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>
76 #include <linux/bpf.h>
77 #include <linux/bpf_trace.h>
78 #include <linux/mutex.h>
80 #include <linux/uaccess.h>
82 /* Uncomment to enable debugging */
83 /* #define TUN_DEBUG 1 */
88 #define tun_debug(level, tun, fmt, args...) \
91 netdev_printk(level, tun->dev, fmt, ##args); \
93 #define DBG1(level, fmt, args...) \
96 printk(level fmt, ##args); \
99 #define tun_debug(level, tun, fmt, args...) \
102 netdev_printk(level, tun->dev, fmt, ##args); \
104 #define DBG1(level, fmt, args...) \
107 printk(level fmt, ##args); \
111 #define TUN_HEADROOM 256
112 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
114 /* TUN device flags */
116 /* IFF_ATTACH_QUEUE is never stored in device flags,
117 * overload it to mean fasync when stored there.
119 #define TUN_FASYNC IFF_ATTACH_QUEUE
120 /* High bits in flags field are unused. */
121 #define TUN_VNET_LE 0x80000000
122 #define TUN_VNET_BE 0x40000000
124 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
125 IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
127 #define GOODCOPY_LEN 128
129 #define FLT_EXACT_COUNT 8
131 unsigned int count
; /* Number of addrs. Zero means disabled */
132 u32 mask
[2]; /* Mask of the hashed addrs */
133 unsigned char addr
[FLT_EXACT_COUNT
][ETH_ALEN
];
136 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
137 * to max number of VCPUs in guest. */
138 #define MAX_TAP_QUEUES 256
139 #define MAX_TAP_FLOWS 4096
141 #define TUN_FLOW_EXPIRE (3 * HZ)
143 struct tun_pcpu_stats
{
148 struct u64_stats_sync syncp
;
154 /* A tun_file connects an open character device to a tuntap netdevice. It
155 * also contains all socket related structures (except sock_fprog and tap_filter)
156 * to serve as one transmit queue for tuntap device. The sock_fprog and
157 * tap_filter were kept in tun_struct since they were used for filtering for the
158 * netdevice not for a specific queue (at least I didn't see the requirement for
162 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
163 * other can only be read while rcu_read_lock or rtnl_lock is held.
167 struct socket socket
;
169 struct tun_struct __rcu
*tun
;
170 struct fasync_struct
*fasync
;
171 /* only used for fasnyc */
175 unsigned int ifindex
;
177 struct napi_struct napi
;
178 struct mutex napi_mutex
; /* Protects access to the above napi */
179 struct list_head next
;
180 struct tun_struct
*detached
;
181 struct skb_array tx_array
;
184 struct tun_flow_entry
{
185 struct hlist_node hash_link
;
187 struct tun_struct
*tun
;
192 unsigned long updated
;
195 #define TUN_NUM_FLOW_ENTRIES 1024
197 /* Since the socket were moved to tun_file, to preserve the behavior of persist
198 * device, socket filter, sndbuf and vnet header size were restore when the
199 * file were attached to a persist device.
202 struct tun_file __rcu
*tfiles
[MAX_TAP_QUEUES
];
203 unsigned int numqueues
;
208 struct net_device
*dev
;
209 netdev_features_t set_features
;
210 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
216 struct tap_filter txflt
;
217 struct sock_fprog fprog
;
218 /* protected by rtnl lock */
219 bool filter_attached
;
224 struct hlist_head flows
[TUN_NUM_FLOW_ENTRIES
];
225 struct timer_list flow_gc_timer
;
226 unsigned long ageing_time
;
227 unsigned int numdisabled
;
228 struct list_head disabled
;
232 struct tun_pcpu_stats __percpu
*pcpu_stats
;
233 struct bpf_prog __rcu
*xdp_prog
;
236 static int tun_napi_receive(struct napi_struct
*napi
, int budget
)
238 struct tun_file
*tfile
= container_of(napi
, struct tun_file
, napi
);
239 struct sk_buff_head
*queue
= &tfile
->sk
.sk_write_queue
;
240 struct sk_buff_head process_queue
;
244 __skb_queue_head_init(&process_queue
);
246 spin_lock(&queue
->lock
);
247 skb_queue_splice_tail_init(queue
, &process_queue
);
248 spin_unlock(&queue
->lock
);
250 while (received
< budget
&& (skb
= __skb_dequeue(&process_queue
))) {
251 napi_gro_receive(napi
, skb
);
255 if (!skb_queue_empty(&process_queue
)) {
256 spin_lock(&queue
->lock
);
257 skb_queue_splice(&process_queue
, queue
);
258 spin_unlock(&queue
->lock
);
264 static int tun_napi_poll(struct napi_struct
*napi
, int budget
)
266 unsigned int received
;
268 received
= tun_napi_receive(napi
, budget
);
270 if (received
< budget
)
271 napi_complete_done(napi
, received
);
276 static void tun_napi_init(struct tun_struct
*tun
, struct tun_file
*tfile
,
280 netif_napi_add(tun
->dev
, &tfile
->napi
, tun_napi_poll
,
282 napi_enable(&tfile
->napi
);
283 mutex_init(&tfile
->napi_mutex
);
287 static void tun_napi_disable(struct tun_struct
*tun
, struct tun_file
*tfile
)
289 if (tun
->flags
& IFF_NAPI
)
290 napi_disable(&tfile
->napi
);
293 static void tun_napi_del(struct tun_struct
*tun
, struct tun_file
*tfile
)
295 if (tun
->flags
& IFF_NAPI
)
296 netif_napi_del(&tfile
->napi
);
299 static bool tun_napi_frags_enabled(const struct tun_struct
*tun
)
301 return READ_ONCE(tun
->flags
) & IFF_NAPI_FRAGS
;
304 #ifdef CONFIG_TUN_VNET_CROSS_LE
305 static inline bool tun_legacy_is_little_endian(struct tun_struct
*tun
)
307 return tun
->flags
& TUN_VNET_BE
? false :
308 virtio_legacy_is_little_endian();
311 static long tun_get_vnet_be(struct tun_struct
*tun
, int __user
*argp
)
313 int be
= !!(tun
->flags
& TUN_VNET_BE
);
315 if (put_user(be
, argp
))
321 static long tun_set_vnet_be(struct tun_struct
*tun
, int __user
*argp
)
325 if (get_user(be
, argp
))
329 tun
->flags
|= TUN_VNET_BE
;
331 tun
->flags
&= ~TUN_VNET_BE
;
336 static inline bool tun_legacy_is_little_endian(struct tun_struct
*tun
)
338 return virtio_legacy_is_little_endian();
341 static long tun_get_vnet_be(struct tun_struct
*tun
, int __user
*argp
)
346 static long tun_set_vnet_be(struct tun_struct
*tun
, int __user
*argp
)
350 #endif /* CONFIG_TUN_VNET_CROSS_LE */
352 static inline bool tun_is_little_endian(struct tun_struct
*tun
)
354 return tun
->flags
& TUN_VNET_LE
||
355 tun_legacy_is_little_endian(tun
);
358 static inline u16
tun16_to_cpu(struct tun_struct
*tun
, __virtio16 val
)
360 return __virtio16_to_cpu(tun_is_little_endian(tun
), val
);
363 static inline __virtio16
cpu_to_tun16(struct tun_struct
*tun
, u16 val
)
365 return __cpu_to_virtio16(tun_is_little_endian(tun
), val
);
368 static inline u32
tun_hashfn(u32 rxhash
)
370 return rxhash
& 0x3ff;
373 static struct tun_flow_entry
*tun_flow_find(struct hlist_head
*head
, u32 rxhash
)
375 struct tun_flow_entry
*e
;
377 hlist_for_each_entry_rcu(e
, head
, hash_link
) {
378 if (e
->rxhash
== rxhash
)
384 static struct tun_flow_entry
*tun_flow_create(struct tun_struct
*tun
,
385 struct hlist_head
*head
,
386 u32 rxhash
, u16 queue_index
)
388 struct tun_flow_entry
*e
= kmalloc(sizeof(*e
), GFP_ATOMIC
);
391 tun_debug(KERN_INFO
, tun
, "create flow: hash %u index %u\n",
392 rxhash
, queue_index
);
393 e
->updated
= jiffies
;
396 e
->queue_index
= queue_index
;
398 hlist_add_head_rcu(&e
->hash_link
, head
);
404 static void tun_flow_delete(struct tun_struct
*tun
, struct tun_flow_entry
*e
)
406 tun_debug(KERN_INFO
, tun
, "delete flow: hash %u index %u\n",
407 e
->rxhash
, e
->queue_index
);
408 hlist_del_rcu(&e
->hash_link
);
413 static void tun_flow_flush(struct tun_struct
*tun
)
417 spin_lock_bh(&tun
->lock
);
418 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
419 struct tun_flow_entry
*e
;
420 struct hlist_node
*n
;
422 hlist_for_each_entry_safe(e
, n
, &tun
->flows
[i
], hash_link
)
423 tun_flow_delete(tun
, e
);
425 spin_unlock_bh(&tun
->lock
);
428 static void tun_flow_delete_by_queue(struct tun_struct
*tun
, u16 queue_index
)
432 spin_lock_bh(&tun
->lock
);
433 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
434 struct tun_flow_entry
*e
;
435 struct hlist_node
*n
;
437 hlist_for_each_entry_safe(e
, n
, &tun
->flows
[i
], hash_link
) {
438 if (e
->queue_index
== queue_index
)
439 tun_flow_delete(tun
, e
);
442 spin_unlock_bh(&tun
->lock
);
445 static void tun_flow_cleanup(unsigned long data
)
447 struct tun_struct
*tun
= (struct tun_struct
*)data
;
448 unsigned long delay
= tun
->ageing_time
;
449 unsigned long next_timer
= jiffies
+ delay
;
450 unsigned long count
= 0;
453 tun_debug(KERN_INFO
, tun
, "tun_flow_cleanup\n");
455 spin_lock_bh(&tun
->lock
);
456 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
457 struct tun_flow_entry
*e
;
458 struct hlist_node
*n
;
460 hlist_for_each_entry_safe(e
, n
, &tun
->flows
[i
], hash_link
) {
461 unsigned long this_timer
;
463 this_timer
= e
->updated
+ delay
;
464 if (time_before_eq(this_timer
, jiffies
))
465 tun_flow_delete(tun
, e
);
466 else if (time_before(this_timer
, next_timer
))
467 next_timer
= this_timer
;
472 mod_timer(&tun
->flow_gc_timer
, round_jiffies_up(next_timer
));
473 spin_unlock_bh(&tun
->lock
);
476 static void tun_flow_update(struct tun_struct
*tun
, u32 rxhash
,
477 struct tun_file
*tfile
)
479 struct hlist_head
*head
;
480 struct tun_flow_entry
*e
;
481 unsigned long delay
= tun
->ageing_time
;
482 u16 queue_index
= tfile
->queue_index
;
487 head
= &tun
->flows
[tun_hashfn(rxhash
)];
491 /* We may get a very small possibility of OOO during switching, not
492 * worth to optimize.*/
493 if (tun
->numqueues
== 1 || tfile
->detached
)
496 e
= tun_flow_find(head
, rxhash
);
498 /* TODO: keep queueing to old queue until it's empty? */
499 e
->queue_index
= queue_index
;
500 e
->updated
= jiffies
;
501 sock_rps_record_flow_hash(e
->rps_rxhash
);
503 spin_lock_bh(&tun
->lock
);
504 if (!tun_flow_find(head
, rxhash
) &&
505 tun
->flow_count
< MAX_TAP_FLOWS
)
506 tun_flow_create(tun
, head
, rxhash
, queue_index
);
508 if (!timer_pending(&tun
->flow_gc_timer
))
509 mod_timer(&tun
->flow_gc_timer
,
510 round_jiffies_up(jiffies
+ delay
));
511 spin_unlock_bh(&tun
->lock
);
519 * Save the hash received in the stack receive path and update the
520 * flow_hash table accordingly.
