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 bool napi_frags_enabled
;
180 struct mutex napi_mutex
; /* Protects access to the above napi */
181 struct list_head next
;
182 struct tun_struct
*detached
;
183 struct skb_array tx_array
;
186 struct tun_flow_entry
{
187 struct hlist_node hash_link
;
189 struct tun_struct
*tun
;
194 unsigned long updated
;
197 #define TUN_NUM_FLOW_ENTRIES 1024
199 /* Since the socket were moved to tun_file, to preserve the behavior of persist
200 * device, socket filter, sndbuf and vnet header size were restore when the
201 * file were attached to a persist device.
204 struct tun_file __rcu
*tfiles
[MAX_TAP_QUEUES
];
205 unsigned int numqueues
;
210 struct net_device
*dev
;
211 netdev_features_t set_features
;
212 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
218 struct tap_filter txflt
;
219 struct sock_fprog fprog
;
220 /* protected by rtnl lock */
221 bool filter_attached
;
226 struct hlist_head flows
[TUN_NUM_FLOW_ENTRIES
];
227 struct timer_list flow_gc_timer
;
228 unsigned long ageing_time
;
229 unsigned int numdisabled
;
230 struct list_head disabled
;
234 struct tun_pcpu_stats __percpu
*pcpu_stats
;
235 struct bpf_prog __rcu
*xdp_prog
;
238 static int tun_napi_receive(struct napi_struct
*napi
, int budget
)
240 struct tun_file
*tfile
= container_of(napi
, struct tun_file
, napi
);
241 struct sk_buff_head
*queue
= &tfile
->sk
.sk_write_queue
;
242 struct sk_buff_head process_queue
;
246 __skb_queue_head_init(&process_queue
);
248 spin_lock(&queue
->lock
);
249 skb_queue_splice_tail_init(queue
, &process_queue
);
250 spin_unlock(&queue
->lock
);
252 while (received
< budget
&& (skb
= __skb_dequeue(&process_queue
))) {
253 napi_gro_receive(napi
, skb
);
257 if (!skb_queue_empty(&process_queue
)) {
258 spin_lock(&queue
->lock
);
259 skb_queue_splice(&process_queue
, queue
);
260 spin_unlock(&queue
->lock
);
266 static int tun_napi_poll(struct napi_struct
*napi
, int budget
)
268 unsigned int received
;
270 received
= tun_napi_receive(napi
, budget
);
272 if (received
< budget
)
273 napi_complete_done(napi
, received
);
278 static void tun_napi_init(struct tun_struct
*tun
, struct tun_file
*tfile
,
279 bool napi_en
, bool napi_frags
)
281 tfile
->napi_enabled
= napi_en
;
282 tfile
->napi_frags_enabled
= napi_en
&& napi_frags
;
284 netif_napi_add(tun
->dev
, &tfile
->napi
, tun_napi_poll
,
286 napi_enable(&tfile
->napi
);
290 static void tun_napi_disable(struct tun_file
*tfile
)
292 if (tfile
->napi_enabled
)
293 napi_disable(&tfile
->napi
);
296 static void tun_napi_del(struct tun_file
*tfile
)
298 if (tfile
->napi_enabled
)
299 netif_napi_del(&tfile
->napi
);
302 static bool tun_napi_frags_enabled(const struct tun_file
*tfile
)
304 return tfile
->napi_frags_enabled
;
307 #ifdef CONFIG_TUN_VNET_CROSS_LE
308 static inline bool tun_legacy_is_little_endian(struct tun_struct
*tun
)
310 return tun
->flags
& TUN_VNET_BE
? false :
311 virtio_legacy_is_little_endian();
314 static long tun_get_vnet_be(struct tun_struct
*tun
, int __user
*argp
)
316 int be
= !!(tun
->flags
& TUN_VNET_BE
);
318 if (put_user(be
, argp
))
324 static long tun_set_vnet_be(struct tun_struct
*tun
, int __user
*argp
)
328 if (get_user(be
, argp
))
332 tun
->flags
|= TUN_VNET_BE
;
334 tun
->flags
&= ~TUN_VNET_BE
;
339 static inline bool tun_legacy_is_little_endian(struct tun_struct
*tun
)
341 return virtio_legacy_is_little_endian();
344 static long tun_get_vnet_be(struct tun_struct
*tun
, int __user
*argp
)
349 static long tun_set_vnet_be(struct tun_struct
*tun
, int __user
*argp
)
353 #endif /* CONFIG_TUN_VNET_CROSS_LE */
355 static inline bool tun_is_little_endian(struct tun_struct
*tun
)
357 return tun
->flags
& TUN_VNET_LE
||
358 tun_legacy_is_little_endian(tun
);
361 static inline u16
tun16_to_cpu(struct tun_struct
*tun
, __virtio16 val
)
363 return __virtio16_to_cpu(tun_is_little_endian(tun
), val
);
366 static inline __virtio16
cpu_to_tun16(struct tun_struct
*tun
, u16 val
)
368 return __cpu_to_virtio16(tun_is_little_endian(tun
), val
);
371 static inline u32
tun_hashfn(u32 rxhash
)
373 return rxhash
& 0x3ff;
376 static struct tun_flow_entry
*tun_flow_find(struct hlist_head
*head
, u32 rxhash
)
378 struct tun_flow_entry
*e
;
380 hlist_for_each_entry_rcu(e
, head
, hash_link
) {
381 if (e
->rxhash
== rxhash
)
387 static struct tun_flow_entry
*tun_flow_create(struct tun_struct
*tun
,
388 struct hlist_head
*head
,
389 u32 rxhash
, u16 queue_index
)
391 struct tun_flow_entry
*e
= kmalloc(sizeof(*e
), GFP_ATOMIC
);
394 tun_debug(KERN_INFO
, tun
, "create flow: hash %u index %u\n",
395 rxhash
, queue_index
);
396 e
->updated
= jiffies
;
399 e
->queue_index
= queue_index
;
401 hlist_add_head_rcu(&e
->hash_link
, head
);
407 static void tun_flow_delete(struct tun_struct
*tun
, struct tun_flow_entry
*e
)
409 tun_debug(KERN_INFO
, tun
, "delete flow: hash %u index %u\n",
410 e
->rxhash
, e
->queue_index
);
411 hlist_del_rcu(&e
->hash_link
);
416 static void tun_flow_flush(struct tun_struct
*tun
)
420 spin_lock_bh(&tun
->lock
);
421 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
422 struct tun_flow_entry
*e
;
423 struct hlist_node
*n
;
425 hlist_for_each_entry_safe(e
, n
, &tun
->flows
[i
], hash_link
)
426 tun_flow_delete(tun
, e
);
428 spin_unlock_bh(&tun
->lock
);
431 static void tun_flow_delete_by_queue(struct tun_struct
*tun
, u16 queue_index
)
435 spin_lock_bh(&tun
->lock
);
436 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
437 struct tun_flow_entry
*e
;
438 struct hlist_node
*n
;
440 hlist_for_each_entry_safe(e
, n
, &tun
->flows
[i
], hash_link
) {
441 if (e
->queue_index
== queue_index
)
442 tun_flow_delete(tun
, e
);
445 spin_unlock_bh(&tun
->lock
);
448 static void tun_flow_cleanup(struct timer_list
*t
)
450 struct tun_struct
*tun
= from_timer(tun
, t
, flow_gc_timer
);
451 unsigned long delay
= tun
->ageing_time
;
452 unsigned long next_timer
= jiffies
+ delay
;
453 unsigned long count
= 0;
456 tun_debug(KERN_INFO
, tun
, "tun_flow_cleanup\n");
458 spin_lock(&tun
->lock
);
459 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
460 struct tun_flow_entry
*e
;
461 struct hlist_node
*n
;
463 hlist_for_each_entry_safe(e
, n
, &tun
->flows
[i
], hash_link
) {
464 unsigned long this_timer
;
466 this_timer
= e
->updated
+ delay
;
467 if (time_before_eq(this_timer
, jiffies
)) {
468 tun_flow_delete(tun
, e
);
472 if (time_before(this_timer
, next_timer
))
473 next_timer
= this_timer
;
478 mod_timer(&tun
->flow_gc_timer
, round_jiffies_up(next_timer
));
479 spin_unlock(&tun
->lock
);
482 static void tun_flow_update(struct tun_struct
*tun
, u32 rxhash
,
483 struct tun_file
*tfile
)
485 struct hlist_head
*head
;
486 struct tun_flow_entry
*e
;
487 unsigned long delay
= tun
->ageing_time
;
488 u16 queue_index
= tfile
->queue_index
;
493 head
= &tun
->flows
[tun_hashfn(rxhash
)];
497 /* We may get a very small possibility of OOO during switching, not
498 * worth to optimize.*/
499 if (tun
->numqueues
== 1 || tfile
->detached
)
502 e
= tun_flow_find(head
, rxhash
);
504 /* TODO: keep queueing to old queue until it's empty? */
505 e
->queue_index
= queue_index
;
506 e
->updated
= jiffies
;
507 sock_rps_record_flow_hash(e
->rps_rxhash
);
509 spin_lock_bh(&tun
->lock
);
510 if (!tun_flow_find(head
, rxhash
) &&
511 tun
->flow_count
< MAX_TAP_FLOWS
)
512 tun_flow_create(tun
, head
, rxhash
, queue_index
);
514 if (!timer_pending(&tun
->flow_gc_timer
))
515 mod_timer(&tun
->flow_gc_timer
,
516 round_jiffies_up(jiffies
+ delay
));
517 spin_unlock_bh(&tun
->lock
);
525 * Save the hash received in the stack receive path and update the
526 * flow_hash table accordingly.
528 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry
*e
, u32 hash
)
530 if (unlikely(e
->rps_rxhash
!= hash
))
531 e
->rps_rxhash
= hash
;
534 /* We try to identify a flow through its rxhash first. The reason that
535 * we do not check rxq no. is because some cards(e.g 82599), chooses
536 * the rxq based on the txq where the last packet of the flow comes. As
537 * the userspace application move between processors, we may get a
538 * different rxq no. here. If we could not get rxhash, then we would
539 * hope the rxq no. may help here.
