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 */
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
;
190 static inline u32
tun_hashfn(u32 rxhash
)
192 return rxhash
& 0x3ff;
195 static struct tun_flow_entry
*tun_flow_find(struct hlist_head
*head
, u32 rxhash
)
197 struct tun_flow_entry
*e
;
198 struct hlist_node
*n
;
200 hlist_for_each_entry_rcu(e
, n
, head
, hash_link
) {
201 if (e
->rxhash
== rxhash
)
207 static struct tun_flow_entry
*tun_flow_create(struct tun_struct
*tun
,
208 struct hlist_head
*head
,
209 u32 rxhash
, u16 queue_index
)
211 struct tun_flow_entry
*e
= kmalloc(sizeof(*e
), GFP_ATOMIC
);
214 tun_debug(KERN_INFO
, tun
, "create flow: hash %u index %u\n",
215 rxhash
, queue_index
);
216 e
->updated
= jiffies
;
218 e
->queue_index
= queue_index
;
220 hlist_add_head_rcu(&e
->hash_link
, head
);
225 static void tun_flow_delete(struct tun_struct
*tun
, struct tun_flow_entry
*e
)
227 tun_debug(KERN_INFO
, tun
, "delete flow: hash %u index %u\n",
228 e
->rxhash
, e
->queue_index
);
229 hlist_del_rcu(&e
->hash_link
);
233 static void tun_flow_flush(struct tun_struct
*tun
)
237 spin_lock_bh(&tun
->lock
);
238 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
239 struct tun_flow_entry
*e
;
240 struct hlist_node
*h
, *n
;
242 hlist_for_each_entry_safe(e
, h
, n
, &tun
->flows
[i
], hash_link
)
243 tun_flow_delete(tun
, e
);
245 spin_unlock_bh(&tun
->lock
);
248 static void tun_flow_delete_by_queue(struct tun_struct
*tun
, u16 queue_index
)
252 spin_lock_bh(&tun
->lock
);
253 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
254 struct tun_flow_entry
*e
;
255 struct hlist_node
*h
, *n
;
257 hlist_for_each_entry_safe(e
, h
, n
, &tun
->flows
[i
], hash_link
) {
258 if (e
->queue_index
== queue_index
)
259 tun_flow_delete(tun
, e
);
262 spin_unlock_bh(&tun
->lock
);
265 static void tun_flow_cleanup(unsigned long data
)
267 struct tun_struct
*tun
= (struct tun_struct
*)data
;
268 unsigned long delay
= tun
->ageing_time
;
269 unsigned long next_timer
= jiffies
+ delay
;
270 unsigned long count
= 0;
273 tun_debug(KERN_INFO
, tun
, "tun_flow_cleanup\n");
275 spin_lock_bh(&tun
->lock
);
276 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
277 struct tun_flow_entry
*e
;
278 struct hlist_node
*h
, *n
;
280 hlist_for_each_entry_safe(e
, h
, n
, &tun
->flows
[i
], hash_link
) {
281 unsigned long this_timer
;
283 this_timer
= e
->updated
+ delay
;
284 if (time_before_eq(this_timer
, jiffies
))
285 tun_flow_delete(tun
, e
);
286 else if (time_before(this_timer
, next_timer
))
287 next_timer
= this_timer
;
292 mod_timer(&tun
->flow_gc_timer
, round_jiffies_up(next_timer
));
293 spin_unlock_bh(&tun
->lock
);
296 static void tun_flow_update(struct tun_struct
*tun
, u32 rxhash
,
299 struct hlist_head
*head
;
300 struct tun_flow_entry
*e
;
301 unsigned long delay
= tun
->ageing_time
;
306 head
= &tun
->flows
[tun_hashfn(rxhash
)];
310 if (tun
->numqueues
== 1)
313 e
= tun_flow_find(head
, rxhash
);
315 /* TODO: keep queueing to old queue until it's empty? */
316 e
->queue_index
= queue_index
;
317 e
->updated
= jiffies
;
319 spin_lock_bh(&tun
->lock
);
320 if (!tun_flow_find(head
, rxhash
))
321 tun_flow_create(tun
, head
, rxhash
, queue_index
);
323 if (!timer_pending(&tun
->flow_gc_timer
))
324 mod_timer(&tun
->flow_gc_timer
,
325 round_jiffies_up(jiffies
+ delay
));
326 spin_unlock_bh(&tun
->lock
);
333 /* We try to identify a flow through its rxhash first. The reason that
334 * we do not check rxq no. is becuase some cards(e.g 82599), chooses
335 * the rxq based on the txq where the last packet of the flow comes. As
336 * the userspace application move between processors, we may get a
337 * different rxq no. here. If we could not get rxhash, then we would
338 * hope the rxq no. may help here.
340 static u16
tun_select_queue(struct net_device
*dev
, struct sk_buff
*skb
)
342 struct tun_struct
*tun
= netdev_priv(dev
);
343 struct tun_flow_entry
*e
;
348 numqueues
= tun
->numqueues
;
350 txq
= skb_get_rxhash(skb
);
352 e
= tun_flow_find(&tun
->flows
[tun_hashfn(txq
)], txq
);
354 txq
= e
->queue_index
;
356 /* use multiply and shift instead of expensive divide */
357 txq
= ((u64
)txq
* numqueues
) >> 32;
358 } else if (likely(skb_rx_queue_recorded(skb
))) {
359 txq
= skb_get_rx_queue(skb
);
360 while (unlikely(txq
>= numqueues
))
368 static inline bool tun_not_capable(struct tun_struct
*tun
)
370 const struct cred
*cred
= current_cred();
371 struct net
*net
= dev_net(tun
->dev
);
373 return ((uid_valid(tun
->owner
) && !uid_eq(cred
->euid
, tun
->owner
)) ||
374 (gid_valid(tun
->group
) && !in_egroup_p(tun
->group
))) &&
375 !ns_capable(net
->user_ns
, CAP_NET_ADMIN
);
378 static void tun_set_real_num_queues(struct tun_struct
*tun
)
380 netif_set_real_num_tx_queues(tun
->dev
, tun
->numqueues
);
381 netif_set_real_num_rx_queues(tun
->dev
, tun
->numqueues
);
384 static void tun_disable_queue(struct tun_struct
*tun
, struct tun_file
*tfile
)
386 tfile
->detached
= tun
;
387 list_add_tail(&tfile
->next
, &tun
->disabled
);
391 static struct tun_struct
*tun_enable_queue(struct tun_file
*tfile
)
393 struct tun_struct
*tun
= tfile
->detached
;
395 tfile
->detached
= NULL
;
396 list_del_init(&tfile
->next
);
401 static void __tun_detach(struct tun_file
*tfile
, bool clean
)
403 struct tun_file
*ntfile
;
404 struct tun_struct
*tun
;
405 struct net_device
*dev
;
407 tun
= rtnl_dereference(tfile
->tun
);
410 u16 index
= tfile
->queue_index
;
411 BUG_ON(index
>= tun
->numqueues
);
414 rcu_assign_pointer(tun
->tfiles
[index
],
415 tun
->tfiles
[tun
->numqueues
- 1]);
416 rcu_assign_pointer(tfile
->tun
, NULL
);
417 ntfile
= rtnl_dereference(tun
->tfiles
[index
]);
418 ntfile
->queue_index
= index
;
422 sock_put(&tfile
->sk
);
424 tun_disable_queue(tun
, tfile
);
427 tun_flow_delete_by_queue(tun
, tun
->numqueues
+ 1);