522 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry
*e
, u32 hash
)
524 if (unlikely(e
->rps_rxhash
!= hash
))
525 e
->rps_rxhash
= hash
;
528 /* We try to identify a flow through its rxhash first. The reason that
529 * we do not check rxq no. is because some cards(e.g 82599), chooses
530 * the rxq based on the txq where the last packet of the flow comes. As
531 * the userspace application move between processors, we may get a
532 * different rxq no. here. If we could not get rxhash, then we would
533 * hope the rxq no. may help here.
535 static u16
tun_select_queue(struct net_device
*dev
, struct sk_buff
*skb
,
536 void *accel_priv
, select_queue_fallback_t fallback
)
538 struct tun_struct
*tun
= netdev_priv(dev
);
539 struct tun_flow_entry
*e
;
544 numqueues
= ACCESS_ONCE(tun
->numqueues
);
546 txq
= __skb_get_hash_symmetric(skb
);
548 e
= tun_flow_find(&tun
->flows
[tun_hashfn(txq
)], txq
);
550 tun_flow_save_rps_rxhash(e
, txq
);
551 txq
= e
->queue_index
;
553 /* use multiply and shift instead of expensive divide */
554 txq
= ((u64
)txq
* numqueues
) >> 32;
555 } else if (likely(skb_rx_queue_recorded(skb
))) {
556 txq
= skb_get_rx_queue(skb
);
557 while (unlikely(txq
>= numqueues
))
565 static inline bool tun_not_capable(struct tun_struct
*tun
)
567 const struct cred
*cred
= current_cred();
568 struct net
*net
= dev_net(tun
->dev
);
570 return ((uid_valid(tun
->owner
) && !uid_eq(cred
->euid
, tun
->owner
)) ||
571 (gid_valid(tun
->group
) && !in_egroup_p(tun
->group
))) &&
572 !ns_capable(net
->user_ns
, CAP_NET_ADMIN
);
575 static void tun_set_real_num_queues(struct tun_struct
*tun
)
577 netif_set_real_num_tx_queues(tun
->dev
, tun
->numqueues
);
578 netif_set_real_num_rx_queues(tun
->dev
, tun
->numqueues
);
581 static void tun_disable_queue(struct tun_struct
*tun
, struct tun_file
*tfile
)
583 tfile
->detached
= tun
;
584 list_add_tail(&tfile
->next
, &tun
->disabled
);
588 static struct tun_struct
*tun_enable_queue(struct tun_file
*tfile
)
590 struct tun_struct
*tun
= tfile
->detached
;
592 tfile
->detached
= NULL
;
593 list_del_init(&tfile
->next
);
598 static void tun_queue_purge(struct tun_file
*tfile
)
602 while ((skb
= skb_array_consume(&tfile
->tx_array
)) != NULL
)
605 skb_queue_purge(&tfile
->sk
.sk_write_queue
);
606 skb_queue_purge(&tfile
->sk
.sk_error_queue
);
609 static void __tun_detach(struct tun_file
*tfile
, bool clean
)
611 struct tun_file
*ntfile
;
612 struct tun_struct
*tun
;
614 tun
= rtnl_dereference(tfile
->tun
);
617 tun_napi_disable(tun
, tfile
);
618 tun_napi_del(tun
, tfile
);
621 if (tun
&& !tfile
->detached
) {
622 u16 index
= tfile
->queue_index
;
623 BUG_ON(index
>= tun
->numqueues
);
625 rcu_assign_pointer(tun
->tfiles
[index
],
626 tun
->tfiles
[tun
->numqueues
- 1]);
627 ntfile
= rtnl_dereference(tun
->tfiles
[index
]);
628 ntfile
->queue_index
= index
;
632 RCU_INIT_POINTER(tfile
->tun
, NULL
);
633 sock_put(&tfile
->sk
);
635 tun_disable_queue(tun
, tfile
);
638 tun_flow_delete_by_queue(tun
, tun
->numqueues
+ 1);
639 /* Drop read queue */
640 tun_queue_purge(tfile
);
641 tun_set_real_num_queues(tun
);
642 } else if (tfile
->detached
&& clean
) {
643 tun
= tun_enable_queue(tfile
);
644 sock_put(&tfile
->sk
);
648 if (tun
&& tun
->numqueues
== 0 && tun
->numdisabled
== 0) {
649 netif_carrier_off(tun
->dev
);
651 if (!(tun
->flags
& IFF_PERSIST
) &&
652 tun
->dev
->reg_state
== NETREG_REGISTERED
)
653 unregister_netdevice(tun
->dev
);
656 skb_array_cleanup(&tfile
->tx_array
);
657 sock_put(&tfile
->sk
);
661 static void tun_detach(struct tun_file
*tfile
, bool clean
)
664 __tun_detach(tfile
, clean
);
668 static void tun_detach_all(struct net_device
*dev
)
670 struct tun_struct
*tun
= netdev_priv(dev
);
671 struct bpf_prog
*xdp_prog
= rtnl_dereference(tun
->xdp_prog
);
672 struct tun_file
*tfile
, *tmp
;
673 int i
, n
= tun
->numqueues
;
675 for (i
= 0; i
< n
; i
++) {
676 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
678 tun_napi_disable(tun
, tfile
);
679 tfile
->socket
.sk
->sk_shutdown
= RCV_SHUTDOWN
;
680 tfile
->socket
.sk
->sk_data_ready(tfile
->socket
.sk
);
681 RCU_INIT_POINTER(tfile
->tun
, NULL
);
684 list_for_each_entry(tfile
, &tun
->disabled
, next
) {
685 tfile
->socket
.sk
->sk_shutdown
= RCV_SHUTDOWN
;
686 tfile
->socket
.sk
->sk_data_ready(tfile
->socket
.sk
);
687 RCU_INIT_POINTER(tfile
->tun
, NULL
);
689 BUG_ON(tun
->numqueues
!= 0);
692 for (i
= 0; i
< n
; i
++) {
693 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
694 tun_napi_del(tun
, tfile
);
695 /* Drop read queue */
696 tun_queue_purge(tfile
);
697 sock_put(&tfile
->sk
);
699 list_for_each_entry_safe(tfile
, tmp
, &tun
->disabled
, next
) {
700 tun_enable_queue(tfile
);
701 tun_queue_purge(tfile
);
702 sock_put(&tfile
->sk
);
704 BUG_ON(tun
->numdisabled
!= 0);
707 bpf_prog_put(xdp_prog
);
709 if (tun
->flags
& IFF_PERSIST
)
710 module_put(THIS_MODULE
);
713 static int tun_attach(struct tun_struct
*tun
, struct file
*file
,
714 bool skip_filter
, bool napi
)
716 struct tun_file
*tfile
= file
->private_data
;
717 struct net_device
*dev
= tun
->dev
;
720 err
= security_tun_dev_attach(tfile
->socket
.sk
, tun
->security
);
725 if (rtnl_dereference(tfile
->tun
) && !tfile
->detached
)
729 if (!(tun
->flags
& IFF_MULTI_QUEUE
) && tun
->numqueues
== 1)
733 if (!tfile
->detached
&&
734 tun
->numqueues
+ tun
->numdisabled
== MAX_TAP_QUEUES
)
739 /* Re-attach the filter to persist device */
740 if (!skip_filter
&& (tun
->filter_attached
== true)) {
741 lock_sock(tfile
->socket
.sk
);
742 err
= sk_attach_filter(&tun
->fprog
, tfile
->socket
.sk
);
743 release_sock(tfile
->socket
.sk
);
748 if (!tfile
->detached
&&
749 skb_array_init(&tfile
->tx_array
, dev
->tx_queue_len
, GFP_KERNEL
)) {
754 tfile
->queue_index
= tun
->numqueues
;
755 tfile
->socket
.sk
->sk_shutdown
&= ~RCV_SHUTDOWN
;
756 rcu_assign_pointer(tfile
->tun
, tun
);
757 rcu_assign_pointer(tun
->tfiles
[tun
->numqueues
], tfile
);
760 if (tfile
->detached
) {
761 tun_enable_queue(tfile
);
763 sock_hold(&tfile
->sk
);
764 tun_napi_init(tun
, tfile
, napi
);
767 tun_set_real_num_queues(tun
);
769 /* device is allowed to go away first, so no need to hold extra
777 static struct tun_struct
*tun_get(struct tun_file
*tfile
)
779 struct tun_struct
*tun
;
782 tun
= rcu_dereference(tfile
->tun
);
790 static void tun_put(struct tun_struct
*tun
)
796 static void addr_hash_set(u32
*mask
, const u8
*addr
)
798 int n
= ether_crc(ETH_ALEN
, addr
) >> 26;
799 mask
[n
>> 5] |= (1 << (n
& 31));
802 static unsigned int addr_hash_test(const u32
*mask
, const u8
*addr
)
804 int n
= ether_crc(ETH_ALEN
, addr
) >> 26;
805 return mask
[n
>> 5] & (1 << (n
& 31));
808 static int update_filter(struct tap_filter
*filter
, void __user
*arg
)
810 struct { u8 u
[ETH_ALEN
]; } *addr
;
811 struct tun_filter uf
;
812 int err
, alen
, n
, nexact
;
814 if (copy_from_user(&uf
, arg
, sizeof(uf
)))
823 alen
= ETH_ALEN
* uf
.count
;
824 addr
= memdup_user(arg
+ sizeof(uf
), alen
);
826 return PTR_ERR(addr
);
828 /* The filter is updated without holding any locks. Which is
829 * perfectly safe. We disable it first and in the worst
830 * case we'll accept a few undesired packets. */
834 /* Use first set of addresses as an exact filter */
835 for (n
= 0; n
< uf
.count
&& n
< FLT_EXACT_COUNT
; n
++)
836 memcpy(filter
->addr
[n
], addr
[n
].u
, ETH_ALEN
);
840 /* Remaining multicast addresses are hashed,
841 * unicast will leave the filter disabled. */
842 memset(filter
->mask
, 0, sizeof(filter
->mask
));
843 for (; n
< uf
.count
; n
++) {
844 if (!is_multicast_ether_addr(addr
[n
].u
)) {
845 err
= 0; /* no filter */
848 addr_hash_set(filter
->mask
, addr
[n
].u
);
851 /* For ALLMULTI just set the mask to all ones.
