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/sched/signal.h>
48 #include <linux/major.h>
49 #include <linux/slab.h>
50 #include <linux/poll.h>
51 #include <linux/fcntl.h>
52 #include <linux/init.h>
53 #include <linux/skbuff.h>
54 #include <linux/netdevice.h>
55 #include <linux/etherdevice.h>
56 #include <linux/miscdevice.h>
57 #include <linux/ethtool.h>
58 #include <linux/rtnetlink.h>
59 #include <linux/compat.h>
61 #include <linux/if_arp.h>
62 #include <linux/if_ether.h>
63 #include <linux/if_tun.h>
64 #include <linux/if_vlan.h>
65 #include <linux/crc32.h>
66 #include <linux/nsproxy.h>
67 #include <linux/virtio_net.h>
68 #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>
74 #include <linux/uio.h>
75 #include <linux/skb_array.h>
77 #include <linux/uaccess.h>
79 /* Uncomment to enable debugging */
80 /* #define TUN_DEBUG 1 */
85 #define tun_debug(level, tun, fmt, args...) \
88 netdev_printk(level, tun->dev, fmt, ##args); \
90 #define DBG1(level, fmt, args...) \
93 printk(level fmt, ##args); \
96 #define tun_debug(level, tun, fmt, args...) \
99 netdev_printk(level, tun->dev, fmt, ##args); \
101 #define DBG1(level, fmt, args...) \
104 printk(level fmt, ##args); \
108 /* TUN device flags */
110 /* IFF_ATTACH_QUEUE is never stored in device flags,
111 * overload it to mean fasync when stored there.
113 #define TUN_FASYNC IFF_ATTACH_QUEUE
114 /* High bits in flags field are unused. */
115 #define TUN_VNET_LE 0x80000000
116 #define TUN_VNET_BE 0x40000000
118 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
120 #define GOODCOPY_LEN 128
122 #define FLT_EXACT_COUNT 8
124 unsigned int count
; /* Number of addrs. Zero means disabled */
125 u32 mask
[2]; /* Mask of the hashed addrs */
126 unsigned char addr
[FLT_EXACT_COUNT
][ETH_ALEN
];
129 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
130 * to max number of VCPUs in guest. */
131 #define MAX_TAP_QUEUES 256
132 #define MAX_TAP_FLOWS 4096
134 #define TUN_FLOW_EXPIRE (3 * HZ)
136 struct tun_pcpu_stats
{
141 struct u64_stats_sync syncp
;
147 /* A tun_file connects an open character device to a tuntap netdevice. It
148 * also contains all socket related structures (except sock_fprog and tap_filter)
149 * to serve as one transmit queue for tuntap device. The sock_fprog and
150 * tap_filter were kept in tun_struct since they were used for filtering for the
151 * netdevice not for a specific queue (at least I didn't see the requirement for
155 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
156 * other can only be read while rcu_read_lock or rtnl_lock is held.
160 struct socket socket
;
162 struct tun_struct __rcu
*tun
;
163 struct fasync_struct
*fasync
;
164 /* only used for fasnyc */
168 unsigned int ifindex
;
170 struct list_head next
;
171 struct tun_struct
*detached
;
172 struct skb_array tx_array
;
175 struct tun_flow_entry
{
176 struct hlist_node hash_link
;
178 struct tun_struct
*tun
;
183 unsigned long updated
;
186 #define TUN_NUM_FLOW_ENTRIES 1024
188 /* Since the socket were moved to tun_file, to preserve the behavior of persist
189 * device, socket filter, sndbuf and vnet header size were restore when the
190 * file were attached to a persist device.
193 struct tun_file __rcu
*tfiles
[MAX_TAP_QUEUES
];
194 unsigned int numqueues
;
199 struct net_device
*dev
;
200 netdev_features_t set_features
;
201 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
202 NETIF_F_TSO6|NETIF_F_UFO)
207 struct tap_filter txflt
;
208 struct sock_fprog fprog
;
209 /* protected by rtnl lock */
210 bool filter_attached
;
215 struct hlist_head flows
[TUN_NUM_FLOW_ENTRIES
];
216 struct timer_list flow_gc_timer
;
217 unsigned long ageing_time
;
218 unsigned int numdisabled
;
219 struct list_head disabled
;
223 struct tun_pcpu_stats __percpu
*pcpu_stats
;
226 #ifdef CONFIG_TUN_VNET_CROSS_LE
227 static inline bool tun_legacy_is_little_endian(struct tun_struct
*tun
)
229 return tun
->flags
& TUN_VNET_BE
? false :
230 virtio_legacy_is_little_endian();
233 static long tun_get_vnet_be(struct tun_struct
*tun
, int __user
*argp
)
235 int be
= !!(tun
->flags
& TUN_VNET_BE
);
237 if (put_user(be
, argp
))
243 static long tun_set_vnet_be(struct tun_struct
*tun
, int __user
*argp
)
247 if (get_user(be
, argp
))
251 tun
->flags
|= TUN_VNET_BE
;
253 tun
->flags
&= ~TUN_VNET_BE
;
258 static inline bool tun_legacy_is_little_endian(struct tun_struct
*tun
)
260 return virtio_legacy_is_little_endian();
263 static long tun_get_vnet_be(struct tun_struct
*tun
, int __user
*argp
)
268 static long tun_set_vnet_be(struct tun_struct
*tun
, int __user
*argp
)
272 #endif /* CONFIG_TUN_VNET_CROSS_LE */
274 static inline bool tun_is_little_endian(struct tun_struct
*tun
)
276 return tun
->flags
& TUN_VNET_LE
||
277 tun_legacy_is_little_endian(tun
);
280 static inline u16
tun16_to_cpu(struct tun_struct
*tun
, __virtio16 val
)
282 return __virtio16_to_cpu(tun_is_little_endian(tun
), val
);
285 static inline __virtio16
cpu_to_tun16(struct tun_struct
*tun
, u16 val
)
287 return __cpu_to_virtio16(tun_is_little_endian(tun
), val
);
290 static inline u32
tun_hashfn(u32 rxhash
)
292 return rxhash
& 0x3ff;
295 static struct tun_flow_entry
*tun_flow_find(struct hlist_head
*head
, u32 rxhash
)
297 struct tun_flow_entry
*e
;
299 hlist_for_each_entry_rcu(e
, head
, hash_link
) {
300 if (e
->rxhash
== rxhash
)
306 static struct tun_flow_entry
*tun_flow_create(struct tun_struct
*tun
,
307 struct hlist_head
*head
,
308 u32 rxhash
, u16 queue_index
)
310 struct tun_flow_entry
*e
= kmalloc(sizeof(*e
), GFP_ATOMIC
);
313 tun_debug(KERN_INFO
, tun
, "create flow: hash %u index %u\n",
314 rxhash
, queue_index
);
315 e
->updated
= jiffies
;
318 e
->queue_index
= queue_index
;
320 hlist_add_head_rcu(&e
->hash_link
, head
);
326 static void tun_flow_delete(struct tun_struct
*tun
, struct tun_flow_entry
*e
)
328 tun_debug(KERN_INFO
, tun
, "delete flow: hash %u index %u\n",
329 e
->rxhash
, e
->queue_index
);
330 hlist_del_rcu(&e
->hash_link
);
335 static void tun_flow_flush(struct tun_struct
*tun
)
339 spin_lock_bh(&tun
->lock
);
340 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
341 struct tun_flow_entry
*e
;
342 struct hlist_node
*n
;
344 hlist_for_each_entry_safe(e
, n
, &tun
->flows
[i
], hash_link
)
345 tun_flow_delete(tun
, e
);
347 spin_unlock_bh(&tun
->lock
);
350 static void tun_flow_delete_by_queue(struct tun_struct
*tun
, u16 queue_index
)
354 spin_lock_bh(&tun
->lock
);
355 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
356 struct tun_flow_entry
*e
;
357 struct hlist_node
*n
;
359 hlist_for_each_entry_safe(e
, n
, &tun
->flows
[i
], hash_link
) {
360 if (e
->queue_index
== queue_index
)
361 tun_flow_delete(tun
, e
);
364 spin_unlock_bh(&tun
->lock
);
367 static void tun_flow_cleanup(unsigned long data
)
369 struct tun_struct
*tun
= (struct tun_struct
*)data
;
370 unsigned long delay
= tun
->ageing_time
;
371 unsigned long next_timer
= jiffies
+ delay
;
372 unsigned long count
= 0;
375 tun_debug(KERN_INFO
, tun
, "tun_flow_cleanup\n");
377 spin_lock_bh(&tun
->lock
);
378 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
379 struct tun_flow_entry
*e
;
380 struct hlist_node
*n
;
382 hlist_for_each_entry_safe(e
, n
, &tun
->flows
[i
], hash_link
) {
383 unsigned long this_timer
;
385 this_timer
= e
->updated
+ delay
;
386 if (time_before_eq(this_timer
, jiffies
))
387 tun_flow_delete(tun
, e
);
388 else if (time_before(this_timer
, next_timer
))
389 next_timer
= this_timer
;
394 mod_timer(&tun
->flow_gc_timer
, round_jiffies_up(next_timer
));
395 spin_unlock_bh(&tun
->lock
);
398 static void tun_flow_update(struct tun_struct
*tun
, u32 rxhash
,
399 struct tun_file
*tfile
)
401 struct hlist_head
*head
;
402 struct tun_flow_entry
*e
;
403 unsigned long delay
= tun
->ageing_time
;
404 u16 queue_index
= tfile
->queue_index
;
409 head
= &tun
->flows
[tun_hashfn(rxhash
)];
413 /* We may get a very small possibility of OOO during switching, not
414 * worth to optimize.*/
415 if (tun
->numqueues
== 1 || tfile
->detached
)
418 e
= tun_flow_find(head
, rxhash
);
420 /* TODO: keep queueing to old queue until it's empty? */
421 e
->queue_index
= queue_index
;
422 e
->updated
= jiffies
;
423 sock_rps_record_flow_hash(e
->rps_rxhash
);
425 spin_lock_bh(&tun
->lock
);
426 if (!tun_flow_find(head
, rxhash
) &&
427 tun
->flow_count
< MAX_TAP_FLOWS
)
428 tun_flow_create(tun
, head
, rxhash
, queue_index
);
430 if (!timer_pending(&tun
->flow_gc_timer
))
431 mod_timer(&tun
->flow_gc_timer
,
432 round_jiffies_up(jiffies
+ delay
));
433 spin_unlock_bh(&tun
->lock
);
441 * Save the hash received in the stack receive path and update the
442 * flow_hash table accordingly.
