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 /* DEFAULT_MAX_NUM_RSS_QUEUES were choosed to let the rx/tx queues allocated for
113 * the netdevice to be fit in one page. So we can make sure the success of
114 * memory allocation. TODO: increase the limit. */
115 #define MAX_TAP_QUEUES DEFAULT_MAX_NUM_RSS_QUEUES
116 #define MAX_TAP_FLOWS 4096
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 */
141 struct list_head next
;
142 struct tun_struct
*detached
;
145 struct tun_flow_entry
{
146 struct hlist_node hash_link
;
148 struct tun_struct
*tun
;
152 unsigned long updated
;
155 #define TUN_NUM_FLOW_ENTRIES 1024
157 /* Since the socket were moved to tun_file, to preserve the behavior of persist
158 * device, socket filter, sndbuf and vnet header size were restore when the
159 * file were attached to a persist device.
162 struct tun_file __rcu
*tfiles
[MAX_TAP_QUEUES
];
163 unsigned int numqueues
;
168 struct net_device
*dev
;
169 netdev_features_t set_features
;
170 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
171 NETIF_F_TSO6|NETIF_F_UFO)
175 struct tap_filter txflt
;
176 struct sock_fprog fprog
;
177 /* protected by rtnl lock */
178 bool filter_attached
;
183 struct hlist_head flows
[TUN_NUM_FLOW_ENTRIES
];
184 struct timer_list flow_gc_timer
;
185 unsigned long ageing_time
;
186 unsigned int numdisabled
;
187 struct list_head disabled
;
192 static inline u32
tun_hashfn(u32 rxhash
)
194 return rxhash
& 0x3ff;
197 static struct tun_flow_entry
*tun_flow_find(struct hlist_head
*head
, u32 rxhash
)
199 struct tun_flow_entry
*e
;
201 hlist_for_each_entry_rcu(e
, head
, hash_link
) {
202 if (e
->rxhash
== rxhash
)
208 static struct tun_flow_entry
*tun_flow_create(struct tun_struct
*tun
,
209 struct hlist_head
*head
,
210 u32 rxhash
, u16 queue_index
)
212 struct tun_flow_entry
*e
= kmalloc(sizeof(*e
), GFP_ATOMIC
);
215 tun_debug(KERN_INFO
, tun
, "create flow: hash %u index %u\n",
216 rxhash
, queue_index
);
217 e
->updated
= jiffies
;
219 e
->queue_index
= queue_index
;
221 hlist_add_head_rcu(&e
->hash_link
, head
);
227 static void tun_flow_delete(struct tun_struct
*tun
, struct tun_flow_entry
*e
)
229 tun_debug(KERN_INFO
, tun
, "delete flow: hash %u index %u\n",
230 e
->rxhash
, e
->queue_index
);
231 hlist_del_rcu(&e
->hash_link
);
236 static void tun_flow_flush(struct tun_struct
*tun
)
240 spin_lock_bh(&tun
->lock
);
241 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
242 struct tun_flow_entry
*e
;
243 struct hlist_node
*n
;
245 hlist_for_each_entry_safe(e
, n
, &tun
->flows
[i
], hash_link
)
246 tun_flow_delete(tun
, e
);
248 spin_unlock_bh(&tun
->lock
);
251 static void tun_flow_delete_by_queue(struct tun_struct
*tun
, u16 queue_index
)
255 spin_lock_bh(&tun
->lock
);
256 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
257 struct tun_flow_entry
*e
;
258 struct hlist_node
*n
;
260 hlist_for_each_entry_safe(e
, n
, &tun
->flows
[i
], hash_link
) {
261 if (e
->queue_index
== queue_index
)
262 tun_flow_delete(tun
, e
);
265 spin_unlock_bh(&tun
->lock
);
268 static void tun_flow_cleanup(unsigned long data
)
270 struct tun_struct
*tun
= (struct tun_struct
*)data
;
271 unsigned long delay
= tun
->ageing_time
;
272 unsigned long next_timer
= jiffies
+ delay
;
273 unsigned long count
= 0;
276 tun_debug(KERN_INFO
, tun
, "tun_flow_cleanup\n");
278 spin_lock_bh(&tun
->lock
);
279 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
280 struct tun_flow_entry
*e
;
281 struct hlist_node
*n
;
283 hlist_for_each_entry_safe(e
, n
, &tun
->flows
[i
], hash_link
) {
284 unsigned long this_timer
;
286 this_timer
= e
->updated
+ delay
;
287 if (time_before_eq(this_timer
, jiffies
))
288 tun_flow_delete(tun
, e
);
289 else if (time_before(this_timer
, next_timer
))
290 next_timer
= this_timer
;
295 mod_timer(&tun
->flow_gc_timer
, round_jiffies_up(next_timer
));
296 spin_unlock_bh(&tun
->lock
);
299 static void tun_flow_update(struct tun_struct
*tun
, u32 rxhash
,
300 struct tun_file
*tfile
)
302 struct hlist_head
*head
;
303 struct tun_flow_entry
*e
;
304 unsigned long delay
= tun
->ageing_time
;
305 u16 queue_index
= tfile
->queue_index
;
310 head
= &tun
->flows
[tun_hashfn(rxhash
)];
314 /* We may get a very small possibility of OOO during switching, not
315 * worth to optimize.*/
316 if (tun
->numqueues
== 1 || tfile
->detached
)
319 e
= tun_flow_find(head
, rxhash
);
321 /* TODO: keep queueing to old queue until it's empty? */
322 e
->queue_index
= queue_index
;
323 e
->updated
= jiffies
;
325 spin_lock_bh(&tun
->lock
);
326 if (!tun_flow_find(head
, rxhash
) &&
327 tun
->flow_count
< MAX_TAP_FLOWS
)
328 tun_flow_create(tun
, head
, rxhash
, queue_index
);
330 if (!timer_pending(&tun
->flow_gc_timer
))
331 mod_timer(&tun
->flow_gc_timer
,
332 round_jiffies_up(jiffies
+ delay
));
333 spin_unlock_bh(&tun
->lock
);
340 /* We try to identify a flow through its rxhash first. The reason that
341 * we do not check rxq no. is becuase some cards(e.g 82599), chooses
342 * the rxq based on the txq where the last packet of the flow comes. As
343 * the userspace application move between processors, we may get a
344 * different rxq no. here. If we could not get rxhash, then we would
345 * hope the rxq no. may help here.