541 static u16
tun_select_queue(struct net_device
*dev
, struct sk_buff
*skb
,
542 void *accel_priv
, select_queue_fallback_t fallback
)
544 struct tun_struct
*tun
= netdev_priv(dev
);
545 struct tun_flow_entry
*e
;
550 numqueues
= READ_ONCE(tun
->numqueues
);
552 txq
= __skb_get_hash_symmetric(skb
);
554 e
= tun_flow_find(&tun
->flows
[tun_hashfn(txq
)], txq
);
556 tun_flow_save_rps_rxhash(e
, txq
);
557 txq
= e
->queue_index
;
559 /* use multiply and shift instead of expensive divide */
560 txq
= ((u64
)txq
* numqueues
) >> 32;
561 } else if (likely(skb_rx_queue_recorded(skb
))) {
562 txq
= skb_get_rx_queue(skb
);
563 while (unlikely(txq
>= numqueues
))
571 static inline bool tun_not_capable(struct tun_struct
*tun
)
573 const struct cred
*cred
= current_cred();
574 struct net
*net
= dev_net(tun
->dev
);
576 return ((uid_valid(tun
->owner
) && !uid_eq(cred
->euid
, tun
->owner
)) ||
577 (gid_valid(tun
->group
) && !in_egroup_p(tun
->group
))) &&
578 !ns_capable(net
->user_ns
, CAP_NET_ADMIN
);
581 static void tun_set_real_num_queues(struct tun_struct
*tun
)
583 netif_set_real_num_tx_queues(tun
->dev
, tun
->numqueues
);
584 netif_set_real_num_rx_queues(tun
->dev
, tun
->numqueues
);
587 static void tun_disable_queue(struct tun_struct
*tun
, struct tun_file
*tfile
)
589 tfile
->detached
= tun
;
590 list_add_tail(&tfile
->next
, &tun
->disabled
);
594 static struct tun_struct
*tun_enable_queue(struct tun_file
*tfile
)
596 struct tun_struct
*tun
= tfile
->detached
;
598 tfile
->detached
= NULL
;
599 list_del_init(&tfile
->next
);
604 static void tun_queue_purge(struct tun_file
*tfile
)
608 while ((skb
= skb_array_consume(&tfile
->tx_array
)) != NULL
)
611 skb_queue_purge(&tfile
->sk
.sk_write_queue
);
612 skb_queue_purge(&tfile
->sk
.sk_error_queue
);
615 static void __tun_detach(struct tun_file
*tfile
, bool clean
)
617 struct tun_file
*ntfile
;
618 struct tun_struct
*tun
;
620 tun
= rtnl_dereference(tfile
->tun
);
623 tun_napi_disable(tfile
);
627 if (tun
&& !tfile
->detached
) {
628 u16 index
= tfile
->queue_index
;
629 BUG_ON(index
>= tun
->numqueues
);
631 rcu_assign_pointer(tun
->tfiles
[index
],
632 tun
->tfiles
[tun
->numqueues
- 1]);
633 ntfile
= rtnl_dereference(tun
->tfiles
[index
]);
634 ntfile
->queue_index
= index
;
638 RCU_INIT_POINTER(tfile
->tun
, NULL
);
639 sock_put(&tfile
->sk
);
641 tun_disable_queue(tun
, tfile
);
644 tun_flow_delete_by_queue(tun
, tun
->numqueues
+ 1);
645 /* Drop read queue */
646 tun_queue_purge(tfile
);
647 tun_set_real_num_queues(tun
);
648 } else if (tfile
->detached
&& clean
) {
649 tun
= tun_enable_queue(tfile
);
650 sock_put(&tfile
->sk
);
654 if (tun
&& tun
->numqueues
== 0 && tun
->numdisabled
== 0) {
655 netif_carrier_off(tun
->dev
);
657 if (!(tun
->flags
& IFF_PERSIST
) &&
658 tun
->dev
->reg_state
== NETREG_REGISTERED
)
659 unregister_netdevice(tun
->dev
);
661 skb_array_cleanup(&tfile
->tx_array
);
662 sock_put(&tfile
->sk
);
666 static void tun_detach(struct tun_file
*tfile
, bool clean
)
669 __tun_detach(tfile
, clean
);
673 static void tun_detach_all(struct net_device
*dev
)
675 struct tun_struct
*tun
= netdev_priv(dev
);
676 struct bpf_prog
*xdp_prog
= rtnl_dereference(tun
->xdp_prog
);
677 struct tun_file
*tfile
, *tmp
;
678 int i
, n
= tun
->numqueues
;
680 for (i
= 0; i
< n
; i
++) {
681 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
683 tun_napi_disable(tfile
);
684 tfile
->socket
.sk
->sk_shutdown
= RCV_SHUTDOWN
;
685 tfile
->socket
.sk
->sk_data_ready(tfile
->socket
.sk
);
686 RCU_INIT_POINTER(tfile
->tun
, NULL
);
689 list_for_each_entry(tfile
, &tun
->disabled
, next
) {
690 tfile
->socket
.sk
->sk_shutdown
= RCV_SHUTDOWN
;
691 tfile
->socket
.sk
->sk_data_ready(tfile
->socket
.sk
);
692 RCU_INIT_POINTER(tfile
->tun
, NULL
);
694 BUG_ON(tun
->numqueues
!= 0);
697 for (i
= 0; i
< n
; i
++) {
698 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
700 /* Drop read queue */
701 tun_queue_purge(tfile
);
702 sock_put(&tfile
->sk
);
704 list_for_each_entry_safe(tfile
, tmp
, &tun
->disabled
, next
) {
705 tun_enable_queue(tfile
);
706 tun_queue_purge(tfile
);
707 sock_put(&tfile
->sk
);
709 BUG_ON(tun
->numdisabled
!= 0);
712 bpf_prog_put(xdp_prog
);
714 if (tun
->flags
& IFF_PERSIST
)
715 module_put(THIS_MODULE
);
718 static int tun_attach(struct tun_struct
*tun
, struct file
*file
,
719 bool skip_filter
, bool napi
, bool napi_frags
)
721 struct tun_file
*tfile
= file
->private_data
;
722 struct net_device
*dev
= tun
->dev
;
725 err
= security_tun_dev_attach(tfile
->socket
.sk
, tun
->security
);
730 if (rtnl_dereference(tfile
->tun
) && !tfile
->detached
)
734 if (!(tun
->flags
& IFF_MULTI_QUEUE
) && tun
->numqueues
== 1)
738 if (!tfile
->detached
&&
739 tun
->numqueues
+ tun
->numdisabled
== MAX_TAP_QUEUES
)
744 /* Re-attach the filter to persist device */
745 if (!skip_filter
&& (tun
->filter_attached
== true)) {
746 lock_sock(tfile
->socket
.sk
);
747 err
= sk_attach_filter(&tun
->fprog
, tfile
->socket
.sk
);
748 release_sock(tfile
->socket
.sk
);
753 if (!tfile
->detached
&&
754 skb_array_resize(&tfile
->tx_array
, dev
->tx_queue_len
, GFP_KERNEL
)) {
759 tfile
->queue_index
= tun
->numqueues
;
760 tfile
->socket
.sk
->sk_shutdown
&= ~RCV_SHUTDOWN
;
761 if (tfile
->detached
) {
762 tun_enable_queue(tfile
);
764 sock_hold(&tfile
->sk
);
765 tun_napi_init(tun
, tfile
, napi
, napi_frags
);
768 /* device is allowed to go away first, so no need to hold extra
772 /* Publish tfile->tun and tun->tfiles only after we've fully
773 * initialized tfile; otherwise we risk using half-initialized
776 rcu_assign_pointer(tfile
->tun
, tun
);
777 rcu_assign_pointer(tun
->tfiles
[tun
->numqueues
], tfile
);
779 tun_set_real_num_queues(tun
);
784 static struct tun_struct
*tun_get(struct tun_file
*tfile
)
786 struct tun_struct
*tun
;
789 tun
= rcu_dereference(tfile
->tun
);
797 static void tun_put(struct tun_struct
*tun
)
803 static void addr_hash_set(u32
*mask
, const u8
*addr
)
805 int n
= ether_crc(ETH_ALEN
, addr
) >> 26;
806 mask
[n
>> 5] |= (1 << (n
& 31));
809 static unsigned int addr_hash_test(const u32
*mask
, const u8
*addr
)
811 int n
= ether_crc(ETH_ALEN
, addr
) >> 26;
812 return mask
[n
>> 5] & (1 << (n
& 31));
815 static int update_filter(struct tap_filter
*filter
, void __user
*arg
)
817 struct { u8 u
[ETH_ALEN
]; } *addr
;
818 struct tun_filter uf
;
819 int err
, alen
, n
, nexact
;
821 if (copy_from_user(&uf
, arg
, sizeof(uf
)))
830 alen
= ETH_ALEN
* uf
.count
;
831 addr
= memdup_user(arg
+ sizeof(uf
), alen
);
833 return PTR_ERR(addr
);
835 /* The filter is updated without holding any locks. Which is
836 * perfectly safe. We disable it first and in the worst
837 * case we'll accept a few undesired packets. */
841 /* Use first set of addresses as an exact filter */
842 for (n
= 0; n
< uf
.count
&& n
< FLT_EXACT_COUNT
; n
++)
843 memcpy(filter
->addr
[n
], addr
[n
].u
, ETH_ALEN
);
847 /* Remaining multicast addresses are hashed,
848 * unicast will leave the filter disabled. */
849 memset(filter
->mask
, 0, sizeof(filter
->mask
));
850 for (; n
< uf
.count
; n
++) {
851 if (!is_multicast_ether_addr(addr
[n
].u
)) {
852 err
= 0; /* no filter */
855 addr_hash_set(filter
->mask
, addr
[n
].u
);
858 /* For ALLMULTI just set the mask to all ones.
859 * This overrides the mask populated above. */
860 if ((uf
.flags
& TUN_FLT_ALLMULTI
))
861 memset(filter
->mask
, ~0, sizeof(filter
->mask
));
863 /* Now enable the filter */
865 filter
->count
= nexact
;
867 /* Return the number of exact filters */
874 /* Returns: 0 - drop, !=0 - accept */
875 static int run_filter(struct tap_filter
*filter
, const struct sk_buff
*skb
)
877 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
879 struct ethhdr
*eh
= (struct ethhdr
*) skb
->data
;
883 for (i
= 0; i
< filter
->count
; i
++)
884 if (ether_addr_equal(eh
->h_dest
, filter
->addr
[i
]))
887 /* Inexact match (multicast only) */
888 if (is_multicast_ether_addr(eh
->h_dest
))
889 return addr_hash_test(filter
->mask
, eh
->h_dest
);
895 * Checks whether the packet is accepted or not.
896 * Returns: 0 - drop, !=0 - accept
898 static int check_filter(struct tap_filter
*filter
, const struct sk_buff
*skb
)
903 return run_filter(filter
, skb
);
906 /* Network device part of the driver */
908 static const struct ethtool_ops tun_ethtool_ops
;
910 /* Net device detach from fd. */
911 static void tun_net_uninit(struct net_device
*dev
)
916 /* Net device open. */
917 static int tun_net_open(struct net_device
*dev
)
919 netif_tx_start_all_queues(dev
);
924 /* Net device close. */
925 static int tun_net_close(struct net_device
*dev
)
927 netif_tx_stop_all_queues(dev
);
931 /* Net device start xmit */
932 static netdev_tx_t
tun_net_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
934 struct tun_struct
*tun
= netdev_priv(dev
);
935 int txq
= skb
->queue_mapping
;
936 struct tun_file
*tfile
;
940 tfile
= rcu_dereference(tun
->tfiles
[txq
]);
941 numqueues
= READ_ONCE(tun
->numqueues
);
943 /* Drop packet if interface is not attached */
944 if (txq
>= numqueues
)
948 if (numqueues
== 1 && static_key_false(&rps_needed
)) {
949 /* Select queue was not called for the skbuff, so we extract the
950 * RPS hash and save it into the flow_table here.