444 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry
*e
, u32 hash
)
446 if (unlikely(e
->rps_rxhash
!= hash
))
447 e
->rps_rxhash
= hash
;
450 /* We try to identify a flow through its rxhash first. The reason that
451 * we do not check rxq no. is because some cards(e.g 82599), chooses
452 * the rxq based on the txq where the last packet of the flow comes. As
453 * the userspace application move between processors, we may get a
454 * different rxq no. here. If we could not get rxhash, then we would
455 * hope the rxq no. may help here.
457 static u16
tun_select_queue(struct net_device
*dev
, struct sk_buff
*skb
,
458 void *accel_priv
, select_queue_fallback_t fallback
)
460 struct tun_struct
*tun
= netdev_priv(dev
);
461 struct tun_flow_entry
*e
;
466 numqueues
= ACCESS_ONCE(tun
->numqueues
);
468 txq
= __skb_get_hash_symmetric(skb
);
470 e
= tun_flow_find(&tun
->flows
[tun_hashfn(txq
)], txq
);
472 tun_flow_save_rps_rxhash(e
, txq
);
473 txq
= e
->queue_index
;
475 /* use multiply and shift instead of expensive divide */
476 txq
= ((u64
)txq
* numqueues
) >> 32;
477 } else if (likely(skb_rx_queue_recorded(skb
))) {
478 txq
= skb_get_rx_queue(skb
);
479 while (unlikely(txq
>= numqueues
))
487 static inline bool tun_not_capable(struct tun_struct
*tun
)
489 const struct cred
*cred
= current_cred();
490 struct net
*net
= dev_net(tun
->dev
);
492 return ((uid_valid(tun
->owner
) && !uid_eq(cred
->euid
, tun
->owner
)) ||
493 (gid_valid(tun
->group
) && !in_egroup_p(tun
->group
))) &&
494 !ns_capable(net
->user_ns
, CAP_NET_ADMIN
);
497 static void tun_set_real_num_queues(struct tun_struct
*tun
)
499 netif_set_real_num_tx_queues(tun
->dev
, tun
->numqueues
);
500 netif_set_real_num_rx_queues(tun
->dev
, tun
->numqueues
);
503 static void tun_disable_queue(struct tun_struct
*tun
, struct tun_file
*tfile
)
505 tfile
->detached
= tun
;
506 list_add_tail(&tfile
->next
, &tun
->disabled
);
510 static struct tun_struct
*tun_enable_queue(struct tun_file
*tfile
)
512 struct tun_struct
*tun
= tfile
->detached
;
514 tfile
->detached
= NULL
;
515 list_del_init(&tfile
->next
);
520 static void tun_queue_purge(struct tun_file
*tfile
)
524 while ((skb
= skb_array_consume(&tfile
->tx_array
)) != NULL
)
527 skb_queue_purge(&tfile
->sk
.sk_write_queue
);
528 skb_queue_purge(&tfile
->sk
.sk_error_queue
);
531 static void __tun_detach(struct tun_file
*tfile
, bool clean
)
533 struct tun_file
*ntfile
;
534 struct tun_struct
*tun
;
536 tun
= rtnl_dereference(tfile
->tun
);
538 if (tun
&& !tfile
->detached
) {
539 u16 index
= tfile
->queue_index
;
540 BUG_ON(index
>= tun
->numqueues
);
542 rcu_assign_pointer(tun
->tfiles
[index
],
543 tun
->tfiles
[tun
->numqueues
- 1]);
544 ntfile
= rtnl_dereference(tun
->tfiles
[index
]);
545 ntfile
->queue_index
= index
;
549 RCU_INIT_POINTER(tfile
->tun
, NULL
);
550 sock_put(&tfile
->sk
);
552 tun_disable_queue(tun
, tfile
);
555 tun_flow_delete_by_queue(tun
, tun
->numqueues
+ 1);
556 /* Drop read queue */
557 tun_queue_purge(tfile
);
558 tun_set_real_num_queues(tun
);
559 } else if (tfile
->detached
&& clean
) {
560 tun
= tun_enable_queue(tfile
);
561 sock_put(&tfile
->sk
);
565 if (tun
&& tun
->numqueues
== 0 && tun
->numdisabled
== 0) {
566 netif_carrier_off(tun
->dev
);
568 if (!(tun
->flags
& IFF_PERSIST
) &&
569 tun
->dev
->reg_state
== NETREG_REGISTERED
)
570 unregister_netdevice(tun
->dev
);
573 skb_array_cleanup(&tfile
->tx_array
);
574 sock_put(&tfile
->sk
);
578 static void tun_detach(struct tun_file
*tfile
, bool clean
)
581 __tun_detach(tfile
, clean
);
585 static void tun_detach_all(struct net_device
*dev
)
587 struct tun_struct
*tun
= netdev_priv(dev
);
588 struct tun_file
*tfile
, *tmp
;
589 int i
, n
= tun
->numqueues
;
591 for (i
= 0; i
< n
; i
++) {
592 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
594 tfile
->socket
.sk
->sk_shutdown
= RCV_SHUTDOWN
;
595 tfile
->socket
.sk
->sk_data_ready(tfile
->socket
.sk
);
596 RCU_INIT_POINTER(tfile
->tun
, NULL
);
599 list_for_each_entry(tfile
, &tun
->disabled
, next
) {
600 tfile
->socket
.sk
->sk_shutdown
= RCV_SHUTDOWN
;
601 tfile
->socket
.sk
->sk_data_ready(tfile
->socket
.sk
);
602 RCU_INIT_POINTER(tfile
->tun
, NULL
);
604 BUG_ON(tun
->numqueues
!= 0);
607 for (i
= 0; i
< n
; i
++) {
608 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
609 /* Drop read queue */
610 tun_queue_purge(tfile
);
611 sock_put(&tfile
->sk
);
613 list_for_each_entry_safe(tfile
, tmp
, &tun
->disabled
, next
) {
614 tun_enable_queue(tfile
);
615 tun_queue_purge(tfile
);
616 sock_put(&tfile
->sk
);
618 BUG_ON(tun
->numdisabled
!= 0);
620 if (tun
->flags
& IFF_PERSIST
)
621 module_put(THIS_MODULE
);
624 static int tun_attach(struct tun_struct
*tun
, struct file
*file
, bool skip_filter
)
626 struct tun_file
*tfile
= file
->private_data
;
627 struct net_device
*dev
= tun
->dev
;
630 err
= security_tun_dev_attach(tfile
->socket
.sk
, tun
->security
);
635 if (rtnl_dereference(tfile
->tun
) && !tfile
->detached
)
639 if (!(tun
->flags
& IFF_MULTI_QUEUE
) && tun
->numqueues
== 1)
643 if (!tfile
->detached
&&
644 tun
->numqueues
+ tun
->numdisabled
== MAX_TAP_QUEUES
)
649 /* Re-attach the filter to persist device */
650 if (!skip_filter
&& (tun
->filter_attached
== true)) {
651 lock_sock(tfile
->socket
.sk
);
652 err
= sk_attach_filter(&tun
->fprog
, tfile
->socket
.sk
);
653 release_sock(tfile
->socket
.sk
);
658 if (!tfile
->detached
&&
659 skb_array_init(&tfile
->tx_array
, dev
->tx_queue_len
, GFP_KERNEL
)) {
664 tfile
->queue_index
= tun
->numqueues
;
665 tfile
->socket
.sk
->sk_shutdown
&= ~RCV_SHUTDOWN
;
666 rcu_assign_pointer(tfile
->tun
, tun
);
667 rcu_assign_pointer(tun
->tfiles
[tun
->numqueues
], tfile
);
671 tun_enable_queue(tfile
);
673 sock_hold(&tfile
->sk
);
675 tun_set_real_num_queues(tun
);
677 /* device is allowed to go away first, so no need to hold extra
685 static struct tun_struct
*__tun_get(struct tun_file
*tfile
)
687 struct tun_struct
*tun
;
690 tun
= rcu_dereference(tfile
->tun
);
698 static struct tun_struct
*tun_get(struct file
*file
)
700 return __tun_get(file
->private_data
);
703 static void tun_put(struct tun_struct
*tun
)
709 static void addr_hash_set(u32
*mask
, const u8
*addr
)
711 int n
= ether_crc(ETH_ALEN
, addr
) >> 26;
712 mask
[n
>> 5] |= (1 << (n
& 31));
715 static unsigned int addr_hash_test(const u32
*mask
, const u8
*addr
)
717 int n
= ether_crc(ETH_ALEN
, addr
) >> 26;
718 return mask
[n
>> 5] & (1 << (n
& 31));
721 static int update_filter(struct tap_filter
*filter
, void __user
*arg
)
723 struct { u8 u
[ETH_ALEN
]; } *addr
;
724 struct tun_filter uf
;
725 int err
, alen
, n
, nexact
;
727 if (copy_from_user(&uf
, arg
, sizeof(uf
)))
736 alen
= ETH_ALEN
* uf
.count
;
737 addr
= memdup_user(arg
+ sizeof(uf
), alen
);
739 return PTR_ERR(addr
);
741 /* The filter is updated without holding any locks. Which is
742 * perfectly safe. We disable it first and in the worst
743 * case we'll accept a few undesired packets. */
747 /* Use first set of addresses as an exact filter */
748 for (n
= 0; n
< uf
.count
&& n
< FLT_EXACT_COUNT
; n
++)
749 memcpy(filter
->addr
[n
], addr
[n
].u
, ETH_ALEN
);
753 /* Remaining multicast addresses are hashed,
754 * unicast will leave the filter disabled. */
755 memset(filter
->mask
, 0, sizeof(filter
->mask
));
756 for (; n
< uf
.count
; n
++) {
757 if (!is_multicast_ether_addr(addr
[n
].u
)) {
758 err
= 0; /* no filter */
761 addr_hash_set(filter
->mask
, addr
[n
].u
);
764 /* For ALLMULTI just set the mask to all ones.
