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>
72 #include <linux/seq_file.h>
73 #include <linux/uio.h>
75 #include <asm/uaccess.h>
77 /* Uncomment to enable debugging */
78 /* #define TUN_DEBUG 1 */
83 #define tun_debug(level, tun, fmt, args...) \
86 netdev_printk(level, tun->dev, fmt, ##args); \
88 #define DBG1(level, fmt, args...) \
91 printk(level fmt, ##args); \
94 #define tun_debug(level, tun, fmt, args...) \
97 netdev_printk(level, tun->dev, fmt, ##args); \
99 #define DBG1(level, fmt, args...) \
102 printk(level fmt, ##args); \
106 /* TUN device flags */
108 /* IFF_ATTACH_QUEUE is never stored in device flags,
109 * overload it to mean fasync when stored there.
111 #define TUN_FASYNC IFF_ATTACH_QUEUE
112 /* High bits in flags field are unused. */
113 #define TUN_VNET_LE 0x80000000
114 #define TUN_VNET_BE 0x40000000
116 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
118 #define GOODCOPY_LEN 128
120 #define FLT_EXACT_COUNT 8
122 unsigned int count
; /* Number of addrs. Zero means disabled */
123 u32 mask
[2]; /* Mask of the hashed addrs */
124 unsigned char addr
[FLT_EXACT_COUNT
][ETH_ALEN
];
127 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
128 * to max number of VCPUs in guest. */
129 #define MAX_TAP_QUEUES 256
130 #define MAX_TAP_FLOWS 4096
132 #define TUN_FLOW_EXPIRE (3 * HZ)
134 struct tun_pcpu_stats
{
139 struct u64_stats_sync syncp
;
145 /* A tun_file connects an open character device to a tuntap netdevice. It
146 * also contains all socket related structures (except sock_fprog and tap_filter)
147 * to serve as one transmit queue for tuntap device. The sock_fprog and
148 * tap_filter were kept in tun_struct since they were used for filtering for the
149 * netdevice not for a specific queue (at least I didn't see the requirement for
153 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
154 * other can only be read while rcu_read_lock or rtnl_lock is held.
158 struct socket socket
;
160 struct tun_struct __rcu
*tun
;
161 struct fasync_struct
*fasync
;
162 /* only used for fasnyc */
166 unsigned int ifindex
;
168 struct list_head next
;
169 struct tun_struct
*detached
;
172 struct tun_flow_entry
{
173 struct hlist_node hash_link
;
175 struct tun_struct
*tun
;
180 unsigned long updated
;
183 #define TUN_NUM_FLOW_ENTRIES 1024
185 /* Since the socket were moved to tun_file, to preserve the behavior of persist
186 * device, socket filter, sndbuf and vnet header size were restore when the
187 * file were attached to a persist device.
190 struct tun_file __rcu
*tfiles
[MAX_TAP_QUEUES
];
191 unsigned int numqueues
;
196 struct net_device
*dev
;
197 netdev_features_t set_features
;
198 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
199 NETIF_F_TSO6|NETIF_F_UFO)
204 struct tap_filter txflt
;
205 struct sock_fprog fprog
;
206 /* protected by rtnl lock */
207 bool filter_attached
;
212 struct hlist_head flows
[TUN_NUM_FLOW_ENTRIES
];
213 struct timer_list flow_gc_timer
;
214 unsigned long ageing_time
;
215 unsigned int numdisabled
;
216 struct list_head disabled
;
219 struct tun_pcpu_stats __percpu
*pcpu_stats
;
222 #ifdef CONFIG_TUN_VNET_CROSS_LE
223 static inline bool tun_legacy_is_little_endian(struct tun_struct
*tun
)
225 return tun
->flags
& TUN_VNET_BE
? false :
226 virtio_legacy_is_little_endian();
229 static long tun_get_vnet_be(struct tun_struct
*tun
, int __user
*argp
)
231 int be
= !!(tun
->flags
& TUN_VNET_BE
);
233 if (put_user(be
, argp
))
239 static long tun_set_vnet_be(struct tun_struct
*tun
, int __user
*argp
)
243 if (get_user(be
, argp
))
247 tun
->flags
|= TUN_VNET_BE
;
249 tun
->flags
&= ~TUN_VNET_BE
;
254 static inline bool tun_legacy_is_little_endian(struct tun_struct
*tun
)
256 return virtio_legacy_is_little_endian();
259 static long tun_get_vnet_be(struct tun_struct
*tun
, int __user
*argp
)
264 static long tun_set_vnet_be(struct tun_struct
*tun
, int __user
*argp
)
268 #endif /* CONFIG_TUN_VNET_CROSS_LE */
270 static inline bool tun_is_little_endian(struct tun_struct
*tun
)
272 return tun
->flags
& TUN_VNET_LE
||
273 tun_legacy_is_little_endian(tun
);
276 static inline u16
tun16_to_cpu(struct tun_struct
*tun
, __virtio16 val
)
278 return __virtio16_to_cpu(tun_is_little_endian(tun
), val
);
281 static inline __virtio16
cpu_to_tun16(struct tun_struct
*tun
, u16 val
)
283 return __cpu_to_virtio16(tun_is_little_endian(tun
), val
);
286 static inline u32
tun_hashfn(u32 rxhash
)
288 return rxhash
& 0x3ff;
291 static struct tun_flow_entry
*tun_flow_find(struct hlist_head
*head
, u32 rxhash
)
293 struct tun_flow_entry
*e
;
295 hlist_for_each_entry_rcu(e
, head
, hash_link
) {
296 if (e
->rxhash
== rxhash
)
302 static struct tun_flow_entry
*tun_flow_create(struct tun_struct
*tun
,
303 struct hlist_head
*head
,
304 u32 rxhash
, u16 queue_index
)
306 struct tun_flow_entry
*e
= kmalloc(sizeof(*e
), GFP_ATOMIC
);
309 tun_debug(KERN_INFO
, tun
, "create flow: hash %u index %u\n",
310 rxhash
, queue_index
);
311 e
->updated
= jiffies
;
314 e
->queue_index
= queue_index
;
316 hlist_add_head_rcu(&e
->hash_link
, head
);
322 static void tun_flow_delete(struct tun_struct
*tun
, struct tun_flow_entry
*e
)
324 tun_debug(KERN_INFO
, tun
, "delete flow: hash %u index %u\n",
325 e
->rxhash
, e
->queue_index
);
326 hlist_del_rcu(&e
->hash_link
);
331 static void tun_flow_flush(struct tun_struct
*tun
)
335 spin_lock_bh(&tun
->lock
);
336 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
337 struct tun_flow_entry
*e
;
338 struct hlist_node
*n
;
340 hlist_for_each_entry_safe(e
, n
, &tun
->flows
[i
], hash_link
)
341 tun_flow_delete(tun
, e
);
343 spin_unlock_bh(&tun
->lock
);
346 static void tun_flow_delete_by_queue(struct tun_struct
*tun
, u16 queue_index
)
350 spin_lock_bh(&tun
->lock
);
351 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
352 struct tun_flow_entry
*e
;
353 struct hlist_node
*n
;
355 hlist_for_each_entry_safe(e
, n
, &tun
->flows
[i
], hash_link
) {
356 if (e
->queue_index
== queue_index
)
357 tun_flow_delete(tun
, e
);
360 spin_unlock_bh(&tun
->lock
);
363 static void tun_flow_cleanup(unsigned long data
)
365 struct tun_struct
*tun
= (struct tun_struct
*)data
;
366 unsigned long delay
= tun
->ageing_time
;
367 unsigned long next_timer
= jiffies
+ delay
;
368 unsigned long count
= 0;
371 tun_debug(KERN_INFO
, tun
, "tun_flow_cleanup\n");
373 spin_lock_bh(&tun
->lock
);
374 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
375 struct tun_flow_entry
*e
;
376 struct hlist_node
*n
;
378 hlist_for_each_entry_safe(e
, n
, &tun
->flows
[i
], hash_link
) {
379 unsigned long this_timer
;
381 this_timer
= e
->updated
+ delay
;
382 if (time_before_eq(this_timer
, jiffies
))
383 tun_flow_delete(tun
, e
);
384 else if (time_before(this_timer
, next_timer
))
385 next_timer
= this_timer
;
390 mod_timer(&tun
->flow_gc_timer
, round_jiffies_up(next_timer
));
391 spin_unlock_bh(&tun
->lock
);
394 static void tun_flow_update(struct tun_struct
*tun
, u32 rxhash
,
395 struct tun_file
*tfile
)
397 struct hlist_head
*head
;
398 struct tun_flow_entry
*e
;
399 unsigned long delay
= tun
->ageing_time
;
400 u16 queue_index
= tfile
->queue_index
;
405 head
= &tun
->flows
[tun_hashfn(rxhash
)];
409 /* We may get a very small possibility of OOO during switching, not
410 * worth to optimize.*/
411 if (tun
->numqueues
== 1 || tfile
->detached
)
414 e
= tun_flow_find(head
, rxhash
);
416 /* TODO: keep queueing to old queue until it's empty? */
417 e
->queue_index
= queue_index
;
418 e
->updated
= jiffies
;
419 sock_rps_record_flow_hash(e
->rps_rxhash
);
421 spin_lock_bh(&tun
->lock
);
422 if (!tun_flow_find(head
, rxhash
) &&
423 tun
->flow_count
< MAX_TAP_FLOWS
)
424 tun_flow_create(tun
, head
, rxhash
, queue_index
);
426 if (!timer_pending(&tun
->flow_gc_timer
))
427 mod_timer(&tun
->flow_gc_timer
,
428 round_jiffies_up(jiffies
+ delay
));
429 spin_unlock_bh(&tun
->lock
);
437 * Save the hash received in the stack receive path and update the
438 * flow_hash table accordingly.