954 rxhash
= __skb_get_hash_symmetric(skb
);
956 struct tun_flow_entry
*e
;
957 e
= tun_flow_find(&tun
->flows
[tun_hashfn(rxhash
)],
960 tun_flow_save_rps_rxhash(e
, rxhash
);
965 tun_debug(KERN_INFO
, tun
, "tun_net_xmit %d\n", skb
->len
);
969 /* Drop if the filter does not like it.
970 * This is a noop if the filter is disabled.
971 * Filter can be enabled only for the TAP devices. */
972 if (!check_filter(&tun
->txflt
, skb
))
975 if (tfile
->socket
.sk
->sk_filter
&&
976 sk_filter(tfile
->socket
.sk
, skb
))
979 if (unlikely(skb_orphan_frags_rx(skb
, GFP_ATOMIC
)))
982 skb_tx_timestamp(skb
);
984 /* Orphan the skb - required as we might hang on to it
985 * for indefinite time.
991 if (skb_array_produce(&tfile
->tx_array
, skb
))
994 /* Notify and wake up reader process */
995 if (tfile
->flags
& TUN_FASYNC
)
996 kill_fasync(&tfile
->fasync
, SIGIO
, POLL_IN
);
997 tfile
->socket
.sk
->sk_data_ready(tfile
->socket
.sk
);
1000 return NETDEV_TX_OK
;
1003 this_cpu_inc(tun
->pcpu_stats
->tx_dropped
);
1007 return NET_XMIT_DROP
;
1010 static void tun_net_mclist(struct net_device
*dev
)
1013 * This callback is supposed to deal with mc filter in
1014 * _rx_ path and has nothing to do with the _tx_ path.
1015 * In rx path we always accept everything userspace gives us.
1019 static netdev_features_t
tun_net_fix_features(struct net_device
*dev
,
1020 netdev_features_t features
)
1022 struct tun_struct
*tun
= netdev_priv(dev
);
1024 return (features
& tun
->set_features
) | (features
& ~TUN_USER_FEATURES
);
1026 #ifdef CONFIG_NET_POLL_CONTROLLER
1027 static void tun_poll_controller(struct net_device
*dev
)
1030 * Tun only receives frames when:
1031 * 1) the char device endpoint gets data from user space
1032 * 2) the tun socket gets a sendmsg call from user space
1033 * If NAPI is not enabled, since both of those are synchronous
1034 * operations, we are guaranteed never to have pending data when we poll
1035 * for it so there is nothing to do here but return.
1036 * We need this though so netpoll recognizes us as an interface that
1037 * supports polling, which enables bridge devices in virt setups to
1038 * still use netconsole
1039 * If NAPI is enabled, however, we need to schedule polling for all
1040 * queues unless we are using napi_gro_frags(), which we call in
1041 * process context and not in NAPI context.
1043 struct tun_struct
*tun
= netdev_priv(dev
);
1045 if (tun
->flags
& IFF_NAPI
) {
1046 struct tun_file
*tfile
;
1050 for (i
= 0; i
< tun
->numqueues
; i
++) {
1051 tfile
= rcu_dereference(tun
->tfiles
[i
]);
1052 if (!tun_napi_frags_enabled(tfile
) &&
1053 tfile
->napi_enabled
)
1054 napi_schedule(&tfile
->napi
);
1062 static void tun_set_headroom(struct net_device
*dev
, int new_hr
)
1064 struct tun_struct
*tun
= netdev_priv(dev
);
1066 if (new_hr
< NET_SKB_PAD
)
1067 new_hr
= NET_SKB_PAD
;
1069 tun
->align
= new_hr
;
1073 tun_net_get_stats64(struct net_device
*dev
, struct rtnl_link_stats64
*stats
)
1075 u32 rx_dropped
= 0, tx_dropped
= 0, rx_frame_errors
= 0;
1076 struct tun_struct
*tun
= netdev_priv(dev
);
1077 struct tun_pcpu_stats
*p
;
1080 for_each_possible_cpu(i
) {
1081 u64 rxpackets
, rxbytes
, txpackets
, txbytes
;
1084 p
= per_cpu_ptr(tun
->pcpu_stats
, i
);
1086 start
= u64_stats_fetch_begin(&p
->syncp
);
1087 rxpackets
= p
->rx_packets
;
1088 rxbytes
= p
->rx_bytes
;
1089 txpackets
= p
->tx_packets
;
1090 txbytes
= p
->tx_bytes
;
1091 } while (u64_stats_fetch_retry(&p
->syncp
, start
));
1093 stats
->rx_packets
+= rxpackets
;
1094 stats
->rx_bytes
+= rxbytes
;
1095 stats
->tx_packets
+= txpackets
;
1096 stats
->tx_bytes
+= txbytes
;
1099 rx_dropped
+= p
->rx_dropped
;
1100 rx_frame_errors
+= p
->rx_frame_errors
;
1101 tx_dropped
+= p
->tx_dropped
;
1103 stats
->rx_dropped
= rx_dropped
;
1104 stats
->rx_frame_errors
= rx_frame_errors
;
1105 stats
->tx_dropped
= tx_dropped
;
1108 static int tun_xdp_set(struct net_device
*dev
, struct bpf_prog
*prog
,
1109 struct netlink_ext_ack
*extack
)
1111 struct tun_struct
*tun
= netdev_priv(dev
);
1112 struct bpf_prog
*old_prog
;
1114 old_prog
= rtnl_dereference(tun
->xdp_prog
);
1115 rcu_assign_pointer(tun
->xdp_prog
, prog
);
1117 bpf_prog_put(old_prog
);
1122 static u32
tun_xdp_query(struct net_device
*dev
)
1124 struct tun_struct
*tun
= netdev_priv(dev
);
1125 const struct bpf_prog
*xdp_prog
;
1127 xdp_prog
= rtnl_dereference(tun
->xdp_prog
);
1129 return xdp_prog
->aux
->id
;
1134 static int tun_xdp(struct net_device
*dev
, struct netdev_bpf
*xdp
)
1136 switch (xdp
->command
) {
1137 case XDP_SETUP_PROG
:
1138 return tun_xdp_set(dev
, xdp
->prog
, xdp
->extack
);
1139 case XDP_QUERY_PROG
:
1140 xdp
->prog_id
= tun_xdp_query(dev
);
1141 xdp
->prog_attached
= !!xdp
->prog_id
;
1148 static int tun_net_change_carrier(struct net_device
*dev
, bool new_carrier
)
1151 struct tun_struct
*tun
= netdev_priv(dev
);
1153 if (!tun
->numqueues
)
1156 netif_carrier_on(dev
);
1158 netif_carrier_off(dev
);
1163 static const struct net_device_ops tun_netdev_ops
= {
1164 .ndo_uninit
= tun_net_uninit
,
1165 .ndo_open
= tun_net_open
,
1166 .ndo_stop
= tun_net_close
,
1167 .ndo_start_xmit
= tun_net_xmit
,
1168 .ndo_fix_features
= tun_net_fix_features
,
1169 .ndo_select_queue
= tun_select_queue
,
1170 #ifdef CONFIG_NET_POLL_CONTROLLER
1171 .ndo_poll_controller
= tun_poll_controller
,
1173 .ndo_set_rx_headroom
= tun_set_headroom
,
1174 .ndo_get_stats64
= tun_net_get_stats64
,
1175 .ndo_change_carrier
= tun_net_change_carrier
,
1178 static const struct net_device_ops tap_netdev_ops
= {
1179 .ndo_uninit
= tun_net_uninit
,
1180 .ndo_open
= tun_net_open
,
1181 .ndo_stop
= tun_net_close
,
1182 .ndo_start_xmit
= tun_net_xmit
,
1183 .ndo_fix_features
= tun_net_fix_features
,
1184 .ndo_set_rx_mode
= tun_net_mclist
,
1185 .ndo_set_mac_address
= eth_mac_addr
,
1186 .ndo_validate_addr
= eth_validate_addr
,
1187 .ndo_select_queue
= tun_select_queue
,
1188 #ifdef CONFIG_NET_POLL_CONTROLLER
1189 .ndo_poll_controller
= tun_poll_controller
,
1191 .ndo_features_check
= passthru_features_check
,
1192 .ndo_set_rx_headroom
= tun_set_headroom
,
1193 .ndo_get_stats64
= tun_net_get_stats64
,
1195 .ndo_change_carrier
= tun_net_change_carrier
,
1198 static void tun_flow_init(struct tun_struct
*tun
)
1202 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++)
1203 INIT_HLIST_HEAD(&tun
->flows
[i
]);
1205 tun
->ageing_time
= TUN_FLOW_EXPIRE
;
1206 timer_setup(&tun
->flow_gc_timer
, tun_flow_cleanup
, 0);
1207 mod_timer(&tun
->flow_gc_timer
,
1208 round_jiffies_up(jiffies
+ tun
->ageing_time
));
1211 static void tun_flow_uninit(struct tun_struct
*tun
)
1213 del_timer_sync(&tun
->flow_gc_timer
);
1214 tun_flow_flush(tun
);
1218 #define MAX_MTU 65535
1220 /* Initialize net device. */
1221 static void tun_net_init(struct net_device
*dev
)
1223 struct tun_struct
*tun
= netdev_priv(dev
);
1225 switch (tun
->flags
& TUN_TYPE_MASK
) {
1227 dev
->netdev_ops
= &tun_netdev_ops
;
1229 /* Point-to-Point TUN Device */
1230 dev
->hard_header_len
= 0;
1234 /* Zero header length */
1235 dev
->type
= ARPHRD_NONE
;
1236 dev
->flags
= IFF_POINTOPOINT
| IFF_NOARP
| IFF_MULTICAST
;
1240 dev
->netdev_ops
= &tap_netdev_ops
;
1241 /* Ethernet TAP Device */
1243 dev
->priv_flags
&= ~IFF_TX_SKB_SHARING
;
1244 dev
->priv_flags
|= IFF_LIVE_ADDR_CHANGE
;
1246 eth_hw_addr_random(dev
);
1251 dev
->min_mtu
= MIN_MTU
;
1252 dev
->max_mtu
= MAX_MTU
- dev
->hard_header_len
;
1255 /* Character device part */
1258 static unsigned int tun_chr_poll(struct file
*file
, poll_table
*wait
)
1260 struct tun_file
*tfile
= file
->private_data
;
1261 struct tun_struct
*tun
= tun_get(tfile
);
1263 unsigned int mask
= 0;
1268 sk
= tfile
->socket
.sk
;
1270 tun_debug(KERN_INFO
, tun
, "tun_chr_poll\n");
1272 poll_wait(file
, sk_sleep(sk
), wait
);
1274 if (!skb_array_empty(&tfile
->tx_array
))
1275 mask
|= POLLIN
| POLLRDNORM
;
1277 if (tun
->dev
->flags
& IFF_UP
&&
1278 (sock_writeable(sk
) ||
1279 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
) &&
1280 sock_writeable(sk
))))
1281 mask
|= POLLOUT
| POLLWRNORM
;
1283 if (tun
->dev
->reg_state
!= NETREG_REGISTERED
)
1290 static struct sk_buff
*tun_napi_alloc_frags(struct tun_file
*tfile
,
1292 const struct iov_iter
*it
)
1294 struct sk_buff
*skb
;
1299 if (it
->nr_segs
> MAX_SKB_FRAGS
+ 1)
1300 return ERR_PTR(-ENOMEM
);
1303 skb
= napi_get_frags(&tfile
->napi
);
1306 return ERR_PTR(-ENOMEM
);
1308 linear
= iov_iter_single_seg_count(it
);
1309 err
= __skb_grow(skb
, linear
);
1314 skb
->data_len
= len
- linear
;
1315 skb
->truesize
+= skb
->data_len
;
1317 for (i
= 1; i
< it
->nr_segs
; i
++) {
1318 struct page_frag
*pfrag
= ¤t
->task_frag
;
1319 size_t fragsz
= it
->iov
[i
].iov_len
;
1321 if (fragsz
== 0 || fragsz
> PAGE_SIZE
) {
1326 if (!skb_page_frag_refill(fragsz
, pfrag
, GFP_KERNEL
)) {
1331 skb_fill_page_desc(skb
, i
- 1, pfrag
->page
,
1332 pfrag
->offset
, fragsz
);
1333 page_ref_inc(pfrag
->page
);
1334 pfrag
->offset
+= fragsz
;
1339 /* frees skb and all frags allocated with napi_alloc_frag() */
1340 napi_free_frags(&tfile
->napi
);
1341 return ERR_PTR(err
);
1344 /* prepad is the amount to reserve at front. len is length after that.