428 /* Drop read queue */
429 skb_queue_purge(&tfile
->sk
.sk_receive_queue
);
430 tun_set_real_num_queues(tun
);
431 } else if (tfile
->detached
&& clean
) {
432 tun
= tun_enable_queue(tfile
);
433 sock_put(&tfile
->sk
);
437 if (tun
&& tun
->numqueues
== 0 && tun
->numdisabled
== 0 &&
438 !(tun
->flags
& TUN_PERSIST
))
439 if (tun
->dev
->reg_state
== NETREG_REGISTERED
)
440 unregister_netdevice(tun
->dev
);
442 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED
,
443 &tfile
->socket
.flags
));
444 sk_release_kernel(&tfile
->sk
);
448 static void tun_detach(struct tun_file
*tfile
, bool clean
)
451 __tun_detach(tfile
, clean
);
455 static void tun_detach_all(struct net_device
*dev
)
457 struct tun_struct
*tun
= netdev_priv(dev
);
458 struct tun_file
*tfile
, *tmp
;
459 int i
, n
= tun
->numqueues
;
461 for (i
= 0; i
< n
; i
++) {
462 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
464 wake_up_all(&tfile
->wq
.wait
);
465 rcu_assign_pointer(tfile
->tun
, NULL
);
468 BUG_ON(tun
->numqueues
!= 0);
471 for (i
= 0; i
< n
; i
++) {
472 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
473 /* Drop read queue */
474 skb_queue_purge(&tfile
->sk
.sk_receive_queue
);
475 sock_put(&tfile
->sk
);
477 list_for_each_entry_safe(tfile
, tmp
, &tun
->disabled
, next
) {
478 tun_enable_queue(tfile
);
479 skb_queue_purge(&tfile
->sk
.sk_receive_queue
);
480 sock_put(&tfile
->sk
);
482 BUG_ON(tun
->numdisabled
!= 0);
484 if (tun
->flags
& TUN_PERSIST
)
485 module_put(THIS_MODULE
);
488 static int tun_attach(struct tun_struct
*tun
, struct file
*file
)
490 struct tun_file
*tfile
= file
->private_data
;
494 if (rtnl_dereference(tfile
->tun
))
498 if (!(tun
->flags
& TUN_TAP_MQ
) && tun
->numqueues
== 1)
502 if (!tfile
->detached
&&
503 tun
->numqueues
+ tun
->numdisabled
== MAX_TAP_QUEUES
)
508 /* Re-attach the filter to presist device */
509 if (tun
->filter_attached
== true) {
510 err
= sk_attach_filter(&tun
->fprog
, tfile
->socket
.sk
);
514 tfile
->queue_index
= tun
->numqueues
;
515 rcu_assign_pointer(tfile
->tun
, tun
);
516 rcu_assign_pointer(tun
->tfiles
[tun
->numqueues
], tfile
);
520 tun_enable_queue(tfile
);
522 sock_hold(&tfile
->sk
);
524 tun_set_real_num_queues(tun
);
526 /* device is allowed to go away first, so no need to hold extra
534 static struct tun_struct
*__tun_get(struct tun_file
*tfile
)
536 struct tun_struct
*tun
;
539 tun
= rcu_dereference(tfile
->tun
);
547 static struct tun_struct
*tun_get(struct file
*file
)
549 return __tun_get(file
->private_data
);
552 static void tun_put(struct tun_struct
*tun
)
558 static void addr_hash_set(u32
*mask
, const u8
*addr
)
560 int n
= ether_crc(ETH_ALEN
, addr
) >> 26;
561 mask
[n
>> 5] |= (1 << (n
& 31));
564 static unsigned int addr_hash_test(const u32
*mask
, const u8
*addr
)
566 int n
= ether_crc(ETH_ALEN
, addr
) >> 26;
567 return mask
[n
>> 5] & (1 << (n
& 31));
570 static int update_filter(struct tap_filter
*filter
, void __user
*arg
)
572 struct { u8 u
[ETH_ALEN
]; } *addr
;
573 struct tun_filter uf
;
574 int err
, alen
, n
, nexact
;
576 if (copy_from_user(&uf
, arg
, sizeof(uf
)))
585 alen
= ETH_ALEN
* uf
.count
;
586 addr
= kmalloc(alen
, GFP_KERNEL
);
590 if (copy_from_user(addr
, arg
+ sizeof(uf
), alen
)) {
595 /* The filter is updated without holding any locks. Which is
596 * perfectly safe. We disable it first and in the worst
597 * case we'll accept a few undesired packets. */
601 /* Use first set of addresses as an exact filter */
602 for (n
= 0; n
< uf
.count
&& n
< FLT_EXACT_COUNT
; n
++)
603 memcpy(filter
->addr
[n
], addr
[n
].u
, ETH_ALEN
);
607 /* Remaining multicast addresses are hashed,
608 * unicast will leave the filter disabled. */
609 memset(filter
->mask
, 0, sizeof(filter
->mask
));
610 for (; n
< uf
.count
; n
++) {
611 if (!is_multicast_ether_addr(addr
[n
].u
)) {
612 err
= 0; /* no filter */
615 addr_hash_set(filter
->mask
, addr
[n
].u
);
618 /* For ALLMULTI just set the mask to all ones.
619 * This overrides the mask populated above. */
620 if ((uf
.flags
& TUN_FLT_ALLMULTI
))
621 memset(filter
->mask
, ~0, sizeof(filter
->mask
));
623 /* Now enable the filter */
625 filter
->count
= nexact
;
627 /* Return the number of exact filters */
635 /* Returns: 0 - drop, !=0 - accept */
636 static int run_filter(struct tap_filter
*filter
, const struct sk_buff
*skb
)
638 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
640 struct ethhdr
*eh
= (struct ethhdr
*) skb
->data
;
644 for (i
= 0; i
< filter
->count
; i
++)
645 if (ether_addr_equal(eh
->h_dest
, filter
->addr
[i
]))
648 /* Inexact match (multicast only) */
649 if (is_multicast_ether_addr(eh
->h_dest
))
650 return addr_hash_test(filter
->mask
, eh
->h_dest
);
656 * Checks whether the packet is accepted or not.
657 * Returns: 0 - drop, !=0 - accept
659 static int check_filter(struct tap_filter
*filter
, const struct sk_buff
*skb
)
664 return run_filter(filter
, skb
);
667 /* Network device part of the driver */
669 static const struct ethtool_ops tun_ethtool_ops
;
671 /* Net device detach from fd. */
672 static void tun_net_uninit(struct net_device
*dev
)
677 /* Net device open. */
678 static int tun_net_open(struct net_device
*dev
)
680 netif_tx_start_all_queues(dev
);
684 /* Net device close. */
685 static int tun_net_close(struct net_device
*dev
)
687 netif_tx_stop_all_queues(dev
);
691 /* Net device start xmit */
692 static netdev_tx_t
tun_net_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
694 struct tun_struct
*tun
= netdev_priv(dev
);
695 int txq
= skb
->queue_mapping
;
696 struct tun_file
*tfile
;
699 tfile
= rcu_dereference(tun
->tfiles
[txq
]);
701 /* Drop packet if interface is not attached */
702 if (txq
>= tun
->numqueues
)
705 tun_debug(KERN_INFO
, tun
, "tun_net_xmit %d\n", skb
->len
);
709 /* Drop if the filter does not like it.
710 * This is a noop if the filter is disabled.
