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/if_vlan.h>
64 #include <linux/crc32.h>
65 #include <linux/nsproxy.h>
66 #include <linux/virtio_net.h>
67 #include <linux/rcupdate.h>
68 #include <net/net_namespace.h>
69 #include <net/netns/generic.h>
70 #include <net/rtnetlink.h>
73 #include <asm/uaccess.h>
75 /* Uncomment to enable debugging */
76 /* #define TUN_DEBUG 1 */
81 #define tun_debug(level, tun, fmt, args...) \
84 netdev_printk(level, tun->dev, fmt, ##args); \
86 #define DBG1(level, fmt, args...) \
89 printk(level fmt, ##args); \
92 #define tun_debug(level, tun, fmt, args...) \
95 netdev_printk(level, tun->dev, fmt, ##args); \
97 #define DBG1(level, fmt, args...) \
100 printk(level fmt, ##args); \
104 #define GOODCOPY_LEN 128
106 #define FLT_EXACT_COUNT 8
108 unsigned int count
; /* Number of addrs. Zero means disabled */
109 u32 mask
[2]; /* Mask of the hashed addrs */
110 unsigned char addr
[FLT_EXACT_COUNT
][ETH_ALEN
];
113 /* DEFAULT_MAX_NUM_RSS_QUEUES were chosen to let the rx/tx queues allocated for
114 * the netdevice to be fit in one page. So we can make sure the success of
115 * memory allocation. TODO: increase the limit. */
116 #define MAX_TAP_QUEUES DEFAULT_MAX_NUM_RSS_QUEUES
117 #define MAX_TAP_FLOWS 4096
119 #define TUN_FLOW_EXPIRE (3 * HZ)
121 /* A tun_file connects an open character device to a tuntap netdevice. It
122 * also contains all socket related structures (except sock_fprog and tap_filter)
123 * to serve as one transmit queue for tuntap device. The sock_fprog and
124 * tap_filter were kept in tun_struct since they were used for filtering for the
125 * netdevice not for a specific queue (at least I didn't see the requirement for
129 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
130 * other can only be read while rcu_read_lock or rtnl_lock is held.
134 struct socket socket
;
136 struct tun_struct __rcu
*tun
;
138 struct fasync_struct
*fasync
;
139 /* only used for fasnyc */
143 unsigned int ifindex
;
145 struct list_head next
;
146 struct tun_struct
*detached
;
149 struct tun_flow_entry
{
150 struct hlist_node hash_link
;
152 struct tun_struct
*tun
;
157 unsigned long updated
;
160 #define TUN_NUM_FLOW_ENTRIES 1024
162 /* Since the socket were moved to tun_file, to preserve the behavior of persist
163 * device, socket filter, sndbuf and vnet header size were restore when the
164 * file were attached to a persist device.
167 struct tun_file __rcu
*tfiles
[MAX_TAP_QUEUES
];
168 unsigned int numqueues
;
173 struct net_device
*dev
;
174 netdev_features_t set_features
;
175 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
176 NETIF_F_TSO6|NETIF_F_UFO)
180 struct tap_filter txflt
;
181 struct sock_fprog fprog
;
182 /* protected by rtnl lock */
183 bool filter_attached
;
188 struct hlist_head flows
[TUN_NUM_FLOW_ENTRIES
];
189 struct timer_list flow_gc_timer
;
190 unsigned long ageing_time
;
191 unsigned int numdisabled
;
192 struct list_head disabled
;
197 static inline u32
tun_hashfn(u32 rxhash
)
199 return rxhash
& 0x3ff;
202 static struct tun_flow_entry
*tun_flow_find(struct hlist_head
*head
, u32 rxhash
)
204 struct tun_flow_entry
*e
;
206 hlist_for_each_entry_rcu(e
, head
, hash_link
) {
207 if (e
->rxhash
== rxhash
)
213 static struct tun_flow_entry
*tun_flow_create(struct tun_struct
*tun
,
214 struct hlist_head
*head
,
215 u32 rxhash
, u16 queue_index
)
217 struct tun_flow_entry
*e
= kmalloc(sizeof(*e
), GFP_ATOMIC
);
220 tun_debug(KERN_INFO
, tun
, "create flow: hash %u index %u\n",
221 rxhash
, queue_index
);
222 e
->updated
= jiffies
;
225 e
->queue_index
= queue_index
;
227 hlist_add_head_rcu(&e
->hash_link
, head
);
233 static void tun_flow_delete(struct tun_struct
*tun
, struct tun_flow_entry
*e
)
235 tun_debug(KERN_INFO
, tun
, "delete flow: hash %u index %u\n",
236 e
->rxhash
, e
->queue_index
);
237 sock_rps_reset_flow_hash(e
->rps_rxhash
);
238 hlist_del_rcu(&e
->hash_link
);
243 static void tun_flow_flush(struct tun_struct
*tun
)
247 spin_lock_bh(&tun
->lock
);
248 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
249 struct tun_flow_entry
*e
;
250 struct hlist_node
*n
;
252 hlist_for_each_entry_safe(e
, n
, &tun
->flows
[i
], hash_link
)
253 tun_flow_delete(tun
, e
);
255 spin_unlock_bh(&tun
->lock
);
258 static void tun_flow_delete_by_queue(struct tun_struct
*tun
, u16 queue_index
)
262 spin_lock_bh(&tun
->lock
);
263 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
264 struct tun_flow_entry
*e
;
265 struct hlist_node
*n
;
267 hlist_for_each_entry_safe(e
, n
, &tun
->flows
[i
], hash_link
) {
268 if (e
->queue_index
== queue_index
)
269 tun_flow_delete(tun
, e
);
272 spin_unlock_bh(&tun
->lock
);
275 static void tun_flow_cleanup(unsigned long data
)
277 struct tun_struct
*tun
= (struct tun_struct
*)data
;
278 unsigned long delay
= tun
->ageing_time
;
279 unsigned long next_timer
= jiffies
+ delay
;
280 unsigned long count
= 0;
283 tun_debug(KERN_INFO
, tun
, "tun_flow_cleanup\n");
285 spin_lock_bh(&tun
->lock
);
286 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
287 struct tun_flow_entry
*e
;
288 struct hlist_node
*n
;
290 hlist_for_each_entry_safe(e
, n
, &tun
->flows
[i
], hash_link
) {
291 unsigned long this_timer
;
293 this_timer
= e
->updated
+ delay
;
294 if (time_before_eq(this_timer
, jiffies
))
295 tun_flow_delete(tun
, e
);
296 else if (time_before(this_timer
, next_timer
))
297 next_timer
= this_timer
;
302 mod_timer(&tun
->flow_gc_timer
, round_jiffies_up(next_timer
));
303 spin_unlock_bh(&tun
->lock
);
306 static void tun_flow_update(struct tun_struct
*tun
, u32 rxhash
,
307 struct tun_file
*tfile
)
309 struct hlist_head
*head
;
310 struct tun_flow_entry
*e
;
311 unsigned long delay
= tun
->ageing_time
;
312 u16 queue_index
= tfile
->queue_index
;
317 head
= &tun
->flows
[tun_hashfn(rxhash
)];
321 /* We may get a very small possibility of OOO during switching, not
322 * worth to optimize.*/
323 if (tun
->numqueues
== 1 || tfile
->detached
)
326 e
= tun_flow_find(head
, rxhash
);
328 /* TODO: keep queueing to old queue until it's empty? */
329 e
->queue_index
= queue_index
;
330 e
->updated
= jiffies
;
331 sock_rps_record_flow_hash(e
->rps_rxhash
);
333 spin_lock_bh(&tun
->lock
);
334 if (!tun_flow_find(head
, rxhash
) &&
335 tun
->flow_count
< MAX_TAP_FLOWS
)
336 tun_flow_create(tun
, head
, rxhash
, queue_index
);
338 if (!timer_pending(&tun
->flow_gc_timer
))
339 mod_timer(&tun
->flow_gc_timer
,
340 round_jiffies_up(jiffies
+ delay
));
341 spin_unlock_bh(&tun
->lock
);
349 * Save the hash received in the stack receive path and update the
350 * flow_hash table accordingly.