1345 * linear is a hint as to how much to copy (usually headers). */
1346 static struct sk_buff
*tun_alloc_skb(struct tun_file
*tfile
,
1347 size_t prepad
, size_t len
,
1348 size_t linear
, int noblock
)
1350 struct sock
*sk
= tfile
->socket
.sk
;
1351 struct sk_buff
*skb
;
1354 /* Under a page? Don't bother with paged skb. */
1355 if (prepad
+ len
< PAGE_SIZE
|| !linear
)
1358 skb
= sock_alloc_send_pskb(sk
, prepad
+ linear
, len
- linear
, noblock
,
1361 return ERR_PTR(err
);
1363 skb_reserve(skb
, prepad
);
1364 skb_put(skb
, linear
);
1365 skb
->data_len
= len
- linear
;
1366 skb
->len
+= len
- linear
;
1371 static void tun_rx_batched(struct tun_struct
*tun
, struct tun_file
*tfile
,
1372 struct sk_buff
*skb
, int more
)
1374 struct sk_buff_head
*queue
= &tfile
->sk
.sk_write_queue
;
1375 struct sk_buff_head process_queue
;
1376 u32 rx_batched
= tun
->rx_batched
;
1379 if (!rx_batched
|| (!more
&& skb_queue_empty(queue
))) {
1381 skb_record_rx_queue(skb
, tfile
->queue_index
);
1382 netif_receive_skb(skb
);
1387 spin_lock(&queue
->lock
);
1388 if (!more
|| skb_queue_len(queue
) == rx_batched
) {
1389 __skb_queue_head_init(&process_queue
);
1390 skb_queue_splice_tail_init(queue
, &process_queue
);
1393 __skb_queue_tail(queue
, skb
);
1395 spin_unlock(&queue
->lock
);
1398 struct sk_buff
*nskb
;
1401 while ((nskb
= __skb_dequeue(&process_queue
))) {
1402 skb_record_rx_queue(nskb
, tfile
->queue_index
);
1403 netif_receive_skb(nskb
);
1405 skb_record_rx_queue(skb
, tfile
->queue_index
);
1406 netif_receive_skb(skb
);
1411 static bool tun_can_build_skb(struct tun_struct
*tun
, struct tun_file
*tfile
,
1412 int len
, int noblock
, bool zerocopy
)
1414 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
1417 if (tfile
->socket
.sk
->sk_sndbuf
!= INT_MAX
)
1426 if (SKB_DATA_ALIGN(len
+ TUN_RX_PAD
) +
1427 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
)) > PAGE_SIZE
)
1433 static struct sk_buff
*tun_build_skb(struct tun_struct
*tun
,
1434 struct tun_file
*tfile
,
1435 struct iov_iter
*from
,
1436 struct virtio_net_hdr
*hdr
,
1437 int len
, int *skb_xdp
)
1439 struct page_frag
*alloc_frag
= ¤t
->task_frag
;
1440 struct sk_buff
*skb
;
1441 struct bpf_prog
*xdp_prog
;
1442 int buflen
= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
1443 unsigned int delta
= 0;
1446 bool xdp_xmit
= false;
1447 int err
, pad
= TUN_RX_PAD
;
1450 xdp_prog
= rcu_dereference(tun
->xdp_prog
);
1452 pad
+= TUN_HEADROOM
;
1453 buflen
+= SKB_DATA_ALIGN(len
+ pad
);
1456 alloc_frag
->offset
= ALIGN((u64
)alloc_frag
->offset
, SMP_CACHE_BYTES
);
1457 if (unlikely(!skb_page_frag_refill(buflen
, alloc_frag
, GFP_KERNEL
)))
1458 return ERR_PTR(-ENOMEM
);
1460 buf
= (char *)page_address(alloc_frag
->page
) + alloc_frag
->offset
;
1461 copied
= copy_page_from_iter(alloc_frag
->page
,
1462 alloc_frag
->offset
+ pad
,
1465 return ERR_PTR(-EFAULT
);
1467 /* There's a small window that XDP may be set after the check
1468 * of xdp_prog above, this should be rare and for simplicity
1469 * we do XDP on skb in case the headroom is not enough.
1471 if (hdr
->gso_type
|| !xdp_prog
)
1478 xdp_prog
= rcu_dereference(tun
->xdp_prog
);
1479 if (xdp_prog
&& !*skb_xdp
) {
1480 struct xdp_buff xdp
;
1484 xdp
.data_hard_start
= buf
;
1485 xdp
.data
= buf
+ pad
;
1486 xdp_set_data_meta_invalid(&xdp
);
1487 xdp
.data_end
= xdp
.data
+ len
;
1488 orig_data
= xdp
.data
;
1489 act
= bpf_prog_run_xdp(xdp_prog
, &xdp
);
1493 get_page(alloc_frag
->page
);
1494 alloc_frag
->offset
+= buflen
;
1495 err
= xdp_do_redirect(tun
->dev
, &xdp
, xdp_prog
);
1506 delta
= orig_data
- xdp
.data
;
1509 bpf_warn_invalid_xdp_action(act
);
1512 trace_xdp_exception(tun
->dev
, xdp_prog
, act
);
1519 skb
= build_skb(buf
, buflen
);
1523 return ERR_PTR(-ENOMEM
);
1526 skb_reserve(skb
, pad
- delta
);
1527 skb_put(skb
, len
+ delta
);
1528 skb_set_owner_w(skb
, tfile
->socket
.sk
);
1529 get_page(alloc_frag
->page
);
1530 alloc_frag
->offset
+= buflen
;
1533 skb
->dev
= tun
->dev
;
1534 generic_xdp_tx(skb
, xdp_prog
);
1546 put_page(alloc_frag
->page
);
1550 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1554 /* Get packet from user space buffer */
1555 static ssize_t
tun_get_user(struct tun_struct
*tun
, struct tun_file
*tfile
,
1556 void *msg_control
, struct iov_iter
*from
,
1557 int noblock
, bool more
)
1559 struct tun_pi pi
= { 0, cpu_to_be16(ETH_P_IP
) };
1560 struct sk_buff
*skb
;
1561 size_t total_len
= iov_iter_count(from
);
1562 size_t len
= total_len
, align
= tun
->align
, linear
;
1563 struct virtio_net_hdr gso
= { 0 };
1564 struct tun_pcpu_stats
*stats
;
1567 bool zerocopy
= false;
1571 bool frags
= tun_napi_frags_enabled(tfile
);
1573 if (!(tun
->flags
& IFF_NO_PI
)) {
1574 if (len
< sizeof(pi
))
1578 if (!copy_from_iter_full(&pi
, sizeof(pi
), from
))
1582 if (tun
->flags
& IFF_VNET_HDR
) {
1583 int vnet_hdr_sz
= READ_ONCE(tun
->vnet_hdr_sz
);
1585 if (len
< vnet_hdr_sz
)
1589 if (!copy_from_iter_full(&gso
, sizeof(gso
), from
))
1592 if ((gso
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) &&
1593 tun16_to_cpu(tun
, gso
.csum_start
) + tun16_to_cpu(tun
, gso
.csum_offset
) + 2 > tun16_to_cpu(tun
, gso
.hdr_len
))
1594 gso
.hdr_len
= cpu_to_tun16(tun
, tun16_to_cpu(tun
, gso
.csum_start
) + tun16_to_cpu(tun
, gso
.csum_offset
) + 2);
1596 if (tun16_to_cpu(tun
, gso
.hdr_len
) > len
)
1598 iov_iter_advance(from
, vnet_hdr_sz
- sizeof(gso
));
1601 if ((tun
->flags
& TUN_TYPE_MASK
) == IFF_TAP
) {
1602 align
+= NET_IP_ALIGN
;
1603 if (unlikely(len
< ETH_HLEN
||
1604 (gso
.hdr_len
&& tun16_to_cpu(tun
, gso
.hdr_len
) < ETH_HLEN
)))
1608 good_linear
= SKB_MAX_HEAD(align
);
1611 struct iov_iter i
= *from
;
1613 /* There are 256 bytes to be copied in skb, so there is
1614 * enough room for skb expand head in case it is used.
1615 * The rest of the buffer is mapped from userspace.
1617 copylen
= gso
.hdr_len
? tun16_to_cpu(tun
, gso
.hdr_len
) : GOODCOPY_LEN
;
1618 if (copylen
> good_linear
)
1619 copylen
= good_linear
;
1621 iov_iter_advance(&i
, copylen
);
1622 if (iov_iter_npages(&i
, INT_MAX
) <= MAX_SKB_FRAGS
)
1626 if (!frags
&& tun_can_build_skb(tun
, tfile
, len
, noblock
, zerocopy
)) {
1627 /* For the packet that is not easy to be processed
1628 * (e.g gso or jumbo packet), we will do it at after
1629 * skb was created with generic XDP routine.