347 static u16
tun_select_queue(struct net_device
*dev
, struct sk_buff
*skb
)
349 struct tun_struct
*tun
= netdev_priv(dev
);
350 struct tun_flow_entry
*e
;
355 numqueues
= ACCESS_ONCE(tun
->numqueues
);
357 txq
= skb_get_rxhash(skb
);
359 e
= tun_flow_find(&tun
->flows
[tun_hashfn(txq
)], txq
);
361 txq
= e
->queue_index
;
363 /* use multiply and shift instead of expensive divide */
364 txq
= ((u64
)txq
* numqueues
) >> 32;
365 } else if (likely(skb_rx_queue_recorded(skb
))) {
366 txq
= skb_get_rx_queue(skb
);
367 while (unlikely(txq
>= numqueues
))
375 static inline bool tun_not_capable(struct tun_struct
*tun
)
377 const struct cred
*cred
= current_cred();
378 struct net
*net
= dev_net(tun
->dev
);
380 return ((uid_valid(tun
->owner
) && !uid_eq(cred
->euid
, tun
->owner
)) ||
381 (gid_valid(tun
->group
) && !in_egroup_p(tun
->group
))) &&
382 !ns_capable(net
->user_ns
, CAP_NET_ADMIN
);
385 static void tun_set_real_num_queues(struct tun_struct
*tun
)
387 netif_set_real_num_tx_queues(tun
->dev
, tun
->numqueues
);
388 netif_set_real_num_rx_queues(tun
->dev
, tun
->numqueues
);
391 static void tun_disable_queue(struct tun_struct
*tun
, struct tun_file
*tfile
)
393 tfile
->detached
= tun
;
394 list_add_tail(&tfile
->next
, &tun
->disabled
);
398 static struct tun_struct
*tun_enable_queue(struct tun_file
*tfile
)
400 struct tun_struct
*tun
= tfile
->detached
;
402 tfile
->detached
= NULL
;
403 list_del_init(&tfile
->next
);
408 static void __tun_detach(struct tun_file
*tfile
, bool clean
)
410 struct tun_file
*ntfile
;
411 struct tun_struct
*tun
;
413 tun
= rtnl_dereference(tfile
->tun
);
415 if (tun
&& !tfile
->detached
) {
416 u16 index
= tfile
->queue_index
;
417 BUG_ON(index
>= tun
->numqueues
);
419 rcu_assign_pointer(tun
->tfiles
[index
],
420 tun
->tfiles
[tun
->numqueues
- 1]);
421 ntfile
= rtnl_dereference(tun
->tfiles
[index
]);
422 ntfile
->queue_index
= index
;
426 rcu_assign_pointer(tfile
->tun
, NULL
);
427 sock_put(&tfile
->sk
);
429 tun_disable_queue(tun
, tfile
);
432 tun_flow_delete_by_queue(tun
, tun
->numqueues
+ 1);
433 /* Drop read queue */
434 skb_queue_purge(&tfile
->sk
.sk_receive_queue
);
435 tun_set_real_num_queues(tun
);
436 } else if (tfile
->detached
&& clean
) {
437 tun
= tun_enable_queue(tfile
);
438 sock_put(&tfile
->sk
);
442 if (tun
&& tun
->numqueues
== 0 && tun
->numdisabled
== 0) {
443 netif_carrier_off(tun
->dev
);
445 if (!(tun
->flags
& TUN_PERSIST
) &&
446 tun
->dev
->reg_state
== NETREG_REGISTERED
)
447 unregister_netdevice(tun
->dev
);
450 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED
,
451 &tfile
->socket
.flags
));
452 sk_release_kernel(&tfile
->sk
);
456 static void tun_detach(struct tun_file
*tfile
, bool clean
)
459 __tun_detach(tfile
, clean
);
463 static void tun_detach_all(struct net_device
*dev
)
465 struct tun_struct
*tun
= netdev_priv(dev
);
466 struct tun_file
*tfile
, *tmp
;
467 int i
, n
= tun
->numqueues
;
469 for (i
= 0; i
< n
; i
++) {
470 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
472 wake_up_all(&tfile
->wq
.wait
);
473 rcu_assign_pointer(tfile
->tun
, NULL
);
476 list_for_each_entry(tfile
, &tun
->disabled
, next
) {
477 wake_up_all(&tfile
->wq
.wait
);
478 rcu_assign_pointer(tfile
->tun
, NULL
);
480 BUG_ON(tun
->numqueues
!= 0);
483 for (i
= 0; i
< n
; i
++) {
484 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
485 /* Drop read queue */
486 skb_queue_purge(&tfile
->sk
.sk_receive_queue
);
487 sock_put(&tfile
->sk
);
489 list_for_each_entry_safe(tfile
, tmp
, &tun
->disabled
, next
) {
490 tun_enable_queue(tfile
);
491 skb_queue_purge(&tfile
->sk
.sk_receive_queue
);
492 sock_put(&tfile
->sk
);
494 BUG_ON(tun
->numdisabled
!= 0);
496 if (tun
->flags
& TUN_PERSIST
)
497 module_put(THIS_MODULE
);
500 static int tun_attach(struct tun_struct
*tun
, struct file
*file
)
502 struct tun_file
*tfile
= file
->private_data
;
505 err
= security_tun_dev_attach(tfile
->socket
.sk
, tun
->security
);
510 if (rtnl_dereference(tfile
->tun
) && !tfile
->detached
)
514 if (!(tun
->flags
& TUN_TAP_MQ
) && tun
->numqueues
== 1)
518 if (!tfile
->detached
&&
519 tun
->numqueues
+ tun
->numdisabled
== MAX_TAP_QUEUES
)
524 /* Re-attach the filter to presist device */
525 if (tun
->filter_attached
== true) {
526 err
= sk_attach_filter(&tun
->fprog
, tfile
->socket
.sk
);
530 tfile
->queue_index
= tun
->numqueues
;
531 rcu_assign_pointer(tfile
->tun
, tun
);
532 rcu_assign_pointer(tun
->tfiles
[tun
->numqueues
], tfile
);
536 tun_enable_queue(tfile
);
538 sock_hold(&tfile
->sk
);
540 tun_set_real_num_queues(tun
);
542 /* device is allowed to go away first, so no need to hold extra
550 static struct tun_struct
*__tun_get(struct tun_file
*tfile
)
552 struct tun_struct
*tun
;
555 tun
= rcu_dereference(tfile
->tun
);
563 static struct tun_struct
*tun_get(struct file
*file
)
565 return __tun_get(file
->private_data
);
568 static void tun_put(struct tun_struct
*tun
)
574 static void addr_hash_set(u32
*mask
, const u8
*addr
)
576 int n
= ether_crc(ETH_ALEN
, addr
) >> 26;
577 mask
[n
>> 5] |= (1 << (n
& 31));
580 static unsigned int addr_hash_test(const u32
*mask
, const u8
*addr
)
582 int n
= ether_crc(ETH_ALEN
, addr
) >> 26;
583 return mask
[n
>> 5] & (1 << (n
& 31));
586 static int update_filter(struct tap_filter
*filter
, void __user
*arg
)
588 struct { u8 u
[ETH_ALEN
]; } *addr
;
589 struct tun_filter uf
;
590 int err
, alen
, n
, nexact
;
592 if (copy_from_user(&uf
, arg
, sizeof(uf
)))
601 alen
= ETH_ALEN
* uf
.count
;
602 addr
= kmalloc(alen
, GFP_KERNEL
);
606 if (copy_from_user(addr
, arg
+ sizeof(uf
), alen
)) {
611 /* The filter is updated without holding any locks. Which is
612 * perfectly safe. We disable it first and in the worst
613 * case we'll accept a few undesired packets. */
617 /* Use first set of addresses as an exact filter */
618 for (n
= 0; n
< uf
.count
&& n
< FLT_EXACT_COUNT
; n
++)
619 memcpy(filter
->addr
[n
], addr
[n
].u
, ETH_ALEN
);
623 /* Remaining multicast addresses are hashed,
624 * unicast will leave the filter disabled. */
625 memset(filter
->mask
, 0, sizeof(filter
->mask
));
626 for (; n
< uf
.count
; n
++) {
627 if (!is_multicast_ether_addr(addr
[n
].u
)) {
628 err
= 0; /* no filter */
631 addr_hash_set(filter
->mask
, addr
[n
].u
);
634 /* For ALLMULTI just set the mask to all ones.
635 * This overrides the mask populated above. */
636 if ((uf
.flags
& TUN_FLT_ALLMULTI
))
637 memset(filter
->mask
, ~0, sizeof(filter
->mask
));
639 /* Now enable the filter */
641 filter
->count
= nexact
;
643 /* Return the number of exact filters */
651 /* Returns: 0 - drop, !=0 - accept */
652 static int run_filter(struct tap_filter
*filter
, const struct sk_buff
*skb
)
654 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
656 struct ethhdr
*eh
= (struct ethhdr
*) skb
->data
;
660 for (i
= 0; i
< filter
->count
; i
++)
661 if (ether_addr_equal(eh
->h_dest
, filter
->addr
[i
]))
664 /* Inexact match (multicast only) */
665 if (is_multicast_ether_addr(eh
->h_dest
))
666 return addr_hash_test(filter
->mask
, eh
->h_dest
);
672 * Checks whether the packet is accepted or not.
