2 * TUN - Universal TUN/TAP device driver.
3 * Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
21 * Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
22 * Add TUNSETLINK ioctl to set the link encapsulation
24 * Mark Smith <markzzzsmith@yahoo.com.au>
25 * Use eth_random_addr() for tap MAC address.
27 * Harald Roelle <harald.roelle@ifi.lmu.de> 2004/04/20
28 * Fixes in packet dropping, queue length setting and queue wakeup.
29 * Increased default tx queue length.
33 * Daniel Podlejski <underley@underley.eu.org>
34 * Modifications for 2.3.99-pre5 kernel.
37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
39 #define DRV_NAME "tun"
40 #define DRV_VERSION "1.6"
41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
42 #define DRV_COPYRIGHT "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
44 #include <linux/module.h>
45 #include <linux/errno.h>
46 #include <linux/kernel.h>
47 #include <linux/major.h>
48 #include <linux/slab.h>
49 #include <linux/poll.h>
50 #include <linux/fcntl.h>
51 #include <linux/init.h>
52 #include <linux/skbuff.h>
53 #include <linux/netdevice.h>
54 #include <linux/etherdevice.h>
55 #include <linux/miscdevice.h>
56 #include <linux/ethtool.h>
57 #include <linux/rtnetlink.h>
58 #include <linux/compat.h>
60 #include <linux/if_arp.h>
61 #include <linux/if_ether.h>
62 #include <linux/if_tun.h>
63 #include <linux/crc32.h>
64 #include <linux/nsproxy.h>
65 #include <linux/virtio_net.h>
66 #include <linux/rcupdate.h>
67 #include <net/net_namespace.h>
68 #include <net/netns/generic.h>
69 #include <net/rtnetlink.h>
72 #include <asm/uaccess.h>
74 /* Uncomment to enable debugging */
75 /* #define TUN_DEBUG 1 */
80 #define tun_debug(level, tun, fmt, args...) \
83 netdev_printk(level, tun->dev, fmt, ##args); \
85 #define DBG1(level, fmt, args...) \
88 printk(level fmt, ##args); \
91 #define tun_debug(level, tun, fmt, args...) \
94 netdev_printk(level, tun->dev, fmt, ##args); \
96 #define DBG1(level, fmt, args...) \
99 printk(level fmt, ##args); \
103 #define GOODCOPY_LEN 128
105 #define FLT_EXACT_COUNT 8
107 unsigned int count
; /* Number of addrs. Zero means disabled */
108 u32 mask
[2]; /* Mask of the hashed addrs */
109 unsigned char addr
[FLT_EXACT_COUNT
][ETH_ALEN
];
112 /* 1024 is probably a high enough limit: modern hypervisors seem to support on
113 * the order of 100-200 CPUs so this leaves us some breathing space if we want
114 * to match a queue per guest CPU.
116 #define MAX_TAP_QUEUES 1024
118 #define TUN_FLOW_EXPIRE (3 * HZ)
120 /* A tun_file connects an open character device to a tuntap netdevice. It
121 * also contains all socket related strctures (except sock_fprog and tap_filter)
122 * to serve as one transmit queue for tuntap device. The sock_fprog and
123 * tap_filter were kept in tun_struct since they were used for filtering for the
124 * netdevice not for a specific queue (at least I didn't see the requirement for
128 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
129 * other can only be read while rcu_read_lock or rtnl_lock is held.
133 struct socket socket
;
135 struct tun_struct __rcu
*tun
;
137 struct fasync_struct
*fasync
;
138 /* only used for fasnyc */
143 struct tun_flow_entry
{
144 struct hlist_node hash_link
;
146 struct tun_struct
*tun
;
150 unsigned long updated
;
153 #define TUN_NUM_FLOW_ENTRIES 1024
155 /* Since the socket were moved to tun_file, to preserve the behavior of persist
156 * device, socket filter, sndbuf and vnet header size were restore when the
157 * file were attached to a persist device.
160 struct tun_file __rcu
*tfiles
[MAX_TAP_QUEUES
];
161 unsigned int numqueues
;
166 struct net_device
*dev
;
167 netdev_features_t set_features
;
168 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
169 NETIF_F_TSO6|NETIF_F_UFO)
173 struct tap_filter txflt
;
174 struct sock_fprog fprog
;
175 /* protected by rtnl lock */
176 bool filter_attached
;
181 struct kmem_cache
*flow_cache
;
182 struct hlist_head flows
[TUN_NUM_FLOW_ENTRIES
];
183 struct timer_list flow_gc_timer
;
184 unsigned long ageing_time
;
187 static inline u32
tun_hashfn(u32 rxhash
)
189 return rxhash
& 0x3ff;
192 static struct tun_flow_entry
*tun_flow_find(struct hlist_head
*head
, u32 rxhash
)
194 struct tun_flow_entry
*e
;
195 struct hlist_node
*n
;
197 hlist_for_each_entry_rcu(e
, n
, head
, hash_link
) {
198 if (e
->rxhash
== rxhash
)
204 static struct tun_flow_entry
*tun_flow_create(struct tun_struct
*tun
,
205 struct hlist_head
*head
,
206 u32 rxhash
, u16 queue_index
)
208 struct tun_flow_entry
*e
= kmem_cache_alloc(tun
->flow_cache
,
211 tun_debug(KERN_INFO
, tun
, "create flow: hash %u index %u\n",
212 rxhash
, queue_index
);
213 e
->updated
= jiffies
;
215 e
->queue_index
= queue_index
;
217 hlist_add_head_rcu(&e
->hash_link
, head
);
222 static void tun_flow_free(struct rcu_head
*head
)
224 struct tun_flow_entry
*e
225 = container_of(head
, struct tun_flow_entry
, rcu
);
226 kmem_cache_free(e
->tun
->flow_cache
, e
);
229 static void tun_flow_delete(struct tun_struct
*tun
, struct tun_flow_entry
*e
)
231 tun_debug(KERN_INFO
, tun
, "delete flow: hash %u index %u\n",
232 e
->rxhash
, e
->queue_index
);
233 hlist_del_rcu(&e
->hash_link
);
234 call_rcu(&e
->rcu
, tun_flow_free
);
237 static void tun_flow_flush(struct tun_struct
*tun
)
241 spin_lock_bh(&tun
->lock
);
242 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
243 struct tun_flow_entry
*e
;
244 struct hlist_node
*h
, *n
;
246 hlist_for_each_entry_safe(e
, h
, n
, &tun
->flows
[i
], hash_link
)
247 tun_flow_delete(tun
, e
);
249 spin_unlock_bh(&tun
->lock
);
252 static void tun_flow_delete_by_queue(struct tun_struct
*tun
, u16 queue_index
)
256 spin_lock_bh(&tun
->lock
);
257 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
258 struct tun_flow_entry
*e
;
259 struct hlist_node
*h
, *n
;
261 hlist_for_each_entry_safe(e
, h
, n
, &tun
->flows
[i
], hash_link
) {
262 if (e
->queue_index
== queue_index
)
263 tun_flow_delete(tun
, e
);
266 spin_unlock_bh(&tun
->lock
);
269 static void tun_flow_cleanup(unsigned long data
)
271 struct tun_struct
*tun
= (struct tun_struct
*)data
;
272 unsigned long delay
= tun
->ageing_time
;
273 unsigned long next_timer
= jiffies
+ delay
;
274 unsigned long count
= 0;
277 tun_debug(KERN_INFO
, tun
, "tun_flow_cleanup\n");
279 spin_lock_bh(&tun
->lock
);
280 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
281 struct tun_flow_entry
*e
;
282 struct hlist_node
*h
, *n
;
284 hlist_for_each_entry_safe(e
, h
, n
, &tun
->flows
[i
], hash_link
) {
285 unsigned long this_timer
;
287 this_timer
= e
->updated
+ delay
;
288 if (time_before_eq(this_timer
, jiffies
))
289 tun_flow_delete(tun
, e
);
290 else if (time_before(this_timer
, next_timer
))
291 next_timer
= this_timer
;
296 mod_timer(&tun
->flow_gc_timer
, round_jiffies_up(next_timer
));
297 spin_unlock_bh(&tun
->lock
);
300 static void tun_flow_update(struct tun_struct
*tun
, struct sk_buff
*skb
,
303 struct hlist_head
*head
;
304 struct tun_flow_entry
*e
;
305 unsigned long delay
= tun
->ageing_time
;
306 u32 rxhash
= skb_get_rxhash(skb
);
311 head
= &tun
->flows
[tun_hashfn(rxhash
)];
315 if (tun
->numqueues
== 1)
318 e
= tun_flow_find(head
, rxhash
);
320 /* TODO: keep queueing to old queue until it's empty? */
321 e
->queue_index
= queue_index
;
322 e
->updated
= jiffies
;
324 spin_lock_bh(&tun
->lock
);
325 if (!tun_flow_find(head
, rxhash
))
326 tun_flow_create(tun
, head
, rxhash
, queue_index
);
328 if (!timer_pending(&tun
->flow_gc_timer
))
329 mod_timer(&tun
->flow_gc_timer
,
330 round_jiffies_up(jiffies
+ delay
));
331 spin_unlock_bh(&tun
->lock
);
338 /* We try to identify a flow through its rxhash first. The reason that
339 * we do not check rxq no. is becuase some cards(e.g 82599), chooses
340 * the rxq based on the txq where the last packet of the flow comes. As
341 * the userspace application move between processors, we may get a
342 * different rxq no. here. If we could not get rxhash, then we would
343 * hope the rxq no. may help here.