1631 skb
= tun_build_skb(tun
, tfile
, from
, &gso
, len
, &skb_xdp
);
1633 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1634 return PTR_ERR(skb
);
1641 if (tun16_to_cpu(tun
, gso
.hdr_len
) > good_linear
)
1642 linear
= good_linear
;
1644 linear
= tun16_to_cpu(tun
, gso
.hdr_len
);
1648 mutex_lock(&tfile
->napi_mutex
);
1649 skb
= tun_napi_alloc_frags(tfile
, copylen
, from
);
1650 /* tun_napi_alloc_frags() enforces a layout for the skb.
1651 * If zerocopy is enabled, then this layout will be
1652 * overwritten by zerocopy_sg_from_iter().
1656 skb
= tun_alloc_skb(tfile
, align
, copylen
, linear
,
1661 if (PTR_ERR(skb
) != -EAGAIN
)
1662 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1664 mutex_unlock(&tfile
->napi_mutex
);
1665 return PTR_ERR(skb
);
1669 err
= zerocopy_sg_from_iter(skb
, from
);
1671 err
= skb_copy_datagram_from_iter(skb
, 0, from
, len
);
1676 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1679 tfile
->napi
.skb
= NULL
;
1680 mutex_unlock(&tfile
->napi_mutex
);
1687 if (virtio_net_hdr_to_skb(skb
, &gso
, tun_is_little_endian(tun
))) {
1688 this_cpu_inc(tun
->pcpu_stats
->rx_frame_errors
);
1691 tfile
->napi
.skb
= NULL
;
1692 mutex_unlock(&tfile
->napi_mutex
);
1698 switch (tun
->flags
& TUN_TYPE_MASK
) {
1700 if (tun
->flags
& IFF_NO_PI
) {
1701 u8 ip_version
= skb
->len
? (skb
->data
[0] >> 4) : 0;
1703 switch (ip_version
) {
1705 pi
.proto
= htons(ETH_P_IP
);
1708 pi
.proto
= htons(ETH_P_IPV6
);
1711 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1717 skb_reset_mac_header(skb
);
1718 skb
->protocol
= pi
.proto
;
1719 skb
->dev
= tun
->dev
;
1723 skb
->protocol
= eth_type_trans(skb
, tun
->dev
);
1727 /* copy skb_ubuf_info for callback when skb has no error */
1729 skb_shinfo(skb
)->destructor_arg
= msg_control
;
1730 skb_shinfo(skb
)->tx_flags
|= SKBTX_DEV_ZEROCOPY
;
1731 skb_shinfo(skb
)->tx_flags
|= SKBTX_SHARED_FRAG
;
1732 } else if (msg_control
) {
1733 struct ubuf_info
*uarg
= msg_control
;
1734 uarg
->callback(uarg
, false);
1737 skb_reset_network_header(skb
);
1738 skb_probe_transport_header(skb
, 0);
1741 struct bpf_prog
*xdp_prog
;
1746 xdp_prog
= rcu_dereference(tun
->xdp_prog
);
1748 ret
= do_xdp_generic(xdp_prog
, skb
);
1749 if (ret
!= XDP_PASS
) {
1759 rxhash
= __skb_get_hash_symmetric(skb
);
1762 if (unlikely(!(tun
->dev
->flags
& IFF_UP
))) {
1769 /* Exercise flow dissector code path. */
1770 u32 headlen
= eth_get_headlen(skb
->data
, skb_headlen(skb
));
1772 if (unlikely(headlen
> skb_headlen(skb
))) {
1773 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1774 napi_free_frags(&tfile
->napi
);
1776 mutex_unlock(&tfile
->napi_mutex
);
1782 napi_gro_frags(&tfile
->napi
);
1784 mutex_unlock(&tfile
->napi_mutex
);
1785 } else if (tfile
->napi_enabled
) {
1786 struct sk_buff_head
*queue
= &tfile
->sk
.sk_write_queue
;
1789 spin_lock_bh(&queue
->lock
);
1790 __skb_queue_tail(queue
, skb
);
1791 queue_len
= skb_queue_len(queue
);
1792 spin_unlock(&queue
->lock
);
1794 if (!more
|| queue_len
> NAPI_POLL_WEIGHT
)
1795 napi_schedule(&tfile
->napi
);
1798 } else if (!IS_ENABLED(CONFIG_4KSTACKS
)) {
1799 tun_rx_batched(tun
, tfile
, skb
, more
);
1805 stats
= get_cpu_ptr(tun
->pcpu_stats
);
1806 u64_stats_update_begin(&stats
->syncp
);
1807 stats
->rx_packets
++;
1808 stats
->rx_bytes
+= len
;
1809 u64_stats_update_end(&stats
->syncp
);
1812 tun_flow_update(tun
, rxhash
, tfile
);
1816 static ssize_t
tun_chr_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
1818 struct file
*file
= iocb
->ki_filp
;
1819 struct tun_file
*tfile
= file
->private_data
;
1820 struct tun_struct
*tun
= tun_get(tfile
);
1826 result
= tun_get_user(tun
, tfile
, NULL
, from
,
1827 file
->f_flags
& O_NONBLOCK
, false);
1833 /* Put packet to the user space buffer */
1834 static ssize_t
tun_put_user(struct tun_struct
*tun
,
1835 struct tun_file
*tfile
,
1836 struct sk_buff
*skb
,
1837 struct iov_iter
*iter
)
1839 struct tun_pi pi
= { 0, skb
->protocol
};
1840 struct tun_pcpu_stats
*stats
;
1842 int vlan_offset
= 0;
1844 int vnet_hdr_sz
= 0;
1846 if (skb_vlan_tag_present(skb
))
1847 vlan_hlen
= VLAN_HLEN
;
1849 if (tun
->flags
& IFF_VNET_HDR
)
1850 vnet_hdr_sz
= READ_ONCE(tun
->vnet_hdr_sz
);
1852 total
= skb
->len
+ vlan_hlen
+ vnet_hdr_sz
;
1854 if (!(tun
->flags
& IFF_NO_PI
)) {
1855 if (iov_iter_count(iter
) < sizeof(pi
))
1858 total
+= sizeof(pi
);
1859 if (iov_iter_count(iter
) < total
) {
1860 /* Packet will be striped */
1861 pi
.flags
|= TUN_PKT_STRIP
;
1864 if (copy_to_iter(&pi
, sizeof(pi
), iter
) != sizeof(pi
))
1869 struct virtio_net_hdr gso
;
1871 if (iov_iter_count(iter
) < vnet_hdr_sz
)
1874 if (virtio_net_hdr_from_skb(skb
, &gso
,
1875 tun_is_little_endian(tun
), true,
1877 struct skb_shared_info
*sinfo
= skb_shinfo(skb
);
1878 pr_err("unexpected GSO type: "
1879 "0x%x, gso_size %d, hdr_len %d\n",
1880 sinfo
->gso_type
, tun16_to_cpu(tun
, gso
.gso_size
),
1881 tun16_to_cpu(tun
, gso
.hdr_len
));
1882 print_hex_dump(KERN_ERR
, "tun: ",
1885 min((int)tun16_to_cpu(tun
, gso
.hdr_len
), 64), true);
1890 if (copy_to_iter(&gso
, sizeof(gso
), iter
) != sizeof(gso
))
1893 iov_iter_advance(iter
, vnet_hdr_sz
- sizeof(gso
));
1899 __be16 h_vlan_proto
;
1903 veth
.h_vlan_proto
= skb
->vlan_proto
;
1904 veth
.h_vlan_TCI
= htons(skb_vlan_tag_get(skb
));
1906 vlan_offset
= offsetof(struct vlan_ethhdr
, h_vlan_proto
);
1908 ret
= skb_copy_datagram_iter(skb
, 0, iter
, vlan_offset
);
1909 if (ret
|| !iov_iter_count(iter
))
1912 ret
= copy_to_iter(&veth
, sizeof(veth
), iter
);
1913 if (ret
!= sizeof(veth
) || !iov_iter_count(iter
))
1917 skb_copy_datagram_iter(skb
, vlan_offset
, iter
, skb
->len
- vlan_offset
);
1920 /* caller is in process context, */
1921 stats
= get_cpu_ptr(tun
->pcpu_stats
);
1922 u64_stats_update_begin(&stats
->syncp
);
1923 stats
->tx_packets
++;
1924 stats
->tx_bytes
+= skb
->len
+ vlan_hlen
;
1925 u64_stats_update_end(&stats
->syncp
);
1926 put_cpu_ptr(tun
->pcpu_stats
);
1931 static struct sk_buff
*tun_ring_recv(struct tun_file
*tfile
, int noblock
,
1934 DECLARE_WAITQUEUE(wait
, current
);
1935 struct sk_buff
*skb
= NULL
;
1938 skb
= skb_array_consume(&tfile
->tx_array
);
1946 add_wait_queue(&tfile
->wq
.wait
, &wait
);
1949 set_current_state(TASK_INTERRUPTIBLE
);
1950 skb
= skb_array_consume(&tfile
->tx_array
);
1953 if (signal_pending(current
)) {
1954 error
= -ERESTARTSYS
;
1957 if (tfile
->socket
.sk
->sk_shutdown
& RCV_SHUTDOWN
) {
1965 __set_current_state(TASK_RUNNING
);
1966 remove_wait_queue(&tfile
->wq
.wait
, &wait
);
1973 static ssize_t
tun_do_read(struct tun_struct
*tun
, struct tun_file
*tfile
,
1974 struct iov_iter
*to
,
1975 int noblock
, struct sk_buff
*skb
)
1980 tun_debug(KERN_INFO
, tun
, "tun_do_read\n");
1982 if (!iov_iter_count(to
)) {
1989 /* Read frames from ring */
1990 skb
= tun_ring_recv(tfile
, noblock
, &err
);
1995 ret
= tun_put_user(tun
, tfile
, skb
, to
);
1996 if (unlikely(ret
< 0))
2004 static ssize_t
tun_chr_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
2006 struct file
*file
= iocb
->ki_filp
;
2007 struct tun_file
*tfile
= file
->private_data
;
2008 struct tun_struct
*tun
= tun_get(tfile
);
2009 ssize_t len
= iov_iter_count(to
), ret
;
2013 ret
= tun_do_read(tun
, tfile
, to
, file
->f_flags
& O_NONBLOCK
, NULL
);
2014 ret
= min_t(ssize_t
, ret
, len
);
2021 static void tun_free_netdev(struct net_device
*dev
)
2023 struct tun_struct
*tun
= netdev_priv(dev
);
2025 BUG_ON(!(list_empty(&tun
->disabled
)));
2026 free_percpu(tun
->pcpu_stats
);
2027 tun_flow_uninit(tun
);
2028 security_tun_dev_free_security(tun
->security
);
2031 static void tun_setup(struct net_device
*dev
)
2033 struct tun_struct
*tun
= netdev_priv(dev
);
2035 tun
->owner
= INVALID_UID
;
2036 tun
->group
= INVALID_GID
;
2038 dev
->ethtool_ops
= &tun_ethtool_ops
;
2039 dev
->needs_free_netdev
= true;
2040 dev
->priv_destructor
= tun_free_netdev
;
2041 /* We prefer our own queue length */
2042 dev
->tx_queue_len
= TUN_READQ_SIZE
;
2045 /* Trivial set of netlink ops to allow deleting tun or tap
2046 * device with netlink.