673 * Returns: 0 - drop, !=0 - accept
675 static int check_filter(struct tap_filter
*filter
, const struct sk_buff
*skb
)
680 return run_filter(filter
, skb
);
683 /* Network device part of the driver */
685 static const struct ethtool_ops tun_ethtool_ops
;
687 /* Net device detach from fd. */
688 static void tun_net_uninit(struct net_device
*dev
)
693 /* Net device open. */
694 static int tun_net_open(struct net_device
*dev
)
696 netif_tx_start_all_queues(dev
);
700 /* Net device close. */
701 static int tun_net_close(struct net_device
*dev
)
703 netif_tx_stop_all_queues(dev
);
707 /* Net device start xmit */
708 static netdev_tx_t
tun_net_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
710 struct tun_struct
*tun
= netdev_priv(dev
);
711 int txq
= skb
->queue_mapping
;
712 struct tun_file
*tfile
;
715 tfile
= rcu_dereference(tun
->tfiles
[txq
]);
717 /* Drop packet if interface is not attached */
718 if (txq
>= tun
->numqueues
)
721 tun_debug(KERN_INFO
, tun
, "tun_net_xmit %d\n", skb
->len
);
725 /* Drop if the filter does not like it.
726 * This is a noop if the filter is disabled.
727 * Filter can be enabled only for the TAP devices. */
728 if (!check_filter(&tun
->txflt
, skb
))
731 if (tfile
->socket
.sk
->sk_filter
&&
732 sk_filter(tfile
->socket
.sk
, skb
))
735 /* Limit the number of packets queued by dividing txq length with the
738 if (skb_queue_len(&tfile
->socket
.sk
->sk_receive_queue
)
739 >= dev
->tx_queue_len
/ tun
->numqueues
)
742 /* Orphan the skb - required as we might hang on to it
743 * for indefinite time. */
744 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
751 skb_queue_tail(&tfile
->socket
.sk
->sk_receive_queue
, skb
);
753 /* Notify and wake up reader process */
754 if (tfile
->flags
& TUN_FASYNC
)
755 kill_fasync(&tfile
->fasync
, SIGIO
, POLL_IN
);
756 wake_up_interruptible_poll(&tfile
->wq
.wait
, POLLIN
|
757 POLLRDNORM
| POLLRDBAND
);
763 dev
->stats
.tx_dropped
++;
770 static void tun_net_mclist(struct net_device
*dev
)
773 * This callback is supposed to deal with mc filter in
774 * _rx_ path and has nothing to do with the _tx_ path.
775 * In rx path we always accept everything userspace gives us.
780 #define MAX_MTU 65535
783 tun_net_change_mtu(struct net_device
*dev
, int new_mtu
)
785 if (new_mtu
< MIN_MTU
|| new_mtu
+ dev
->hard_header_len
> MAX_MTU
)
791 static netdev_features_t
tun_net_fix_features(struct net_device
*dev
,
792 netdev_features_t features
)
794 struct tun_struct
*tun
= netdev_priv(dev
);
796 return (features
& tun
->set_features
) | (features
& ~TUN_USER_FEATURES
);
798 #ifdef CONFIG_NET_POLL_CONTROLLER
799 static void tun_poll_controller(struct net_device
*dev
)
802 * Tun only receives frames when:
803 * 1) the char device endpoint gets data from user space
804 * 2) the tun socket gets a sendmsg call from user space
805 * Since both of those are syncronous operations, we are guaranteed
806 * never to have pending data when we poll for it
807 * so theres nothing to do here but return.
808 * We need this though so netpoll recognizes us as an interface that
809 * supports polling, which enables bridge devices in virt setups to
810 * still use netconsole
815 static const struct net_device_ops tun_netdev_ops
= {
816 .ndo_uninit
= tun_net_uninit
,
817 .ndo_open
= tun_net_open
,
818 .ndo_stop
= tun_net_close
,
819 .ndo_start_xmit
= tun_net_xmit
,
820 .ndo_change_mtu
= tun_net_change_mtu
,
821 .ndo_fix_features
= tun_net_fix_features
,
822 .ndo_select_queue
= tun_select_queue
,
823 #ifdef CONFIG_NET_POLL_CONTROLLER
824 .ndo_poll_controller
= tun_poll_controller
,
828 static const struct net_device_ops tap_netdev_ops
= {
829 .ndo_uninit
= tun_net_uninit
,
830 .ndo_open
= tun_net_open
,
831 .ndo_stop
= tun_net_close
,
832 .ndo_start_xmit
= tun_net_xmit
,
833 .ndo_change_mtu
= tun_net_change_mtu
,
834 .ndo_fix_features
= tun_net_fix_features
,
835 .ndo_set_rx_mode
= tun_net_mclist
,
836 .ndo_set_mac_address
= eth_mac_addr
,
837 .ndo_validate_addr
= eth_validate_addr
,
838 .ndo_select_queue
= tun_select_queue
,
839 #ifdef CONFIG_NET_POLL_CONTROLLER
840 .ndo_poll_controller
= tun_poll_controller
,
844 static void tun_flow_init(struct tun_struct
*tun
)
848 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++)
849 INIT_HLIST_HEAD(&tun
->flows
[i
]);
851 tun
->ageing_time
= TUN_FLOW_EXPIRE
;
852 setup_timer(&tun
->flow_gc_timer
, tun_flow_cleanup
, (unsigned long)tun
);
853 mod_timer(&tun
->flow_gc_timer
,
854 round_jiffies_up(jiffies
+ tun
->ageing_time
));
857 static void tun_flow_uninit(struct tun_struct
*tun
)
859 del_timer_sync(&tun
->flow_gc_timer
);
863 /* Initialize net device. */
864 static void tun_net_init(struct net_device
*dev
)
866 struct tun_struct
*tun
= netdev_priv(dev
);
868 switch (tun
->flags
& TUN_TYPE_MASK
) {
870 dev
->netdev_ops
= &tun_netdev_ops
;
872 /* Point-to-Point TUN Device */
873 dev
->hard_header_len
= 0;
877 /* Zero header length */
878 dev
->type
= ARPHRD_NONE
;
879 dev
->flags
= IFF_POINTOPOINT
| IFF_NOARP
| IFF_MULTICAST
;
880 dev
->tx_queue_len
= TUN_READQ_SIZE
; /* We prefer our own queue length */
884 dev
->netdev_ops
= &tap_netdev_ops
;
885 /* Ethernet TAP Device */
887 dev
->priv_flags
&= ~IFF_TX_SKB_SHARING
;
888 dev
->priv_flags
|= IFF_LIVE_ADDR_CHANGE
;
890 eth_hw_addr_random(dev
);
892 dev
->tx_queue_len
= TUN_READQ_SIZE
; /* We prefer our own queue length */
897 /* Character device part */
900 static unsigned int tun_chr_poll(struct file
*file
, poll_table
*wait
)
902 struct tun_file
*tfile
= file
->private_data
;
903 struct tun_struct
*tun
= __tun_get(tfile
);
905 unsigned int mask
= 0;
910 sk
= tfile
->socket
.sk
;
912 tun_debug(KERN_INFO
, tun
, "tun_chr_poll\n");
914 poll_wait(file
, &tfile
->wq
.wait
, wait
);
916 if (!skb_queue_empty(&sk
->sk_receive_queue
))
917 mask
|= POLLIN
| POLLRDNORM
;
919 if (sock_writeable(sk
) ||
920 (!test_and_set_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
) &&
922 mask
|= POLLOUT
| POLLWRNORM
;
924 if (tun
->dev
->reg_state
!= NETREG_REGISTERED
)
931 /* prepad is the amount to reserve at front. len is length after that.