345 static u16
tun_select_queue(struct net_device
*dev
, struct sk_buff
*skb
)
347 struct tun_struct
*tun
= netdev_priv(dev
);
348 struct tun_flow_entry
*e
;
353 numqueues
= tun
->numqueues
;
355 txq
= skb_get_rxhash(skb
);
357 e
= tun_flow_find(&tun
->flows
[tun_hashfn(txq
)], txq
);
359 txq
= e
->queue_index
;
361 /* use multiply and shift instead of expensive divide */
362 txq
= ((u64
)txq
* numqueues
) >> 32;
363 } else if (likely(skb_rx_queue_recorded(skb
))) {
364 txq
= skb_get_rx_queue(skb
);
365 while (unlikely(txq
>= numqueues
))
373 static inline bool tun_not_capable(struct tun_struct
*tun
)
375 const struct cred
*cred
= current_cred();
376 struct net
*net
= dev_net(tun
->dev
);
378 return ((uid_valid(tun
->owner
) && !uid_eq(cred
->euid
, tun
->owner
)) ||
379 (gid_valid(tun
->group
) && !in_egroup_p(tun
->group
))) &&
380 !ns_capable(net
->user_ns
, CAP_NET_ADMIN
);
383 static void tun_set_real_num_queues(struct tun_struct
*tun
)
385 netif_set_real_num_tx_queues(tun
->dev
, tun
->numqueues
);
386 netif_set_real_num_rx_queues(tun
->dev
, tun
->numqueues
);
389 static void __tun_detach(struct tun_file
*tfile
, bool clean
)
391 struct tun_file
*ntfile
;
392 struct tun_struct
*tun
;
393 struct net_device
*dev
;
395 tun
= rcu_dereference_protected(tfile
->tun
,
396 lockdep_rtnl_is_held());
398 u16 index
= tfile
->queue_index
;
399 BUG_ON(index
>= tun
->numqueues
);
402 rcu_assign_pointer(tun
->tfiles
[index
],
403 tun
->tfiles
[tun
->numqueues
- 1]);
404 rcu_assign_pointer(tfile
->tun
, NULL
);
405 ntfile
= rcu_dereference_protected(tun
->tfiles
[index
],
406 lockdep_rtnl_is_held());
407 ntfile
->queue_index
= index
;
410 sock_put(&tfile
->sk
);
413 tun_flow_delete_by_queue(tun
, tun
->numqueues
+ 1);
414 /* Drop read queue */
415 skb_queue_purge(&tfile
->sk
.sk_receive_queue
);
416 tun_set_real_num_queues(tun
);
418 if (tun
->numqueues
== 0 && !(tun
->flags
& TUN_PERSIST
))
419 if (dev
->reg_state
== NETREG_REGISTERED
)
420 unregister_netdevice(dev
);
424 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED
,
425 &tfile
->socket
.flags
));
426 sk_release_kernel(&tfile
->sk
);
430 static void tun_detach(struct tun_file
*tfile
, bool clean
)
433 __tun_detach(tfile
, clean
);
437 static void tun_detach_all(struct net_device
*dev
)
439 struct tun_struct
*tun
= netdev_priv(dev
);
440 struct tun_file
*tfile
;
441 int i
, n
= tun
->numqueues
;
443 for (i
= 0; i
< n
; i
++) {
444 tfile
= rcu_dereference_protected(tun
->tfiles
[i
],
445 lockdep_rtnl_is_held());
447 wake_up_all(&tfile
->wq
.wait
);
448 rcu_assign_pointer(tfile
->tun
, NULL
);
451 BUG_ON(tun
->numqueues
!= 0);
454 for (i
= 0; i
< n
; i
++) {
455 tfile
= rcu_dereference_protected(tun
->tfiles
[i
],
456 lockdep_rtnl_is_held());
457 /* Drop read queue */
458 skb_queue_purge(&tfile
->sk
.sk_receive_queue
);
459 sock_put(&tfile
->sk
);
463 static int tun_attach(struct tun_struct
*tun
, struct file
*file
)
465 struct tun_file
*tfile
= file
->private_data
;
469 if (rcu_dereference_protected(tfile
->tun
, lockdep_rtnl_is_held()))
473 if (!(tun
->flags
& TUN_TAP_MQ
) && tun
->numqueues
== 1)
477 if (tun
->numqueues
== MAX_TAP_QUEUES
)
482 /* Re-attach the filter to presist device */
483 if (tun
->filter_attached
== true) {
484 err
= sk_attach_filter(&tun
->fprog
, tfile
->socket
.sk
);
488 tfile
->queue_index
= tun
->numqueues
;
489 rcu_assign_pointer(tfile
->tun
, tun
);
490 rcu_assign_pointer(tun
->tfiles
[tun
->numqueues
], tfile
);
491 sock_hold(&tfile
->sk
);
494 tun_set_real_num_queues(tun
);
496 /* device is allowed to go away first, so no need to hold extra
504 static struct tun_struct
*__tun_get(struct tun_file
*tfile
)
506 struct tun_struct
*tun
;
509 tun
= rcu_dereference(tfile
->tun
);
517 static struct tun_struct
*tun_get(struct file
*file
)
519 return __tun_get(file
->private_data
);
522 static void tun_put(struct tun_struct
*tun
)
528 static void addr_hash_set(u32
*mask
, const u8
*addr
)
530 int n
= ether_crc(ETH_ALEN
, addr
) >> 26;
531 mask
[n
>> 5] |= (1 << (n
& 31));
534 static unsigned int addr_hash_test(const u32
*mask
, const u8
*addr
)
536 int n
= ether_crc(ETH_ALEN
, addr
) >> 26;
537 return mask
[n
>> 5] & (1 << (n
& 31));
540 static int update_filter(struct tap_filter
*filter
, void __user
*arg
)
542 struct { u8 u
[ETH_ALEN
]; } *addr
;
543 struct tun_filter uf
;
544 int err
, alen
, n
, nexact
;
546 if (copy_from_user(&uf
, arg
, sizeof(uf
)))
555 alen
= ETH_ALEN
* uf
.count
;
556 addr
= kmalloc(alen
, GFP_KERNEL
);
560 if (copy_from_user(addr
, arg
+ sizeof(uf
), alen
)) {
565 /* The filter is updated without holding any locks. Which is
566 * perfectly safe. We disable it first and in the worst
567 * case we'll accept a few undesired packets. */
571 /* Use first set of addresses as an exact filter */
572 for (n
= 0; n
< uf
.count
&& n
< FLT_EXACT_COUNT
; n
++)
573 memcpy(filter
->addr
[n
], addr
[n
].u
, ETH_ALEN
);
577 /* Remaining multicast addresses are hashed,
578 * unicast will leave the filter disabled. */
579 memset(filter
->mask
, 0, sizeof(filter
->mask
));
580 for (; n
< uf
.count
; n
++) {
581 if (!is_multicast_ether_addr(addr
[n
].u
)) {
582 err
= 0; /* no filter */
585 addr_hash_set(filter
->mask
, addr
[n
].u
);
588 /* For ALLMULTI just set the mask to all ones.