440 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry
*e
, u32 hash
)
442 if (unlikely(e
->rps_rxhash
!= hash
))
443 e
->rps_rxhash
= hash
;
446 /* We try to identify a flow through its rxhash first. The reason that
447 * we do not check rxq no. is because some cards(e.g 82599), chooses
448 * the rxq based on the txq where the last packet of the flow comes. As
449 * the userspace application move between processors, we may get a
450 * different rxq no. here. If we could not get rxhash, then we would
451 * hope the rxq no. may help here.
453 static u16
tun_select_queue(struct net_device
*dev
, struct sk_buff
*skb
,
454 void *accel_priv
, select_queue_fallback_t fallback
)
456 struct tun_struct
*tun
= netdev_priv(dev
);
457 struct tun_flow_entry
*e
;
462 numqueues
= ACCESS_ONCE(tun
->numqueues
);
464 txq
= skb_get_hash(skb
);
466 e
= tun_flow_find(&tun
->flows
[tun_hashfn(txq
)], txq
);
468 tun_flow_save_rps_rxhash(e
, txq
);
469 txq
= e
->queue_index
;
471 /* use multiply and shift instead of expensive divide */
472 txq
= ((u64
)txq
* numqueues
) >> 32;
473 } else if (likely(skb_rx_queue_recorded(skb
))) {
474 txq
= skb_get_rx_queue(skb
);
475 while (unlikely(txq
>= numqueues
))
483 static inline bool tun_not_capable(struct tun_struct
*tun
)
485 const struct cred
*cred
= current_cred();
486 struct net
*net
= dev_net(tun
->dev
);
488 return ((uid_valid(tun
->owner
) && !uid_eq(cred
->euid
, tun
->owner
)) ||
489 (gid_valid(tun
->group
) && !in_egroup_p(tun
->group
))) &&
490 !ns_capable(net
->user_ns
, CAP_NET_ADMIN
);
493 static void tun_set_real_num_queues(struct tun_struct
*tun
)
495 netif_set_real_num_tx_queues(tun
->dev
, tun
->numqueues
);
496 netif_set_real_num_rx_queues(tun
->dev
, tun
->numqueues
);
499 static void tun_disable_queue(struct tun_struct
*tun
, struct tun_file
*tfile
)
501 tfile
->detached
= tun
;
502 list_add_tail(&tfile
->next
, &tun
->disabled
);
506 static struct tun_struct
*tun_enable_queue(struct tun_file
*tfile
)
508 struct tun_struct
*tun
= tfile
->detached
;
510 tfile
->detached
= NULL
;
511 list_del_init(&tfile
->next
);
516 static void tun_queue_purge(struct tun_file
*tfile
)
518 skb_queue_purge(&tfile
->sk
.sk_receive_queue
);
519 skb_queue_purge(&tfile
->sk
.sk_error_queue
);
522 static void __tun_detach(struct tun_file
*tfile
, bool clean
)
524 struct tun_file
*ntfile
;
525 struct tun_struct
*tun
;
527 tun
= rtnl_dereference(tfile
->tun
);
529 if (tun
&& !tfile
->detached
) {
530 u16 index
= tfile
->queue_index
;
531 BUG_ON(index
>= tun
->numqueues
);
533 rcu_assign_pointer(tun
->tfiles
[index
],
534 tun
->tfiles
[tun
->numqueues
- 1]);
535 ntfile
= rtnl_dereference(tun
->tfiles
[index
]);
536 ntfile
->queue_index
= index
;
540 RCU_INIT_POINTER(tfile
->tun
, NULL
);
541 sock_put(&tfile
->sk
);
543 tun_disable_queue(tun
, tfile
);
546 tun_flow_delete_by_queue(tun
, tun
->numqueues
+ 1);
547 /* Drop read queue */
548 tun_queue_purge(tfile
);
549 tun_set_real_num_queues(tun
);
550 } else if (tfile
->detached
&& clean
) {
551 tun
= tun_enable_queue(tfile
);
552 sock_put(&tfile
->sk
);
556 if (tun
&& tun
->numqueues
== 0 && tun
->numdisabled
== 0) {
557 netif_carrier_off(tun
->dev
);
559 if (!(tun
->flags
& IFF_PERSIST
) &&
560 tun
->dev
->reg_state
== NETREG_REGISTERED
)
561 unregister_netdevice(tun
->dev
);
563 sock_put(&tfile
->sk
);
567 static void tun_detach(struct tun_file
*tfile
, bool clean
)
570 __tun_detach(tfile
, clean
);
574 static void tun_detach_all(struct net_device
*dev
)
576 struct tun_struct
*tun
= netdev_priv(dev
);
577 struct tun_file
*tfile
, *tmp
;
578 int i
, n
= tun
->numqueues
;
580 for (i
= 0; i
< n
; i
++) {
581 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
583 tfile
->socket
.sk
->sk_shutdown
= RCV_SHUTDOWN
;
584 tfile
->socket
.sk
->sk_data_ready(tfile
->socket
.sk
);
585 RCU_INIT_POINTER(tfile
->tun
, NULL
);
588 list_for_each_entry(tfile
, &tun
->disabled
, next
) {
589 tfile
->socket
.sk
->sk_shutdown
= RCV_SHUTDOWN
;
590 tfile
->socket
.sk
->sk_data_ready(tfile
->socket
.sk
);
591 RCU_INIT_POINTER(tfile
->tun
, NULL
);
593 BUG_ON(tun
->numqueues
!= 0);
596 for (i
= 0; i
< n
; i
++) {
597 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
598 /* Drop read queue */
599 tun_queue_purge(tfile
);
600 sock_put(&tfile
->sk
);
602 list_for_each_entry_safe(tfile
, tmp
, &tun
->disabled
, next
) {
603 tun_enable_queue(tfile
);
604 tun_queue_purge(tfile
);
605 sock_put(&tfile
->sk
);
607 BUG_ON(tun
->numdisabled
!= 0);
609 if (tun
->flags
& IFF_PERSIST
)
610 module_put(THIS_MODULE
);
613 static int tun_attach(struct tun_struct
*tun
, struct file
*file
, bool skip_filter
)
615 struct tun_file
*tfile
= file
->private_data
;
618 err
= security_tun_dev_attach(tfile
->socket
.sk
, tun
->security
);
623 if (rtnl_dereference(tfile
->tun
) && !tfile
->detached
)
627 if (!(tun
->flags
& IFF_MULTI_QUEUE
) && tun
->numqueues
== 1)
631 if (!tfile
->detached
&&
632 tun
->numqueues
+ tun
->numdisabled
== MAX_TAP_QUEUES
)
637 /* Re-attach the filter to persist device */
638 if (!skip_filter
&& (tun
->filter_attached
== true)) {
639 lock_sock(tfile
->socket
.sk
);
640 err
= sk_attach_filter(&tun
->fprog
, tfile
->socket
.sk
);
641 release_sock(tfile
->socket
.sk
);
645 tfile
->queue_index
= tun
->numqueues
;
646 tfile
->socket
.sk
->sk_shutdown
&= ~RCV_SHUTDOWN
;
647 rcu_assign_pointer(tfile
->tun
, tun
);
648 rcu_assign_pointer(tun
->tfiles
[tun
->numqueues
], tfile
);
652 tun_enable_queue(tfile
);
654 sock_hold(&tfile
->sk
);
656 tun_set_real_num_queues(tun
);
658 /* device is allowed to go away first, so no need to hold extra
666 static struct tun_struct
*__tun_get(struct tun_file
*tfile
)
668 struct tun_struct
*tun
;
671 tun
= rcu_dereference(tfile
->tun
);
679 static struct tun_struct
*tun_get(struct file
*file
)
681 return __tun_get(file
->private_data
);
684 static void tun_put(struct tun_struct
*tun
)
690 static void addr_hash_set(u32
*mask
, const u8
*addr
)
692 int n
= ether_crc(ETH_ALEN
, addr
) >> 26;
693 mask
[n
>> 5] |= (1 << (n
& 31));
696 static unsigned int addr_hash_test(const u32
*mask
, const u8
*addr
)
698 int n
= ether_crc(ETH_ALEN
, addr
) >> 26;
699 return mask
[n
>> 5] & (1 << (n
& 31));
702 static int update_filter(struct tap_filter
*filter
, void __user
*arg
)
704 struct { u8 u
[ETH_ALEN
]; } *addr
;
705 struct tun_filter uf
;
706 int err
, alen
, n
, nexact
;
708 if (copy_from_user(&uf
, arg
, sizeof(uf
)))
717 alen
= ETH_ALEN
* uf
.count
;
718 addr
= kmalloc(alen
, GFP_KERNEL
);
722 if (copy_from_user(addr
, arg
+ sizeof(uf
), alen
)) {
727 /* The filter is updated without holding any locks. Which is
728 * perfectly safe. We disable it first and in the worst
729 * case we'll accept a few undesired packets. */
733 /* Use first set of addresses as an exact filter */
734 for (n
= 0; n
< uf
.count
&& n
< FLT_EXACT_COUNT
; n
++)
735 memcpy(filter
->addr
[n
], addr
[n
].u
, ETH_ALEN
);
739 /* Remaining multicast addresses are hashed,
740 * unicast will leave the filter disabled. */
741 memset(filter
->mask
, 0, sizeof(filter
->mask
));
742 for (; n
< uf
.count
; n
++) {
743 if (!is_multicast_ether_addr(addr
[n
].u
)) {
744 err
= 0; /* no filter */
747 addr_hash_set(filter
->mask
, addr
[n
].u
);
750 /* For ALLMULTI just set the mask to all ones.
