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>
69 #include <net/net_namespace.h>
70 #include <net/netns/generic.h>
71 #include <net/rtnetlink.h>
73 #include <linux/seq_file.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 #define GOODCOPY_LEN 128
108 #define FLT_EXACT_COUNT 8
110 unsigned int count
; /* Number of addrs. Zero means disabled */
111 u32 mask
[2]; /* Mask of the hashed addrs */
112 unsigned char addr
[FLT_EXACT_COUNT
][ETH_ALEN
];
115 /* DEFAULT_MAX_NUM_RSS_QUEUES were chosen to let the rx/tx queues allocated for
116 * the netdevice to be fit in one page. So we can make sure the success of
117 * memory allocation. TODO: increase the limit. */
118 #define MAX_TAP_QUEUES DEFAULT_MAX_NUM_RSS_QUEUES
119 #define MAX_TAP_FLOWS 4096
121 #define TUN_FLOW_EXPIRE (3 * HZ)
123 /* A tun_file connects an open character device to a tuntap netdevice. It
124 * also contains all socket related structures (except sock_fprog and tap_filter)
125 * to serve as one transmit queue for tuntap device. The sock_fprog and
126 * tap_filter were kept in tun_struct since they were used for filtering for the
127 * netdevice not for a specific queue (at least I didn't see the requirement for
131 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
132 * other can only be read while rcu_read_lock or rtnl_lock is held.
136 struct socket socket
;
138 struct tun_struct __rcu
*tun
;
140 struct fasync_struct
*fasync
;
141 /* only used for fasnyc */
145 unsigned int ifindex
;
147 struct list_head next
;
148 struct tun_struct
*detached
;
151 struct tun_flow_entry
{
152 struct hlist_node hash_link
;
154 struct tun_struct
*tun
;
159 unsigned long updated
;
162 #define TUN_NUM_FLOW_ENTRIES 1024
164 /* Since the socket were moved to tun_file, to preserve the behavior of persist
165 * device, socket filter, sndbuf and vnet header size were restore when the
166 * file were attached to a persist device.
169 struct tun_file __rcu
*tfiles
[MAX_TAP_QUEUES
];
170 unsigned int numqueues
;
175 struct net_device
*dev
;
176 netdev_features_t set_features
;
177 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
182 struct tap_filter txflt
;
183 struct sock_fprog fprog
;
184 /* protected by rtnl lock */
185 bool filter_attached
;
190 struct hlist_head flows
[TUN_NUM_FLOW_ENTRIES
];
191 struct timer_list flow_gc_timer
;
192 unsigned long ageing_time
;
193 unsigned int numdisabled
;
194 struct list_head disabled
;
199 static inline u32
tun_hashfn(u32 rxhash
)
201 return rxhash
& 0x3ff;
204 static struct tun_flow_entry
*tun_flow_find(struct hlist_head
*head
, u32 rxhash
)
206 struct tun_flow_entry
*e
;
208 hlist_for_each_entry_rcu(e
, head
, hash_link
) {
209 if (e
->rxhash
== rxhash
)
215 static struct tun_flow_entry
*tun_flow_create(struct tun_struct
*tun
,
216 struct hlist_head
*head
,
217 u32 rxhash
, u16 queue_index
)
219 struct tun_flow_entry
*e
= kmalloc(sizeof(*e
), GFP_ATOMIC
);
222 tun_debug(KERN_INFO
, tun
, "create flow: hash %u index %u\n",
223 rxhash
, queue_index
);
224 e
->updated
= jiffies
;
227 e
->queue_index
= queue_index
;
229 hlist_add_head_rcu(&e
->hash_link
, head
);
235 static void tun_flow_delete(struct tun_struct
*tun
, struct tun_flow_entry
*e
)
237 tun_debug(KERN_INFO
, tun
, "delete flow: hash %u index %u\n",
238 e
->rxhash
, e
->queue_index
);
239 sock_rps_reset_flow_hash(e
->rps_rxhash
);
240 hlist_del_rcu(&e
->hash_link
);
245 static void tun_flow_flush(struct tun_struct
*tun
)
249 spin_lock_bh(&tun
->lock
);
250 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
251 struct tun_flow_entry
*e
;
252 struct hlist_node
*n
;
254 hlist_for_each_entry_safe(e
, n
, &tun
->flows
[i
], hash_link
)
255 tun_flow_delete(tun
, e
);
257 spin_unlock_bh(&tun
->lock
);
260 static void tun_flow_delete_by_queue(struct tun_struct
*tun
, u16 queue_index
)
264 spin_lock_bh(&tun
->lock
);
265 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
266 struct tun_flow_entry
*e
;
267 struct hlist_node
*n
;
269 hlist_for_each_entry_safe(e
, n
, &tun
->flows
[i
], hash_link
) {
270 if (e
->queue_index
== queue_index
)
271 tun_flow_delete(tun
, e
);
274 spin_unlock_bh(&tun
->lock
);
277 static void tun_flow_cleanup(unsigned long data
)
279 struct tun_struct
*tun
= (struct tun_struct
*)data
;
280 unsigned long delay
= tun
->ageing_time
;
281 unsigned long next_timer
= jiffies
+ delay
;
282 unsigned long count
= 0;
285 tun_debug(KERN_INFO
, tun
, "tun_flow_cleanup\n");
287 spin_lock_bh(&tun
->lock
);
288 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
289 struct tun_flow_entry
*e
;
290 struct hlist_node
*n
;
292 hlist_for_each_entry_safe(e
, n
, &tun
->flows
[i
], hash_link
) {
293 unsigned long this_timer
;
295 this_timer
= e
->updated
+ delay
;
296 if (time_before_eq(this_timer
, jiffies
))
297 tun_flow_delete(tun
, e
);
298 else if (time_before(this_timer
, next_timer
))
299 next_timer
= this_timer
;
304 mod_timer(&tun
->flow_gc_timer
, round_jiffies_up(next_timer
));
305 spin_unlock_bh(&tun
->lock
);
308 static void tun_flow_update(struct tun_struct
*tun
, u32 rxhash
,
309 struct tun_file
*tfile
)
311 struct hlist_head
*head
;
312 struct tun_flow_entry
*e
;
313 unsigned long delay
= tun
->ageing_time
;
314 u16 queue_index
= tfile
->queue_index
;
319 head
= &tun
->flows
[tun_hashfn(rxhash
)];
323 /* We may get a very small possibility of OOO during switching, not
324 * worth to optimize.*/
325 if (tun
->numqueues
== 1 || tfile
->detached
)
328 e
= tun_flow_find(head
, rxhash
);
330 /* TODO: keep queueing to old queue until it's empty? */
331 e
->queue_index
= queue_index
;
332 e
->updated
= jiffies
;
333 sock_rps_record_flow_hash(e
->rps_rxhash
);
335 spin_lock_bh(&tun
->lock
);
336 if (!tun_flow_find(head
, rxhash
) &&
337 tun
->flow_count
< MAX_TAP_FLOWS
)
338 tun_flow_create(tun
, head
, rxhash
, queue_index
);
340 if (!timer_pending(&tun
->flow_gc_timer
))
341 mod_timer(&tun
->flow_gc_timer
,
342 round_jiffies_up(jiffies
+ delay
));
343 spin_unlock_bh(&tun
->lock
);
351 * Save the hash received in the stack receive path and update the
352 * flow_hash table accordingly.
354 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry
*e
, u32 hash
)
356 if (unlikely(e
->rps_rxhash
!= hash
)) {
357 sock_rps_reset_flow_hash(e
->rps_rxhash
);
358 e
->rps_rxhash
= hash
;
362 /* We try to identify a flow through its rxhash first. The reason that
363 * we do not check rxq no. is because some cards(e.g 82599), chooses
364 * the rxq based on the txq where the last packet of the flow comes. As
365 * the userspace application move between processors, we may get a
366 * different rxq no. here. If we could not get rxhash, then we would
367 * hope the rxq no. may help here.
