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
;
179 struct mutex napi_mutex
; /* Protects access to the above napi */
180 struct list_head next
;
181 struct tun_struct
*detached
;
182 struct skb_array tx_array
;
185 struct tun_flow_entry
{
186 struct hlist_node hash_link
;
188 struct tun_struct
*tun
;
193 unsigned long updated
;
196 #define TUN_NUM_FLOW_ENTRIES 1024
198 /* Since the socket were moved to tun_file, to preserve the behavior of persist
199 * device, socket filter, sndbuf and vnet header size were restore when the
200 * file were attached to a persist device.
203 struct tun_file __rcu
*tfiles
[MAX_TAP_QUEUES
];
204 unsigned int numqueues
;
209 struct net_device
*dev
;
210 netdev_features_t set_features
;
211 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
217 struct tap_filter txflt
;
218 struct sock_fprog fprog
;
219 /* protected by rtnl lock */
220 bool filter_attached
;
225 struct hlist_head flows
[TUN_NUM_FLOW_ENTRIES
];
226 struct timer_list flow_gc_timer
;
227 unsigned long ageing_time
;
228 unsigned int numdisabled
;
229 struct list_head disabled
;
233 struct tun_pcpu_stats __percpu
*pcpu_stats
;
234 struct bpf_prog __rcu
*xdp_prog
;
237 static int tun_napi_receive(struct napi_struct
*napi
, int budget
)
239 struct tun_file
*tfile
= container_of(napi
, struct tun_file
, napi
);
240 struct sk_buff_head
*queue
= &tfile
->sk
.sk_write_queue
;
241 struct sk_buff_head process_queue
;
245 __skb_queue_head_init(&process_queue
);
247 spin_lock(&queue
->lock
);
248 skb_queue_splice_tail_init(queue
, &process_queue
);
249 spin_unlock(&queue
->lock
);
251 while (received
< budget
&& (skb
= __skb_dequeue(&process_queue
))) {
252 napi_gro_receive(napi
, skb
);
256 if (!skb_queue_empty(&process_queue
)) {
257 spin_lock(&queue
->lock
);
258 skb_queue_splice(&process_queue
, queue
);
259 spin_unlock(&queue
->lock
);
265 static int tun_napi_poll(struct napi_struct
*napi
, int budget
)
267 unsigned int received
;
269 received
= tun_napi_receive(napi
, budget
);
271 if (received
< budget
)
272 napi_complete_done(napi
, received
);
277 static void tun_napi_init(struct tun_struct
*tun
, struct tun_file
*tfile
,
280 tfile
->napi_enabled
= napi_en
;
282 netif_napi_add(tun
->dev
, &tfile
->napi
, tun_napi_poll
,
284 napi_enable(&tfile
->napi
);
285 mutex_init(&tfile
->napi_mutex
);
289 static void tun_napi_disable(struct tun_struct
*tun
, struct tun_file
*tfile
)
291 if (tfile
->napi_enabled
)
292 napi_disable(&tfile
->napi
);
295 static void tun_napi_del(struct tun_struct
*tun
, struct tun_file
*tfile
)
297 if (tfile
->napi_enabled
)
298 netif_napi_del(&tfile
->napi
);
301 static bool tun_napi_frags_enabled(const struct tun_struct
*tun
)
303 return READ_ONCE(tun
->flags
) & IFF_NAPI_FRAGS
;
306 #ifdef CONFIG_TUN_VNET_CROSS_LE
307 static inline bool tun_legacy_is_little_endian(struct tun_struct
*tun
)
309 return tun
->flags
& TUN_VNET_BE
? false :
310 virtio_legacy_is_little_endian();
313 static long tun_get_vnet_be(struct tun_struct
*tun
, int __user
*argp
)
315 int be
= !!(tun
->flags
& TUN_VNET_BE
);
317 if (put_user(be
, argp
))
323 static long tun_set_vnet_be(struct tun_struct
*tun
, int __user
*argp
)
327 if (get_user(be
, argp
))
331 tun
->flags
|= TUN_VNET_BE
;
333 tun
->flags
&= ~TUN_VNET_BE
;
338 static inline bool tun_legacy_is_little_endian(struct tun_struct
*tun
)
340 return virtio_legacy_is_little_endian();
343 static long tun_get_vnet_be(struct tun_struct
*tun
, int __user
*argp
)
348 static long tun_set_vnet_be(struct tun_struct
*tun
, int __user
*argp
)
352 #endif /* CONFIG_TUN_VNET_CROSS_LE */
354 static inline bool tun_is_little_endian(struct tun_struct
*tun
)
356 return tun
->flags
& TUN_VNET_LE
||
357 tun_legacy_is_little_endian(tun
);
360 static inline u16
tun16_to_cpu(struct tun_struct
*tun
, __virtio16 val
)
362 return __virtio16_to_cpu(tun_is_little_endian(tun
), val
);
365 static inline __virtio16
cpu_to_tun16(struct tun_struct
*tun
, u16 val
)
367 return __cpu_to_virtio16(tun_is_little_endian(tun
), val
);
370 static inline u32
tun_hashfn(u32 rxhash
)
372 return rxhash
& 0x3ff;
375 static struct tun_flow_entry
*tun_flow_find(struct hlist_head
*head
, u32 rxhash
)
377 struct tun_flow_entry
*e
;
379 hlist_for_each_entry_rcu(e
, head
, hash_link
) {
380 if (e
->rxhash
== rxhash
)
386 static struct tun_flow_entry
*tun_flow_create(struct tun_struct
*tun
,
387 struct hlist_head
*head
,
388 u32 rxhash
, u16 queue_index
)
390 struct tun_flow_entry
*e
= kmalloc(sizeof(*e
), GFP_ATOMIC
);
393 tun_debug(KERN_INFO
, tun
, "create flow: hash %u index %u\n",
394 rxhash
, queue_index
);
395 e
->updated
= jiffies
;
398 e
->queue_index
= queue_index
;
400 hlist_add_head_rcu(&e
->hash_link
, head
);
406 static void tun_flow_delete(struct tun_struct
*tun
, struct tun_flow_entry
*e
)
408 tun_debug(KERN_INFO
, tun
, "delete flow: hash %u index %u\n",
409 e
->rxhash
, e
->queue_index
);
410 hlist_del_rcu(&e
->hash_link
);
415 static void tun_flow_flush(struct tun_struct
*tun
)
419 spin_lock_bh(&tun
->lock
);
420 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
421 struct tun_flow_entry
*e
;
422 struct hlist_node
*n
;
424 hlist_for_each_entry_safe(e
, n
, &tun
->flows
[i
], hash_link
)
425 tun_flow_delete(tun
, e
);
427 spin_unlock_bh(&tun
->lock
);
430 static void tun_flow_delete_by_queue(struct tun_struct
*tun
, u16 queue_index
)
434 spin_lock_bh(&tun
->lock
);
435 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
436 struct tun_flow_entry
*e
;
437 struct hlist_node
*n
;
439 hlist_for_each_entry_safe(e
, n
, &tun
->flows
[i
], hash_link
) {
440 if (e
->queue_index
== queue_index
)
441 tun_flow_delete(tun
, e
);
444 spin_unlock_bh(&tun
->lock
);
447 static void tun_flow_cleanup(struct timer_list
*t
)
449 struct tun_struct
*tun
= from_timer(tun
, t
, flow_gc_timer
);
450 unsigned long delay
= tun
->ageing_time
;
451 unsigned long next_timer
= jiffies
+ delay
;
452 unsigned long count
= 0;
455 tun_debug(KERN_INFO
, tun
, "tun_flow_cleanup\n");
457 spin_lock(&tun
->lock
);
458 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
459 struct tun_flow_entry
*e
;
460 struct hlist_node
*n
;
462 hlist_for_each_entry_safe(e
, n
, &tun
->flows
[i
], hash_link
) {
463 unsigned long this_timer
;
465 this_timer
= e
->updated
+ delay
;
466 if (time_before_eq(this_timer
, jiffies
)) {
467 tun_flow_delete(tun
, e
);
471 if (time_before(this_timer
, next_timer
))
472 next_timer
= this_timer
;
477 mod_timer(&tun
->flow_gc_timer
, round_jiffies_up(next_timer
));
478 spin_unlock(&tun
->lock
);
481 static void tun_flow_update(struct tun_struct
*tun
, u32 rxhash
,
482 struct tun_file
*tfile
)
484 struct hlist_head
*head
;
485 struct tun_flow_entry
*e
;
486 unsigned long delay
= tun
->ageing_time
;
487 u16 queue_index
= tfile
->queue_index
;
492 head
= &tun
->flows
[tun_hashfn(rxhash
)];
496 /* We may get a very small possibility of OOO during switching, not
497 * worth to optimize.*/
498 if (tun
->numqueues
== 1 || tfile
->detached
)
501 e
= tun_flow_find(head
, rxhash
);
503 /* TODO: keep queueing to old queue until it's empty? */
504 e
->queue_index
= queue_index
;
505 e
->updated
= jiffies
;
506 sock_rps_record_flow_hash(e
->rps_rxhash
);
508 spin_lock_bh(&tun
->lock
);
509 if (!tun_flow_find(head
, rxhash
) &&
510 tun
->flow_count
< MAX_TAP_FLOWS
)
511 tun_flow_create(tun
, head
, rxhash
, queue_index
);
513 if (!timer_pending(&tun
->flow_gc_timer
))
514 mod_timer(&tun
->flow_gc_timer
,
515 round_jiffies_up(jiffies
+ delay
));
516 spin_unlock_bh(&tun
->lock
);
524 * Save the hash received in the stack receive path and update the
525 * flow_hash table accordingly.
527 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry
*e
, u32 hash
)
529 if (unlikely(e
->rps_rxhash
!= hash
))
530 e
->rps_rxhash
= hash
;
533 /* We try to identify a flow through its rxhash first. The reason that
534 * we do not check rxq no. is because some cards(e.g 82599), chooses
535 * the rxq based on the txq where the last packet of the flow comes. As
536 * the userspace application move between processors, we may get a
537 * different rxq no. here. If we could not get rxhash, then we would
538 * hope the rxq no. may help here.
