post_create(&tfile
->sk
);
1608 if (tun_flow_init(tun
))
1611 dev
->hw_features
= NETIF_F_SG
| NETIF_F_FRAGLIST
|
1613 dev
->features
= dev
->hw_features
;
1615 err
= register_netdevice(tun
->dev
);
1619 if (device_create_file(&tun
->dev
->dev
, &dev_attr_tun_flags
) ||
1620 device_create_file(&tun
->dev
->dev
, &dev_attr_owner
) ||
1621 device_create_file(&tun
->dev
->dev
, &dev_attr_group
))
1622 pr_err("Failed to create tun sysfs files\n");
1624 err
= tun_attach(tun
, file
);
1629 tun_debug(KERN_INFO
, tun
, "tun_set_iff\n");
1631 if (ifr
->ifr_flags
& IFF_NO_PI
)
1632 tun
->flags
|= TUN_NO_PI
;
1634 tun
->flags
&= ~TUN_NO_PI
;
1636 if (ifr
->ifr_flags
& IFF_ONE_QUEUE
)
1637 tun
->flags
|= TUN_ONE_QUEUE
;
1639 tun
->flags
&= ~TUN_ONE_QUEUE
;
1641 if (ifr
->ifr_flags
& IFF_VNET_HDR
)
1642 tun
->flags
|= TUN_VNET_HDR
;
1644 tun
->flags
&= ~TUN_VNET_HDR
;
1646 if (ifr
->ifr_flags
& IFF_MULTI_QUEUE
)
1647 tun
->flags
|= TUN_TAP_MQ
;
1649 tun
->flags
&= ~TUN_TAP_MQ
;
1651 /* Make sure persistent devices do not get stuck in
1654 if (netif_running(tun
->dev
))
1655 netif_tx_wake_all_queues(tun
->dev
);
1657 strcpy(ifr
->ifr_name
, tun
->dev
->name
);
1665 static void tun_get_iff(struct net
*net
, struct tun_struct
*tun
,
1668 tun_debug(KERN_INFO
, tun
, "tun_get_iff\n");
1670 strcpy(ifr
->ifr_name
, tun
->dev
->name
);
1672 ifr
->ifr_flags
= tun_flags(tun
);
1676 /* This is like a cut-down ethtool ops, except done via tun fd so no
1677 * privs required. */
1678 static int set_offload(struct tun_struct
*tun
, unsigned long arg
)
1680 netdev_features_t features
= 0;
1682 if (arg
& TUN_F_CSUM
) {
1683 features
|= NETIF_F_HW_CSUM
;
1686 if (arg
& (TUN_F_TSO4
|TUN_F_TSO6
)) {
1687 if (arg
& TUN_F_TSO_ECN
) {
1688 features
|= NETIF_F_TSO_ECN
;
1689 arg
&= ~TUN_F_TSO_ECN
;
1691 if (arg
& TUN_F_TSO4
)
1692 features
|= NETIF_F_TSO
;
1693 if (arg
& TUN_F_TSO6
)
1694 features
|= NETIF_F_TSO6
;
1695 arg
&= ~(TUN_F_TSO4
|TUN_F_TSO6
);
1698 if (arg
& TUN_F_UFO
) {
1699 features
|= NETIF_F_UFO
;
1704 /* This gives the user a way to test for new features in future by
1705 * trying to set them. */
1709 tun
->set_features
= features
;
1710 netdev_update_features(tun
->dev
);
1715 static void tun_detach_filter(struct tun_struct
*tun
, int n
)
1718 struct tun_file
*tfile
;
1720 for (i
= 0; i
< n
; i
++) {
1721 tfile
= rcu_dereference_protected(tun
->tfiles
[i
],
1722 lockdep_rtnl_is_held());
1723 sk_detach_filter(tfile
->socket
.sk
);
1726 tun
->filter_attached
= false;
1729 static int tun_attach_filter(struct tun_struct
*tun
)
1732 struct tun_file
*tfile
;
1734 for (i
= 0; i
< tun
->numqueues
; i
++) {
1735 tfile
= rcu_dereference_protected(tun
->tfiles
[i
],
1736 lockdep_rtnl_is_held());
1737 ret
= sk_attach_filter(&tun
->fprog
, tfile
->socket
.sk
);
1739 tun_detach_filter(tun
, i
);
1744 tun
->filter_attached
= true;
1748 static void tun_set_sndbuf(struct tun_struct
*tun
)
1750 struct tun_file
*tfile
;
1753 for (i
= 0; i
< tun
->numqueues
; i
++) {
1754 tfile
= rcu_dereference_protected(tun
->tfiles
[i
],
1755 lockdep_rtnl_is_held());
1756 tfile
->socket
.sk
->sk_sndbuf
= tun
->sndbuf
;
1760 static int tun_set_queue(struct file
*file
, struct ifreq
*ifr
)
1762 struct tun_file