852 * This overrides the mask populated above. */
853 if ((uf
.flags
& TUN_FLT_ALLMULTI
))
854 memset(filter
->mask
, ~0, sizeof(filter
->mask
));
856 /* Now enable the filter */
858 filter
->count
= nexact
;
860 /* Return the number of exact filters */
867 /* Returns: 0 - drop, !=0 - accept */
868 static int run_filter(struct tap_filter
*filter
, const struct sk_buff
*skb
)
870 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
872 struct ethhdr
*eh
= (struct ethhdr
*) skb
->data
;
876 for (i
= 0; i
< filter
->count
; i
++)
877 if (ether_addr_equal(eh
->h_dest
, filter
->addr
[i
]))
880 /* Inexact match (multicast only) */
881 if (is_multicast_ether_addr(eh
->h_dest
))
882 return addr_hash_test(filter
->mask
, eh
->h_dest
);
888 * Checks whether the packet is accepted or not.
889 * Returns: 0 - drop, !=0 - accept
891 static int check_filter(struct tap_filter
*filter
, const struct sk_buff
*skb
)
896 return run_filter(filter
, skb
);
899 /* Network device part of the driver */
901 static const struct ethtool_ops tun_ethtool_ops
;
903 /* Net device detach from fd. */
904 static void tun_net_uninit(struct net_device
*dev
)
909 /* Net device open. */
910 static int tun_net_open(struct net_device
*dev
)
912 struct tun_struct
*tun
= netdev_priv(dev
);
915 netif_tx_start_all_queues(dev
);
917 for (i
= 0; i
< tun
->numqueues
; i
++) {
918 struct tun_file
*tfile
;
920 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
921 tfile
->socket
.sk
->sk_write_space(tfile
->socket
.sk
);
927 /* Net device close. */
928 static int tun_net_close(struct net_device
*dev
)
930 netif_tx_stop_all_queues(dev
);
934 /* Net device start xmit */
935 static netdev_tx_t
tun_net_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
937 struct tun_struct
*tun
= netdev_priv(dev
);
938 int txq
= skb
->queue_mapping
;
939 struct tun_file
*tfile
;
943 tfile
= rcu_dereference(tun
->tfiles
[txq
]);
944 numqueues
= ACCESS_ONCE(tun
->numqueues
);
946 /* Drop packet if interface is not attached */
947 if (txq
>= numqueues
)
951 if (numqueues
== 1 && static_key_false(&rps_needed
)) {
952 /* Select queue was not called for the skbuff, so we extract the
953 * RPS hash and save it into the flow_table here.
957 rxhash
= __skb_get_hash_symmetric(skb
);
959 struct tun_flow_entry
*e
;
960 e
= tun_flow_find(&tun
->flows
[tun_hashfn(rxhash
)],
963 tun_flow_save_rps_rxhash(e
, rxhash
);
968 tun_debug(KERN_INFO
, tun
, "tun_net_xmit %d\n", skb
->len
);
972 /* Drop if the filter does not like it.
973 * This is a noop if the filter is disabled.
974 * Filter can be enabled only for the TAP devices. */
975 if (!check_filter(&tun
->txflt
, skb
))
978 if (tfile
->socket
.sk
->sk_filter
&&
979 sk_filter(tfile
->socket
.sk
, skb
))
982 if (unlikely(skb_orphan_frags_rx(skb
, GFP_ATOMIC
)))
985 skb_tx_timestamp(skb
);
987 /* Orphan the skb - required as we might hang on to it
988 * for indefinite time.
994 if (skb_array_produce(&tfile
->tx_array
, skb
))
997 /* Notify and wake up reader process */
998 if (tfile
->flags
& TUN_FASYNC
)
999 kill_fasync(&tfile
->fasync
, SIGIO
, POLL_IN
);
1000 tfile
->socket
.sk
->sk_data_ready(tfile
->socket
.sk
);
1003 return NETDEV_TX_OK
;
1006 this_cpu_inc(tun
->pcpu_stats
->tx_dropped
);
1010 return NET_XMIT_DROP
;
1013 static void tun_net_mclist(struct net_device
*dev
)
1016 * This callback is supposed to deal with mc filter in
1017 * _rx_ path and has nothing to do with the _tx_ path.
1018 * In rx path we always accept everything userspace gives us.
1022 static netdev_features_t
tun_net_fix_features(struct net_device
*dev
,
1023 netdev_features_t features
)
1025 struct tun_struct
*tun
= netdev_priv(dev
);
1027 return (features
& tun
->set_features
) | (features
& ~TUN_USER_FEATURES
);
1029 #ifdef CONFIG_NET_POLL_CONTROLLER
1030 static void tun_poll_controller(struct net_device
*dev
)
1033 * Tun only receives frames when:
1034 * 1) the char device endpoint gets data from user space
1035 * 2) the tun socket gets a sendmsg call from user space
1036 * If NAPI is not enabled, since both of those are synchronous
1037 * operations, we are guaranteed never to have pending data when we poll
1038 * for it so there is nothing to do here but return.
1039 * We need this though so netpoll recognizes us as an interface that
1040 * supports polling, which enables bridge devices in virt setups to
1041 * still use netconsole
1042 * If NAPI is enabled, however, we need to schedule polling for all
1043 * queues unless we are using napi_gro_frags(), which we call in
1044 * process context and not in NAPI context.
1046 struct tun_struct
*tun
= netdev_priv(dev
);
1048 if (tun
->flags
& IFF_NAPI
) {
1049 struct tun_file
*tfile
;
1052 if (tun_napi_frags_enabled(tun
))
1056 for (i
= 0; i
< tun
->numqueues
; i
++) {
1057 tfile
= rcu_dereference(tun
->tfiles
[i
]);
1058 napi_schedule(&tfile
->napi
);
1066 static void tun_set_headroom(struct net_device
*dev
, int new_hr
)
1068 struct tun_struct
*tun
= netdev_priv(dev
);
1070 if (new_hr
< NET_SKB_PAD
)
1071 new_hr
= NET_SKB_PAD
;
1073 tun
->align
= new_hr
;
1077 tun_net_get_stats64(struct net_device
*dev
, struct rtnl_link_stats64
*stats
)
1079 u32 rx_dropped
= 0, tx_dropped
= 0, rx_frame_errors
= 0;
1080 struct tun_struct
*tun
= netdev_priv(dev
);
1081 struct tun_pcpu_stats
*p
;
1084 for_each_possible_cpu(i
) {
1085 u64 rxpackets
, rxbytes
, txpackets
, txbytes
;
1088 p
= per_cpu_ptr(tun
->pcpu_stats
, i
);
1090 start
= u64_stats_fetch_begin(&p
->syncp
);
1091 rxpackets
= p
->rx_packets
;
1092 rxbytes
= p
->rx_bytes
;
1093 txpackets
= p
->tx_packets
;
1094 txbytes
= p
->tx_bytes
;
1095 } while (u64_stats_fetch_retry(&p
->syncp
, start
));
1097 stats
->rx_packets
+= rxpackets
;
1098 stats
->rx_bytes
+= rxbytes
;
1099 stats
->tx_packets
+= txpackets
;
1100 stats
->tx_bytes
+= txbytes
;
1103 rx_dropped
+= p
->rx_dropped
;
1104 rx_frame_errors
+= p
->rx_frame_errors
;
1105 tx_dropped
+= p
->tx_dropped
;
1107 stats
->rx_dropped
= rx_dropped
;
1108 stats
->rx_frame_errors
= rx_frame_errors
;
1109 stats
->tx_dropped
= tx_dropped
;
1112 static int tun_xdp_set(struct net_device
*dev
, struct bpf_prog
*prog
,
1113 struct netlink_ext_ack
*extack
)
1115 struct tun_struct
*tun
= netdev_priv(dev
);
1116 struct bpf_prog
*old_prog
;
1118 old_prog
= rtnl_dereference(tun
->xdp_prog
);
1119 rcu_assign_pointer(tun
->xdp_prog
, prog
);
1121 bpf_prog_put(old_prog
);
1126 static u32
tun_xdp_query(struct net_device
*dev
)
1128 struct tun_struct
*tun
= netdev_priv(dev
);
1129 const struct bpf_prog
*xdp_prog
;
1131 xdp_prog
= rtnl_dereference(tun
->xdp_prog
);
1133 return xdp_prog
->aux
->id
;
1138 static int tun_xdp(struct net_device
*dev
, struct netdev_xdp
*xdp
)
1140 switch (xdp
->command
) {
1141 case XDP_SETUP_PROG
:
1142 return tun_xdp_set(dev
, xdp
->prog
, xdp
->extack
);
1143 case XDP_QUERY_PROG
:
1144 xdp
->prog_id
= tun_xdp_query(dev
);
1145 xdp
->prog_attached
= !!xdp
->prog_id
;
1152 static const struct net_device_ops tun_netdev_ops
= {
1153 .ndo_uninit
= tun_net_uninit
,
1154 .ndo_open
= tun_net_open
,
1155 .ndo_stop
= tun_net_close
,
1156 .ndo_start_xmit
= tun_net_xmit
,
1157 .ndo_fix_features
= tun_net_fix_features
,
1158 .ndo_select_queue
= tun_select_queue
,
1159 #ifdef CONFIG_NET_POLL_CONTROLLER
1160 .ndo_poll_controller
= tun_poll_controller
,
1162 .ndo_set_rx_headroom
= tun_set_headroom
,
1163 .ndo_get_stats64
= tun_net_get_stats64
,
1166 static const struct net_device_ops tap_netdev_ops
= {
1167 .ndo_uninit
= tun_net_uninit
,
1168 .ndo_open
= tun_net_open
,
1169 .ndo_stop
= tun_net_close
,
1170 .ndo_start_xmit
= tun_net_xmit
,
1171 .ndo_fix_features
= tun_net_fix_features
,
1172 .ndo_set_rx_mode
= tun_net_mclist
,
1173 .ndo_set_mac_address
= eth_mac_addr
,
1174 .ndo_validate_addr
= eth_validate_addr
,
1175 .ndo_select_queue
= tun_select_queue
,
1176 #ifdef CONFIG_NET_POLL_CONTROLLER
1177 .ndo_poll_controller
= tun_poll_controller
,
1179 .ndo_features_check
= passthru_features_check
,
1180 .ndo_set_rx_headroom
= tun_set_headroom
,
1181 .ndo_get_stats64
= tun_net_get_stats64
,
1185 static void tun_flow_init(struct tun_struct
*tun
)
1189 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++)
1190 INIT_HLIST_HEAD(&tun
->flows
[i
]);
1192 tun
->ageing_time
= TUN_FLOW_EXPIRE
;
1193 setup_timer(&tun