540 static u16
tun_select_queue(struct net_device
*dev
, struct sk_buff
*skb
,
541 void *accel_priv
, select_queue_fallback_t fallback
)
543 struct tun_struct
*tun
= netdev_priv(dev
);
544 struct tun_flow_entry
*e
;
549 numqueues
= READ_ONCE(tun
->numqueues
);
551 txq
= __skb_get_hash_symmetric(skb
);
553 e
= tun_flow_find(&tun
->flows
[tun_hashfn(txq
)], txq
);
555 tun_flow_save_rps_rxhash(e
, txq
);
556 txq
= e
->queue_index
;
558 /* use multiply and shift instead of expensive divide */
559 txq
= ((u64
)txq
* numqueues
) >> 32;
560 } else if (likely(skb_rx_queue_recorded(skb
))) {
561 txq
= skb_get_rx_queue(skb
);
562 while (unlikely(txq
>= numqueues
))
570 static inline bool tun_not_capable(struct tun_struct
*tun
)
572 const struct cred
*cred
= current_cred();
573 struct net
*net
= dev_net(tun
->dev
);
575 return ((uid_valid(tun
->owner
) && !uid_eq(cred
->euid
, tun
->owner
)) ||
576 (gid_valid(tun
->group
) && !in_egroup_p(tun
->group
))) &&
577 !ns_capable(net
->user_ns
, CAP_NET_ADMIN
);
580 static void tun_set_real_num_queues(struct tun_struct
*tun
)
582 netif_set_real_num_tx_queues(tun
->dev
, tun
->numqueues
);
583 netif_set_real_num_rx_queues(tun
->dev
, tun
->numqueues
);
586 static void tun_disable_queue(struct tun_struct
*tun
, struct tun_file
*tfile
)
588 tfile
->detached
= tun
;
589 list_add_tail(&tfile
->next
, &tun
->disabled
);
593 static struct tun_struct
*tun_enable_queue(struct tun_file
*tfile
)
595 struct tun_struct
*tun
= tfile
->detached
;
597 tfile
->detached
= NULL
;
598 list_del_init(&tfile
->next
);
603 static void tun_queue_purge(struct tun_file
*tfile
)
607 while ((skb
= skb_array_consume(&tfile
->tx_array
)) != NULL
)
610 skb_queue_purge(&tfile
->sk
.sk_write_queue
);
611 skb_queue_purge(&tfile
->sk
.sk_error_queue
);
614 static void __tun_detach(struct tun_file
*tfile
, bool clean
)
616 struct tun_file
*ntfile
;
617 struct tun_struct
*tun
;
619 tun
= rtnl_dereference(tfile
->tun
);
622 tun_napi_disable(tun
, tfile
);
623 tun_napi_del(tun
, tfile
);
626 if (tun
&& !tfile
->detached
) {
627 u16 index
= tfile
->queue_index
;
628 BUG_ON(index
>= tun
->numqueues
);
630 rcu_assign_pointer(tun
->tfiles
[index
],
631 tun
->tfiles
[tun
->numqueues
- 1]);
632 ntfile
= rtnl_dereference(tun
->tfiles
[index
]);
633 ntfile
->queue_index
= index
;
637 RCU_INIT_POINTER(tfile
->tun
, NULL
);
638 sock_put(&tfile
->sk
);
640 tun_disable_queue(tun
, tfile
);
643 tun_flow_delete_by_queue(tun
, tun
->numqueues
+ 1);
644 /* Drop read queue */
645 tun_queue_purge(tfile
);
646 tun_set_real_num_queues(tun
);
647 } else if (tfile
->detached
&& clean
) {
648 tun
= tun_enable_queue(tfile
);
649 sock_put(&tfile
->sk
);
653 if (tun
&& tun
->numqueues
== 0 && tun
->numdisabled
== 0) {
654 netif_carrier_off(tun
->dev
);
656 if (!(tun
->flags
& IFF_PERSIST
) &&
657 tun
->dev
->reg_state
== NETREG_REGISTERED
)
658 unregister_netdevice(tun
->dev
);
660 sock_put(&tfile
->sk
);
664 static void tun_detach(struct tun_file
*tfile
, bool clean
)
667 __tun_detach(tfile
, clean
);
671 static void tun_detach_all(struct net_device
*dev
)
673 struct tun_struct
*tun
= netdev_priv(dev
);
674 struct bpf_prog
*xdp_prog
= rtnl_dereference(tun
->xdp_prog
);
675 struct tun_file
*tfile
, *tmp
;
676 int i
, n
= tun
->numqueues
;
678 for (i
= 0; i
< n
; i
++) {
679 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
681 tun_napi_disable(tun
, tfile
);
682 tfile
->socket
.sk
->sk_shutdown
= RCV_SHUTDOWN
;
683 tfile
->socket
.sk
->sk_data_ready(tfile
->socket
.sk
);
684 RCU_INIT_POINTER(tfile
->tun
, NULL
);
687 list_for_each_entry(tfile
, &tun
->disabled
, next
) {
688 tfile
->socket
.sk
->sk_shutdown
= RCV_SHUTDOWN
;
689 tfile
->socket
.sk
->sk_data_ready(tfile
->socket
.sk
);
690 RCU_INIT_POINTER(tfile
->tun
, NULL
);
692 BUG_ON(tun
->numqueues
!= 0);
695 for (i
= 0; i
< n
; i
++) {
696 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
697 tun_napi_del(tun
, tfile
);
698 /* Drop read queue */
699 tun_queue_purge(tfile
);
700 sock_put(&tfile
->sk
);
702 list_for_each_entry_safe(tfile
, tmp
, &tun
->disabled
, next
) {
703 tun_enable_queue(tfile
);
704 tun_queue_purge(tfile
);
705 sock_put(&tfile
->sk
);
707 BUG_ON(tun
->numdisabled
!= 0);
710 bpf_prog_put(xdp_prog
);
712 if (tun
->flags
& IFF_PERSIST
)
713 module_put(THIS_MODULE
);
716 static int tun_attach(struct tun_struct
*tun
, struct file
*file
,
717 bool skip_filter
, bool napi
)
719 struct tun_file
*tfile
= file
->private_data
;
720 struct net_device
*dev
= tun
->dev
;
723 err
= security_tun_dev_attach(tfile
->socket
.sk
, tun
->security
);
728 if (rtnl_dereference(tfile
->tun
) && !tfile
->detached
)
732 if (!(tun
->flags
& IFF_MULTI_QUEUE
) && tun
->numqueues
== 1)
736 if (!tfile
->detached
&&
737 tun
->numqueues
+ tun
->numdisabled
== MAX_TAP_QUEUES
)
742 /* Re-attach the filter to persist device */
743 if (!skip_filter
&& (tun
->filter_attached
== true)) {
744 lock_sock(tfile
->socket
.sk
);
745 err
= sk_attach_filter(&tun
->fprog
, tfile
->socket
.sk
);
746 release_sock(tfile
->socket
.sk
);
751 if (!tfile
->detached
&&
752 skb_array_resize(&tfile
->tx_array
, dev
->tx_queue_len
, GFP_KERNEL
)) {
757 tfile
->queue_index
= tun
->numqueues
;
758 tfile
->socket
.sk
->sk_shutdown
&= ~RCV_SHUTDOWN
;
759 rcu_assign_pointer(tfile
->tun
, tun
);
760 rcu_assign_pointer(tun
->tfiles
[tun
->numqueues
], tfile
);
763 if (tfile
->detached
) {
764 tun_enable_queue(tfile
);
766 sock_hold(&tfile
->sk
);
767 tun_napi_init(tun
, tfile
, napi
);
770 tun_set_real_num_queues(tun
);
772 /* device is allowed to go away first, so no need to hold extra
780 static struct tun_struct
*tun_get(struct tun_file
*tfile
)
782 struct tun_struct
*tun
;
785 tun
= rcu_dereference(tfile
->tun
);
793 static void tun_put(struct tun_struct
*tun
)
799 static void addr_hash_set(u32
*mask
, const u8
*addr
)
801 int n
= ether_crc(ETH_ALEN
, addr
) >> 26;
802 mask
[n
>> 5] |= (1 << (n
& 31));
805 static unsigned int addr_hash_test(const u32
*mask
, const u8
*addr
)
807 int n
= ether_crc(ETH_ALEN
, addr
) >> 26;
808 return mask
[n
>> 5] & (1 << (n
& 31));
811 static int update_filter(struct tap_filter
*filter
, void __user
*arg
)
813 struct { u8 u
[ETH_ALEN
]; } *addr
;
814 struct tun_filter uf
;
815 int err
, alen
, n
, nexact
;
817 if (copy_from_user(&uf
, arg
, sizeof(uf
)))
826 alen
= ETH_ALEN
* uf
.count
;
827 addr
= memdup_user(arg
+ sizeof(uf
), alen
);
829 return PTR_ERR(addr
);
831 /* The filter is updated without holding any locks. Which is
832 * perfectly safe. We disable it first and in the worst
833 * case we'll accept a few undesired packets. */
837 /* Use first set of addresses as an exact filter */
838 for (n
= 0; n
< uf
.count
&& n
< FLT_EXACT_COUNT
; n
++)
839 memcpy(filter
->addr
[n
], addr
[n
].u
, ETH_ALEN
);
843 /* Remaining multicast addresses are hashed,
844 * unicast will leave the filter disabled. */
845 memset(filter
->mask
, 0, sizeof(filter
->mask
));
846 for (; n
< uf
.count
; n
++) {
847 if (!is_multicast_ether_addr(addr
[n
].u
)) {
848 err
= 0; /* no filter */
851 addr_hash_set(filter
->mask
, addr
[n
].u
);
854 /* For ALLMULTI just set the mask to all ones.
855 * This overrides the mask populated above. */
856 if ((uf
.flags
& TUN_FLT_ALLMULTI
))
857 memset(filter
->mask
, ~0, sizeof(filter
->mask
));
859 /* Now enable the filter */
861 filter
->count
= nexact
;
863 /* Return the number of exact filters */
870 /* Returns: 0 - drop, !=0 - accept */
871 static int run_filter(struct tap_filter
*filter
, const struct sk_buff
*skb
)
873 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
875 struct ethhdr
*eh
= (struct ethhdr
*) skb
->data
;
879 for (i
= 0; i
< filter
->count
; i
++)
880 if (ether_addr_equal(eh
->h_dest
, filter
->addr
[i
]))
883 /* Inexact match (multicast only) */
884 if (is_multicast_ether_addr(eh
->h_dest
))
885 return addr_hash_test(filter
->mask
, eh
->h_dest
);
891 * Checks whether the packet is accepted or not.