352 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry
*e
, u32 hash
)
354 if (unlikely(e
->rps_rxhash
!= hash
)) {
355 sock_rps_reset_flow_hash(e
->rps_rxhash
);
356 e
->rps_rxhash
= hash
;
360 /* We try to identify a flow through its rxhash first. The reason that
361 * we do not check rxq no. is because some cards(e.g 82599), chooses
362 * the rxq based on the txq where the last packet of the flow comes. As
363 * the userspace application move between processors, we may get a
364 * different rxq no. here. If we could not get rxhash, then we would
365 * hope the rxq no. may help here.
367 static u16
tun_select_queue(struct net_device
*dev
, struct sk_buff
*skb
,
370 struct tun_struct
*tun
= netdev_priv(dev
);
371 struct tun_flow_entry
*e
;
376 numqueues
= ACCESS_ONCE(tun
->numqueues
);
378 txq
= skb_get_hash(skb
);
380 e
= tun_flow_find(&tun
->flows
[tun_hashfn(txq
)], txq
);
382 tun_flow_save_rps_rxhash(e
, txq
);
383 txq
= e
->queue_index
;
385 /* use multiply and shift instead of expensive divide */
386 txq
= ((u64
)txq
* numqueues
) >> 32;
387 } else if (likely(skb_rx_queue_recorded(skb
))) {
388 txq
= skb_get_rx_queue(skb
);
389 while (unlikely(txq
>= numqueues
))
397 static inline bool tun_not_capable(struct tun_struct
*tun
)
399 const struct cred
*cred
= current_cred();
400 struct net
*net
= dev_net(tun
->dev
);
402 return ((uid_valid(tun
->owner
) && !uid_eq(cred
->euid
, tun
->owner
)) ||
403 (gid_valid(tun
->group
) && !in_egroup_p(tun
->group
))) &&
404 !ns_capable(net
->user_ns
, CAP_NET_ADMIN
);
407 static void tun_set_real_num_queues(struct tun_struct
*tun
)
409 netif_set_real_num_tx_queues(tun
->dev
, tun
->numqueues
);
410 netif_set_real_num_rx_queues(tun
->dev
, tun
->numqueues
);
413 static void tun_disable_queue(struct tun_struct
*tun
, struct tun_file
*tfile
)
415 tfile
->detached
= tun
;
416 list_add_tail(&tfile
->next
, &tun
->disabled
);
420 static struct tun_struct
*tun_enable_queue(struct tun_file
*tfile
)
422 struct tun_struct
*tun
= tfile
->detached
;
424 tfile
->detached
= NULL
;
425 list_del_init(&tfile
->next
);
430 static void tun_queue_purge(struct tun_file
*tfile
)
432 skb_queue_purge(&tfile
->sk
.sk_receive_queue
);
433 skb_queue_purge(&tfile
->sk
.sk_error_queue
);
436 static void __tun_detach(struct tun_file
*tfile
, bool clean
)
438 struct tun_file
*ntfile
;
439 struct tun_struct
*tun
;
441 tun
= rtnl_dereference(tfile
->tun
);
443 if (tun
&& !tfile
->detached
) {
444 u16 index
= tfile
->queue_index
;
445 BUG_ON(index
>= tun
->numqueues
);
447 rcu_assign_pointer(tun
->tfiles
[index
],
448 tun
->tfiles
[tun
->numqueues
- 1]);
449 ntfile
= rtnl_dereference(tun
->tfiles
[index
]);
450 ntfile
->queue_index
= index
;
454 rcu_assign_pointer(tfile
->tun
, NULL
);
455 sock_put(&tfile
->sk
);
457 tun_disable_queue(tun
, tfile
);
460 tun_flow_delete_by_queue(tun
, tun
->numqueues
+ 1);
461 /* Drop read queue */
462 tun_queue_purge(tfile
);
463 tun_set_real_num_queues(tun
);
464 } else if (tfile
->detached
&& clean
) {
465 tun
= tun_enable_queue(tfile
);
466 sock_put(&tfile
->sk
);
470 if (tun
&& tun
->numqueues
== 0 && tun
->numdisabled
== 0) {
471 netif_carrier_off(tun
->dev
);
473 if (!(tun
->flags
& TUN_PERSIST
) &&
474 tun
->dev
->reg_state
== NETREG_REGISTERED
)
475 unregister_netdevice(tun
->dev
);
478 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED
,
479 &tfile
->socket
.flags
));
480 sk_release_kernel(&tfile
->sk
);
484 static void tun_detach(struct tun_file
*tfile
, bool clean
)
487 __tun_detach(tfile
, clean
);
491 static void tun_detach_all(struct net_device
*dev
)
493 struct tun_struct
*tun
= netdev_priv(dev
);
494 struct tun_file
*tfile
, *tmp
;
495 int i
, n
= tun
->numqueues
;
497 for (i
= 0; i
< n
; i
++) {
498 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
500 wake_up_all(&tfile
->wq
.wait
);
501 rcu_assign_pointer(tfile
->tun
, NULL
);
504 list_for_each_entry(tfile
, &tun
->disabled
, next
) {
505 wake_up_all(&tfile
->wq
.wait
);
506 rcu_assign_pointer(tfile
->tun
, NULL
);
508 BUG_ON(tun
->numqueues
!= 0);
511 for (i
= 0; i
< n
; i
++) {
512 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
513 /* Drop read queue */
514 tun_queue_purge(tfile
);
515 sock_put(&tfile
->sk
);
517 list_for_each_entry_safe(tfile
, tmp
, &tun
->disabled
, next
) {
518 tun_enable_queue(tfile
);
519 tun_queue_purge(tfile
);
520 sock_put(&tfile
->sk
);
522 BUG_ON(tun
->numdisabled
!= 0);
524 if (tun
->flags
& TUN_PERSIST
)
525 module_put(THIS_MODULE
);
528 static int tun_attach(struct tun_struct
*tun
, struct file
*file
, bool skip_filter
)
530 struct tun_file
*tfile
= file
->private_data
;
533 err
= security_tun_dev_attach(tfile
->socket
.sk
, tun
->security
);
538 if (rtnl_dereference(tfile
->tun
) && !tfile
->detached
)
542 if (!(tun
->flags
& TUN_TAP_MQ
) && tun
->numqueues
== 1)
546 if (!tfile
->detached
&&
547 tun
->numqueues
+ tun
->numdisabled
== MAX_TAP_QUEUES
)
552 /* Re-attach the filter to persist device */
553 if (!skip_filter
&& (tun
->filter_attached
== true)) {
554 err
= sk_attach_filter(&tun
->fprog
, tfile
->socket
.sk
);
558 tfile
->queue_index
= tun
->numqueues
;
559 rcu_assign_pointer(tfile
->tun
, tun
);
560 rcu_assign_pointer(tun
->tfiles
[tun
->numqueues
], tfile
);
564 tun_enable_queue(tfile
);
566 sock_hold(&tfile
->sk
);
568 tun_set_real_num_queues(tun
);
570 /* device is allowed to go away first, so no need to hold extra
578 static struct tun_struct
*__tun_get(struct tun_file
*tfile
)
580 struct tun_struct
*tun
;
583 tun
= rcu_dereference(tfile
->tun
);
591 static struct tun_struct
*tun_get(struct file
*file
)
593 return __tun_get(file
->private_data
);
596 static void tun_put(struct tun_struct
*tun
)
602 static void addr_hash_set(u32
*mask
, const u8
*addr
)
604 int n
= ether_crc(ETH_ALEN
, addr
) >> 26;
605 mask
[n
>> 5] |= (1 << (n
& 31));
608 static unsigned int addr_hash_test(const u32
*mask
, const u8
*addr
)
610 int n
= ether_crc(ETH_ALEN
, addr
) >> 26;
611 return mask
[n
>> 5] & (1 << (n
& 31));
614 static int update_filter(struct tap_filter
*filter
, void __user
*arg
)
616 struct { u8 u
[ETH_ALEN
]; } *addr
;
617 struct tun_filter uf
;
618 int err
, alen
, n
, nexact
;
620 if (copy_from_user(&uf
, arg
, sizeof(uf
)))
629 alen
= ETH_ALEN
* uf
.count
;
630 addr
= kmalloc(alen
, GFP_KERNEL
);
634 if (copy_from_user(addr
, arg
+ sizeof(uf
), alen
)) {
639 /* The filter is updated without holding any locks. Which is
640 * perfectly safe. We disable it first and in the worst
641 * case we'll accept a few undesired packets. */
645 /* Use first set of addresses as an exact filter */
646 for (n
= 0; n
< uf
.count
&& n
< FLT_EXACT_COUNT
; n
++)
647 memcpy(filter
->addr
[n
], addr
[n
].u
, ETH_ALEN
);
651 /* Remaining multicast addresses are hashed,
652 * unicast will leave the filter disabled. */
653 memset(filter
->mask
, 0, sizeof(filter
->mask
));
654 for (; n
< uf
.count
; n
++) {
655 if (!is_multicast_ether_addr(addr
[n
].u
)) {
656 err
= 0; /* no filter */
659 addr_hash_set(filter
->mask
, addr
[n
].u
);
662 /* For ALLMULTI just set the mask to all ones.