*tfile
= file
->private_data
;
1763 struct tun_struct
*tun
;
1764 struct net_device
*dev
;
1769 if (ifr
->ifr_flags
& IFF_ATTACH_QUEUE
) {
1770 dev
= __dev_get_by_name(tfile
->net
, ifr
->ifr_name
);
1776 tun
= netdev_priv(dev
);
1777 if (dev
->netdev_ops
!= &tap_netdev_ops
&&
1778 dev
->netdev_ops
!= &tun_netdev_ops
)
1780 else if (tun_not_capable(tun
))
1783 ret
= tun_attach(tun
, file
);
1784 } else if (ifr
->ifr_flags
& IFF_DETACH_QUEUE
)
1785 __tun_detach(tfile
, false);
1794 static long __tun_chr_ioctl(struct file
*file
, unsigned int cmd
,
1795 unsigned long arg
, int ifreq_len
)
1797 struct tun_file
*tfile
= file
->private_data
;
1798 struct tun_struct
*tun
;
1799 void __user
* argp
= (void __user
*)arg
;
1807 if (cmd
== TUNSETIFF
|| cmd
== TUNSETQUEUE
|| _IOC_TYPE(cmd
) == 0x89) {
1808 if (copy_from_user(&ifr
, argp
, ifreq_len
))
1811 memset(&ifr
, 0, sizeof(ifr
));
1813 if (cmd
== TUNGETFEATURES
) {
1814 /* Currently this just means: "what IFF flags are valid?".
1815 * This is needed because we never checked for invalid flags on
1817 return put_user(IFF_TUN
| IFF_TAP
| IFF_NO_PI
| IFF_ONE_QUEUE
|
1818 IFF_VNET_HDR
| IFF_MULTI_QUEUE
,
1819 (unsigned int __user
*)argp
);
1820 } else if (cmd
== TUNSETQUEUE
)
1821 return tun_set_queue(file
, &ifr
);
1826 tun
= __tun_get(tfile
);
1827 if (cmd
== TUNSETIFF
&& !tun
) {
1828 ifr
.ifr_name
[IFNAMSIZ
-1] = '\0';
1830 ret
= tun_set_iff(tfile
->net
, file
, &ifr
);
1835 if (copy_to_user(argp
, &ifr
, ifreq_len
))
1844 tun_debug(KERN_INFO
, tun
, "tun_chr_ioctl cmd %u\n", cmd
);
1849 tun_get_iff(current
->nsproxy
->net_ns
, tun
, &ifr
);
1851 if (copy_to_user(argp
, &ifr
, ifreq_len
))
1856 /* Disable/Enable checksum */
1858 /* [unimplemented] */
1859 tun_debug(KERN_INFO
, tun
, "ignored: set checksum %s\n",
1860 arg
? "disabled" : "enabled");
1864 /* Disable/Enable persist mode. Keep an extra reference to the
1865 * module to prevent the module being unprobed.
1868 tun
->flags
|= TUN_PERSIST
;
1869 __module_get(THIS_MODULE
);
1871 tun
->flags
&= ~TUN_PERSIST
;
1872 module_put(THIS_MODULE
);
1875 tun_debug(KERN_INFO
, tun
, "persist %s\n",
1876 arg
? "enabled" : "disabled");
1880 /* Set owner of the device */
1881 owner
= make_kuid(current_user_ns(), arg
);
1882 if (!uid_valid(owner
)) {
1887 tun_debug(KERN_INFO
, tun
, "owner set to %u\n",
1888 from_kuid(&init_user_ns
, tun
->owner
));
1892 /* Set group of the device */
1893 group
= make_kgid(current_user_ns(), arg
);
1894 if (!gid_valid(group
)) {
1899 tun_debug(KERN_INFO
, tun
, "group set to %u\n",
1900 from_kgid(&init_user_ns
, tun
->group
));
1904 /* Only allow setting the type when the interface is down */
1905 if (tun
->dev
->flags
& IFF_UP
) {
1906 tun_debug(KERN_INFO
, tun
,
1907 "Linktype set failed because interface is up\n");
1910 tun
->dev
->type
= (int) arg
;
1911 tun_debug(KERN_INFO
, tun
, "linktype set to %d\n",
1923 ret
= set_offload(tun
, arg
);
1926 case TUNSETTXFILTER
:
1927 /* Can be set only for TAPs */
1929 if ((tun
->flags
& TUN_TYPE_MASK
) != TUN_TAP_DEV
)
1931 ret
= update_filter(&tun
->txflt
, (void __user
*)arg
);
1935 /* Get hw address */
1936 memcpy(ifr
.ifr_hwaddr
.sa_data
, tun
->dev
->dev_addr
, ETH_ALEN
);
1937 ifr