711 * Filter can be enabled only for the TAP devices. */
712 if (!check_filter(&tun
->txflt
, skb
))
715 if (tfile
->socket
.sk
->sk_filter
&&
716 sk_filter(tfile
->socket
.sk
, skb
))
719 /* Limit the number of packets queued by dividing txq length with the
722 if (skb_queue_len(&tfile
->socket
.sk
->sk_receive_queue
)
723 >= dev
->tx_queue_len
/ tun
->numqueues
)
726 /* Orphan the skb - required as we might hang on to it
727 * for indefinite time. */
728 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
733 skb_queue_tail(&tfile
->socket
.sk
->sk_receive_queue
, skb
);
735 /* Notify and wake up reader process */
736 if (tfile
->flags
& TUN_FASYNC
)
737 kill_fasync(&tfile
->fasync
, SIGIO
, POLL_IN
);
738 wake_up_interruptible_poll(&tfile
->wq
.wait
, POLLIN
|
739 POLLRDNORM
| POLLRDBAND
);
745 dev
->stats
.tx_dropped
++;
752 static void tun_net_mclist(struct net_device
*dev
)
755 * This callback is supposed to deal with mc filter in
756 * _rx_ path and has nothing to do with the _tx_ path.
757 * In rx path we always accept everything userspace gives us.
762 #define MAX_MTU 65535
765 tun_net_change_mtu(struct net_device
*dev
, int new_mtu
)
767 if (new_mtu
< MIN_MTU
|| new_mtu
+ dev
->hard_header_len
> MAX_MTU
)
773 static netdev_features_t
tun_net_fix_features(struct net_device
*dev
,
774 netdev_features_t features
)
776 struct tun_struct
*tun
= netdev_priv(dev
);
778 return (features
& tun
->set_features
) | (features
& ~TUN_USER_FEATURES
);
780 #ifdef CONFIG_NET_POLL_CONTROLLER
781 static void tun_poll_controller(struct net_device
*dev
)
784 * Tun only receives frames when:
785 * 1) the char device endpoint gets data from user space
786 * 2) the tun socket gets a sendmsg call from user space
787 * Since both of those are syncronous operations, we are guaranteed
788 * never to have pending data when we poll for it
789 * so theres nothing to do here but return.
790 * We need this though so netpoll recognizes us as an interface that
791 * supports polling, which enables bridge devices in virt setups to
792 * still use netconsole
797 static const struct net_device_ops tun_netdev_ops
= {
798 .ndo_uninit
= tun_net_uninit
,
799 .ndo_open
= tun_net_open
,
800 .ndo_stop
= tun_net_close
,
801 .ndo_start_xmit
= tun_net_xmit
,
802 .ndo_change_mtu
= tun_net_change_mtu
,
803 .ndo_fix_features
= tun_net_fix_features
,
804 .ndo_select_queue
= tun_select_queue
,
805 #ifdef CONFIG_NET_POLL_CONTROLLER
806 .ndo_poll_controller
= tun_poll_controller
,
810 static const struct net_device_ops tap_netdev_ops
= {
811 .ndo_uninit
= tun_net_uninit
,
812 .ndo_open
= tun_net_open
,
813 .ndo_stop
= tun_net_close
,
814 .ndo_start_xmit
= tun_net_xmit
,
815 .ndo_change_mtu
= tun_net_change_mtu
,
816 .ndo_fix_features
= tun_net_fix_features
,
817 .ndo_set_rx_mode
= tun_net_mclist
,
818 .ndo_set_mac_address
= eth_mac_addr
,
819 .ndo_validate_addr
= eth_validate_addr
,
820 .ndo_select_queue
= tun_select_queue
,
821 #ifdef CONFIG_NET_POLL_CONTROLLER
822 .ndo_poll_controller
= tun_poll_controller
,
826 static int tun_flow_init(struct tun_struct
*tun
)
830 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++)
831 INIT_HLIST_HEAD(&tun
->flows
[i
]);
833 tun
->ageing_time
= TUN_FLOW_EXPIRE
;
834 setup_timer(&tun
->flow_gc_timer
, tun_flow_cleanup
, (unsigned long)tun
);
835 mod_timer(&tun
->flow_gc_timer
,
836 round_jiffies_up(jiffies
+ tun
->ageing_time
));
841 static void tun_flow_uninit(struct tun_struct
*tun
)
843 del_timer_sync(&tun
->flow_gc_timer
);
847 /* Initialize net device. */
848 static void tun_net_init(struct net_device
*dev
)
850 struct tun_struct
*tun
= netdev_priv(dev
);
852 switch (tun
->flags
& TUN_TYPE_MASK
) {
854 dev
->netdev_ops
= &tun_netdev_ops
;
856 /* Point-to-Point TUN Device */
857 dev
->hard_header_len
= 0;
861 /* Zero header length */
862 dev
->type
= ARPHRD_NONE
;
863 dev
->flags
= IFF_POINTOPOINT
| IFF_NOARP
| IFF_MULTICAST
;
864 dev
->tx_queue_len
= TUN_READQ_SIZE
; /* We prefer our own queue length */
868 dev
->netdev_ops
= &tap_netdev_ops
;
869 /* Ethernet TAP Device */
871 dev
->priv_flags
&= ~IFF_TX_SKB_SHARING
;
872 dev
->priv_flags
|= IFF_LIVE_ADDR_CHANGE
;
874 eth_hw_addr_random(dev
);
876 dev
->tx_queue_len
= TUN_READQ_SIZE
; /* We prefer our own queue length */
881 /* Character device part */
884 static unsigned int tun_chr_poll(struct file
*file
, poll_table
*wait
)
886 struct tun_file
*tfile
= file
->private_data
;
887 struct tun_struct
*tun
= __tun_get(tfile
);
889 unsigned int mask
= 0;
894 sk
= tfile
->socket
.sk
;
896 tun_debug(KERN_INFO
, tun
, "tun_chr_poll\n");
898 poll_wait(file
, &tfile
->wq
.wait
, wait
);
900 if (!skb_queue_empty(&sk
->sk_receive_queue
))
901 mask
|= POLLIN
| POLLRDNORM
;
903 if (sock_writeable(sk
) ||
904 (!test_and_set_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
) &&
906 mask
|= POLLOUT
| POLLWRNORM
;
908 if (tun
->dev
->reg_state
!= NETREG_REGISTERED
)
915 /* prepad is the amount to reserve at front. len is length after that.