765 * This overrides the mask populated above. */
766 if ((uf
.flags
& TUN_FLT_ALLMULTI
))
767 memset(filter
->mask
, ~0, sizeof(filter
->mask
));
769 /* Now enable the filter */
771 filter
->count
= nexact
;
773 /* Return the number of exact filters */
780 /* Returns: 0 - drop, !=0 - accept */
781 static int run_filter(struct tap_filter
*filter
, const struct sk_buff
*skb
)
783 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
785 struct ethhdr
*eh
= (struct ethhdr
*) skb
->data
;
789 for (i
= 0; i
< filter
->count
; i
++)
790 if (ether_addr_equal(eh
->h_dest
, filter
->addr
[i
]))
793 /* Inexact match (multicast only) */
794 if (is_multicast_ether_addr(eh
->h_dest
))
795 return addr_hash_test(filter
->mask
, eh
->h_dest
);
801 * Checks whether the packet is accepted or not.
802 * Returns: 0 - drop, !=0 - accept
804 static int check_filter(struct tap_filter
*filter
, const struct sk_buff
*skb
)
809 return run_filter(filter
, skb
);
812 /* Network device part of the driver */
814 static const struct ethtool_ops tun_ethtool_ops
;
816 /* Net device detach from fd. */
817 static void tun_net_uninit(struct net_device
*dev
)
822 /* Net device open. */
823 static int tun_net_open(struct net_device
*dev
)
825 struct tun_struct
*tun
= netdev_priv(dev
);
828 netif_tx_start_all_queues(dev
);
830 for (i
= 0; i
< tun
->numqueues
; i
++) {
831 struct tun_file
*tfile
;
833 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
834 tfile
->socket
.sk
->sk_write_space(tfile
->socket
.sk
);
840 /* Net device close. */
841 static int tun_net_close(struct net_device
*dev
)
843 netif_tx_stop_all_queues(dev
);
847 /* Net device start xmit */
848 static netdev_tx_t
tun_net_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
850 struct tun_struct
*tun
= netdev_priv(dev
);
851 int txq
= skb
->queue_mapping
;
852 struct tun_file
*tfile
;
856 tfile
= rcu_dereference(tun
->tfiles
[txq
]);
857 numqueues
= ACCESS_ONCE(tun
->numqueues
);
859 /* Drop packet if interface is not attached */
860 if (txq
>= numqueues
)
864 if (numqueues
== 1 && static_key_false(&rps_needed
)) {
865 /* Select queue was not called for the skbuff, so we extract the
866 * RPS hash and save it into the flow_table here.
870 rxhash
= __skb_get_hash_symmetric(skb
);
872 struct tun_flow_entry
*e
;
873 e
= tun_flow_find(&tun
->flows
[tun_hashfn(rxhash
)],
876 tun_flow_save_rps_rxhash(e
, rxhash
);
881 tun_debug(KERN_INFO
, tun
, "tun_net_xmit %d\n", skb
->len
);
885 /* Drop if the filter does not like it.
886 * This is a noop if the filter is disabled.
887 * Filter can be enabled only for the TAP devices. */
888 if (!check_filter(&tun
->txflt
, skb
))
891 if (tfile
->socket
.sk
->sk_filter
&&
892 sk_filter(tfile
->socket
.sk
, skb
))
895 if (unlikely(skb_orphan_frags(skb
, GFP_ATOMIC
)))
898 skb_tx_timestamp(skb
);
900 /* Orphan the skb - required as we might hang on to it
901 * for indefinite time.
907 if (skb_array_produce(&tfile
->tx_array
, skb
))
910 /* Notify and wake up reader process */
911 if (tfile
->flags
& TUN_FASYNC
)
912 kill_fasync(&tfile
->fasync
, SIGIO
, POLL_IN
);
913 tfile
->socket
.sk
->sk_data_ready(tfile
->socket
.sk
);
919 this_cpu_inc(tun
->pcpu_stats
->tx_dropped
);
923 return NET_XMIT_DROP
;
926 static void tun_net_mclist(struct net_device
*dev
)
929 * This callback is supposed to deal with mc filter in
930 * _rx_ path and has nothing to do with the _tx_ path.
931 * In rx path we always accept everything userspace gives us.
935 static netdev_features_t
tun_net_fix_features(struct net_device
*dev
,
936 netdev_features_t features
)
938 struct tun_struct
*tun
= netdev_priv(dev
);
940 return (features
& tun
->set_features
) | (features
& ~TUN_USER_FEATURES
);
942 #ifdef CONFIG_NET_POLL_CONTROLLER
943 static void tun_poll_controller(struct net_device
*dev
)
946 * Tun only receives frames when:
947 * 1) the char device endpoint gets data from user space
948 * 2) the tun socket gets a sendmsg call from user space
949 * Since both of those are synchronous operations, we are guaranteed
950 * never to have pending data when we poll for it
951 * so there is nothing to do here but return.
952 * We need this though so netpoll recognizes us as an interface that
953 * supports polling, which enables bridge devices in virt setups to
954 * still use netconsole
960 static void tun_set_headroom(struct net_device
*dev
, int new_hr
)
962 struct tun_struct
*tun
= netdev_priv(dev
);
964 if (new_hr
< NET_SKB_PAD
)
965 new_hr
= NET_SKB_PAD
;
971 tun_net_get_stats64(struct net_device
*dev
, struct rtnl_link_stats64
*stats
)
973 u32 rx_dropped
= 0, tx_dropped
= 0, rx_frame_errors
= 0;
974 struct tun_struct
*tun
= netdev_priv(dev
);
975 struct tun_pcpu_stats
*p
;
978 for_each_possible_cpu(i
) {
979 u64 rxpackets
, rxbytes
, txpackets
, txbytes
;
982 p
= per_cpu_ptr(tun
->pcpu_stats
, i
);
984 start
= u64_stats_fetch_begin(&p
->syncp
);
985 rxpackets
= p
->rx_packets
;
986 rxbytes
= p
->rx_bytes
;
987 txpackets
= p
->tx_packets
;
988 txbytes
= p
->tx_bytes
;
989 } while (u64_stats_fetch_retry(&p
->syncp
, start
));
991 stats
->rx_packets
+= rxpackets
;
992 stats
->rx_bytes
+= rxbytes
;
993 stats
->tx_packets
+= txpackets
;
994 stats
->tx_bytes
+= txbytes
;
997 rx_dropped
+= p
->rx_dropped
;
998 rx_frame_errors
+= p
->rx_frame_errors
;
999 tx_dropped
+= p
->tx_dropped
;
1001 stats
->rx_dropped
= rx_dropped
;
1002 stats
->rx_frame_errors
= rx_frame_errors
;
1003 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_fix_features
= tun_net_fix_features
,
1012 .ndo_select_queue
= tun_select_queue
,
1013 #ifdef CONFIG_NET_POLL_CONTROLLER
1014 .ndo_poll_controller
= tun_poll_controller
,
1016 .ndo_set_rx_headroom
= tun_set_headroom
,
1017 .ndo_get_stats64
= tun_net_get_stats64
,