369 static u16
tun_select_queue(struct net_device
*dev
, struct sk_buff
*skb
,
370 void *accel_priv
, select_queue_fallback_t fallback
)
372 struct tun_struct
*tun
= netdev_priv(dev
);
373 struct tun_flow_entry
*e
;
378 numqueues
= ACCESS_ONCE(tun
->numqueues
);
380 txq
= skb_get_hash(skb
);
382 e
= tun_flow_find(&tun
->flows
[tun_hashfn(txq
)], txq
);
384 tun_flow_save_rps_rxhash(e
, txq
);
385 txq
= e
->queue_index
;
387 /* use multiply and shift instead of expensive divide */
388 txq
= ((u64
)txq
* numqueues
) >> 32;
389 } else if (likely(skb_rx_queue_recorded(skb
))) {
390 txq
= skb_get_rx_queue(skb
);
391 while (unlikely(txq
>= numqueues
))
399 static inline bool tun_not_capable(struct tun_struct
*tun
)
401 const struct cred
*cred
= current_cred();
402 struct net
*net
= dev_net(tun
->dev
);
404 return ((uid_valid(tun
->owner
) && !uid_eq(cred
->euid
, tun
->owner
)) ||
405 (gid_valid(tun
->group
) && !in_egroup_p(tun
->group
))) &&
406 !ns_capable(net
->user_ns
, CAP_NET_ADMIN
);
409 static void tun_set_real_num_queues(struct tun_struct
*tun
)
411 netif_set_real_num_tx_queues(tun
->dev
, tun
->numqueues
);
412 netif_set_real_num_rx_queues(tun
->dev
, tun
->numqueues
);
415 static void tun_disable_queue(struct tun_struct
*tun
, struct tun_file
*tfile
)
417 tfile
->detached
= tun
;
418 list_add_tail(&tfile
->next
, &tun
->disabled
);
422 static struct tun_struct
*tun_enable_queue(struct tun_file
*tfile
)
424 struct tun_struct
*tun
= tfile
->detached
;
426 tfile
->detached
= NULL
;
427 list_del_init(&tfile
->next
);
432 static void tun_queue_purge(struct tun_file
*tfile
)
434 skb_queue_purge(&tfile
->sk
.sk_receive_queue
);
435 skb_queue_purge(&tfile
->sk
.sk_error_queue
);
438 static void __tun_detach(struct tun_file
*tfile
, bool clean
)
440 struct tun_file
*ntfile
;
441 struct tun_struct
*tun
;
443 tun
= rtnl_dereference(tfile
->tun
);
445 if (tun
&& !tfile
->detached
) {
446 u16 index
= tfile
->queue_index
;
447 BUG_ON(index
>= tun
->numqueues
);
449 rcu_assign_pointer(tun
->tfiles
[index
],
450 tun
->tfiles
[tun
->numqueues
- 1]);
451 ntfile
= rtnl_dereference(tun
->tfiles
[index
]);
452 ntfile
->queue_index
= index
;
456 RCU_INIT_POINTER(tfile
->tun
, NULL
);
457 sock_put(&tfile
->sk
);
459 tun_disable_queue(tun
, tfile
);
462 tun_flow_delete_by_queue(tun
, tun
->numqueues
+ 1);
463 /* Drop read queue */
464 tun_queue_purge(tfile
);
465 tun_set_real_num_queues(tun
);
466 } else if (tfile
->detached
&& clean
) {
467 tun
= tun_enable_queue(tfile
);
468 sock_put(&tfile
->sk
);
472 if (tun
&& tun
->numqueues
== 0 && tun
->numdisabled
== 0) {
473 netif_carrier_off(tun
->dev
);
475 if (!(tun
->flags
& TUN_PERSIST
) &&
476 tun
->dev
->reg_state
== NETREG_REGISTERED
)
477 unregister_netdevice(tun
->dev
);
480 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED
,
481 &tfile
->socket
.flags
));
482 sk_release_kernel(&tfile
->sk
);
486 static void tun_detach(struct tun_file
*tfile
, bool clean
)
489 __tun_detach(tfile
, clean
);
493 static void tun_detach_all(struct net_device
*dev
)
495 struct tun_struct
*tun
= netdev_priv(dev
);
496 struct tun_file
*tfile
, *tmp
;
497 int i
, n
= tun
->numqueues
;
499 for (i
= 0; i
< n
; i
++) {
500 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
502 tfile
->socket
.sk
->sk_data_ready(tfile
->socket
.sk
);
503 RCU_INIT_POINTER(tfile
->tun
, NULL
);
506 list_for_each_entry(tfile
, &tun
->disabled
, next
) {
507 tfile
->socket
.sk
->sk_data_ready(tfile
->socket
.sk
);
508 RCU_INIT_POINTER(tfile
->tun
, NULL
);
510 BUG_ON(tun
->numqueues
!= 0);
513 for (i
= 0; i
< n
; i
++) {
514 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
515 /* Drop read queue */
516 tun_queue_purge(tfile
);
517 sock_put(&tfile
->sk
);
519 list_for_each_entry_safe(tfile
, tmp
, &tun
->disabled
, next
) {
520 tun_enable_queue(tfile
);
521 tun_queue_purge(tfile
);
522 sock_put(&tfile
->sk
);
524 BUG_ON(tun
->numdisabled
!= 0);
526 if (tun
->flags
& TUN_PERSIST
)
527 module_put(THIS_MODULE
);
530 static int tun_attach(struct tun_struct
*tun
, struct file
*file
, bool skip_filter
)
532 struct tun_file
*tfile
= file
->private_data
;
535 err
= security_tun_dev_attach(tfile
->socket
.sk
, tun
->security
);
540 if (rtnl_dereference(tfile
->tun
) && !tfile
->detached
)
544 if (!(tun
->flags
& TUN_TAP_MQ
) && tun
->numqueues
== 1)
548 if (!tfile
->detached
&&
549 tun
->numqueues
+ tun
->numdisabled
== MAX_TAP_QUEUES
)
554 /* Re-attach the filter to persist device */
555 if (!skip_filter
&& (tun
->filter_attached
== true)) {
556 err
= sk_attach_filter(&tun
->fprog
, tfile
->socket
.sk
);
560 tfile
->queue_index
= tun
->numqueues
;
561 rcu_assign_pointer(tfile
->tun
, tun
);
562 rcu_assign_pointer(tun
->tfiles
[tun
->numqueues
], tfile
);
566 tun_enable_queue(tfile
);
568 sock_hold(&tfile
->sk
);
570 tun_set_real_num_queues(tun
);
572 /* device is allowed to go away first, so no need to hold extra
580 static struct tun_struct
*__tun_get(struct tun_file
*tfile
)
582 struct tun_struct
*tun
;
585 tun
= rcu_dereference(tfile
->tun
);
593 static struct tun_struct
*tun_get(struct file
*file
)
595 return __tun_get(file
->private_data
);
598 static void tun_put(struct tun_struct
*tun
)
604 static void addr_hash_set(u32
*mask
, const u8
*addr
)
606 int n
= ether_crc(ETH_ALEN
, addr
) >> 26;
607 mask
[n
>> 5] |= (1 << (n
& 31));
610 static unsigned int addr_hash_test(const u32
*mask
, const u8
*addr
)
612 int n
= ether_crc(ETH_ALEN
, addr
) >> 26;
613 return mask
[n
>> 5] & (1 << (n
& 31));
616 static int update_filter(struct tap_filter
*filter
, void __user
*arg
)
618 struct { u8 u
[ETH_ALEN
]; } *addr
;
619 struct tun_filter uf
;
620 int err
, alen
, n
, nexact
;
622 if (copy_from_user(&uf
, arg
, sizeof(uf
)))
631 alen
= ETH_ALEN
* uf
.count
;
632 addr
= kmalloc(alen
, GFP_KERNEL
);
636 if (copy_from_user(addr
, arg
+ sizeof(uf
), alen
)) {
641 /* The filter is updated without holding any locks. Which is
642 * perfectly safe. We disable it first and in the worst
643 * case we'll accept a few undesired packets. */
647 /* Use first set of addresses as an exact filter */
648 for (n
= 0; n
< uf
.count
&& n
< FLT_EXACT_COUNT
; n
++)
649 memcpy(filter
->addr
[n
], addr
[n
].u
, ETH_ALEN
);
653 /* Remaining multicast addresses are hashed,
654 * unicast will leave the filter disabled. */
655 memset(filter
->mask
, 0, sizeof(filter
->mask
));
656 for (; n
< uf
.count
; n
++) {
657 if (!is_multicast_ether_addr(addr
[n
].u
)) {
658 err
= 0; /* no filter */
661 addr_hash_set(filter
->mask
, addr
[n
].u
);
664 /* For ALLMULTI just set the mask to all ones.