916 * linear is a hint as to how much to copy (usually headers). */
917 static struct sk_buff
*tun_alloc_skb(struct tun_file
*tfile
,
918 size_t prepad
, size_t len
,
919 size_t linear
, int noblock
)
921 struct sock
*sk
= tfile
->socket
.sk
;
925 /* Under a page? Don't bother with paged skb. */
926 if (prepad
+ len
< PAGE_SIZE
|| !linear
)
929 skb
= sock_alloc_send_pskb(sk
, prepad
+ linear
, len
- linear
, noblock
,
934 skb_reserve(skb
, prepad
);
935 skb_put(skb
, linear
);
936 skb
->data_len
= len
- linear
;
937 skb
->len
+= len
- linear
;
942 /* set skb frags from iovec, this can move to core network code for reuse */
943 static int zerocopy_sg_from_iovec(struct sk_buff
*skb
, const struct iovec
*from
,
944 int offset
, size_t count
)
946 int len
= iov_length(from
, count
) - offset
;
947 int copy
= skb_headlen(skb
);
948 int size
, offset1
= 0;
951 /* Skip over from offset */
952 while (count
&& (offset
>= from
->iov_len
)) {
953 offset
-= from
->iov_len
;
958 /* copy up to skb headlen */
959 while (count
&& (copy
> 0)) {
960 size
= min_t(unsigned int, copy
, from
->iov_len
- offset
);
961 if (copy_from_user(skb
->data
+ offset1
, from
->iov_base
+ offset
,
978 struct page
*page
[MAX_SKB_FRAGS
];
981 unsigned long truesize
;
983 len
= from
->iov_len
- offset
;
989 base
= (unsigned long)from
->iov_base
+ offset
;
990 size
= ((base
& ~PAGE_MASK
) + len
+ ~PAGE_MASK
) >> PAGE_SHIFT
;
991 if (i
+ size
> MAX_SKB_FRAGS
)
993 num_pages
= get_user_pages_fast(base
, size
, 0, &page
[i
]);
994 if (num_pages
!= size
) {
995 for (i
= 0; i
< num_pages
; i
++)
999 truesize
= size
* PAGE_SIZE
;
1000 skb
->data_len
+= len
;
1002 skb
->truesize
+= truesize
;
1003 atomic_add(truesize
, &skb
->sk
->sk_wmem_alloc
);
1005 int off
= base
& ~PAGE_MASK
;
1006 int size
= min_t(int, len
, PAGE_SIZE
- off
);
1007 __skb_fill_page_desc(skb
, i
, page
[i
], off
, size
);
1008 skb_shinfo(skb
)->nr_frags
++;
1009 /* increase sk_wmem_alloc */
1020 /* Get packet from user space buffer */
1021 static ssize_t
tun_get_user(struct tun_struct
*tun
, struct tun_file
*tfile
,
1022 void *msg_control
, const struct iovec
*iv
,
1023 size_t total_len
, size_t count
, int noblock
)
1025 struct tun_pi pi
= { 0, cpu_to_be16(ETH_P_IP
) };
1026 struct sk_buff
*skb
;
1027 size_t len
= total_len
, align
= NET_SKB_PAD
;
1028 struct virtio_net_hdr gso
= { 0 };
1031 bool zerocopy
= false;
1035 if (!(tun
->flags
& TUN_NO_PI
)) {
1036 if ((len
-= sizeof(pi
)) > total_len
)
1039 if (memcpy_fromiovecend((void *)&pi
, iv
, 0, sizeof(pi
)))
1041 offset
+= sizeof(pi
);
1044 if (tun
->flags
& TUN_VNET_HDR
) {
1045 if ((len
-= tun
->vnet_hdr_sz
) > total_len
)
1048 if (memcpy_fromiovecend((void *)&gso
, iv
, offset
, sizeof(gso
)))
1051 if ((gso
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) &&
1052 gso
.csum_start
+ gso
.csum_offset
+ 2 > gso
.hdr_len
)
1053 gso
.hdr_len
= gso
.csum_start
+ gso
.csum_offset
+ 2;
1055 if (gso
.hdr_len
> len
)
1057 offset
+= tun
->vnet_hdr_sz
;
1060 if ((tun
->flags
& TUN_TYPE_MASK
) == TUN_TAP_DEV
) {
1061 align
+= NET_IP_ALIGN
;
1062 if (unlikely(len
< ETH_HLEN
||
1063 (gso
.hdr_len
&& gso
.hdr_len
< ETH_HLEN
)))
1071 /* Userspace may produce vectors with count greater than
1072 * MAX_SKB_FRAGS, so we need to linearize parts of the skb
1073 * to let the rest of data to be fit in the frags.
1075 if (count
> MAX_SKB_FRAGS
) {
1076 copylen
= iov_length(iv
, count
- MAX_SKB_FRAGS
);
1077 if (copylen
< offset
)
1083 /* There are 256 bytes to be copied in skb, so there is enough
1084 * room for skb expand head in case it is used.
1085 * The rest of the buffer is mapped from userspace.
1087 if (copylen
< gso
.hdr_len
)
1088 copylen
= gso
.hdr_len
;
1090 copylen
= GOODCOPY_LEN
;
1094 skb
= tun_alloc_skb(tfile
, align
, copylen
, gso
.hdr_len
, noblock
);
1096 if (PTR_ERR(skb
) != -EAGAIN
)
1097 tun
->dev
->stats
.rx_dropped
++;
1098 return PTR_ERR(skb
);
1102 err
= zerocopy_sg_from_iovec(skb
, iv
, offset
, count
);
1104 err
= skb_copy_datagram_from_iovec(skb
, 0, iv
, offset
, len
);
1107 tun
->dev
->stats
.rx_dropped
++;
1112 if (gso
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) {
1113 if (!skb_partial_csum_set(skb
, gso
.csum_start
,
1115 tun
->dev
->stats
.rx_frame_errors
++;
1121 switch (tun
->flags
& TUN_TYPE_MASK
) {
1123 if (tun
->flags
& TUN_NO_PI
) {
1124 switch (skb
->data
[0] & 0xf0) {
1126 pi
.proto
= htons(ETH_P_IP
);
1129 pi
.proto
= htons(ETH_P_IPV6
);
1132 tun
->dev
->stats
.rx_dropped
++;
1138 skb_reset_mac_header(skb
);
1139 skb
->protocol
= pi
.proto
;
1140 skb
->dev
= tun
->dev
;
1143 skb
->protocol
= eth_type_trans(skb
, tun
->dev
);
1147 if (gso
.gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
1149 switch (gso
.gso_type
& ~VIRTIO_NET_HDR_GSO_ECN
) {
1150 case VIRTIO_NET_HDR_GSO_TCPV4
:
1151 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV4
;
1153 case VIRTIO_NET_HDR_GSO_TCPV6
:
1154 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV6
;
1156 case VIRTIO_NET_HDR_GSO_UDP
:
1157 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
1160 tun
->dev
->stats
.rx_frame_errors
++;
1165 if (gso
.gso_type
& VIRTIO_NET_HDR_GSO_ECN
)
1166 skb_shinfo(skb
)->gso_type
|= SKB_GSO_TCP_ECN
;
1168 skb_shinfo(skb
)->gso_size
= gso
.gso_size
;
1169 if (skb_shinfo(skb
)->gso_size
== 0) {
1170 tun
->dev
->stats
.rx_frame_errors
++;
1175 /* Header must be checked, and gso_segs computed. */
1176 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
1177 skb_shinfo(skb
)->gso_segs
= 0;
1180 /* copy skb_ubuf_info for callback when skb has no error */
1182 skb_shinfo(skb
)->destructor_arg
= msg_control
;
1183 skb_shinfo(skb
)->tx_flags
|= SKBTX_DEV_ZEROCOPY
;
1186 skb_reset_network_header(skb
);
1187 rxhash
= skb_get_rxhash(skb
);
1190 tun
->dev
->stats
.rx_packets
++;
1191 tun
->dev
->stats
.rx_bytes
+= len
;
1193 tun_flow_update(tun
, rxhash
, tfile
->queue_index
);
1197 static ssize_t
tun_chr_aio_write(struct kiocb
*iocb
, const struct iovec
*iv
,
1198 unsigned long count
, loff_t pos
)
1200 struct file
*file
= iocb
->ki_filp
;
1201 struct tun_struct
*tun
= tun_get(file
);
1202 struct tun_file
*tfile
= file
->private_data
;
1208 tun_debug(KERN_INFO
, tun
, "tun_chr_write %ld\n", count
);
1210 result
= tun_get_user(tun
, tfile
, NULL
, iv
, iov_length(iv
, count
),
1211 count
, file
->f_flags
& O_NONBLOCK
);
1217 /* Put packet to the user space buffer */
1218 static ssize_t
tun_put_user(struct tun_struct
*tun
,
1219 struct tun_file
*tfile
,
1220 struct sk_buff
*skb
,