892 * Returns: 0 - drop, !=0 - accept
894 static int check_filter(struct tap_filter
*filter
, const struct sk_buff
*skb
)
899 return run_filter(filter
, skb
);
902 /* Network device part of the driver */
904 static const struct ethtool_ops tun_ethtool_ops
;
906 /* Net device detach from fd. */
907 static void tun_net_uninit(struct net_device
*dev
)
912 /* Net device open. */
913 static int tun_net_open(struct net_device
*dev
)
915 struct tun_struct
*tun
= netdev_priv(dev
);
918 netif_tx_start_all_queues(dev
);
920 for (i
= 0; i
< tun
->numqueues
; i
++) {
921 struct tun_file
*tfile
;
923 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
924 tfile
->socket
.sk
->sk_write_space(tfile
->socket
.sk
);
930 /* Net device close. */
931 static int tun_net_close(struct net_device
*dev
)
933 netif_tx_stop_all_queues(dev
);
937 /* Net device start xmit */
938 static netdev_tx_t
tun_net_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
940 struct tun_struct
*tun
= netdev_priv(dev
);
941 int txq
= skb
->queue_mapping
;
942 struct tun_file
*tfile
;
946 tfile
= rcu_dereference(tun
->tfiles
[txq
]);
947 numqueues
= READ_ONCE(tun
->numqueues
);
949 /* Drop packet if interface is not attached */
950 if (txq
>= numqueues
)
954 if (numqueues
== 1 && static_key_false(&rps_needed
)) {
955 /* Select queue was not called for the skbuff, so we extract the
956 * RPS hash and save it into the flow_table here.
960 rxhash
= __skb_get_hash_symmetric(skb
);
962 struct tun_flow_entry
*e
;
963 e
= tun_flow_find(&tun
->flows
[tun_hashfn(rxhash
)],
966 tun_flow_save_rps_rxhash(e
, rxhash
);
971 tun_debug(KERN_INFO
, tun
, "tun_net_xmit %d\n", skb
->len
);
975 /* Drop if the filter does not like it.
976 * This is a noop if the filter is disabled.
977 * Filter can be enabled only for the TAP devices. */
978 if (!check_filter(&tun
->txflt
, skb
))
981 if (tfile
->socket
.sk
->sk_filter
&&
982 sk_filter(tfile
->socket
.sk
, skb
))
985 if (unlikely(skb_orphan_frags_rx(skb
, GFP_ATOMIC
)))
988 skb_tx_timestamp(skb
);
990 /* Orphan the skb - required as we might hang on to it
991 * for indefinite time.
997 if (skb_array_produce(&tfile
->tx_array
, skb
))
1000 /* Notify and wake up reader process */
1001 if (tfile
->flags
& TUN_FASYNC
)
1002 kill_fasync(&tfile
->fasync
, SIGIO
, POLL_IN
);
1003 tfile
->socket
.sk
->sk_data_ready(tfile
->socket
.sk
);
1006 return NETDEV_TX_OK
;
1009 this_cpu_inc(tun
->pcpu_stats
->tx_dropped
);
1013 return NET_XMIT_DROP
;
1016 static void tun_net_mclist(struct net_device
*dev
)
1019 * This callback is supposed to deal with mc filter in
1020 * _rx_ path and has nothing to do with the _tx_ path.
1021 * In rx path we always accept everything userspace gives us.
1025 static netdev_features_t
tun_net_fix_features(struct net_device
*dev
,
1026 netdev_features_t features
)
1028 struct tun_struct
*tun
= netdev_priv(dev
);
1030 return (features
& tun
->set_features
) | (features
& ~TUN_USER_FEATURES
);
1032 #ifdef CONFIG_NET_POLL_CONTROLLER
1033 static void tun_poll_controller(struct net_device
*dev
)
1036 * Tun only receives frames when:
1037 * 1) the char device endpoint gets data from user space
1038 * 2) the tun socket gets a sendmsg call from user space
1039 * If NAPI is not enabled, since both of those are synchronous
1040 * operations, we are guaranteed never to have pending data when we poll
1041 * for it so there is nothing to do here but return.
1042 * We need this though so netpoll recognizes us as an interface that
1043 * supports polling, which enables bridge devices in virt setups to
1044 * still use netconsole
1045 * If NAPI is enabled, however, we need to schedule polling for all
1046 * queues unless we are using napi_gro_frags(), which we call in
1047 * process context and not in NAPI context.
1049 struct tun_struct
*tun
= netdev_priv(dev
);
1051 if (tun
->flags
& IFF_NAPI
) {
1052 struct tun_file
*tfile
;
1055 if (tun_napi_frags_enabled(tun
))
1059 for (i
= 0; i
< tun
->numqueues
; i
++) {
1060 tfile
= rcu_dereference(tun
->tfiles
[i
]);
1061 if (tfile
->napi_enabled
)
1062 napi_schedule(&tfile
->napi
);
1070 static void tun_set_headroom(struct net_device
*dev
, int new_hr
)
1072 struct tun_struct
*tun
= netdev_priv(dev
);
1074 if (new_hr
< NET_SKB_PAD
)
1075 new_hr
= NET_SKB_PAD
;
1077 tun
->align
= new_hr
;
1081 tun_net_get_stats64(struct net_device
*dev
, struct rtnl_link_stats64
*stats
)
1083 u32 rx_dropped
= 0, tx_dropped
= 0, rx_frame_errors
= 0;
1084 struct tun_struct
*tun
= netdev_priv(dev
);
1085 struct tun_pcpu_stats
*p
;
1088 for_each_possible_cpu(i
) {
1089 u64 rxpackets
, rxbytes
, txpackets
, txbytes
;
1092 p
= per_cpu_ptr(tun
->pcpu_stats
, i
);
1094 start
= u64_stats_fetch_begin(&p
->syncp
);
1095 rxpackets
= p
->rx_packets
;
1096 rxbytes
= p
->rx_bytes
;
1097 txpackets
= p
->tx_packets
;
1098 txbytes
= p
->tx_bytes
;
1099 } while (u64_stats_fetch_retry(&p
->syncp
, start
));
1101 stats
->rx_packets
+= rxpackets
;
1102 stats
->rx_bytes
+= rxbytes
;
1103 stats
->tx_packets
+= txpackets
;
1104 stats
->tx_bytes
+= txbytes
;
1107 rx_dropped
+= p
->rx_dropped
;
1108 rx_frame_errors
+= p
->rx_frame_errors
;
1109 tx_dropped
+= p
->tx_dropped
;
1111 stats
->rx_dropped
= rx_dropped
;
1112 stats
->rx_frame_errors
= rx_frame_errors
;
1113 stats
->tx_dropped
= tx_dropped
;
1116 static int tun_xdp_set(struct net_device
*dev
, struct bpf_prog
*prog
,
1117 struct netlink_ext_ack
*extack
)
1119 struct tun_struct
*tun
= netdev_priv(dev
);
1120 struct bpf_prog
*old_prog
;
1122 old_prog
= rtnl_dereference(tun
->xdp_prog
);
1123 rcu_assign_pointer(tun
->xdp_prog
, prog
);
1125 bpf_prog_put(old_prog
);
1130 static u32
tun_xdp_query(struct net_device
*dev
)
1132 struct tun_struct
*tun
= netdev_priv(dev
);
1133 const struct bpf_prog
*xdp_prog
;
1135 xdp_prog
= rtnl_dereference(tun
->xdp_prog
);
1137 return xdp_prog
->aux
->id
;
1142 static int tun_xdp(struct net_device
*dev
, struct netdev_bpf
*xdp
)
1144 switch (xdp
->command
) {
1145 case XDP_SETUP_PROG
:
1146 return tun_xdp_set(dev
, xdp
->prog
, xdp
->extack
);
1147 case XDP_QUERY_PROG
:
1148 xdp
->prog_id
= tun_xdp_query(dev
);
1149 xdp
->prog_attached
= !!xdp
->prog_id
;
1156 static int tun_net_change_carrier(struct net_device
*dev
, bool new_carrier
)
1159 struct tun_struct
*tun
= netdev_priv(dev
);
1161 if (!tun
->numqueues
)
1164 netif_carrier_on(dev
);
1166 netif_carrier_off(dev
);
1171 static const struct net_device_ops tun_netdev_ops
= {
1172 .ndo_uninit
= tun_net_uninit
,
1173 .ndo_open
= tun_net_open
,
1174 .ndo_stop
= tun_net_close
,
1175 .ndo_start_xmit
= tun_net_xmit
,
1176 .ndo_fix_features
= tun_net_fix_features
,
1177 .ndo_select_queue
= tun_select_queue
,
1178 #ifdef CONFIG_NET_POLL_CONTROLLER
1179 .ndo_poll_controller
= tun_poll_controller
,
1181 .ndo_set_rx_headroom
= tun_set_headroom
,
1182 .ndo_get_stats64
= tun_net_get_stats64
,
1183 .ndo_change_carrier
= tun_net_change_carrier
,
1186 static const struct net_device_ops tap_netdev_ops
= {
1187 .ndo_uninit
= tun_net_uninit
,
1188 .ndo_open
= tun_net_open
,
1189 .ndo_stop
= tun_net_close
,
1190 .ndo_start_xmit
= tun_net_xmit
,
1191 .ndo_fix_features
= tun_net_fix_features
,
1192 .ndo_set_rx_mode
= tun_net_mclist
,
1193 .ndo_set_mac_address
= eth_mac_addr
,
1194 .ndo_validate_addr
= eth_validate_addr
,
1195 .ndo_select_queue
= tun_select_queue
,
1196 #ifdef CONFIG_NET_POLL_CONTROLLER
1197 .ndo_poll_controller
= tun_poll_controller
,
1199 .ndo_features_check
= passthru_features_check
,
1200 .ndo_set_rx_headroom
= tun_set_headroom
,
1201 .ndo_get_stats64
= tun_net_get_stats64
,
1203 .ndo_change_carrier
= tun_net_change_carrier
,
1206 static void tun_flow_init(struct tun_struct
*tun
)
1210 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++)
1211 INIT_HLIST_HEAD(&tun
->flows
[i
]);
1213 tun
->ageing_time
= TUN_FLOW_EXPIRE
;
1214 timer_setup(&tun
->flow_gc_timer
, tun_flow_cleanup
, 0);
1215 mod_timer(&tun
->flow_gc_timer
,
1216 round_jiffies_up(jiffies
+ tun
->ageing_time
));
1219 static void tun_flow_uninit(struct tun_struct
*tun
)
1221 del_timer_sync(&tun
->flow_gc_timer
);
1222 tun_flow_flush(tun
);
1226 #define MAX_MTU 65535
1228 /* Initialize net device. */
1229 static void tun_net_init(struct net_device
*dev
)
1231 struct tun_struct
*tun
= netdev_priv(dev
);
1233 switch (tun
->flags
& TUN_TYPE_MASK
) {
1235 dev
->netdev_ops
= &tun_netdev_ops
;
1237 /* Point-to-Point TUN Device */
1238 dev
->hard_header_len
= 0;
1242 /* Zero header length */
1243 dev
->type
= ARPHRD_NONE
;
1244 dev
->flags
= IFF_POINTOPOINT
| IFF_NOARP
| IFF_MULTICAST
;
1248 dev
->netdev_ops
= &tap_netdev_ops
;
1249 /* Ethernet TAP Device */
1251 dev
->priv_flags
&= ~IFF_TX_SKB_SHARING
;
1252 dev
->priv_flags
|= IFF_LIVE_ADDR_CHANGE
;
1254 eth_hw_addr_random(dev
);
1259 dev
->min_mtu
= MIN_MTU
;
1260 dev
->max_mtu
= MAX_MTU
- dev
->hard_header_len
;
1263 /* Character device part */
1266 static unsigned int tun_chr_poll(struct file
*file
, poll_table
*wait
)
1268 struct tun_file
*tfile
= file
->private_data
;
1269 struct tun_struct
*tun
= tun_get(tfile
);
1271 unsigned int mask
= 0;
1276 sk
= tfile
->socket
.sk
;
1278 tun_debug(KERN_INFO
, tun
, "tun_chr_poll\n");
1280 poll_wait(file
, sk_sleep(sk
), wait
);
1282 if (!skb_array_empty(&tfile
->tx_array
))
1283 mask
|= POLLIN
| POLLRDNORM
;
1285 if (tun
->dev
->flags
& IFF_UP
&&
1286 (sock_writeable(sk
) ||
1287 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
) &&
1288 sock_writeable(sk
))))
1289 mask
|= POLLOUT
| POLLWRNORM
;
1291 if (tun
->dev
->reg_state
!= NETREG_REGISTERED
)
1298 static struct sk_buff
*tun_napi_alloc_frags(struct tun_file
*tfile
,
1300 const struct iov_iter
*it
)
1302 struct sk_buff
*skb
;
1307 if (it
->nr_segs
> MAX_SKB_FRAGS
+ 1)
1308 return ERR_PTR(-ENOMEM
);
1311 skb
= napi_get_frags(&tfile
->napi
);
1314 return ERR_PTR(-ENOMEM
);
1316 linear
= iov_iter_single_seg_count(it
);
1317 err
= __skb_grow(skb
, linear
);
1322 skb
->data_len
= len
- linear
;
1323 skb
->truesize
+= skb
->data_len
;
1325 for (i
= 1; i
< it
->nr_segs
; i
++) {
1326 struct page_frag
*pfrag
= ¤t
->task_frag
;
1327 size_t fragsz
= it
->iov
[i
].iov_len
;
1329 if (fragsz
== 0 || fragsz
> PAGE_SIZE
) {
1334 if (!skb_page_frag_refill(fragsz
, pfrag
, GFP_KERNEL
)) {
1339 skb_fill_page_desc(skb
, i
- 1, pfrag
->page
,
1340 pfrag
->offset
, fragsz
);
1341 page_ref_inc(pfrag
->page
);
1342 pfrag
->offset
+= fragsz
;
1347 /* frees skb and all frags allocated with napi_alloc_frag() */
1348 napi_free_frags(&tfile
->napi
);
1349 return ERR_PTR(err
);
1352 /* prepad is the amount to reserve at front. len is length after that.