2048 static int tun_validate(struct nlattr
*tb
[], struct nlattr
*data
[],
2049 struct netlink_ext_ack
*extack
)
2051 NL_SET_ERR_MSG(extack
,
2052 "tun/tap creation via rtnetlink is not supported.");
2056 static struct rtnl_link_ops tun_link_ops __read_mostly
= {
2058 .priv_size
= sizeof(struct tun_struct
),
2060 .validate
= tun_validate
,
2063 static void tun_sock_write_space(struct sock
*sk
)
2065 struct tun_file
*tfile
;
2066 wait_queue_head_t
*wqueue
;
2068 if (!sock_writeable(sk
))
2071 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
))
2074 wqueue
= sk_sleep(sk
);
2075 if (wqueue
&& waitqueue_active(wqueue
))
2076 wake_up_interruptible_sync_poll(wqueue
, POLLOUT
|
2077 POLLWRNORM
| POLLWRBAND
);
2079 tfile
= container_of(sk
, struct tun_file
, sk
);
2080 kill_fasync(&tfile
->fasync
, SIGIO
, POLL_OUT
);
2083 static int tun_sendmsg(struct socket
*sock
, struct msghdr
*m
, size_t total_len
)
2086 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
2087 struct tun_struct
*tun
= tun_get(tfile
);
2092 ret
= tun_get_user(tun
, tfile
, m
->msg_control
, &m
->msg_iter
,
2093 m
->msg_flags
& MSG_DONTWAIT
,
2094 m
->msg_flags
& MSG_MORE
);
2099 static int tun_recvmsg(struct socket
*sock
, struct msghdr
*m
, size_t total_len
,
2102 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
2103 struct tun_struct
*tun
= tun_get(tfile
);
2104 struct sk_buff
*skb
= m
->msg_control
;
2112 if (flags
& ~(MSG_DONTWAIT
|MSG_TRUNC
|MSG_ERRQUEUE
)) {
2116 if (flags
& MSG_ERRQUEUE
) {
2117 ret
= sock_recv_errqueue(sock
->sk
, m
, total_len
,
2118 SOL_PACKET
, TUN_TX_TIMESTAMP
);
2121 ret
= tun_do_read(tun
, tfile
, &m
->msg_iter
, flags
& MSG_DONTWAIT
, skb
);
2122 if (ret
> (ssize_t
)total_len
) {
2123 m
->msg_flags
|= MSG_TRUNC
;
2124 ret
= flags
& MSG_TRUNC
? ret
: total_len
;
2138 static int tun_peek_len(struct socket
*sock
)
2140 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
2141 struct tun_struct
*tun
;
2144 tun
= tun_get(tfile
);
2148 ret
= skb_array_peek_len(&tfile
->tx_array
);
2154 /* Ops structure to mimic raw sockets with tun */
2155 static const struct proto_ops tun_socket_ops
= {
2156 .peek_len
= tun_peek_len
,
2157 .sendmsg
= tun_sendmsg
,
2158 .recvmsg
= tun_recvmsg
,
2161 static struct proto tun_proto
= {
2163 .owner
= THIS_MODULE
,
2164 .obj_size
= sizeof(struct tun_file
),
2167 static int tun_flags(struct tun_struct
*tun
)
2169 return tun
->flags
& (TUN_FEATURES
| IFF_PERSIST
| IFF_TUN
| IFF_TAP
);
2172 static ssize_t
tun_show_flags(struct device
*dev
, struct device_attribute
*attr
,
2175 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
2176 return sprintf(buf
, "0x%x\n", tun_flags(tun
));
2179 static ssize_t
tun_show_owner(struct device
*dev
, struct device_attribute
*attr
,
2182 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
2183 return uid_valid(tun
->owner
)?
2184 sprintf(buf
, "%u\n",
2185 from_kuid_munged(current_user_ns(), tun
->owner
)):
2186 sprintf(buf
, "-1\n");
2189 static ssize_t
tun_show_group(struct device
*dev
, struct device_attribute
*attr
,
2192 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
2193 return gid_valid(tun
->group
) ?
2194 sprintf(buf
, "%u\n",
2195 from_kgid_munged(current_user_ns(), tun
->group
)):
2196 sprintf(buf
, "-1\n");
2199 static DEVICE_ATTR(tun_flags
, 0444, tun_show_flags
, NULL
);
2200 static DEVICE_ATTR(owner
, 0444, tun_show_owner
, NULL
);
2201 static DEVICE_ATTR(group
, 0444, tun_show_group
, NULL
);
2203 static struct attribute
*tun_dev_attrs
[] = {
2204 &dev_attr_tun_flags
.attr
,
2205 &dev_attr_owner
.attr
,
2206 &dev_attr_group
.attr
,
2210 static const struct attribute_group tun_attr_group
= {
2211 .attrs
= tun_dev_attrs
2214 static int tun_set_iff(struct net
*net
, struct file
*file
, struct ifreq
*ifr
)
2216 struct tun_struct
*tun
;
2217 struct tun_file
*tfile
= file
->private_data
;
2218 struct net_device
*dev
;
2221 if (tfile
->detached
)
2224 if ((ifr
->ifr_flags
& IFF_NAPI_FRAGS
)) {
2225 if (!capable(CAP_NET_ADMIN
))
2228 if (!(ifr
->ifr_flags
& IFF_NAPI
) ||
2229 (ifr
->ifr_flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2233 dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
2235 if (ifr
->ifr_flags
& IFF_TUN_EXCL
)
2237 if ((ifr
->ifr_flags
& IFF_TUN
) && dev
->netdev_ops
== &tun_netdev_ops
)
2238 tun
= netdev_priv(dev
);
2239 else if ((ifr
->ifr_flags
& IFF_TAP
) && dev
->netdev_ops
== &tap_netdev_ops
)
2240 tun
= netdev_priv(dev
);
2244 if (!!(ifr
->ifr_flags
& IFF_MULTI_QUEUE
) !=
2245 !!(tun
->flags
& IFF_MULTI_QUEUE
))
2248 if (tun_not_capable(tun
))
2250 err
= security_tun_dev_open(tun
->security
);
2254 err
= tun_attach(tun
, file
, ifr
->ifr_flags
& IFF_NOFILTER
,
2255 ifr
->ifr_flags
& IFF_NAPI
,
2256 ifr
->ifr_flags
& IFF_NAPI_FRAGS
);
2260 if (tun
->flags
& IFF_MULTI_QUEUE
&&
2261 (tun
->numqueues
+ tun
->numdisabled
> 1)) {
2262 /* One or more queue has already been attached, no need
2263 * to initialize the device again.
2270 unsigned long flags
= 0;
2271 int queues
= ifr
->ifr_flags
& IFF_MULTI_QUEUE
?
2274 if (!ns_capable(net
->user_ns
, CAP_NET_ADMIN
))
2276 err
= security_tun_dev_create();
2281 if (ifr
->ifr_flags
& IFF_TUN
) {
2285 } else if (ifr
->ifr_flags
& IFF_TAP
) {
2293 name
= ifr
->ifr_name
;
2295 dev
= alloc_netdev_mqs(sizeof(struct tun_struct
), name
,
2296 NET_NAME_UNKNOWN
, tun_setup
, queues
,
2301 err
= dev_get_valid_name(net
, dev
, name
);
2305 dev_net_set(dev
, net
);
2306 dev
->rtnl_link_ops
= &tun_link_ops
;
2307 dev
->ifindex
= tfile
->ifindex
;
2308 dev
->sysfs_groups
[0] = &tun_attr_group
;
2310 tun
= netdev_priv(dev
);
2313 tun
->txflt
.count
= 0;
2314 tun
->vnet_hdr_sz
= sizeof(struct virtio_net_hdr
);
2316 tun
->align
= NET_SKB_PAD
;
2317 tun
->filter_attached
= false;
2318 tun
->sndbuf
= tfile
->socket
.sk
->sk_sndbuf
;
2319 tun
->rx_batched
= 0;
2321 tun
->pcpu_stats
= netdev_alloc_pcpu_stats(struct tun_pcpu_stats
);
2322 if (!tun
->pcpu_stats
) {
2327 spin_lock_init(&tun
->lock
);
2329 err
= security_tun_dev_alloc_security(&tun
->security
);
2336 dev
->hw_features
= NETIF_F_SG
| NETIF_F_FRAGLIST
|
2337 TUN_USER_FEATURES
| NETIF_F_HW_VLAN_CTAG_TX
|
2338 NETIF_F_HW_VLAN_STAG_TX
;
2339 dev
->features
= dev
->hw_features
| NETIF_F_LLTX
;
2340 dev
->vlan_features
= dev
->features
&
2341 ~(NETIF_F_HW_VLAN_CTAG_TX
|
2342 NETIF_F_HW_VLAN_STAG_TX
);
2344 INIT_LIST_HEAD(&tun
->disabled
);
2345 err
= tun_attach(tun
, file
, false, ifr
->ifr_flags
& IFF_NAPI
,
2346 ifr
->ifr_flags
& IFF_NAPI_FRAGS
);
2350 err
= register_netdevice(tun
->dev
);
2355 netif_carrier_on(tun
->dev
);
2357 tun_debug(KERN_INFO
, tun
, "tun_set_iff\n");
2359 tun
->flags
= (tun
->flags
& ~TUN_FEATURES
) |
2360 (ifr
->ifr_flags
& TUN_FEATURES
);
2362 /* Make sure persistent devices do not get stuck in
2365 if (netif_running(tun
->dev
))
2366 netif_tx_wake_all_queues(tun
->dev
);
2368 strcpy(ifr
->ifr_name
, tun
->dev
->name
);
2372 tun_detach_all(dev
);
2373 /* register_netdevice() already called tun_free_netdev() */
2377 tun_flow_uninit(tun
);
2378 security_tun_dev_free_security(tun
->security
);
2380 free_percpu(tun
->pcpu_stats
);
2386 static void tun_get_iff(struct net
*net
, struct tun_struct
*tun
,
2389 tun_debug(KERN_INFO
, tun
, "tun_get_iff\n");
2391 strcpy(ifr
->ifr_name
, tun
->dev
->name
);
2393 ifr
->ifr_flags
= tun_flags(tun
);
2397 /* This is like a cut-down ethtool ops, except done via tun fd so no
2398 * privs required. */
2399 static int set_offload(struct tun_struct
*tun
, unsigned long arg
)
2401 netdev_features_t features
= 0;
2403 if (arg
& TUN_F_CSUM
) {
2404 features
|= NETIF_F_HW_CSUM