751 * This overrides the mask populated above. */
752 if ((uf
.flags
& TUN_FLT_ALLMULTI
))
753 memset(filter
->mask
, ~0, sizeof(filter
->mask
));
755 /* Now enable the filter */
757 filter
->count
= nexact
;
759 /* Return the number of exact filters */
767 /* Returns: 0 - drop, !=0 - accept */
768 static int run_filter(struct tap_filter
*filter
, const struct sk_buff
*skb
)
770 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
772 struct ethhdr
*eh
= (struct ethhdr
*) skb
->data
;
776 for (i
= 0; i
< filter
->count
; i
++)
777 if (ether_addr_equal(eh
->h_dest
, filter
->addr
[i
]))
780 /* Inexact match (multicast only) */
781 if (is_multicast_ether_addr(eh
->h_dest
))
782 return addr_hash_test(filter
->mask
, eh
->h_dest
);
788 * Checks whether the packet is accepted or not.
789 * Returns: 0 - drop, !=0 - accept
791 static int check_filter(struct tap_filter
*filter
, const struct sk_buff
*skb
)
796 return run_filter(filter
, skb
);
799 /* Network device part of the driver */
801 static const struct ethtool_ops tun_ethtool_ops
;
803 /* Net device detach from fd. */
804 static void tun_net_uninit(struct net_device
*dev
)
809 /* Net device open. */
810 static int tun_net_open(struct net_device
*dev
)
812 netif_tx_start_all_queues(dev
);
816 /* Net device close. */
817 static int tun_net_close(struct net_device
*dev
)
819 netif_tx_stop_all_queues(dev
);
823 /* Net device start xmit */
824 static netdev_tx_t
tun_net_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
826 struct tun_struct
*tun
= netdev_priv(dev
);
827 int txq
= skb
->queue_mapping
;
828 struct tun_file
*tfile
;
832 tfile
= rcu_dereference(tun
->tfiles
[txq
]);
833 numqueues
= ACCESS_ONCE(tun
->numqueues
);
835 /* Drop packet if interface is not attached */
836 if (txq
>= numqueues
)
840 if (numqueues
== 1 && static_key_false(&rps_needed
)) {
841 /* Select queue was not called for the skbuff, so we extract the
842 * RPS hash and save it into the flow_table here.
846 rxhash
= skb_get_hash(skb
);
848 struct tun_flow_entry
*e
;
849 e
= tun_flow_find(&tun
->flows
[tun_hashfn(rxhash
)],
852 tun_flow_save_rps_rxhash(e
, rxhash
);
857 tun_debug(KERN_INFO
, tun
, "tun_net_xmit %d\n", skb
->len
);
861 /* Drop if the filter does not like it.
862 * This is a noop if the filter is disabled.
863 * Filter can be enabled only for the TAP devices. */
864 if (!check_filter(&tun
->txflt
, skb
))
867 if (tfile
->socket
.sk
->sk_filter
&&
868 sk_filter(tfile
->socket
.sk
, skb
))
871 /* Limit the number of packets queued by dividing txq length with the
874 if (skb_queue_len(&tfile
->socket
.sk
->sk_receive_queue
) * numqueues
875 >= dev
->tx_queue_len
)
878 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
881 if (skb
->sk
&& sk_fullsock(skb
->sk
)) {
882 sock_tx_timestamp(skb
->sk
, skb
->sk
->sk_tsflags
,
883 &skb_shinfo(skb
)->tx_flags
);
884 sw_tx_timestamp(skb
);
887 /* Orphan the skb - required as we might hang on to it
888 * for indefinite time.
895 skb_queue_tail(&tfile
->socket
.sk
->sk_receive_queue
, skb
);
897 /* Notify and wake up reader process */
898 if (tfile
->flags
& TUN_FASYNC
)
899 kill_fasync(&tfile
->fasync
, SIGIO
, POLL_IN
);
900 tfile
->socket
.sk
->sk_data_ready(tfile
->socket
.sk
);
906 this_cpu_inc(tun
->pcpu_stats
->tx_dropped
);
910 return NET_XMIT_DROP
;
913 static void tun_net_mclist(struct net_device
*dev
)
916 * This callback is supposed to deal with mc filter in
917 * _rx_ path and has nothing to do with the _tx_ path.
918 * In rx path we always accept everything userspace gives us.
923 #define MAX_MTU 65535
926 tun_net_change_mtu(struct net_device
*dev
, int new_mtu
)
928 if (new_mtu
< MIN_MTU
|| new_mtu
+ dev
->hard_header_len
> MAX_MTU
)
934 static netdev_features_t
tun_net_fix_features(struct net_device
*dev
,
935 netdev_features_t features
)
937 struct tun_struct
*tun
= netdev_priv(dev
);
939 return (features
& tun
->set_features
) | (features
& ~TUN_USER_FEATURES
);
941 #ifdef CONFIG_NET_POLL_CONTROLLER
942 static void tun_poll_controller(struct net_device
*dev
)
945 * Tun only receives frames when:
946 * 1) the char device endpoint gets data from user space
947 * 2) the tun socket gets a sendmsg call from user space
948 * Since both of those are synchronous operations, we are guaranteed
949 * never to have pending data when we poll for it
950 * so there is nothing to do here but return.