665 * This overrides the mask populated above. */
666 if ((uf
.flags
& TUN_FLT_ALLMULTI
))
667 memset(filter
->mask
, ~0, sizeof(filter
->mask
));
669 /* Now enable the filter */
671 filter
->count
= nexact
;
673 /* Return the number of exact filters */
681 /* Returns: 0 - drop, !=0 - accept */
682 static int run_filter(struct tap_filter
*filter
, const struct sk_buff
*skb
)
684 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
686 struct ethhdr
*eh
= (struct ethhdr
*) skb
->data
;
690 for (i
= 0; i
< filter
->count
; i
++)
691 if (ether_addr_equal(eh
->h_dest
, filter
->addr
[i
]))
694 /* Inexact match (multicast only) */
695 if (is_multicast_ether_addr(eh
->h_dest
))
696 return addr_hash_test(filter
->mask
, eh
->h_dest
);
702 * Checks whether the packet is accepted or not.
703 * Returns: 0 - drop, !=0 - accept
705 static int check_filter(struct tap_filter
*filter
, const struct sk_buff
*skb
)
710 return run_filter(filter
, skb
);
713 /* Network device part of the driver */
715 static const struct ethtool_ops tun_ethtool_ops
;
717 /* Net device detach from fd. */
718 static void tun_net_uninit(struct net_device
*dev
)
723 /* Net device open. */
724 static int tun_net_open(struct net_device
*dev
)
726 netif_tx_start_all_queues(dev
);
730 /* Net device close. */
731 static int tun_net_close(struct net_device
*dev
)
733 netif_tx_stop_all_queues(dev
);
737 /* Net device start xmit */
738 static netdev_tx_t
tun_net_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
740 struct tun_struct
*tun
= netdev_priv(dev
);
741 int txq
= skb
->queue_mapping
;
742 struct tun_file
*tfile
;
746 tfile
= rcu_dereference(tun
->tfiles
[txq
]);
747 numqueues
= ACCESS_ONCE(tun
->numqueues
);
749 /* Drop packet if interface is not attached */
750 if (txq
>= numqueues
)
753 if (numqueues
== 1) {
754 /* Select queue was not called for the skbuff, so we extract the
755 * RPS hash and save it into the flow_table here.
759 rxhash
= skb_get_hash(skb
);
761 struct tun_flow_entry
*e
;
762 e
= tun_flow_find(&tun
->flows
[tun_hashfn(rxhash
)],
765 tun_flow_save_rps_rxhash(e
, rxhash
);
769 tun_debug(KERN_INFO
, tun
, "tun_net_xmit %d\n", skb
->len
);
773 /* Drop if the filter does not like it.
774 * This is a noop if the filter is disabled.
775 * Filter can be enabled only for the TAP devices. */
776 if (!check_filter(&tun
->txflt
, skb
))
779 if (tfile
->socket
.sk
->sk_filter
&&
780 sk_filter(tfile
->socket
.sk
, skb
))
783 /* Limit the number of packets queued by dividing txq length with the
786 if (skb_queue_len(&tfile
->socket
.sk
->sk_receive_queue
) * numqueues
787 >= dev
->tx_queue_len
)
790 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
794 sock_tx_timestamp(skb
->sk
, &skb_shinfo(skb
)->tx_flags
);
795 sw_tx_timestamp(skb
);
798 /* Orphan the skb - required as we might hang on to it
799 * for indefinite time.
806 skb_queue_tail(&tfile
->socket
.sk
->sk_receive_queue
, skb
);
808 /* Notify and wake up reader process */
809 if (tfile
->flags
& TUN_FASYNC
)
810 kill_fasync(&tfile
->fasync
, SIGIO
, POLL_IN
);
811 tfile
->socket
.sk
->sk_data_ready(tfile
->socket
.sk
);
817 dev
->stats
.tx_dropped
++;
824 static void tun_net_mclist(struct net_device
*dev
)
827 * This callback is supposed to deal with mc filter in
828 * _rx_ path and has nothing to do with the _tx_ path.
829 * In rx path we always accept everything userspace gives us.
834 #define MAX_MTU 65535
837 tun_net_change_mtu(struct net_device
*dev
, int new_mtu
)
839 if (new_mtu
< MIN_MTU
|| new_mtu
+ dev
->hard_header_len
> MAX_MTU
)
845 static netdev_features_t
tun_net_fix_features(struct net_device
*dev
,
846 netdev_features_t features
)
848 struct tun_struct
*tun
= netdev_priv(dev
);
850 return (features
& tun
->set_features
) | (features
& ~TUN_USER_FEATURES
);
852 #ifdef CONFIG_NET_POLL_CONTROLLER
853 static void tun_poll_controller(struct net_device
*dev
)
856 * Tun only receives frames when:
857 * 1) the char device endpoint gets data from user space
858 * 2) the tun socket gets a sendmsg call from user space
859 * Since both of those are synchronous operations, we are guaranteed
860 * never to have pending data when we poll for it
861 * so there is nothing to do here but return.
862 * We need this though so netpoll recognizes us as an interface that
863 * supports polling, which enables bridge devices in virt setups to
864 * still use netconsole
869 static const struct net_device_ops tun_netdev_ops
= {
870 .ndo_uninit
= tun_net_uninit
,
871 .ndo_open
= tun_net_open
,
872 .ndo_stop
= tun_net_close
,
873 .ndo_start_xmit
= tun_net_xmit
,
874 .ndo_change_mtu
= tun_net_change_mtu
,
875 .ndo_fix_features
= tun_net_fix_features
,
876 .ndo_select_queue
= tun_select_queue
,
877 #ifdef CONFIG_NET_POLL_CONTROLLER
878 .ndo_poll_controller
= tun_poll_controller
,
882 static const struct net_device_ops tap_netdev_ops
= {
883 .ndo_uninit
= tun_net_uninit
,
884 .ndo_open
= tun_net_open
,
885 .ndo_stop
= tun_net_close
,
886 .ndo_start_xmit
= tun_net_xmit
,
887 .ndo_change_mtu
= tun_net_change_mtu
,
888 .ndo_fix_features
= tun_net_fix_features
,
889 .ndo_set_rx_mode
= tun_net_mclist
,
890 .ndo_set_mac_address
= eth_mac_addr
,
891 .ndo_validate_addr
= eth_validate_addr
,
892 .ndo_select_queue
= tun_select_queue
,
893 #ifdef CONFIG_NET_POLL_CONTROLLER
894 .ndo_poll_controller
= tun_poll_controller
,
898 static void tun_flow_init(struct tun_struct
*tun
)
902 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++)
903 INIT_HLIST_HEAD(&tun
->flows
[i
]);
905 tun
->ageing_time
= TUN_FLOW_EXPIRE
;
906 setup_timer(&tun
->flow_gc_timer
, tun_flow_cleanup
, (unsigned long)tun
);
907 mod_timer(&tun
->flow_gc_timer
,
908 round_jiffies_up(jiffies
+ tun
->ageing_time
));
911 static void tun_flow_uninit(struct tun_struct
*tun
)
913 del_timer_sync(&tun
->flow_gc_timer
);
917 /* Initialize net device. */
918 static void tun_net_init(struct net_device
*dev
)
920 struct tun_struct
*tun
= netdev_priv(dev
);
922 switch (tun
->flags
& TUN_TYPE_MASK
) {
924 dev
->netdev_ops
= &tun_netdev_ops
;
926 /* Point-to-Point TUN Device */
927 dev
->hard_header_len
= 0;
931 /* Zero header length */
932 dev
->type
= ARPHRD_NONE
;
933 dev
->flags
= IFF_POINTOPOINT
| IFF_NOARP
| IFF_MULTICAST
;
934 dev
->tx_queue_len
= TUN_READQ_SIZE
; /* We prefer our own queue length */
938 dev
->netdev_ops
= &tap_netdev_ops
;
939 /* Ethernet TAP Device */
941 dev
->priv_flags
&= ~IFF_TX_SKB_SHARING
;
942 dev
->priv_flags
|= IFF_LIVE_ADDR_CHANGE
;
944 eth_hw_addr_random(dev
);
946 dev
->tx_queue_len
= TUN_READQ_SIZE
; /* We prefer our own queue length */
951 /* Character device part */
954 static unsigned int tun_chr_poll(struct file
*file
, poll_table
*wait
)
956 struct tun_file
*tfile
= file
->private_data
;
957 struct tun_struct
*tun
= __tun_get(tfile
);
959 unsigned int mask
= 0;
964 sk
= tfile
->socket
.sk
;
966 tun_debug(KERN_INFO
, tun
, "tun_chr_poll\n");
968 poll_wait(file
, sk_sleep(sk
), wait
);
970 if (!skb_queue_empty(&sk
->sk_receive_queue
))
971 mask
|= POLLIN
| POLLRDNORM
;
973 if (sock_writeable(sk
) ||
974 (!test_and_set_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
) &&
976 mask
|= POLLOUT
| POLLWRNORM
;
978 if (tun
->dev
->reg_state
!= NETREG_REGISTERED
)
985 /* prepad is the amount to reserve at front. len is length after that.