->flow_gc_timer
, tun_flow_cleanup
, (unsigned long)tun
);
1194 mod_timer(&tun
->flow_gc_timer
,
1195 round_jiffies_up(jiffies
+ tun
->ageing_time
));
1198 static void tun_flow_uninit(struct tun_struct
*tun
)
1200 del_timer_sync(&tun
->flow_gc_timer
);
1201 tun_flow_flush(tun
);
1205 #define MAX_MTU 65535
1207 /* Initialize net device. */
1208 static void tun_net_init(struct net_device
*dev
)
1210 struct tun_struct
*tun
= netdev_priv(dev
);
1212 switch (tun
->flags
& TUN_TYPE_MASK
) {
1214 dev
->netdev_ops
= &tun_netdev_ops
;
1216 /* Point-to-Point TUN Device */
1217 dev
->hard_header_len
= 0;
1221 /* Zero header length */
1222 dev
->type
= ARPHRD_NONE
;
1223 dev
->flags
= IFF_POINTOPOINT
| IFF_NOARP
| IFF_MULTICAST
;
1227 dev
->netdev_ops
= &tap_netdev_ops
;
1228 /* Ethernet TAP Device */
1230 dev
->priv_flags
&= ~IFF_TX_SKB_SHARING
;
1231 dev
->priv_flags
|= IFF_LIVE_ADDR_CHANGE
;
1233 eth_hw_addr_random(dev
);
1238 dev
->min_mtu
= MIN_MTU
;
1239 dev
->max_mtu
= MAX_MTU
- dev
->hard_header_len
;
1242 /* Character device part */
1245 static unsigned int tun_chr_poll(struct file
*file
, poll_table
*wait
)
1247 struct tun_file
*tfile
= file
->private_data
;
1248 struct tun_struct
*tun
= tun_get(tfile
);
1250 unsigned int mask
= 0;
1255 sk
= tfile
->socket
.sk
;
1257 tun_debug(KERN_INFO
, tun
, "tun_chr_poll\n");
1259 poll_wait(file
, sk_sleep(sk
), wait
);
1261 if (!skb_array_empty(&tfile
->tx_array
))
1262 mask
|= POLLIN
| POLLRDNORM
;
1264 if (tun
->dev
->flags
& IFF_UP
&&
1265 (sock_writeable(sk
) ||
1266 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
) &&
1267 sock_writeable(sk
))))
1268 mask
|= POLLOUT
| POLLWRNORM
;
1270 if (tun
->dev
->reg_state
!= NETREG_REGISTERED
)
1277 static struct sk_buff
*tun_napi_alloc_frags(struct tun_file
*tfile
,
1279 const struct iov_iter
*it
)
1281 struct sk_buff
*skb
;
1286 if (it
->nr_segs
> MAX_SKB_FRAGS
+ 1)
1287 return ERR_PTR(-ENOMEM
);
1290 skb
= napi_get_frags(&tfile
->napi
);
1293 return ERR_PTR(-ENOMEM
);
1295 linear
= iov_iter_single_seg_count(it
);
1296 err
= __skb_grow(skb
, linear
);
1301 skb
->data_len
= len
- linear
;
1302 skb
->truesize
+= skb
->data_len
;
1304 for (i
= 1; i
< it
->nr_segs
; i
++) {
1305 size_t fragsz
= it
->iov
[i
].iov_len
;
1306 unsigned long offset
;
1310 if (fragsz
== 0 || fragsz
> PAGE_SIZE
) {
1316 data
= napi_alloc_frag(fragsz
);
1323 page
= virt_to_head_page(data
);
1324 offset
= data
- page_address(page
);
1325 skb_fill_page_desc(skb
, i
- 1, page
, offset
, fragsz
);
1330 /* frees skb and all frags allocated with napi_alloc_frag() */
1331 napi_free_frags(&tfile
->napi
);
1332 return ERR_PTR(err
);
1335 /* prepad is the amount to reserve at front. len is length after that.
1336 * linear is a hint as to how much to copy (usually headers). */
1337 static struct sk_buff
*tun_alloc_skb(struct tun_file
*tfile
,
1338 size_t prepad
, size_t len
,
1339 size_t linear
, int noblock
)
1341 struct sock
*sk
= tfile
->socket
.sk
;
1342 struct sk_buff
*skb
;
1345 /* Under a page? Don't bother with paged skb. */
1346 if (prepad
+ len
< PAGE_SIZE
|| !linear
)
1349 skb
= sock_alloc_send_pskb(sk
, prepad
+ linear
, len
- linear
, noblock
,
1352 return ERR_PTR(err
);
1354 skb_reserve(skb
, prepad
);
1355 skb_put(skb
, linear
);
1356 skb
->data_len
= len
- linear
;
1357 skb
->len
+= len
- linear
;
1362 static void tun_rx_batched(struct tun_struct
*tun
, struct tun_file
*tfile
,
1363 struct sk_buff
*skb
, int more
)
1365 struct sk_buff_head
*queue
= &tfile
->sk
.sk_write_queue
;
1366 struct sk_buff_head process_queue
;
1367 u32 rx_batched
= tun
->rx_batched
;
1370 if (!rx_batched
|| (!more
&& skb_queue_empty(queue
))) {
1372 netif_receive_skb(skb
);
1377 spin_lock(&queue
->lock
);
1378 if (!more
|| skb_queue_len(queue
) == rx_batched
) {
1379 __skb_queue_head_init(&process_queue
);
1380 skb_queue_splice_tail_init(queue
, &process_queue
);
1383 __skb_queue_tail(queue
, skb
);
1385 spin_unlock(&queue
->lock
);
1388 struct sk_buff
*nskb
;
1391 while ((nskb
= __skb_dequeue(&process_queue
)))
1392 netif_receive_skb(nskb
);
1393 netif_receive_skb(skb
);
1398 static bool tun_can_build_skb(struct tun_struct
*tun
, struct tun_file
*tfile
,
1399 int len
, int noblock
, bool zerocopy
)
1401 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
1404 if (tfile
->socket
.sk
->sk_sndbuf
!= INT_MAX
)
1413 if (SKB_DATA_ALIGN(len
+ TUN_RX_PAD
) +
1414 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
)) > PAGE_SIZE
)
1420 static struct sk_buff
*tun_build_skb(struct tun_struct
*tun
,
1421 struct tun_file
*tfile
,
1422 struct iov_iter
*from
,
1423 struct virtio_net_hdr
*hdr
,
1424 int len
, int *skb_xdp
)
1426 struct page_frag
*alloc_frag
= ¤t
->task_frag
;
1427 struct sk_buff
*skb
;
1428 struct bpf_prog
*xdp_prog
;
1429 int buflen
= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
1430 unsigned int delta
= 0;
1433 bool xdp_xmit
= false;
1434 int err
, pad
= TUN_RX_PAD
;
1437 xdp_prog
= rcu_dereference(tun
->xdp_prog
);
1439 pad
+= TUN_HEADROOM
;
1440 buflen
+= SKB_DATA_ALIGN(len
+ pad
);
1443 if (unlikely(!skb_page_frag_refill(buflen
, alloc_frag
, GFP_KERNEL
)))
1444 return ERR_PTR(-ENOMEM
);
1446 buf
= (char *)page_address(alloc_frag
->page
) + alloc_frag
->offset
;
1447 copied
= copy_page_from_iter(alloc_frag
->page
,
1448 alloc_frag
->offset
+ pad
,
1451 return ERR_PTR(-EFAULT
);
1453 /* There's a small window that XDP may be set after the check
1454 * of xdp_prog above, this should be rare and for simplicity
1455 * we do XDP on skb in case the headroom is not enough.
1457 if (hdr
->gso_type
|| !xdp_prog
)
1463 xdp_prog
= rcu_dereference(tun
->xdp_prog
);
1464 if (xdp_prog
&& !*skb_xdp
) {
1465 struct xdp_buff xdp
;
1469 xdp
.data_hard_start
= buf
;
1470 xdp
.data
= buf
+ pad
;
1471 xdp_set_data_meta_invalid(&xdp
);
1472 xdp
.data_end
= xdp
.data
+ len
;
1473 orig_data
= xdp
.data
;
1474 act
= bpf_prog_run_xdp(xdp_prog
, &xdp
);
1478 get_page(alloc_frag
->page
);
1479 alloc_frag
->offset
+= buflen
;
1480 err
= xdp_do_redirect(tun
->dev
, &xdp
, xdp_prog
);
1488 delta
= orig_data
- xdp
.data
;
1491 bpf_warn_invalid_xdp_action(act
);
1494 trace_xdp_exception(tun
->dev
, xdp_prog
, act
);
1501 skb
= build_skb(buf
, buflen
);
1504 return ERR_PTR(-ENOMEM
);
1507 skb_reserve(skb
, pad
- delta
);
1508 skb_put(skb
, len
+ delta
);
1509 get_page(alloc_frag
->page
);
1510 alloc_frag
->offset
+= buflen
;
1513 skb
->dev
= tun
->dev
;
1514 generic_xdp_tx(skb
, xdp_prog
);
1524 put_page(alloc_frag
->page
);
1527 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1531 /* Get packet from user space buffer */
1532 static ssize_t
tun_get_user(struct tun_struct
*tun
, struct tun_file
*tfile
,
1533 void *msg_control
, struct iov_iter
*from
,
1534 int noblock
, bool more
)
1536 struct tun_pi pi
= { 0, cpu_to_be16(ETH_P_IP
) };
1537 struct sk_buff
*skb
;
1538 size_t total_len
= iov_iter_count(from
);
1539 size_t len
= total_len
, align
= tun
->align
, linear
;
1540 struct virtio_net_hdr gso
= { 0 };
1541 struct tun_pcpu_stats
*stats
;
1544 bool zerocopy
= false;
1548 bool frags
= tun_napi_frags_enabled(tun
);
1550 if (!(tun
->dev
->flags
& IFF_UP
))
1553 if (!(tun
->flags
& IFF_NO_PI
)) {
1554 if (len
< sizeof(pi
))
1558 if (!copy_from_iter_full(&pi
, sizeof(pi
), from
))
1562 if (tun
->flags
& IFF_VNET_HDR
) {
1563 int vnet_hdr_sz
= READ_ONCE(tun
->vnet_hdr_sz
);
1565 if (len
< vnet_hdr_sz
)
1569 if (!copy_from_iter_full(&gso
, sizeof(gso
), from
))
1572 if ((gso
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) &&
1573 tun16_to_cpu(tun
, gso
.csum_start
) + tun16_to_cpu(tun
, gso
.csum_offset
) + 2 > tun16_to_cpu(tun
, gso
.hdr_len
))
1574 gso
.hdr_len
= cpu_to_tun16(tun
, tun16_to_cpu(tun
, gso
.csum_start
) + tun16_to_cpu(tun
, gso
.csum_offset
) + 2);
1576 if (tun16_to_cpu(tun
, gso
.hdr_len
) > len
)
1578 iov_iter_advance(from
, vnet_hdr_sz
- sizeof(gso
));
1581 if ((tun
->flags
& TUN_TYPE_MASK
) == IFF_TAP
) {
1582 align
+= NET_IP_ALIGN
;
1583 if (unlikely(len
< ETH_HLEN
||
1584 (gso
.hdr_len
&& tun16_to_cpu(tun
, gso
.hdr_len
) < ETH_HLEN
)))
1588 good_linear
= SKB_MAX_HEAD(align
);
1591 struct iov_iter i
= *from
;
1593 /* There are 256 bytes to be copied in skb, so there is
1594 * enough room for skb expand head in case it is used.