1353 * linear is a hint as to how much to copy (usually headers). */
1354 static struct sk_buff
*tun_alloc_skb(struct tun_file
*tfile
,
1355 size_t prepad
, size_t len
,
1356 size_t linear
, int noblock
)
1358 struct sock
*sk
= tfile
->socket
.sk
;
1359 struct sk_buff
*skb
;
1362 /* Under a page? Don't bother with paged skb. */
1363 if (prepad
+ len
< PAGE_SIZE
|| !linear
)
1366 skb
= sock_alloc_send_pskb(sk
, prepad
+ linear
, len
- linear
, noblock
,
1369 return ERR_PTR(err
);
1371 skb_reserve(skb
, prepad
);
1372 skb_put(skb
, linear
);
1373 skb
->data_len
= len
- linear
;
1374 skb
->len
+= len
- linear
;
1379 static void tun_rx_batched(struct tun_struct
*tun
, struct tun_file
*tfile
,
1380 struct sk_buff
*skb
, int more
)
1382 struct sk_buff_head
*queue
= &tfile
->sk
.sk_write_queue
;
1383 struct sk_buff_head process_queue
;
1384 u32 rx_batched
= tun
->rx_batched
;
1387 if (!rx_batched
|| (!more
&& skb_queue_empty(queue
))) {
1389 netif_receive_skb(skb
);
1394 spin_lock(&queue
->lock
);
1395 if (!more
|| skb_queue_len(queue
) == rx_batched
) {
1396 __skb_queue_head_init(&process_queue
);
1397 skb_queue_splice_tail_init(queue
, &process_queue
);
1400 __skb_queue_tail(queue
, skb
);
1402 spin_unlock(&queue
->lock
);
1405 struct sk_buff
*nskb
;
1408 while ((nskb
= __skb_dequeue(&process_queue
)))
1409 netif_receive_skb(nskb
);
1410 netif_receive_skb(skb
);
1415 static bool tun_can_build_skb(struct tun_struct
*tun
, struct tun_file
*tfile
,
1416 int len
, int noblock
, bool zerocopy
)
1418 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
1421 if (tfile
->socket
.sk
->sk_sndbuf
!= INT_MAX
)
1430 if (SKB_DATA_ALIGN(len
+ TUN_RX_PAD
) +
1431 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
)) > PAGE_SIZE
)
1437 static struct sk_buff
*tun_build_skb(struct tun_struct
*tun
,
1438 struct tun_file
*tfile
,
1439 struct iov_iter
*from
,
1440 struct virtio_net_hdr
*hdr
,
1441 int len
, int *skb_xdp
)
1443 struct page_frag
*alloc_frag
= ¤t
->task_frag
;
1444 struct sk_buff
*skb
;
1445 struct bpf_prog
*xdp_prog
;
1446 int buflen
= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
1447 unsigned int delta
= 0;
1450 bool xdp_xmit
= false;
1451 int err
, pad
= TUN_RX_PAD
;
1454 xdp_prog
= rcu_dereference(tun
->xdp_prog
);
1456 pad
+= TUN_HEADROOM
;
1457 buflen
+= SKB_DATA_ALIGN(len
+ pad
);
1460 alloc_frag
->offset
= ALIGN((u64
)alloc_frag
->offset
, SMP_CACHE_BYTES
);
1461 if (unlikely(!skb_page_frag_refill(buflen
, alloc_frag
, GFP_KERNEL
)))
1462 return ERR_PTR(-ENOMEM
);
1464 buf
= (char *)page_address(alloc_frag
->page
) + alloc_frag
->offset
;
1465 copied
= copy_page_from_iter(alloc_frag
->page
,
1466 alloc_frag
->offset
+ pad
,
1469 return ERR_PTR(-EFAULT
);
1471 /* There's a small window that XDP may be set after the check
1472 * of xdp_prog above, this should be rare and for simplicity
1473 * we do XDP on skb in case the headroom is not enough.
1475 if (hdr
->gso_type
|| !xdp_prog
)
1482 xdp_prog
= rcu_dereference(tun
->xdp_prog
);
1483 if (xdp_prog
&& !*skb_xdp
) {
1484 struct xdp_buff xdp
;
1488 xdp
.data_hard_start
= buf
;
1489 xdp
.data
= buf
+ pad
;
1490 xdp_set_data_meta_invalid(&xdp
);
1491 xdp
.data_end
= xdp
.data
+ len
;
1492 orig_data
= xdp
.data
;
1493 act
= bpf_prog_run_xdp(xdp_prog
, &xdp
);
1497 get_page(alloc_frag
->page
);
1498 alloc_frag
->offset
+= buflen
;
1499 err
= xdp_do_redirect(tun
->dev
, &xdp
, xdp_prog
);
1510 delta
= orig_data
- xdp
.data
;
1513 bpf_warn_invalid_xdp_action(act
);
1516 trace_xdp_exception(tun
->dev
, xdp_prog
, act
);
1523 skb
= build_skb(buf
, buflen
);
1527 return ERR_PTR(-ENOMEM
);
1530 skb_reserve(skb
, pad
- delta
);
1531 skb_put(skb
, len
+ delta
);
1532 get_page(alloc_frag
->page
);
1533 alloc_frag
->offset
+= buflen
;
1536 skb
->dev
= tun
->dev
;
1537 generic_xdp_tx(skb
, xdp_prog
);
1549 put_page(alloc_frag
->page
);
1553 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1557 /* Get packet from user space buffer */
1558 static ssize_t
tun_get_user(struct tun_struct
*tun
, struct tun_file
*tfile
,
1559 void *msg_control
, struct iov_iter
*from
,
1560 int noblock
, bool more
)
1562 struct tun_pi pi
= { 0, cpu_to_be16(ETH_P_IP
) };
1563 struct sk_buff
*skb
;
1564 size_t total_len
= iov_iter_count(from
);
1565 size_t len
= total_len
, align
= tun
->align
, linear
;
1566 struct virtio_net_hdr gso
= { 0 };
1567 struct tun_pcpu_stats
*stats
;
1570 bool zerocopy
= false;
1574 bool frags
= tun_napi_frags_enabled(tun
);
1576 if (!(tun
->dev
->flags
& IFF_UP
))
1579 if (!(tun
->flags
& IFF_NO_PI
)) {
1580 if (len
< sizeof(pi
))
1584 if (!copy_from_iter_full(&pi
, sizeof(pi
), from
))
1588 if (tun
->flags
& IFF_VNET_HDR
) {
1589 int vnet_hdr_sz
= READ_ONCE(tun
->vnet_hdr_sz
);
1591 if (len
< vnet_hdr_sz
)
1595 if (!copy_from_iter_full(&gso
, sizeof(gso
), from
))
1598 if ((gso
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) &&
1599 tun16_to_cpu(tun
, gso
.csum_start
) + tun16_to_cpu(tun
, gso
.csum_offset
) + 2 > tun16_to_cpu(tun
, gso
.hdr_len
))
1600 gso
.hdr_len
= cpu_to_tun16(tun
, tun16_to_cpu(tun
, gso
.csum_start
) + tun16_to_cpu(tun
, gso
.csum_offset
) + 2);
1602 if (tun16_to_cpu(tun
, gso
.hdr_len
) > len
)
1604 iov_iter_advance(from
, vnet_hdr_sz
- sizeof(gso
));
1607 if ((tun
->flags
& TUN_TYPE_MASK
) == IFF_TAP
) {
1608 align
+= NET_IP_ALIGN
;
1609 if (unlikely(len
< ETH_HLEN
||
1610 (gso
.hdr_len
&& tun16_to_cpu(tun
, gso
.hdr_len
) < ETH_HLEN
)))
1614 good_linear
= SKB_MAX_HEAD(align
);
1617 struct iov_iter i
= *from
;
1619 /* There are 256 bytes to be copied in skb, so there is
1620 * enough room for skb expand head in case it is used.
1621 * The rest of the buffer is mapped from userspace.