986 * linear is a hint as to how much to copy (usually headers). */
987 static struct sk_buff
*tun_alloc_skb(struct tun_file
*tfile
,
988 size_t prepad
, size_t len
,
989 size_t linear
, int noblock
)
991 struct sock
*sk
= tfile
->socket
.sk
;
995 /* Under a page? Don't bother with paged skb. */
996 if (prepad
+ len
< PAGE_SIZE
|| !linear
)
999 skb
= sock_alloc_send_pskb(sk
, prepad
+ linear
, len
- linear
, noblock
,
1002 return ERR_PTR(err
);
1004 skb_reserve(skb
, prepad
);
1005 skb_put(skb
, linear
);
1006 skb
->data_len
= len
- linear
;
1007 skb
->len
+= len
- linear
;
1012 /* Get packet from user space buffer */
1013 static ssize_t
tun_get_user(struct tun_struct
*tun
, struct tun_file
*tfile
,
1014 void *msg_control
, const struct iovec
*iv
,
1015 size_t total_len
, size_t count
, int noblock
)
1017 struct tun_pi pi
= { 0, cpu_to_be16(ETH_P_IP
) };
1018 struct sk_buff
*skb
;
1019 size_t len
= total_len
, align
= NET_SKB_PAD
, linear
;
1020 struct virtio_net_hdr gso
= { 0 };
1024 bool zerocopy
= false;
1028 if (!(tun
->flags
& TUN_NO_PI
)) {
1029 if (len
< sizeof(pi
))
1033 if (memcpy_fromiovecend((void *)&pi
, iv
, 0, sizeof(pi
)))
1035 offset
+= sizeof(pi
);
1038 if (tun
->flags
& TUN_VNET_HDR
) {
1039 if (len
< tun
->vnet_hdr_sz
)
1041 len
-= tun
->vnet_hdr_sz
;
1043 if (memcpy_fromiovecend((void *)&gso
, iv
, offset
, sizeof(gso
)))
1046 if ((gso
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) &&
1047 gso
.csum_start
+ gso
.csum_offset
+ 2 > gso
.hdr_len
)
1048 gso
.hdr_len
= gso
.csum_start
+ gso
.csum_offset
+ 2;
1050 if (gso
.hdr_len
> len
)
1052 offset
+= tun
->vnet_hdr_sz
;
1055 if ((tun
->flags
& TUN_TYPE_MASK
) == TUN_TAP_DEV
) {
1056 align
+= NET_IP_ALIGN
;
1057 if (unlikely(len
< ETH_HLEN
||
1058 (gso
.hdr_len
&& gso
.hdr_len
< ETH_HLEN
)))
1062 good_linear
= SKB_MAX_HEAD(align
);
1065 /* There are 256 bytes to be copied in skb, so there is
1066 * enough room for skb expand head in case it is used.
1067 * The rest of the buffer is mapped from userspace.
1069 copylen
= gso
.hdr_len
? gso
.hdr_len
: GOODCOPY_LEN
;
1070 if (copylen
> good_linear
)
1071 copylen
= good_linear
;
1073 if (iov_pages(iv
, offset
+ copylen
, count
) <= MAX_SKB_FRAGS
)
1079 if (gso
.hdr_len
> good_linear
)
1080 linear
= good_linear
;
1082 linear
= gso
.hdr_len
;
1085 skb
= tun_alloc_skb(tfile
, align
, copylen
, linear
, noblock
);
1087 if (PTR_ERR(skb
) != -EAGAIN
)
1088 tun
->dev
->stats
.rx_dropped
++;
1089 return PTR_ERR(skb
);
1093 err
= zerocopy_sg_from_iovec(skb
, iv
, offset
, count
);
1095 err
= skb_copy_datagram_from_iovec(skb
, 0, iv
, offset
, len
);
1096 if (!err
&& msg_control
) {
1097 struct ubuf_info
*uarg
= msg_control
;
1098 uarg
->callback(uarg
, false);
1103 tun
->dev
->stats
.rx_dropped
++;
1108 if (gso
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) {
1109 if (!skb_partial_csum_set(skb
, gso
.csum_start
,
1111 tun
->dev
->stats
.rx_frame_errors
++;
1117 switch (tun
->flags
& TUN_TYPE_MASK
) {
1119 if (tun
->flags
& TUN_NO_PI
) {
1120 switch (skb
->data
[0] & 0xf0) {
1122 pi
.proto
= htons(ETH_P_IP
);
1125 pi
.proto
= htons(ETH_P_IPV6
);
1128 tun
->dev
->stats
.rx_dropped
++;
1134 skb_reset_mac_header(skb
);
1135 skb
->protocol
= pi
.proto
;
1136 skb
->dev
= tun
->dev
;
1139 skb
->protocol
= eth_type_trans(skb
, tun
->dev
);
1143 skb_reset_network_header(skb
);
1145 if (gso
.gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
1147 switch (gso
.gso_type
& ~VIRTIO_NET_HDR_GSO_ECN
) {
1148 case VIRTIO_NET_HDR_GSO_TCPV4
:
1149 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV4
;
1151 case VIRTIO_NET_HDR_GSO_TCPV6
:
1152 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV6
;
1154 case VIRTIO_NET_HDR_GSO_UDP
:
1160 netdev_warn(tun
->dev
,
1161 "%s: using disabled UFO feature; please fix this program\n",
1164 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
1165 if (skb
->protocol
== htons(ETH_P_IPV6
))
1166 ipv6_proxy_select_ident(skb
);
1170 tun
->dev
->stats
.rx_frame_errors
++;
1175 if (gso
.gso_type
& VIRTIO_NET_HDR_GSO_ECN
)
1176 skb_shinfo(skb
)->gso_type
|= SKB_GSO_TCP_ECN
;
1178 skb_shinfo(skb
)->gso_size
= gso
.gso_size
;
1179 if (skb_shinfo(skb
)->gso_size
== 0) {
1180 tun
->dev
->stats
.rx_frame_errors
++;
1185 /* Header must be checked, and gso_segs computed. */
1186 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
1187 skb_shinfo(skb
)->gso_segs
= 0;
1190 /* copy skb_ubuf_info for callback when skb has no error */
1192 skb_shinfo(skb
)->destructor_arg
= msg_control
;
1193 skb_shinfo(skb
)->tx_flags
|= SKBTX_DEV_ZEROCOPY
;
1194 skb_shinfo(skb
)->tx_flags
|= SKBTX_SHARED_FRAG
;
1197 skb_probe_transport_header(skb
, 0);
1199 rxhash
= skb_get_hash(skb
);
1202 tun
->dev
->stats
.rx_packets
++;
1203 tun
->dev
->stats
.rx_bytes
+= len
;
1205 tun_flow_update(tun
, rxhash
, tfile
);
1209 static ssize_t
tun_chr_aio_write(struct kiocb
*iocb
, const struct iovec
*iv
,
1210 unsigned long count
, loff_t pos
)
1212 struct file
*file
= iocb
->ki_filp
;
1213 struct tun_struct
*tun
= tun_get(file
);
1214 struct tun_file
*tfile
= file
->private_data
;
1220 tun_debug(KERN_INFO
, tun
, "tun_chr_write %ld\n", count
);
1222 result
= tun_get_user(tun
, tfile
, NULL
, iv
, iov_length(iv
, count
),
1223 count
, file
->f_flags
& O_NONBLOCK
);
1229 /* Put packet to the user space buffer */
1230 static ssize_t
tun_put_user(struct tun_struct
*tun
,
1231 struct tun_file
*tfile
,
1232 struct sk_buff
*skb
,
1233 const struct iovec
*iv
, int len
)
1235 struct tun_pi pi
= { 0, skb
->protocol
};
1237 int vlan_offset
= 0, copied
;
1239 int vnet_hdr_sz
= 0;
1241 if (vlan_tx_tag_present(skb
))
1242 vlan_hlen
= VLAN_HLEN
;
1244 if (tun
->flags
& TUN_VNET_HDR
)
1245 vnet_hdr_sz
= tun
->vnet_hdr_sz
;
1247 if (!(tun
->flags
& TUN_NO_PI
)) {
1248 if ((len
-= sizeof(pi
)) < 0)
1251 if (len
< skb
->len
+ vlan_hlen
+ vnet_hdr_sz
) {
1252 /* Packet will be striped */
1253 pi
.flags
|= TUN_PKT_STRIP
;
1256 if (memcpy_toiovecend(iv
, (void *) &pi
, 0, sizeof(pi
)))
1258 total
+= sizeof(pi
);
1262 struct virtio_net_hdr gso
= { 0 }; /* no info leak */
1263 if ((len
-= vnet_hdr_sz
) < 0)
1266 if (skb_is_gso(skb
)) {
1267 struct skb_shared_info
*sinfo
= skb_shinfo(skb
);