589 * This overrides the mask populated above. */
590 if ((uf
.flags
& TUN_FLT_ALLMULTI
))
591 memset(filter
->mask
, ~0, sizeof(filter
->mask
));
593 /* Now enable the filter */
595 filter
->count
= nexact
;
597 /* Return the number of exact filters */
605 /* Returns: 0 - drop, !=0 - accept */
606 static int run_filter(struct tap_filter
*filter
, const struct sk_buff
*skb
)
608 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
610 struct ethhdr
*eh
= (struct ethhdr
*) skb
->data
;
614 for (i
= 0; i
< filter
->count
; i
++)
615 if (ether_addr_equal(eh
->h_dest
, filter
->addr
[i
]))
618 /* Inexact match (multicast only) */
619 if (is_multicast_ether_addr(eh
->h_dest
))
620 return addr_hash_test(filter
->mask
, eh
->h_dest
);
626 * Checks whether the packet is accepted or not.
627 * Returns: 0 - drop, !=0 - accept
629 static int check_filter(struct tap_filter
*filter
, const struct sk_buff
*skb
)
634 return run_filter(filter
, skb
);
637 /* Network device part of the driver */
639 static const struct ethtool_ops tun_ethtool_ops
;
641 /* Net device detach from fd. */
642 static void tun_net_uninit(struct net_device
*dev
)
647 /* Net device open. */
648 static int tun_net_open(struct net_device
*dev
)
650 netif_tx_start_all_queues(dev
);
654 /* Net device close. */
655 static int tun_net_close(struct net_device
*dev
)
657 netif_tx_stop_all_queues(dev
);
661 /* Net device start xmit */
662 static netdev_tx_t
tun_net_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
664 struct tun_struct
*tun
= netdev_priv(dev
);
665 int txq
= skb
->queue_mapping
;
666 struct tun_file
*tfile
;
669 tfile
= rcu_dereference(tun
->tfiles
[txq
]);
671 /* Drop packet if interface is not attached */
672 if (txq
>= tun
->numqueues
)
675 tun_debug(KERN_INFO
, tun
, "tun_net_xmit %d\n", skb
->len
);
679 /* Drop if the filter does not like it.
680 * This is a noop if the filter is disabled.
681 * Filter can be enabled only for the TAP devices. */
682 if (!check_filter(&tun
->txflt
, skb
))
685 if (tfile
->socket
.sk
->sk_filter
&&
686 sk_filter(tfile
->socket
.sk
, skb
))
689 /* Limit the number of packets queued by dividing txq length with the
692 if (skb_queue_len(&tfile
->socket
.sk
->sk_receive_queue
)
693 >= dev
->tx_queue_len
/ tun
->numqueues
)
696 /* Orphan the skb - required as we might hang on to it
697 * for indefinite time. */
698 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
703 skb_queue_tail(&tfile
->socket
.sk
->sk_receive_queue
, skb
);
705 /* Notify and wake up reader process */
706 if (tfile
->flags
& TUN_FASYNC
)
707 kill_fasync(&tfile
->fasync
, SIGIO
, POLL_IN
);
708 wake_up_interruptible_poll(&tfile
->wq
.wait
, POLLIN
|
709 POLLRDNORM
| POLLRDBAND
);
715 dev
->stats
.tx_dropped
++;
722 static void tun_net_mclist(struct net_device
*dev
)
725 * This callback is supposed to deal with mc filter in
726 * _rx_ path and has nothing to do with the _tx_ path.
727 * In rx path we always accept everything userspace gives us.
732 #define MAX_MTU 65535
735 tun_net_change_mtu(struct net_device
*dev
, int new_mtu
)
737 if (new_mtu
< MIN_MTU
|| new_mtu
+ dev
->hard_header_len
> MAX_MTU
)
743 static netdev_features_t
tun_net_fix_features(struct net_device
*dev
,
744 netdev_features_t features
)
746 struct tun_struct
*tun
= netdev_priv(dev
);
748 return (features
& tun
->set_features
) | (features
& ~TUN_USER_FEATURES
);
750 #ifdef CONFIG_NET_POLL_CONTROLLER
751 static void tun_poll_controller(struct net_device
*dev
)
754 * Tun only receives frames when:
755 * 1) the char device endpoint gets data from user space
756 * 2) the tun socket gets a sendmsg call from user space
757 * Since both of those are syncronous operations, we are guaranteed
758 * never to have pending data when we poll for it
759 * so theres nothing to do here but return.
760 * We need this though so netpoll recognizes us as an interface that
761 * supports polling, which enables bridge devices in virt setups to
762 * still use netconsole
767 static const struct net_device_ops tun_netdev_ops
= {
768 .ndo_uninit
= tun_net_uninit
,
769 .ndo_open
= tun_net_open
,
770 .ndo_stop
= tun_net_close
,
771 .ndo_start_xmit
= tun_net_xmit
,
772 .ndo_change_mtu
= tun_net_change_mtu
,
773 .ndo_fix_features
= tun_net_fix_features
,
774 .ndo_select_queue
= tun_select_queue
,
775 #ifdef CONFIG_NET_POLL_CONTROLLER
776 .ndo_poll_controller
= tun_poll_controller
,
780 static const struct net_device_ops tap_netdev_ops
= {
781 .ndo_uninit
= tun_net_uninit
,
782 .ndo_open
= tun_net_open
,
783 .ndo_stop
= tun_net_close
,
784 .ndo_start_xmit
= tun_net_xmit
,
785 .ndo_change_mtu
= tun_net_change_mtu
,
786 .ndo_fix_features
= tun_net_fix_features
,
787 .ndo_set_rx_mode
= tun_net_mclist
,
788 .ndo_set_mac_address
= eth_mac_addr
,
789 .ndo_validate_addr
= eth_validate_addr
,
790 .ndo_select_queue
= tun_select_queue
,
791 #ifdef CONFIG_NET_POLL_CONTROLLER
792 .ndo_poll_controller
= tun_poll_controller
,
796 static int tun_flow_init(struct tun_struct
*tun
)
800 tun
->flow_cache
= kmem_cache_create("tun_flow_cache",
801 sizeof(struct tun_flow_entry
), 0, 0,
803 if (!tun
->flow_cache
)
806 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++)
807 INIT_HLIST_HEAD(&tun
->flows
[i
]);
809 tun
->ageing_time
= TUN_FLOW_EXPIRE
;
810 setup_timer(&tun
->flow_gc_timer
, tun_flow_cleanup
, (unsigned long)tun
);
811 mod_timer(&tun
->flow_gc_timer
,
812 round_jiffies_up(jiffies
+ tun
->ageing_time
));
817 static void tun_flow_uninit(struct tun_struct
*tun
)
819 del_timer_sync(&tun
->flow_gc_timer
);
822 /* Wait for completion of call_rcu()'s */
824 kmem_cache_destroy(tun
->flow_cache
);
827 /* Initialize net device. */
828 static void tun_net_init(struct net_device
*dev
)
830 struct tun_struct
*tun
= netdev_priv(dev
);
832 switch (tun
->flags
& TUN_TYPE_MASK
) {
834 dev
->netdev_ops
= &tun_netdev_ops
;
836 /* Point-to-Point TUN Device */
837 dev
->hard_header_len
= 0;
841 /* Zero header length */
842 dev
->type
= ARPHRD_NONE
;
843 dev
->flags
= IFF_POINTOPOINT
| IFF_NOARP
| IFF_MULTICAST
;
844 dev
->tx_queue_len
= TUN_READQ_SIZE
; /* We prefer our own queue length */
848 dev
->netdev_ops
= &tap_netdev_ops
;
849 /* Ethernet TAP Device */
851 dev
->priv_flags
&= ~IFF_TX_SKB_SHARING
;
853 eth_hw_addr_random(dev
);
855 dev
->tx_queue_len
= TUN_READQ_SIZE
; /* We prefer our own queue length */
860 /* Character device part */
863 static unsigned int tun_chr_poll(struct file
*file
, poll_table
*wait
)
865 struct tun_file
*tfile
= file
->private_data
;
866 struct tun_struct
*tun
= __tun_get(tfile
);
868 unsigned int mask
= 0;
873 sk
= tfile
->socket
.sk
;
875 tun_debug(KERN_INFO
, tun
, "tun_chr_poll\n");
877 poll_wait(file
, &tfile
->wq
.wait
, wait
);
879 if (!skb_queue_empty(&sk
->sk_receive_queue
))
880 mask
|= POLLIN
| POLLRDNORM
;
882 if (sock_writeable(sk
) ||
883 (!test_and_set_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
) &&
885 mask
|= POLLOUT
| POLLWRNORM
;
887 if (tun
->dev
->reg_state
!= NETREG_REGISTERED
)
894 /* prepad is the amount to reserve at front. len is length after that.