951 * We need this though so netpoll recognizes us as an interface that
952 * supports polling, which enables bridge devices in virt setups to
953 * still use netconsole
959 static void tun_set_headroom(struct net_device
*dev
, int new_hr
)
961 struct tun_struct
*tun
= netdev_priv(dev
);
963 if (new_hr
< NET_SKB_PAD
)
964 new_hr
= NET_SKB_PAD
;
969 static struct rtnl_link_stats64
*
970 tun_net_get_stats64(struct net_device
*dev
, struct rtnl_link_stats64
*stats
)
972 u32 rx_dropped
= 0, tx_dropped
= 0, rx_frame_errors
= 0;
973 struct tun_struct
*tun
= netdev_priv(dev
);
974 struct tun_pcpu_stats
*p
;
977 for_each_possible_cpu(i
) {
978 u64 rxpackets
, rxbytes
, txpackets
, txbytes
;
981 p
= per_cpu_ptr(tun
->pcpu_stats
, i
);
983 start
= u64_stats_fetch_begin(&p
->syncp
);
984 rxpackets
= p
->rx_packets
;
985 rxbytes
= p
->rx_bytes
;
986 txpackets
= p
->tx_packets
;
987 txbytes
= p
->tx_bytes
;
988 } while (u64_stats_fetch_retry(&p
->syncp
, start
));
990 stats
->rx_packets
+= rxpackets
;
991 stats
->rx_bytes
+= rxbytes
;
992 stats
->tx_packets
+= txpackets
;
993 stats
->tx_bytes
+= txbytes
;
996 rx_dropped
+= p
->rx_dropped
;
997 rx_frame_errors
+= p
->rx_frame_errors
;
998 tx_dropped
+= p
->tx_dropped
;
1000 stats
->rx_dropped
= rx_dropped
;
1001 stats
->rx_frame_errors
= rx_frame_errors
;
1002 stats
->tx_dropped
= tx_dropped
;
1006 static const struct net_device_ops tun_netdev_ops
= {
1007 .ndo_uninit
= tun_net_uninit
,
1008 .ndo_open
= tun_net_open
,
1009 .ndo_stop
= tun_net_close
,
1010 .ndo_start_xmit
= tun_net_xmit
,
1011 .ndo_change_mtu
= tun_net_change_mtu
,
1012 .ndo_fix_features
= tun_net_fix_features
,
1013 .ndo_select_queue
= tun_select_queue
,
1014 #ifdef CONFIG_NET_POLL_CONTROLLER
1015 .ndo_poll_controller
= tun_poll_controller
,
1017 .ndo_set_rx_headroom
= tun_set_headroom
,
1018 .ndo_get_stats64
= tun_net_get_stats64
,
1021 static const struct net_device_ops tap_netdev_ops
= {
1022 .ndo_uninit
= tun_net_uninit
,
1023 .ndo_open
= tun_net_open
,
1024 .ndo_stop
= tun_net_close
,
1025 .ndo_start_xmit
= tun_net_xmit
,
1026 .ndo_change_mtu
= tun_net_change_mtu
,
1027 .ndo_fix_features
= tun_net_fix_features
,
1028 .ndo_set_rx_mode
= tun_net_mclist
,
1029 .ndo_set_mac_address
= eth_mac_addr
,
1030 .ndo_validate_addr
= eth_validate_addr
,
1031 .ndo_select_queue
= tun_select_queue
,
1032 #ifdef CONFIG_NET_POLL_CONTROLLER
1033 .ndo_poll_controller
= tun_poll_controller
,
1035 .ndo_features_check
= passthru_features_check
,
1036 .ndo_set_rx_headroom
= tun_set_headroom
,
1037 .ndo_get_stats64
= tun_net_get_stats64
,
1040 static void tun_flow_init(struct tun_struct
*tun
)
1044 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++)
1045 INIT_HLIST_HEAD(&tun
->flows
[i
]);
1047 tun
->ageing_time
= TUN_FLOW_EXPIRE
;
1048 setup_timer(&tun
->flow_gc_timer
, tun_flow_cleanup
, (unsigned long)tun
);
1049 mod_timer(&tun
->flow_gc_timer
,
1050 round_jiffies_up(jiffies
+ tun
->ageing_time
));
1053 static void tun_flow_uninit(struct tun_struct
*tun
)
1055 del_timer_sync(&tun
->flow_gc_timer
);
1056 tun_flow_flush(tun
);
1059 /* Initialize net device. */
1060 static void tun_net_init(struct net_device
*dev
)
1062 struct tun_struct
*tun
= netdev_priv(dev
);
1064 switch (tun
->flags
& TUN_TYPE_MASK
) {
1066 dev
->netdev_ops
= &tun_netdev_ops
;
1068 /* Point-to-Point TUN Device */
1069 dev
->hard_header_len
= 0;
1073 /* Zero header length */
1074 dev
->type
= ARPHRD_NONE
;
1075 dev
->flags
= IFF_POINTOPOINT
| IFF_NOARP
| IFF_MULTICAST
;
1079 dev
->netdev_ops
= &tap_netdev_ops
;
1080 /* Ethernet TAP Device */
1082 dev
->priv_flags
&= ~IFF_TX_SKB_SHARING
;
1083 dev
->priv_flags
|= IFF_LIVE_ADDR_CHANGE
;
1085 eth_hw_addr_random(dev
);
1091 /* Character device part */
1094 static unsigned int tun_chr_poll(struct file
*file
, poll_table
*wait
)
1096 struct tun_file
*tfile
= file
->private_data
;
1097 struct tun_struct
*tun
= __tun_get(tfile
);
1099 unsigned int mask
= 0;
1104 sk
= tfile
->socket
.sk
;
1106 tun_debug(KERN_INFO
, tun
, "tun_chr_poll\n");
1108 poll_wait(file
, sk_sleep(sk
), wait
);
1110 if (!skb_queue_empty(&sk
->sk_receive_queue
))
1111 mask
|= POLLIN
| POLLRDNORM
;
1113 if (sock_writeable(sk
) ||
1114 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
) &&
1115 sock_writeable(sk
)))
1116 mask
|= POLLOUT
| POLLWRNORM
;
1118 if (tun
->dev
->reg_state
!= NETREG_REGISTERED
)
1125 /* prepad is the amount to reserve at front. len is length after that.
1126 * linear is a hint as to how much to copy (usually headers). */
1127 static struct sk_buff
*tun_alloc_skb(struct tun_file
*tfile
,
1128 size_t prepad
, size_t len
,
1129 size_t linear
, int noblock
)
1131 struct sock
*sk
= tfile
->socket
.sk
;
1132 struct sk_buff
*skb
;
1135 /* Under a page? Don't bother with paged skb. */
1136 if (prepad
+ len
< PAGE_SIZE
|| !linear
)
1139 skb
= sock_alloc_send_pskb(sk
, prepad
+ linear
, len
- linear
, noblock
,
1142 return ERR_PTR(err
);
1144 skb_reserve(skb
, prepad
);
1145 skb_put(skb
, linear
);
1146 skb
->data_len
= len
- linear
;
1147 skb
->len
+= len
- linear
;
1152 /* Get packet from user space buffer */
1153 static ssize_t
tun_get_user(struct tun_struct
*tun
, struct tun_file
*tfile
,
1154 void *msg_control
, struct iov_iter
*from
,
1157 struct tun_pi pi
= { 0, cpu_to_be16(ETH_P_IP
) };
1158 struct sk_buff
*skb
;
1159 size_t total_len
= iov_iter_count(from
);
1160 size_t len
= total_len
, align
= tun
->align
, linear
;
1161 struct virtio_net_hdr gso
= { 0 };
1162 struct tun_pcpu_stats
*stats
;
1165 bool zerocopy
= false;
1170 if (!(tun
->dev
->flags
& IFF_UP
))
1173 if (!(tun
->flags
& IFF_NO_PI
)) {
1174 if (len
< sizeof(pi
))
1178 n
= copy_from_iter(&pi
, sizeof(pi
), from
);
1179 if (n
!= sizeof(pi
))
1183 if (tun
->flags
& IFF_VNET_HDR
) {
1184 if (len
< tun
->vnet_hdr_sz
)
1186 len
-= tun
->vnet_hdr_sz
;
1188 n
= copy_from_iter(&gso
, sizeof(gso
), from
);
1189 if (n
!= sizeof(gso
))
1192 if ((gso
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) &&
1193 tun16_to_cpu(tun
, gso
.csum_start
) + tun16_to_cpu(tun
, gso
.csum_offset
) + 2 > tun16_to_cpu(tun
, gso
.hdr_len
))
1194 gso
.hdr_len
= cpu_to_tun16(tun
, tun16_to_cpu(tun
, gso
.csum_start
) + tun16_to_cpu(tun
, gso
.csum_offset
) + 2);
1196 if (tun16_to_cpu(tun
, gso
.hdr_len
) > len
)
1198 iov_iter_advance(from
, tun
->vnet_hdr_sz
- sizeof(gso
));
1201 if ((tun
->flags
& TUN_TYPE_MASK
) == IFF_TAP
) {
1202 align
+= NET_IP_ALIGN
;
1203 if (unlikely(len
< ETH_HLEN
||
1204 (gso
.hdr_len
&& tun16_to_cpu(tun
, gso
.hdr_len
) < ETH_HLEN
)))
1208 good_linear
= SKB_MAX_HEAD(align
);
1211 struct iov_iter i
= *from
;
1213 /* There are 256 bytes to be copied in skb, so there is
1214 * enough room for skb expand head in case it is used.
1215 * The rest of the buffer is mapped from userspace.