1269 /* This is a hint as to how much should be linear. */
1270 gso
.hdr_len
= skb_headlen(skb
);
1271 gso
.gso_size
= sinfo
->gso_size
;
1272 if (sinfo
->gso_type
& SKB_GSO_TCPV4
)
1273 gso
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV4
;
1274 else if (sinfo
->gso_type
& SKB_GSO_TCPV6
)
1275 gso
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV6
;
1277 pr_err("unexpected GSO type: "
1278 "0x%x, gso_size %d, hdr_len %d\n",
1279 sinfo
->gso_type
, gso
.gso_size
,
1281 print_hex_dump(KERN_ERR
, "tun: ",
1284 min((int)gso
.hdr_len
, 64), true);
1288 if (sinfo
->gso_type
& SKB_GSO_TCP_ECN
)
1289 gso
.gso_type
|= VIRTIO_NET_HDR_GSO_ECN
;
1291 gso
.gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
1293 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1294 gso
.flags
= VIRTIO_NET_HDR_F_NEEDS_CSUM
;
1295 gso
.csum_start
= skb_checksum_start_offset(skb
) +
1297 gso
.csum_offset
= skb
->csum_offset
;
1298 } else if (skb
->ip_summed
== CHECKSUM_UNNECESSARY
) {
1299 gso
.flags
= VIRTIO_NET_HDR_F_DATA_VALID
;
1300 } /* else everything is zero */
1302 if (unlikely(memcpy_toiovecend(iv
, (void *)&gso
, total
,
1305 total
+= vnet_hdr_sz
;
1309 len
= min_t(int, skb
->len
+ vlan_hlen
, len
);
1310 total
+= skb
->len
+ vlan_hlen
;
1314 __be16 h_vlan_proto
;
1318 veth
.h_vlan_proto
= skb
->vlan_proto
;
1319 veth
.h_vlan_TCI
= htons(vlan_tx_tag_get(skb
));
1321 vlan_offset
= offsetof(struct vlan_ethhdr
, h_vlan_proto
);
1323 copy
= min_t(int, vlan_offset
, len
);
1324 ret
= skb_copy_datagram_const_iovec(skb
, 0, iv
, copied
, copy
);
1330 copy
= min_t(int, sizeof(veth
), len
);
1331 ret
= memcpy_toiovecend(iv
, (void *)&veth
, copied
, copy
);
1338 skb_copy_datagram_const_iovec(skb
, vlan_offset
, iv
, copied
, len
);
1341 tun
->dev
->stats
.tx_packets
++;
1342 tun
->dev
->stats
.tx_bytes
+= len
;
1347 static ssize_t
tun_do_read(struct tun_struct
*tun
, struct tun_file
*tfile
,
1348 const struct iovec
*iv
, ssize_t len
, int noblock
)
1350 struct sk_buff
*skb
;
1352 int peeked
, err
, off
= 0;
1354 tun_debug(KERN_INFO
, tun
, "tun_do_read\n");
1359 if (tun
->dev
->reg_state
!= NETREG_REGISTERED
)
1362 /* Read frames from queue */
1363 skb
= __skb_recv_datagram(tfile
->socket
.sk
, noblock
? MSG_DONTWAIT
: 0,
1364 &peeked
, &off
, &err
);
1366 ret
= tun_put_user(tun
, tfile
, skb
, iv
, len
);
1374 static ssize_t
tun_chr_aio_read(struct kiocb
*iocb
, const struct iovec
*iv
,
1375 unsigned long count
, loff_t pos
)
1377 struct file
*file
= iocb
->ki_filp
;
1378 struct tun_file
*tfile
= file
->private_data
;
1379 struct tun_struct
*tun
= __tun_get(tfile
);
1384 len
= iov_length(iv
, count
);
1390 ret
= tun_do_read(tun
, tfile
, iv
, len
,
1391 file
->f_flags
& O_NONBLOCK
);
1392 ret
= min_t(ssize_t
, ret
, len
);
1400 static void tun_free_netdev(struct net_device
*dev
)
1402 struct tun_struct
*tun
= netdev_priv(dev
);
1404 BUG_ON(!(list_empty(&tun
->disabled
)));
1405 tun_flow_uninit(tun
);
1406 security_tun_dev_free_security(tun
->security
);
1410 static void tun_setup(struct net_device
*dev
)
1412 struct tun_struct
*tun
= netdev_priv(dev
);
1414 tun
->owner
= INVALID_UID
;
1415 tun
->group
= INVALID_GID
;
1417 dev
->ethtool_ops
= &tun_ethtool_ops
;
1418 dev
->destructor
= tun_free_netdev
;
1421 /* Trivial set of netlink ops to allow deleting tun or tap
1422 * device with netlink.
1424 static int tun_validate(struct nlattr
*tb
[], struct nlattr
*data
[])
1429 static struct rtnl_link_ops tun_link_ops __read_mostly
= {
1431 .priv_size
= sizeof(struct tun_struct
),
1433 .validate
= tun_validate
,
1436 static void tun_sock_write_space(struct sock
*sk
)
1438 struct tun_file
*tfile
;
1439 wait_queue_head_t
*wqueue
;
1441 if (!sock_writeable(sk
))
1444 if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
))
1447 wqueue
= sk_sleep(sk
);
1448 if (wqueue
&& waitqueue_active(wqueue
))
1449 wake_up_interruptible_sync_poll(wqueue
, POLLOUT
|
1450 POLLWRNORM
| POLLWRBAND
);
1452 tfile
= container_of(sk
, struct tun_file
, sk
);
1453 kill_fasync(&tfile
->fasync
, SIGIO
, POLL_OUT
);
1456 static int tun_sendmsg(struct kiocb
*iocb
, struct socket
*sock
,
1457 struct msghdr
*m
, size_t total_len
)
1460 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
1461 struct tun_struct
*tun
= __tun_get(tfile
);
1465 ret
= tun_get_user(tun
, tfile
, m
->msg_control
, m
->msg_iov
, total_len
,
1466 m
->msg_iovlen
, m
->msg_flags
& MSG_DONTWAIT
);
1471 static int tun_recvmsg(struct kiocb
*iocb
, struct socket
*sock
,
1472 struct msghdr
*m
, size_t total_len
,
1475 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
1476 struct tun_struct
*tun
= __tun_get(tfile
);
1482 if (flags
& ~(MSG_DONTWAIT
|MSG_TRUNC
|MSG_ERRQUEUE
)) {
1486 if (flags
& MSG_ERRQUEUE
) {
1487 ret
= sock_recv_errqueue(sock
->sk
, m
, total_len
,
1488 SOL_PACKET
, TUN_TX_TIMESTAMP
);
1491 ret
= tun_do_read(tun
, tfile
, m
->msg_iov
, total_len
,
1492 flags
& MSG_DONTWAIT
);
1493 if (ret
> total_len
) {
1494 m
->msg_flags
|= MSG_TRUNC
;
1495 ret
= flags
& MSG_TRUNC
? ret
: total_len
;
1502 static int tun_release(struct socket
*sock
)
1509 /* Ops structure to mimic raw sockets with tun */
1510 static const struct proto_ops tun_socket_ops
= {
1511 .sendmsg
= tun_sendmsg
,
1512 .recvmsg
= tun_recvmsg
,
1513 .release
= tun_release
,
1516 static struct proto tun_proto
= {
1518 .owner
= THIS_MODULE
,
1519 .obj_size
= sizeof(struct tun_file
),
1522 static int tun_flags(struct tun_struct
*tun
)
1526 if (tun
->flags
& TUN_TUN_DEV
)
1531 if (tun
->flags
& TUN_NO_PI
)
1534 /* This flag has no real effect. We track the value for backwards
1537 if (tun
->flags
& TUN_ONE_QUEUE
)
1538 flags
|= IFF_ONE_QUEUE
;
1540 if (tun
->flags
& TUN_VNET_HDR
)
1541 flags
|= IFF_VNET_HDR
;
1543 if (tun
->flags
& TUN_TAP_MQ
)
1544 flags
|= IFF_MULTI_QUEUE
;
1546 if (tun
->flags
& TUN_PERSIST
)
1547 flags
|= IFF_PERSIST
;
1552 static ssize_t
tun_show_flags(struct device
*dev
, struct device_attribute
*attr
,
1555 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
1556 return sprintf(buf
, "0x%x\n", tun_flags(tun
));
1559 static ssize_t
tun_show_owner(struct device
*dev
, struct device_attribute
*attr
,
1562 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
1563 return uid_valid(tun
->owner
)?