663 * This overrides the mask populated above. */
664 if ((uf
.flags
& TUN_FLT_ALLMULTI
))
665 memset(filter
->mask
, ~0, sizeof(filter
->mask
));
667 /* Now enable the filter */
669 filter
->count
= nexact
;
671 /* Return the number of exact filters */
679 /* Returns: 0 - drop, !=0 - accept */
680 static int run_filter(struct tap_filter
*filter
, const struct sk_buff
*skb
)
682 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
684 struct ethhdr
*eh
= (struct ethhdr
*) skb
->data
;
688 for (i
= 0; i
< filter
->count
; i
++)
689 if (ether_addr_equal(eh
->h_dest
, filter
->addr
[i
]))
692 /* Inexact match (multicast only) */
693 if (is_multicast_ether_addr(eh
->h_dest
))
694 return addr_hash_test(filter
->mask
, eh
->h_dest
);
700 * Checks whether the packet is accepted or not.
701 * Returns: 0 - drop, !=0 - accept
703 static int check_filter(struct tap_filter
*filter
, const struct sk_buff
*skb
)
708 return run_filter(filter
, skb
);
711 /* Network device part of the driver */
713 static const struct ethtool_ops tun_ethtool_ops
;
715 /* Net device detach from fd. */
716 static void tun_net_uninit(struct net_device
*dev
)
721 /* Net device open. */
722 static int tun_net_open(struct net_device
*dev
)
724 netif_tx_start_all_queues(dev
);
728 /* Net device close. */
729 static int tun_net_close(struct net_device
*dev
)
731 netif_tx_stop_all_queues(dev
);
735 /* Net device start xmit */
736 static netdev_tx_t
tun_net_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
738 struct tun_struct
*tun
= netdev_priv(dev
);
739 int txq
= skb
->queue_mapping
;
740 struct tun_file
*tfile
;
744 tfile
= rcu_dereference(tun
->tfiles
[txq
]);
745 numqueues
= ACCESS_ONCE(tun
->numqueues
);
747 /* Drop packet if interface is not attached */
748 if (txq
>= numqueues
)
751 if (numqueues
== 1) {
752 /* Select queue was not called for the skbuff, so we extract the
753 * RPS hash and save it into the flow_table here.
757 rxhash
= skb_get_hash(skb
);
759 struct tun_flow_entry
*e
;
760 e
= tun_flow_find(&tun
->flows
[tun_hashfn(rxhash
)],
763 tun_flow_save_rps_rxhash(e
, rxhash
);
767 tun_debug(KERN_INFO
, tun
, "tun_net_xmit %d\n", skb
->len
);
771 /* Drop if the filter does not like it.
772 * This is a noop if the filter is disabled.
773 * Filter can be enabled only for the TAP devices. */
774 if (!check_filter(&tun
->txflt
, skb
))
777 if (tfile
->socket
.sk
->sk_filter
&&
778 sk_filter(tfile
->socket
.sk
, skb
))
781 /* Limit the number of packets queued by dividing txq length with the
784 if (skb_queue_len(&tfile
->socket
.sk
->sk_receive_queue
) * numqueues
785 >= dev
->tx_queue_len
)
788 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
792 sock_tx_timestamp(skb
->sk
, &skb_shinfo(skb
)->tx_flags
);
793 sw_tx_timestamp(skb
);
796 /* Orphan the skb - required as we might hang on to it
797 * for indefinite time.
804 skb_queue_tail(&tfile
->socket
.sk
->sk_receive_queue
, skb
);
806 /* Notify and wake up reader process */
807 if (tfile
->flags
& TUN_FASYNC
)
808 kill_fasync(&tfile
->fasync
, SIGIO
, POLL_IN
);
809 wake_up_interruptible_poll(&tfile
->wq
.wait
, POLLIN
|
810 POLLRDNORM
| POLLRDBAND
);
816 dev
->stats
.tx_dropped
++;
823 static void tun_net_mclist(struct net_device
*dev
)
826 * This callback is supposed to deal with mc filter in
827 * _rx_ path and has nothing to do with the _tx_ path.
828 * In rx path we always accept everything userspace gives us.
833 #define MAX_MTU 65535
836 tun_net_change_mtu(struct net_device
*dev
, int new_mtu
)
838 if (new_mtu
< MIN_MTU
|| new_mtu
+ dev
->hard_header_len
> MAX_MTU
)
844 static netdev_features_t
tun_net_fix_features(struct net_device
*dev
,
845 netdev_features_t features
)
847 struct tun_struct
*tun
= netdev_priv(dev
);
849 return (features
& tun
->set_features
) | (features
& ~TUN_USER_FEATURES
);
851 #ifdef CONFIG_NET_POLL_CONTROLLER
852 static void tun_poll_controller(struct net_device
*dev
)
855 * Tun only receives frames when:
856 * 1) the char device endpoint gets data from user space
857 * 2) the tun socket gets a sendmsg call from user space
858 * Since both of those are synchronous operations, we are guaranteed
859 * never to have pending data when we poll for it
860 * so there is nothing to do here but return.
861 * We need this though so netpoll recognizes us as an interface that
862 * supports polling, which enables bridge devices in virt setups to
863 * still use netconsole
868 static const struct net_device_ops tun_netdev_ops
= {
869 .ndo_uninit
= tun_net_uninit
,
870 .ndo_open
= tun_net_open
,
871 .ndo_stop
= tun_net_close
,
872 .ndo_start_xmit
= tun_net_xmit
,
873 .ndo_change_mtu
= tun_net_change_mtu
,
874 .ndo_fix_features
= tun_net_fix_features
,
875 .ndo_select_queue
= tun_select_queue
,
876 #ifdef CONFIG_NET_POLL_CONTROLLER
877 .ndo_poll_controller
= tun_poll_controller
,
881 static const struct net_device_ops tap_netdev_ops
= {
882 .ndo_uninit
= tun_net_uninit
,
883 .ndo_open
= tun_net_open
,
884 .ndo_stop
= tun_net_close
,
885 .ndo_start_xmit
= tun_net_xmit
,
886 .ndo_change_mtu
= tun_net_change_mtu
,
887 .ndo_fix_features
= tun_net_fix_features
,
888 .ndo_set_rx_mode
= tun_net_mclist
,
889 .ndo_set_mac_address
= eth_mac_addr
,
890 .ndo_validate_addr
= eth_validate_addr
,
891 .ndo_select_queue
= tun_select_queue
,
892 #ifdef CONFIG_NET_POLL_CONTROLLER
893 .ndo_poll_controller
= tun_poll_controller
,
897 static void tun_flow_init(struct tun_struct
*tun
)
901 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++)
902 INIT_HLIST_HEAD(&tun
->flows
[i
]);
904 tun
->ageing_time
= TUN_FLOW_EXPIRE
;
905 setup_timer(&tun
->flow_gc_timer
, tun_flow_cleanup
, (unsigned long)tun
);
906 mod_timer(&tun
->flow_gc_timer
,
907 round_jiffies_up(jiffies
+ tun
->ageing_time
));
910 static void tun_flow_uninit(struct tun_struct
*tun
)
912 del_timer_sync(&tun
->flow_gc_timer
);
916 /* Initialize net device. */
917 static void tun_net_init(struct net_device
*dev
)
919 struct tun_struct
*tun
= netdev_priv(dev
);
921 switch (tun
->flags
& TUN_TYPE_MASK
) {
923 dev
->netdev_ops
= &tun_netdev_ops
;
925 /* Point-to-Point TUN Device */
926 dev
->hard_header_len
= 0;
930 /* Zero header length */
931 dev
->type
= ARPHRD_NONE
;
932 dev
->flags
= IFF_POINTOPOINT
| IFF_NOARP
| IFF_MULTICAST
;
933 dev
->tx_queue_len
= TUN_READQ_SIZE
; /* We prefer our own queue length */
937 dev
->netdev_ops
= &tap_netdev_ops
;
938 /* Ethernet TAP Device */
940 dev
->priv_flags
&= ~IFF_TX_SKB_SHARING
;
941 dev
->priv_flags
|= IFF_LIVE_ADDR_CHANGE
;
943 eth_hw_addr_random(dev
);
945 dev
->tx_queue_len
= TUN_READQ_SIZE
; /* We prefer our own queue length */
950 /* Character device part */
953 static unsigned int tun_chr_poll(struct file
*file
, poll_table
*wait
)
955 struct tun_file
*tfile
= file
->private_data
;
956 struct tun_struct
*tun
= __tun_get(tfile
);
958 unsigned int mask
= 0;
963 sk
= tfile
->socket
.sk
;
965 tun_debug(KERN_INFO
, tun
, "tun_chr_poll\n");
967 poll_wait(file
, &tfile
->wq
.wait
, wait
);
969 if (!skb_queue_empty(&sk
->sk_receive_queue
))
970 mask
|= POLLIN
| POLLRDNORM
;
972 if (sock_writeable(sk
) ||
973 (!test_and_set_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
) &&
975 mask
|= POLLOUT
| POLLWRNORM
;
977 if (tun
->dev
->reg_state
!= NETREG_REGISTERED
)
984 /* prepad is the amount to reserve at front. len is length after that.