1217 copylen
= gso
.hdr_len
? tun16_to_cpu(tun
, gso
.hdr_len
) : GOODCOPY_LEN
;
1218 if (copylen
> good_linear
)
1219 copylen
= good_linear
;
1221 iov_iter_advance(&i
, copylen
);
1222 if (iov_iter_npages(&i
, INT_MAX
) <= MAX_SKB_FRAGS
)
1228 if (tun16_to_cpu(tun
, gso
.hdr_len
) > good_linear
)
1229 linear
= good_linear
;
1231 linear
= tun16_to_cpu(tun
, gso
.hdr_len
);
1234 skb
= tun_alloc_skb(tfile
, align
, copylen
, linear
, noblock
);
1236 if (PTR_ERR(skb
) != -EAGAIN
)
1237 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1238 return PTR_ERR(skb
);
1242 err
= zerocopy_sg_from_iter(skb
, from
);
1244 err
= skb_copy_datagram_from_iter(skb
, 0, from
, len
);
1245 if (!err
&& msg_control
) {
1246 struct ubuf_info
*uarg
= msg_control
;
1247 uarg
->callback(uarg
, false);
1252 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1257 err
= virtio_net_hdr_to_skb(skb
, &gso
, tun_is_little_endian(tun
));
1259 this_cpu_inc(tun
->pcpu_stats
->rx_frame_errors
);
1264 switch (tun
->flags
& TUN_TYPE_MASK
) {
1266 if (tun
->flags
& IFF_NO_PI
) {
1267 switch (skb
->data
[0] & 0xf0) {
1269 pi
.proto
= htons(ETH_P_IP
);
1272 pi
.proto
= htons(ETH_P_IPV6
);
1275 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1281 skb_reset_mac_header(skb
);
1282 skb
->protocol
= pi
.proto
;
1283 skb
->dev
= tun
->dev
;
1286 skb
->protocol
= eth_type_trans(skb
, tun
->dev
);
1290 /* copy skb_ubuf_info for callback when skb has no error */
1292 skb_shinfo(skb
)->destructor_arg
= msg_control
;
1293 skb_shinfo(skb
)->tx_flags
|= SKBTX_DEV_ZEROCOPY
;
1294 skb_shinfo(skb
)->tx_flags
|= SKBTX_SHARED_FRAG
;
1297 skb_reset_network_header(skb
);
1298 skb_probe_transport_header(skb
, 0);
1300 rxhash
= skb_get_hash(skb
);
1303 stats
= get_cpu_ptr(tun
->pcpu_stats
);
1304 u64_stats_update_begin(&stats
->syncp
);
1305 stats
->rx_packets
++;
1306 stats
->rx_bytes
+= len
;
1307 u64_stats_update_end(&stats
->syncp
);
1310 tun_flow_update(tun
, rxhash
, tfile
);
1314 static ssize_t
tun_chr_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
1316 struct file
*file
= iocb
->ki_filp
;
1317 struct tun_struct
*tun
= tun_get(file
);
1318 struct tun_file
*tfile
= file
->private_data
;
1324 result
= tun_get_user(tun
, tfile
, NULL
, from
, file
->f_flags
& O_NONBLOCK
);
1330 /* Put packet to the user space buffer */
1331 static ssize_t
tun_put_user(struct tun_struct
*tun
,
1332 struct tun_file
*tfile
,
1333 struct sk_buff
*skb
,
1334 struct iov_iter
*iter
)
1336 struct tun_pi pi
= { 0, skb
->protocol
};
1337 struct tun_pcpu_stats
*stats
;
1339 int vlan_offset
= 0;
1341 int vnet_hdr_sz
= 0;
1343 if (skb_vlan_tag_present(skb
))
1344 vlan_hlen
= VLAN_HLEN
;
1346 if (tun
->flags
& IFF_VNET_HDR
)
1347 vnet_hdr_sz
= tun
->vnet_hdr_sz
;
1349 total
= skb
->len
+ vlan_hlen
+ vnet_hdr_sz
;
1351 if (!(tun
->flags
& IFF_NO_PI
)) {
1352 if (iov_iter_count(iter
) < sizeof(pi
))
1355 total
+= sizeof(pi
);
1356 if (iov_iter_count(iter
) < total
) {
1357 /* Packet will be striped */
1358 pi
.flags
|= TUN_PKT_STRIP
;
1361 if (copy_to_iter(&pi
, sizeof(pi
), iter
) != sizeof(pi
))
1366 struct virtio_net_hdr gso
= { 0 }; /* no info leak */
1369 if (iov_iter_count(iter
) < vnet_hdr_sz
)
1372 ret
= virtio_net_hdr_from_skb(skb
, &gso
,
1373 tun_is_little_endian(tun
));
1375 struct skb_shared_info
*sinfo
= skb_shinfo(skb
);
1376 pr_err("unexpected GSO type: "
1377 "0x%x, gso_size %d, hdr_len %d\n",
1378 sinfo
->gso_type
, tun16_to_cpu(tun
, gso
.gso_size
),
1379 tun16_to_cpu(tun
, gso
.hdr_len
));
1380 print_hex_dump(KERN_ERR
, "tun: ",
1383 min((int)tun16_to_cpu(tun
, gso
.hdr_len
), 64), true);
1388 if (copy_to_iter(&gso
, sizeof(gso
), iter
) != sizeof(gso
))
1391 iov_iter_advance(iter
, vnet_hdr_sz
- sizeof(gso
));
1397 __be16 h_vlan_proto
;
1401 veth
.h_vlan_proto
= skb
->vlan_proto
;
1402 veth
.h_vlan_TCI
= htons(skb_vlan_tag_get(skb
));
1404 vlan_offset
= offsetof(struct vlan_ethhdr
, h_vlan_proto
);
1406 ret
= skb_copy_datagram_iter(skb
, 0, iter
, vlan_offset
);
1407 if (ret
|| !iov_iter_count(iter
))
1410 ret
= copy_to_iter(&veth
, sizeof(veth
), iter
);
1411 if (ret
!= sizeof(veth
) || !iov_iter_count(iter
))
1415 skb_copy_datagram_iter(skb
, vlan_offset
, iter
, skb
->len
- vlan_offset
);
1418 /* caller is in process context, */
1419 stats
= get_cpu_ptr(tun
->pcpu_stats
);
1420 u64_stats_update_begin(&stats
->syncp
);
1421 stats
->tx_packets
++;
1422 stats
->tx_bytes
+= skb
->len
+ vlan_hlen
;
1423 u64_stats_update_end(&stats
->syncp
);
1424 put_cpu_ptr(tun
->pcpu_stats
);
1429 static ssize_t
tun_do_read(struct tun_struct
*tun
, struct tun_file
*tfile
,
1430 struct iov_iter
*to
,
1433 struct sk_buff
*skb
;
1435 int peeked
, err
, off
= 0;
1437 tun_debug(KERN_INFO
, tun
, "tun_do_read\n");
1439 if (!iov_iter_count(to
))
1442 /* Read frames from queue */
1443 skb
= __skb_recv_datagram(tfile
->socket
.sk
, noblock
? MSG_DONTWAIT
: 0,
1444 &peeked
, &off
, &err
);
1448 ret
= tun_put_user(tun
, tfile
, skb
, to
);
1449 if (unlikely(ret
< 0))
1457 static ssize_t
tun_chr_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
1459 struct file
*file
= iocb
->ki_filp
;
1460 struct tun_file
*tfile
= file
->private_data
;
1461 struct tun_struct
*tun
= __tun_get(tfile
);
1462 ssize_t len
= iov_iter_count(to
), ret
;
1466 ret
= tun_do_read(tun
, tfile
, to
, file
->f_flags
& O_NONBLOCK
);
1467 ret
= min_t(ssize_t
, ret
, len
);
1474 static void tun_free_netdev(struct net_device
*dev
)
1476 struct tun_struct
*tun
= netdev_priv(dev
);
1478 BUG_ON(!(list_empty(&tun
->disabled
)));
1479 free_percpu(tun
->pcpu_stats
);
1480 tun_flow_uninit(tun
);
1481 security_tun_dev_free_security(tun
->security
);
1485 static void tun_setup(struct net_device
*dev
)
1487 struct tun_struct
*tun
= netdev_priv(dev
);
1489 tun
->owner
= INVALID_UID
;
1490 tun
->group
= INVALID_GID
;
1492 dev
->ethtool_ops
= &tun_ethtool_ops
;
1493 dev
->destructor
= tun_free_netdev
;
1494 /* We prefer our own queue length */
1495 dev
->tx_queue_len
= TUN_READQ_SIZE
;
1498 /* Trivial set of netlink ops to allow deleting tun or tap
1499 * device with netlink.