1564 sprintf(buf
, "%u\n",
1565 from_kuid_munged(current_user_ns(), tun
->owner
)):
1566 sprintf(buf
, "-1\n");
1569 static ssize_t
tun_show_group(struct device
*dev
, struct device_attribute
*attr
,
1572 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
1573 return gid_valid(tun
->group
) ?
1574 sprintf(buf
, "%u\n",
1575 from_kgid_munged(current_user_ns(), tun
->group
)):
1576 sprintf(buf
, "-1\n");
1579 static DEVICE_ATTR(tun_flags
, 0444, tun_show_flags
, NULL
);
1580 static DEVICE_ATTR(owner
, 0444, tun_show_owner
, NULL
);
1581 static DEVICE_ATTR(group
, 0444, tun_show_group
, NULL
);
1583 static int tun_set_iff(struct net
*net
, struct file
*file
, struct ifreq
*ifr
)
1585 struct tun_struct
*tun
;
1586 struct tun_file
*tfile
= file
->private_data
;
1587 struct net_device
*dev
;
1590 if (tfile
->detached
)
1593 dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
1595 if (ifr
->ifr_flags
& IFF_TUN_EXCL
)
1597 if ((ifr
->ifr_flags
& IFF_TUN
) && dev
->netdev_ops
== &tun_netdev_ops
)
1598 tun
= netdev_priv(dev
);
1599 else if ((ifr
->ifr_flags
& IFF_TAP
) && dev
->netdev_ops
== &tap_netdev_ops
)
1600 tun
= netdev_priv(dev
);
1604 if (!!(ifr
->ifr_flags
& IFF_MULTI_QUEUE
) !=
1605 !!(tun
->flags
& TUN_TAP_MQ
))
1608 if (tun_not_capable(tun
))
1610 err
= security_tun_dev_open(tun
->security
);
1614 err
= tun_attach(tun
, file
, ifr
->ifr_flags
& IFF_NOFILTER
);
1618 if (tun
->flags
& TUN_TAP_MQ
&&
1619 (tun
->numqueues
+ tun
->numdisabled
> 1)) {
1620 /* One or more queue has already been attached, no need
1621 * to initialize the device again.
1628 unsigned long flags
= 0;
1629 int queues
= ifr
->ifr_flags
& IFF_MULTI_QUEUE
?
1632 if (!ns_capable(net
->user_ns
, CAP_NET_ADMIN
))
1634 err
= security_tun_dev_create();
1639 if (ifr
->ifr_flags
& IFF_TUN
) {
1641 flags
|= TUN_TUN_DEV
;
1643 } else if (ifr
->ifr_flags
& IFF_TAP
) {
1645 flags
|= TUN_TAP_DEV
;
1651 name
= ifr
->ifr_name
;
1653 dev
= alloc_netdev_mqs(sizeof(struct tun_struct
), name
,
1654 NET_NAME_UNKNOWN
, tun_setup
, queues
,
1660 dev_net_set(dev
, net
);
1661 dev
->rtnl_link_ops
= &tun_link_ops
;
1662 dev
->ifindex
= tfile
->ifindex
;
1664 tun
= netdev_priv(dev
);
1667 tun
->txflt
.count
= 0;
1668 tun
->vnet_hdr_sz
= sizeof(struct virtio_net_hdr
);
1670 tun
->filter_attached
= false;
1671 tun
->sndbuf
= tfile
->socket
.sk
->sk_sndbuf
;
1673 spin_lock_init(&tun
->lock
);
1675 err
= security_tun_dev_alloc_security(&tun
->security
);
1682 dev
->hw_features
= NETIF_F_SG
| NETIF_F_FRAGLIST
|
1683 TUN_USER_FEATURES
| NETIF_F_HW_VLAN_CTAG_TX
|
1684 NETIF_F_HW_VLAN_STAG_TX
;
1685 dev
->features
= dev
->hw_features
;
1686 dev
->vlan_features
= dev
->features
&
1687 ~(NETIF_F_HW_VLAN_CTAG_TX
|
1688 NETIF_F_HW_VLAN_STAG_TX
);
1690 INIT_LIST_HEAD(&tun
->disabled
);
1691 err
= tun_attach(tun
, file
, false);
1695 err
= register_netdevice(tun
->dev
);
1699 if (device_create_file(&tun
->dev
->dev
, &dev_attr_tun_flags
) ||
1700 device_create_file(&tun
->dev
->dev
, &dev_attr_owner
) ||
1701 device_create_file(&tun
->dev
->dev
, &dev_attr_group
))
1702 pr_err("Failed to create tun sysfs files\n");
1705 netif_carrier_on(tun
->dev
);
1707 tun_debug(KERN_INFO
, tun
, "tun_set_iff\n");
1709 if (ifr
->ifr_flags
& IFF_NO_PI
)
1710 tun
->flags
|= TUN_NO_PI
;
1712 tun
->flags
&= ~TUN_NO_PI
;
1714 /* This flag has no real effect. We track the value for backwards
1717 if (ifr
->ifr_flags
& IFF_ONE_QUEUE
)
1718 tun
->flags
|= TUN_ONE_QUEUE
;
1720 tun
->flags
&= ~TUN_ONE_QUEUE
;
1722 if (ifr
->ifr_flags
& IFF_VNET_HDR
)
1723 tun
->flags
|= TUN_VNET_HDR
;
1725 tun
->flags
&= ~TUN_VNET_HDR
;
1727 if (ifr
->ifr_flags
& IFF_MULTI_QUEUE
)
1728 tun
->flags
|= TUN_TAP_MQ
;
1730 tun
->flags
&= ~TUN_TAP_MQ
;
1732 /* Make sure persistent devices do not get stuck in
1735 if (netif_running(tun
->dev
))
1736 netif_tx_wake_all_queues(tun
->dev
);
1738 strcpy(ifr
->ifr_name
, tun
->dev
->name
);
1742 tun_detach_all(dev
);
1744 tun_flow_uninit(tun
);
1745 security_tun_dev_free_security(tun
->security
);
1751 static void tun_get_iff(struct net
*net
, struct tun_struct
*tun
,
1754 tun_debug(KERN_INFO
, tun
, "tun_get_iff\n");
1756 strcpy(ifr
->ifr_name
, tun
->dev
->name
);
1758 ifr
->ifr_flags
= tun_flags(tun
);
1762 /* This is like a cut-down ethtool ops, except done via tun fd so no
1763 * privs required. */
1764 static int set_offload(struct tun_struct
*tun
, unsigned long arg
)
1766 netdev_features_t features
= 0;
1768 if (arg
& TUN_F_CSUM
) {
1769 features
|= NETIF_F_HW_CSUM
;
1772 if (arg
& (TUN_F_TSO4
|TUN_F_TSO6
)) {
1773 if (arg
& TUN_F_TSO_ECN
) {
1774 features
|= NETIF_F_TSO_ECN
;
1775 arg
&= ~TUN_F_TSO_ECN
;
1777 if (arg
& TUN_F_TSO4
)
1778 features
|= NETIF_F_TSO
;
1779 if (arg
& TUN_F_TSO6
)
1780 features
|= NETIF_F_TSO6
;
1781 arg
&= ~(TUN_F_TSO4
|TUN_F_TSO6
);
1785 /* This gives the user a way to test for new features in future by
1786 * trying to set them. */
1790 tun
->set_features
= features
;
1791 netdev_update_features(tun
->dev
);
1796 static void tun_detach_filter(struct tun_struct
*tun
, int n
)
1799 struct tun_file
*tfile
;
1801 for (i
= 0; i
< n
; i
++) {
1802 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
1803 sk_detach_filter(tfile
->socket
.sk
);
1806 tun
->filter_attached
= false;
1809 static int tun_attach_filter(struct tun_struct
*tun
)
1812 struct tun_file
*tfile
;
1814 for (i
= 0; i
< tun
->numqueues
; i
++) {
1815 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
1816 ret
= sk_attach_filter(&tun
->fprog
, tfile
->socket
.sk