1221 const struct iovec
*iv
, int len
)
1223 struct tun_pi pi
= { 0, skb
->protocol
};
1226 if (!(tun
->flags
& TUN_NO_PI
)) {
1227 if ((len
-= sizeof(pi
)) < 0)
1230 if (len
< skb
->len
) {
1231 /* Packet will be striped */
1232 pi
.flags
|= TUN_PKT_STRIP
;
1235 if (memcpy_toiovecend(iv
, (void *) &pi
, 0, sizeof(pi
)))
1237 total
+= sizeof(pi
);
1240 if (tun
->flags
& TUN_VNET_HDR
) {
1241 struct virtio_net_hdr gso
= { 0 }; /* no info leak */
1242 if ((len
-= tun
->vnet_hdr_sz
) < 0)
1245 if (skb_is_gso(skb
)) {
1246 struct skb_shared_info
*sinfo
= skb_shinfo(skb
);
1248 /* This is a hint as to how much should be linear. */
1249 gso
.hdr_len
= skb_headlen(skb
);
1250 gso
.gso_size
= sinfo
->gso_size
;
1251 if (sinfo
->gso_type
& SKB_GSO_TCPV4
)
1252 gso
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV4
;
1253 else if (sinfo
->gso_type
& SKB_GSO_TCPV6
)
1254 gso
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV6
;
1255 else if (sinfo
->gso_type
& SKB_GSO_UDP
)
1256 gso
.gso_type
= VIRTIO_NET_HDR_GSO_UDP
;
1258 pr_err("unexpected GSO type: "
1259 "0x%x, gso_size %d, hdr_len %d\n",
1260 sinfo
->gso_type
, gso
.gso_size
,
1262 print_hex_dump(KERN_ERR
, "tun: ",
1265 min((int)gso
.hdr_len
, 64), true);
1269 if (sinfo
->gso_type
& SKB_GSO_TCP_ECN
)
1270 gso
.gso_type
|= VIRTIO_NET_HDR_GSO_ECN
;
1272 gso
.gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
1274 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1275 gso
.flags
= VIRTIO_NET_HDR_F_NEEDS_CSUM
;
1276 gso
.csum_start
= skb_checksum_start_offset(skb
);
1277 gso
.csum_offset
= skb
->csum_offset
;
1278 } else if (skb
->ip_summed
== CHECKSUM_UNNECESSARY
) {
1279 gso
.flags
= VIRTIO_NET_HDR_F_DATA_VALID
;
1280 } /* else everything is zero */
1282 if (unlikely(memcpy_toiovecend(iv
, (void *)&gso
, total
,
1285 total
+= tun
->vnet_hdr_sz
;
1288 len
= min_t(int, skb
->len
, len
);
1290 skb_copy_datagram_const_iovec(skb
, 0, iv
, total
, len
);
1293 tun
->dev
->stats
.tx_packets
++;
1294 tun
->dev
->stats
.tx_bytes
+= len
;
1299 static ssize_t
tun_do_read(struct tun_struct
*tun
, struct tun_file
*tfile
,
1300 struct kiocb
*iocb
, const struct iovec
*iv
,
1301 ssize_t len
, int noblock
)
1303 DECLARE_WAITQUEUE(wait
, current
);
1304 struct sk_buff
*skb
;
1307 tun_debug(KERN_INFO
, tun
, "tun_do_read\n");
1309 if (unlikely(!noblock
))
1310 add_wait_queue(&tfile
->wq
.wait
, &wait
);
1312 current
->state
= TASK_INTERRUPTIBLE
;
1314 /* Read frames from the queue */
1315 if (!(skb
= skb_dequeue(&tfile
->socket
.sk
->sk_receive_queue
))) {
1320 if (signal_pending(current
)) {
1324 if (tun
->dev
->reg_state
!= NETREG_REGISTERED
) {
1329 /* Nothing to read, let's sleep */
1334 ret
= tun_put_user(tun
, tfile
, skb
, iv
, len
);
1339 current
->state
= TASK_RUNNING
;
1340 if (unlikely(!noblock
))
1341 remove_wait_queue(&tfile
->wq
.wait
, &wait
);
1346 static ssize_t
tun_chr_aio_read(struct kiocb
*iocb
, const struct iovec
*iv
,
1347 unsigned long count
, loff_t pos
)
1349 struct file
*file
= iocb
->ki_filp
;
1350 struct tun_file
*tfile
= file
->private_data
;
1351 struct tun_struct
*tun
= __tun_get(tfile
);
1356 len
= iov_length(iv
, count
);
1362 ret
= tun_do_read(tun
, tfile
, iocb
, iv
, len
,
1363 file
->f_flags
& O_NONBLOCK
);
1364 ret
= min_t(ssize_t
, ret
, len
);
1370 static void tun_free_netdev(struct net_device
*dev
)
1372 struct tun_struct
*tun
= netdev_priv(dev
);
1374 BUG_ON(!(list_empty(&tun
->disabled
)));
1375 tun_flow_uninit(tun
);
1379 static void tun_setup(struct net_device
*dev
)
1381 struct tun_struct
*tun
= netdev_priv(dev
);
1383 tun
->owner
= INVALID_UID
;
1384 tun
->group
= INVALID_GID
;
1386 dev
->ethtool_ops
= &tun_ethtool_ops
;
1387 dev
->destructor
= tun_free_netdev
;
1390 /* Trivial set of netlink ops to allow deleting tun or tap
1391 * device with netlink.
1393 static int tun_validate(struct nlattr
*tb
[], struct nlattr
*data
[])
1398 static struct rtnl_link_ops tun_link_ops __read_mostly
= {
1400 .priv_size
= sizeof(struct tun_struct
),
1402 .validate
= tun_validate
,
1405 static void tun_sock_write_space(struct sock
*sk
)
1407 struct tun_file
*tfile
;
1408 wait_queue_head_t
*wqueue
;
1410 if (!sock_writeable(sk
))
1413 if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
))
1416 wqueue
= sk_sleep(sk
);
1417 if (wqueue
&& waitqueue_active(wqueue
))
1418 wake_up_interruptible_sync_poll(wqueue
, POLLOUT
|
1419 POLLWRNORM
| POLLWRBAND
);
1421 tfile
= container_of(sk
, struct tun_file
, sk
);
1422 kill_fasync(&tfile
->fasync
, SIGIO
, POLL_OUT
);
1425 static int tun_sendmsg(struct kiocb
*iocb
, struct socket
*sock
,
1426 struct msghdr
*m
, size_t total_len
)
1429 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
1430 struct tun_struct
*tun
= __tun_get(tfile
);
1434 ret
= tun_get_user(tun
, tfile
, m
->msg_control
, m
->msg_iov
, total_len
,
1435 m
->msg_iovlen
, m
->msg_flags
& MSG_DONTWAIT
);
1441 static int tun_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
1442 struct msghdr
*m
, size_t total_len
,
1445 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
1446 struct tun_struct
*tun
= __tun_get(tfile
);
1452 if (flags
& ~(MSG_DONTWAIT
|MSG_TRUNC
))
1454 ret
= tun_do_read(tun
, tfile
, iocb
, m
->msg_iov
, total_len
,
1455 flags
& MSG_DONTWAIT
);
1456 if (ret
> total_len
) {
1457 m
->msg_flags
|= MSG_TRUNC
;
1458 ret
= flags
& MSG_TRUNC
? ret
: total_len
;
1464 static int tun_release(struct socket
*sock
)
1471 /* Ops structure to mimic raw sockets with tun */
1472 static const struct proto_ops tun_socket_ops
= {
1473 .sendmsg
= tun_sendmsg
,
1474 .recvmsg
= tun_recvmsg
,
1475 .release
= tun_release
,
1478 static struct proto tun_proto
= {
1480 .owner
= THIS_MODULE
,
1481 .obj_size
= sizeof(struct tun_file
),
1484 static int tun_flags(struct tun_struct
*tun
)
1488 if (tun
->flags
& TUN_TUN_DEV
)
1493 if (tun
->flags
& TUN_NO_PI
)
1496 /* This flag has no real effect. We track the value for backwards
1499 if (tun
->flags
& TUN_ONE_QUEUE
)
1500 flags
|= IFF_ONE_QUEUE
;
1502 if (tun
->flags
& TUN_VNET_HDR
)
1503 flags
|= IFF_VNET_HDR
;
1505 if (tun
->flags
& TUN_TAP_MQ
)
1506 flags
|= IFF_MULTI_QUEUE
;
1511 static ssize_t
tun_show_flags(struct device
*dev
, struct device_attribute
*attr
,
1514 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
1515 return sprintf(buf
, "0x%x\n", tun_flags(tun
));
1518 static ssize_t
tun_show_owner(struct device
*dev
, struct device_attribute
*attr
,
1521 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
1522 return uid_valid(tun
->owner
)?
