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_tx_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
,
722 struct tun_file
*tfile
= file
->private_data
;
723 struct net_device
*dev
= tun
->dev
;
726 err
= security_tun_dev_attach(tfile
->socket
.sk
, tun
->security
);
731 if (rtnl_dereference(tfile
->tun
) && !tfile
->detached
)
735 if (!(tun
->flags
& IFF_MULTI_QUEUE
) && tun
->numqueues
== 1)
739 if (!tfile
->detached
&&
740 tun
->numqueues
+ tun
->numdisabled
== MAX_TAP_QUEUES
)
745 /* Re-attach the filter to persist device */
746 if (!skip_filter
&& (tun
->filter_attached
== true)) {
747 lock_sock(tfile
->socket
.sk
);
748 err
= sk_attach_filter(&tun
->fprog
, tfile
->socket
.sk
);
749 release_sock(tfile
->socket
.sk
);
754 if (!tfile
->detached
&&
755 skb_array_resize(&tfile
->tx_array
, dev
->tx_queue_len
, GFP_KERNEL
)) {
760 tfile
->queue_index
= tun
->numqueues
;
761 tfile
->socket
.sk
->sk_shutdown
&= ~RCV_SHUTDOWN
;
762 if (tfile
->detached
) {
763 tun_enable_queue(tfile
);
765 sock_hold(&tfile
->sk
);
766 tun_napi_init(tun
, tfile
, napi
, napi_frags
);
769 /* device is allowed to go away first, so no need to hold extra
773 /* Publish tfile->tun and tun->tfiles only after we've fully
774 * initialized tfile; otherwise we risk using half-initialized
778 rcu_assign_pointer(tfile
->tun
, tun
);
779 rcu_assign_pointer(tun
->tfiles
[tun
->numqueues
], tfile
);
781 tun_set_real_num_queues(tun
);
786 static struct tun_struct
*tun_get(struct tun_file
*tfile
)
788 struct tun_struct
*tun
;
791 tun
= rcu_dereference(tfile
->tun
);
799 static void tun_put(struct tun_struct
*tun
)
805 static void addr_hash_set(u32
*mask
, const u8
*addr
)
807 int n
= ether_crc(ETH_ALEN
, addr
) >> 26;
808 mask
[n
>> 5] |= (1 << (n
& 31));
811 static unsigned int addr_hash_test(const u32
*mask
, const u8
*addr
)
813 int n
= ether_crc(ETH_ALEN
, addr
) >> 26;
814 return mask
[n
>> 5] & (1 << (n
& 31));
817 static int update_filter(struct tap_filter
*filter
, void __user
*arg
)
819 struct { u8 u
[ETH_ALEN
]; } *addr
;
820 struct tun_filter uf
;
821 int err
, alen
, n
, nexact
;
823 if (copy_from_user(&uf
, arg
, sizeof(uf
)))
832 alen
= ETH_ALEN
* uf
.count
;
833 addr
= memdup_user(arg
+ sizeof(uf
), alen
);
835 return PTR_ERR(addr
);
837 /* The filter is updated without holding any locks. Which is
838 * perfectly safe. We disable it first and in the worst
839 * case we'll accept a few undesired packets. */
843 /* Use first set of addresses as an exact filter */
844 for (n
= 0; n
< uf
.count
&& n
< FLT_EXACT_COUNT
; n
++)
845 memcpy(filter
->addr
[n
], addr
[n
].u
, ETH_ALEN
);
849 /* Remaining multicast addresses are hashed,
850 * unicast will leave the filter disabled. */
851 memset(filter
->mask
, 0, sizeof(filter
->mask
));
852 for (; n
< uf
.count
; n
++) {
853 if (!is_multicast_ether_addr(addr
[n
].u
)) {
854 err
= 0; /* no filter */
857 addr_hash_set(filter
->mask
, addr
[n
].u
);
860 /* For ALLMULTI just set the mask to all ones.