1020 static const struct net_device_ops tap_netdev_ops
= {
1021 .ndo_uninit
= tun_net_uninit
,
1022 .ndo_open
= tun_net_open
,
1023 .ndo_stop
= tun_net_close
,
1024 .ndo_start_xmit
= tun_net_xmit
,
1025 .ndo_fix_features
= tun_net_fix_features
,
1026 .ndo_set_rx_mode
= tun_net_mclist
,
1027 .ndo_set_mac_address
= eth_mac_addr
,
1028 .ndo_validate_addr
= eth_validate_addr
,
1029 .ndo_select_queue
= tun_select_queue
,
1030 #ifdef CONFIG_NET_POLL_CONTROLLER
1031 .ndo_poll_controller
= tun_poll_controller
,
1033 .ndo_features_check
= passthru_features_check
,
1034 .ndo_set_rx_headroom
= tun_set_headroom
,
1035 .ndo_get_stats64
= tun_net_get_stats64
,
1038 static void tun_flow_init(struct tun_struct
*tun
)
1042 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++)
1043 INIT_HLIST_HEAD(&tun
->flows
[i
]);
1045 tun
->ageing_time
= TUN_FLOW_EXPIRE
;
1046 setup_timer(&tun
->flow_gc_timer
, tun_flow_cleanup
, (unsigned long)tun
);
1047 mod_timer(&tun
->flow_gc_timer
,
1048 round_jiffies_up(jiffies
+ tun
->ageing_time
));
1051 static void tun_flow_uninit(struct tun_struct
*tun
)
1053 del_timer_sync(&tun
->flow_gc_timer
);
1054 tun_flow_flush(tun
);
1058 #define MAX_MTU 65535
1060 /* Initialize net device. */
1061 static void tun_net_init(struct net_device
*dev
)
1063 struct tun_struct
*tun
= netdev_priv(dev
);
1065 switch (tun
->flags
& TUN_TYPE_MASK
) {
1067 dev
->netdev_ops
= &tun_netdev_ops
;
1069 /* Point-to-Point TUN Device */
1070 dev
->hard_header_len
= 0;
1074 /* Zero header length */
1075 dev
->type
= ARPHRD_NONE
;
1076 dev
->flags
= IFF_POINTOPOINT
| IFF_NOARP
| IFF_MULTICAST
;
1080 dev
->netdev_ops
= &tap_netdev_ops
;
1081 /* Ethernet TAP Device */
1083 dev
->priv_flags
&= ~IFF_TX_SKB_SHARING
;
1084 dev
->priv_flags
|= IFF_LIVE_ADDR_CHANGE
;
1086 eth_hw_addr_random(dev
);
1091 dev
->min_mtu
= MIN_MTU
;
1092 dev
->max_mtu
= MAX_MTU
- dev
->hard_header_len
;
1095 /* Character device part */
1098 static unsigned int tun_chr_poll(struct file
*file
, poll_table
*wait
)
1100 struct tun_file
*tfile
= file
->private_data
;
1101 struct tun_struct
*tun
= __tun_get(tfile
);
1103 unsigned int mask
= 0;
1108 sk
= tfile
->socket
.sk
;
1110 tun_debug(KERN_INFO
, tun
, "tun_chr_poll\n");
1112 poll_wait(file
, sk_sleep(sk
), wait
);
1114 if (!skb_array_empty(&tfile
->tx_array
))
1115 mask
|= POLLIN
| POLLRDNORM
;
1117 if (tun
->dev
->flags
& IFF_UP
&&
1118 (sock_writeable(sk
) ||
1119 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
) &&
1120 sock_writeable(sk
))))
1121 mask
|= POLLOUT
| POLLWRNORM
;
1123 if (tun
->dev
->reg_state
!= NETREG_REGISTERED
)
1130 /* prepad is the amount to reserve at front. len is length after that.
1131 * linear is a hint as to how much to copy (usually headers). */
1132 static struct sk_buff
*tun_alloc_skb(struct tun_file
*tfile
,
1133 size_t prepad
, size_t len
,
1134 size_t linear
, int noblock
)
1136 struct sock
*sk
= tfile
->socket
.sk
;
1137 struct sk_buff
*skb
;
1140 /* Under a page? Don't bother with paged skb. */
1141 if (prepad
+ len
< PAGE_SIZE
|| !linear
)
1144 skb
= sock_alloc_send_pskb(sk
, prepad
+ linear
, len
- linear
, noblock
,
1147 return ERR_PTR(err
);
1149 skb_reserve(skb
, prepad
);
1150 skb_put(skb
, linear
);
1151 skb
->data_len
= len
- linear
;
1152 skb
->len
+= len
- linear
;
1157 static void tun_rx_batched(struct tun_struct
*tun
, struct tun_file
*tfile
,
1158 struct sk_buff
*skb
, int more
)
1160 struct sk_buff_head
*queue
= &tfile
->sk
.sk_write_queue
;
1161 struct sk_buff_head process_queue
;
1162 u32 rx_batched
= tun
->rx_batched
;
1165 if (!rx_batched
|| (!more
&& skb_queue_empty(queue
))) {
1167 netif_receive_skb(skb
);
1172 spin_lock(&queue
->lock
);
1173 if (!more
|| skb_queue_len(queue
) == rx_batched
) {
1174 __skb_queue_head_init(&process_queue
);
1175 skb_queue_splice_tail_init(queue
, &process_queue
);
1178 __skb_queue_tail(queue
, skb
);
1180 spin_unlock(&queue
->lock
);
1183 struct sk_buff
*nskb
;
1186 while ((nskb
= __skb_dequeue(&process_queue
)))
1187 netif_receive_skb(nskb
);
1188 netif_receive_skb(skb
);
1193 /* Get packet from user space buffer */
1194 static ssize_t
tun_get_user(struct tun_struct
*tun
, struct tun_file
*tfile
,
1195 void *msg_control
, struct iov_iter
*from
,
1196 int noblock
, bool more
)
1198 struct tun_pi pi
= { 0, cpu_to_be16(ETH_P_IP
) };
1199 struct sk_buff
*skb
;
1200 size_t total_len
= iov_iter_count(from
);
1201 size_t len
= total_len
, align
= tun
->align
, linear
;
1202 struct virtio_net_hdr gso
= { 0 };
1203 struct tun_pcpu_stats
*stats
;
1206 bool zerocopy
= false;
1210 if (!(tun
->dev
->flags
& IFF_UP
))
1213 if (!(tun
->flags
& IFF_NO_PI
)) {
1214 if (len
< sizeof(pi
))
1218 if (!copy_from_iter_full(&pi
, sizeof(pi
), from
))
1222 if (tun
->flags
& IFF_VNET_HDR
) {
1223 int vnet_hdr_sz
= READ_ONCE(tun
->vnet_hdr_sz
);
1225 if (len
< vnet_hdr_sz
)
1229 if (!copy_from_iter_full(&gso
, sizeof(gso
), from
))
1232 if ((gso
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) &&
1233 tun16_to_cpu(tun
, gso
.csum_start
) + tun16_to_cpu(tun
, gso
.csum_offset
) + 2 > tun16_to_cpu(tun
, gso
.hdr_len
))
1234 gso
.hdr_len
= cpu_to_tun16(tun
, tun16_to_cpu(tun
, gso
.csum_start
) + tun16_to_cpu(tun
, gso
.csum_offset
) + 2);
1236 if (tun16_to_cpu(tun
, gso
.hdr_len
) > len
)
1238 iov_iter_advance(from
, vnet_hdr_sz
- sizeof(gso
));
1241 if ((tun
->flags
& TUN_TYPE_MASK
) == IFF_TAP
) {
1242 align
+= NET_IP_ALIGN
;
1243 if (unlikely(len
< ETH_HLEN
||
1244 (gso
.hdr_len
&& tun16_to_cpu(tun
, gso
.hdr_len
) < ETH_HLEN
)))
1248 good_linear
= SKB_MAX_HEAD(align
);
1251 struct iov_iter i
= *from
;
1253 /* There are 256 bytes to be copied in skb, so there is
1254 * enough room for skb expand head in case it is used.
1255 * The rest of the buffer is mapped from userspace.