.ifr_hwaddr
.sa_family
= tun
->dev
->type
;
1938 if (copy_to_user(argp
, &ifr
, ifreq_len
))
1943 /* Set hw address */
1944 tun_debug(KERN_DEBUG
, tun
, "set hw address: %pM\n",
1945 ifr
.ifr_hwaddr
.sa_data
);
1947 ret
= dev_set_mac_address(tun
->dev
, &ifr
.ifr_hwaddr
);
1951 sndbuf
= tfile
->socket
.sk
->sk_sndbuf
;
1952 if (copy_to_user(argp
, &sndbuf
, sizeof(sndbuf
)))
1957 if (copy_from_user(&sndbuf
, argp
, sizeof(sndbuf
))) {
1962 tun
->sndbuf
= sndbuf
;
1963 tun_set_sndbuf(tun
);
1966 case TUNGETVNETHDRSZ
:
1967 vnet_hdr_sz
= tun
->vnet_hdr_sz
;
1968 if (copy_to_user(argp
, &vnet_hdr_sz
, sizeof(vnet_hdr_sz
)))
1972 case TUNSETVNETHDRSZ
:
1973 if (copy_from_user(&vnet_hdr_sz
, argp
, sizeof(vnet_hdr_sz
))) {
1977 if (vnet_hdr_sz
< (int)sizeof(struct virtio_net_hdr
)) {
1982 tun
->vnet_hdr_sz
= vnet_hdr_sz
;
1985 case TUNATTACHFILTER
:
1986 /* Can be set only for TAPs */
1988 if ((tun
->flags
& TUN_TYPE_MASK
) != TUN_TAP_DEV
)
1991 if (copy_from_user(&tun
->fprog
, argp
, sizeof(tun
->fprog
)))
1994 ret
= tun_attach_filter(tun
);
1997 case TUNDETACHFILTER
:
1998 /* Can be set only for TAPs */
2000 if ((tun
->flags
& TUN_TYPE_MASK
) != TUN_TAP_DEV
)
2003 tun_detach_filter(tun
, tun
->numqueues
);
2018 static long tun_chr_ioctl(struct file
*file
,
2019 unsigned int cmd
, unsigned long arg
)
2021 return __tun_chr_ioctl(file
, cmd
, arg
, sizeof (struct ifreq
));
2024 #ifdef CONFIG_COMPAT
2025 static long tun_chr_compat_ioctl(struct file
*file
,
2026 unsigned int cmd
, unsigned long arg
)
2031 case TUNSETTXFILTER
:
2036 arg
= (unsigned long)compat_ptr(arg
);
2039 arg
= (compat_ulong_t
)arg
;
2044 * compat_ifreq is shorter than ifreq, so we must not access beyond
2045 * the end of that structure. All fields that are used in this
2046 * driver are compatible though, we don't need to convert the
2049 return __tun_chr_ioctl(file
, cmd
, arg
, sizeof(struct compat_ifreq
));
2051 #endif /* CONFIG_COMPAT */
2053 static int tun_chr_fasync(int fd
, struct file
*file
, int on
)
2055 struct tun_file
*tfile
= file
->private_data
;
2058 if ((ret
= fasync_helper(fd
, file
, on
, &tfile
->fasync
)) < 0)
2062 ret
= __f_setown(file
, task_pid(current
), PIDTYPE_PID
, 0);
2065 tfile
->flags
|= TUN_FASYNC
;
2067 tfile
->flags
&= ~TUN_FASYNC
;
2073 static int tun_chr_open(struct inode
*inode
, struct file
* file
)
2075 struct tun_file
*tfile
;
2077 DBG1(KERN_INFO
, "tunX: tun_chr_open\n");
2079 tfile
= (struct tun_file
*)sk_alloc(&init_net
, AF_UNSPEC
, GFP_KERNEL
,
2083 rcu_assign_pointer(tfile
->tun
, NULL
);
2084 tfile
->net
= get_net(current
->nsproxy
->net_ns
);
2087 rcu_assign_pointer(tfile
->socket
.wq
, &tfile
->wq
);
2088 init_waitqueue_head(&tfile
->wq
.wait
);
2090 tfile
->socket
.file
= file
;
2091 tfile
->socket
.ops
= &tun_socket_ops
;
2093 sock_init_data(&tfile
->socket
, &tfile
->sk
);
2094 sk_change_net(&tfile
->sk
, tfile
->net
);
2096 tfile
->sk
.sk_write_space
= tun_sock_write_space
;
2097 tfile
->sk
.sk_sndbuf
= INT_MAX
;
2099 file
->private_data
= tfile
;
2100 set_bit(SOCK_EXTERNALLY_ALLOCATED
, &tfile
->socket
.flags
);
2105 static int tun_chr_close(struct inode
*inode
, struct file
*file
)
2107 struct tun_file
*tfile
= file
->private_data
;