1523 sprintf(buf
, "%u\n",
1524 from_kuid_munged(current_user_ns(), tun
->owner
)):
1525 sprintf(buf
, "-1\n");
1528 static ssize_t
tun_show_group(struct device
*dev
, struct device_attribute
*attr
,
1531 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
1532 return gid_valid(tun
->group
) ?
1533 sprintf(buf
, "%u\n",
1534 from_kgid_munged(current_user_ns(), tun
->group
)):
1535 sprintf(buf
, "-1\n");
1538 static DEVICE_ATTR(tun_flags
, 0444, tun_show_flags
, NULL
);
1539 static DEVICE_ATTR(owner
, 0444, tun_show_owner
, NULL
);
1540 static DEVICE_ATTR(group
, 0444, tun_show_group
, NULL
);
1542 static int tun_set_iff(struct net
*net
, struct file
*file
, struct ifreq
*ifr
)
1544 struct tun_struct
*tun
;
1545 struct tun_file
*tfile
= file
->private_data
;
1546 struct net_device
*dev
;
1549 if (tfile
->detached
)
1552 dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
1554 if (ifr
->ifr_flags
& IFF_TUN_EXCL
)
1556 if ((ifr
->ifr_flags
& IFF_TUN
) && dev
->netdev_ops
== &tun_netdev_ops
)
1557 tun
= netdev_priv(dev
);
1558 else if ((ifr
->ifr_flags
& IFF_TAP
) && dev
->netdev_ops
== &tap_netdev_ops
)
1559 tun
= netdev_priv(dev
);
1563 if (tun_not_capable(tun
))
1565 err
= security_tun_dev_attach(tfile
->socket
.sk
);
1569 err
= tun_attach(tun
, file
);
1573 if (tun
->flags
& TUN_TAP_MQ
&&
1574 (tun
->numqueues
+ tun
->numdisabled
> 1))
1579 unsigned long flags
= 0;
1581 if (!ns_capable(net
->user_ns
, CAP_NET_ADMIN
))
1583 err
= security_tun_dev_create();
1588 if (ifr
->ifr_flags
& IFF_TUN
) {
1590 flags
|= TUN_TUN_DEV
;
1592 } else if (ifr
->ifr_flags
& IFF_TAP
) {
1594 flags
|= TUN_TAP_DEV
;
1600 name
= ifr
->ifr_name
;
1602 dev
= alloc_netdev_mqs(sizeof(struct tun_struct
), name
,
1604 MAX_TAP_QUEUES
, MAX_TAP_QUEUES
);
1608 dev_net_set(dev
, net
);
1609 dev
->rtnl_link_ops
= &tun_link_ops
;
1611 tun
= netdev_priv(dev
);
1614 tun
->txflt
.count
= 0;
1615 tun
->vnet_hdr_sz
= sizeof(struct virtio_net_hdr
);
1617 tun
->filter_attached
= false;
1618 tun
->sndbuf
= tfile
->socket
.sk
->sk_sndbuf
;
1620 spin_lock_init(&tun
->lock
);
1622 security_tun_dev_post_create(&tfile
->sk
);
1626 err
= tun_flow_init(tun
);
1630 dev
->hw_features
= NETIF_F_SG
| NETIF_F_FRAGLIST
|
1632 dev
->features
= dev
->hw_features
;
1634 INIT_LIST_HEAD(&tun
->disabled
);
1635 err
= tun_attach(tun
, file
);
1639 err
= register_netdevice(tun
->dev
);
1643 if (device_create_file(&tun
->dev
->dev
, &dev_attr_tun_flags
) ||
1644 device_create_file(&tun
->dev
->dev
, &dev_attr_owner
) ||
1645 device_create_file(&tun
->dev
->dev
, &dev_attr_group
))
1646 pr_err("Failed to create tun sysfs files\n");
1648 netif_carrier_on(tun
->dev
);
1651 tun_debug(KERN_INFO
, tun
, "tun_set_iff\n");
1653 if (ifr
->ifr_flags
& IFF_NO_PI
)
1654 tun
->flags
|= TUN_NO_PI
;
1656 tun
->flags
&= ~TUN_NO_PI
;
1658 /* This flag has no real effect. We track the value for backwards
1661 if (ifr
->ifr_flags
& IFF_ONE_QUEUE
)
1662 tun
->flags
|= TUN_ONE_QUEUE
;
1664 tun
->flags
&= ~TUN_ONE_QUEUE
;
1666 if (ifr
->ifr_flags
& IFF_VNET_HDR
)
1667 tun
->flags
|= TUN_VNET_HDR
;
1669 tun
->flags
&= ~TUN_VNET_HDR
;
1671 if (ifr
->ifr_flags
& IFF_MULTI_QUEUE
)
1672 tun
->flags
|= TUN_TAP_MQ
;
1674 tun
->flags
&= ~TUN_TAP_MQ
;
1676 /* Make sure persistent devices do not get stuck in
1679 if (netif_running(tun
->dev
))
1680 netif_tx_wake_all_queues(tun
->dev
);
1682 strcpy(ifr
->ifr_name
, tun
->dev
->name
);
1690 static void tun_get_iff(struct net
*net
, struct tun_struct
*tun
,
1693 tun_debug(KERN_INFO
, tun
, "tun_get_iff\n");
1695 strcpy(ifr
->ifr_name
, tun
->dev
->name
);
1697 ifr
->ifr_flags
= tun_flags(tun
);
1701 /* This is like a cut-down ethtool ops, except done via tun fd so no
1702 * privs required. */
1703 static int set_offload(struct tun_struct
*tun
, unsigned long arg
)
1705 netdev_features_t features
= 0;
1707 if (arg
& TUN_F_CSUM
) {
1708 features
|= NETIF_F_HW_CSUM
;
1711 if (arg
& (TUN_F_TSO4
|TUN_F_TSO6
)) {
1712 if (arg
& TUN_F_TSO_ECN
) {
1713 features
|= NETIF_F_TSO_ECN
;
1714 arg
&= ~TUN_F_TSO_ECN
;
1716 if (arg
& TUN_F_TSO4
)
1717 features
|= NETIF_F_TSO
;
1718 if (arg
& TUN_F_TSO6
)
1719 features
|= NETIF_F_TSO6
;
1720 arg
&= ~(TUN_F_TSO4
|TUN_F_TSO6
);
1723 if (arg
& TUN_F_UFO
) {
1724 features
|= NETIF_F_UFO
;
1729 /* This gives the user a way to test for new features in future by
1730 * trying to set them. */
1734 tun
->set_features
= features
;
1735 netdev_update_features(tun
->dev
);
1740 static void tun_detach_filter(struct tun_struct
*tun
, int n
)
1743 struct tun_file
*tfile
;
1745 for (i
= 0; i
< n
; i
++) {
1746 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
1747 sk_detach_filter(tfile
->socket
.sk
);
1750 tun
->filter_attached
= false;
1753 static int tun_attach_filter(struct tun_struct
*tun
)
1756 struct tun_file
*tfile
;
1758 for (i
= 0; i
< tun
->numqueues
; i
++) {
1759 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
1760 ret
= sk_attach_filter(&tun
->fprog
, tfile
->socket
.sk
);
1762 tun_detach_filter(tun
, i
);
1767 tun