1595 * The rest of the buffer is mapped from userspace.
1597 copylen
= gso
.hdr_len
? tun16_to_cpu(tun
, gso
.hdr_len
) : GOODCOPY_LEN
;
1598 if (copylen
> good_linear
)
1599 copylen
= good_linear
;
1601 iov_iter_advance(&i
, copylen
);
1602 if (iov_iter_npages(&i
, INT_MAX
) <= MAX_SKB_FRAGS
)
1606 if (!frags
&& tun_can_build_skb(tun
, tfile
, len
, noblock
, zerocopy
)) {
1607 /* For the packet that is not easy to be processed
1608 * (e.g gso or jumbo packet), we will do it at after
1609 * skb was created with generic XDP routine.
1611 skb
= tun_build_skb(tun
, tfile
, from
, &gso
, len
, &skb_xdp
);
1613 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1614 return PTR_ERR(skb
);
1621 if (tun16_to_cpu(tun
, gso
.hdr_len
) > good_linear
)
1622 linear
= good_linear
;
1624 linear
= tun16_to_cpu(tun
, gso
.hdr_len
);
1628 mutex_lock(&tfile
->napi_mutex
);
1629 skb
= tun_napi_alloc_frags(tfile
, copylen
, from
);
1630 /* tun_napi_alloc_frags() enforces a layout for the skb.
1631 * If zerocopy is enabled, then this layout will be
1632 * overwritten by zerocopy_sg_from_iter().
1636 skb
= tun_alloc_skb(tfile
, align
, copylen
, linear
,
1641 if (PTR_ERR(skb
) != -EAGAIN
)
1642 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1644 mutex_unlock(&tfile
->napi_mutex
);
1645 return PTR_ERR(skb
);
1649 err
= zerocopy_sg_from_iter(skb
, from
);
1651 err
= skb_copy_datagram_from_iter(skb
, 0, from
, len
);
1654 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1657 tfile
->napi
.skb
= NULL
;
1658 mutex_unlock(&tfile
->napi_mutex
);
1665 if (virtio_net_hdr_to_skb(skb
, &gso
, tun_is_little_endian(tun
))) {
1666 this_cpu_inc(tun
->pcpu_stats
->rx_frame_errors
);
1669 tfile
->napi
.skb
= NULL
;
1670 mutex_unlock(&tfile
->napi_mutex
);
1676 switch (tun
->flags
& TUN_TYPE_MASK
) {
1678 if (tun
->flags
& IFF_NO_PI
) {
1679 switch (skb
->data
[0] & 0xf0) {
1681 pi
.proto
= htons(ETH_P_IP
);
1684 pi
.proto
= htons(ETH_P_IPV6
);
1687 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1693 skb_reset_mac_header(skb
);
1694 skb
->protocol
= pi
.proto
;
1695 skb
->dev
= tun
->dev
;
1699 skb
->protocol
= eth_type_trans(skb
, tun
->dev
);
1703 /* copy skb_ubuf_info for callback when skb has no error */
1705 skb_shinfo(skb
)->destructor_arg
= msg_control
;
1706 skb_shinfo(skb
)->tx_flags
|= SKBTX_DEV_ZEROCOPY
;
1707 skb_shinfo(skb
)->tx_flags
|= SKBTX_SHARED_FRAG
;
1708 } else if (msg_control
) {
1709 struct ubuf_info
*uarg
= msg_control
;
1710 uarg
->callback(uarg
, false);
1713 skb_reset_network_header(skb
);
1714 skb_probe_transport_header(skb
, 0);
1717 struct bpf_prog
*xdp_prog
;
1721 xdp_prog
= rcu_dereference(tun
->xdp_prog
);
1723 ret
= do_xdp_generic(xdp_prog
, skb
);
1724 if (ret
!= XDP_PASS
) {
1732 rxhash
= __skb_get_hash_symmetric(skb
);
1735 /* Exercise flow dissector code path. */
1736 u32 headlen
= eth_get_headlen(skb
->data
, skb_headlen(skb
));
1738 if (headlen
> skb_headlen(skb
) || headlen
< ETH_HLEN
) {
1739 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1740 napi_free_frags(&tfile
->napi
);
1741 mutex_unlock(&tfile
->napi_mutex
);
1747 napi_gro_frags(&tfile
->napi
);
1749 mutex_unlock(&tfile
->napi_mutex
);
1750 } else if (tun
->flags
& IFF_NAPI
) {
1751 struct sk_buff_head
*queue
= &tfile
->sk
.sk_write_queue
;
1754 spin_lock_bh(&queue
->lock
);
1755 __skb_queue_tail(queue
, skb
);
1756 queue_len
= skb_queue_len(queue
);
1757 spin_unlock(&queue
->lock
);
1759 if (!more
|| queue_len
> NAPI_POLL_WEIGHT
)
1760 napi_schedule(&tfile
->napi
);
1763 } else if (!IS_ENABLED(CONFIG_4KSTACKS
)) {
1764 tun_rx_batched(tun
, tfile
, skb
, more
);
1769 stats
= get_cpu_ptr(tun
->pcpu_stats
);
1770 u64_stats_update_begin(&stats
->syncp
);
1771 stats
->rx_packets
++;
1772 stats
->rx_bytes
+= len
;
1773 u64_stats_update_end(&stats
->syncp
);
1776 tun_flow_update(tun
, rxhash
, tfile
);
1780 static ssize_t
tun_chr_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
1782 struct file
*file
= iocb
->ki_filp
;
1783 struct tun_file
*tfile
= file
->private_data
;
1784 struct tun_struct
*tun
= tun_get(tfile
);
1790 result
= tun_get_user(tun
, tfile
, NULL
, from
,
1791 file
->f_flags
& O_NONBLOCK
, false);
1797 /* Put packet to the user space buffer */
1798 static ssize_t
tun_put_user(struct tun_struct
*tun
,
1799 struct tun_file
*tfile
,
1800 struct sk_buff
*skb
,
1801 struct iov_iter
*iter
)
1803 struct tun_pi pi
= { 0, skb
->protocol
};
1804 struct tun_pcpu_stats
*stats
;
1806 int vlan_offset
= 0;
1808 int vnet_hdr_sz
= 0;
1810 if (skb_vlan_tag_present(skb
))
1811 vlan_hlen
= VLAN_HLEN
;
1813 if (tun
->flags
& IFF_VNET_HDR
)
1814 vnet_hdr_sz
= READ_ONCE(tun
->vnet_hdr_sz
);
1816 total
= skb
->len
+ vlan_hlen
+ vnet_hdr_sz
;
1818 if (!(tun
->flags
& IFF_NO_PI
)) {
1819 if (iov_iter_count(iter
) < sizeof(pi
))
1822 total
+= sizeof(pi
);
1823 if (iov_iter_count(iter
) < total
) {
1824 /* Packet will be striped */
1825 pi
.flags
|= TUN_PKT_STRIP
;
1828 if (copy_to_iter(&pi
, sizeof(pi
), iter
) != sizeof(pi
))
1833 struct virtio_net_hdr gso
;
1835 if (iov_iter_count(iter
) < vnet_hdr_sz
)
1838 if (virtio_net_hdr_from_skb(skb
, &gso
,
1839 tun_is_little_endian(tun
), true)) {
1840 struct skb_shared_info
*sinfo
= skb_shinfo(skb
);
1841 pr_err("unexpected GSO type: "
1842 "0x%x, gso_size %d, hdr_len %d\n",
1843 sinfo
->gso_type
, tun16_to_cpu(tun
, gso
.gso_size
),
1844 tun16_to_cpu(tun
, gso
.hdr_len
));
1845 print_hex_dump(KERN_ERR
, "tun: ",
1848 min((int)tun16_to_cpu(tun
, gso
.hdr_len
), 64), true);
1853 if (copy_to_iter(&gso
, sizeof(gso
), iter
) != sizeof(gso
))
1856 iov_iter_advance(iter
, vnet_hdr_sz
- sizeof(gso
));
1862 __be16 h_vlan_proto
;
1866 veth
.h_vlan_proto
= skb
->vlan_proto
;
1867 veth
.h_vlan_TCI
= htons(skb_vlan_tag_get(skb
));
1869 vlan_offset
= offsetof(struct vlan_ethhdr
, h_vlan_proto
);
1871 ret
= skb_copy_datagram_iter(skb
, 0, iter
, vlan_offset
);
1872 if (ret
|| !iov_iter_count(iter
))
1875 ret
= copy_to_iter(&veth
, sizeof(veth
), iter
);
1876 if (ret
!= sizeof(veth
) || !iov_iter_count(iter
))
1880 skb_copy_datagram_iter(skb
, vlan_offset
, iter
, skb
->len
- vlan_offset
);
1883 /* caller is in process context, */
1884 stats
= get_cpu_ptr(tun
->pcpu_stats
);
1885 u64_stats_update_begin(&stats
->syncp
);
1886 stats
->tx_packets
++;
1887 stats
->tx_bytes
+= skb
->len
+ vlan_hlen
;
1888 u64_stats_update_end(&stats
->syncp
);
1889 put_cpu_ptr(tun
->pcpu_stats
);
1894 static struct sk_buff
*tun_ring_recv(struct tun_file
*tfile
, int noblock
,
1897 DECLARE_WAITQUEUE(wait
, current
);
1898 struct sk_buff
*skb
= NULL
;
1901 skb
= skb_array_consume(&tfile
->tx_array
);
1909 add_wait_queue(&tfile
->wq
.wait
, &wait
);
1910 current
->state
= TASK_INTERRUPTIBLE
;
1913 skb
= skb_array_consume(&tfile
->tx_array
);
1916 if (signal_pending(current
)) {
1917 error
= -ERESTARTSYS
;
1920 if (tfile
->socket
.sk
->sk_shutdown
& RCV_SHUTDOWN
) {
1928 current
->state
= TASK_RUNNING
;
1929 remove_wait_queue(&tfile
->wq
.wait
, &wait
);
1936 static ssize_t
tun_do_read(struct tun_struct
*tun
, struct tun_file
*tfile
,
1937 struct iov_iter
*to
,
1938 int noblock
, struct sk_buff
*skb
)
1943 tun_debug(KERN_INFO
, tun
, "tun_do_read\n");
1945 if (!iov_iter_count(to
))
1949 /* Read frames from ring */
1950 skb
= tun_ring_recv(tfile
, noblock
, &err
);
1955 ret
= tun_put_user(tun
, tfile
, skb
, to
);
1956 if (unlikely(ret
< 0))
1964 static ssize_t
tun_chr_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
1966 struct file
*file
= iocb
->ki_filp
;
1967 struct tun_file
*tfile
= file
->private_data
;
1968 struct tun_struct
*tun
= tun_get(tfile
);
1969 ssize_t len
= iov_iter_count(to
), ret
;
1973 ret
= tun_do_read(tun
, tfile
, to
, file
->f_flags
& O_NONBLOCK
, NULL
);
1974 ret
= min_t(ssize_t
, ret
, len
);
1981 static void tun_free_netdev(struct net_device
*dev
)
1983 struct tun_struct
*tun
= netdev_priv(dev
);
1985 BUG_ON(!(list_empty(&tun
->disabled
)));
1986 free_percpu(tun
->pcpu_stats
);
1987 tun_flow_uninit(tun
);
1988 security_tun_dev_free_security(tun
->security
);
1991 static void tun_setup(struct net_device
*dev
)
1993 struct tun_struct
*tun
= netdev_priv(dev
);
1995 tun
->owner
= INVALID_UID
;
1996 tun
->group
= INVALID_GID
;
1998 dev
->ethtool_ops
= &tun_ethtool_ops
;
1999 dev
->needs_free_netdev
= true;
2000 dev
->priv_destructor
= tun_free_netdev
;
2001 /* We prefer our own queue length */
2002 dev
->tx_queue_len
= TUN_READQ_SIZE
;
2005 /* Trivial set of netlink ops to allow deleting tun or tap
2006 * device with netlink.