895 * linear is a hint as to how much to copy (usually headers). */
896 static struct sk_buff
*tun_alloc_skb(struct tun_file
*tfile
,
897 size_t prepad
, size_t len
,
898 size_t linear
, int noblock
)
900 struct sock
*sk
= tfile
->socket
.sk
;
904 /* Under a page? Don't bother with paged skb. */
905 if (prepad
+ len
< PAGE_SIZE
|| !linear
)
908 skb
= sock_alloc_send_pskb(sk
, prepad
+ linear
, len
- linear
, noblock
,
913 skb_reserve(skb
, prepad
);
914 skb_put(skb
, linear
);
915 skb
->data_len
= len
- linear
;
916 skb
->len
+= len
- linear
;
921 /* set skb frags from iovec, this can move to core network code for reuse */
922 static int zerocopy_sg_from_iovec(struct sk_buff
*skb
, const struct iovec
*from
,
923 int offset
, size_t count
)
925 int len
= iov_length(from
, count
) - offset
;
926 int copy
= skb_headlen(skb
);
927 int size
, offset1
= 0;
930 /* Skip over from offset */
931 while (count
&& (offset
>= from
->iov_len
)) {
932 offset
-= from
->iov_len
;
937 /* copy up to skb headlen */
938 while (count
&& (copy
> 0)) {
939 size
= min_t(unsigned int, copy
, from
->iov_len
- offset
);
940 if (copy_from_user(skb
->data
+ offset1
, from
->iov_base
+ offset
,
957 struct page
*page
[MAX_SKB_FRAGS
];
960 unsigned long truesize
;
962 len
= from
->iov_len
- offset
;
968 base
= (unsigned long)from
->iov_base
+ offset
;
969 size
= ((base
& ~PAGE_MASK
) + len
+ ~PAGE_MASK
) >> PAGE_SHIFT
;
970 if (i
+ size
> MAX_SKB_FRAGS
)
972 num_pages
= get_user_pages_fast(base
, size
, 0, &page
[i
]);
973 if (num_pages
!= size
) {
974 for (i
= 0; i
< num_pages
; i
++)
978 truesize
= size
* PAGE_SIZE
;
979 skb
->data_len
+= len
;
981 skb
->truesize
+= truesize
;
982 atomic_add(truesize
, &skb
->sk
->sk_wmem_alloc
);
984 int off
= base
& ~PAGE_MASK
;
985 int size
= min_t(int, len
, PAGE_SIZE
- off
);
986 __skb_fill_page_desc(skb
, i
, page
[i
], off
, size
);
987 skb_shinfo(skb
)->nr_frags
++;
988 /* increase sk_wmem_alloc */
999 /* Get packet from user space buffer */
1000 static ssize_t
tun_get_user(struct tun_struct
*tun
, struct tun_file
*tfile
,
1001 void *msg_control
, const struct iovec
*iv
,
1002 size_t total_len
, size_t count
, int noblock
)
1004 struct tun_pi pi
= { 0, cpu_to_be16(ETH_P_IP
) };
1005 struct sk_buff
*skb
;
1006 size_t len
= total_len
, align
= NET_SKB_PAD
;
1007 struct virtio_net_hdr gso
= { 0 };
1010 bool zerocopy
= false;
1013 if (!(tun
->flags
& TUN_NO_PI
)) {
1014 if ((len
-= sizeof(pi
)) > total_len
)
1017 if (memcpy_fromiovecend((void *)&pi
, iv
, 0, sizeof(pi
)))
1019 offset
+= sizeof(pi
);
1022 if (tun
->flags
& TUN_VNET_HDR
) {
1023 if ((len
-= tun
->vnet_hdr_sz
) > total_len
)
1026 if (memcpy_fromiovecend((void *)&gso
, iv
, offset
, sizeof(gso
)))
1029 if ((gso
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) &&
1030 gso
.csum_start
+ gso
.csum_offset
+ 2 > gso
.hdr_len
)
1031 gso
.hdr_len
= gso
.csum_start
+ gso
.csum_offset
+ 2;
1033 if (gso
.hdr_len
> len
)
1035 offset
+= tun
->vnet_hdr_sz
;
1038 if ((tun
->flags
& TUN_TYPE_MASK
) == TUN_TAP_DEV
) {
1039 align
+= NET_IP_ALIGN
;
1040 if (unlikely(len
< ETH_HLEN
||
1041 (gso
.hdr_len
&& gso
.hdr_len
< ETH_HLEN
)))
1049 /* Userspace may produce vectors with count greater than
1050 * MAX_SKB_FRAGS, so we need to linearize parts of the skb
1051 * to let the rest of data to be fit in the frags.
1053 if (count
> MAX_SKB_FRAGS
) {
1054 copylen
= iov_length(iv
, count
- MAX_SKB_FRAGS
);
1055 if (copylen
< offset
)
1061 /* There are 256 bytes to be copied in skb, so there is enough
1062 * room for skb expand head in case it is used.
1063 * The rest of the buffer is mapped from userspace.