->filter_attached
= true;
1771 static void tun_set_sndbuf(struct tun_struct
*tun
)
1773 struct tun_file
*tfile
;
1776 for (i
= 0; i
< tun
->numqueues
; i
++) {
1777 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
1778 tfile
->socket
.sk
->sk_sndbuf
= tun
->sndbuf
;
1782 static int tun_set_queue(struct file
*file
, struct ifreq
*ifr
)
1784 struct tun_file
*tfile
= file
->private_data
;
1785 struct tun_struct
*tun
;
1790 if (ifr
->ifr_flags
& IFF_ATTACH_QUEUE
) {
1791 tun
= tfile
->detached
;
1795 ret
= tun_attach(tun
, file
);
1796 } else if (ifr
->ifr_flags
& IFF_DETACH_QUEUE
) {
1797 tun
= rtnl_dereference(tfile
->tun
);
1798 if (!tun
|| !(tun
->flags
& TUN_TAP_MQ
))
1801 __tun_detach(tfile
, false);
1809 static long __tun_chr_ioctl(struct file
*file
, unsigned int cmd
,
1810 unsigned long arg
, int ifreq_len
)
1812 struct tun_file
*tfile
= file
->private_data
;
1813 struct tun_struct
*tun
;
1814 void __user
* argp
= (void __user
*)arg
;
1822 if (cmd
== TUNSETIFF
|| cmd
== TUNSETQUEUE
|| _IOC_TYPE(cmd
) == 0x89) {
1823 if (copy_from_user(&ifr
, argp
, ifreq_len
))
1826 memset(&ifr
, 0, sizeof(ifr
));
1828 if (cmd
== TUNGETFEATURES
) {
1829 /* Currently this just means: "what IFF flags are valid?".
1830 * This is needed because we never checked for invalid flags on
1832 return put_user(IFF_TUN
| IFF_TAP
| IFF_NO_PI
| IFF_ONE_QUEUE
|
1833 IFF_VNET_HDR
| IFF_MULTI_QUEUE
,
1834 (unsigned int __user
*)argp
);
1835 } else if (cmd
== TUNSETQUEUE
)
1836 return tun_set_queue(file
, &ifr
);
1841 tun
= __tun_get(tfile
);
1842 if (cmd
== TUNSETIFF
&& !tun
) {
1843 ifr
.ifr_name
[IFNAMSIZ
-1] = '\0';
1845 ret
= tun_set_iff(tfile
->net
, file
, &ifr
);
1850 if (copy_to_user(argp
, &ifr
, ifreq_len
))
1859 tun_debug(KERN_INFO
, tun
, "tun_chr_ioctl cmd %u\n", cmd
);
1864 tun_get_iff(current
->nsproxy
->net_ns
, tun
, &ifr
);
1866 if (copy_to_user(argp
, &ifr
, ifreq_len
))
1871 /* Disable/Enable checksum */
1873 /* [unimplemented] */
1874 tun_debug(KERN_INFO
, tun
, "ignored: set checksum %s\n",
1875 arg
? "disabled" : "enabled");
1879 /* Disable/Enable persist mode. Keep an extra reference to the
1880 * module to prevent the module being unprobed.
1882 if (arg
&& !(tun
->flags
& TUN_PERSIST
)) {
1883 tun
->flags
|= TUN_PERSIST
;
1884 __module_get(THIS_MODULE
);
1886 if (!arg
&& (tun
->flags
& TUN_PERSIST
)) {
1887 tun
->flags
&= ~TUN_PERSIST
;
1888 module_put(THIS_MODULE
);
1891 tun_debug(KERN_INFO
, tun
, "persist %s\n",
1892 arg
? "enabled" : "disabled");
1896 /* Set owner of the device */
1897 owner
= make_kuid(current_user_ns(), arg
);
1898 if (!uid_valid(owner
)) {
1903 tun_debug(KERN_INFO
, tun
, "owner set to %u\n",
1904 from_kuid(&init_user_ns
, tun
->owner
));
1908 /* Set group of the device */
1909 group
= make_kgid(current_user_ns(), arg
);
1910 if (!gid_valid(group
)) {
1915 tun_debug(KERN_INFO
, tun
, "group set to %u\n",
1916 from_kgid(&init_user_ns
, tun
->group
));
1920 /* Only allow setting the type when the interface is down */
1921 if (tun
->dev
->flags
& IFF_UP
) {
1922 tun_debug(KERN_INFO
, tun
,
1923 "Linktype set failed because interface is up\n");
1926 tun
->dev
->type
= (int) arg
;
1927 tun_debug(KERN_INFO
, tun
, "linktype set to %d\n",
1939 ret
= set_offload(tun
, arg
);
1942 case TUNSETTXFILTER
:
1943 /* Can be set only for TAPs */
1945 if ((tun
->flags
& TUN_TYPE_MASK
) != TUN_TAP_DEV
)
1947 ret
= update_filter(&tun
->txflt
, (void __user
*)arg
);
1951 /* Get hw address */
1952 memcpy(ifr
.ifr_hwaddr
.sa_data
, tun
->dev
->dev_addr
, ETH_ALEN
);
1953 ifr
.ifr_hwaddr
.sa_family
= tun
->dev
->type
;
1954 if (copy_to_user(argp
, &ifr
, ifreq_len
))
1959 /* Set hw address */
1960 tun_debug(KERN_DEBUG
, tun
, "set hw address: %pM\n",
1961 ifr
.ifr_hwaddr
.sa_data
);
1963 ret
= dev_set_mac_address(tun
->dev
, &ifr
.ifr_hwaddr
);
1967 sndbuf
= tfile
->socket
.sk
->sk_sndbuf
;
1968 if (copy_to_user(argp
, &sndbuf
, sizeof(sndbuf
)))
1973 if (copy_from_user(&sndbuf
, argp
, sizeof(sndbuf
))) {
1978 tun
->sndbuf
= sndbuf
;
1979 tun_set_sndbuf(tun
);
1982 case TUNGETVNETHDRSZ
:
1983 vnet_hdr_sz
= tun
->vnet_hdr_sz
;
1984 if (copy_to_user(argp
, &vnet_hdr_sz
, sizeof(vnet_hdr_sz
)))
1988 case TUNSETVNETHDRSZ
:
1989 if (copy_from_user(&vnet_hdr_sz
, argp
, sizeof(vnet_hdr_sz
))) {
1993 if (vnet_hdr_sz
< (int)sizeof(struct virtio_net_hdr
)) {
1998 tun
->vnet_hdr_sz
= vnet_hdr_sz
;
2001 case TUNATTACHFILTER
:
2002 /* Can be set only for TAPs */
2004 if ((tun
->flags
& TUN_TYPE_MASK
) != TUN_TAP_DEV
)
2007 if (copy_from_user(&tun
->fprog
, argp
, sizeof(tun
->fprog
)))
2010 ret
= tun_attach_filter(tun
);
2013 case TUNDETACHFILTER
:
2014 /* Can be set only for TAPs */
2016 if ((tun
->flags
& TUN_TYPE_MASK
) != TUN_TAP_DEV
)
2019 tun_detach_filter(tun
, tun
->numqueues
);
2034 static long tun_chr_ioctl(struct file
*file
,
2035 unsigned int cmd
, unsigned long arg
)
2037 return __tun_chr_ioctl(file
, cmd
, arg
, sizeof (struct ifreq
));
2040 #ifdef CONFIG_COMPAT
2041 static long tun_chr_compat_ioctl(struct file