1623 copylen
= gso
.hdr_len
? tun16_to_cpu(tun
, gso
.hdr_len
) : GOODCOPY_LEN
;
1624 if (copylen
> good_linear
)
1625 copylen
= good_linear
;
1627 iov_iter_advance(&i
, copylen
);
1628 if (iov_iter_npages(&i
, INT_MAX
) <= MAX_SKB_FRAGS
)
1632 if (!frags
&& tun_can_build_skb(tun
, tfile
, len
, noblock
, zerocopy
)) {
1633 /* For the packet that is not easy to be processed
1634 * (e.g gso or jumbo packet), we will do it at after
1635 * skb was created with generic XDP routine.
1637 skb
= tun_build_skb(tun
, tfile
, from
, &gso
, len
, &skb_xdp
);
1639 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1640 return PTR_ERR(skb
);
1647 if (tun16_to_cpu(tun
, gso
.hdr_len
) > good_linear
)
1648 linear
= good_linear
;
1650 linear
= tun16_to_cpu(tun
, gso
.hdr_len
);
1654 mutex_lock(&tfile
->napi_mutex
);
1655 skb
= tun_napi_alloc_frags(tfile
, copylen
, from
);
1656 /* tun_napi_alloc_frags() enforces a layout for the skb.
1657 * If zerocopy is enabled, then this layout will be
1658 * overwritten by zerocopy_sg_from_iter().
1662 skb
= tun_alloc_skb(tfile
, align
, copylen
, linear
,
1667 if (PTR_ERR(skb
) != -EAGAIN
)
1668 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1670 mutex_unlock(&tfile
->napi_mutex
);
1671 return PTR_ERR(skb
);
1675 err
= zerocopy_sg_from_iter(skb
, from
);
1677 err
= skb_copy_datagram_from_iter(skb
, 0, from
, len
);
1680 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1683 tfile
->napi
.skb
= NULL
;
1684 mutex_unlock(&tfile
->napi_mutex
);
1691 if (virtio_net_hdr_to_skb(skb
, &gso
, tun_is_little_endian(tun
))) {
1692 this_cpu_inc(tun
->pcpu_stats
->rx_frame_errors
);
1695 tfile
->napi
.skb
= NULL
;
1696 mutex_unlock(&tfile
->napi_mutex
);
1702 switch (tun
->flags
& TUN_TYPE_MASK
) {
1704 if (tun
->flags
& IFF_NO_PI
) {
1705 u8 ip_version
= skb
->len
? (skb
->data
[0] >> 4) : 0;
1707 switch (ip_version
) {
1709 pi
.proto
= htons(ETH_P_IP
);
1712 pi
.proto
= htons(ETH_P_IPV6
);
1715 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1721 skb_reset_mac_header(skb
);
1722 skb
->protocol
= pi
.proto
;
1723 skb
->dev
= tun
->dev
;
1727 skb
->protocol
= eth_type_trans(skb
, tun
->dev
);
1731 /* copy skb_ubuf_info for callback when skb has no error */
1733 skb_shinfo(skb
)->destructor_arg
= msg_control
;
1734 skb_shinfo(skb
)->tx_flags
|= SKBTX_DEV_ZEROCOPY
;
1735 skb_shinfo(skb
)->tx_flags
|= SKBTX_SHARED_FRAG
;
1736 } else if (msg_control
) {
1737 struct ubuf_info
*uarg
= msg_control
;
1738 uarg
->callback(uarg
, false);
1741 skb_reset_network_header(skb
);
1742 skb_probe_transport_header(skb
, 0);
1745 struct bpf_prog
*xdp_prog
;
1750 xdp_prog
= rcu_dereference(tun
->xdp_prog
);
1752 ret
= do_xdp_generic(xdp_prog
, skb
);
1753 if (ret
!= XDP_PASS
) {
1763 rxhash
= __skb_get_hash_symmetric(skb
);
1766 /* Exercise flow dissector code path. */
1767 u32 headlen
= eth_get_headlen(skb
->data
, skb_headlen(skb
));
1769 if (unlikely(headlen
> skb_headlen(skb
))) {
1770 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1771 napi_free_frags(&tfile
->napi
);
1772 mutex_unlock(&tfile
->napi_mutex
);
1778 napi_gro_frags(&tfile
->napi
);
1780 mutex_unlock(&tfile
->napi_mutex
);
1781 } else if (tfile
->napi_enabled
) {
1782 struct sk_buff_head
*queue
= &tfile
->sk
.sk_write_queue
;
1785 spin_lock_bh(&queue
->lock
);
1786 __skb_queue_tail(queue
, skb
);
1787 queue_len
= skb_queue_len(queue
);
1788 spin_unlock(&queue
->lock
);
1790 if (!more
|| queue_len
> NAPI_POLL_WEIGHT
)
1791 napi_schedule(&tfile
->napi
);
1794 } else if (!IS_ENABLED(CONFIG_4KSTACKS
)) {
1795 tun_rx_batched(tun
, tfile
, skb
, more
);
1800 stats
= get_cpu_ptr(tun
->pcpu_stats
);
1801 u64_stats_update_begin(&stats
->syncp
);
1802 stats
->rx_packets
++;
1803 stats
->rx_bytes
+= len
;
1804 u64_stats_update_end(&stats
->syncp
);
1807 tun_flow_update(tun
, rxhash
, tfile
);
1811 static ssize_t
tun_chr_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
1813 struct file
*file
= iocb
->ki_filp
;
1814 struct tun_file
*tfile
= file
->private_data
;
1815 struct tun_struct
*tun
= tun_get(tfile
);
1821 result
= tun_get_user(tun
, tfile
, NULL
, from
,
1822 file
->f_flags
& O_NONBLOCK
, false);
1828 /* Put packet to the user space buffer */
1829 static ssize_t
tun_put_user(struct tun_struct
*tun
,
1830 struct tun_file
*tfile
,
1831 struct sk_buff
*skb
,
1832 struct iov_iter
*iter
)
1834 struct tun_pi pi
= { 0, skb
->protocol
};
1835 struct tun_pcpu_stats
*stats
;
1837 int vlan_offset
= 0;
1839 int vnet_hdr_sz
= 0;
1841 if (skb_vlan_tag_present(skb
))
1842 vlan_hlen
= VLAN_HLEN
;
1844 if (tun
->flags
& IFF_VNET_HDR
)
1845 vnet_hdr_sz
= READ_ONCE(tun
->vnet_hdr_sz
);
1847 total
= skb
->len
+ vlan_hlen
+ vnet_hdr_sz
;
1849 if (!(tun
->flags
& IFF_NO_PI
)) {
1850 if (iov_iter_count(iter
) < sizeof(pi
))
1853 total
+= sizeof(pi
);
1854 if (iov_iter_count(iter
) < total
) {
1855 /* Packet will be striped */
1856 pi
.flags
|= TUN_PKT_STRIP
;
1859 if (copy_to_iter(&pi
, sizeof(pi
), iter
) != sizeof(pi
))
1864 struct virtio_net_hdr gso
;
1866 if (iov_iter_count(iter
) < vnet_hdr_sz
)
1869 if (virtio_net_hdr_from_skb(skb
, &gso
,
1870 tun_is_little_endian(tun
), true,
1872 struct skb_shared_info
*sinfo
= skb_shinfo(skb
);
1873 pr_err("unexpected GSO type: "
1874 "0x%x, gso_size %d, hdr_len %d\n",
1875 sinfo
->gso_type
, tun16_to_cpu(tun
, gso
.gso_size
),
1876 tun16_to_cpu(tun
, gso
.hdr_len
));
1877 print_hex_dump(KERN_ERR
, "tun: ",
1880 min((int)tun16_to_cpu(tun
, gso
.hdr_len
), 64), true);
1885 if (copy_to_iter(&gso
, sizeof(gso
), iter
) != sizeof(gso
))
1888 iov_iter_advance(iter
, vnet_hdr_sz
- sizeof(gso
));
1894 __be16 h_vlan_proto
;
1898 veth
.h_vlan_proto
= skb
->vlan_proto
;
1899 veth
.h_vlan_TCI
= htons(skb_vlan_tag_get(skb
));
1901 vlan_offset
= offsetof(struct vlan_ethhdr
, h_vlan_proto
);
1903 ret
= skb_copy_datagram_iter(skb
, 0, iter
, vlan_offset
);
1904 if (ret
|| !iov_iter_count(iter
))
1907 ret
= copy_to_iter(&veth
, sizeof(veth
), iter
);
1908 if (ret
!= sizeof(veth
) || !iov_iter_count(iter
))
1912 skb_copy_datagram_iter(skb
, vlan_offset
, iter
, skb
->len
- vlan_offset
);
1915 /* caller is in process context, */
1916 stats
= get_cpu_ptr(tun
->pcpu_stats
);
1917 u64_stats_update_begin(&stats
->syncp
);
1918 stats
->tx_packets
++;
1919 stats
->tx_bytes
+= skb
->len
+ vlan_hlen
;
1920 u64_stats_update_end(&stats
->syncp
);
1921 put_cpu_ptr(tun
->pcpu_stats
);
1926 static struct sk_buff
*tun_ring_recv(struct tun_file
*tfile
, int noblock
,
1929 DECLARE_WAITQUEUE(wait
, current
);
1930 struct sk_buff
*skb
= NULL
;
1933 skb
= skb_array_consume(&tfile
->tx_array
);
1941 add_wait_queue(&tfile
->wq
.wait
, &wait
);
1942 current
->state
= TASK_INTERRUPTIBLE
;
1945 skb
= skb_array_consume(&tfile
->tx_array
);
1948 if (signal_pending(current
)) {
1949 error
= -ERESTARTSYS
;
1952 if (tfile
->socket
.sk
->sk_shutdown
& RCV_SHUTDOWN
) {
1960 current
->state
= TASK_RUNNING
;
1961 remove_wait_queue(&tfile
->wq
.wait
, &wait
);
1968 static ssize_t
tun_do_read(struct tun_struct
*tun
, struct tun_file
*tfile
,
1969 struct iov_iter
*to
,
1970 int noblock
, struct sk_buff
*skb
)
1975 tun_debug(KERN_INFO
, tun
, "tun_do_read\n");
1977 if (!iov_iter_count(to
)) {
1984 /* Read frames from ring */
1985 skb
= tun_ring_recv(tfile
, noblock
, &err
);
1990 ret
= tun_put_user(tun
, tfile
, skb
, to
);
1991 if (unlikely(ret
< 0))
1999 static ssize_t
tun_chr_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
2001 struct file
*file
= iocb
->ki_filp
;
2002 struct tun_file
*tfile
= file
->private_data
;
2003 struct tun_struct
*tun
= tun_get(tfile
);
2004 ssize_t len
= iov_iter_count(to
), ret
;
2008 ret
= tun_do_read(tun
, tfile
, to
, file
->f_flags
& O_NONBLOCK
, NULL
);
2009 ret
= min_t(ssize_t
, ret
, len
);
2016 static void tun_free_netdev(struct net_device
*dev
)
2018 struct tun_struct
*tun
= netdev_priv(dev
);
2020 BUG_ON(!(list_empty(&tun
->disabled
)));
2021 free_percpu(tun
->pcpu_stats
);
2022 tun_flow_uninit(tun
);
2023 security_tun_dev_free_security(tun
->security
);
2026 static void tun_setup(struct net_device
*dev
)
2028 struct tun_struct
*tun
= netdev_priv(dev
);
2030 tun
->owner
= INVALID_UID
;
2031 tun
->group
= INVALID_GID
;
2033 dev
->ethtool_ops
= &tun_ethtool_ops
;
2034 dev
->needs_free_netdev
= true;
2035 dev
->priv_destructor
= tun_free_netdev
;
2036 /* We prefer our own queue length */
2037 dev
->tx_queue_len
= TUN_READQ_SIZE
;
2040 /* Trivial set of netlink ops to allow deleting tun or tap
2041 * device with netlink.