932 * linear is a hint as to how much to copy (usually headers). */
933 static struct sk_buff
*tun_alloc_skb(struct tun_file
*tfile
,
934 size_t prepad
, size_t len
,
935 size_t linear
, int noblock
)
937 struct sock
*sk
= tfile
->socket
.sk
;
941 /* Under a page? Don't bother with paged skb. */
942 if (prepad
+ len
< PAGE_SIZE
|| !linear
)
945 skb
= sock_alloc_send_pskb(sk
, prepad
+ linear
, len
- linear
, noblock
,
950 skb_reserve(skb
, prepad
);
951 skb_put(skb
, linear
);
952 skb
->data_len
= len
- linear
;
953 skb
->len
+= len
- linear
;
958 /* set skb frags from iovec, this can move to core network code for reuse */
959 static int zerocopy_sg_from_iovec(struct sk_buff
*skb
, const struct iovec
*from
,
960 int offset
, size_t count
)
962 int len
= iov_length(from
, count
) - offset
;
963 int copy
= skb_headlen(skb
);
964 int size
, offset1
= 0;
967 /* Skip over from offset */
968 while (count
&& (offset
>= from
->iov_len
)) {
969 offset
-= from
->iov_len
;
974 /* copy up to skb headlen */
975 while (count
&& (copy
> 0)) {
976 size
= min_t(unsigned int, copy
, from
->iov_len
- offset
);
977 if (copy_from_user(skb
->data
+ offset1
, from
->iov_base
+ offset
,
994 struct page
*page
[MAX_SKB_FRAGS
];
997 unsigned long truesize
;
999 len
= from
->iov_len
- offset
;
1005 base
= (unsigned long)from
->iov_base
+ offset
;
1006 size
= ((base
& ~PAGE_MASK
) + len
+ ~PAGE_MASK
) >> PAGE_SHIFT
;
1007 if (i
+ size
> MAX_SKB_FRAGS
)
1009 num_pages
= get_user_pages_fast(base
, size
, 0, &page
[i
]);
1010 if (num_pages
!= size
) {
1013 for (j
= 0; j
< num_pages
; j
++)
1014 put_page(page
[i
+ j
]);
1017 truesize
= size
* PAGE_SIZE
;
1018 skb
->data_len
+= len
;
1020 skb
->truesize
+= truesize
;
1021 atomic_add(truesize
, &skb
->sk
->sk_wmem_alloc
);
1023 int off
= base
& ~PAGE_MASK
;
1024 int size
= min_t(int, len
, PAGE_SIZE
- off
);
1025 __skb_fill_page_desc(skb
, i
, page
[i
], off
, size
);
1026 skb_shinfo(skb
)->nr_frags
++;
1027 /* increase sk_wmem_alloc */
1038 static unsigned long iov_pages(const struct iovec
*iv
, int offset
,
1039 unsigned long nr_segs
)
1041 unsigned long seg
, base
;
1042 int pages
= 0, len
, size
;
1044 while (nr_segs
&& (offset
>= iv
->iov_len
)) {
1045 offset
-= iv
->iov_len
;
1050 for (seg
= 0; seg
< nr_segs
; seg
++) {
1051 base
= (unsigned long)iv
[seg
].iov_base
+ offset
;
1052 len
= iv
[seg
].iov_len
- offset
;
1053 size
= ((base
& ~PAGE_MASK
) + len
+ ~PAGE_MASK
) >> PAGE_SHIFT
;
1061 /* Get packet from user space buffer */
1062 static ssize_t
tun_get_user(struct tun_struct
*tun
, struct tun_file
*tfile
,
1063 void *msg_control
, const struct iovec
*iv
,
1064 size_t total_len
, size_t count
, int noblock
)
1066 struct tun_pi pi
= { 0, cpu_to_be16(ETH_P_IP
) };
1067 struct sk_buff
*skb
;
1068 size_t len
= total_len
, align
= NET_SKB_PAD
, linear
;
1069 struct virtio_net_hdr gso
= { 0 };
1072 bool zerocopy
= false;
1076 if (!(tun
->flags
& TUN_NO_PI
)) {
1077 if (len
< sizeof(pi
))
1081 if (memcpy_fromiovecend((void *)&pi
, iv
, 0, sizeof(pi
)))
1083 offset
+= sizeof(pi
);
1086 if (tun
->flags
& TUN_VNET_HDR
) {
1087 if (len
< tun
->vnet_hdr_sz
)
1089 len
-= tun
->vnet_hdr_sz
;
1091 if (memcpy_fromiovecend((void *)&gso
, iv
, offset
, sizeof(gso
)))
1094 if ((gso
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) &&
1095 gso
.csum_start
+ gso
.csum_offset
+ 2 > gso
.hdr_len
)
1096 gso
.hdr_len
= gso
.csum_start
+ gso
.csum_offset
+ 2;
1098 if (gso
.hdr_len
> len
)
1100 offset
+= tun
->vnet_hdr_sz
;
1103 if ((tun
->flags
& TUN_TYPE_MASK
) == TUN_TAP_DEV
) {
1104 align
+= NET_IP_ALIGN
;
1105 if (unlikely(len
< ETH_HLEN
||
1106 (gso
.hdr_len
&& gso
.hdr_len
< ETH_HLEN
)))
1111 /* There are 256 bytes to be copied in skb, so there is
1112 * enough room for skb expand head in case it is used.
1113 * The rest of the buffer is mapped from userspace.
1115 copylen
= gso
.hdr_len
? gso
.hdr_len
: GOODCOPY_LEN
;
1117 if (iov_pages(iv
, offset
+ copylen
, count
) <= MAX_SKB_FRAGS
)
1123 linear
= gso
.hdr_len
;
1126 skb
= tun_alloc_skb(tfile
, align
, copylen
, linear
, noblock
);
1128 if (PTR_ERR(skb
) != -EAGAIN
)
1129 tun
->dev
->stats
.rx_dropped
++;
1130 return PTR_ERR(skb
);
1134 err
= zerocopy_sg_from_iovec(skb
, iv
, offset
, count
);
1136 err
= skb_copy_datagram_from_iovec(skb
, 0, iv
, offset
, len
);
1137 if (!err
&& msg_control
) {
1138 struct ubuf_info
*uarg
= msg_control
;
1139 uarg
->callback(uarg
, false);
1144 tun
->dev
->stats
.rx_dropped
++;
1149 if (gso
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) {
1150 if (!skb_partial_csum_set(skb
, gso
.csum_start
,
1152 tun
->dev
->stats
.rx_frame_errors
++;
1158 switch (tun
->flags
& TUN_TYPE_MASK
) {
1160 if (tun
->flags
& TUN_NO_PI
) {
1161 switch (skb
->data
[0] & 0xf0) {
1163 pi
.proto
= htons(ETH_P_IP
);
1166 pi
.proto
= htons(ETH_P_IPV6
);
1169 tun
->dev
->stats
.rx_dropped
++;
1175 skb_reset_mac_header(skb
);
1176 skb
->protocol
= pi
.proto
;
1177 skb
->dev
= tun
->dev
;
1180 skb
->protocol
= eth_type_trans(skb
, tun
->dev
);
1184 if (gso
.gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
1186 switch (gso
.gso_type
& ~VIRTIO_NET_HDR_GSO_ECN
) {
1187 case VIRTIO_NET_HDR_GSO_TCPV4
:
1188 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV4
;
1190 case VIRTIO_NET_HDR_GSO_TCPV6
:
1191 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV6
;
1193 case VIRTIO_NET_HDR_GSO_UDP
:
1194 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