1257 copylen
= gso
.hdr_len
? tun16_to_cpu(tun
, gso
.hdr_len
) : GOODCOPY_LEN
;
1258 if (copylen
> good_linear
)
1259 copylen
= good_linear
;
1261 iov_iter_advance(&i
, copylen
);
1262 if (iov_iter_npages(&i
, INT_MAX
) <= MAX_SKB_FRAGS
)
1268 if (tun16_to_cpu(tun
, gso
.hdr_len
) > good_linear
)
1269 linear
= good_linear
;
1271 linear
= tun16_to_cpu(tun
, gso
.hdr_len
);
1274 skb
= tun_alloc_skb(tfile
, align
, copylen
, linear
, noblock
);
1276 if (PTR_ERR(skb
) != -EAGAIN
)
1277 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1278 return PTR_ERR(skb
);
1282 err
= zerocopy_sg_from_iter(skb
, from
);
1284 err
= skb_copy_datagram_from_iter(skb
, 0, from
, len
);
1287 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1292 if (virtio_net_hdr_to_skb(skb
, &gso
, tun_is_little_endian(tun
))) {
1293 this_cpu_inc(tun
->pcpu_stats
->rx_frame_errors
);
1298 switch (tun
->flags
& TUN_TYPE_MASK
) {
1300 if (tun
->flags
& IFF_NO_PI
) {
1301 u8 ip_version
= skb
->len
? (skb
->data
[0] >> 4) : 0;
1303 switch (ip_version
) {
1305 pi
.proto
= htons(ETH_P_IP
);
1308 pi
.proto
= htons(ETH_P_IPV6
);
1311 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1317 skb_reset_mac_header(skb
);
1318 skb
->protocol
= pi
.proto
;
1319 skb
->dev
= tun
->dev
;
1322 skb
->protocol
= eth_type_trans(skb
, tun
->dev
);
1326 /* copy skb_ubuf_info for callback when skb has no error */
1328 skb_shinfo(skb
)->destructor_arg
= msg_control
;
1329 skb_shinfo(skb
)->tx_flags
|= SKBTX_DEV_ZEROCOPY
;
1330 skb_shinfo(skb
)->tx_flags
|= SKBTX_SHARED_FRAG
;
1331 } else if (msg_control
) {
1332 struct ubuf_info
*uarg
= msg_control
;
1333 uarg
->callback(uarg
, false);
1336 skb_reset_network_header(skb
);
1337 skb_probe_transport_header(skb
, 0);
1339 rxhash
= __skb_get_hash_symmetric(skb
);
1340 #ifndef CONFIG_4KSTACKS
1341 tun_rx_batched(tun
, tfile
, skb
, more
);
1346 stats
= get_cpu_ptr(tun
->pcpu_stats
);
1347 u64_stats_update_begin(&stats
->syncp
);
1348 stats
->rx_packets
++;
1349 stats
->rx_bytes
+= len
;
1350 u64_stats_update_end(&stats
->syncp
);
1353 tun_flow_update(tun
, rxhash
, tfile
);
1357 static ssize_t
tun_chr_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
1359 struct file
*file
= iocb
->ki_filp
;
1360 struct tun_struct
*tun
= tun_get(file
);
1361 struct tun_file
*tfile
= file
->private_data
;
1367 result
= tun_get_user(tun
, tfile
, NULL
, from
,
1368 file
->f_flags
& O_NONBLOCK
, false);
1374 /* Put packet to the user space buffer */
1375 static ssize_t
tun_put_user(struct tun_struct
*tun
,
1376 struct tun_file
*tfile
,
1377 struct sk_buff
*skb
,
1378 struct iov_iter
*iter
)
1380 struct tun_pi pi
= { 0, skb
->protocol
};
1381 struct tun_pcpu_stats
*stats
;
1383 int vlan_offset
= 0;
1385 int vnet_hdr_sz
= 0;
1387 if (skb_vlan_tag_present(skb
))
1388 vlan_hlen
= VLAN_HLEN
;
1390 if (tun
->flags
& IFF_VNET_HDR
)
1391 vnet_hdr_sz
= READ_ONCE(tun
->vnet_hdr_sz
);
1393 total
= skb
->len
+ vlan_hlen
+ vnet_hdr_sz
;
1395 if (!(tun
->flags
& IFF_NO_PI
)) {
1396 if (iov_iter_count(iter
) < sizeof(pi
))
1399 total
+= sizeof(pi
);
1400 if (iov_iter_count(iter
) < total
) {
1401 /* Packet will be striped */
1402 pi
.flags
|= TUN_PKT_STRIP
;
1405 if (copy_to_iter(&pi
, sizeof(pi
), iter
) != sizeof(pi
))
1410 struct virtio_net_hdr gso
;
1412 if (iov_iter_count(iter
) < vnet_hdr_sz
)
1415 if (virtio_net_hdr_from_skb(skb
, &gso
,
1416 tun_is_little_endian(tun
), true)) {
1417 struct skb_shared_info
*sinfo
= skb_shinfo(skb
);
1418 pr_err("unexpected GSO type: "
1419 "0x%x, gso_size %d, hdr_len %d\n",
1420 sinfo
->gso_type
, tun16_to_cpu(tun
, gso
.gso_size
),
1421 tun16_to_cpu(tun
, gso
.hdr_len
));
1422 print_hex_dump(KERN_ERR
, "tun: ",
1425 min((int)tun16_to_cpu(tun
, gso
.hdr_len
), 64), true);
1430 if (copy_to_iter(&gso
, sizeof(gso
), iter
) != sizeof(gso
))
1433 iov_iter_advance(iter
, vnet_hdr_sz
- sizeof(gso
));
1439 __be16 h_vlan_proto
;
1443 veth
.h_vlan_proto
= skb
->vlan_proto
;
1444 veth
.h_vlan_TCI
= htons(skb_vlan_tag_get(skb
));
1446 vlan_offset
= offsetof(struct vlan_ethhdr
, h_vlan_proto
);
1448 ret
= skb_copy_datagram_iter(skb
, 0, iter
, vlan_offset
);
1449 if (ret
|| !iov_iter_count(iter
))
1452 ret
= copy_to_iter(&veth
, sizeof(veth
), iter
);
1453 if (ret
!= sizeof(veth
) || !iov_iter_count(iter
))
1457 skb_copy_datagram_iter(skb
, vlan_offset
, iter
, skb
->len
- vlan_offset
);
1460 /* caller is in process context, */
1461 stats
= get_cpu_ptr(tun
->pcpu_stats
);
1462 u64_stats_update_begin(&stats
->syncp
);
1463 stats
->tx_packets
++;
1464 stats
->tx_bytes
+= skb
->len
+ vlan_hlen
;
1465 u64_stats_update_end(&stats
->syncp
);
1466 put_cpu_ptr(tun
->pcpu_stats
);
1471 static struct sk_buff
*tun_ring_recv(struct tun_file
*tfile
, int noblock
,
1474 DECLARE_WAITQUEUE(wait
, current
);
1475 struct sk_buff
*skb
= NULL
;
1478 skb
= skb_array_consume(&tfile
->tx_array
);
1486 add_wait_queue(&tfile
->wq
.wait
, &wait
);
1487 current
->state
= TASK_INTERRUPTIBLE
;
1490 skb
= skb_array_consume(&tfile
->tx_array
);
1493 if (signal_pending(current
)) {
1494 error
= -ERESTARTSYS
;
1497 if (tfile
->socket
.sk
->sk_shutdown
& RCV_SHUTDOWN
) {
1505 current
->state
= TASK_RUNNING
;
1506 remove_wait_queue(&tfile
->wq
.wait
, &wait
);
1513 static ssize_t
tun_do_read(struct tun_struct
*tun
, struct tun_file
*tfile
,
1514 struct iov_iter
*to
,
1515 int noblock
, struct sk_buff
*skb
)
1520 tun_debug(KERN_INFO
, tun
, "tun_do_read\n");
1522 if (!iov_iter_count(to
))
1526 /* Read frames from ring */
1527 skb
= tun_ring_recv(tfile
, noblock
, &err
);
1532 ret
= tun_put_user(tun
, tfile
, skb
, to
);
1533 if (unlikely(ret
< 0))
1541 static ssize_t
tun_chr_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
1543 struct file
*file
= iocb
->ki_filp
;
1544 struct tun_file
*tfile
= file
->private_data
;
1545 struct tun_struct
*tun
= __tun_get(tfile
);
1546 ssize_t len
= iov_iter_count(to
), ret
;
1550 ret
= tun_do_read(tun
, tfile
, to
, file
->f_flags
& O_NONBLOCK
, NULL
);
1551 ret
= min_t(ssize_t
, ret
, len
);
1558 static void tun_free_netdev(struct net_device
*dev
)
1560 struct tun_struct
*tun
= netdev_priv(dev
);
1562 BUG_ON(!(list_empty(&tun
->disabled
)));
1563 free_percpu(tun
->pcpu_stats
);
1564 tun_flow_uninit(tun
);
1565 security_tun_dev_free_security(tun
->security
);
1568 static void tun_setup(struct net_device
*dev
)
1570 struct tun_struct
*tun
= netdev_priv(dev
);
1572 tun
->owner
= INVALID_UID
;
1573 tun
->group
= INVALID_GID
;
1575 dev
->ethtool_ops
= &tun_ethtool_ops
;
1576 dev
->needs_free_netdev
= true;
1577 dev
->priv_destructor
= tun_free_netdev
;
1578 /* We prefer our own queue length */
1579 dev
->tx_queue_len
= TUN_READQ_SIZE
;
1582 /* Trivial set of netlink ops to allow deleting tun or tap
1583 * device with netlink.
1585 static int tun_validate(struct nlattr
*tb
[], struct nlattr
*data
[],
1586 struct netlink_ext_ack
*extack
)
1591 static struct rtnl_link_ops tun_link_ops __read_mostly
= {
1593 .priv_size
= sizeof(struct tun_struct
),
1595 .validate
= tun_validate
,
1598 static void tun_sock_write_space(struct sock
*sk
)
1600 struct tun_file
*tfile
;
1601 wait_queue_head_t
*wqueue
;
1603 if (!sock_writeable(sk
))
1606 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
))
1609 wqueue
= sk_sleep(sk
);
1610 if (wqueue
&& waitqueue_active(wqueue
))
1611 wake_up_interruptible_sync_poll(wqueue
, POLLOUT
|
1612 POLLWRNORM
| POLLWRBAND
);
1614 tfile
= container_of(sk
, struct tun_file
, sk
);
1615 kill_fasync(&tfile
->fasync
, SIGIO
, POLL_OUT
);
1618 static int tun_sendmsg(struct socket
*sock
, struct msghdr
*m
, size_t total_len
)
1621 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
1622 struct tun_struct
*tun
= __tun_get(tfile
);
1627 ret
= tun_get_user(tun
, tfile
, m
->msg_control
, &m
->msg_iter
,
1628 m
->msg_flags
& MSG_DONTWAIT
,
1629 m
->msg_flags
& MSG_MORE
);
1634 static int tun_recvmsg(struct socket
*sock
, struct msghdr
*m
, size_t total_len
,
1637 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
1638 struct tun_struct
*tun
= __tun_get(tfile
);
1644 if (flags
& ~(MSG_DONTWAIT
|MSG_TRUNC
|MSG_ERRQUEUE
)) {
1648 if (flags
& MSG_ERRQUEUE
) {
1649 ret
= sock_recv_errqueue(sock
->sk
, m
, total_len
,
1650 SOL_PACKET
, TUN_TX_TIMESTAMP
);
1653 ret
= tun_do_read(tun
, tfile
, &m
->msg_iter
, flags
& MSG_DONTWAIT
,
1655 if (ret
> (ssize_t
)total_len
) {
1656 m
->msg_flags
|= MSG_TRUNC
;
1657 ret
= flags
& MSG_TRUNC
? ret
: total_len
;
1664 static int tun_peek_len(struct socket
*sock
)
1666 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
1667 struct tun_struct
*tun
;
1670 tun
= __tun_get(tfile
);
1674 ret
= skb_array_peek_len(&tfile
->tx_array
);
1680 /* Ops structure to mimic raw sockets with tun */
1681 static const struct proto_ops tun_socket_ops
= {
1682 .peek_len
= tun_peek_len
,
1683 .sendmsg
= tun_sendmsg
,
1684 .recvmsg
= tun_recvmsg
,
1687 static struct proto tun_proto
= {
1689 .owner
= THIS_MODULE
,
1690 .obj_size
= sizeof(struct tun_file
),
1693 static int tun_flags(struct tun_struct
*tun
)
1695 return tun
->flags
& (TUN_FEATURES
| IFF_PERSIST
| IFF_TUN
| IFF_TAP
);
1698 static ssize_t
tun_show_flags(struct device
*dev
, struct device_attribute
*attr
,
1701 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
1702 return sprintf(buf
, "0x%x\n", tun_flags(tun
));
1705 static ssize_t
tun_show_owner(struct device
*dev
, struct device_attribute
*attr
,
1708 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
1709 return uid_valid(tun
->owner
)?