2108 struct net
*net
= tfile
->net
;
2110 tun_detach(tfile
, true);
2116 static const struct file_operations tun_fops
= {
2117 .owner
= THIS_MODULE
,
2118 .llseek
= no_llseek
,
2119 .read
= do_sync_read
,
2120 .aio_read
= tun_chr_aio_read
,
2121 .write
= do_sync_write
,
2122 .aio_write
= tun_chr_aio_write
,
2123 .poll
= tun_chr_poll
,
2124 .unlocked_ioctl
= tun_chr_ioctl
,
2125 #ifdef CONFIG_COMPAT
2126 .compat_ioctl
= tun_chr_compat_ioctl
,
2128 .open
= tun_chr_open
,
2129 .release
= tun_chr_close
,
2130 .fasync
= tun_chr_fasync
2133 static struct miscdevice tun_miscdev
= {
2136 .nodename
= "net/tun",
2140 /* ethtool interface */
2142 static int tun_get_settings(struct net_device
*dev
, struct ethtool_cmd
*cmd
)
2145 cmd
->advertising
= 0;
2146 ethtool_cmd_speed_set(cmd
, SPEED_10
);
2147 cmd
->duplex
= DUPLEX_FULL
;
2148 cmd
->port
= PORT_TP
;
2149 cmd
->phy_address
= 0;
2150 cmd
->transceiver
= XCVR_INTERNAL
;
2151 cmd
->autoneg
= AUTONEG_DISABLE
;
2157 static void tun_get_drvinfo(struct net_device
*dev
, struct ethtool_drvinfo
*info
)
2159 struct tun_struct
*tun
= netdev_priv(dev
);
2161 strlcpy(info
->driver
, DRV_NAME
, sizeof(info
->driver
));
2162 strlcpy(info
->version
, DRV_VERSION
, sizeof(info
->version
));
2164 switch (tun
->flags
& TUN_TYPE_MASK
) {
2166 strlcpy(info
->bus_info
, "tun", sizeof(info
->bus_info
));
2169 strlcpy(info
->bus_info
, "tap", sizeof(info
->bus_info
));
2174 static u32
tun_get_msglevel(struct net_device
*dev
)
2177 struct tun_struct
*tun
= netdev_priv(dev
);
2184 static void tun_set_msglevel(struct net_device
*dev
, u32 value
)
2187 struct tun_struct
*tun
= netdev_priv(dev
);
2192 static const struct ethtool_ops tun_ethtool_ops
= {
2193 .get_settings
= tun_get_settings
,
2194 .get_drvinfo
= tun_get_drvinfo
,
2195 .get_msglevel
= tun_get_msglevel
,
2196 .set_msglevel
= tun_set_msglevel
,
2197 .get_link
= ethtool_op_get_link
,
2201 static int __init
tun_init(void)
2205 pr_info("%s, %s\n", DRV_DESCRIPTION
, DRV_VERSION
);
2206 pr_info("%s\n", DRV_COPYRIGHT
);
2208 ret
= rtnl_link_register(&tun_link_ops
);
2210 pr_err("Can't register link_ops\n");
2214 ret
= misc_register(&tun_miscdev
);
2216 pr_err("Can't register misc device %d\n", TUN_MINOR
);
2221 rtnl_link_unregister(&tun_link_ops
);
2226 static void tun_cleanup(void)
2228 misc_deregister(&tun_miscdev
);
2229 rtnl_link_unregister(&tun_link_ops
);
2232 /* Get an underlying socket object from tun file. Returns error unless file is
2233 * attached to a device. The returned object works like a packet socket, it
2234 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
2235 * holding a reference to the file for as long as the socket is in use. */
2236 struct socket
*tun_get_socket(struct file
*file
)
2238 struct tun_file
*tfile
;
2239 if (file
->f_op
!= &tun_fops
)
2240 return ERR_PTR(-EINVAL
);
2241 tfile
= file
->private_data
;
2243 return ERR_PTR(-EBADFD
);
2244 return &tfile
->socket
;
2246 EXPORT_SYMBOL_GPL(tun_get_socket
);
2248 module_init(tun_init
);
2249 module_exit(tun_cleanup
);
2250 MODULE_DESCRIPTION(DRV_DESCRIPTION
);
2251 MODULE_AUTHOR(DRV_COPYRIGHT
);
2252 MODULE_LICENSE("GPL");
2253 MODULE_ALIAS_MISCDEV(TUN_MINOR
);
2254 MODULE_ALIAS("devname:net/tun");