1065 if (copylen
< gso
.hdr_len
)
1066 copylen
= gso
.hdr_len
;
1068 copylen
= GOODCOPY_LEN
;
1072 skb
= tun_alloc_skb(tfile
, align
, copylen
, gso
.hdr_len
, noblock
);
1074 if (PTR_ERR(skb
) != -EAGAIN
)
1075 tun
->dev
->stats
.rx_dropped
++;
1076 return PTR_ERR(skb
);
1080 err
= zerocopy_sg_from_iovec(skb
, iv
, offset
, count
);
1082 err
= skb_copy_datagram_from_iovec(skb
, 0, iv
, offset
, len
);
1085 tun
->dev
->stats
.rx_dropped
++;
1090 if (gso
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) {
1091 if (!skb_partial_csum_set(skb
, gso
.csum_start
,
1093 tun
->dev
->stats
.rx_frame_errors
++;
1099 switch (tun
->flags
& TUN_TYPE_MASK
) {
1101 if (tun
->flags
& TUN_NO_PI
) {
1102 switch (skb
->data
[0] & 0xf0) {
1104 pi
.proto
= htons(ETH_P_IP
);
1107 pi
.proto
= htons(ETH_P_IPV6
);
1110 tun
->dev
->stats
.rx_dropped
++;
1116 skb_reset_mac_header(skb
);
1117 skb
->protocol
= pi
.proto
;
1118 skb
->dev
= tun
->dev
;
1121 skb
->protocol
= eth_type_trans(skb
, tun
->dev
);
1125 if (gso
.gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
1127 switch (gso
.gso_type
& ~VIRTIO_NET_HDR_GSO_ECN
) {
1128 case VIRTIO_NET_HDR_GSO_TCPV4
:
1129 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV4
;
1131 case VIRTIO_NET_HDR_GSO_TCPV6
:
1132 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV6
;
1134 case VIRTIO_NET_HDR_GSO_UDP
:
1135 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
1138 tun
->dev
->stats
.rx_frame_errors
++;
1143 if (gso
.gso_type
& VIRTIO_NET_HDR_GSO_ECN
)
1144 skb_shinfo(skb
)->gso_type
|= SKB_GSO_TCP_ECN
;
1146 skb_shinfo(skb
)->gso_size
= gso
.gso_size
;
1147 if (skb_shinfo(skb
)->gso_size
== 0) {
1148 tun
->dev
->stats
.rx_frame_errors
++;
1153 /* Header must be checked, and gso_segs computed. */
1154 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
1155 skb_shinfo(skb
)->gso_segs
= 0;
1158 /* copy skb_ubuf_info for callback when skb has no error */
1160 skb_shinfo(skb
)->destructor_arg
= msg_control
;
1161 skb_shinfo(skb
)->tx_flags
|= SKBTX_DEV_ZEROCOPY
;
1166 tun
->dev
->stats
.rx_packets
++;
1167 tun
->dev
->stats
.rx_bytes
+= len
;
1169 tun_flow_update(tun
, skb
, tfile
->queue_index
);
1173 static ssize_t
tun_chr_aio_write(struct kiocb
*iocb
, const struct iovec
*iv
,
1174 unsigned long count
, loff_t pos
)
1176 struct file
*file
= iocb
->ki_filp
;
1177 struct tun_struct
*tun
= tun_get(file
);
1178 struct tun_file
*tfile
= file
->private_data
;
1184 tun_debug(KERN_INFO
, tun
, "tun_chr_write %ld\n", count
);
1186 result
= tun_get_user(tun
, tfile
, NULL
, iv
, iov_length(iv
, count
),
1187 count
, file
->f_flags
& O_NONBLOCK
);
1193 /* Put packet to the user space buffer */
1194 static ssize_t
tun_put_user(struct tun_struct
*tun
,
1195 struct tun_file
*tfile
,
1196 struct sk_buff
*skb
,
1197 const struct iovec
*iv
, int len
)
1199 struct tun_pi pi
= { 0, skb
->protocol
};
1202 if (!(tun
->flags
& TUN_NO_PI
)) {
1203 if ((len
-= sizeof(pi
)) < 0)
1206 if (len
< skb
->len
) {
1207 /* Packet will be striped */
1208 pi
.flags
|= TUN_PKT_STRIP
;
1211 if (memcpy_toiovecend(iv
, (void *) &pi
, 0, sizeof(pi
)))
1213 total
+= sizeof(pi
);
1216 if (tun
->flags
& TUN_VNET_HDR
) {
1217 struct virtio_net_hdr gso
= { 0 }; /* no info leak */
1218 if ((len
-= tun
->vnet_hdr_sz
) < 0)
1221 if (skb_is_gso(skb
)) {
1222 struct skb_shared_info
*sinfo
= skb_shinfo(skb
);
1224 /* This is a hint as to how much should be linear. */
1225 gso
.hdr_len
= skb_headlen(skb
);
1226 gso
.gso_size
= sinfo
->gso_size
;
1227 if (sinfo
->gso_type
& SKB_GSO_TCPV4
)
1228 gso
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV4
;
1229 else if (sinfo
->gso_type
& SKB_GSO_TCPV6
)
1230 gso
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV6
;
1231 else if (sinfo
->gso_type
& SKB_GSO_UDP
)
1232 gso
.gso_type
= VIRTIO_NET_HDR_GSO_UDP
;
1234 pr_err("unexpected GSO type: "
1235 "0x%x, gso_size %d, hdr_len %d\n",
1236 sinfo
->gso_type
, gso
.gso_size
,
1238 print_hex_dump(KERN_ERR
, "tun: ",
1241 min((int)gso
.hdr_len
, 64), true);
1245 if (sinfo
->gso_type
& SKB_GSO_TCP_ECN
)
1246 gso
.gso_type
|= VIRTIO_NET_HDR_GSO_ECN
;
1248 gso
.gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
1250 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1251 gso
.flags
= VIRTIO_NET_HDR_F_NEEDS_CSUM
;
1252 gso
.csum_start
= skb_checksum_start_offset(skb
);
1253 gso
.csum_offset
= skb
->csum_offset
;
1254 } else if (skb
->ip_summed
== CHECKSUM_UNNECESSARY
) {
1255 gso
.flags
= VIRTIO_NET_HDR_F_DATA_VALID
;
1256 } /* else everything is zero */
1258 if (unlikely(memcpy_toiovecend(iv
, (void *)&gso
, total
,
1261 total
+= tun
->vnet_hdr_sz
;
1264 len
= min_t(int, skb
->len
, len
);
1266 skb_copy_datagram_const_iovec(skb
, 0, iv
, total
, len
);
1269 tun
->dev
->stats
.tx_packets
++;
1270 tun
->dev
->stats
.tx_bytes
+= len
;
1275 static ssize_t
tun_do_read(struct tun_struct
*tun
, struct tun_file
*tfile
,
1276 struct kiocb
*iocb
, const struct iovec
*iv
,
1277 ssize_t len
, int noblock
)
1279 DECLARE_WAITQUEUE(wait
, current
);
1280 struct sk_buff
*skb
;
1283 tun_debug(KERN_INFO
, tun
, "tun_do_read\n");
1285 if (unlikely(!noblock
))
1286 add_wait_queue(&tfile
->wq
.wait
, &wait
);
1288 current
->state
= TASK_INTERRUPTIBLE
;
1290 /* Read frames from the queue */
1291 if (!(skb
= skb_dequeue(&tfile
->socket
.sk
->sk_receive_queue
))) {
1296 if (signal_pending(current
)) {
1300 if (tun
->dev
->reg_state
!= NETREG_REGISTERED
) {
1305 /* Nothing to read, let's sleep */
1310 ret
= tun_put_user(tun
, tfile
, skb
, iv
, len
);
1315 current
->state
= TASK_RUNNING
;
1316 if (unlikely(!noblock
))
1317 remove_wait_queue(&tfile
->wq
.wait
, &wait
);
1322 static ssize_t
tun_chr_aio_read(struct kiocb
*iocb
, const struct iovec
*iv
,
1323 unsigned long count
, loff_t pos
)
1325 struct file
*file
= iocb
->ki_filp
;
1326 struct tun_file
*tfile
= file
->private_data
;
1327 struct tun_struct
*tun
= __tun_get(tfile
);
1332 len
= iov_length(iv
, count
);
1338 ret
= tun_do_read(tun
, tfile
, iocb
, iv
, len
,
1339 file
->f_flags
& O_NONBLOCK
);
1340 ret
= min_t(ssize_t
, ret
, len
);
1346 static void tun_free_netdev(struct net_device
*dev
)
1348 struct tun_struct
*tun
= netdev_priv(dev
);
1350 tun_flow_uninit(tun
);
1354 static void tun_setup(struct net_device
*dev
)
1356 struct tun_struct
*tun
= netdev_priv(dev
);
1358 tun
->owner
= INVALID_UID
;
1359 tun
->group
= INVALID_GID
;
1361 dev
->ethtool_ops
= &tun_ethtool_ops
;
1362 dev
->destructor
= tun_free_netdev
;
1365 /* Trivial set of netlink ops to allow deleting tun or tap
1366 * device with netlink.