985 * linear is a hint as to how much to copy (usually headers). */
986 static struct sk_buff
*tun_alloc_skb(struct tun_file
*tfile
,
987 size_t prepad
, size_t len
,
988 size_t linear
, int noblock
)
990 struct sock
*sk
= tfile
->socket
.sk
;
994 /* Under a page? Don't bother with paged skb. */
995 if (prepad
+ len
< PAGE_SIZE
|| !linear
)
998 skb
= sock_alloc_send_pskb(sk
, prepad
+ linear
, len
- linear
, noblock
,
1001 return ERR_PTR(err
);
1003 skb_reserve(skb
, prepad
);
1004 skb_put(skb
, linear
);
1005 skb
->data_len
= len
- linear
;
1006 skb
->len
+= len
- linear
;
1011 /* Get packet from user space buffer */
1012 static ssize_t
tun_get_user(struct tun_struct
*tun
, struct tun_file
*tfile
,
1013 void *msg_control
, const struct iovec
*iv
,
1014 size_t total_len
, size_t count
, int noblock
)
1016 struct tun_pi pi
= { 0, cpu_to_be16(ETH_P_IP
) };
1017 struct sk_buff
*skb
;
1018 size_t len
= total_len
, align
= NET_SKB_PAD
, linear
;
1019 struct virtio_net_hdr gso
= { 0 };
1023 bool zerocopy
= false;
1027 if (!(tun
->flags
& TUN_NO_PI
)) {
1028 if (len
< sizeof(pi
))
1032 if (memcpy_fromiovecend((void *)&pi
, iv
, 0, sizeof(pi
)))
1034 offset
+= sizeof(pi
);
1037 if (tun
->flags
& TUN_VNET_HDR
) {
1038 if (len
< tun
->vnet_hdr_sz
)
1040 len
-= tun
->vnet_hdr_sz
;
1042 if (memcpy_fromiovecend((void *)&gso
, iv
, offset
, sizeof(gso
)))
1045 if ((gso
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) &&
1046 gso
.csum_start
+ gso
.csum_offset
+ 2 > gso
.hdr_len
)
1047 gso
.hdr_len
= gso
.csum_start
+ gso
.csum_offset
+ 2;
1049 if (gso
.hdr_len
> len
)
1051 offset
+= tun
->vnet_hdr_sz
;
1054 if ((tun
->flags
& TUN_TYPE_MASK
) == TUN_TAP_DEV
) {
1055 align
+= NET_IP_ALIGN
;
1056 if (unlikely(len
< ETH_HLEN
||
1057 (gso
.hdr_len
&& gso
.hdr_len
< ETH_HLEN
)))
1061 good_linear
= SKB_MAX_HEAD(align
);
1064 /* There are 256 bytes to be copied in skb, so there is
1065 * enough room for skb expand head in case it is used.
1066 * The rest of the buffer is mapped from userspace.
1068 copylen
= gso
.hdr_len
? gso
.hdr_len
: GOODCOPY_LEN
;
1069 if (copylen
> good_linear
)
1070 copylen
= good_linear
;
1072 if (iov_pages(iv
, offset
+ copylen
, count
) <= MAX_SKB_FRAGS
)
1078 if (gso
.hdr_len
> good_linear
)
1079 linear
= good_linear
;
1081 linear
= gso
.hdr_len
;
1084 skb
= tun_alloc_skb(tfile
, align
, copylen
, linear
, noblock
);
1086 if (PTR_ERR(skb
) != -EAGAIN
)
1087 tun
->dev
->stats
.rx_dropped
++;
1088 return PTR_ERR(skb
);
1092 err
= zerocopy_sg_from_iovec(skb
, iv
, offset
, count
);
1094 err
= skb_copy_datagram_from_iovec(skb
, 0, iv
, offset
, len
);
1095 if (!err
&& msg_control
) {
1096 struct ubuf_info
*uarg
= msg_control
;
1097 uarg
->callback(uarg
, false);
1102 tun
->dev
->stats
.rx_dropped
++;
1107 if (gso
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) {
1108 if (!skb_partial_csum_set(skb
, gso
.csum_start
,
1110 tun
->dev
->stats
.rx_frame_errors
++;
1116 switch (tun
->flags
& TUN_TYPE_MASK
) {
1118 if (tun
->flags
& TUN_NO_PI
) {
1119 switch (skb
->data
[0] & 0xf0) {
1121 pi
.proto
= htons(ETH_P_IP
);
1124 pi
.proto
= htons(ETH_P_IPV6
);
1127 tun
->dev
->stats
.rx_dropped
++;
1133 skb_reset_mac_header(skb
);
1134 skb
->protocol
= pi
.proto
;
1135 skb
->dev
= tun
->dev
;
1138 skb
->protocol
= eth_type_trans(skb
, tun
->dev
);
1142 if (gso
.gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
1144 switch (gso
.gso_type
& ~VIRTIO_NET_HDR_GSO_ECN
) {
1145 case VIRTIO_NET_HDR_GSO_TCPV4
:
1146 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV4
;
1148 case VIRTIO_NET_HDR_GSO_TCPV6
:
1149 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV6
;
1151 case VIRTIO_NET_HDR_GSO_UDP
:
1152 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
1155 tun
->dev
->stats
.rx_frame_errors
++;
1160 if (gso
.gso_type
& VIRTIO_NET_HDR_GSO_ECN
)
1161 skb_shinfo(skb
)->gso_type
|= SKB_GSO_TCP_ECN
;
1163 skb_shinfo(skb
)->gso_size
= gso
.gso_size
;
1164 if (skb_shinfo(skb
)->gso_size
== 0) {
1165 tun
->dev
->stats
.rx_frame_errors
++;
1170 /* Header must be checked, and gso_segs computed. */
1171 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
1172 skb_shinfo(skb
)->gso_segs
= 0;
1175 /* copy skb_ubuf_info for callback when skb has no error */
1177 skb_shinfo(skb
)->destructor_arg
= msg_control
;
1178 skb_shinfo(skb
)->tx_flags
|= SKBTX_DEV_ZEROCOPY
;
1179 skb_shinfo(skb
)->tx_flags
|= SKBTX_SHARED_FRAG
;
1182 skb_reset_network_header(skb
);
1183 skb_probe_transport_header(skb
, 0);
1185 rxhash
= skb_get_hash(skb
);
1188 tun
->dev
->stats
.rx_packets
++;
1189 tun
->dev
->stats
.rx_bytes
+= len
;
1191 tun_flow_update(tun
, rxhash
, tfile
);
1195 static ssize_t
tun_chr_aio_write(struct kiocb
*iocb
, const struct iovec
*iv
,
1196 unsigned long count
, loff_t pos
)
1198 struct file
*file
= iocb
->ki_filp
;
1199 struct tun_struct
*tun
= tun_get(file
);
1200 struct tun_file
*tfile
= file
->private_data
;
1206 tun_debug(KERN_INFO
, tun
, "tun_chr_write %ld\n", count
);
1208 result
= tun_get_user(tun
, tfile
, NULL
, iv
, iov_length(iv
, count
),
1209 count
, file
->f_flags
& O_NONBLOCK
);
1215 /* Put packet to the user space buffer */
1216 static ssize_t
tun_put_user(struct tun_struct
*tun
,
1217 struct tun_file
*tfile
,
1218 struct sk_buff
*skb
,
1219 const struct iovec
*iv
, int len
)
1221 struct tun_pi pi
= { 0, skb
->protocol
};
1223 int vlan_offset
= 0, copied
;
1225 if (!(tun
->flags
& TUN_NO_PI
)) {
1226 if ((len
-= sizeof(pi
)) < 0)
1229 if (len
< skb
->len
) {
1230 /* Packet will be striped */
1231 pi
.flags
|= TUN_PKT_STRIP
;
1234 if (memcpy_toiovecend(iv
, (void *) &pi
, 0, sizeof(pi
)))
1236 total
+= sizeof(pi
);
1239 if (tun
->flags
& TUN_VNET_HDR
) {
1240 struct virtio_net_hdr gso
= { 0 }; /* no info leak */
1241 if ((len
-= tun
->vnet_hdr_sz
) < 0)
1244 if (skb_is_gso(skb
)) {
1245 struct skb_shared_info
*sinfo
= skb_shinfo(skb
);
1247 /* This is a hint as to how much should be linear. */
1248 gso
.hdr_len
= skb_headlen(skb
);
1249 gso
.gso_size
= sinfo
->gso_size
;
1250 if (sinfo
->gso_type
& SKB_GSO_TCPV4
)
1251 gso
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV4
;
1252 else if (sinfo
->gso_type
& SKB_GSO_TCPV6
)
1253 gso
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV6
;
1254 else if (sinfo
->gso_type
& SKB_GSO_UDP
)
1255 gso