2008 static int tun_validate(struct nlattr
*tb
[], struct nlattr
*data
[],
2009 struct netlink_ext_ack
*extack
)
2014 static struct rtnl_link_ops tun_link_ops __read_mostly
= {
2016 .priv_size
= sizeof(struct tun_struct
),
2018 .validate
= tun_validate
,
2021 static void tun_sock_write_space(struct sock
*sk
)
2023 struct tun_file
*tfile
;
2024 wait_queue_head_t
*wqueue
;
2026 if (!sock_writeable(sk
))
2029 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
))
2032 wqueue
= sk_sleep(sk
);
2033 if (wqueue
&& waitqueue_active(wqueue
))
2034 wake_up_interruptible_sync_poll(wqueue
, POLLOUT
|
2035 POLLWRNORM
| POLLWRBAND
);
2037 tfile
= container_of(sk
, struct tun_file
, sk
);
2038 kill_fasync(&tfile
->fasync
, SIGIO
, POLL_OUT
);
2041 static int tun_sendmsg(struct socket
*sock
, struct msghdr
*m
, size_t total_len
)
2044 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
2045 struct tun_struct
*tun
= tun_get(tfile
);
2050 ret
= tun_get_user(tun
, tfile
, m
->msg_control
, &m
->msg_iter
,
2051 m
->msg_flags
& MSG_DONTWAIT
,
2052 m
->msg_flags
& MSG_MORE
);
2057 static int tun_recvmsg(struct socket
*sock
, struct msghdr
*m
, size_t total_len
,
2060 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
2061 struct tun_struct
*tun
= tun_get(tfile
);
2067 if (flags
& ~(MSG_DONTWAIT
|MSG_TRUNC
|MSG_ERRQUEUE
)) {
2071 if (flags
& MSG_ERRQUEUE
) {
2072 ret
= sock_recv_errqueue(sock
->sk
, m
, total_len
,
2073 SOL_PACKET
, TUN_TX_TIMESTAMP
);
2076 ret
= tun_do_read(tun
, tfile
, &m
->msg_iter
, flags
& MSG_DONTWAIT
,
2078 if (ret
> (ssize_t
)total_len
) {
2079 m
->msg_flags
|= MSG_TRUNC
;
2080 ret
= flags
& MSG_TRUNC
? ret
: total_len
;
2087 static int tun_peek_len(struct socket
*sock
)
2089 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
2090 struct tun_struct
*tun
;
2093 tun
= tun_get(tfile
);
2097 ret
= skb_array_peek_len(&tfile
->tx_array
);
2103 /* Ops structure to mimic raw sockets with tun */
2104 static const struct proto_ops tun_socket_ops
= {
2105 .peek_len
= tun_peek_len
,
2106 .sendmsg
= tun_sendmsg
,
2107 .recvmsg
= tun_recvmsg
,
2110 static struct proto tun_proto
= {
2112 .owner
= THIS_MODULE
,
2113 .obj_size
= sizeof(struct tun_file
),
2116 static int tun_flags(struct tun_struct
*tun
)
2118 return tun
->flags
& (TUN_FEATURES
| IFF_PERSIST
| IFF_TUN
| IFF_TAP
);
2121 static ssize_t
tun_show_flags(struct device
*dev
, struct device_attribute
*attr
,
2124 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
2125 return sprintf(buf
, "0x%x\n", tun_flags(tun
));
2128 static ssize_t
tun_show_owner(struct device
*dev
, struct device_attribute
*attr
,
2131 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
2132 return uid_valid(tun
->owner
)?
2133 sprintf(buf
, "%u\n",
2134 from_kuid_munged(current_user_ns(), tun
->owner
)):
2135 sprintf(buf
, "-1\n");
2138 static ssize_t
tun_show_group(struct device
*dev
, struct device_attribute
*attr
,
2141 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
2142 return gid_valid(tun
->group
) ?
2143 sprintf(buf
, "%u\n",
2144 from_kgid_munged(current_user_ns(), tun
->group
)):
2145 sprintf(buf
, "-1\n");
2148 static DEVICE_ATTR(tun_flags
, 0444, tun_show_flags
, NULL
);
2149 static DEVICE_ATTR(owner
, 0444, tun_show_owner
, NULL
);
2150 static DEVICE_ATTR(group
, 0444, tun_show_group
, NULL
);
2152 static struct attribute
*tun_dev_attrs
[] = {
2153 &dev_attr_tun_flags
.attr
,
2154 &dev_attr_owner
.attr
,
2155 &dev_attr_group
.attr
,
2159 static const struct attribute_group tun_attr_group
= {
2160 .attrs
= tun_dev_attrs
2163 static int tun_set_iff(struct net
*net
, struct file
*file
, struct ifreq
*ifr
)
2165 struct tun_struct
*tun
;
2166 struct tun_file
*tfile
= file
->private_data
;
2167 struct net_device
*dev
;
2170 if (tfile
->detached
)
2173 if ((ifr
->ifr_flags
& IFF_NAPI_FRAGS
)) {
2174 if (!capable(CAP_NET_ADMIN
))
2177 if (!(ifr
->ifr_flags
& IFF_NAPI
) ||
2178 (ifr
->ifr_flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2182 dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
2184 if (ifr
->ifr_flags
& IFF_TUN_EXCL
)
2186 if ((ifr
->ifr_flags
& IFF_TUN
) && dev
->netdev_ops
== &tun_netdev_ops
)
2187 tun
= netdev_priv(dev
);
2188 else if ((ifr
->ifr_flags
& IFF_TAP
) && dev
->netdev_ops
== &tap_netdev_ops
)
2189 tun
= netdev_priv(dev
);
2193 if (!!(ifr
->ifr_flags
& IFF_MULTI_QUEUE
) !=
2194 !!(tun
->flags
& IFF_MULTI_QUEUE
))
2197 if (tun_not_capable(tun
))
2199 err
= security_tun_dev_open(tun
->security
);
2203 err
= tun_attach(tun
, file
, ifr
->ifr_flags
& IFF_NOFILTER
,
2204 ifr
->ifr_flags
& IFF_NAPI
);
2208 if (tun
->flags
& IFF_MULTI_QUEUE
&&
2209 (tun
->numqueues
+ tun
->numdisabled
> 1)) {
2210 /* One or more queue has already been attached, no need
2211 * to initialize the device again.
2218 unsigned long flags
= 0;
2219 int queues
= ifr
->ifr_flags
& IFF_MULTI_QUEUE
?