1197 tun
->dev
->stats
.rx_frame_errors
++;
1202 if (gso
.gso_type
& VIRTIO_NET_HDR_GSO_ECN
)
1203 skb_shinfo(skb
)->gso_type
|= SKB_GSO_TCP_ECN
;
1205 skb_shinfo(skb
)->gso_size
= gso
.gso_size
;
1206 if (skb_shinfo(skb
)->gso_size
== 0) {
1207 tun
->dev
->stats
.rx_frame_errors
++;
1212 /* Header must be checked, and gso_segs computed. */
1213 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
1214 skb_shinfo(skb
)->gso_segs
= 0;
1217 /* copy skb_ubuf_info for callback when skb has no error */
1219 skb_shinfo(skb
)->destructor_arg
= msg_control
;
1220 skb_shinfo(skb
)->tx_flags
|= SKBTX_DEV_ZEROCOPY
;
1221 skb_shinfo(skb
)->tx_flags
|= SKBTX_SHARED_FRAG
;
1224 skb_reset_network_header(skb
);
1225 skb_probe_transport_header(skb
, 0);
1227 rxhash
= skb_get_rxhash(skb
);
1230 tun
->dev
->stats
.rx_packets
++;
1231 tun
->dev
->stats
.rx_bytes
+= len
;
1233 tun_flow_update(tun
, rxhash
, tfile
);
1237 static ssize_t
tun_chr_aio_write(struct kiocb
*iocb
, const struct iovec
*iv
,
1238 unsigned long count
, loff_t pos
)
1240 struct file
*file
= iocb
->ki_filp
;
1241 struct tun_struct
*tun
= tun_get(file
);
1242 struct tun_file
*tfile
= file
->private_data
;
1248 tun_debug(KERN_INFO
, tun
, "tun_chr_write %ld\n", count
);
1250 result
= tun_get_user(tun
, tfile
, NULL
, iv
, iov_length(iv
, count
),
1251 count
, file
->f_flags
& O_NONBLOCK
);
1257 /* Put packet to the user space buffer */
1258 static ssize_t
tun_put_user(struct tun_struct
*tun
,
1259 struct tun_file
*tfile
,
1260 struct sk_buff
*skb
,
1261 const struct iovec
*iv
, int len
)
1263 struct tun_pi pi
= { 0, skb
->protocol
};
1266 if (!(tun
->flags
& TUN_NO_PI
)) {
1267 if ((len
-= sizeof(pi
)) < 0)
1270 if (len
< skb
->len
) {
1271 /* Packet will be striped */
1272 pi
.flags
|= TUN_PKT_STRIP
;
1275 if (memcpy_toiovecend(iv
, (void *) &pi
, 0, sizeof(pi
)))
1277 total
+= sizeof(pi
);
1280 if (tun
->flags
& TUN_VNET_HDR
) {
1281 struct virtio_net_hdr gso
= { 0 }; /* no info leak */
1282 if ((len
-= tun
->vnet_hdr_sz
) < 0)
1285 if (skb_is_gso(skb
)) {
1286 struct skb_shared_info
*sinfo
= skb_shinfo(skb
);
1288 /* This is a hint as to how much should be linear. */
1289 gso
.hdr_len
= skb_headlen(skb
);
1290 gso
.gso_size
= sinfo
->gso_size
;
1291 if (sinfo
->gso_type
& SKB_GSO_TCPV4
)
1292 gso
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV4
;
1293 else if (sinfo
->gso_type
& SKB_GSO_TCPV6
)
1294 gso
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV6
;
1295 else if (sinfo
->gso_type
& SKB_GSO_UDP
)
1296 gso
.gso_type
= VIRTIO_NET_HDR_GSO_UDP
;
1298 pr_err("unexpected GSO type: "
1299 "0x%x, gso_size %d, hdr_len %d\n",
1300 sinfo
->gso_type
, gso
.gso_size
,
1302 print_hex_dump(KERN_ERR
, "tun: ",
1305 min((int)gso
.hdr_len
, 64), true);
1309 if (sinfo
->gso_type
& SKB_GSO_TCP_ECN
)
1310 gso
.gso_type
|= VIRTIO_NET_HDR_GSO_ECN
;
1312 gso
.gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
1314 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1315 gso
.flags
= VIRTIO_NET_HDR_F_NEEDS_CSUM
;
1316 gso
.csum_start
= skb_checksum_start_offset(skb
);
1317 gso
.csum_offset
= skb
->csum_offset
;
1318 } else if (skb
->ip_summed
== CHECKSUM_UNNECESSARY
) {
1319 gso
.flags
= VIRTIO_NET_HDR_F_DATA_VALID
;
1320 } /* else everything is zero */
1322 if (unlikely(memcpy_toiovecend(iv
, (void *)&gso
, total
,
1325 total
+= tun
->vnet_hdr_sz
;
1328 len
= min_t(int, skb
->len
, len
);
1330 skb_copy_datagram_const_iovec(skb
, 0, iv
, total
, len
);
1333 tun
->dev
->stats
.tx_packets
++;
1334 tun
->dev
->stats
.tx_bytes
+= len
;
1339 static ssize_t
tun_do_read(struct tun_struct
*tun
, struct tun_file
*tfile
,
1340 struct kiocb
*iocb
, const struct iovec
*iv
,
1341 ssize_t len
, int noblock
)
1343 DECLARE_WAITQUEUE(wait
, current
);
1344 struct sk_buff
*skb
;
1347 tun_debug(KERN_INFO
, tun
, "tun_do_read\n");
1349 if (unlikely(!noblock
))
1350 add_wait_queue(&tfile
->wq
.wait
, &wait
);
1352 current
->state
= TASK_INTERRUPTIBLE
;
1354 /* Read frames from the queue */
1355 if (!(skb
= skb_dequeue(&tfile
->socket
.sk
->sk_receive_queue
))) {
1360 if (signal_pending(current
)) {
1364 if (tun
->dev
->reg_state
!= NETREG_REGISTERED
) {
1369 /* Nothing to read, let's sleep */
1374 ret
= tun_put_user(tun
, tfile
, skb
, iv
, len
);
1379 current
->state
= TASK_RUNNING
;
1380 if (unlikely(!noblock
))
1381 remove_wait_queue(&tfile
->wq
.wait
, &wait
);
1386 static ssize_t
tun_chr_aio_read(struct kiocb
*iocb
, const struct iovec
*iv
,
1387 unsigned long count
, loff_t pos
)
1389 struct file
*file
= iocb
->ki_filp
;
1390 struct tun_file
*tfile
= file
->private_data
;
1391 struct tun_struct
*tun
= __tun_get(tfile
);
1396 len
= iov_length(iv
, count
);
1402 ret
= tun_do_read(tun
, tfile
, iocb
, iv
, len
,
1403 file
->f_flags
& O_NONBLOCK
);
1404 ret
= min_t(ssize_t
, ret
, len
);
1410 static void tun_free_netdev(struct net_device
*dev
)
1412 struct tun_struct
*tun
= netdev_priv(dev
);
1414 BUG_ON(!(list_empty(&tun
->disabled
)));
1415 tun_flow_uninit(tun
);
1416 security_tun_dev_free_security(tun
->security
);
1420 static void tun_setup(struct net_device
*dev
)
1422 struct tun_struct
*tun
= netdev_priv(dev
);
1424 tun
->owner
= INVALID_UID
;
1425 tun
->group
= INVALID_GID
;
1427 dev
->ethtool_ops
= &tun_ethtool_ops
;
1428 dev
->destructor
= tun_free_netdev
;
1431 /* Trivial set of netlink ops to allow deleting tun or tap
1432 * device with netlink.