.gso_type
= VIRTIO_NET_HDR_GSO_UDP
;
1257 pr_err("unexpected GSO type: "
1258 "0x%x, gso_size %d, hdr_len %d\n",
1259 sinfo
->gso_type
, gso
.gso_size
,
1261 print_hex_dump(KERN_ERR
, "tun: ",
1264 min((int)gso
.hdr_len
, 64), true);
1268 if (sinfo
->gso_type
& SKB_GSO_TCP_ECN
)
1269 gso
.gso_type
|= VIRTIO_NET_HDR_GSO_ECN
;
1271 gso
.gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
1273 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1274 gso
.flags
= VIRTIO_NET_HDR_F_NEEDS_CSUM
;
1275 gso
.csum_start
= skb_checksum_start_offset(skb
);
1276 gso
.csum_offset
= skb
->csum_offset
;
1277 } else if (skb
->ip_summed
== CHECKSUM_UNNECESSARY
) {
1278 gso
.flags
= VIRTIO_NET_HDR_F_DATA_VALID
;
1279 } /* else everything is zero */
1281 if (unlikely(memcpy_toiovecend(iv
, (void *)&gso
, total
,
1284 total
+= tun
->vnet_hdr_sz
;
1289 if (!vlan_tx_tag_present(skb
)) {
1290 len
= min_t(int, skb
->len
, len
);
1294 __be16 h_vlan_proto
;
1298 veth
.h_vlan_proto
= skb
->vlan_proto
;
1299 veth
.h_vlan_TCI
= htons(vlan_tx_tag_get(skb
));
1301 vlan_offset
= offsetof(struct vlan_ethhdr
, h_vlan_proto
);
1302 len
= min_t(int, skb
->len
+ VLAN_HLEN
, len
);
1305 copy
= min_t(int, vlan_offset
, len
);
1306 ret
= skb_copy_datagram_const_iovec(skb
, 0, iv
, copied
, copy
);
1312 copy
= min_t(int, sizeof(veth
), len
);
1313 ret
= memcpy_toiovecend(iv
, (void *)&veth
, copied
, copy
);
1320 skb_copy_datagram_const_iovec(skb
, vlan_offset
, iv
, copied
, len
);
1323 tun
->dev
->stats
.tx_packets
++;
1324 tun
->dev
->stats
.tx_bytes
+= len
;
1329 static ssize_t
tun_do_read(struct tun_struct
*tun
, struct tun_file
*tfile
,
1330 const struct iovec
*iv
, ssize_t len
, int noblock
)
1332 DECLARE_WAITQUEUE(wait
, current
);
1333 struct sk_buff
*skb
;
1336 tun_debug(KERN_INFO
, tun
, "tun_do_read\n");
1338 if (unlikely(!noblock
))
1339 add_wait_queue(&tfile
->wq
.wait
, &wait
);
1341 if (unlikely(!noblock
))
1342 current
->state
= TASK_INTERRUPTIBLE
;
1344 /* Read frames from the queue */
1345 if (!(skb
= skb_dequeue(&tfile
->socket
.sk
->sk_receive_queue
))) {
1350 if (signal_pending(current
)) {
1354 if (tun
->dev
->reg_state
!= NETREG_REGISTERED
) {
1359 /* Nothing to read, let's sleep */
1364 ret
= tun_put_user(tun
, tfile
, skb
, iv
, len
);
1369 if (unlikely(!noblock
)) {
1370 current
->state
= TASK_RUNNING
;
1371 remove_wait_queue(&tfile
->wq
.wait
, &wait
);
1377 static ssize_t
tun_chr_aio_read(struct kiocb
*iocb
, const struct iovec
*iv
,
1378 unsigned long count
, loff_t pos
)
1380 struct file
*file
= iocb
->ki_filp
;
1381 struct tun_file
*tfile
= file
->private_data
;
1382 struct tun_struct
*tun
= __tun_get(tfile
);
1387 len
= iov_length(iv
, count
);
1393 ret
= tun_do_read(tun
, tfile
, iv
, len
,
1394 file
->f_flags
& O_NONBLOCK
);
1395 ret
= min_t(ssize_t
, ret
, len
);
1403 static void tun_free_netdev(struct net_device
*dev
)
1405 struct tun_struct
*tun
= netdev_priv(dev
);
1407 BUG_ON(!(list_empty(&tun
->disabled
)));
1408 tun_flow_uninit(tun
);
1409 security_tun_dev_free_security(tun
->security
);
1413 static void tun_setup(struct net_device
*dev
)
1415 struct tun_struct
*tun
= netdev_priv(dev
);
1417 tun
->owner
= INVALID_UID
;
1418 tun
->group
= INVALID_GID
;
1420 dev
->ethtool_ops
= &tun_ethtool_ops
;
1421 dev
->destructor
= tun_free_netdev
;
1424 /* Trivial set of netlink ops to allow deleting tun or tap
1425 * device with netlink.
1427 static int tun_validate(struct nlattr
*tb
[], struct nlattr
*data
[])
1432 static struct rtnl_link_ops tun_link_ops __read_mostly
= {
1434 .priv_size
= sizeof(struct tun_struct
),
1436 .validate
= tun_validate
,
1439 static void tun_sock_write_space(struct sock
*sk
)
1441 struct tun_file
*tfile
;
1442 wait_queue_head_t
*wqueue
;
1444 if (!sock_writeable(sk
))
1447 if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
))
1450 wqueue
= sk_sleep(sk
);
1451 if (wqueue
&& waitqueue_active(wqueue
))
1452 wake_up_interruptible_sync_poll(wqueue
, POLLOUT
|
1453 POLLWRNORM
| POLLWRBAND
);
1455 tfile
= container_of(sk
, struct tun_file
, sk
);
1456 kill_fasync(&tfile
->fasync
, SIGIO
, POLL_OUT
);
1459 static int tun_sendmsg(struct kiocb
*iocb
, struct socket
*sock
,
1460 struct msghdr
*m
, size_t total_len
)
1463 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
1464 struct tun_struct
*tun
= __tun_get(tfile
);
1468 ret
= tun_get_user(tun
, tfile
, m
->msg_control
, m
->msg_iov
, total_len
,
1469 m
->msg_iovlen
, m
->msg_flags
& MSG_DONTWAIT
);
1474 static int tun_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
1475 struct msghdr
*m
, size_t total_len
,
1478 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
1479 struct tun_struct
*tun
= __tun_get(tfile
);
1485 if (flags
& ~(MSG_DONTWAIT
|MSG_TRUNC
|MSG_ERRQUEUE
)) {
1489 if (flags
& MSG_ERRQUEUE
) {
1490 ret
= sock_recv_errqueue(sock
->sk
, m
, total_len
,
1491 SOL_PACKET
, TUN_TX_TIMESTAMP
);
1494 ret
= tun_do_read(tun
, tfile
, m
->msg_iov
, total_len
,
1495 flags
& MSG_DONTWAIT
);
1496 if (ret
> total_len
) {
1497 m
->msg_flags
|= MSG_TRUNC
;
1498 ret
= flags
& MSG_TRUNC
? ret
: total_len
;
1505 static int tun_release(struct socket
*sock
)
1512 /* Ops structure to mimic raw sockets with tun */
1513 static const struct proto_ops tun_socket_ops
= {
1514 .sendmsg
= tun_sendmsg
,
1515 .recvmsg
= tun_recvmsg
,
1516 .release
= tun_release
,
1519 static struct proto tun_proto
= {
1521 .owner
= THIS_MODULE
,
1522 .obj_size
= sizeof(struct tun_file
),
1525 static int tun_flags(struct tun_struct
*tun
)
1529 if (tun
->flags
& TUN_TUN_DEV
)
1534 if (tun
->flags
& TUN_NO_PI
)
1537 /* This flag has no real effect. We track the value for backwards
1540 if (tun
->flags
& TUN_ONE_QUEUE
)
1541 flags
|= IFF_ONE_QUEUE
;
1543 if (tun
->flags
& TUN_VNET_HDR
)
1544 flags
|= IFF_VNET_HDR
;
1546 if (tun
->flags
& TUN_TAP_MQ
)
1547 flags
|= IFF_MULTI_QUEUE
;
1549 if (tun
->flags
& TUN_PERSIST
)
1550 flags
|= IFF_PERSIST
;
1555 static ssize_t
tun_show_flags(struct device
*dev
, struct device_attribute
*attr
,
1558 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
1559 return sprintf(buf
, "0x%x\n", tun_flags(tun
));
1562 static ssize_t
tun_show_owner(struct device
*dev
, struct device_attribute
*attr
,
1565 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
1566 return uid_valid(tun
->owner
)?