2222 if (!ns_capable(net
->user_ns
, CAP_NET_ADMIN
))
2224 err
= security_tun_dev_create();
2229 if (ifr
->ifr_flags
& IFF_TUN
) {
2233 } else if (ifr
->ifr_flags
& IFF_TAP
) {
2241 name
= ifr
->ifr_name
;
2243 dev
= alloc_netdev_mqs(sizeof(struct tun_struct
), name
,
2244 NET_NAME_UNKNOWN
, tun_setup
, queues
,
2250 dev_net_set(dev
, net
);
2251 dev
->rtnl_link_ops
= &tun_link_ops
;
2252 dev
->ifindex
= tfile
->ifindex
;
2253 dev
->sysfs_groups
[0] = &tun_attr_group
;
2255 tun
= netdev_priv(dev
);
2258 tun
->txflt
.count
= 0;
2259 tun
->vnet_hdr_sz
= sizeof(struct virtio_net_hdr
);
2261 tun
->align
= NET_SKB_PAD
;
2262 tun
->filter_attached
= false;
2263 tun
->sndbuf
= tfile
->socket
.sk
->sk_sndbuf
;
2264 tun
->rx_batched
= 0;
2266 tun
->pcpu_stats
= netdev_alloc_pcpu_stats(struct tun_pcpu_stats
);
2267 if (!tun
->pcpu_stats
) {
2272 spin_lock_init(&tun
->lock
);
2274 err
= security_tun_dev_alloc_security(&tun
->security
);
2281 dev
->hw_features
= NETIF_F_SG
| NETIF_F_FRAGLIST
|
2282 TUN_USER_FEATURES
| NETIF_F_HW_VLAN_CTAG_TX
|
2283 NETIF_F_HW_VLAN_STAG_TX
;
2284 dev
->features
= dev
->hw_features
| NETIF_F_LLTX
;
2285 dev
->vlan_features
= dev
->features
&
2286 ~(NETIF_F_HW_VLAN_CTAG_TX
|
2287 NETIF_F_HW_VLAN_STAG_TX
);
2289 INIT_LIST_HEAD(&tun
->disabled
);
2290 err
= tun_attach(tun
, file
, false, ifr
->ifr_flags
& IFF_NAPI
);
2294 err
= register_netdevice(tun
->dev
);
2299 netif_carrier_on(tun
->dev
);
2301 tun_debug(KERN_INFO
, tun
, "tun_set_iff\n");
2303 tun
->flags
= (tun
->flags
& ~TUN_FEATURES
) |
2304 (ifr
->ifr_flags
& TUN_FEATURES
);
2306 /* Make sure persistent devices do not get stuck in
2309 if (netif_running(tun
->dev
))
2310 netif_tx_wake_all_queues(tun
->dev
);
2312 strcpy(ifr
->ifr_name
, tun
->dev
->name
);
2316 tun_detach_all(dev
);
2317 /* register_netdevice() already called tun_free_netdev() */
2321 tun_flow_uninit(tun
);
2322 security_tun_dev_free_security(tun
->security
);
2324 free_percpu(tun
->pcpu_stats
);
2330 static void tun_get_iff(struct net
*net
, struct tun_struct
*tun
,
2333 tun_debug(KERN_INFO
, tun
, "tun_get_iff\n");
2335 strcpy(ifr
->ifr_name
, tun
->dev
->name
);
2337 ifr
->ifr_flags
= tun_flags(tun
);
2341 /* This is like a cut-down ethtool ops, except done via tun fd so no
2342 * privs required. */
2343 static int set_offload(struct tun_struct
*tun
, unsigned long arg
)
2345 netdev_features_t features
= 0;
2347 if (arg
& TUN_F_CSUM
) {
2348 features
|= NETIF_F_HW_CSUM
;
2351 if (arg
& (TUN_F_TSO4
|TUN_F_TSO6
)) {
2352 if (arg
& TUN_F_TSO_ECN
) {
2353 features
|= NETIF_F_TSO_ECN
;
2354 arg
&= ~TUN_F_TSO_ECN
;
2356 if (arg
& TUN_F_TSO4
)
2357 features
|= NETIF_F_TSO
;
2358 if (arg
& TUN_F_TSO6
)
2359 features
|= NETIF_F_TSO6
;
2360 arg
&= ~(TUN_F_TSO4
|TUN_F_TSO6
);
2364 /* This gives the user a way to test for new features in future by
2365 * trying to set them. */
2369 tun
->set_features
= features
;
2370 tun
->dev
->wanted_features
&= ~TUN_USER_FEATURES
;
2371 tun
->dev
->wanted_features
|= features
;
2372 netdev_update_features(tun
->dev
);
2377 static void tun_detach_filter(struct tun_struct
*tun
, int n
)
2380 struct tun_file
*tfile
;
2382 for (i
= 0; i
< n
; i
++) {
2383 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
2384 lock_sock(tfile
->socket
.sk
);
2385 sk_detach_filter(tfile
->socket
.sk
);
2386 release_sock(tfile
->socket
.sk
);
2389 tun
->filter_attached
= false;
2392 static int tun_attach_filter(struct tun_struct
*tun
)
2395 struct tun_file
*tfile
;
2397 for (i
= 0; i
< tun
->numqueues
; i
++) {
2398 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
2399 lock_sock(tfile
->socket
.sk
);
2400 ret
= sk_attach_filter(&tun
->fprog
, tfile
->socket
.sk
);
2401 release_sock(tfile
->socket
.sk
);
2403 tun_detach_filter(tun
, i
);
2408 tun
->filter_attached
= true;
2412 static void tun_set_sndbuf(struct tun_struct
*tun
)
2414 struct tun_file
*tfile
;
2417 for (i
= 0; i
< tun
->numqueues
; i
++) {
2418 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
2419 tfile
->socket
.sk
->sk_sndbuf
= tun
->sndbuf
;
2423 static int tun_set_queue(struct file
*file
, struct ifreq
*ifr
)
2425 struct tun_file
*tfile
= file
->private_data
;
2426 struct tun_struct
*tun
;
2431 if (ifr
->ifr_flags
& IFF_ATTACH_QUEUE
) {
2432 tun
= tfile
->detached
;
2437 ret
= security_tun_dev_attach_queue(tun
->security
);
2440 ret
= tun_attach(tun
, file
, false, tun
->flags
& IFF_NAPI
);
2441 } else if (ifr
->ifr_flags
& IFF_DETACH_QUEUE
) {
2442 tun
= rtnl_dereference(tfile
->tun
);
2443 if (!tun
|| !(tun
->flags
& IFF_MULTI_QUEUE
) || tfile
->detached
)
2446 __tun_detach(tfile
, false);
2455 static long __tun_chr_ioctl(struct file
*file
, unsigned int cmd
,
2456 unsigned long arg
, int ifreq_len
)
2458 struct tun_file
*tfile
= file
->private_data
;
2459 struct tun_struct
*tun
;
2460 void __user
* argp
= (void __user
*)arg
;
2466 unsigned int ifindex
;
2470 if (cmd
== TUNSETIFF
|| cmd
== TUNSETQUEUE
|| _IOC_TYPE(cmd
) == SOCK_IOC_TYPE
) {
2471 if (copy_from_user(&ifr
, argp
, ifreq_len
))
2474 memset(&ifr
, 0, sizeof(ifr
));
2476 if (cmd
== TUNGETFEATURES
) {
2477 /* Currently this just means: "what IFF flags are valid?".
2478 * This is needed because we never checked for invalid flags on
2481 return put_user(IFF_TUN
| IFF_TAP
| TUN_FEATURES
,
2482 (unsigned int __user
*)argp
);
2483 } else if (cmd
== TUNSETQUEUE
)
2484 return tun_set_queue(file
, &ifr
);
2489 tun
= tun_get(tfile
);
2490 if (cmd
== TUNSETIFF
) {
2495 ifr
.ifr_name
[IFNAMSIZ
-1] = '\0';
2497 ret
= tun_set_iff(sock_net(&tfile
->sk
), file
, &ifr
);
2502 if (copy_to_user(argp
, &ifr
, ifreq_len
))
2506 if (cmd
== TUNSETIFINDEX
) {
2512 if (copy_from_user(&ifindex
, argp
, sizeof(ifindex
)))
2516 tfile
->ifindex
= ifindex
;
2524 tun_debug(KERN_INFO
, tun
, "tun_chr_ioctl cmd %u\n", cmd
);
2529 tun_get_iff(current
->nsproxy
->net_ns
, tun
, &ifr
);
2531 if (tfile
->detached
)
2532 ifr
.ifr_flags
|= IFF_DETACH_QUEUE
;
2533 if (!tfile
->socket
.sk
->sk_filter
)
2534 ifr
.ifr_flags
|= IFF_NOFILTER
;
2536 if (copy_to_user(argp
, &ifr
, ifreq_len
))
2541 /* Disable/Enable checksum */
2543 /* [unimplemented] */
2544 tun_debug(KERN_INFO
, tun
, "ignored: set checksum %s\n",
2545 arg
? "disabled" : "enabled");
2549 /* Disable/Enable persist mode. Keep an extra reference to the
2550 * module to prevent the module being unprobed.
2552 if (arg
&& !(tun
->flags
& IFF_PERSIST
)) {
2553 tun
->flags
|= IFF_PERSIST
;
2554 __module_get(THIS_MODULE
);
2556 if (!arg
&& (tun
->flags
& IFF_PERSIST
)) {
2557 tun
->flags
&= ~IFF_PERSIST
;
2558 module_put(THIS_MODULE
);
2561 tun_debug(KERN_INFO
, tun
, "persist %s\n",
2562 arg
? "enabled" : "disabled");
2566 /* Set owner of the device */
2567 owner
= make_kuid(current_user_ns(), arg
);
2568 if (!uid_valid(owner
)) {
2573 tun_debug(KERN_INFO
, tun
, "owner set to %u\n",
2574 from_kuid(&init_user_ns
, tun
->owner
));
2578 /* Set group of the device */
2579 group
= make_kgid(current_user_ns(), arg
);
2580 if (!gid_valid(group
)) {
2585 tun_debug(KERN_INFO
, tun
, "group set to %u\n",
2586 from_kgid(&init_user_ns
, tun
->group
));
2590 /* Only allow setting the type when the interface is down */
2591 if (tun
->dev
->flags
& IFF_UP
) {
2592 tun_debug(KERN_INFO
, tun
,
2593 "Linktype set failed because interface is up\n");
2596 tun
->dev
->type
= (int) arg
;
2597 tun_debug(KERN_INFO
, tun
, "linktype set to %d\n",
2609 ret
= set_offload(tun
, arg
);
2612 case TUNSETTXFILTER
:
2613 /* Can be set only for TAPs */
2615 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2617 ret
= update_filter(&tun
->txflt
, (void __user
*)arg
);
2621 /* Get hw address */
2622 memcpy(ifr
.ifr_hwaddr
.sa_data
, tun
->dev
->dev_addr
, ETH_ALEN
);
2623 ifr
.ifr_hwaddr
.sa_family
= tun
->dev
->type
;
2624 if (copy_to_user(argp
, &ifr
, ifreq_len
))
2629 /* Set hw address */
2630 tun_debug(KERN_DEBUG
, tun
, "set hw address: %pM\n",
2631 ifr
.ifr_hwaddr
.sa_data
);
2633 ret
= dev_set_mac_address(tun
->dev
, &ifr
.ifr_hwaddr
);
2637 sndbuf
= tfile
->socket
.sk
->sk_sndbuf
;
2638 if (copy_to_user(argp
, &sndbuf
, sizeof(sndbuf
)))
2643 if (copy_from_user(&sndbuf
, argp
, sizeof(sndbuf
))) {
2648 tun
->sndbuf
= sndbuf
;
2649 tun_set_sndbuf(tun
);
2652 case TUNGETVNETHDRSZ
:
2653 vnet_hdr_sz
= tun
->vnet_hdr_sz
;
2654 if (copy_to_user(argp
, &vnet_hdr_sz
, sizeof(vnet_hdr_sz
)))
2658 case TUNSETVNETHDRSZ
:
2659 if (copy_from_user(&vnet_hdr_sz
, argp
, sizeof(vnet_hdr_sz
))) {
2663 if (vnet_hdr_sz
< (int)sizeof(struct virtio_net_hdr
)) {
2668 tun
->vnet_hdr_sz
= vnet_hdr_sz
;
2672 le
= !!(tun