);
1818 tun_detach_filter(tun
, i
);
1823 tun
->filter_attached
= true;
1827 static void tun_set_sndbuf(struct tun_struct
*tun
)
1829 struct tun_file
*tfile
;
1832 for (i
= 0; i
< tun
->numqueues
; i
++) {
1833 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
1834 tfile
->socket
.sk
->sk_sndbuf
= tun
->sndbuf
;
1838 static int tun_set_queue(struct file
*file
, struct ifreq
*ifr
)
1840 struct tun_file
*tfile
= file
->private_data
;
1841 struct tun_struct
*tun
;
1846 if (ifr
->ifr_flags
& IFF_ATTACH_QUEUE
) {
1847 tun
= tfile
->detached
;
1852 ret
= security_tun_dev_attach_queue(tun
->security
);
1855 ret
= tun_attach(tun
, file
, false);
1856 } else if (ifr
->ifr_flags
& IFF_DETACH_QUEUE
) {
1857 tun
= rtnl_dereference(tfile
->tun
);
1858 if (!tun
|| !(tun
->flags
& TUN_TAP_MQ
) || tfile
->detached
)
1861 __tun_detach(tfile
, false);
1870 static long __tun_chr_ioctl(struct file
*file
, unsigned int cmd
,
1871 unsigned long arg
, int ifreq_len
)
1873 struct tun_file
*tfile
= file
->private_data
;
1874 struct tun_struct
*tun
;
1875 void __user
* argp
= (void __user
*)arg
;
1881 unsigned int ifindex
;
1884 if (cmd
== TUNSETIFF
|| cmd
== TUNSETQUEUE
|| _IOC_TYPE(cmd
) == 0x89) {
1885 if (copy_from_user(&ifr
, argp
, ifreq_len
))
1888 memset(&ifr
, 0, sizeof(ifr
));
1890 if (cmd
== TUNGETFEATURES
) {
1891 /* Currently this just means: "what IFF flags are valid?".
1892 * This is needed because we never checked for invalid flags on
1894 return put_user(IFF_TUN
| IFF_TAP
| IFF_NO_PI
| IFF_ONE_QUEUE
|
1895 IFF_VNET_HDR
| IFF_MULTI_QUEUE
,
1896 (unsigned int __user
*)argp
);
1897 } else if (cmd
== TUNSETQUEUE
)
1898 return tun_set_queue(file
, &ifr
);
1903 tun
= __tun_get(tfile
);
1904 if (cmd
== TUNSETIFF
&& !tun
) {
1905 ifr
.ifr_name
[IFNAMSIZ
-1] = '\0';
1907 ret
= tun_set_iff(tfile
->net
, file
, &ifr
);
1912 if (copy_to_user(argp
, &ifr
, ifreq_len
))
1916 if (cmd
== TUNSETIFINDEX
) {
1922 if (copy_from_user(&ifindex
, argp
, sizeof(ifindex
)))
1926 tfile
->ifindex
= ifindex
;
1934 tun_debug(KERN_INFO
, tun
, "tun_chr_ioctl cmd %u\n", cmd
);
1939 tun_get_iff(current
->nsproxy
->net_ns
, tun
, &ifr
);
1941 if (tfile
->detached
)
1942 ifr
.ifr_flags
|= IFF_DETACH_QUEUE
;
1943 if (!tfile
->socket
.sk
->sk_filter
)
1944 ifr
.ifr_flags
|= IFF_NOFILTER
;
1946 if (copy_to_user(argp
, &ifr
, ifreq_len
))
1951 /* Disable/Enable checksum */
1953 /* [unimplemented] */
1954 tun_debug(KERN_INFO
, tun
, "ignored: set checksum %s\n",
1955 arg
? "disabled" : "enabled");
1959 /* Disable/Enable persist mode. Keep an extra reference to the
1960 * module to prevent the module being unprobed.
1962 if (arg
&& !(tun
->flags
& TUN_PERSIST
)) {
1963 tun
->flags
|= TUN_PERSIST
;
1964 __module_get(THIS_MODULE
);
1966 if (!arg
&& (tun
->flags
& TUN_PERSIST
)) {
1967 tun
->flags
&= ~TUN_PERSIST
;
1968 module_put(THIS_MODULE
);
1971 tun_debug(KERN_INFO
, tun
, "persist %s\n",
1972 arg
? "enabled" : "disabled");
1976 /* Set owner of the device */
1977 owner
= make_kuid(current_user_ns(), arg
);
1978 if (!uid_valid(owner
)) {
1983 tun_debug(KERN_INFO
, tun
, "owner set to %u\n",
1984 from_kuid(&init_user_ns
, tun
->owner
));
1988 /* Set group of the device */
1989 group
= make_kgid(current_user_ns(), arg
);
1990 if (!gid_valid(group
)) {
1995 tun_debug(KERN_INFO
, tun
, "group set to %u\n",
1996 from_kgid(&init_user_ns
, tun
->group
));
2000 /* Only allow setting the type when the interface is down */
2001 if (tun
->dev
->flags
& IFF_UP
) {
2002 tun_debug(KERN_INFO
, tun
,
2003 "Linktype set failed because interface is up\n");
2006 tun
->dev
->type
= (int) arg
;
2007 tun_debug(KERN_INFO
, tun
, "linktype set to %d\n",
2019 ret
= set_offload(tun
, arg
);
2022 case TUNSETTXFILTER
:
2023 /* Can be set only for TAPs */
2025 if ((tun
->flags
& TUN_TYPE_MASK
) != TUN_TAP_DEV
)
2027 ret
= update_filter(&tun
->txflt
, (void __user
*)arg
);
2031 /* Get hw address */
2032 memcpy(ifr
.ifr_hwaddr
.sa_data
, tun
->dev
->dev_addr
, ETH_ALEN
);
2033 ifr
.ifr_hwaddr
.sa_family
= tun
->dev
->type
;
2034 if (copy_to_user(argp
, &ifr
, ifreq_len
))
2039 /* Set hw address */
2040 tun_debug(KERN_DEBUG
, tun
, "set hw address: %pM\n",
2041 ifr
.ifr_hwaddr
.sa_data
);
2043 ret
= dev_set_mac_address(tun
->dev
, &ifr
.ifr_hwaddr
);
2047 sndbuf
= tfile
->socket
.sk
->sk_sndbuf
;
2048 if (copy_to_user(argp
, &sndbuf
, sizeof(sndbuf
)))
2053 if (copy_from_user(&sndbuf
, argp
, sizeof(sndbuf
))) {
2058 tun
->sndbuf
= sndbuf
;
2059 tun_set_sndbuf(tun
);
2062 case TUNGETVNETHDRSZ
:
2063 vnet_hdr_sz
= tun
->vnet_hdr_sz
;
2064 if (copy_to_user(argp
, &vnet_hdr_sz
, sizeof(vnet_hdr_sz
)))
2068 case TUNSETVNETHDRSZ
:
2069 if (copy_from_user(&vnet_hdr_sz
, argp
, sizeof(vnet_hdr_sz
))) {
2073 if (vnet_hdr_sz
< (int)sizeof(struct virtio_net_hdr
)) {
2078 tun
->vnet_hdr_sz
= vnet_hdr_sz
;
2081 case TUNATTACHFILTER
:
2082 /* Can be set only for TAPs */
2084 if ((tun
->flags
& TUN_TYPE_MASK
) != TUN_TAP_DEV
)
2087 if (copy_from_user(&tun
->fprog
, argp
, sizeof(tun
->fprog
)))
2090 ret
= tun_attach_filter(tun
);
2093 case TUNDETACHFILTER
:
2094 /* Can be set only for TAPs */
2096 if ((tun
->flags
& TUN_TYPE_MASK
) != TUN_TAP_DEV
)
2099 tun_detach_filter(tun
, tun
->numqueues
);
2104 if ((tun
->flags
& TUN_TYPE_MASK
) != TUN_TAP_DEV
)
2107 if (copy_to_user(argp
, &tun
->fprog
, sizeof(tun
->fprog
)))
2124 static long tun_chr_ioctl(struct file
*file
,
2125 unsigned int cmd
, unsigned long arg
)
2127 return __tun_chr_ioctl(file
, cmd
, arg
, sizeof (struct ifreq
));