1710 sprintf(buf
, "%u\n",
1711 from_kuid_munged(current_user_ns(), tun
->owner
)):
1712 sprintf(buf
, "-1\n");
1715 static ssize_t
tun_show_group(struct device
*dev
, struct device_attribute
*attr
,
1718 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
1719 return gid_valid(tun
->group
) ?
1720 sprintf(buf
, "%u\n",
1721 from_kgid_munged(current_user_ns(), tun
->group
)):
1722 sprintf(buf
, "-1\n");
1725 static DEVICE_ATTR(tun_flags
, 0444, tun_show_flags
, NULL
);
1726 static DEVICE_ATTR(owner
, 0444, tun_show_owner
, NULL
);
1727 static DEVICE_ATTR(group
, 0444, tun_show_group
, NULL
);
1729 static struct attribute
*tun_dev_attrs
[] = {
1730 &dev_attr_tun_flags
.attr
,
1731 &dev_attr_owner
.attr
,
1732 &dev_attr_group
.attr
,
1736 static const struct attribute_group tun_attr_group
= {
1737 .attrs
= tun_dev_attrs
1740 static int tun_set_iff(struct net
*net
, struct file
*file
, struct ifreq
*ifr
)
1742 struct tun_struct
*tun
;
1743 struct tun_file
*tfile
= file
->private_data
;
1744 struct net_device
*dev
;
1747 if (tfile
->detached
)
1750 dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
1752 if (ifr
->ifr_flags
& IFF_TUN_EXCL
)
1754 if ((ifr
->ifr_flags
& IFF_TUN
) && dev
->netdev_ops
== &tun_netdev_ops
)
1755 tun
= netdev_priv(dev
);
1756 else if ((ifr
->ifr_flags
& IFF_TAP
) && dev
->netdev_ops
== &tap_netdev_ops
)
1757 tun
= netdev_priv(dev
);
1761 if (!!(ifr
->ifr_flags
& IFF_MULTI_QUEUE
) !=
1762 !!(tun
->flags
& IFF_MULTI_QUEUE
))
1765 if (tun_not_capable(tun
))
1767 err
= security_tun_dev_open(tun
->security
);
1771 err
= tun_attach(tun
, file
, ifr
->ifr_flags
& IFF_NOFILTER
);
1775 if (tun
->flags
& IFF_MULTI_QUEUE
&&
1776 (tun
->numqueues
+ tun
->numdisabled
> 1)) {
1777 /* One or more queue has already been attached, no need
1778 * to initialize the device again.
1785 unsigned long flags
= 0;
1786 int queues
= ifr
->ifr_flags
& IFF_MULTI_QUEUE
?
1789 if (!ns_capable(net
->user_ns
, CAP_NET_ADMIN
))
1791 err
= security_tun_dev_create();
1796 if (ifr
->ifr_flags
& IFF_TUN
) {
1800 } else if (ifr
->ifr_flags
& IFF_TAP
) {
1808 name
= ifr
->ifr_name
;
1810 dev
= alloc_netdev_mqs(sizeof(struct tun_struct
), name
,
1811 NET_NAME_UNKNOWN
, tun_setup
, queues
,
1816 err
= dev_get_valid_name(net
, dev
, name
);
1820 dev_net_set(dev
, net
);
1821 dev
->rtnl_link_ops
= &tun_link_ops
;
1822 dev
->ifindex
= tfile
->ifindex
;
1823 dev
->sysfs_groups
[0] = &tun_attr_group
;
1825 tun
= netdev_priv(dev
);
1828 tun
->txflt
.count
= 0;
1829 tun
->vnet_hdr_sz
= sizeof(struct virtio_net_hdr
);
1831 tun
->align
= NET_SKB_PAD
;
1832 tun
->filter_attached
= false;
1833 tun
->sndbuf
= tfile
->socket
.sk
->sk_sndbuf
;
1834 tun
->rx_batched
= 0;
1836 tun
->pcpu_stats
= netdev_alloc_pcpu_stats(struct tun_pcpu_stats
);
1837 if (!tun
->pcpu_stats
) {
1842 spin_lock_init(&tun
->lock
);
1844 err
= security_tun_dev_alloc_security(&tun
->security
);
1851 dev
->hw_features
= NETIF_F_SG
| NETIF_F_FRAGLIST
|
1852 TUN_USER_FEATURES
| NETIF_F_HW_VLAN_CTAG_TX
|
1853 NETIF_F_HW_VLAN_STAG_TX
;
1854 dev
->features
= dev
->hw_features
| NETIF_F_LLTX
;
1855 dev
->vlan_features
= dev
->features
&
1856 ~(NETIF_F_HW_VLAN_CTAG_TX
|
1857 NETIF_F_HW_VLAN_STAG_TX
);
1859 INIT_LIST_HEAD(&tun
->disabled
);
1860 err
= tun_attach(tun
, file
, false);
1864 err
= register_netdevice(tun
->dev
);
1869 netif_carrier_on(tun
->dev
);
1871 tun_debug(KERN_INFO
, tun
, "tun_set_iff\n");
1873 tun
->flags
= (tun
->flags
& ~TUN_FEATURES
) |
1874 (ifr
->ifr_flags
& TUN_FEATURES
);
1876 /* Make sure persistent devices do not get stuck in
1879 if (netif_running(tun
->dev
))
1880 netif_tx_wake_all_queues(tun
->dev
);
1882 strcpy(ifr
->ifr_name
, tun
->dev
->name
);
1886 tun_detach_all(dev
);
1887 /* register_netdevice() already called tun_free_netdev() */
1891 tun_flow_uninit(tun
);
1892 security_tun_dev_free_security(tun
->security
);
1894 free_percpu(tun
->pcpu_stats
);
1900 static void tun_get_iff(struct net
*net
, struct tun_struct
*tun
,
1903 tun_debug(KERN_INFO
, tun
, "tun_get_iff\n");
1905 strcpy(ifr
->ifr_name
, tun
->dev
->name
);
1907 ifr
->ifr_flags
= tun_flags(tun
);
1911 /* This is like a cut-down ethtool ops, except done via tun fd so no
1912 * privs required. */
1913 static int set_offload(struct tun_struct
*tun
, unsigned long arg
)
1915 netdev_features_t features
= 0;
1917 if (arg
& TUN_F_CSUM
) {
1918 features
|= NETIF_F_HW_CSUM
;
1921 if (arg
& (TUN_F_TSO4
|TUN_F_TSO6
)) {
1922 if (arg
& TUN_F_TSO_ECN
) {
1923 features
|= NETIF_F_TSO_ECN
;
1924 arg
&= ~TUN_F_TSO_ECN
;
1926 if (arg
& TUN_F_TSO4
)
1927 features
|= NETIF_F_TSO
;
1928 if (arg
& TUN_F_TSO6
)
1929 features
|= NETIF_F_TSO6
;
1930 arg
&= ~(TUN_F_TSO4
|TUN_F_TSO6
);
1933 if (arg
& TUN_F_UFO
) {
1934 features
|= NETIF_F_UFO
;
1939 /* This gives the user a way to test for new features in future by
1940 * trying to set them. */
1944 tun
->set_features
= features
;
1945 tun
->dev
->wanted_features
&= ~TUN_USER_FEATURES
;
1946 tun
->dev
->wanted_features
|= features
;
1947 netdev_update_features(tun
->dev
);
1952 static void tun_detach_filter(struct tun_struct
*tun
, int n
)
1955 struct tun_file
*tfile
;
1957 for (i
= 0; i
< n
; i
++) {
1958 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
1959 lock_sock(tfile
->socket
.sk
);
1960 sk_detach_filter(tfile
->socket
.sk
);
1961 release_sock(tfile
->socket
.sk
);
1964 tun
->filter_attached
= false;
1967 static int tun_attach_filter(struct tun_struct
*tun
)
1970 struct tun_file
*tfile
;
1972 for (i
= 0; i
< tun
->numqueues
; i
++) {
1973 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
1974 lock_sock(tfile
->socket
.sk
);
1975 ret
= sk_attach_filter(&tun
->fprog
, tfile
->socket
.sk
);
1976 release_sock(tfile
->socket
.sk
);
1978 tun_detach_filter(tun
, i
);
1983 tun
->filter_attached
= true;
1987 static void tun_set_sndbuf(struct tun_struct
*tun
)
1989 struct tun_file
*tfile
;
1992 for (i
= 0; i
< tun
->numqueues
; i
++) {
1993 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
1994 tfile
->socket
.sk
->sk_sndbuf
= tun
->sndbuf
;
1998 static int tun_set_queue(struct file
*file
, struct ifreq
*ifr
)
2000 struct tun_file
*tfile
= file
->private_data
;
2001 struct tun_struct
*tun
;
2006 if (ifr
->ifr_flags
& IFF_ATTACH_QUEUE
) {
2007 tun
= tfile
->detached
;
2012 ret
= security_tun_dev_attach_queue(tun
->security
);
2015 ret
= tun_attach(tun
, file
, false);
2016 } else if (ifr
->ifr_flags
& IFF_DETACH_QUEUE
) {
2017 tun
= rtnl_dereference(tfile
->tun
);
2018 if (!tun
|| !(tun
->flags
& IFF_MULTI_QUEUE
) || tfile
->detached
)
2021 __tun_detach(tfile
, false);
2030 static long __tun_chr_ioctl(struct file
*file
, unsigned int cmd
,
2031 unsigned long arg
, int ifreq_len
)
2033 struct tun_file
*tfile
= file
->private_data
;
2034 struct tun_struct
*tun
;
2035 void __user
* argp
= (void __user
*)arg
;
2041 unsigned int ifindex
;
2045 if (cmd
== TUNSETIFF
|| cmd
== TUNSETQUEUE
|| _IOC_TYPE(cmd
) == SOCK_IOC_TYPE
) {
2046 if (copy_from_user(&ifr
, argp
, ifreq_len
))
2049 memset(&ifr
, 0, sizeof(ifr
));
2051 if (cmd
== TUNGETFEATURES
) {
2052 /* Currently this just means: "what IFF flags are valid?".