->flags
& TUN_VNET_LE
);
2673 if (put_user(le
, (int __user
*)argp
))
2678 if (get_user(le
, (int __user
*)argp
)) {
2683 tun
->flags
|= TUN_VNET_LE
;
2685 tun
->flags
&= ~TUN_VNET_LE
;
2689 ret
= tun_get_vnet_be(tun
, argp
);
2693 ret
= tun_set_vnet_be(tun
, argp
);
2696 case TUNATTACHFILTER
:
2697 /* Can be set only for TAPs */
2699 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2702 if (copy_from_user(&tun
->fprog
, argp
, sizeof(tun
->fprog
)))
2705 ret
= tun_attach_filter(tun
);
2708 case TUNDETACHFILTER
:
2709 /* Can be set only for TAPs */
2711 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2714 tun_detach_filter(tun
, tun
->numqueues
);
2719 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2722 if (copy_to_user(argp
, &tun
->fprog
, sizeof(tun
->fprog
)))
2739 static long tun_chr_ioctl(struct file
*file
,
2740 unsigned int cmd
, unsigned long arg
)
2742 return __tun_chr_ioctl(file
, cmd
, arg
, sizeof (struct ifreq
));
2745 #ifdef CONFIG_COMPAT
2746 static long tun_chr_compat_ioctl(struct file
*file
,
2747 unsigned int cmd
, unsigned long arg
)
2752 case TUNSETTXFILTER
:
2757 arg
= (unsigned long)compat_ptr(arg
);
2760 arg
= (compat_ulong_t
)arg
;
2765 * compat_ifreq is shorter than ifreq, so we must not access beyond
2766 * the end of that structure. All fields that are used in this
2767 * driver are compatible though, we don't need to convert the
2770 return __tun_chr_ioctl(file
, cmd
, arg
, sizeof(struct compat_ifreq
));
2772 #endif /* CONFIG_COMPAT */
2774 static int tun_chr_fasync(int fd
, struct file
*file
, int on
)
2776 struct tun_file
*tfile
= file
->private_data
;
2779 if ((ret
= fasync_helper(fd
, file
, on
, &tfile
->fasync
)) < 0)
2783 __f_setown(file
, task_pid(current
), PIDTYPE_PID
, 0);
2784 tfile
->flags
|= TUN_FASYNC
;
2786 tfile
->flags
&= ~TUN_FASYNC
;
2792 static int tun_chr_open(struct inode
*inode
, struct file
* file
)
2794 struct net
*net
= current
->nsproxy
->net_ns
;
2795 struct tun_file
*tfile
;
2797 DBG1(KERN_INFO
, "tunX: tun_chr_open\n");
2799 tfile
= (struct tun_file
*)sk_alloc(net
, AF_UNSPEC
, GFP_KERNEL
,
2803 RCU_INIT_POINTER(tfile
->tun
, NULL
);
2807 init_waitqueue_head(&tfile
->wq
.wait
);
2808 RCU_INIT_POINTER(tfile
->socket
.wq
, &tfile
->wq
);
2810 tfile
->socket
.file
= file
;
2811 tfile
->socket
.ops
= &tun_socket_ops
;
2813 sock_init_data(&tfile
->socket
, &tfile
->sk
);
2815 tfile
->sk
.sk_write_space
= tun_sock_write_space
;
2816 tfile
->sk
.sk_sndbuf
= INT_MAX
;
2818 file
->private_data
= tfile
;
2819 INIT_LIST_HEAD(&tfile
->next
);
2821 sock_set_flag(&tfile
->sk
, SOCK_ZEROCOPY
);
2826 static int tun_chr_close(struct inode
*inode
, struct file
*file
)
2828 struct tun_file
*tfile
= file
->private_data
;
2830 tun_detach(tfile
, true);
2835 #ifdef CONFIG_PROC_FS
2836 static void tun_chr_show_fdinfo(struct seq_file
*m
, struct file
*file
)
2838 struct tun_file
*tfile
= file
->private_data
;
2839 struct tun_struct
*tun
;
2842 memset(&ifr
, 0, sizeof(ifr
));
2845 tun
= tun_get(tfile
);
2847 tun_get_iff(current
->nsproxy
->net_ns
, tun
, &ifr
);
2853 seq_printf(m
, "iff:\t%s\n", ifr
.ifr_name
);
2857 static const struct file_operations tun_fops
= {
2858 .owner
= THIS_MODULE
,
2859 .llseek
= no_llseek
,
2860 .read_iter
= tun_chr_read_iter
,
2861 .write_iter
= tun_chr_write_iter
,
2862 .poll
= tun_chr_poll
,
2863 .unlocked_ioctl
= tun_chr_ioctl
,
2864 #ifdef CONFIG_COMPAT
2865 .compat_ioctl
= tun_chr_compat_ioctl
,
2867 .open
= tun_chr_open
,
2868 .release
= tun_chr_close
,
2869 .fasync
= tun_chr_fasync
,
2870 #ifdef CONFIG_PROC_FS
2871 .show_fdinfo
= tun_chr_show_fdinfo
,
2875 static struct miscdevice tun_miscdev
= {
2878 .nodename
= "net/tun",
2882 /* ethtool interface */
2884 static int tun_get_link_ksettings(struct net_device
*dev
,
2885 struct ethtool_link_ksettings
*cmd
)
2887 ethtool_link_ksettings_zero_link_mode(cmd
, supported
);
2888 ethtool_link_ksettings_zero_link_mode(cmd
, advertising
);
2889 cmd
->base
.speed
= SPEED_10
;
2890 cmd
->base
.duplex
= DUPLEX_FULL
;
2891 cmd
->base
.port
= PORT_TP
;
2892 cmd
->base
.phy_address
= 0;
2893 cmd
->base
.autoneg
= AUTONEG_DISABLE
;
2897 static void tun_get_drvinfo(struct net_device
*dev
, struct ethtool_drvinfo
*info
)
2899 struct tun_struct
*tun
= netdev_priv(dev
);
2901 strlcpy(info
->driver
, DRV_NAME
, sizeof(info
->driver
));
2902 strlcpy(info
->version
, DRV_VERSION
, sizeof(info
->version
));
2904 switch (tun
->flags
& TUN_TYPE_MASK
) {
2906 strlcpy(info
->bus_info
, "tun", sizeof(info
->bus_info
));
2909 strlcpy(info
->bus_info
, "tap", sizeof(info
->bus_info
));
2914 static u32
tun_get_msglevel(struct net_device
*dev
)
2917 struct tun_struct
*tun
= netdev_priv(dev
);
2924 static void tun_set_msglevel(struct net_device
*dev
, u32 value
)
2927 struct tun_struct
*tun
= netdev_priv(dev
);
2932 static int tun_get_coalesce(struct net_device
*dev
,
2933 struct ethtool_coalesce
*ec
)
2935 struct tun_struct
*tun
= netdev_priv(dev
);
2937 ec
->rx_max_coalesced_frames
= tun
->rx_batched
;
2942 static int tun_set_coalesce(struct net_device
*dev
,
2943 struct ethtool_coalesce
*ec
)
2945 struct tun_struct
*tun
= netdev_priv(dev
);
2947 if (ec
->rx_max_coalesced_frames
> NAPI_POLL_WEIGHT
)
2948 tun
->rx_batched
= NAPI_POLL_WEIGHT
;
2950 tun
->rx_batched
= ec
->rx_max_coalesced_frames
;
2955 static const struct ethtool_ops tun_ethtool_ops
= {
2956 .get_drvinfo
= tun_get_drvinfo
,
2957 .get_msglevel
= tun_get_msglevel
,
2958 .set_msglevel
= tun_set_msglevel
,
2959 .get_link
= ethtool_op_get_link
,
2960 .get_ts_info
= ethtool_op_get_ts_info
,
2961 .get_coalesce
= tun_get_coalesce
,
2962 .set_coalesce
= tun_set_coalesce
,
2963 .get_link_ksettings
= tun_get_link_ksettings
,
2966 static int tun_queue_resize(struct tun_struct
*tun
)
2968 struct net_device
*dev
= tun
->dev
;
2969 struct tun_file
*tfile
;
2970 struct skb_array
**arrays
;
2971 int n
= tun
->numqueues
+ tun
->numdisabled
;
2974 arrays
= kmalloc_array(n
, sizeof(*arrays
), GFP_KERNEL
);
2978 for (i
= 0; i
< tun
->numqueues
; i
++) {
2979 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
2980 arrays
[i
] = &tfile
->tx_array
;
2982 list_for_each_entry(tfile
, &tun
->disabled
, next
)
2983 arrays
[i
++] = &tfile
->tx_array
;
2985 ret
= skb_array_resize_multiple(arrays
, n
,
2986 dev
->tx_queue_len
, GFP_KERNEL
);
2992 static int tun_device_event(struct notifier_block
*unused
,
2993 unsigned long event
, void *ptr
)
2995 struct net_device
*dev
= netdev_notifier_info_to_dev(ptr
);
2996 struct tun_struct
*tun
= netdev_priv(dev
);
2998 if (dev
->rtnl_link_ops
!= &tun_link_ops
)
3002 case NETDEV_CHANGE_TX_QUEUE_LEN
:
3003 if (tun_queue_resize(tun
))
3013 static struct notifier_block tun_notifier_block __read_mostly
= {
3014 .notifier_call
= tun_device_event
,
3017 static int __init
tun_init(void)
3021 pr_info("%s, %s\n", DRV_DESCRIPTION
, DRV_VERSION
);
3023 ret
= rtnl_link_register(&tun_link_ops
);
3025 pr_err("Can't register link_ops\n");
3029 ret
= misc_register(&tun_miscdev
);
3031 pr_err("Can't register misc device %d\n", TUN_MINOR
);
3035 ret
= register_netdevice_notifier(&tun_notifier_block
);
3037 pr_err("Can't register netdevice notifier\n");
3044 misc_deregister(&tun_miscdev
);
3046 rtnl_link_unregister(&tun_link_ops
);
3051 static void tun_cleanup(void)
3053 misc_deregister(&tun_miscdev
);
3054 rtnl_link_unregister(&tun_link_ops
);
3055 unregister_netdevice_notifier(&tun_notifier_block
);
3058 /* Get an underlying socket object from tun file. Returns error unless file is
3059 * attached to a device. The returned object works like a packet socket, it
3060 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
3061 * holding a reference to the file for as long as the socket is in use. */
3062 struct socket
*tun_get_socket(struct file
*file
)
3064 struct tun_file
*tfile
;
3065 if (file
->f_op
!= &tun_fops
)
3066 return ERR_PTR(-EINVAL
);
3067 tfile
= file
->private_data
;
3069 return ERR_PTR(-EBADFD
);
3070 return &tfile
->socket
;
3072 EXPORT_SYMBOL_GPL(tun_get_socket
);
3074 struct skb_array
*tun_get_skb_array(struct file
*file
)
3076 struct tun_file
*tfile
;
3078 if (file
->f_op
!= &tun_fops
)
3079 return ERR_PTR(-EINVAL
);
3080 tfile
= file
->private_data
;
3082 return ERR_PTR(-EBADFD
);
3083 return &tfile
->tx_array
;
3085 EXPORT_SYMBOL_GPL(tun_get_skb_array
);
3087 module_init(tun_init
);
3088 module_exit(tun_cleanup
);
3089 MODULE_DESCRIPTION(DRV_DESCRIPTION
);
3090 MODULE_AUTHOR(DRV_COPYRIGHT
);
3091 MODULE_LICENSE("GPL");
3092 MODULE_ALIAS_MISCDEV(TUN_MINOR
);
3093 MODULE_ALIAS("devname:net/tun");