1368 static int tun_validate(struct nlattr
*tb
[], struct nlattr
*data
[])
1373 static struct rtnl_link_ops tun_link_ops __read_mostly
= {
1375 .priv_size
= sizeof(struct tun_struct
),
1377 .validate
= tun_validate
,
1380 static void tun_sock_write_space(struct sock
*sk
)
1382 struct tun_file
*tfile
;
1383 wait_queue_head_t
*wqueue
;
1385 if (!sock_writeable(sk
))
1388 if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
))
1391 wqueue
= sk_sleep(sk
);
1392 if (wqueue
&& waitqueue_active(wqueue
))
1393 wake_up_interruptible_sync_poll(wqueue
, POLLOUT
|
1394 POLLWRNORM
| POLLWRBAND
);
1396 tfile
= container_of(sk
, struct tun_file
, sk
);
1397 kill_fasync(&tfile
->fasync
, SIGIO
, POLL_OUT
);
1400 static int tun_sendmsg(struct kiocb
*iocb
, struct socket
*sock
,
1401 struct msghdr
*m
, size_t total_len
)
1404 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
1405 struct tun_struct
*tun
= __tun_get(tfile
);
1409 ret
= tun_get_user(tun
, tfile
, m
->msg_control
, m
->msg_iov
, total_len
,
1410 m
->msg_iovlen
, m
->msg_flags
& MSG_DONTWAIT
);
1416 static int tun_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
1417 struct msghdr
*m
, size_t total_len
,
1420 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
1421 struct tun_struct
*tun
= __tun_get(tfile
);
1427 if (flags
& ~(MSG_DONTWAIT
|MSG_TRUNC
))
1429 ret
= tun_do_read(tun
, tfile
, iocb
, m
->msg_iov
, total_len
,
1430 flags
& MSG_DONTWAIT
);
1431 if (ret
> total_len
) {
1432 m
->msg_flags
|= MSG_TRUNC
;
1433 ret
= flags
& MSG_TRUNC
? ret
: total_len
;
1439 static int tun_release(struct socket
*sock
)
1446 /* Ops structure to mimic raw sockets with tun */
1447 static const struct proto_ops tun_socket_ops
= {
1448 .sendmsg
= tun_sendmsg
,
1449 .recvmsg
= tun_recvmsg
,
1450 .release
= tun_release
,
1453 static struct proto tun_proto
= {
1455 .owner
= THIS_MODULE
,
1456 .obj_size
= sizeof(struct tun_file
),
1459 static int tun_flags(struct tun_struct
*tun
)
1463 if (tun
->flags
& TUN_TUN_DEV
)
1468 if (tun
->flags
& TUN_NO_PI
)
1471 /* This flag has no real effect. We track the value for backwards
1474 if (tun
->flags
& TUN_ONE_QUEUE
)
1475 flags
|= IFF_ONE_QUEUE
;
1477 if (tun
->flags
& TUN_VNET_HDR
)
1478 flags
|= IFF_VNET_HDR
;
1480 if (tun
->flags
& TUN_TAP_MQ
)
1481 flags
|= IFF_MULTI_QUEUE
;
1486 static ssize_t
tun_show_flags(struct device
*dev
, struct device_attribute
*attr
,
1489 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
1490 return sprintf(buf
, "0x%x\n", tun_flags(tun
));
1493 static ssize_t
tun_show_owner(struct device
*dev
, struct device_attribute
*attr
,
1496 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
1497 return uid_valid(tun
->owner
)?
1498 sprintf(buf
, "%u\n",
1499 from_kuid_munged(current_user_ns(), tun
->owner
)):
1500 sprintf(buf
, "-1\n");
1503 static ssize_t
tun_show_group(struct device
*dev
, struct device_attribute
*attr
,
1506 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
1507 return gid_valid(tun
->group
) ?
1508 sprintf(buf
, "%u\n",
1509 from_kgid_munged(current_user_ns(), tun
->group
)):
1510 sprintf(buf
, "-1\n");
1513 static DEVICE_ATTR(tun_flags
, 0444, tun_show_flags
, NULL
);
1514 static DEVICE_ATTR(owner
, 0444, tun_show_owner
, NULL
);
1515 static DEVICE_ATTR(group
, 0444, tun_show_group
, NULL
);
1517 static int tun_set_iff(struct net
*net
, struct file
*file
, struct ifreq
*ifr
)
1519 struct tun_struct
*tun
;
1520 struct tun_file
*tfile
= file
->private_data
;
1521 struct net_device
*dev
;
1524 dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
1526 if (ifr
->ifr_flags
& IFF_TUN_EXCL
)
1528 if ((ifr
->ifr_flags
& IFF_TUN
) && dev
->netdev_ops
== &tun_netdev_ops
)
1529 tun
= netdev_priv(dev
);
1530 else if ((ifr
->ifr_flags
& IFF_TAP
) && dev
->netdev_ops
== &tap_netdev_ops
)
1531 tun
= netdev_priv(dev
);
1535 if (tun_not_capable(tun
))
1537 err
= security_tun_dev_attach(tfile
->socket
.sk
);
1541 err
= tun_attach(tun
, file
);
1547 unsigned long flags
= 0;
1549 if (!ns_capable(net
->user_ns
, CAP_NET_ADMIN
))
1551 err
= security_tun_dev_create();
1556 if (ifr
->ifr_flags
& IFF_TUN
) {
1558 flags
|= TUN_TUN_DEV
;
1560 } else if (ifr
->ifr_flags
& IFF_TAP
) {
1562 flags
|= TUN_TAP_DEV
;
1568 name
= ifr
->ifr_name
;
1570 dev
= alloc_netdev_mqs(sizeof(struct tun_struct
), name
,
1572 MAX_TAP_QUEUES
, MAX_TAP_QUEUES
);
1576 dev_net_set(dev
, net
);
1577 dev
->rtnl_link_ops
= &tun_link_ops
;
1579 tun
= netdev_priv(dev
);
1582 tun
->txflt
.count
= 0;
1583 tun
->vnet_hdr_sz
= sizeof(struct virtio_net_hdr
);
1585 tun
->filter_attached
= false;
1586 tun
->sndbuf
= tfile
->socket
.sk
->sk_sndbuf
;
1588 spin_lock_init(&tun
->lock
);
1590 security_tun_dev_post_create(&tfile
->sk
);
1594 if (tun_flow_init(tun
))
1597 dev
->hw_features
= NETIF_F_SG
| NETIF_F_FRAGLIST
|
1599 dev
->features
= dev
->hw_features
;
1601 err
= tun_attach(tun
, file
);
1605 err
= register_netdevice(tun
->dev
);
1609 if (device_create_file(&tun
->dev
->dev
, &dev_attr_tun_flags
) ||
1610 device_create_file(&tun
->dev
->dev
, &dev_attr_owner
) ||
1611 device_create_file(&tun
->dev
->dev
, &dev_attr_group
))
1612 pr_err("Failed to create tun sysfs files\n");
1614 netif_carrier_on(tun
->dev
);
1617 tun_debug(KERN_INFO
, tun
, "tun_set_iff\n");
1619 if (ifr
->ifr_flags
& IFF_NO_PI
)
1620 tun
->flags
|= TUN_NO_PI
;
1622 tun
->flags
&= ~TUN_NO_PI
;
1624 /* This flag has no real effect. We track the value for backwards
1627 if (ifr
->ifr_flags
& IFF_ONE_QUEUE
)
1628 tun
->flags
|= TUN_ONE_QUEUE
;
1630 tun
->flags
&= ~TUN_ONE_QUEUE
;
1632 if (ifr
->ifr_flags
& IFF_VNET_HDR
)
1633 tun
->flags
|= TUN_VNET_HDR
;
1635 tun
->flags
&= ~TUN_VNET_HDR
;
1637 if (ifr
->ifr_flags
& IFF_MULTI_QUEUE
)
1638 tun
->flags
|= TUN_TAP_MQ
;
1640 tun
->flags
&= ~TUN_TAP_MQ
;
1642 /* Make sure persistent devices do not get stuck in
1645 if (netif_running(tun
->dev
))
1646 netif_tx_wake_all_queues(tun