2053 * This is needed because we never checked for invalid flags on
2056 return put_user(IFF_TUN
| IFF_TAP
| TUN_FEATURES
,
2057 (unsigned int __user
*)argp
);
2058 } else if (cmd
== TUNSETQUEUE
)
2059 return tun_set_queue(file
, &ifr
);
2064 tun
= __tun_get(tfile
);
2065 if (cmd
== TUNSETIFF
) {
2070 ifr
.ifr_name
[IFNAMSIZ
-1] = '\0';
2072 ret
= tun_set_iff(sock_net(&tfile
->sk
), file
, &ifr
);
2077 if (copy_to_user(argp
, &ifr
, ifreq_len
))
2081 if (cmd
== TUNSETIFINDEX
) {
2087 if (copy_from_user(&ifindex
, argp
, sizeof(ifindex
)))
2091 tfile
->ifindex
= ifindex
;
2099 tun_debug(KERN_INFO
, tun
, "tun_chr_ioctl cmd %u\n", cmd
);
2104 tun_get_iff(current
->nsproxy
->net_ns
, tun
, &ifr
);
2106 if (tfile
->detached
)
2107 ifr
.ifr_flags
|= IFF_DETACH_QUEUE
;
2108 if (!tfile
->socket
.sk
->sk_filter
)
2109 ifr
.ifr_flags
|= IFF_NOFILTER
;
2111 if (copy_to_user(argp
, &ifr
, ifreq_len
))
2116 /* Disable/Enable checksum */
2118 /* [unimplemented] */
2119 tun_debug(KERN_INFO
, tun
, "ignored: set checksum %s\n",
2120 arg
? "disabled" : "enabled");
2124 /* Disable/Enable persist mode. Keep an extra reference to the
2125 * module to prevent the module being unprobed.
2127 if (arg
&& !(tun
->flags
& IFF_PERSIST
)) {
2128 tun
->flags
|= IFF_PERSIST
;
2129 __module_get(THIS_MODULE
);
2131 if (!arg
&& (tun
->flags
& IFF_PERSIST
)) {
2132 tun
->flags
&= ~IFF_PERSIST
;
2133 module_put(THIS_MODULE
);
2136 tun_debug(KERN_INFO
, tun
, "persist %s\n",
2137 arg
? "enabled" : "disabled");
2141 /* Set owner of the device */
2142 owner
= make_kuid(current_user_ns(), arg
);
2143 if (!uid_valid(owner
)) {
2148 tun_debug(KERN_INFO
, tun
, "owner set to %u\n",
2149 from_kuid(&init_user_ns
, tun
->owner
));
2153 /* Set group of the device */
2154 group
= make_kgid(current_user_ns(), arg
);
2155 if (!gid_valid(group
)) {
2160 tun_debug(KERN_INFO
, tun
, "group set to %u\n",
2161 from_kgid(&init_user_ns
, tun
->group
));
2165 /* Only allow setting the type when the interface is down */
2166 if (tun
->dev
->flags
& IFF_UP
) {
2167 tun_debug(KERN_INFO
, tun
,
2168 "Linktype set failed because interface is up\n");
2171 tun
->dev
->type
= (int) arg
;
2172 tun_debug(KERN_INFO
, tun
, "linktype set to %d\n",
2184 ret
= set_offload(tun
, arg
);
2187 case TUNSETTXFILTER
:
2188 /* Can be set only for TAPs */
2190 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2192 ret
= update_filter(&tun
->txflt
, (void __user
*)arg
);
2196 /* Get hw address */
2197 memcpy(ifr
.ifr_hwaddr
.sa_data
, tun
->dev
->dev_addr
, ETH_ALEN
);
2198 ifr
.ifr_hwaddr
.sa_family
= tun
->dev
->type
;
2199 if (copy_to_user(argp
, &ifr
, ifreq_len
))
2204 /* Set hw address */
2205 tun_debug(KERN_DEBUG
, tun
, "set hw address: %pM\n",
2206 ifr
.ifr_hwaddr
.sa_data
);
2208 ret
= dev_set_mac_address(tun
->dev
, &ifr
.ifr_hwaddr
);
2212 sndbuf
= tfile
->socket
.sk
->sk_sndbuf
;
2213 if (copy_to_user(argp
, &sndbuf
, sizeof(sndbuf
)))
2218 if (copy_from_user(&sndbuf
, argp
, sizeof(sndbuf
))) {
2227 tun
->sndbuf
= sndbuf
;
2228 tun_set_sndbuf(tun
);
2231 case TUNGETVNETHDRSZ
:
2232 vnet_hdr_sz
= tun
->vnet_hdr_sz
;
2233 if (copy_to_user(argp
, &vnet_hdr_sz
, sizeof(vnet_hdr_sz
)))
2237 case TUNSETVNETHDRSZ
:
2238 if (copy_from_user(&vnet_hdr_sz
, argp
, sizeof(vnet_hdr_sz
))) {
2242 if (vnet_hdr_sz
< (int)sizeof(struct virtio_net_hdr
)) {
2247 tun
->vnet_hdr_sz
= vnet_hdr_sz
;
2251 le
= !!(tun
->flags
& TUN_VNET_LE
);
2252 if (put_user(le
, (int __user
*)argp
))
2257 if (get_user(le
, (int __user
*)argp
)) {
2262 tun
->flags
|= TUN_VNET_LE
;
2264 tun
->flags
&= ~TUN_VNET_LE
;
2268 ret
= tun_get_vnet_be(tun
, argp
);
2272 ret
= tun_set_vnet_be(tun
, argp
);
2275 case TUNATTACHFILTER
:
2276 /* Can be set only for TAPs */
2278 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2281 if (copy_from_user(&tun
->fprog
, argp
, sizeof(tun
->fprog
)))
2284 ret
= tun_attach_filter(tun
);
2287 case TUNDETACHFILTER
:
2288 /* Can be set only for TAPs */
2290 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2293 tun_detach_filter(tun
, tun
->numqueues
);
2298 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2301 if (copy_to_user(argp
, &tun
->fprog
, sizeof(tun
->fprog
)))
2318 static long tun_chr_ioctl(struct file
*file
,
2319 unsigned int cmd
, unsigned long arg
)
2321 return __tun_chr_ioctl(file
, cmd
, arg
, sizeof (struct ifreq
));
2324 #ifdef CONFIG_COMPAT
2325 static long tun_chr_compat_ioctl(struct file
*file
,
2326 unsigned int cmd
, unsigned long arg
)
2331 case TUNSETTXFILTER
:
2336 arg
= (unsigned long)compat_ptr(arg
);
2339 arg
= (compat_ulong_t
)arg
;
2344 * compat_ifreq is shorter than ifreq, so we must not access beyond
2345 * the end of that structure. All fields that are used in this
2346 * driver are compatible though, we don't need to convert the
2349 return __tun_chr_ioctl(file
, cmd
, arg
, sizeof(struct compat_ifreq
));
2351 #endif /* CONFIG_COMPAT */
2353 static int tun_chr_fasync(int fd
, struct file
*file
, int on
)
2355 struct tun_file
*tfile
= file
->private_data
;
2358 if ((ret
= fasync_helper(fd
, file
, on
, &tfile
->fasync
)) < 0)
2362 __f_setown(file
, task_pid(current
), PIDTYPE_PID
, 0);
2363 tfile
->flags
|= TUN_FASYNC
;
2365 tfile
->flags
&= ~TUN_FASYNC
;
2371 static int tun_chr_open(struct inode
*inode
, struct file
* file
)
2373 struct net
*net
= current
->nsproxy
->net_ns
;
2374 struct tun_file
*tfile
;
2376 DBG1(KERN_INFO
, "tunX: tun_chr_open\n");
2378 tfile
= (struct tun_file
*)sk_alloc(net
, AF_UNSPEC
, GFP_KERNEL
,
2382 RCU_INIT_POINTER(tfile
->tun
, NULL
);
2386 init_waitqueue_head(&tfile
->wq
.wait
);
2387 RCU_INIT_POINTER(tfile
->socket
.wq
, &tfile
->wq
);
2389 tfile
->socket
.file
= file
;
2390 tfile
->socket
.ops
= &tun_socket_ops
;
2392 sock_init_data(&tfile
->socket
, &tfile
->sk
);
2394 tfile
->sk
.sk_write_space
= tun_sock_write_space
;
2395 tfile
->sk
.sk_sndbuf
= INT_MAX
;
2397 file
->private_data
= tfile
;
2398 INIT_LIST_HEAD(&tfile