1434 static int tun_validate(struct nlattr
*tb
[], struct nlattr
*data
[])
1439 static struct rtnl_link_ops tun_link_ops __read_mostly
= {
1441 .priv_size
= sizeof(struct tun_struct
),
1443 .validate
= tun_validate
,
1446 static void tun_sock_write_space(struct sock
*sk
)
1448 struct tun_file
*tfile
;
1449 wait_queue_head_t
*wqueue
;
1451 if (!sock_writeable(sk
))
1454 if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
))
1457 wqueue
= sk_sleep(sk
);
1458 if (wqueue
&& waitqueue_active(wqueue
))
1459 wake_up_interruptible_sync_poll(wqueue
, POLLOUT
|
1460 POLLWRNORM
| POLLWRBAND
);
1462 tfile
= container_of(sk
, struct tun_file
, sk
);
1463 kill_fasync(&tfile
->fasync
, SIGIO
, POLL_OUT
);
1466 static int tun_sendmsg(struct kiocb
*iocb
, struct socket
*sock
,
1467 struct msghdr
*m
, size_t total_len
)
1470 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
1471 struct tun_struct
*tun
= __tun_get(tfile
);
1475 ret
= tun_get_user(tun
, tfile
, m
->msg_control
, m
->msg_iov
, total_len
,
1476 m
->msg_iovlen
, m
->msg_flags
& MSG_DONTWAIT
);
1482 static int tun_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
1483 struct msghdr
*m
, size_t total_len
,
1486 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
1487 struct tun_struct
*tun
= __tun_get(tfile
);
1493 if (flags
& ~(MSG_DONTWAIT
|MSG_TRUNC
)) {
1497 ret
= tun_do_read(tun
, tfile
, iocb
, m
->msg_iov
, total_len
,
1498 flags
& MSG_DONTWAIT
);
1499 if (ret
> total_len
) {
1500 m
->msg_flags
|= MSG_TRUNC
;
1501 ret
= flags
& MSG_TRUNC
? ret
: total_len
;
1508 static int tun_release(struct socket
*sock
)
1515 /* Ops structure to mimic raw sockets with tun */
1516 static const struct proto_ops tun_socket_ops
= {
1517 .sendmsg
= tun_sendmsg
,
1518 .recvmsg
= tun_recvmsg
,
1519 .release
= tun_release
,
1522 static struct proto tun_proto
= {
1524 .owner
= THIS_MODULE
,
1525 .obj_size
= sizeof(struct tun_file
),
1528 static int tun_flags(struct tun_struct
*tun
)
1532 if (tun
->flags
& TUN_TUN_DEV
)
1537 if (tun
->flags
& TUN_NO_PI
)
1540 /* This flag has no real effect. We track the value for backwards
1543 if (tun
->flags
& TUN_ONE_QUEUE
)
1544 flags
|= IFF_ONE_QUEUE
;
1546 if (tun
->flags
& TUN_VNET_HDR
)
1547 flags
|= IFF_VNET_HDR
;
1549 if (tun
->flags
& TUN_TAP_MQ
)
1550 flags
|= IFF_MULTI_QUEUE
;
1552 if (tun
->flags
& TUN_PERSIST
)
1553 flags
|= IFF_PERSIST
;
1558 static ssize_t
tun_show_flags(struct device
*dev
, struct device_attribute
*attr
,
1561 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
1562 return sprintf(buf
, "0x%x\n", tun_flags(tun
));
1565 static ssize_t
tun_show_owner(struct device
*dev
, struct device_attribute
*attr
,
1568 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
1569 return uid_valid(tun
->owner
)?
1570 sprintf(buf
, "%u\n",
1571 from_kuid_munged(current_user_ns(), tun
->owner
)):
1572 sprintf(buf
, "-1\n");
1575 static ssize_t
tun_show_group(struct device
*dev
, struct device_attribute
*attr
,
1578 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
1579 return gid_valid(tun
->group
) ?
1580 sprintf(buf
, "%u\n",
1581 from_kgid_munged(current_user_ns(), tun
->group
)):
1582 sprintf(buf
, "-1\n");
1585 static DEVICE_ATTR(tun_flags
, 0444, tun_show_flags
, NULL
);
1586 static DEVICE_ATTR(owner
, 0444, tun_show_owner
, NULL
);
1587 static DEVICE_ATTR(group
, 0444, tun_show_group
, NULL
);
1589 static int tun_set_iff(struct net
*net
, struct file
*file
, struct ifreq
*ifr
)
1591 struct tun_struct
*tun
;
1592 struct tun_file
*tfile
= file
->private_data
;
1593 struct net_device
*dev
;
1596 if (tfile
->detached
)
1599 dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
1601 if (ifr
->ifr_flags
& IFF_TUN_EXCL
)
1603 if ((ifr
->ifr_flags
& IFF_TUN
) && dev
->netdev_ops
== &tun_netdev_ops
)
1604 tun
= netdev_priv(dev
);
1605 else if ((ifr
->ifr_flags
& IFF_TAP
) && dev
->netdev_ops
== &tap_netdev_ops
)
1606 tun
= netdev_priv(dev
);
1610 if (!!(ifr
->ifr_flags
& IFF_MULTI_QUEUE
) !=
1611 !!(tun
->flags
& TUN_TAP_MQ
))
1614 if (tun_not_capable(tun
))
1616 err
= security_tun_dev_open(tun
->security
);
1620 err
= tun_attach(tun
, file
);
1624 if (tun
->flags
& TUN_TAP_MQ
&&
1625 (tun
->numqueues
+ tun
->numdisabled
> 1)) {
1626 /* One or more queue has already been attached, no need
1627 * to initialize the device again.
1634 unsigned long flags
= 0;
1635 int queues
= ifr
->ifr_flags
& IFF_MULTI_QUEUE
?
1638 if (!ns_capable(net
->user_ns
, CAP_NET_ADMIN
))
1640 err
= security_tun_dev_create();
1645 if (ifr
->ifr_flags
& IFF_TUN
) {
1647 flags
|= TUN_TUN_DEV
;
1649 } else if (ifr
->ifr_flags
& IFF_TAP
) {
1651 flags
|= TUN_TAP_DEV
;
1657 name
= ifr
->ifr_name
;
1659 dev
= alloc_netdev_mqs(sizeof(struct tun_struct
), name
,
1660 tun_setup
, queues
, queues
);
1665 dev_net_set(dev
, net
);
1666 dev
->rtnl_link_ops
= &tun_link_ops
;
1668 tun
= netdev_priv(dev
);
1671 tun
->txflt
.count
= 0;
1672 tun
->vnet_hdr_sz
= sizeof(struct virtio_net_hdr
);
1674 tun
->filter_attached
= false;
1675 tun
->sndbuf
= tfile
->socket
.sk
->sk_sndbuf
;
1677 spin_lock_init(&tun
->lock
);
1679 err
= security_tun_dev_alloc_security(&tun
->security
);
1686 dev
->hw_features
= NETIF_F_SG
| NETIF_F_FRAGLIST
|
1688 dev
->features
= dev
->hw_features
;
1689 dev
->vlan_features
= dev
->features
;
1691 INIT_LIST_HEAD(&tun
->disabled
);
1692 err
= tun_attach(tun
, file
);
1696 err
= register_netdevice(tun
->dev
);
1700 if (device_create_file(&tun
->dev
->dev
, &dev_attr_tun_flags
) ||
1701 device_create_file(&tun
->dev
->dev
, &dev_attr_owner
) ||
1702 device_create_file(&tun
->dev
->dev
, &dev_attr_group
))
1703 pr_err("Failed to create tun sysfs files\n");
1706 netif_carrier_on(tun
->dev
);
1708 tun_debug(KERN_INFO
, tun
, "tun_set_iff\n");
1710 if (ifr
->ifr_flags
& IFF_NO_PI
)
1711 tun
->flags
|= TUN_NO_PI
;
1713 tun
->flags
&= ~TUN_NO_PI
;
1715 /* This flag has no real effect. We track the value for backwards
1718 if (ifr
->ifr_flags
& IFF_ONE_QUEUE
)
1719 tun
->flags
|= TUN_ONE_QUEUE
;
1721 tun
->flags
&= ~TUN_ONE_QUEUE
;
1723 if (ifr
->ifr_flags
& IFF_VNET_HDR
)