*file
,
2042 unsigned int cmd
, unsigned long arg
)
2047 case TUNSETTXFILTER
:
2052 arg
= (unsigned long)compat_ptr(arg
);
2055 arg
= (compat_ulong_t
)arg
;
2060 * compat_ifreq is shorter than ifreq, so we must not access beyond
2061 * the end of that structure. All fields that are used in this
2062 * driver are compatible though, we don't need to convert the
2065 return __tun_chr_ioctl(file
, cmd
, arg
, sizeof(struct compat_ifreq
));
2067 #endif /* CONFIG_COMPAT */
2069 static int tun_chr_fasync(int fd
, struct file
*file
, int on
)
2071 struct tun_file
*tfile
= file
->private_data
;
2074 if ((ret
= fasync_helper(fd
, file
, on
, &tfile
->fasync
)) < 0)
2078 ret
= __f_setown(file
, task_pid(current
), PIDTYPE_PID
, 0);
2081 tfile
->flags
|= TUN_FASYNC
;
2083 tfile
->flags
&= ~TUN_FASYNC
;
2089 static int tun_chr_open(struct inode
*inode
, struct file
* file
)
2091 struct tun_file
*tfile
;
2093 DBG1(KERN_INFO
, "tunX: tun_chr_open\n");
2095 tfile
= (struct tun_file
*)sk_alloc(&init_net
, AF_UNSPEC
, GFP_KERNEL
,
2099 rcu_assign_pointer(tfile
->tun
, NULL
);
2100 tfile
->net
= get_net(current
->nsproxy
->net_ns
);
2103 rcu_assign_pointer(tfile
->socket
.wq
, &tfile
->wq
);
2104 init_waitqueue_head(&tfile
->wq
.wait
);
2106 tfile
->socket
.file
= file
;
2107 tfile
->socket
.ops
= &tun_socket_ops
;
2109 sock_init_data(&tfile
->socket
, &tfile
->sk
);
2110 sk_change_net(&tfile
->sk
, tfile
->net
);
2112 tfile
->sk
.sk_write_space
= tun_sock_write_space
;
2113 tfile
->sk
.sk_sndbuf
= INT_MAX
;
2115 file
->private_data
= tfile
;
2116 set_bit(SOCK_EXTERNALLY_ALLOCATED
, &tfile
->socket
.flags
);
2117 INIT_LIST_HEAD(&tfile
->next
);
2122 static int tun_chr_close(struct inode
*inode
, struct file
*file
)
2124 struct tun_file
*tfile
= file
->private_data
;
2125 struct net
*net
= tfile
->net
;
2127 tun_detach(tfile
, true);
2133 static const struct file_operations tun_fops
= {
2134 .owner
= THIS_MODULE
,
2135 .llseek
= no_llseek
,
2136 .read
= do_sync_read
,
2137 .aio_read
= tun_chr_aio_read
,
2138 .write
= do_sync_write
,
2139 .aio_write
= tun_chr_aio_write
,
2140 .poll
= tun_chr_poll
,
2141 .unlocked_ioctl
= tun_chr_ioctl
,
2142 #ifdef CONFIG_COMPAT
2143 .compat_ioctl
= tun_chr_compat_ioctl
,
2145 .open
= tun_chr_open
,
2146 .release
= tun_chr_close
,
2147 .fasync
= tun_chr_fasync
2150 static struct miscdevice tun_miscdev
= {
2153 .nodename
= "net/tun",
2157 /* ethtool interface */
2159 static int tun_get_settings(struct net_device
*dev
, struct ethtool_cmd
*cmd
)
2162 cmd
->advertising
= 0;
2163 ethtool_cmd_speed_set(cmd
, SPEED_10
);
2164 cmd
->duplex
= DUPLEX_FULL
;
2165 cmd
->port
= PORT_TP
;
2166 cmd
->phy_address
= 0;
2167 cmd
->transceiver
= XCVR_INTERNAL
;
2168 cmd
->autoneg
= AUTONEG_DISABLE
;
2174 static void tun_get_drvinfo(struct net_device
*dev
, struct ethtool_drvinfo
*info
)
2176 struct tun_struct
*tun
= netdev_priv(dev
);
2178 strlcpy(info
->driver
, DRV_NAME
, sizeof(info
->driver
));
2179 strlcpy(info
->version
, DRV_VERSION
, sizeof(info
->version
));
2181 switch (tun
->flags
& TUN_TYPE_MASK
) {
2183 strlcpy(info
->bus_info
, "tun", sizeof(info
->bus_info
));
2186 strlcpy(info
->bus_info
, "tap", sizeof(info
->bus_info
));
2191 static u32
tun_get_msglevel(struct net_device
*dev
)
2194 struct tun_struct
*tun
= netdev_priv(dev
);
2201 static void tun_set_msglevel(struct net_device
*dev
, u32 value
)
2204 struct tun_struct
*tun
= netdev_priv(dev
);
2209 static const struct ethtool_ops tun_ethtool_ops
= {
2210 .get_settings
= tun_get_settings
,
2211 .get_drvinfo
= tun_get_drvinfo
,
2212 .get_msglevel
= tun_get_msglevel
,
2213 .set_msglevel
= tun_set_msglevel
,
2214 .get_link
= ethtool_op_get_link
,
2218 static int __init
tun_init(void)
2222 pr_info("%s, %s\n", DRV_DESCRIPTION
, DRV_VERSION
);
2223 pr_info("%s\n", DRV_COPYRIGHT
);
2225 ret
= rtnl_link_register(&tun_link_ops
);
2227 pr_err("Can't register link_ops\n");
2231 ret
= misc_register(&tun_miscdev
);
2233 pr_err("Can't register misc device %d\n", TUN_MINOR
);
2238 rtnl_link_unregister(&tun_link_ops
);
2243 static void tun_cleanup(void)
2245 misc_deregister(&tun_miscdev
);
2246 rtnl_link_unregister(&tun_link_ops
);
2249 /* Get an underlying socket object from tun file. Returns error unless file is
2250 * attached to a device. The returned object works like a packet socket, it
2251 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
2252 * holding a reference to the file for as long as the socket is in use. */
2253 struct socket
*tun_get_socket(struct file
*file
)
2255 struct tun_file
*tfile
;
2256 if (file
->f_op
!= &tun_fops
)
2257 return ERR_PTR(-EINVAL
);
2258 tfile
= file
->private_data
;
2260 return ERR_PTR(-EBADFD
);
2261 return &tfile
->socket
;
2263 EXPORT_SYMBOL_GPL(tun_get_socket
);
2265 module_init(tun_init
);
2266 module_exit(tun_cleanup
);
2267 MODULE_DESCRIPTION(DRV_DESCRIPTION
);
2268 MODULE_AUTHOR(DRV_COPYRIGHT
);
2269 MODULE_LICENSE("GPL");
2270 MODULE_ALIAS_MISCDEV(TUN_MINOR
);
2271 MODULE_ALIAS("devname:net/tun");