2043 static int tun_validate(struct nlattr
*tb
[], struct nlattr
*data
[],
2044 struct netlink_ext_ack
*extack
)
2049 static struct rtnl_link_ops tun_link_ops __read_mostly
= {
2051 .priv_size
= sizeof(struct tun_struct
),
2053 .validate
= tun_validate
,
2056 static void tun_sock_write_space(struct sock
*sk
)
2058 struct tun_file
*tfile
;
2059 wait_queue_head_t
*wqueue
;
2061 if (!sock_writeable(sk
))
2064 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
))
2067 wqueue
= sk_sleep(sk
);
2068 if (wqueue
&& waitqueue_active(wqueue
))
2069 wake_up_interruptible_sync_poll(wqueue
, POLLOUT
|
2070 POLLWRNORM
| POLLWRBAND
);
2072 tfile
= container_of(sk
, struct tun_file
, sk
);
2073 kill_fasync(&tfile
->fasync
, SIGIO
, POLL_OUT
);
2076 static int tun_sendmsg(struct socket
*sock
, struct msghdr
*m
, size_t total_len
)
2079 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
2080 struct tun_struct
*tun
= tun_get(tfile
);
2085 ret
= tun_get_user(tun
, tfile
, m
->msg_control
, &m
->msg_iter
,
2086 m
->msg_flags
& MSG_DONTWAIT
,
2087 m
->msg_flags
& MSG_MORE
);
2092 static int tun_recvmsg(struct socket
*sock
, struct msghdr
*m
, size_t total_len
,
2095 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
2096 struct tun_struct
*tun
= tun_get(tfile
);
2097 struct sk_buff
*skb
= m
->msg_control
;
2105 if (flags
& ~(MSG_DONTWAIT
|MSG_TRUNC
|MSG_ERRQUEUE
)) {
2109 if (flags
& MSG_ERRQUEUE
) {
2110 ret
= sock_recv_errqueue(sock
->sk
, m
, total_len
,
2111 SOL_PACKET
, TUN_TX_TIMESTAMP
);
2114 ret
= tun_do_read(tun
, tfile
, &m
->msg_iter
, flags
& MSG_DONTWAIT
, skb
);
2115 if (ret
> (ssize_t
)total_len
) {
2116 m
->msg_flags
|= MSG_TRUNC
;
2117 ret
= flags
& MSG_TRUNC
? ret
: total_len
;
2131 static int tun_peek_len(struct socket
*sock
)
2133 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
2134 struct tun_struct
*tun
;
2137 tun
= tun_get(tfile
);
2141 ret
= skb_array_peek_len(&tfile
->tx_array
);
2147 /* Ops structure to mimic raw sockets with tun */
2148 static const struct proto_ops tun_socket_ops
= {
2149 .peek_len
= tun_peek_len
,
2150 .sendmsg
= tun_sendmsg
,
2151 .recvmsg
= tun_recvmsg
,
2154 static struct proto tun_proto
= {
2156 .owner
= THIS_MODULE
,
2157 .obj_size
= sizeof(struct tun_file
),
2160 static int tun_flags(struct tun_struct
*tun
)
2162 return tun
->flags
& (TUN_FEATURES
| IFF_PERSIST
| IFF_TUN
| IFF_TAP
);
2165 static ssize_t
tun_show_flags(struct device
*dev
, struct device_attribute
*attr
,
2168 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
2169 return sprintf(buf
, "0x%x\n", tun_flags(tun
));
2172 static ssize_t
tun_show_owner(struct device
*dev
, struct device_attribute
*attr
,
2175 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
2176 return uid_valid(tun
->owner
)?
2177 sprintf(buf
, "%u\n",
2178 from_kuid_munged(current_user_ns(), tun
->owner
)):
2179 sprintf(buf
, "-1\n");
2182 static ssize_t
tun_show_group(struct device
*dev
, struct device_attribute
*attr
,
2185 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
2186 return gid_valid(tun
->group
) ?
2187 sprintf(buf
, "%u\n",
2188 from_kgid_munged(current_user_ns(), tun
->group
)):
2189 sprintf(buf
, "-1\n");
2192 static DEVICE_ATTR(tun_flags
, 0444, tun_show_flags
, NULL
);
2193 static DEVICE_ATTR(owner
, 0444, tun_show_owner
, NULL
);
2194 static DEVICE_ATTR(group
, 0444, tun_show_group
, NULL
);
2196 static struct attribute
*tun_dev_attrs
[] = {
2197 &dev_attr_tun_flags
.attr
,
2198 &dev_attr_owner
.attr
,
2199 &dev_attr_group
.attr
,
2203 static const struct attribute_group tun_attr_group
= {
2204 .attrs
= tun_dev_attrs
2207 static int tun_set_iff(struct net
*net
, struct file
*file
, struct ifreq
*ifr
)
2209 struct tun_struct
*tun
;
2210 struct tun_file
*tfile
= file
->private_data
;
2211 struct net_device
*dev
;
2214 if (tfile
->detached
)
2217 if ((ifr
->ifr_flags
& IFF_NAPI_FRAGS
)) {
2218 if (!capable(CAP_NET_ADMIN
))
2221 if (!(ifr
->ifr_flags
& IFF_NAPI
) ||
2222 (ifr
->ifr_flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2226 dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
2228 if (ifr
->ifr_flags
& IFF_TUN_EXCL
)
2230 if ((ifr
->ifr_flags
& IFF_TUN
) && dev
->netdev_ops
== &tun_netdev_ops
)
2231 tun
= netdev_priv(dev
);
2232 else if ((ifr
->ifr_flags
& IFF_TAP
) && dev
->netdev_ops
== &tap_netdev_ops
)
2233 tun
= netdev_priv(dev
);
2237 if (!!(ifr
->ifr_flags
& IFF_MULTI_QUEUE
) !=
2238 !!(tun
->flags
& IFF_MULTI_QUEUE
))
2241 if (tun_not_capable(tun
))
2243 err
= security_tun_dev_open(tun
->security
);
2247 err
= tun_attach(tun
, file
, ifr
->ifr_flags
& IFF_NOFILTER
,
2248 ifr
->ifr_flags
& IFF_NAPI
);
2252 if (tun
->flags
& IFF_MULTI_QUEUE
&&
2253 (tun
->numqueues
+ tun
->numdisabled
> 1)) {
2254 /* One or more queue has already been attached, no need
2255 * to initialize the device again.
2262 unsigned long flags
= 0;
2263 int queues
= ifr
->ifr_flags
& IFF_MULTI_QUEUE
?