1567 sprintf(buf
, "%u\n",
1568 from_kuid_munged(current_user_ns(), tun
->owner
)):
1569 sprintf(buf
, "-1\n");
1572 static ssize_t
tun_show_group(struct device
*dev
, struct device_attribute
*attr
,
1575 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
1576 return gid_valid(tun
->group
) ?
1577 sprintf(buf
, "%u\n",
1578 from_kgid_munged(current_user_ns(), tun
->group
)):
1579 sprintf(buf
, "-1\n");
1582 static DEVICE_ATTR(tun_flags
, 0444, tun_show_flags
, NULL
);
1583 static DEVICE_ATTR(owner
, 0444, tun_show_owner
, NULL
);
1584 static DEVICE_ATTR(group
, 0444, tun_show_group
, NULL
);
1586 static int tun_set_iff(struct net
*net
, struct file
*file
, struct ifreq
*ifr
)
1588 struct tun_struct
*tun
;
1589 struct tun_file
*tfile
= file
->private_data
;
1590 struct net_device
*dev
;
1593 if (tfile
->detached
)
1596 dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
1598 if (ifr
->ifr_flags
& IFF_TUN_EXCL
)
1600 if ((ifr
->ifr_flags
& IFF_TUN
) && dev
->netdev_ops
== &tun_netdev_ops
)
1601 tun
= netdev_priv(dev
);
1602 else if ((ifr
->ifr_flags
& IFF_TAP
) && dev
->netdev_ops
== &tap_netdev_ops
)
1603 tun
= netdev_priv(dev
);
1607 if (!!(ifr
->ifr_flags
& IFF_MULTI_QUEUE
) !=
1608 !!(tun
->flags
& TUN_TAP_MQ
))
1611 if (tun_not_capable(tun
))
1613 err
= security_tun_dev_open(tun
->security
);
1617 err
= tun_attach(tun
, file
, ifr
->ifr_flags
& IFF_NOFILTER
);
1621 if (tun
->flags
& TUN_TAP_MQ
&&
1622 (tun
->numqueues
+ tun
->numdisabled
> 1)) {
1623 /* One or more queue has already been attached, no need
1624 * to initialize the device again.
1631 unsigned long flags
= 0;
1632 int queues
= ifr
->ifr_flags
& IFF_MULTI_QUEUE
?
1635 if (!ns_capable(net
->user_ns
, CAP_NET_ADMIN
))
1637 err
= security_tun_dev_create();
1642 if (ifr
->ifr_flags
& IFF_TUN
) {
1644 flags
|= TUN_TUN_DEV
;
1646 } else if (ifr
->ifr_flags
& IFF_TAP
) {
1648 flags
|= TUN_TAP_DEV
;
1654 name
= ifr
->ifr_name
;
1656 dev
= alloc_netdev_mqs(sizeof(struct tun_struct
), name
,
1657 tun_setup
, queues
, queues
);
1662 dev_net_set(dev
, net
);
1663 dev
->rtnl_link_ops
= &tun_link_ops
;
1664 dev
->ifindex
= tfile
->ifindex
;
1666 tun
= netdev_priv(dev
);
1669 tun
->txflt
.count
= 0;
1670 tun
->vnet_hdr_sz
= sizeof(struct virtio_net_hdr
);
1672 tun
->filter_attached
= false;
1673 tun
->sndbuf
= tfile
->socket
.sk
->sk_sndbuf
;
1675 spin_lock_init(&tun
->lock
);
1677 err
= security_tun_dev_alloc_security(&tun
->security
);
1684 dev
->hw_features
= NETIF_F_SG
| NETIF_F_FRAGLIST
|
1685 TUN_USER_FEATURES
| NETIF_F_HW_VLAN_CTAG_TX
|
1686 NETIF_F_HW_VLAN_STAG_TX
;
1687 dev
->features
= dev
->hw_features
;
1688 dev
->vlan_features
= dev
->features
;
1690 INIT_LIST_HEAD(&tun
->disabled
);
1691 err
= tun_attach(tun
, file
, false);
1695 err
= register_netdevice(tun
->dev
);
1699 if (device_create_file(&tun
->dev
->dev
, &dev_attr_tun_flags
) ||
1700 device_create_file(&tun
->dev
->dev
, &dev_attr_owner
) ||
1701 device_create_file(&tun
->dev
->dev
, &dev_attr_group
))
1702 pr_err("Failed to create tun sysfs files\n");
1705 netif_carrier_on(tun
->dev
);
1707 tun_debug(KERN_INFO
, tun
, "tun_set_iff\n");
1709 if (ifr
->ifr_flags
& IFF_NO_PI
)
1710 tun
->flags
|= TUN_NO_PI
;
1712 tun
->flags
&= ~TUN_NO_PI
;
1714 /* This flag has no real effect. We track the value for backwards
1717 if (ifr
->ifr_flags
& IFF_ONE_QUEUE
)
1718 tun
->flags
|= TUN_ONE_QUEUE
;
1720 tun
->flags
&= ~TUN_ONE_QUEUE
;
1722 if (ifr
->ifr_flags
& IFF_VNET_HDR
)
1723 tun
->flags
|= TUN_VNET_HDR
;
1725 tun
->flags
&= ~TUN_VNET_HDR
;
1727 if (ifr
->ifr_flags
& IFF_MULTI_QUEUE
)
1728 tun
->flags
|= TUN_TAP_MQ
;
1730 tun
->flags
&= ~TUN_TAP_MQ
;
1732 /* Make sure persistent devices do not get stuck in
1735 if (netif_running(tun
->dev
))
1736 netif_tx_wake_all_queues(tun
->dev
);
1738 strcpy(ifr
->ifr_name
, tun
->dev
->name
);
1742 tun_detach_all(dev
);
1744 tun_flow_uninit(tun
);
1745 security_tun_dev_free_security(tun
->security
);
1751 static void tun_get_iff(struct net
*net
, struct tun_struct
*tun
,
1754 tun_debug(KERN_INFO
, tun
, "tun_get_iff\n");
1756 strcpy(ifr
->ifr_name
, tun
->dev
->name
);
1758 ifr
->ifr_flags
= tun_flags(tun
);
1762 /* This is like a cut-down ethtool ops, except done via tun fd so no
1763 * privs required. */
1764 static int set_offload(struct tun_struct
*tun
, unsigned long arg
)
1766 netdev_features_t features
= 0;
1768 if (arg
& TUN_F_CSUM
) {
1769 features
|= NETIF_F_HW_CSUM
;
1772 if (arg
& (TUN_F_TSO4
|TUN_F_TSO6
)) {
1773 if (arg
& TUN_F_TSO_ECN
) {
1774 features
|= NETIF_F_TSO_ECN
;
1775 arg
&= ~TUN_F_TSO_ECN
;
1777 if (arg
& TUN_F_TSO4
)
1778 features
|= NETIF_F_TSO
;
1779 if (arg
& TUN_F_TSO6
)
1780 features
|= NETIF_F_TSO6
;
1781 arg
&= ~(TUN_F_TSO4
|TUN_F_TSO6
);
1784 if (arg
& TUN_F_UFO
) {
1785 features
|= NETIF_F_UFO
;
1790 /* This gives the user a way to test for new features in future by
1791 * trying to set them. */
1795 tun
->set_features
= features
;
1796 netdev_update_features(tun
->dev
);
1801 static void tun_detach_filter(struct tun_struct
*tun
, int n
)
1804 struct tun_file
*tfile
;
1806 for (i
= 0; i
< n
; i
++) {
1807 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
1808 sk_detach_filter(tfile
->socket
.sk
);
1811 tun
->filter_attached
= false;
1814 static int tun_attach_filter(struct tun_struct
*tun
)
1817 struct tun_file
*tfile
;
1819 for (i
= 0; i
< tun
->numqueues
; i
++) {