->dev
);
1648 strcpy(ifr
->ifr_name
, tun
->dev
->name
);
1656 static void tun_get_iff(struct net
*net
, struct tun_struct
*tun
,
1659 tun_debug(KERN_INFO
, tun
, "tun_get_iff\n");
1661 strcpy(ifr
->ifr_name
, tun
->dev
->name
);
1663 ifr
->ifr_flags
= tun_flags(tun
);
1667 /* This is like a cut-down ethtool ops, except done via tun fd so no
1668 * privs required. */
1669 static int set_offload(struct tun_struct
*tun
, unsigned long arg
)
1671 netdev_features_t features
= 0;
1673 if (arg
& TUN_F_CSUM
) {
1674 features
|= NETIF_F_HW_CSUM
;
1677 if (arg
& (TUN_F_TSO4
|TUN_F_TSO6
)) {
1678 if (arg
& TUN_F_TSO_ECN
) {
1679 features
|= NETIF_F_TSO_ECN
;
1680 arg
&= ~TUN_F_TSO_ECN
;
1682 if (arg
& TUN_F_TSO4
)
1683 features
|= NETIF_F_TSO
;
1684 if (arg
& TUN_F_TSO6
)
1685 features
|= NETIF_F_TSO6
;
1686 arg
&= ~(TUN_F_TSO4
|TUN_F_TSO6
);
1689 if (arg
& TUN_F_UFO
) {
1690 features
|= NETIF_F_UFO
;
1695 /* This gives the user a way to test for new features in future by
1696 * trying to set them. */
1700 tun
->set_features
= features
;
1701 netdev_update_features(tun
->dev
);
1706 static void tun_detach_filter(struct tun_struct
*tun
, int n
)
1709 struct tun_file
*tfile
;
1711 for (i
= 0; i
< n
; i
++) {
1712 tfile
= rcu_dereference_protected(tun
->tfiles
[i
],
1713 lockdep_rtnl_is_held());
1714 sk_detach_filter(tfile
->socket
.sk
);
1717 tun
->filter_attached
= false;
1720 static int tun_attach_filter(struct tun_struct
*tun
)
1723 struct tun_file
*tfile
;
1725 for (i
= 0; i
< tun
->numqueues
; i
++) {
1726 tfile
= rcu_dereference_protected(tun
->tfiles
[i
],
1727 lockdep_rtnl_is_held());
1728 ret
= sk_attach_filter(&tun
->fprog
, tfile
->socket
.sk
);
1730 tun_detach_filter(tun
, i
);
1735 tun
->filter_attached
= true;
1739 static void tun_set_sndbuf(struct tun_struct
*tun
)
1741 struct tun_file
*tfile
;
1744 for (i
= 0; i
< tun
->numqueues
; i
++) {
1745 tfile
= rcu_dereference_protected(tun
->tfiles
[i
],
1746 lockdep_rtnl_is_held());
1747 tfile
->socket
.sk
->sk_sndbuf
= tun
->sndbuf
;
1751 static int tun_set_queue(struct file
*file
, struct ifreq
*ifr
)
1753 struct tun_file
*tfile
= file
->private_data
;
1754 struct tun_struct
*tun
;
1755 struct net_device
*dev
;
1760 if (ifr
->ifr_flags
& IFF_ATTACH_QUEUE
) {
1761 dev
= __dev_get_by_name(tfile
->net
, ifr
->ifr_name
);
1767 tun
= netdev_priv(dev
);
1768 if (dev
->netdev_ops
!= &tap_netdev_ops
&&
1769 dev
->netdev_ops
!= &tun_netdev_ops
)
1771 else if (tun_not_capable(tun
))
1774 ret
= tun_attach(tun
, file
);
1775 } else if (ifr
->ifr_flags
& IFF_DETACH_QUEUE
)
1776 __tun_detach(tfile
, false);
1785 static long __tun_chr_ioctl(struct file
*file
, unsigned int cmd
,
1786 unsigned long arg
, int ifreq_len
)
1788 struct tun_file
*tfile
= file
->private_data
;
1789 struct tun_struct
*tun
;
1790 void __user
* argp
= (void __user
*)arg
;
1798 if (cmd
== TUNSETIFF
|| cmd
== TUNSETQUEUE
|| _IOC_TYPE(cmd
) == 0x89) {
1799 if (copy_from_user(&ifr
, argp
, ifreq_len
))
1802 memset(&ifr
, 0, sizeof(ifr
));
1804 if (cmd
== TUNGETFEATURES
) {
1805 /* Currently this just means: "what IFF flags are valid?".
1806 * This is needed because we never checked for invalid flags on
1808 return put_user(IFF_TUN
| IFF_TAP
| IFF_NO_PI
| IFF_ONE_QUEUE
|
1809 IFF_VNET_HDR
| IFF_MULTI_QUEUE
,
1810 (unsigned int __user
*)argp
);
1811 } else if (cmd
== TUNSETQUEUE
)
1812 return tun_set_queue(file
, &ifr
);
1817 tun
= __tun_get(tfile
);
1818 if (cmd
== TUNSETIFF
&& !tun
) {
1819 ifr
.ifr_name
[IFNAMSIZ
-1] = '\0';
1821 ret
= tun_set_iff(tfile
->net
, file
, &ifr
);
1826 if (copy_to_user(argp
, &ifr
, ifreq_len
))
1835 tun_debug(KERN_INFO
, tun
, "tun_chr_ioctl cmd %u\n", cmd
);
1840 tun_get_iff(current
->nsproxy
->net_ns
, tun
, &ifr
);
1842 if (copy_to_user(argp
, &ifr
, ifreq_len
))
1847 /* Disable/Enable checksum */
1849 /* [unimplemented] */
1850 tun_debug(KERN_INFO
, tun
, "ignored: set checksum %s\n",
1851 arg
? "disabled" : "enabled");
1855 /* Disable/Enable persist mode. Keep an extra reference to the
1856 * module to prevent the module being unprobed.
1859 tun
->flags
|= TUN_PERSIST
;
1860 __module_get(THIS_MODULE
);
1862 tun
->flags
&= ~TUN_PERSIST
;
1863 module_put(THIS_MODULE
);
1866 tun_debug(KERN_INFO
, tun
, "persist %s\n",
1867 arg
? "enabled" : "disabled");
1871 /* Set owner of the device */
1872 owner
= make_kuid(current_user_ns(), arg
);
1873 if (!uid_valid(owner
)) {
1878 tun_debug(KERN_INFO
, tun
, "owner set to %u\n",
1879 from_kuid(&init_user_ns
, tun
->owner
));
1883 /* Set group of the device */
1884 group
= make_kgid(current_user_ns(), arg
);
1885 if (!gid_valid(group
)) {
1890 tun_debug(KERN_INFO
, tun
, "group set to %u\n",
1891 from_kgid(&init_user_ns
, tun
->group
));
1895 /* Only allow setting the type when the interface is down */
1896 if (tun
->dev
->flags
& IFF_UP
) {
1897 tun_debug(KERN_INFO
, tun
,
1898 "Linktype set failed because interface is up\n");
1901 tun
->dev
->type
= (int) arg
;
1902 tun_debug(KERN_INFO
, tun
, "linktype set to %d\n",
1914 ret
= set_offload(tun
, arg
);
1917 case TUNSETTXFILTER
:
1918 /* Can be set only for TAPs */
1920 if ((tun
->flags
& TUN_TYPE_MASK
) != TUN_TAP_DEV
)
1922 ret
= update_filter(&tun
->txflt
, (void __user
*)arg
);
1926 /* Get hw address */
1927 memcpy(ifr
.ifr_hwaddr
.sa_data
, tun
->dev
->dev_addr
, ETH_ALEN
);
1928 ifr
.ifr_hwaddr
.sa_family
= tun
->dev
->type
;
1929 if (copy_to_user(argp
, &ifr
, ifreq_len
))
1934 /* Set hw address */
1935 tun_debug(KERN_DEBUG
, tun
, "set hw address: %pM\n",
1936 ifr
.ifr_hwaddr
.sa_data
);
1938 ret
= dev_set_mac_address(tun
->dev
, &ifr
.ifr_hwaddr
);
1942 sndbuf
= tfile
->socket
.sk
->sk_sndbuf
;
1943 if (copy_to_user(argp
, &sndbuf
, sizeof(sndbuf
)))
1948 if (copy_from_user(&sndbuf
, argp
, sizeof(sndbuf
))) {
1953 tun
->sndbuf
= sndbuf
;
1954 tun_set_sndbuf(tun
);
1957 case TUNGETVNETHDRSZ
:
1958 vnet_hdr_sz
= tun