2266 if (!ns_capable(net
->user_ns
, CAP_NET_ADMIN
))
2268 err
= security_tun_dev_create();
2273 if (ifr
->ifr_flags
& IFF_TUN
) {
2277 } else if (ifr
->ifr_flags
& IFF_TAP
) {
2285 name
= ifr
->ifr_name
;
2287 dev
= alloc_netdev_mqs(sizeof(struct tun_struct
), name
,
2288 NET_NAME_UNKNOWN
, tun_setup
, queues
,
2293 err
= dev_get_valid_name(net
, dev
, name
);
2297 dev_net_set(dev
, net
);
2298 dev
->rtnl_link_ops
= &tun_link_ops
;
2299 dev
->ifindex
= tfile
->ifindex
;
2300 dev
->sysfs_groups
[0] = &tun_attr_group
;
2302 tun
= netdev_priv(dev
);
2305 tun
->txflt
.count
= 0;
2306 tun
->vnet_hdr_sz
= sizeof(struct virtio_net_hdr
);
2308 tun
->align
= NET_SKB_PAD
;
2309 tun
->filter_attached
= false;
2310 tun
->sndbuf
= tfile
->socket
.sk
->sk_sndbuf
;
2311 tun
->rx_batched
= 0;
2313 tun
->pcpu_stats
= netdev_alloc_pcpu_stats(struct tun_pcpu_stats
);
2314 if (!tun
->pcpu_stats
) {
2319 spin_lock_init(&tun
->lock
);
2321 err
= security_tun_dev_alloc_security(&tun
->security
);
2328 dev
->hw_features
= NETIF_F_SG
| NETIF_F_FRAGLIST
|
2329 TUN_USER_FEATURES
| NETIF_F_HW_VLAN_CTAG_TX
|
2330 NETIF_F_HW_VLAN_STAG_TX
;
2331 dev
->features
= dev
->hw_features
| NETIF_F_LLTX
;
2332 dev
->vlan_features
= dev
->features
&
2333 ~(NETIF_F_HW_VLAN_CTAG_TX
|
2334 NETIF_F_HW_VLAN_STAG_TX
);
2336 INIT_LIST_HEAD(&tun
->disabled
);
2337 err
= tun_attach(tun
, file
, false, ifr
->ifr_flags
& IFF_NAPI
);
2341 err
= register_netdevice(tun
->dev
);
2346 netif_carrier_on(tun
->dev
);
2348 tun_debug(KERN_INFO
, tun
, "tun_set_iff\n");
2350 tun
->flags
= (tun
->flags
& ~TUN_FEATURES
) |
2351 (ifr
->ifr_flags
& TUN_FEATURES
);
2353 /* Make sure persistent devices do not get stuck in
2356 if (netif_running(tun
->dev
))
2357 netif_tx_wake_all_queues(tun
->dev
);
2359 strcpy(ifr
->ifr_name
, tun
->dev
->name
);
2363 tun_detach_all(dev
);
2364 /* register_netdevice() already called tun_free_netdev() */
2368 tun_flow_uninit(tun
);
2369 security_tun_dev_free_security(tun
->security
);
2371 free_percpu(tun
->pcpu_stats
);
2377 static void tun_get_iff(struct net
*net
, struct tun_struct
*tun
,
2380 tun_debug(KERN_INFO
, tun
, "tun_get_iff\n");
2382 strcpy(ifr
->ifr_name
, tun
->dev
->name
);
2384 ifr
->ifr_flags
= tun_flags(tun
);
2388 /* This is like a cut-down ethtool ops, except done via tun fd so no
2389 * privs required. */
2390 static int set_offload(struct tun_struct
*tun
, unsigned long arg
)
2392 netdev_features_t features
= 0;
2394 if (arg
& TUN_F_CSUM
) {
2395 features
|= NETIF_F_HW_CSUM
;
2398 if (arg
& (TUN_F_TSO4
|TUN_F_TSO6
)) {
2399 if (arg
& TUN_F_TSO_ECN
) {
2400 features
|= NETIF_F_TSO_ECN
;
2401 arg
&= ~TUN_F_TSO_ECN
;
2403 if (arg
& TUN_F_TSO4
)
2404 features
|= NETIF_F_TSO
;
2405 if (arg
& TUN_F_TSO6
)
2406 features
|= NETIF_F_TSO6
;
2407 arg
&= ~(TUN_F_TSO4
|TUN_F_TSO6
);
2413 /* This gives the user a way to test for new features in future by
2414 * trying to set them. */
2418 tun
->set_features
= features
;
2419 tun
->dev
->wanted_features
&= ~TUN_USER_FEATURES
;
2420 tun
->dev
->wanted_features
|= features
;
2421 netdev_update_features(tun
->dev
);
2426 static void tun_detach_filter(struct tun_struct
*tun
, int n
)
2429 struct tun_file
*tfile
;
2431 for (i
= 0; i
< n
; i
++) {
2432 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
2433 lock_sock(tfile
->socket
.sk
);
2434 sk_detach_filter(tfile
->socket
.sk
);
2435 release_sock(tfile
->socket
.sk
);
2438 tun
->filter_attached
= false;
2441 static int tun_attach_filter(struct tun_struct
*tun
)
2444 struct tun_file
*tfile
;
2446 for (i
= 0; i
< tun
->numqueues
; i
++) {
2447 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
2448 lock_sock(tfile
->socket
.sk
);
2449 ret
= sk_attach_filter(&tun
->fprog
, tfile
->socket
.sk
);
2450 release_sock(tfile
->socket
.sk
);
2452 tun_detach_filter(tun
, i
);
2457 tun
->filter_attached
= true;
2461 static void tun_set_sndbuf(struct tun_struct
*tun
)
2463 struct tun_file
*tfile
;
2466 for (i
= 0; i
< tun
->numqueues
; i
++) {
2467 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
2468 tfile
->socket
.sk
->sk_sndbuf
= tun
->sndbuf
;
2472 static int tun_set_queue(struct file
*file
, struct ifreq
*ifr
)
2474 struct tun_file
*tfile
= file
->private_data
;
2475 struct tun_struct
*tun
;
2480 if (ifr
->ifr_flags
& IFF_ATTACH_QUEUE
) {
2481 tun
= tfile
->detached
;
2486 ret
= security_tun_dev_attach_queue(tun
->security
);
2489 ret
= tun_attach(tun
, file
, false, tun
->flags
& IFF_NAPI
);
2490 } else if (ifr
->ifr_flags
& IFF_DETACH_QUEUE
) {
2491 tun
= rtnl_dereference(tfile
->tun
);
2492 if (!tun
|| !(tun
->flags
& IFF_MULTI_QUEUE
) || tfile
->detached
)
2495 __tun_detach(tfile
, false);
2504 static long __tun_chr_ioctl(struct file
*file
, unsigned int cmd
,
2505 unsigned long arg
, int ifreq_len
)
2507 struct tun_file
*tfile
= file
->private_data
;
2508 struct tun_struct
*tun
;
2509 void __user
* argp
= (void __user
*)arg
;
2510 unsigned int ifindex
, carrier
;
2519 if (cmd
== TUNSETIFF
|| cmd
== TUNSETQUEUE
|| _IOC_TYPE(cmd
) == SOCK_IOC_TYPE
) {
2520 if (copy_from_user(&ifr
, argp
, ifreq_len
))
2523 memset(&ifr
, 0, sizeof(ifr
));
2525 if (cmd
== TUNGETFEATURES
) {
2526 /* Currently this just means: "what IFF flags are valid?".
2527 * This is needed because we never checked for invalid flags on
2530 return put_user(IFF_TUN
| IFF_TAP
| TUN_FEATURES
,
2531 (unsigned int __user
*)argp
);
2532 } else if (cmd
== TUNSETQUEUE
)
2533 return tun_set_queue(file
, &ifr
);
2538 tun
= tun_get(tfile
);
2539 if (cmd
== TUNSETIFF
) {
2544 ifr
.ifr_name
[IFNAMSIZ
-1] = '\0';
2546 ret
= tun_set_iff(sock_net(&tfile
->sk
), file
, &ifr
);
2551 if (copy_to_user(argp
, &ifr
, ifreq_len
))
2555 if (cmd
== TUNSETIFINDEX
) {
2561 if (copy_from_user(&ifindex
, argp
, sizeof(ifindex
)))
2565 tfile
->ifindex
= ifindex
;
2573 tun_debug(KERN_INFO
, tun
, "tun_chr_ioctl cmd %u\n", cmd
);
2578 tun_get_iff(current
->nsproxy
->net_ns
, tun
, &ifr
);
2580 if (tfile
->detached
)
2581 ifr
.ifr_flags
|= IFF_DETACH_QUEUE
;
2582 if (!tfile
->socket
.sk
->sk_filter
)
2583 ifr
.ifr_flags
|= IFF_NOFILTER
;
2585 if (copy_to_user(argp
, &ifr
, ifreq_len
))
2590 /* Disable/Enable checksum */
2592 /* [unimplemented] */
2593 tun_debug(KERN_INFO
, tun
, "ignored: set checksum %s\n",
2594 arg
? "disabled" : "enabled");
2598 /* Disable/Enable persist mode. Keep an extra reference to the
2599 * module to prevent the module being unprobed.
2601 if (arg
&& !(tun
->flags
& IFF_PERSIST
)) {
2602 tun
->flags
|= IFF_PERSIST
;
2603 __module_get(THIS_MODULE
);
2605 if (!arg
&& (tun
->flags
& IFF_PERSIST
)) {
2606 tun
->flags
&= ~IFF_PERSIST
;
2607 module_put(THIS_MODULE
);
2610 tun_debug(KERN_INFO
, tun
, "persist %s\n",
2611 arg
? "enabled" : "disabled");
2615 /* Set owner of the device */
2616 owner
= make_kuid(current_user_ns(), arg
);
2617 if (!uid_valid(owner
)) {
2622 tun_debug(KERN_INFO
, tun
, "owner set to %u\n",
2623 from_kuid(&init_user_ns
, tun
->owner
));
2627 /* Set group of the device */
2628 group
= make_kgid(current_user_ns(), arg
);
2629 if (!gid_valid(group
)) {
2634 tun_debug(KERN_INFO
, tun
, "group set to %u\n",
2635 from_kgid(&init_user_ns
, tun
->group
));
2639 /* Only allow setting the type when the interface is down */
2640 if (tun
->dev
->flags
& IFF_UP
) {
2641 tun_debug(KERN_INFO
, tun
,
2642 "Linktype set failed because interface is up\n");
2645 tun
->dev
->type
= (int) arg
;
2646 tun_debug(KERN_INFO
, tun
, "linktype set to %d\n",
2658 ret
= set_offload(tun
, arg
);
2661 case TUNSETTXFILTER
:
2662 /* Can be set only for TAPs */
2664 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2666 ret
= update_filter(&tun
->txflt
, (void __user
*)arg
);
2670 /* Get hw address */
2671 memcpy(ifr
.ifr_hwaddr
.sa_data
, tun
->dev
->dev_addr
, ETH_ALEN
);
2672 ifr
.ifr_hwaddr
.sa_family
= tun
->dev
->type
;
2673 if (copy_to_user(argp
, &ifr
, ifreq_len
))
2678 /* Set hw address */
2679 tun_debug(KERN_DEBUG
, tun
, "set hw address: %pM\n",
2680 ifr
.ifr_hwaddr
.sa_data
);
2682 ret
= dev_set_mac_address(tun
->dev
, &ifr
.ifr_hwaddr
);
2686 sndbuf
= tfile
->socket
.sk
->sk_sndbuf
;
2687 if (copy_to_user(argp
, &sndbuf
, sizeof(sndbuf
)))
2692 if (copy_from_user(&sndbuf
, argp
, sizeof(sndbuf
))) {
2701 tun
->sndbuf
= sndbuf
;
2702 tun_set_sndbuf(tun
);
2705 case TUNGETVNETHDRSZ
:
2706 vnet_hdr_sz
= tun
->vnet_hdr_sz
;
2707 if (copy_to_user(argp
, &vnet_hdr_sz
, sizeof(vnet_hdr_sz
)))
2711 case TUNSETVNETHDRSZ
:
2712 if (copy_from_user(&vnet_hdr_sz
, argp
, sizeof(vnet_hdr_sz
))) {
2716 if (vnet_hdr_sz
< (int)sizeof(struct virtio_net_hdr
)) {
2721 tun
->vnet_hdr_sz
= vnet_hdr_sz
;
2725 le
= !!(tun
->flags
& TUN_VNET_LE
);
2726 if (put_user(le
, (int __user
*)argp
))
2731 if (get_user(le
, (int __user