1820 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
1821 ret
= sk_attach_filter(&tun
->fprog
, tfile
->socket
.sk
);
1823 tun_detach_filter(tun
, i
);
1828 tun
->filter_attached
= true;
1832 static void tun_set_sndbuf(struct tun_struct
*tun
)
1834 struct tun_file
*tfile
;
1837 for (i
= 0; i
< tun
->numqueues
; i
++) {
1838 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
1839 tfile
->socket
.sk
->sk_sndbuf
= tun
->sndbuf
;
1843 static int tun_set_queue(struct file
*file
, struct ifreq
*ifr
)
1845 struct tun_file
*tfile
= file
->private_data
;
1846 struct tun_struct
*tun
;
1851 if (ifr
->ifr_flags
& IFF_ATTACH_QUEUE
) {
1852 tun
= tfile
->detached
;
1857 ret
= security_tun_dev_attach_queue(tun
->security
);
1860 ret
= tun_attach(tun
, file
, false);
1861 } else if (ifr
->ifr_flags
& IFF_DETACH_QUEUE
) {
1862 tun
= rtnl_dereference(tfile
->tun
);
1863 if (!tun
|| !(tun
->flags
& TUN_TAP_MQ
) || tfile
->detached
)
1866 __tun_detach(tfile
, false);
1875 static long __tun_chr_ioctl(struct file
*file
, unsigned int cmd
,
1876 unsigned long arg
, int ifreq_len
)
1878 struct tun_file
*tfile
= file
->private_data
;
1879 struct tun_struct
*tun
;
1880 void __user
* argp
= (void __user
*)arg
;
1886 unsigned int ifindex
;
1889 if (cmd
== TUNSETIFF
|| cmd
== TUNSETQUEUE
|| _IOC_TYPE(cmd
) == 0x89) {
1890 if (copy_from_user(&ifr
, argp
, ifreq_len
))
1893 memset(&ifr
, 0, sizeof(ifr
));
1895 if (cmd
== TUNGETFEATURES
) {
1896 /* Currently this just means: "what IFF flags are valid?".
1897 * This is needed because we never checked for invalid flags on
1899 return put_user(IFF_TUN
| IFF_TAP
| IFF_NO_PI
| IFF_ONE_QUEUE
|
1900 IFF_VNET_HDR
| IFF_MULTI_QUEUE
,
1901 (unsigned int __user
*)argp
);
1902 } else if (cmd
== TUNSETQUEUE
)
1903 return tun_set_queue(file
, &ifr
);
1908 tun
= __tun_get(tfile
);
1909 if (cmd
== TUNSETIFF
&& !tun
) {
1910 ifr
.ifr_name
[IFNAMSIZ
-1] = '\0';
1912 ret
= tun_set_iff(tfile
->net
, file
, &ifr
);
1917 if (copy_to_user(argp
, &ifr
, ifreq_len
))
1921 if (cmd
== TUNSETIFINDEX
) {
1927 if (copy_from_user(&ifindex
, argp
, sizeof(ifindex
)))
1931 tfile
->ifindex
= ifindex
;
1939 tun_debug(KERN_INFO
, tun
, "tun_chr_ioctl cmd %u\n", cmd
);
1944 tun_get_iff(current
->nsproxy
->net_ns
, tun
, &ifr
);
1946 if (tfile
->detached
)
1947 ifr
.ifr_flags
|= IFF_DETACH_QUEUE
;
1948 if (!tfile
->socket
.sk
->sk_filter
)
1949 ifr
.ifr_flags
|= IFF_NOFILTER
;
1951 if (copy_to_user(argp
, &ifr
, ifreq_len
))
1956 /* Disable/Enable checksum */
1958 /* [unimplemented] */
1959 tun_debug(KERN_INFO
, tun
, "ignored: set checksum %s\n",
1960 arg
? "disabled" : "enabled");
1964 /* Disable/Enable persist mode. Keep an extra reference to the
1965 * module to prevent the module being unprobed.
1967 if (arg
&& !(tun
->flags
& TUN_PERSIST
)) {
1968 tun
->flags
|= TUN_PERSIST
;
1969 __module_get(THIS_MODULE
);
1971 if (!arg
&& (tun
->flags
& TUN_PERSIST
)) {
1972 tun
->flags
&= ~TUN_PERSIST
;
1973 module_put(THIS_MODULE
);
1976 tun_debug(KERN_INFO
, tun
, "persist %s\n",
1977 arg
? "enabled" : "disabled");
1981 /* Set owner of the device */
1982 owner
= make_kuid(current_user_ns(), arg
);
1983 if (!uid_valid(owner
)) {
1988 tun_debug(KERN_INFO
, tun
, "owner set to %u\n",
1989 from_kuid(&init_user_ns
, tun
->owner
));
1993 /* Set group of the device */
1994 group
= make_kgid(current_user_ns(), arg
);
1995 if (!gid_valid(group
)) {
2000 tun_debug(KERN_INFO
, tun
, "group set to %u\n",
2001 from_kgid(&init_user_ns
, tun
->group
));
2005 /* Only allow setting the type when the interface is down */
2006 if (tun
->dev
->flags
& IFF_UP
) {
2007 tun_debug(KERN_INFO
, tun
,
2008 "Linktype set failed because interface is up\n");
2011 tun
->dev
->type
= (int) arg
;
2012 tun_debug(KERN_INFO
, tun
, "linktype set to %d\n",
2024 ret
= set_offload(tun
, arg
);
2027 case TUNSETTXFILTER
:
2028 /* Can be set only for TAPs */
2030 if ((tun
->flags
& TUN_TYPE_MASK
) != TUN_TAP_DEV
)
2032 ret
= update_filter(&tun
->txflt
, (void __user
*)arg
);
2036 /* Get hw address */
2037 memcpy(ifr
.ifr_hwaddr
.sa_data
, tun
->dev
->dev_addr
, ETH_ALEN
);
2038 ifr
.ifr_hwaddr
.sa_family
= tun
->dev
->type
;
2039 if (copy_to_user(argp
, &ifr
, ifreq_len
))
2044 /* Set hw address */
2045 tun_debug(KERN_DEBUG
, tun
, "set hw address: %pM\n",
2046 ifr
.ifr_hwaddr
.sa_data
);
2048 ret
= dev_set_mac_address(tun
->dev
, &ifr
.ifr_hwaddr
);
2052 sndbuf
= tfile
->socket
.sk
->sk_sndbuf
;
2053 if (copy_to_user(argp
, &sndbuf
, sizeof(sndbuf
)))
2058 if (copy_from_user(&sndbuf
, argp
, sizeof(sndbuf
))) {
2063 tun
->sndbuf
= sndbuf
;
2064 tun_set_sndbuf(tun
);
2067 case TUNGETVNETHDRSZ
:
2068 vnet_hdr_sz
= tun
->vnet_hdr_sz
;
2069 if (copy_to_user(argp
, &vnet_hdr_sz
, sizeof(vnet_hdr_sz
)))
2073 case TUNSETVNETHDRSZ
:
2074 if (copy_from_user(&vnet_hdr_sz
, argp
, sizeof(vnet_hdr_sz
))) {
2078 if (vnet_hdr_sz
< (int)sizeof(struct virtio_net_hdr
)) {
2083 tun
->vnet_hdr_sz
= vnet_hdr_sz
;
2086 case TUNATTACHFILTER
:
2087 /* Can be set only for TAPs */
2089 if ((tun
->flags
& TUN_TYPE_MASK
) != TUN_TAP_DEV
)
2092 if (copy_from_user(&tun
->fprog
, argp
, sizeof(tun
->fprog
)))
2095 ret
= tun_attach_filter(tun
);
2098 case TUNDETACHFILTER
:
2099 /* Can be set only for TAPs */
2101 if ((tun
->flags
& TUN_TYPE_MASK
) != TUN_TAP_DEV
)
2104 tun_detach_filter(tun
, tun
->numqueues
);
2109 if ((tun
->flags
& TUN_TYPE_MASK
) != TUN_TAP_DEV
)
2112 if (copy_to_user(argp