1501 static int tun_validate(struct nlattr
*tb
[], struct nlattr
*data
[])
1506 static struct rtnl_link_ops tun_link_ops __read_mostly
= {
1508 .priv_size
= sizeof(struct tun_struct
),
1510 .validate
= tun_validate
,
1513 static void tun_sock_write_space(struct sock
*sk
)
1515 struct tun_file
*tfile
;
1516 wait_queue_head_t
*wqueue
;
1518 if (!sock_writeable(sk
))
1521 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
))
1524 wqueue
= sk_sleep(sk
);
1525 if (wqueue
&& waitqueue_active(wqueue
))
1526 wake_up_interruptible_sync_poll(wqueue
, POLLOUT
|
1527 POLLWRNORM
| POLLWRBAND
);
1529 tfile
= container_of(sk
, struct tun_file
, sk
);
1530 kill_fasync(&tfile
->fasync
, SIGIO
, POLL_OUT
);
1533 static int tun_sendmsg(struct socket
*sock
, struct msghdr
*m
, size_t total_len
)
1536 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
1537 struct tun_struct
*tun
= __tun_get(tfile
);
1542 ret
= tun_get_user(tun
, tfile
, m
->msg_control
, &m
->msg_iter
,
1543 m
->msg_flags
& MSG_DONTWAIT
);
1548 static int tun_recvmsg(struct socket
*sock
, struct msghdr
*m
, size_t total_len
,
1551 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
1552 struct tun_struct
*tun
= __tun_get(tfile
);
1558 if (flags
& ~(MSG_DONTWAIT
|MSG_TRUNC
|MSG_ERRQUEUE
)) {
1562 if (flags
& MSG_ERRQUEUE
) {
1563 ret
= sock_recv_errqueue(sock
->sk
, m
, total_len
,
1564 SOL_PACKET
, TUN_TX_TIMESTAMP
);
1567 ret
= tun_do_read(tun
, tfile
, &m
->msg_iter
, flags
& MSG_DONTWAIT
);
1568 if (ret
> (ssize_t
)total_len
) {
1569 m
->msg_flags
|= MSG_TRUNC
;
1570 ret
= flags
& MSG_TRUNC
? ret
: total_len
;
1577 /* Ops structure to mimic raw sockets with tun */
1578 static const struct proto_ops tun_socket_ops
= {
1579 .sendmsg
= tun_sendmsg
,
1580 .recvmsg
= tun_recvmsg
,
1583 static struct proto tun_proto
= {
1585 .owner
= THIS_MODULE
,
1586 .obj_size
= sizeof(struct tun_file
),
1589 static int tun_flags(struct tun_struct
*tun
)
1591 return tun
->flags
& (TUN_FEATURES
| IFF_PERSIST
| IFF_TUN
| IFF_TAP
);
1594 static ssize_t
tun_show_flags(struct device
*dev
, struct device_attribute
*attr
,
1597 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
1598 return sprintf(buf
, "0x%x\n", tun_flags(tun
));
1601 static ssize_t
tun_show_owner(struct device
*dev
, struct device_attribute
*attr
,
1604 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
1605 return uid_valid(tun
->owner
)?
1606 sprintf(buf
, "%u\n",
1607 from_kuid_munged(current_user_ns(), tun
->owner
)):
1608 sprintf(buf
, "-1\n");
1611 static ssize_t
tun_show_group(struct device
*dev
, struct device_attribute
*attr
,
1614 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
1615 return gid_valid(tun
->group
) ?
1616 sprintf(buf
, "%u\n",
1617 from_kgid_munged(current_user_ns(), tun
->group
)):
1618 sprintf(buf
, "-1\n");
1621 static DEVICE_ATTR(tun_flags
, 0444, tun_show_flags
, NULL
);
1622 static DEVICE_ATTR(owner
, 0444, tun_show_owner
, NULL
);
1623 static DEVICE_ATTR(group
, 0444, tun_show_group
, NULL
);
1625 static struct attribute
*tun_dev_attrs
[] = {
1626 &dev_attr_tun_flags
.attr
,
1627 &dev_attr_owner
.attr
,
1628 &dev_attr_group
.attr
,
1632 static const struct attribute_group tun_attr_group
= {
1633 .attrs
= tun_dev_attrs
1636 static int tun_set_iff(struct net
*net
, struct file
*file
, struct ifreq
*ifr
)
1638 struct tun_struct
*tun
;
1639 struct tun_file
*tfile
= file
->private_data
;
1640 struct net_device
*dev
;
1643 if (tfile
->detached
)
1646 dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
1648 if (ifr
->ifr_flags
& IFF_TUN_EXCL
)
1650 if ((ifr
->ifr_flags
& IFF_TUN
) && dev
->netdev_ops
== &tun_netdev_ops
)
1651 tun
= netdev_priv(dev
);
1652 else if ((ifr
->ifr_flags
& IFF_TAP
) && dev
->netdev_ops
== &tap_netdev_ops
)
1653 tun
= netdev_priv(dev
);
1657 if (!!(ifr
->ifr_flags
& IFF_MULTI_QUEUE
) !=
1658 !!(tun
->flags
& IFF_MULTI_QUEUE
))
1661 if (tun_not_capable(tun
))
1663 err
= security_tun_dev_open(tun
->security
);
1667 err
= tun_attach(tun
, file
, ifr
->ifr_flags
& IFF_NOFILTER
);
1671 if (tun
->flags
& IFF_MULTI_QUEUE
&&
1672 (tun
->numqueues
+ tun
->numdisabled
> 1)) {
1673 /* One or more queue has already been attached, no need
1674 * to initialize the device again.
1681 unsigned long flags
= 0;
1682 int queues
= ifr
->ifr_flags
& IFF_MULTI_QUEUE
?
1685 if (!ns_capable(net
->user_ns
, CAP_NET_ADMIN
))
1687 err
= security_tun_dev_create();
1692 if (ifr
->ifr_flags
& IFF_TUN
) {
1696 } else if (ifr
->ifr_flags
& IFF_TAP
) {
1704 name
= ifr
->ifr_name
;
1706 dev
= alloc_netdev_mqs(sizeof(struct tun_struct
), name
,
1707 NET_NAME_UNKNOWN
, tun_setup
, queues
,
1713 dev_net_set(dev
, net
);
1714 dev
->rtnl_link_ops
= &tun_link_ops
;
1715 dev
->ifindex
= tfile
->ifindex
;
1716 dev
->sysfs_groups
[0] = &tun_attr_group
;
1718 tun
= netdev_priv(dev
);
1721 tun
->txflt
.count
= 0;
1722 tun
->vnet_hdr_sz
= sizeof(struct virtio_net_hdr
);
1724 tun
->align
= NET_SKB_PAD
;
1725 tun
->filter_attached
= false;
1726 tun
->sndbuf
= tfile
->socket
.sk
->sk_sndbuf
;
1728 tun
->pcpu_stats
= netdev_alloc_pcpu_stats(struct tun_pcpu_stats
);
1729 if (!tun
->pcpu_stats
) {
1734 spin_lock_init(&tun
->lock
);
1736 err
= security_tun_dev_alloc_security(&tun
->security
);
1743 dev
->hw_features
= NETIF_F_SG
| NETIF_F_FRAGLIST
|
1744 TUN_USER_FEATURES
| NETIF_F_HW_VLAN_CTAG_TX
|
1745 NETIF_F_HW_VLAN_STAG_TX
;
1746 dev
->features
= dev
->hw_features
| NETIF_F_LLTX
;
1747 dev
->vlan_features
= dev
->features
&
1748 ~(NETIF_F_HW_VLAN_CTAG_TX
|
1749 NETIF_F_HW_VLAN_STAG_TX
);
1751 INIT_LIST_HEAD(&tun
->disabled
);
1752 err
= tun_attach(tun
, file
, false);
1756 err
= register_netdevice(tun
->dev
);
1761 netif_carrier_on(tun
->dev
);
1763 tun_debug(KERN_INFO
, tun
, "tun_set_iff\n");
1765 tun
->flags
= (tun
->flags
& ~TUN_FEATURES
) |
1766 (ifr
->ifr_flags
& TUN_FEATURES
);
1768 /* Make sure persistent devices do not get stuck in
1771 if (netif_running(tun
->dev
))
1772 netif_tx_wake_all_queues(tun
->dev
);
1774 strcpy(ifr
->ifr_name
, tun
->dev
->name
);
1778 tun_detach_all(dev
);
1780 tun_flow_uninit(tun
);
1781 security_tun_dev_free_security(tun
->security
);
1783 free_percpu(tun
->pcpu_stats
);
1789 static void tun_get_iff(struct net
*net
, struct tun_struct
*tun
,
1792 tun_debug(KERN_INFO
, tun
, "tun_get_iff\n");
1794 strcpy(ifr
->ifr_name
, tun