*)argp
)) {
2736 tun
->flags
|= TUN_VNET_LE
;
2738 tun
->flags
&= ~TUN_VNET_LE
;
2742 ret
= tun_get_vnet_be(tun
, argp
);
2746 ret
= tun_set_vnet_be(tun
, argp
);
2749 case TUNATTACHFILTER
:
2750 /* Can be set only for TAPs */
2752 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2755 if (copy_from_user(&tun
->fprog
, argp
, sizeof(tun
->fprog
)))
2758 ret
= tun_attach_filter(tun
);
2761 case TUNDETACHFILTER
:
2762 /* Can be set only for TAPs */
2764 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2767 tun_detach_filter(tun
, tun
->numqueues
);
2772 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2775 if (copy_to_user(argp
, &tun
->fprog
, sizeof(tun
->fprog
)))
2782 if (copy_from_user(&carrier
, argp
, sizeof(carrier
)))
2785 ret
= tun_net_change_carrier(tun
->dev
, (bool)carrier
);
2800 static long tun_chr_ioctl(struct file
*file
,
2801 unsigned int cmd
, unsigned long arg
)
2803 return __tun_chr_ioctl(file
, cmd
, arg
, sizeof (struct ifreq
));
2806 #ifdef CONFIG_COMPAT
2807 static long tun_chr_compat_ioctl(struct file
*file
,
2808 unsigned int cmd
, unsigned long arg
)
2813 case TUNSETTXFILTER
:
2818 arg
= (unsigned long)compat_ptr(arg
);
2821 arg
= (compat_ulong_t
)arg
;
2826 * compat_ifreq is shorter than ifreq, so we must not access beyond
2827 * the end of that structure. All fields that are used in this
2828 * driver are compatible though, we don't need to convert the
2831 return __tun_chr_ioctl(file
, cmd
, arg
, sizeof(struct compat_ifreq
));
2833 #endif /* CONFIG_COMPAT */
2835 static int tun_chr_fasync(int fd
, struct file
*file
, int on
)
2837 struct tun_file
*tfile
= file
->private_data
;
2840 if ((ret
= fasync_helper(fd
, file
, on
, &tfile
->fasync
)) < 0)
2844 __f_setown(file
, task_pid(current
), PIDTYPE_PID
, 0);
2845 tfile
->flags
|= TUN_FASYNC
;
2847 tfile
->flags
&= ~TUN_FASYNC
;
2853 static int tun_chr_open(struct inode
*inode
, struct file
* file
)
2855 struct net
*net
= current
->nsproxy
->net_ns
;
2856 struct tun_file
*tfile
;
2858 DBG1(KERN_INFO
, "tunX: tun_chr_open\n");
2860 tfile
= (struct tun_file
*)sk_alloc(net
, AF_UNSPEC
, GFP_KERNEL
,
2864 if (skb_array_init(&tfile
->tx_array
, 0, GFP_KERNEL
)) {
2865 sk_free(&tfile
->sk
);
2869 RCU_INIT_POINTER(tfile
->tun
, NULL
);
2873 init_waitqueue_head(&tfile
->wq
.wait
);
2874 RCU_INIT_POINTER(tfile
->socket
.wq
, &tfile
->wq
);
2876 tfile
->socket
.file
= file
;
2877 tfile
->socket
.ops
= &tun_socket_ops
;
2879 sock_init_data(&tfile
->socket
, &tfile
->sk
);
2881 tfile
->sk
.sk_write_space
= tun_sock_write_space
;
2882 tfile
->sk
.sk_sndbuf
= INT_MAX
;
2884 file
->private_data
= tfile
;
2885 INIT_LIST_HEAD(&tfile
->next
);
2887 sock_set_flag(&tfile
->sk
, SOCK_ZEROCOPY
);
2892 static int tun_chr_close(struct inode
*inode
, struct file
*file
)
2894 struct tun_file
*tfile
= file
->private_data
;
2896 tun_detach(tfile
, true);
2897 skb_array_cleanup(&tfile
->tx_array
);
2902 #ifdef CONFIG_PROC_FS
2903 static void tun_chr_show_fdinfo(struct seq_file
*m
, struct file
*file
)
2905 struct tun_file
*tfile
= file
->private_data
;
2906 struct tun_struct
*tun
;
2909 memset(&ifr
, 0, sizeof(ifr
));
2912 tun
= tun_get(tfile
);
2914 tun_get_iff(current
->nsproxy
->net_ns
, tun
, &ifr
);
2920 seq_printf(m
, "iff:\t%s\n", ifr
.ifr_name
);
2924 static const struct file_operations tun_fops
= {
2925 .owner
= THIS_MODULE
,
2926 .llseek
= no_llseek
,
2927 .read_iter
= tun_chr_read_iter
,
2928 .write_iter
= tun_chr_write_iter
,
2929 .poll
= tun_chr_poll
,
2930 .unlocked_ioctl
= tun_chr_ioctl
,
2931 #ifdef CONFIG_COMPAT
2932 .compat_ioctl
= tun_chr_compat_ioctl
,
2934 .open
= tun_chr_open
,
2935 .release
= tun_chr_close
,
2936 .fasync
= tun_chr_fasync
,
2937 #ifdef CONFIG_PROC_FS
2938 .show_fdinfo
= tun_chr_show_fdinfo
,
2942 static struct miscdevice tun_miscdev
= {
2945 .nodename
= "net/tun",
2949 /* ethtool interface */
2951 static int tun_get_link_ksettings(struct net_device
*dev
,
2952 struct ethtool_link_ksettings
*cmd
)
2954 ethtool_link_ksettings_zero_link_mode(cmd
, supported
);
2955 ethtool_link_ksettings_zero_link_mode(cmd
, advertising
);
2956 cmd
->base
.speed
= SPEED_10
;
2957 cmd
->base
.duplex
= DUPLEX_FULL
;
2958 cmd
->base
.port
= PORT_TP
;
2959 cmd
->base
.phy_address
= 0;
2960 cmd
->base
.autoneg
= AUTONEG_DISABLE
;
2964 static void tun_get_drvinfo(struct net_device
*dev
, struct ethtool_drvinfo
*info
)
2966 struct tun_struct
*tun
= netdev_priv(dev
);
2968 strlcpy(info
->driver
, DRV_NAME
, sizeof(info
->driver
));
2969 strlcpy(info
->version
, DRV_VERSION
, sizeof(info
->version
));
2971 switch (tun
->flags
& TUN_TYPE_MASK
) {
2973 strlcpy(info
->bus_info
, "tun", sizeof(info
->bus_info
));
2976 strlcpy(info
->bus_info
, "tap", sizeof(info
->bus_info
));
2981 static u32
tun_get_msglevel(struct net_device
*dev
)
2984 struct tun_struct
*tun
= netdev_priv(dev
);
2991 static void tun_set_msglevel(struct net_device
*dev
, u32 value
)
2994 struct tun_struct
*tun
= netdev_priv(dev
);
2999 static int tun_get_coalesce(struct net_device
*dev
,
3000 struct ethtool_coalesce
*ec
)
3002 struct tun_struct
*tun
= netdev_priv(dev
);
3004 ec
->rx_max_coalesced_frames
= tun
->rx_batched
;
3009 static int tun_set_coalesce(struct net_device
*dev
,
3010 struct ethtool_coalesce
*ec
)
3012 struct tun_struct
*tun
= netdev_priv(dev
);
3014 if (ec
->rx_max_coalesced_frames
> NAPI_POLL_WEIGHT
)
3015 tun
->rx_batched
= NAPI_POLL_WEIGHT
;
3017 tun
->rx_batched
= ec
->rx_max_coalesced_frames
;
3022 static const struct ethtool_ops tun_ethtool_ops
= {
3023 .get_drvinfo
= tun_get_drvinfo
,
3024 .get_msglevel
= tun_get_msglevel
,
3025 .set_msglevel
= tun_set_msglevel
,
3026 .get_link
= ethtool_op_get_link
,
3027 .get_ts_info
= ethtool_op_get_ts_info
,
3028 .get_coalesce
= tun_get_coalesce
,
3029 .set_coalesce
= tun_set_coalesce
,
3030 .get_link_ksettings
= tun_get_link_ksettings
,
3033 static int tun_queue_resize(struct tun_struct
*tun
)
3035 struct net_device
*dev
= tun
->dev
;
3036 struct tun_file
*tfile
;
3037 struct skb_array
**arrays
;
3038 int n
= tun
->numqueues
+ tun
->numdisabled
;
3041 arrays
= kmalloc_array(n
, sizeof(*arrays
), GFP_KERNEL
);
3045 for (i
= 0; i
< tun
->numqueues
; i
++) {
3046 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
3047 arrays
[i
] = &tfile
->tx_array
;
3049 list_for_each_entry(tfile
, &tun
->disabled
, next
)
3050 arrays
[i
++] = &tfile
->tx_array
;
3052 ret
= skb_array_resize_multiple(arrays
, n
,
3053 dev
->tx_queue_len
, GFP_KERNEL
);
3059 static int tun_device_event(struct notifier_block
*unused
,
3060 unsigned long event
, void *ptr
)
3062 struct net_device
*dev
= netdev_notifier_info_to_dev(ptr
);
3063 struct tun_struct
*tun
= netdev_priv(dev
);
3065 if (dev
->rtnl_link_ops
!= &tun_link_ops
)
3069 case NETDEV_CHANGE_TX_QUEUE_LEN
:
3070 if (tun_queue_resize(tun
))
3080 static struct notifier_block tun_notifier_block __read_mostly
= {
3081 .notifier_call
= tun_device_event
,
3084 static int __init
tun_init(void)
3088 pr_info("%s, %s\n", DRV_DESCRIPTION
, DRV_VERSION
);
3090 ret
= rtnl_link_register(&tun_link_ops
);
3092 pr_err("Can't register link_ops\n");
3096 ret
= misc_register(&tun_miscdev
);
3098 pr_err("Can't register misc device %d\n", TUN_MINOR
);
3102 ret
= register_netdevice_notifier(&tun_notifier_block
);
3104 pr_err("Can't register netdevice notifier\n");
3111 misc_deregister(&tun_miscdev
);
3113 rtnl_link_unregister(&tun_link_ops
);
3118 static void tun_cleanup(void)
3120 misc_deregister(&tun_miscdev
);
3121 rtnl_link_unregister(&tun_link_ops
);
3122 unregister_netdevice_notifier(&tun_notifier_block
);
3125 /* Get an underlying socket object from tun file. Returns error unless file is
3126 * attached to a device. The returned object works like a packet socket, it
3127 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
3128 * holding a reference to the file for as long as the socket is in use. */
3129 struct socket
*tun_get_socket(struct file
*file
)
3131 struct tun_file
*tfile
;
3132 if (file
->f_op
!= &tun_fops
)
3133 return ERR_PTR(-EINVAL
);
3134 tfile
= file
->private_data
;
3136 return ERR_PTR(-EBADFD
);
3137 return &tfile
->socket
;
3139 EXPORT_SYMBOL_GPL(tun_get_socket
);
3141 struct skb_array
*tun_get_skb_array(struct file
*file
)
3143 struct tun_file
*tfile
;
3145 if (file
->f_op
!= &tun_fops
)
3146 return ERR_PTR(-EINVAL
);
3147 tfile
= file
->private_data
;
3149 return ERR_PTR(-EBADFD
);
3150 return &tfile
->tx_array
;
3152 EXPORT_SYMBOL_GPL(tun_get_skb_array
);
3154 module_init(tun_init
);
3155 module_exit(tun_cleanup
);
3156 MODULE_DESCRIPTION(DRV_DESCRIPTION
);
3157 MODULE_AUTHOR(DRV_COPYRIGHT
);
3158 MODULE_LICENSE("GPL");
3159 MODULE_ALIAS_MISCDEV(TUN_MINOR
);
3160 MODULE_ALIAS("devname:net/tun");