, &tun
->fprog
, sizeof(tun
->fprog
)))
2129 static long tun_chr_ioctl(struct file
*file
,
2130 unsigned int cmd
, unsigned long arg
)
2132 return __tun_chr_ioctl(file
, cmd
, arg
, sizeof (struct ifreq
));
2135 #ifdef CONFIG_COMPAT
2136 static long tun_chr_compat_ioctl(struct file
*file
,
2137 unsigned int cmd
, unsigned long arg
)
2142 case TUNSETTXFILTER
:
2147 arg
= (unsigned long)compat_ptr(arg
);
2150 arg
= (compat_ulong_t
)arg
;
2155 * compat_ifreq is shorter than ifreq, so we must not access beyond
2156 * the end of that structure. All fields that are used in this
2157 * driver are compatible though, we don't need to convert the
2160 return __tun_chr_ioctl(file
, cmd
, arg
, sizeof(struct compat_ifreq
));
2162 #endif /* CONFIG_COMPAT */
2164 static int tun_chr_fasync(int fd
, struct file
*file
, int on
)
2166 struct tun_file
*tfile
= file
->private_data
;
2169 if ((ret
= fasync_helper(fd
, file
, on
, &tfile
->fasync
)) < 0)
2173 ret
= __f_setown(file
, task_pid(current
), PIDTYPE_PID
, 0);
2176 tfile
->flags
|= TUN_FASYNC
;
2178 tfile
->flags
&= ~TUN_FASYNC
;
2184 static int tun_chr_open(struct inode
*inode
, struct file
* file
)
2186 struct tun_file
*tfile
;
2188 DBG1(KERN_INFO
, "tunX: tun_chr_open\n");
2190 tfile
= (struct tun_file
*)sk_alloc(&init_net
, AF_UNSPEC
, GFP_KERNEL
,
2194 rcu_assign_pointer(tfile
->tun
, NULL
);
2195 tfile
->net
= get_net(current
->nsproxy
->net_ns
);
2199 rcu_assign_pointer(tfile
->socket
.wq
, &tfile
->wq
);
2200 init_waitqueue_head(&tfile
->wq
.wait
);
2202 tfile
->socket
.file
= file
;
2203 tfile
->socket
.ops
= &tun_socket_ops
;
2205 sock_init_data(&tfile
->socket
, &tfile
->sk
);
2206 sk_change_net(&tfile
->sk
, tfile
->net
);
2208 tfile
->sk
.sk_write_space
= tun_sock_write_space
;
2209 tfile
->sk
.sk_sndbuf
= INT_MAX
;
2211 file
->private_data
= tfile
;
2212 set_bit(SOCK_EXTERNALLY_ALLOCATED
, &tfile
->socket
.flags
);
2213 INIT_LIST_HEAD(&tfile
->next
);
2215 sock_set_flag(&tfile
->sk
, SOCK_ZEROCOPY
);
2220 static int tun_chr_close(struct inode
*inode
, struct file
*file
)
2222 struct tun_file
*tfile
= file
->private_data
;
2223 struct net
*net
= tfile
->net
;
2225 tun_detach(tfile
, true);
2231 static const struct file_operations tun_fops
= {
2232 .owner
= THIS_MODULE
,
2233 .llseek
= no_llseek
,
2234 .read
= do_sync_read
,
2235 .aio_read
= tun_chr_aio_read
,
2236 .write
= do_sync_write
,
2237 .aio_write
= tun_chr_aio_write
,
2238 .poll
= tun_chr_poll
,
2239 .unlocked_ioctl
= tun_chr_ioctl
,
2240 #ifdef CONFIG_COMPAT
2241 .compat_ioctl
= tun_chr_compat_ioctl
,
2243 .open
= tun_chr_open
,
2244 .release
= tun_chr_close
,
2245 .fasync
= tun_chr_fasync
2248 static struct miscdevice tun_miscdev
= {
2251 .nodename
= "net/tun",
2255 /* ethtool interface */
2257 static int tun_get_settings(struct net_device
*dev
, struct ethtool_cmd
*cmd
)
2260 cmd
->advertising
= 0;
2261 ethtool_cmd_speed_set(cmd
, SPEED_10
);
2262 cmd
->duplex
= DUPLEX_FULL
;
2263 cmd
->port
= PORT_TP
;
2264 cmd
->phy_address
= 0;
2265 cmd
->transceiver
= XCVR_INTERNAL
;
2266 cmd
->autoneg
= AUTONEG_DISABLE
;
2272 static void tun_get_drvinfo(struct net_device
*dev
, struct ethtool_drvinfo
*info
)
2274 struct tun_struct
*tun
= netdev_priv(dev
);
2276 strlcpy(info
->driver
, DRV_NAME
, sizeof(info
->driver
));
2277 strlcpy(info
->version
, DRV_VERSION
, sizeof(info
->version
));
2279 switch (tun
->flags
& TUN_TYPE_MASK
) {
2281 strlcpy(info
->bus_info
, "tun", sizeof(info
->bus_info
));
2284 strlcpy(info
->bus_info
, "tap", sizeof(info
->bus_info
));
2289 static u32
tun_get_msglevel(struct net_device
*dev
)
2292 struct tun_struct
*tun
= netdev_priv(dev
);
2299 static void tun_set_msglevel(struct net_device
*dev
, u32 value
)
2302 struct tun_struct
*tun
= netdev_priv(dev
);
2307 static const struct ethtool_ops tun_ethtool_ops
= {
2308 .get_settings
= tun_get_settings
,
2309 .get_drvinfo
= tun_get_drvinfo
,
2310 .get_msglevel
= tun_get_msglevel
,
2311 .set_msglevel
= tun_set_msglevel
,
2312 .get_link
= ethtool_op_get_link
,
2313 .get_ts_info
= ethtool_op_get_ts_info
,
2317 static int __init
tun_init(void)
2321 pr_info("%s, %s\n", DRV_DESCRIPTION
, DRV_VERSION
);
2322 pr_info("%s\n", DRV_COPYRIGHT
);
2324 ret
= rtnl_link_register(&tun_link_ops
);
2326 pr_err("Can't register link_ops\n");
2330 ret
= misc_register(&tun_miscdev
);
2332 pr_err("Can't register misc device %d\n", TUN_MINOR
);
2337 rtnl_link_unregister(&tun_link_ops
);
2342 static void tun_cleanup(void)
2344 misc_deregister(&tun_miscdev
);
2345 rtnl_link_unregister(&tun_link_ops
);
2348 /* Get an underlying socket object from tun file. Returns error unless file is
2349 * attached to a device. The returned object works like a packet socket, it
2350 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
2351 * holding a reference to the file for as long as the socket is in use. */
2352 struct socket
*tun_get_socket(struct file
*file
)
2354 struct tun_file
*tfile
;
2355 if (file
->f_op
!= &tun_fops
)
2356 return ERR_PTR(-EINVAL
);
2357 tfile
= file
->private_data
;
2359 return ERR_PTR(-EBADFD
);
2360 return &tfile
->socket
;
2362 EXPORT_SYMBOL_GPL(tun_get_socket
);
2364 module_init(tun_init
);
2365 module_exit(tun_cleanup
);
2366 MODULE_DESCRIPTION(DRV_DESCRIPTION
);
2367 MODULE_AUTHOR(DRV_COPYRIGHT
);
2368 MODULE_LICENSE("GPL");
2369 MODULE_ALIAS_MISCDEV(TUN_MINOR
);
2370 MODULE_ALIAS("devname:net/tun");