->dev
->name
);
1796 ifr
->ifr_flags
= tun_flags(tun
);
1800 /* This is like a cut-down ethtool ops, except done via tun fd so no
1801 * privs required. */
1802 static int set_offload(struct tun_struct
*tun
, unsigned long arg
)
1804 netdev_features_t features
= 0;
1806 if (arg
& TUN_F_CSUM
) {
1807 features
|= NETIF_F_HW_CSUM
;
1810 if (arg
& (TUN_F_TSO4
|TUN_F_TSO6
)) {
1811 if (arg
& TUN_F_TSO_ECN
) {
1812 features
|= NETIF_F_TSO_ECN
;
1813 arg
&= ~TUN_F_TSO_ECN
;
1815 if (arg
& TUN_F_TSO4
)
1816 features
|= NETIF_F_TSO
;
1817 if (arg
& TUN_F_TSO6
)
1818 features
|= NETIF_F_TSO6
;
1819 arg
&= ~(TUN_F_TSO4
|TUN_F_TSO6
);
1822 if (arg
& TUN_F_UFO
) {
1823 features
|= NETIF_F_UFO
;
1828 /* This gives the user a way to test for new features in future by
1829 * trying to set them. */
1833 tun
->set_features
= features
;
1834 netdev_update_features(tun
->dev
);
1839 static void tun_detach_filter(struct tun_struct
*tun
, int n
)
1842 struct tun_file
*tfile
;
1844 for (i
= 0; i
< n
; i
++) {
1845 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
1846 lock_sock(tfile
->socket
.sk
);
1847 sk_detach_filter(tfile
->socket
.sk
);
1848 release_sock(tfile
->socket
.sk
);
1851 tun
->filter_attached
= false;
1854 static int tun_attach_filter(struct tun_struct
*tun
)
1857 struct tun_file
*tfile
;
1859 for (i
= 0; i
< tun
->numqueues
; i
++) {
1860 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
1861 lock_sock(tfile
->socket
.sk
);
1862 ret
= sk_attach_filter(&tun
->fprog
, tfile
->socket
.sk
);
1863 release_sock(tfile
->socket
.sk
);
1865 tun_detach_filter(tun
, i
);
1870 tun
->filter_attached
= true;
1874 static void tun_set_sndbuf(struct tun_struct
*tun
)
1876 struct tun_file
*tfile
;
1879 for (i
= 0; i
< tun
->numqueues
; i
++) {
1880 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
1881 tfile
->socket
.sk
->sk_sndbuf
= tun
->sndbuf
;
1885 static int tun_set_queue(struct file
*file
, struct ifreq
*ifr
)
1887 struct tun_file
*tfile
= file
->private_data
;
1888 struct tun_struct
*tun
;
1893 if (ifr
->ifr_flags
& IFF_ATTACH_QUEUE
) {
1894 tun
= tfile
->detached
;
1899 ret
= security_tun_dev_attach_queue(tun
->security
);
1902 ret
= tun_attach(tun
, file
, false);
1903 } else if (ifr
->ifr_flags
& IFF_DETACH_QUEUE
) {
1904 tun
= rtnl_dereference(tfile
->tun
);
1905 if (!tun
|| !(tun
->flags
& IFF_MULTI_QUEUE
) || tfile
->detached
)
1908 __tun_detach(tfile
, false);
1917 static long __tun_chr_ioctl(struct file
*file
, unsigned int cmd
,
1918 unsigned long arg
, int ifreq_len
)
1920 struct tun_file
*tfile
= file
->private_data
;
1921 struct tun_struct
*tun
;
1922 void __user
* argp
= (void __user
*)arg
;
1928 unsigned int ifindex
;
1932 if (cmd
== TUNSETIFF
|| cmd
== TUNSETQUEUE
|| _IOC_TYPE(cmd
) == 0x89) {
1933 if (copy_from_user(&ifr
, argp
, ifreq_len
))
1936 memset(&ifr
, 0, sizeof(ifr
));
1938 if (cmd
== TUNGETFEATURES
) {
1939 /* Currently this just means: "what IFF flags are valid?".
1940 * This is needed because we never checked for invalid flags on
1943 return put_user(IFF_TUN
| IFF_TAP
| TUN_FEATURES
,
1944 (unsigned int __user
*)argp
);
1945 } else if (cmd
== TUNSETQUEUE
)
1946 return tun_set_queue(file
, &ifr
);
1951 tun
= __tun_get(tfile
);
1952 if (cmd
== TUNSETIFF
&& !tun
) {
1953 ifr
.ifr_name
[IFNAMSIZ
-1] = '\0';
1955 ret
= tun_set_iff(sock_net(&tfile
->sk
), file
, &ifr
);
1960 if (copy_to_user(argp
, &ifr
, ifreq_len
))
1964 if (cmd
== TUNSETIFINDEX
) {
1970 if (copy_from_user(&ifindex
, argp
, sizeof(ifindex
)))
1974 tfile
->ifindex
= ifindex
;
1982 tun_debug(KERN_INFO
, tun
, "tun_chr_ioctl cmd %u\n", cmd
);
1987 tun_get_iff(current
->nsproxy
->net_ns
, tun
, &ifr
);
1989 if (tfile
->detached
)
1990 ifr
.ifr_flags
|= IFF_DETACH_QUEUE
;
1991 if (!tfile
->socket
.sk
->sk_filter
)
1992 ifr
.ifr_flags
|= IFF_NOFILTER
;
1994 if (copy_to_user(argp
, &ifr
, ifreq_len
))
1999 /* Disable/Enable checksum */
2001 /* [unimplemented] */
2002 tun_debug(KERN_INFO
, tun
, "ignored: set checksum %s\n",
2003 arg
? "disabled" : "enabled");
2007 /* Disable/Enable persist mode. Keep an extra reference to the
2008 * module to prevent the module being unprobed.
2010 if (arg
&& !(tun
->flags
& IFF_PERSIST
)) {
2011 tun
->flags
|= IFF_PERSIST
;
2012 __module_get(THIS_MODULE
);
2014 if (!arg
&& (tun
->flags
& IFF_PERSIST
)) {
2015 tun
->flags
&= ~IFF_PERSIST
;
2016 module_put(THIS_MODULE
);
2019 tun_debug(KERN_INFO
, tun
, "persist %s\n",
2020 arg
? "enabled" : "disabled");
2024 /* Set owner of the device */
2025 owner
= make_kuid(current_user_ns(), arg
);
2026 if (!uid_valid(owner
)) {
2031 tun_debug(KERN_INFO
, tun
, "owner set to %u\n",
2032 from_kuid(&init_user_ns
, tun
->owner
));
2036 /* Set group of the device */
2037 group
= make_kgid(current_user_ns(), arg
);
2038 if (!gid_valid(group
)) {
2043 tun_debug(KERN_INFO
, tun
, "group set to %u\n",
2044 from_kgid(&init_user_ns
, tun
->group
));
2048 /* Only allow setting the type when the interface is down */
2049 if (tun
->dev
->flags
& IFF_UP
) {
2050 tun_debug(KERN_INFO
, tun
,
2051 "Linktype set failed because interface is up\n");
2054 tun
->dev
->type
= (int) arg
;
2055 tun_debug(KERN_INFO
, tun
, "linktype set to %d\n",
2067 ret
= set_offload(tun
, arg
);
2070 case TUNSETTXFILTER
:
2071 /* Can be set only for TAPs */
2073 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2075 ret
= update_filter(&tun
->txflt
, (void __user
*)arg
);
2079 /* Get hw address */
2080 memcpy(ifr
.ifr_hwaddr
.sa_data
, tun
->dev
->dev_addr
, ETH_ALEN
);
2081 ifr
.ifr_hwaddr
.sa_family
= tun
->dev
->type
;
2082 if (copy_to_user(argp
, &ifr
, ifreq_len
))
2087 /* Set hw address */
2088 tun_debug(KERN_DEBUG
, tun
, "set hw address: %pM\n",
2089 ifr
.ifr_hwaddr
.sa_data
);
2091 ret
= dev_set_mac_address(tun
->dev
, &ifr
.ifr_hwaddr
);
2095 sndbuf
= tfile
->socket
.sk
->sk_sndbuf
;
2096 if (copy_to_user(argp
, &sndbuf
, sizeof(sndbuf
)))
2101 if (copy_from_user(&sndbuf
, argp
, sizeof(sndbuf
))) {
2106 tun
->sndbuf
= sndbuf
;
2107 tun_set_sndbuf(tun
);
2110 case TUNGETVNETHDRSZ
:
2111 vnet_hdr_sz
= tun
->vnet_hdr_sz
;
2112 if (copy_to_user(argp
, &vnet_hdr_sz
, sizeof(vnet_hdr_sz
)))
2116 case TUNSETVNETHDRSZ
:
2117 if (copy_from_user(&vnet_hdr_sz
, argp
, sizeof(vnet_hdr_sz
))) {
2121 if (vnet_hdr_sz
< (int)sizeof(struct virtio_net_hdr
)) {