->vnet_hdr_sz
;
1959 if (copy_to_user(argp
, &vnet_hdr_sz
, sizeof(vnet_hdr_sz
)))
1963 case TUNSETVNETHDRSZ
:
1964 if (copy_from_user(&vnet_hdr_sz
, argp
, sizeof(vnet_hdr_sz
))) {
1968 if (vnet_hdr_sz
< (int)sizeof(struct virtio_net_hdr
)) {
1973 tun
->vnet_hdr_sz
= vnet_hdr_sz
;
1976 case TUNATTACHFILTER
:
1977 /* Can be set only for TAPs */
1979 if ((tun
->flags
& TUN_TYPE_MASK
) != TUN_TAP_DEV
)
1982 if (copy_from_user(&tun
->fprog
, argp
, sizeof(tun
->fprog
)))
1985 ret
= tun_attach_filter(tun
);
1988 case TUNDETACHFILTER
:
1989 /* Can be set only for TAPs */
1991 if ((tun
->flags
& TUN_TYPE_MASK
) != TUN_TAP_DEV
)
1994 tun_detach_filter(tun
, tun
->numqueues
);
2009 static long tun_chr_ioctl(struct file
*file
,
2010 unsigned int cmd
, unsigned long arg
)
2012 return __tun_chr_ioctl(file
, cmd
, arg
, sizeof (struct ifreq
));
2015 #ifdef CONFIG_COMPAT
2016 static long tun_chr_compat_ioctl(struct file
*file
,
2017 unsigned int cmd
, unsigned long arg
)
2022 case TUNSETTXFILTER
:
2027 arg
= (unsigned long)compat_ptr(arg
);
2030 arg
= (compat_ulong_t
)arg
;
2035 * compat_ifreq is shorter than ifreq, so we must not access beyond
2036 * the end of that structure. All fields that are used in this
2037 * driver are compatible though, we don't need to convert the
2040 return __tun_chr_ioctl(file
, cmd
, arg
, sizeof(struct compat_ifreq
));
2042 #endif /* CONFIG_COMPAT */
2044 static int tun_chr_fasync(int fd
, struct file
*file
, int on
)
2046 struct tun_file
*tfile
= file
->private_data
;
2049 if ((ret
= fasync_helper(fd
, file
, on
, &tfile
->fasync
)) < 0)
2053 ret
= __f_setown(file
, task_pid(current
), PIDTYPE_PID
, 0);
2056 tfile
->flags
|= TUN_FASYNC
;
2058 tfile
->flags
&= ~TUN_FASYNC
;
2064 static int tun_chr_open(struct inode
*inode
, struct file
* file
)
2066 struct tun_file
*tfile
;
2068 DBG1(KERN_INFO
, "tunX: tun_chr_open\n");
2070 tfile
= (struct tun_file
*)sk_alloc(&init_net
, AF_UNSPEC
, GFP_KERNEL
,
2074 rcu_assign_pointer(tfile
->tun
, NULL
);
2075 tfile
->net
= get_net(current
->nsproxy
->net_ns
);
2078 rcu_assign_pointer(tfile
->socket
.wq
, &tfile
->wq
);
2079 init_waitqueue_head(&tfile
->wq
.wait
);
2081 tfile
->socket
.file
= file
;
2082 tfile
->socket
.ops
= &tun_socket_ops
;
2084 sock_init_data(&tfile
->socket
, &tfile
->sk
);
2085 sk_change_net(&tfile
->sk
, tfile
->net
);
2087 tfile
->sk
.sk_write_space
= tun_sock_write_space
;
2088 tfile
->sk
.sk_sndbuf
= INT_MAX
;
2090 file
->private_data
= tfile
;
2091 set_bit(SOCK_EXTERNALLY_ALLOCATED
, &tfile
->socket
.flags
);
2096 static int tun_chr_close(struct inode
*inode
, struct file
*file
)
2098 struct tun_file
*tfile
= file
->private_data
;
2099 struct net
*net
= tfile
->net
;
2101 tun_detach(tfile
, true);
2107 static const struct file_operations tun_fops
= {
2108 .owner
= THIS_MODULE
,
2109 .llseek
= no_llseek
,
2110 .read
= do_sync_read
,
2111 .aio_read
= tun_chr_aio_read
,
2112 .write
= do_sync_write
,
2113 .aio_write
= tun_chr_aio_write
,
2114 .poll
= tun_chr_poll
,
2115 .unlocked_ioctl
= tun_chr_ioctl
,
2116 #ifdef CONFIG_COMPAT
2117 .compat_ioctl
= tun_chr_compat_ioctl
,
2119 .open
= tun_chr_open
,
2120 .release
= tun_chr_close
,
2121 .fasync
= tun_chr_fasync
2124 static struct miscdevice tun_miscdev
= {
2127 .nodename
= "net/tun",
2131 /* ethtool interface */
2133 static int tun_get_settings(struct net_device
*dev
, struct ethtool_cmd
*cmd
)
2136 cmd
->advertising
= 0;
2137 ethtool_cmd_speed_set(cmd
, SPEED_10
);
2138 cmd
->duplex
= DUPLEX_FULL
;
2139 cmd
->port
= PORT_TP
;
2140 cmd
->phy_address
= 0;
2141 cmd
->transceiver
= XCVR_INTERNAL
;
2142 cmd
->autoneg
= AUTONEG_DISABLE
;
2148 static void tun_get_drvinfo(struct net_device
*dev
, struct ethtool_drvinfo
*info
)
2150 struct tun_struct
*tun
= netdev_priv(dev
);
2152 strlcpy(info
->driver
, DRV_NAME
, sizeof(info
->driver
));
2153 strlcpy(info
->version
, DRV_VERSION
, sizeof(info
->version
));
2155 switch (tun
->flags
& TUN_TYPE_MASK
) {
2157 strlcpy(info
->bus_info
, "tun", sizeof(info
->bus_info
));
2160 strlcpy(info
->bus_info
, "tap", sizeof(info
->bus_info
));
2165 static u32
tun_get_msglevel(struct net_device
*dev
)
2168 struct tun_struct
*tun
= netdev_priv(dev
);
2175 static void tun_set_msglevel(struct net_device
*dev
, u32 value
)
2178 struct tun_struct
*tun
= netdev_priv(dev
);
2183 static const struct ethtool_ops tun_ethtool_ops
= {
2184 .get_settings
= tun_get_settings
,
2185 .get_drvinfo
= tun_get_drvinfo
,
2186 .get_msglevel
= tun_get_msglevel
,
2187 .set_msglevel
= tun_set_msglevel
,
2188 .get_link
= ethtool_op_get_link
,
2192 static int __init
tun_init(void)
2196 pr_info("%s, %s\n", DRV_DESCRIPTION
, DRV_VERSION
);
2197 pr_info("%s\n", DRV_COPYRIGHT
);
2199 ret
= rtnl_link_register(&tun_link_ops
);
2201 pr_err("Can't register link_ops\n");
2205 ret
= misc_register(&tun_miscdev
);
2207 pr_err("Can't register misc device %d\n", TUN_MINOR
);
2212 rtnl_link_unregister(&tun_link_ops
);
2217 static void tun_cleanup(void)
2219 misc_deregister(&tun_miscdev
);
2220 rtnl_link_unregister(&tun_link_ops
);
2223 /* Get an underlying socket object from tun file. Returns error unless file is
2224 * attached to a device. The returned object works like a packet socket, it
2225 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
2226 * holding a reference to the file for as long as the socket is in use. */
2227 struct socket
*tun_get_socket(struct file
*file
)
2229 struct tun_file
*tfile
;
2230 if (file
->f_op
!= &tun_fops
)
2231 return ERR_PTR(-EINVAL
);
2232 tfile
= file
->private_data
;
2234 return ERR_PTR(-EBADFD
);
2235 return &tfile
->socket
;
2237 EXPORT_SYMBOL_GPL(tun_get_socket
);
2239 module_init(tun_init
);
2240 module_exit(tun_cleanup
);
2241 MODULE_DESCRIPTION(DRV_DESCRIPTION
);
2242 MODULE_AUTHOR(DRV_COPYRIGHT
);
2243 MODULE_LICENSE("GPL");
2244 MODULE_ALIAS_MISCDEV(TUN_MINOR
);
2245 MODULE_ALIAS("devname:net/tun");