;
2407 if (arg
& (TUN_F_TSO4
|TUN_F_TSO6
)) {
2408 if (arg
& TUN_F_TSO_ECN
) {
2409 features
|= NETIF_F_TSO_ECN
;
2410 arg
&= ~TUN_F_TSO_ECN
;
2412 if (arg
& TUN_F_TSO4
)
2413 features
|= NETIF_F_TSO
;
2414 if (arg
& TUN_F_TSO6
)
2415 features
|= NETIF_F_TSO6
;
2416 arg
&= ~(TUN_F_TSO4
|TUN_F_TSO6
);
2422 /* This gives the user a way to test for new features in future by
2423 * trying to set them. */
2427 tun
->set_features
= features
;
2428 tun
->dev
->wanted_features
&= ~TUN_USER_FEATURES
;
2429 tun
->dev
->wanted_features
|= features
;
2430 netdev_update_features(tun
->dev
);
2435 static void tun_detach_filter(struct tun_struct
*tun
, int n
)
2438 struct tun_file
*tfile
;
2440 for (i
= 0; i
< n
; i
++) {
2441 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
2442 lock_sock(tfile
->socket
.sk
);
2443 sk_detach_filter(tfile
->socket
.sk
);
2444 release_sock(tfile
->socket
.sk
);
2447 tun
->filter_attached
= false;
2450 static int tun_attach_filter(struct tun_struct
*tun
)
2453 struct tun_file
*tfile
;
2455 for (i
= 0; i
< tun
->numqueues
; i
++) {
2456 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
2457 lock_sock(tfile
->socket
.sk
);
2458 ret
= sk_attach_filter(&tun
->fprog
, tfile
->socket
.sk
);
2459 release_sock(tfile
->socket
.sk
);
2461 tun_detach_filter(tun
, i
);
2466 tun
->filter_attached
= true;
2470 static void tun_set_sndbuf(struct tun_struct
*tun
)
2472 struct tun_file
*tfile
;
2475 for (i
= 0; i
< tun
->numqueues
; i
++) {
2476 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
2477 tfile
->socket
.sk
->sk_sndbuf
= tun
->sndbuf
;
2481 static int tun_set_queue(struct file
*file
, struct ifreq
*ifr
)
2483 struct tun_file
*tfile
= file
->private_data
;
2484 struct tun_struct
*tun
;
2489 if (ifr
->ifr_flags
& IFF_ATTACH_QUEUE
) {
2490 tun
= tfile
->detached
;
2495 ret
= security_tun_dev_attach_queue(tun
->security
);
2498 ret
= tun_attach(tun
, file
, false, tun
->flags
& IFF_NAPI
,
2499 tun
->flags
& IFF_NAPI_FRAGS
);
2500 } else if (ifr
->ifr_flags
& IFF_DETACH_QUEUE
) {
2501 tun
= rtnl_dereference(tfile
->tun
);
2502 if (!tun
|| !(tun
->flags
& IFF_MULTI_QUEUE
) || tfile
->detached
)
2505 __tun_detach(tfile
, false);
2514 static long __tun_chr_ioctl(struct file
*file
, unsigned int cmd
,
2515 unsigned long arg
, int ifreq_len
)
2517 struct tun_file
*tfile
= file
->private_data
;
2518 struct tun_struct
*tun
;
2519 void __user
* argp
= (void __user
*)arg
;
2520 unsigned int ifindex
, carrier
;
2529 if (cmd
== TUNSETIFF
|| cmd
== TUNSETQUEUE
|| _IOC_TYPE(cmd
) == SOCK_IOC_TYPE
) {
2530 if (copy_from_user(&ifr
, argp
, ifreq_len
))
2533 memset(&ifr
, 0, sizeof(ifr
));
2535 if (cmd
== TUNGETFEATURES
) {
2536 /* Currently this just means: "what IFF flags are valid?".
2537 * This is needed because we never checked for invalid flags on
2540 return put_user(IFF_TUN
| IFF_TAP
| TUN_FEATURES
,
2541 (unsigned int __user
*)argp
);
2542 } else if (cmd
== TUNSETQUEUE
)
2543 return tun_set_queue(file
, &ifr
);
2548 tun
= tun_get(tfile
);
2549 if (cmd
== TUNSETIFF
) {
2554 ifr
.ifr_name
[IFNAMSIZ
-1] = '\0';
2556 ret
= tun_set_iff(sock_net(&tfile
->sk
), file
, &ifr
);
2561 if (copy_to_user(argp
, &ifr
, ifreq_len
))
2565 if (cmd
== TUNSETIFINDEX
) {
2571 if (copy_from_user(&ifindex
, argp
, sizeof(ifindex
)))
2575 tfile
->ifindex
= ifindex
;
2583 tun_debug(KERN_INFO
, tun
, "tun_chr_ioctl cmd %u\n", cmd
);
2588 tun_get_iff(current
->nsproxy
->net_ns
, tun
, &ifr
);
2590 if (tfile
->detached
)
2591 ifr
.ifr_flags
|= IFF_DETACH_QUEUE
;
2592 if (!tfile
->socket
.sk
->sk_filter
)
2593 ifr
.ifr_flags
|= IFF_NOFILTER
;
2595 if (copy_to_user(argp
, &ifr
, ifreq_len
))
2600 /* Disable/Enable checksum */
2602 /* [unimplemented] */
2603 tun_debug(KERN_INFO
, tun
, "ignored: set checksum %s\n",
2604 arg
? "disabled" : "enabled");
2608 /* Disable/Enable persist mode. Keep an extra reference to the
2609 * module to prevent the module being unprobed.
2611 if (arg
&& !(tun
->flags
& IFF_PERSIST
)) {
2612 tun
->flags
|= IFF_PERSIST
;
2613 __module_get(THIS_MODULE
);
2615 if (!arg
&& (tun
->flags
& IFF_PERSIST
)) {
2616 tun
->flags
&= ~IFF_PERSIST
;
2617 module_put(THIS_MODULE
);
2620 tun_debug(KERN_INFO
, tun
, "persist %s\n",
2621 arg
? "enabled" : "disabled");
2625 /* Set owner of the device */
2626 owner
= make_kuid(current_user_ns(), arg
);
2627 if (!uid_valid(owner
)) {
2632 tun_debug(KERN_INFO
, tun
, "owner set to %u\n",
2633 from_kuid(&init_user_ns
, tun
->owner
));
2637 /* Set group of the device */
2638 group
= make_kgid(current_user_ns(), arg
);
2639 if (!gid_valid(group
)) {
2644 tun_debug(KERN_INFO
, tun
, "group set to %u\n",
2645 from_kgid(&init_user_ns
, tun
->group
));
2649 /* Only allow setting the type when the interface is down */
2650 if (tun
->dev
->flags
& IFF_UP
) {
2651 tun_debug(KERN_INFO
, tun
,
2652 "Linktype set failed because interface is up\n");
2655 tun
->dev
->type
= (int) arg
;
2656 tun_debug(KERN_INFO
, tun
, "linktype set to %d\n",
2668 ret
= set_offload(tun
, arg
);
2671 case TUNSETTXFILTER
:
2672 /* Can be set only for TAPs */
2674 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2676 ret
= update_filter(&tun
->txflt
, (void __user
*)arg
);
2680 /* Get hw address */
2681 memcpy(ifr
.ifr_hwaddr
.sa_data
, tun
->dev
->dev_addr
, ETH_ALEN
);
2682 ifr
.ifr_hwaddr
.sa_family
= tun
->dev
->type
;
2683 if (copy_to_user(argp
, &ifr
, ifreq_len
))
2688 /* Set hw address */
2689 tun_debug(KERN_DEBUG
, tun
, "set hw address: %pM\n",
2690 ifr
.ifr_hwaddr
.sa_data
);
2692 ret
= dev_set_mac_address(tun
->dev
, &ifr
.ifr_hwaddr
);
2696 sndbuf
= tfile
->socket
.sk
->sk_sndbuf
;
2697 if (copy_to_user(argp
, &sndbuf
, sizeof(sndbuf
)))
2702 if (copy_from_user(&sndbuf
, argp
, sizeof(sndbuf
))) {
2711 tun
->sndbuf
= sndbuf
;
2712 tun_set_sndbuf(tun
);
2715 case TUNGETVNETHDRSZ
:
2716 vnet_hdr_sz
= tun
->vnet_hdr_sz
;
2717 if (copy_to_user(argp
, &vnet_hdr_sz
, sizeof(vnet_hdr_sz
)))
2721 case TUNSETVNETHDRSZ
:
2722 if (copy_from_user(&vnet_hdr_sz
, argp
, sizeof(vnet_hdr_sz
))) {
2726 if (vnet_hdr_sz
< (int)sizeof(struct virtio_net_hdr
)) {
2731 tun
->vnet_hdr_sz
= vnet_hdr_sz
;
2735 le
= !!(tun
->flags
& TUN_VNET_LE
);
2736 if (put_user(le
, (int __user
*)argp
))
2741 if (get_user(le
, (int __user
*)argp
)) {
2746 tun
->flags
|= TUN_VNET_LE
;
2748 tun
->flags
&= ~TUN_VNET_LE
;
2752 ret
= tun_get_vnet_be(tun
, argp
);
2756 ret
= tun_set_vnet_be(tun
, argp
);
2759 case TUNATTACHFILTER
:
2760 /* Can be set only for TAPs */
2762 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2765 if (copy_from_user(&tun
->fprog
, argp
, sizeof(tun
->fprog
)))
2768 ret
= tun_attach_filter(tun
);
2771 case TUNDETACHFILTER
:
2772 /* Can be set only for TAPs */
2774 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2777 tun_detach_filter(tun
, tun
->numqueues
);
2782 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2785 if (copy_to_user(argp
, &tun
->fprog
, sizeof(tun
->fprog
)))
2792 if (copy_from_user(&carrier
, argp
, sizeof(carrier
)))
2795 ret
= tun_net_change_carrier(tun
->dev
, (bool)carrier
);
2810 static long tun_chr_ioctl(struct file
*file
,
2811 unsigned int cmd
, unsigned long arg
)
2813 return __tun_chr_ioctl(file
, cmd
, arg
, sizeof (struct ifreq
));
2816 #ifdef CONFIG_COMPAT
2817 static long tun_chr_compat_ioctl(struct file
*file
,
2818 unsigned int cmd
, unsigned long arg
)
2823 case TUNSETTXFILTER
:
2828 arg
= (unsigned long)compat_ptr(arg
);
2831 arg
= (compat_ulong_t
)arg
;
2836 * compat_ifreq is shorter than ifreq, so we must not access beyond
2837 * the end of that structure. All fields that are used in this