1724 tun
->flags
|= TUN_VNET_HDR
;
1726 tun
->flags
&= ~TUN_VNET_HDR
;
1728 if (ifr
->ifr_flags
& IFF_MULTI_QUEUE
)
1729 tun
->flags
|= TUN_TAP_MQ
;
1731 tun
->flags
&= ~TUN_TAP_MQ
;
1733 /* Make sure persistent devices do not get stuck in
1736 if (netif_running(tun
->dev
))
1737 netif_tx_wake_all_queues(tun
->dev
);
1739 strcpy(ifr
->ifr_name
, tun
->dev
->name
);
1747 static void tun_get_iff(struct net
*net
, struct tun_struct
*tun
,
1750 tun_debug(KERN_INFO
, tun
, "tun_get_iff\n");
1752 strcpy(ifr
->ifr_name
, tun
->dev
->name
);
1754 ifr
->ifr_flags
= tun_flags(tun
);
1758 /* This is like a cut-down ethtool ops, except done via tun fd so no
1759 * privs required. */
1760 static int set_offload(struct tun_struct
*tun
, unsigned long arg
)
1762 netdev_features_t features
= 0;
1764 if (arg
& TUN_F_CSUM
) {
1765 features
|= NETIF_F_HW_CSUM
;
1768 if (arg
& (TUN_F_TSO4
|TUN_F_TSO6
)) {
1769 if (arg
& TUN_F_TSO_ECN
) {
1770 features
|= NETIF_F_TSO_ECN
;
1771 arg
&= ~TUN_F_TSO_ECN
;
1773 if (arg
& TUN_F_TSO4
)
1774 features
|= NETIF_F_TSO
;
1775 if (arg
& TUN_F_TSO6
)
1776 features
|= NETIF_F_TSO6
;
1777 arg
&= ~(TUN_F_TSO4
|TUN_F_TSO6
);
1780 if (arg
& TUN_F_UFO
) {
1781 features
|= NETIF_F_UFO
;
1786 /* This gives the user a way to test for new features in future by
1787 * trying to set them. */
1791 tun
->set_features
= features
;
1792 netdev_update_features(tun
->dev
);
1797 static void tun_detach_filter(struct tun_struct
*tun
, int n
)
1800 struct tun_file
*tfile
;
1802 for (i
= 0; i
< n
; i
++) {
1803 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
1804 sk_detach_filter(tfile
->socket
.sk
);
1807 tun
->filter_attached
= false;
1810 static int tun_attach_filter(struct tun_struct
*tun
)
1813 struct tun_file
*tfile
;
1815 for (i
= 0; i
< tun
->numqueues
; i
++) {
1816 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
1817 ret
= sk_attach_filter(&tun
->fprog
, tfile
->socket
.sk
);
1819 tun_detach_filter(tun
, i
);
1824 tun
->filter_attached
= true;
1828 static void tun_set_sndbuf(struct tun_struct
*tun
)
1830 struct tun_file
*tfile
;
1833 for (i
= 0; i
< tun
->numqueues
; i
++) {
1834 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
1835 tfile
->socket
.sk
->sk_sndbuf
= tun
->sndbuf
;
1839 static int tun_set_queue(struct file
*file
, struct ifreq
*ifr
)
1841 struct tun_file
*tfile
= file
->private_data
;
1842 struct tun_struct
*tun
;
1847 if (ifr
->ifr_flags
& IFF_ATTACH_QUEUE
) {
1848 tun
= tfile
->detached
;
1853 ret
= security_tun_dev_attach_queue(tun
->security
);
1856 ret
= tun_attach(tun
, file
);
1857 } else if (ifr
->ifr_flags
& IFF_DETACH_QUEUE
) {
1858 tun
= rtnl_dereference(tfile
->tun
);
1859 if (!tun
|| !(tun
->flags
& TUN_TAP_MQ
) || tfile
->detached
)
1862 __tun_detach(tfile
, false);
1871 static long __tun_chr_ioctl(struct file
*file
, unsigned int cmd
,
1872 unsigned long arg
, int ifreq_len
)
1874 struct tun_file
*tfile
= file
->private_data
;
1875 struct tun_struct
*tun
;
1876 void __user
* argp
= (void __user
*)arg
;
1884 if (cmd
== TUNSETIFF
|| cmd
== TUNSETQUEUE
|| _IOC_TYPE(cmd
) == 0x89) {
1885 if (copy_from_user(&ifr
, argp
, ifreq_len
))
1888 memset(&ifr
, 0, sizeof(ifr
));
1890 if (cmd
== TUNGETFEATURES
) {
1891 /* Currently this just means: "what IFF flags are valid?".
1892 * This is needed because we never checked for invalid flags on
1894 return put_user(IFF_TUN
| IFF_TAP
| IFF_NO_PI
| IFF_ONE_QUEUE
|
1895 IFF_VNET_HDR
| IFF_MULTI_QUEUE
,
1896 (unsigned int __user
*)argp
);
1897 } else if (cmd
== TUNSETQUEUE
)
1898 return tun_set_queue(file
, &ifr
);
1903 tun
= __tun_get(tfile
);
1904 if (cmd
== TUNSETIFF
&& !tun
) {
1905 ifr
.ifr_name
[IFNAMSIZ
-1] = '\0';
1907 ret
= tun_set_iff(tfile
->net
, file
, &ifr
);
1912 if (copy_to_user(argp
, &ifr
, ifreq_len
))
1921 tun_debug(KERN_INFO
, tun
, "tun_chr_ioctl cmd %u\n", cmd
);
1926 tun_get_iff(current
->nsproxy
->net_ns
, tun
, &ifr
);
1928 if (copy_to_user(argp
, &ifr
, ifreq_len
))
1933 /* Disable/Enable checksum */
1935 /* [unimplemented] */
1936 tun_debug(KERN_INFO
, tun
, "ignored: set checksum %s\n",
1937 arg
? "disabled" : "enabled");
1941 /* Disable/Enable persist mode. Keep an extra reference to the
1942 * module to prevent the module being unprobed.
1944 if (arg
&& !(tun
->flags
& TUN_PERSIST
)) {
1945 tun
->flags
|= TUN_PERSIST
;
1946 __module_get(THIS_MODULE
);
1948 if (!arg
&& (tun
->flags
& TUN_PERSIST
)) {
1949 tun
->flags
&= ~TUN_PERSIST
;
1950 module_put(THIS_MODULE
);
1953 tun_debug(KERN_INFO
, tun
, "persist %s\n",
1954 arg
? "enabled" : "disabled");
1958 /* Set owner of the device */
1959 owner
= make_kuid(current_user_ns(), arg
);
1960 if (!uid_valid(owner
)) {
1965 tun_debug(KERN_INFO
, tun
, "owner set to %u\n",
1966 from_kuid(&init_user_ns
, tun
->owner
));
1970 /* Set group of the device */
1971 group
= make_kgid(current_user_ns(), arg
);
1972 if (!gid_valid(group
)) {
1977 tun_debug(KERN_INFO
, tun
, "group set to %u\n",
1978 from_kgid(&init_user_ns
, tun
->group
));
1982 /* Only allow setting the type when the interface is down */
1983 if (tun
->dev
->flags
& IFF_UP
) {
1984 tun_debug(KERN_INFO
, tun
,
1985 "Linktype set failed because interface is up\n");
1988 tun
->dev
->type
= (int) arg
;
1989 tun_debug(KERN_INFO
, tun
, "linktype set to %d\n",
2001 ret
= set_offload(tun
, arg
);
2004 case TUNSETTXFILTER
:
2005 /* Can be set only for TAPs */
2007 if ((tun
->flags
& TUN_TYPE_MASK
) != TUN_TAP_DEV
)
2009 ret
= update_filter(&tun
->txflt
, (void __user
*)arg
);
2013 /* Get hw address */
2014 memcpy(ifr
.ifr_hwaddr
.sa_data
, tun
->dev
->dev_addr
, ETH_ALEN
);
2015 ifr
.ifr_hwaddr
.sa_family
= tun
->dev
->type
;
2016 if (copy_to_user(argp
, &ifr
, ifreq_len
))
2021 /* Set hw address */
2022 tun_debug(KERN_DEBUG
, tun
, "set hw address: %pM\n",
2023 ifr
.ifr_hwaddr
.sa_data
);
2025 ret
= dev_set_mac_address(tun
->dev
, &ifr
.ifr_hwaddr
);
2029 sndbuf
= tfile
->socket
.sk
->sk_sndbuf
;
2030 if (copy_to_user(argp
, &sndbuf
, sizeof(sndbuf
)))
2035 if (copy_from_user(&sndbuf
, argp
, sizeof(sndbuf
))) {
2040 tun
->sndbuf