->next
);
2400 sock_set_flag(&tfile
->sk
, SOCK_ZEROCOPY
);
2405 static int tun_chr_close(struct inode
*inode
, struct file
*file
)
2407 struct tun_file
*tfile
= file
->private_data
;
2409 tun_detach(tfile
, true);
2414 #ifdef CONFIG_PROC_FS
2415 static void tun_chr_show_fdinfo(struct seq_file
*m
, struct file
*f
)
2417 struct tun_struct
*tun
;
2420 memset(&ifr
, 0, sizeof(ifr
));
2425 tun_get_iff(current
->nsproxy
->net_ns
, tun
, &ifr
);
2431 seq_printf(m
, "iff:\t%s\n", ifr
.ifr_name
);
2435 static const struct file_operations tun_fops
= {
2436 .owner
= THIS_MODULE
,
2437 .llseek
= no_llseek
,
2438 .read_iter
= tun_chr_read_iter
,
2439 .write_iter
= tun_chr_write_iter
,
2440 .poll
= tun_chr_poll
,
2441 .unlocked_ioctl
= tun_chr_ioctl
,
2442 #ifdef CONFIG_COMPAT
2443 .compat_ioctl
= tun_chr_compat_ioctl
,
2445 .open
= tun_chr_open
,
2446 .release
= tun_chr_close
,
2447 .fasync
= tun_chr_fasync
,
2448 #ifdef CONFIG_PROC_FS
2449 .show_fdinfo
= tun_chr_show_fdinfo
,
2453 static struct miscdevice tun_miscdev
= {
2456 .nodename
= "net/tun",
2460 /* ethtool interface */
2462 static int tun_get_link_ksettings(struct net_device
*dev
,
2463 struct ethtool_link_ksettings
*cmd
)
2465 ethtool_link_ksettings_zero_link_mode(cmd
, supported
);
2466 ethtool_link_ksettings_zero_link_mode(cmd
, advertising
);
2467 cmd
->base
.speed
= SPEED_10
;
2468 cmd
->base
.duplex
= DUPLEX_FULL
;
2469 cmd
->base
.port
= PORT_TP
;
2470 cmd
->base
.phy_address
= 0;
2471 cmd
->base
.autoneg
= AUTONEG_DISABLE
;
2475 static void tun_get_drvinfo(struct net_device
*dev
, struct ethtool_drvinfo
*info
)
2477 struct tun_struct
*tun
= netdev_priv(dev
);
2479 strlcpy(info
->driver
, DRV_NAME
, sizeof(info
->driver
));
2480 strlcpy(info
->version
, DRV_VERSION
, sizeof(info
->version
));
2482 switch (tun
->flags
& TUN_TYPE_MASK
) {
2484 strlcpy(info
->bus_info
, "tun", sizeof(info
->bus_info
));
2487 strlcpy(info
->bus_info
, "tap", sizeof(info
->bus_info
));
2492 static u32
tun_get_msglevel(struct net_device
*dev
)
2495 struct tun_struct
*tun
= netdev_priv(dev
);
2502 static void tun_set_msglevel(struct net_device
*dev
, u32 value
)
2505 struct tun_struct
*tun
= netdev_priv(dev
);
2510 static int tun_get_coalesce(struct net_device
*dev
,
2511 struct ethtool_coalesce
*ec
)
2513 struct tun_struct
*tun
= netdev_priv(dev
);
2515 ec
->rx_max_coalesced_frames
= tun
->rx_batched
;
2520 static int tun_set_coalesce(struct net_device
*dev
,
2521 struct ethtool_coalesce
*ec
)
2523 struct tun_struct
*tun
= netdev_priv(dev
);
2525 if (ec
->rx_max_coalesced_frames
> NAPI_POLL_WEIGHT
)
2526 tun
->rx_batched
= NAPI_POLL_WEIGHT
;
2528 tun
->rx_batched
= ec
->rx_max_coalesced_frames
;
2533 static const struct ethtool_ops tun_ethtool_ops
= {
2534 .get_drvinfo
= tun_get_drvinfo
,
2535 .get_msglevel
= tun_get_msglevel
,
2536 .set_msglevel
= tun_set_msglevel
,
2537 .get_link
= ethtool_op_get_link
,
2538 .get_ts_info
= ethtool_op_get_ts_info
,
2539 .get_coalesce
= tun_get_coalesce
,
2540 .set_coalesce
= tun_set_coalesce
,
2541 .get_link_ksettings
= tun_get_link_ksettings
,
2544 static int tun_queue_resize(struct tun_struct
*tun
)
2546 struct net_device
*dev
= tun
->dev
;
2547 struct tun_file
*tfile
;
2548 struct skb_array
**arrays
;
2549 int n
= tun
->numqueues
+ tun
->numdisabled
;
2552 arrays
= kmalloc(sizeof *arrays
* n
, GFP_KERNEL
);
2556 for (i
= 0; i
< tun
->numqueues
; i
++) {
2557 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
2558 arrays
[i
] = &tfile
->tx_array
;
2560 list_for_each_entry(tfile
, &tun
->disabled
, next
)
2561 arrays
[i
++] = &tfile
->tx_array
;
2563 ret
= skb_array_resize_multiple(arrays
, n
,
2564 dev
->tx_queue_len
, GFP_KERNEL
);
2570 static int tun_device_event(struct notifier_block
*unused
,
2571 unsigned long event
, void *ptr
)
2573 struct net_device
*dev
= netdev_notifier_info_to_dev(ptr
);
2574 struct tun_struct
*tun
= netdev_priv(dev
);
2576 if (dev
->rtnl_link_ops
!= &tun_link_ops
)
2580 case NETDEV_CHANGE_TX_QUEUE_LEN
:
2581 if (tun_queue_resize(tun
))
2591 static struct notifier_block tun_notifier_block __read_mostly
= {
2592 .notifier_call
= tun_device_event
,
2595 static int __init
tun_init(void)
2599 pr_info("%s, %s\n", DRV_DESCRIPTION
, DRV_VERSION
);
2601 ret
= rtnl_link_register(&tun_link_ops
);
2603 pr_err("Can't register link_ops\n");
2607 ret
= misc_register(&tun_miscdev
);
2609 pr_err("Can't register misc device %d\n", TUN_MINOR
);
2613 ret
= register_netdevice_notifier(&tun_notifier_block
);
2615 pr_err("Can't register netdevice notifier\n");
2622 misc_deregister(&tun_miscdev
);
2624 rtnl_link_unregister(&tun_link_ops
);
2629 static void tun_cleanup(void)
2631 misc_deregister(&tun_miscdev
);
2632 rtnl_link_unregister(&tun_link_ops
);
2633 unregister_netdevice_notifier(&tun_notifier_block
);
2636 /* Get an underlying socket object from tun file. Returns error unless file is
2637 * attached to a device. The returned object works like a packet socket, it
2638 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
2639 * holding a reference to the file for as long as the socket is in use. */
2640 struct socket
*tun_get_socket(struct file
*file
)
2642 struct tun_file
*tfile
;
2643 if (file
->f_op
!= &tun_fops
)
2644 return ERR_PTR(-EINVAL
);
2645 tfile
= file
->private_data
;
2647 return ERR_PTR(-EBADFD
);
2648 return &tfile
->socket
;
2650 EXPORT_SYMBOL_GPL(tun_get_socket
);
2652 struct skb_array
*tun_get_skb_array(struct file
*file
)
2654 struct tun_file
*tfile
;
2656 if (file
->f_op
!= &tun_fops
)
2657 return ERR_PTR(-EINVAL
);
2658 tfile
= file
->private_data
;
2660 return ERR_PTR(-EBADFD
);
2661 return &tfile
->tx_array
;
2663 EXPORT_SYMBOL_GPL(tun_get_skb_array
);
2665 module_init(tun_init
);
2666 module_exit(tun_cleanup
);
2667 MODULE_DESCRIPTION(DRV_DESCRIPTION
);
2668 MODULE_AUTHOR(DRV_COPYRIGHT
);
2669 MODULE_LICENSE("GPL");
2670 MODULE_ALIAS_MISCDEV(TUN_MINOR
);
2671 MODULE_ALIAS("devname:net/tun");