2126 tun
->vnet_hdr_sz
= vnet_hdr_sz
;
2130 le
= !!(tun
->flags
& TUN_VNET_LE
);
2131 if (put_user(le
, (int __user
*)argp
))
2136 if (get_user(le
, (int __user
*)argp
)) {
2141 tun
->flags
|= TUN_VNET_LE
;
2143 tun
->flags
&= ~TUN_VNET_LE
;
2147 ret
= tun_get_vnet_be(tun
, argp
);
2151 ret
= tun_set_vnet_be(tun
, argp
);
2154 case TUNATTACHFILTER
:
2155 /* Can be set only for TAPs */
2157 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2160 if (copy_from_user(&tun
->fprog
, argp
, sizeof(tun
->fprog
)))
2163 ret
= tun_attach_filter(tun
);
2166 case TUNDETACHFILTER
:
2167 /* Can be set only for TAPs */
2169 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2172 tun_detach_filter(tun
, tun
->numqueues
);
2177 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2180 if (copy_to_user(argp
, &tun
->fprog
, sizeof(tun
->fprog
)))
2197 static long tun_chr_ioctl(struct file
*file
,
2198 unsigned int cmd
, unsigned long arg
)
2200 return __tun_chr_ioctl(file
, cmd
, arg
, sizeof (struct ifreq
));
2203 #ifdef CONFIG_COMPAT
2204 static long tun_chr_compat_ioctl(struct file
*file
,
2205 unsigned int cmd
, unsigned long arg
)
2210 case TUNSETTXFILTER
:
2215 arg
= (unsigned long)compat_ptr(arg
);
2218 arg
= (compat_ulong_t
)arg
;
2223 * compat_ifreq is shorter than ifreq, so we must not access beyond
2224 * the end of that structure. All fields that are used in this
2225 * driver are compatible though, we don't need to convert the
2228 return __tun_chr_ioctl(file
, cmd
, arg
, sizeof(struct compat_ifreq
));
2230 #endif /* CONFIG_COMPAT */
2232 static int tun_chr_fasync(int fd
, struct file
*file
, int on
)
2234 struct tun_file
*tfile
= file
->private_data
;
2237 if ((ret
= fasync_helper(fd
, file
, on
, &tfile
->fasync
)) < 0)
2241 __f_setown(file
, task_pid(current
), PIDTYPE_PID
, 0);
2242 tfile
->flags
|= TUN_FASYNC
;
2244 tfile
->flags
&= ~TUN_FASYNC
;
2250 static int tun_chr_open(struct inode
*inode
, struct file
* file
)
2252 struct net
*net
= current
->nsproxy
->net_ns
;
2253 struct tun_file
*tfile
;
2255 DBG1(KERN_INFO
, "tunX: tun_chr_open\n");
2257 tfile
= (struct tun_file
*)sk_alloc(net
, AF_UNSPEC
, GFP_KERNEL
,
2261 RCU_INIT_POINTER(tfile
->tun
, NULL
);
2265 init_waitqueue_head(&tfile
->wq
.wait
);
2266 RCU_INIT_POINTER(tfile
->socket
.wq
, &tfile
->wq
);
2268 tfile
->socket
.file
= file
;
2269 tfile
->socket
.ops
= &tun_socket_ops
;
2271 sock_init_data(&tfile
->socket
, &tfile
->sk
);
2273 tfile
->sk
.sk_write_space
= tun_sock_write_space
;
2274 tfile
->sk
.sk_sndbuf
= INT_MAX
;
2276 file
->private_data
= tfile
;
2277 INIT_LIST_HEAD(&tfile
->next
);
2279 sock_set_flag(&tfile
->sk
, SOCK_ZEROCOPY
);
2284 static int tun_chr_close(struct inode
*inode
, struct file
*file
)
2286 struct tun_file
*tfile
= file
->private_data
;
2288 tun_detach(tfile
, true);
2293 #ifdef CONFIG_PROC_FS
2294 static void tun_chr_show_fdinfo(struct seq_file
*m
, struct file
*f
)
2296 struct tun_struct
*tun
;
2299 memset(&ifr
, 0, sizeof(ifr
));
2304 tun_get_iff(current
->nsproxy
->net_ns
, tun
, &ifr
);
2310 seq_printf(m
, "iff:\t%s\n", ifr
.ifr_name
);
2314 static const struct file_operations tun_fops
= {
2315 .owner
= THIS_MODULE
,
2316 .llseek
= no_llseek
,
2317 .read_iter
= tun_chr_read_iter
,
2318 .write_iter
= tun_chr_write_iter
,
2319 .poll
= tun_chr_poll
,
2320 .unlocked_ioctl
= tun_chr_ioctl
,
2321 #ifdef CONFIG_COMPAT
2322 .compat_ioctl
= tun_chr_compat_ioctl
,
2324 .open
= tun_chr_open
,
2325 .release
= tun_chr_close
,
2326 .fasync
= tun_chr_fasync
,
2327 #ifdef CONFIG_PROC_FS
2328 .show_fdinfo
= tun_chr_show_fdinfo
,
2332 static struct miscdevice tun_miscdev
= {
2335 .nodename
= "net/tun",
2339 /* ethtool interface */
2341 static int tun_get_settings(struct net_device
*dev
, struct ethtool_cmd
*cmd
)
2344 cmd
->advertising
= 0;
2345 ethtool_cmd_speed_set(cmd
, SPEED_10
);
2346 cmd
->duplex
= DUPLEX_FULL
;
2347 cmd
->port
= PORT_TP
;
2348 cmd
->phy_address
= 0;
2349 cmd
->transceiver
= XCVR_INTERNAL
;
2350 cmd
->autoneg
= AUTONEG_DISABLE
;
2356 static void tun_get_drvinfo(struct net_device
*dev
, struct ethtool_drvinfo
*info
)
2358 struct tun_struct
*tun
= netdev_priv(dev
);
2360 strlcpy(info
->driver
, DRV_NAME
, sizeof(info
->driver
));
2361 strlcpy(info
->version
, DRV_VERSION
, sizeof(info
->version
));
2363 switch (tun
->flags
& TUN_TYPE_MASK
) {
2365 strlcpy(info
->bus_info
, "tun", sizeof(info
->bus_info
));
2368 strlcpy(info
->bus_info
, "tap", sizeof(info
->bus_info
));
2373 static u32
tun_get_msglevel(struct net_device
*dev
)
2376 struct tun_struct
*tun
= netdev_priv(dev
);
2383 static void tun_set_msglevel(struct net_device
*dev
, u32 value
)
2386 struct tun_struct
*tun
= netdev_priv(dev
);
2391 static const struct ethtool_ops tun_ethtool_ops
= {
2392 .get_settings
= tun_get_settings
,
2393 .get_drvinfo
= tun_get_drvinfo
,
2394 .get_msglevel
= tun_get_msglevel
,
2395 .set_msglevel
= tun_set_msglevel
,
2396 .get_link
= ethtool_op_get_link
,
2397 .get_ts_info
= ethtool_op_get_ts_info
,
2401 static int __init
tun_init(void)
2405 pr_info("%s, %s\n", DRV_DESCRIPTION
, DRV_VERSION
);
2406 pr_info("%s\n", DRV_COPYRIGHT
);
2408 ret
= rtnl_link_register(&tun_link_ops
);
2410 pr_err("Can't register link_ops\n");
2414 ret
= misc_register(&tun_miscdev
);
2416 pr_err("Can't register misc device %d\n", TUN_MINOR
);
2421 rtnl_link_unregister(&tun_link_ops
);
2426 static void tun_cleanup(void)
2428 misc_deregister(&tun_miscdev
);
2429 rtnl_link_unregister(&tun_link_ops
);
2432 /* Get an underlying socket object from tun file. Returns error unless file is
2433 * attached to a device. The returned object works like a packet socket, it
2434 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
2435 * holding a reference to the file for as long as the socket is in use. */
2436 struct socket
*tun_get_socket(struct file
*file
)
2438 struct tun_file
*tfile
;
2439 if (file
->f_op
!= &tun_fops
)
2440 return ERR_PTR(-EINVAL
);
2441 tfile
= file
->private_data
;
2443 return ERR_PTR(-EBADFD
);
2444 return &tfile
->socket
;
2446 EXPORT_SYMBOL_GPL(tun_get_socket
);
2448 module_init(tun_init
);
2449 module_exit(tun_cleanup
);
2450 MODULE_DESCRIPTION(DRV_DESCRIPTION
);
2451 MODULE_AUTHOR(DRV_COPYRIGHT
);
2452 MODULE_LICENSE("GPL");
2453 MODULE_ALIAS_MISCDEV(TUN_MINOR
);
2454 MODULE_ALIAS("devname:net/tun");