2838 * driver are compatible though, we don't need to convert the
2841 return __tun_chr_ioctl(file
, cmd
, arg
, sizeof(struct compat_ifreq
));
2843 #endif /* CONFIG_COMPAT */
2845 static int tun_chr_fasync(int fd
, struct file
*file
, int on
)
2847 struct tun_file
*tfile
= file
->private_data
;
2850 if ((ret
= fasync_helper(fd
, file
, on
, &tfile
->fasync
)) < 0)
2854 __f_setown(file
, task_pid(current
), PIDTYPE_PID
, 0);
2855 tfile
->flags
|= TUN_FASYNC
;
2857 tfile
->flags
&= ~TUN_FASYNC
;
2863 static int tun_chr_open(struct inode
*inode
, struct file
* file
)
2865 struct net
*net
= current
->nsproxy
->net_ns
;
2866 struct tun_file
*tfile
;
2868 DBG1(KERN_INFO
, "tunX: tun_chr_open\n");
2870 tfile
= (struct tun_file
*)sk_alloc(net
, AF_UNSPEC
, GFP_KERNEL
,
2874 if (skb_array_init(&tfile
->tx_array
, 0, GFP_KERNEL
)) {
2875 sk_free(&tfile
->sk
);
2879 mutex_init(&tfile
->napi_mutex
);
2880 RCU_INIT_POINTER(tfile
->tun
, NULL
);
2884 init_waitqueue_head(&tfile
->wq
.wait
);
2885 RCU_INIT_POINTER(tfile
->socket
.wq
, &tfile
->wq
);
2887 tfile
->socket
.file
= file
;
2888 tfile
->socket
.ops
= &tun_socket_ops
;
2890 sock_init_data(&tfile
->socket
, &tfile
->sk
);
2892 tfile
->sk
.sk_write_space
= tun_sock_write_space
;
2893 tfile
->sk
.sk_sndbuf
= INT_MAX
;
2895 file
->private_data
= tfile
;
2896 INIT_LIST_HEAD(&tfile
->next
);
2898 sock_set_flag(&tfile
->sk
, SOCK_ZEROCOPY
);
2903 static int tun_chr_close(struct inode
*inode
, struct file
*file
)
2905 struct tun_file
*tfile
= file
->private_data
;
2907 tun_detach(tfile
, true);
2912 #ifdef CONFIG_PROC_FS
2913 static void tun_chr_show_fdinfo(struct seq_file
*m
, struct file
*file
)
2915 struct tun_file
*tfile
= file
->private_data
;
2916 struct tun_struct
*tun
;
2919 memset(&ifr
, 0, sizeof(ifr
));
2922 tun
= tun_get(tfile
);
2924 tun_get_iff(current
->nsproxy
->net_ns
, tun
, &ifr
);
2930 seq_printf(m
, "iff:\t%s\n", ifr
.ifr_name
);
2934 static const struct file_operations tun_fops
= {
2935 .owner
= THIS_MODULE
,
2936 .llseek
= no_llseek
,
2937 .read_iter
= tun_chr_read_iter
,
2938 .write_iter
= tun_chr_write_iter
,
2939 .poll
= tun_chr_poll
,
2940 .unlocked_ioctl
= tun_chr_ioctl
,
2941 #ifdef CONFIG_COMPAT
2942 .compat_ioctl
= tun_chr_compat_ioctl
,
2944 .open
= tun_chr_open
,
2945 .release
= tun_chr_close
,
2946 .fasync
= tun_chr_fasync
,
2947 #ifdef CONFIG_PROC_FS
2948 .show_fdinfo
= tun_chr_show_fdinfo
,
2952 static struct miscdevice tun_miscdev
= {
2955 .nodename
= "net/tun",
2959 /* ethtool interface */
2961 static int tun_get_link_ksettings(struct net_device
*dev
,
2962 struct ethtool_link_ksettings
*cmd
)
2964 ethtool_link_ksettings_zero_link_mode(cmd
, supported
);
2965 ethtool_link_ksettings_zero_link_mode(cmd
, advertising
);
2966 cmd
->base
.speed
= SPEED_10
;
2967 cmd
->base
.duplex
= DUPLEX_FULL
;
2968 cmd
->base
.port
= PORT_TP
;
2969 cmd
->base
.phy_address
= 0;
2970 cmd
->base
.autoneg
= AUTONEG_DISABLE
;
2974 static void tun_get_drvinfo(struct net_device
*dev
, struct ethtool_drvinfo
*info
)
2976 struct tun_struct
*tun
= netdev_priv(dev
);
2978 strlcpy(info
->driver
, DRV_NAME
, sizeof(info
->driver
));
2979 strlcpy(info
->version
, DRV_VERSION
, sizeof(info
->version
));
2981 switch (tun
->flags
& TUN_TYPE_MASK
) {
2983 strlcpy(info
->bus_info
, "tun", sizeof(info
->bus_info
));
2986 strlcpy(info
->bus_info
, "tap", sizeof(info
->bus_info
));
2991 static u32
tun_get_msglevel(struct net_device
*dev
)
2994 struct tun_struct
*tun
= netdev_priv(dev
);
3001 static void tun_set_msglevel(struct net_device
*dev
, u32 value
)
3004 struct tun_struct
*tun
= netdev_priv(dev
);
3009 static int tun_get_coalesce(struct net_device
*dev
,
3010 struct ethtool_coalesce
*ec
)
3012 struct tun_struct
*tun
= netdev_priv(dev
);
3014 ec
->rx_max_coalesced_frames
= tun
->rx_batched
;
3019 static int tun_set_coalesce(struct net_device
*dev
,
3020 struct ethtool_coalesce
*ec
)
3022 struct tun_struct
*tun
= netdev_priv(dev
);
3024 if (ec
->rx_max_coalesced_frames
> NAPI_POLL_WEIGHT
)
3025 tun
->rx_batched
= NAPI_POLL_WEIGHT
;
3027 tun
->rx_batched
= ec
->rx_max_coalesced_frames
;
3032 static const struct ethtool_ops tun_ethtool_ops
= {
3033 .get_drvinfo
= tun_get_drvinfo
,
3034 .get_msglevel
= tun_get_msglevel
,
3035 .set_msglevel
= tun_set_msglevel
,
3036 .get_link
= ethtool_op_get_link
,
3037 .get_ts_info
= ethtool_op_get_ts_info
,
3038 .get_coalesce
= tun_get_coalesce
,
3039 .set_coalesce
= tun_set_coalesce
,
3040 .get_link_ksettings
= tun_get_link_ksettings
,
3043 static int tun_queue_resize(struct tun_struct
*tun
)
3045 struct net_device
*dev
= tun
->dev
;
3046 struct tun_file
*tfile
;
3047 struct skb_array
**arrays
;
3048 int n
= tun
->numqueues
+ tun
->numdisabled
;
3051 arrays
= kmalloc_array(n
, sizeof(*arrays
), GFP_KERNEL
);
3055 for (i
= 0; i
< tun
->numqueues
; i
++) {
3056 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
3057 arrays
[i
] = &tfile
->tx_array
;
3059 list_for_each_entry(tfile
, &tun
->disabled
, next
)
3060 arrays
[i
++] = &tfile
->tx_array
;
3062 ret
= skb_array_resize_multiple(arrays
, n
,
3063 dev
->tx_queue_len
, GFP_KERNEL
);
3069 static int tun_device_event(struct notifier_block
*unused
,
3070 unsigned long event
, void *ptr
)
3072 struct net_device
*dev
= netdev_notifier_info_to_dev(ptr
);
3073 struct tun_struct
*tun
= netdev_priv(dev
);
3076 if (dev
->rtnl_link_ops
!= &tun_link_ops
)
3080 case NETDEV_CHANGE_TX_QUEUE_LEN
:
3081 if (tun_queue_resize(tun
))
3085 for (i
= 0; i
< tun
->numqueues
; i
++) {
3086 struct tun_file
*tfile
;
3088 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
3089 tfile
->socket
.sk
->sk_write_space(tfile
->socket
.sk
);
3099 static struct notifier_block tun_notifier_block __read_mostly
= {
3100 .notifier_call
= tun_device_event
,
3103 static int __init
tun_init(void)
3107 pr_info("%s, %s\n", DRV_DESCRIPTION
, DRV_VERSION
);
3109 ret
= rtnl_link_register(&tun_link_ops
);
3111 pr_err("Can't register link_ops\n");
3115 ret
= misc_register(&tun_miscdev
);
3117 pr_err("Can't register misc device %d\n", TUN_MINOR
);
3121 ret
= register_netdevice_notifier(&tun_notifier_block
);
3123 pr_err("Can't register netdevice notifier\n");
3130 misc_deregister(&tun_miscdev
);
3132 rtnl_link_unregister(&tun_link_ops
);
3137 static void tun_cleanup(void)
3139 misc_deregister(&tun_miscdev
);
3140 rtnl_link_unregister(&tun_link_ops
);
3141 unregister_netdevice_notifier(&tun_notifier_block
);
3144 /* Get an underlying socket object from tun file. Returns error unless file is
3145 * attached to a device. The returned object works like a packet socket, it
3146 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
3147 * holding a reference to the file for as long as the socket is in use. */
3148 struct socket
*tun_get_socket(struct file
*file
)
3150 struct tun_file
*tfile
;
3151 if (file
->f_op
!= &tun_fops
)
3152 return ERR_PTR(-EINVAL
);
3153 tfile
= file
->private_data
;
3155 return ERR_PTR(-EBADFD
);
3156 return &tfile
->socket
;
3158 EXPORT_SYMBOL_GPL(tun_get_socket
);
3160 struct skb_array
*tun_get_skb_array(struct file
*file
)
3162 struct tun_file
*tfile
;
3164 if (file
->f_op
!= &tun_fops
)
3165 return ERR_PTR(-EINVAL
);
3166 tfile
= file
->private_data
;
3168 return ERR_PTR(-EBADFD
);
3169 return &tfile
->tx_array
;
3171 EXPORT_SYMBOL_GPL(tun_get_skb_array
);
3173 module_init(tun_init
);
3174 module_exit(tun_cleanup
);
3175 MODULE_DESCRIPTION(DRV_DESCRIPTION
);
3176 MODULE_AUTHOR(DRV_COPYRIGHT
);
3177 MODULE_LICENSE("GPL");
3178 MODULE_ALIAS_MISCDEV(TUN_MINOR
);
3179 MODULE_ALIAS("devname:net/tun");