2130 #ifdef CONFIG_COMPAT
2131 static long tun_chr_compat_ioctl(struct file
*file
,
2132 unsigned int cmd
, unsigned long arg
)
2137 case TUNSETTXFILTER
:
2142 arg
= (unsigned long)compat_ptr(arg
);
2145 arg
= (compat_ulong_t
)arg
;
2150 * compat_ifreq is shorter than ifreq, so we must not access beyond
2151 * the end of that structure. All fields that are used in this
2152 * driver are compatible though, we don't need to convert the
2155 return __tun_chr_ioctl(file
, cmd
, arg
, sizeof(struct compat_ifreq
));
2157 #endif /* CONFIG_COMPAT */
2159 static int tun_chr_fasync(int fd
, struct file
*file
, int on
)
2161 struct tun_file
*tfile
= file
->private_data
;
2164 if ((ret
= fasync_helper(fd
, file
, on
, &tfile
->fasync
)) < 0)
2168 __f_setown(file
, task_pid(current
), PIDTYPE_PID
, 0);
2169 tfile
->flags
|= TUN_FASYNC
;
2171 tfile
->flags
&= ~TUN_FASYNC
;
2177 static int tun_chr_open(struct inode
*inode
, struct file
* file
)
2179 struct tun_file
*tfile
;
2181 DBG1(KERN_INFO
, "tunX: tun_chr_open\n");
2183 tfile
= (struct tun_file
*)sk_alloc(&init_net
, AF_UNSPEC
, GFP_KERNEL
,
2187 RCU_INIT_POINTER(tfile
->tun
, NULL
);
2188 tfile
->net
= get_net(current
->nsproxy
->net_ns
);
2192 init_waitqueue_head(&tfile
->wq
.wait
);
2193 RCU_INIT_POINTER(tfile
->socket
.wq
, &tfile
->wq
);
2195 tfile
->socket
.file
= file
;
2196 tfile
->socket
.ops
= &tun_socket_ops
;
2198 sock_init_data(&tfile
->socket
, &tfile
->sk
);
2199 sk_change_net(&tfile
->sk
, tfile
->net
);
2201 tfile
->sk
.sk_write_space
= tun_sock_write_space
;
2202 tfile
->sk
.sk_sndbuf
= INT_MAX
;
2204 file
->private_data
= tfile
;
2205 set_bit(SOCK_EXTERNALLY_ALLOCATED
, &tfile
->socket
.flags
);
2206 INIT_LIST_HEAD(&tfile
->next
);
2208 sock_set_flag(&tfile
->sk
, SOCK_ZEROCOPY
);
2213 static int tun_chr_close(struct inode
*inode
, struct file
*file
)
2215 struct tun_file
*tfile
= file
->private_data
;
2216 struct net
*net
= tfile
->net
;
2218 tun_detach(tfile
, true);
2224 #ifdef CONFIG_PROC_FS
2225 static int tun_chr_show_fdinfo(struct seq_file
*m
, struct file
*f
)
2227 struct tun_struct
*tun
;
2230 memset(&ifr
, 0, sizeof(ifr
));
2235 tun_get_iff(current
->nsproxy
->net_ns
, tun
, &ifr
);
2241 return seq_printf(m
, "iff:\t%s\n", ifr
.ifr_name
);
2245 static const struct file_operations tun_fops
= {
2246 .owner
= THIS_MODULE
,
2247 .llseek
= no_llseek
,
2248 .read
= do_sync_read
,
2249 .aio_read
= tun_chr_aio_read
,
2250 .write
= do_sync_write
,
2251 .aio_write
= tun_chr_aio_write
,
2252 .poll
= tun_chr_poll
,
2253 .unlocked_ioctl
= tun_chr_ioctl
,
2254 #ifdef CONFIG_COMPAT
2255 .compat_ioctl
= tun_chr_compat_ioctl
,
2257 .open
= tun_chr_open
,
2258 .release
= tun_chr_close
,
2259 .fasync
= tun_chr_fasync
,
2260 #ifdef CONFIG_PROC_FS
2261 .show_fdinfo
= tun_chr_show_fdinfo
,
2265 static struct miscdevice tun_miscdev
= {
2268 .nodename
= "net/tun",
2272 /* ethtool interface */
2274 static int tun_get_settings(struct net_device
*dev
, struct ethtool_cmd
*cmd
)
2277 cmd
->advertising
= 0;
2278 ethtool_cmd_speed_set(cmd
, SPEED_10
);
2279 cmd
->duplex
= DUPLEX_FULL
;
2280 cmd
->port
= PORT_TP
;
2281 cmd
->phy_address
= 0;
2282 cmd
->transceiver
= XCVR_INTERNAL
;
2283 cmd
->autoneg
= AUTONEG_DISABLE
;
2289 static void tun_get_drvinfo(struct net_device
*dev
, struct ethtool_drvinfo
*info
)
2291 struct tun_struct
*tun
= netdev_priv(dev
);
2293 strlcpy(info
->driver
, DRV_NAME
, sizeof(info
->driver
));
2294 strlcpy(info
->version
, DRV_VERSION
, sizeof(info
->version
));
2296 switch (tun
->flags
& TUN_TYPE_MASK
) {
2298 strlcpy(info
->bus_info
, "tun", sizeof(info
->bus_info
));
2301 strlcpy(info
->bus_info
, "tap", sizeof(info
->bus_info
));
2306 static u32
tun_get_msglevel(struct net_device
*dev
)
2309 struct tun_struct
*tun
= netdev_priv(dev
);
2316 static void tun_set_msglevel(struct net_device
*dev
, u32 value
)
2319 struct tun_struct
*tun
= netdev_priv(dev
);
2324 static const struct ethtool_ops tun_ethtool_ops
= {
2325 .get_settings
= tun_get_settings
,
2326 .get_drvinfo
= tun_get_drvinfo
,
2327 .get_msglevel
= tun_get_msglevel
,
2328 .set_msglevel
= tun_set_msglevel
,
2329 .get_link
= ethtool_op_get_link
,
2330 .get_ts_info
= ethtool_op_get_ts_info
,
2334 static int __init
tun_init(void)
2338 pr_info("%s, %s\n", DRV_DESCRIPTION
, DRV_VERSION
);
2339 pr_info("%s\n", DRV_COPYRIGHT
);
2341 ret
= rtnl_link_register(&tun_link_ops
);
2343 pr_err("Can't register link_ops\n");
2347 ret
= misc_register(&tun_miscdev
);
2349 pr_err("Can't register misc device %d\n", TUN_MINOR
);
2354 rtnl_link_unregister(&tun_link_ops
);
2359 static void tun_cleanup(void)
2361 misc_deregister(&tun_miscdev
);
2362 rtnl_link_unregister(&tun_link_ops
);
2365 /* Get an underlying socket object from tun file. Returns error unless file is
2366 * attached to a device. The returned object works like a packet socket, it
2367 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
2368 * holding a reference to the file for as long as the socket is in use. */
2369 struct socket
*tun_get_socket(struct file
*file
)
2371 struct tun_file
*tfile
;
2372 if (file
->f_op
!= &tun_fops
)
2373 return ERR_PTR(-EINVAL
);
2374 tfile
= file
->private_data
;
2376 return ERR_PTR(-EBADFD
);
2377 return &tfile
->socket
;
2379 EXPORT_SYMBOL_GPL(tun_get_socket
);
2381 module_init(tun_init
);
2382 module_exit(tun_cleanup
);
2383 MODULE_DESCRIPTION(DRV_DESCRIPTION
);
2384 MODULE_AUTHOR(DRV_COPYRIGHT
);
2385 MODULE_LICENSE("GPL");
2386 MODULE_ALIAS_MISCDEV(TUN_MINOR
);
2387 MODULE_ALIAS("devname:net/tun");