= sndbuf
;
2041 tun_set_sndbuf(tun
);
2044 case TUNGETVNETHDRSZ
:
2045 vnet_hdr_sz
= tun
->vnet_hdr_sz
;
2046 if (copy_to_user(argp
, &vnet_hdr_sz
, sizeof(vnet_hdr_sz
)))
2050 case TUNSETVNETHDRSZ
:
2051 if (copy_from_user(&vnet_hdr_sz
, argp
, sizeof(vnet_hdr_sz
))) {
2055 if (vnet_hdr_sz
< (int)sizeof(struct virtio_net_hdr
)) {
2060 tun
->vnet_hdr_sz
= vnet_hdr_sz
;
2063 case TUNATTACHFILTER
:
2064 /* Can be set only for TAPs */
2066 if ((tun
->flags
& TUN_TYPE_MASK
) != TUN_TAP_DEV
)
2069 if (copy_from_user(&tun
->fprog
, argp
, sizeof(tun
->fprog
)))
2072 ret
= tun_attach_filter(tun
);
2075 case TUNDETACHFILTER
:
2076 /* Can be set only for TAPs */
2078 if ((tun
->flags
& TUN_TYPE_MASK
) != TUN_TAP_DEV
)
2081 tun_detach_filter(tun
, tun
->numqueues
);
2096 static long tun_chr_ioctl(struct file
*file
,
2097 unsigned int cmd
, unsigned long arg
)
2099 return __tun_chr_ioctl(file
, cmd
, arg
, sizeof (struct ifreq
));
2102 #ifdef CONFIG_COMPAT
2103 static long tun_chr_compat_ioctl(struct file
*file
,
2104 unsigned int cmd
, unsigned long arg
)
2109 case TUNSETTXFILTER
:
2114 arg
= (unsigned long)compat_ptr(arg
);
2117 arg
= (compat_ulong_t
)arg
;
2122 * compat_ifreq is shorter than ifreq, so we must not access beyond
2123 * the end of that structure. All fields that are used in this
2124 * driver are compatible though, we don't need to convert the
2127 return __tun_chr_ioctl(file
, cmd
, arg
, sizeof(struct compat_ifreq
));
2129 #endif /* CONFIG_COMPAT */
2131 static int tun_chr_fasync(int fd
, struct file
*file
, int on
)
2133 struct tun_file
*tfile
= file
->private_data
;
2136 if ((ret
= fasync_helper(fd
, file
, on
, &tfile
->fasync
)) < 0)
2140 ret
= __f_setown(file
, task_pid(current
), PIDTYPE_PID
, 0);
2143 tfile
->flags
|= TUN_FASYNC
;
2145 tfile
->flags
&= ~TUN_FASYNC
;
2151 static int tun_chr_open(struct inode
*inode
, struct file
* file
)
2153 struct tun_file
*tfile
;
2155 DBG1(KERN_INFO
, "tunX: tun_chr_open\n");
2157 tfile
= (struct tun_file
*)sk_alloc(&init_net
, AF_UNSPEC
, GFP_KERNEL
,
2161 rcu_assign_pointer(tfile
->tun
, NULL
);
2162 tfile
->net
= get_net(current
->nsproxy
->net_ns
);
2165 rcu_assign_pointer(tfile
->socket
.wq
, &tfile
->wq
);
2166 init_waitqueue_head(&tfile
->wq
.wait
);
2168 tfile
->socket
.file
= file
;
2169 tfile
->socket
.ops
= &tun_socket_ops
;
2171 sock_init_data(&tfile
->socket
, &tfile
->sk
);
2172 sk_change_net(&tfile
->sk
, tfile
->net
);
2174 tfile
->sk
.sk_write_space
= tun_sock_write_space
;
2175 tfile
->sk
.sk_sndbuf
= INT_MAX
;
2177 file
->private_data
= tfile
;
2178 set_bit(SOCK_EXTERNALLY_ALLOCATED
, &tfile
->socket
.flags
);
2179 INIT_LIST_HEAD(&tfile
->next
);
2181 sock_set_flag(&tfile
->sk
, SOCK_ZEROCOPY
);
2186 static int tun_chr_close(struct inode
*inode
, struct file
*file
)
2188 struct tun_file
*tfile
= file
->private_data
;
2189 struct net
*net
= tfile
->net
;
2191 tun_detach(tfile
, true);
2197 static const struct file_operations tun_fops
= {
2198 .owner
= THIS_MODULE
,
2199 .llseek
= no_llseek
,
2200 .read
= do_sync_read
,
2201 .aio_read
= tun_chr_aio_read
,
2202 .write
= do_sync_write
,
2203 .aio_write
= tun_chr_aio_write
,
2204 .poll
= tun_chr_poll
,
2205 .unlocked_ioctl
= tun_chr_ioctl
,
2206 #ifdef CONFIG_COMPAT
2207 .compat_ioctl
= tun_chr_compat_ioctl
,
2209 .open
= tun_chr_open
,
2210 .release
= tun_chr_close
,
2211 .fasync
= tun_chr_fasync
2214 static struct miscdevice tun_miscdev
= {
2217 .nodename
= "net/tun",
2221 /* ethtool interface */
2223 static int tun_get_settings(struct net_device
*dev
, struct ethtool_cmd
*cmd
)
2226 cmd
->advertising
= 0;
2227 ethtool_cmd_speed_set(cmd
, SPEED_10
);
2228 cmd
->duplex
= DUPLEX_FULL
;
2229 cmd
->port
= PORT_TP
;
2230 cmd
->phy_address
= 0;
2231 cmd
->transceiver
= XCVR_INTERNAL
;
2232 cmd
->autoneg
= AUTONEG_DISABLE
;
2238 static void tun_get_drvinfo(struct net_device
*dev
, struct ethtool_drvinfo
*info
)
2240 struct tun_struct
*tun
= netdev_priv(dev
);
2242 strlcpy(info
->driver
, DRV_NAME
, sizeof(info
->driver
));
2243 strlcpy(info
->version
, DRV_VERSION
, sizeof(info
->version
));
2245 switch (tun
->flags
& TUN_TYPE_MASK
) {
2247 strlcpy(info
->bus_info
, "tun", sizeof(info
->bus_info
));
2250 strlcpy(info
->bus_info
, "tap", sizeof(info
->bus_info
));
2255 static u32
tun_get_msglevel(struct net_device
*dev
)
2258 struct tun_struct
*tun
= netdev_priv(dev
);
2265 static void tun_set_msglevel(struct net_device
*dev
, u32 value
)
2268 struct tun_struct
*tun
= netdev_priv(dev
);
2273 static const struct ethtool_ops tun_ethtool_ops
= {
2274 .get_settings
= tun_get_settings
,
2275 .get_drvinfo
= tun_get_drvinfo
,
2276 .get_msglevel
= tun_get_msglevel
,
2277 .set_msglevel
= tun_set_msglevel
,
2278 .get_link
= ethtool_op_get_link
,
2282 static int __init
tun_init(void)
2286 pr_info("%s, %s\n", DRV_DESCRIPTION
, DRV_VERSION
);
2287 pr_info("%s\n", DRV_COPYRIGHT
);
2289 ret
= rtnl_link_register(&tun_link_ops
);
2291 pr_err("Can't register link_ops\n");
2295 ret
= misc_register(&tun_miscdev
);
2297 pr_err("Can't register misc device %d\n", TUN_MINOR
);
2302 rtnl_link_unregister(&tun_link_ops
);
2307 static void tun_cleanup(void)
2309 misc_deregister(&tun_miscdev
);
2310 rtnl_link_unregister(&tun_link_ops
);
2313 /* Get an underlying socket object from tun file. Returns error unless file is
2314 * attached to a device. The returned object works like a packet socket, it
2315 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
2316 * holding a reference to the file for as long as the socket is in use. */
2317 struct socket
*tun_get_socket(struct file
*file
)
2319 struct tun_file
*tfile
;
2320 if (file
->f_op
!= &tun_fops
)
2321 return ERR_PTR(-EINVAL
);
2322 tfile
= file
->private_data
;
2324 return ERR_PTR(-EBADFD
);
2325 return &tfile
->socket
;
2327 EXPORT_SYMBOL_GPL(tun_get_socket
);
2329 module_init(tun_init
);
2330 module_exit(tun_cleanup
);
2331 MODULE_DESCRIPTION(DRV_DESCRIPTION
);
2332 MODULE_AUTHOR(DRV_COPYRIGHT
);
2333 MODULE_LICENSE("GPL");
2334 MODULE_ALIAS_MISCDEV(TUN_MINOR
);
2335 MODULE_ALIAS("devname:net/tun");