861 * This overrides the mask populated above. */
862 if ((uf
.flags
& TUN_FLT_ALLMULTI
))
863 memset(filter
->mask
, ~0, sizeof(filter
->mask
));
865 /* Now enable the filter */
867 filter
->count
= nexact
;
869 /* Return the number of exact filters */
876 /* Returns: 0 - drop, !=0 - accept */
877 static int run_filter(struct tap_filter
*filter
, const struct sk_buff
*skb
)
879 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
881 struct ethhdr
*eh
= (struct ethhdr
*) skb
->data
;
885 for (i
= 0; i
< filter
->count
; i
++)
886 if (ether_addr_equal(eh
->h_dest
, filter
->addr
[i
]))
889 /* Inexact match (multicast only) */
890 if (is_multicast_ether_addr(eh
->h_dest
))
891 return addr_hash_test(filter
->mask
, eh
->h_dest
);
897 * Checks whether the packet is accepted or not.
898 * Returns: 0 - drop, !=0 - accept
900 static int check_filter(struct tap_filter
*filter
, const struct sk_buff
*skb
)
905 return run_filter(filter
, skb
);
908 /* Network device part of the driver */
910 static const struct ethtool_ops tun_ethtool_ops
;
912 /* Net device detach from fd. */
913 static void tun_net_uninit(struct net_device
*dev
)
918 /* Net device open. */
919 static int tun_net_open(struct net_device
*dev
)
921 netif_tx_start_all_queues(dev
);
926 /* Net device close. */
927 static int tun_net_close(struct net_device
*dev
)
929 netif_tx_stop_all_queues(dev
);
933 /* Net device start xmit */
934 static netdev_tx_t
tun_net_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
936 struct tun_struct
*tun
= netdev_priv(dev
);
937 int txq
= skb
->queue_mapping
;
938 struct tun_file
*tfile
;
942 tfile
= rcu_dereference(tun
->tfiles
[txq
]);
943 numqueues
= READ_ONCE(tun
->numqueues
);
945 /* Drop packet if interface is not attached */
946 if (txq
>= numqueues
)
950 if (numqueues
== 1 && static_key_false(&rps_needed
)) {
951 /* Select queue was not called for the skbuff, so we extract the
952 * RPS hash and save it into the flow_table here.
956 rxhash
= __skb_get_hash_symmetric(skb
);
958 struct tun_flow_entry
*e
;
959 e
= tun_flow_find(&tun
->flows
[tun_hashfn(rxhash
)],
962 tun_flow_save_rps_rxhash(e
, rxhash
);
967 tun_debug(KERN_INFO
, tun
, "tun_net_xmit %d\n", skb
->len
);
971 /* Drop if the filter does not like it.
972 * This is a noop if the filter is disabled.
973 * Filter can be enabled only for the TAP devices. */
974 if (!check_filter(&tun
->txflt
, skb
))
977 if (tfile
->socket
.sk
->sk_filter
&&
978 sk_filter(tfile
->socket
.sk
, skb
))
981 if (unlikely(skb_orphan_frags_rx(skb
, GFP_ATOMIC
)))
984 skb_tx_timestamp(skb
);
986 /* Orphan the skb - required as we might hang on to it
987 * for indefinite time.
993 if (skb_array_produce(&tfile
->tx_array
, skb
))
996 /* Notify and wake up reader process */
997 if (tfile
->flags
& TUN_FASYNC
)
998 kill_fasync(&tfile
->fasync
, SIGIO
, POLL_IN
);
999 tfile
->socket
.sk
->sk_data_ready(tfile
->socket
.sk
);
1002 return NETDEV_TX_OK
;
1005 this_cpu_inc(tun
->pcpu_stats
->tx_dropped
);
1009 return NET_XMIT_DROP
;
1012 static void tun_net_mclist(struct net_device
*dev
)
1015 * This callback is supposed to deal with mc filter in
1016 * _rx_ path and has nothing to do with the _tx_ path.
1017 * In rx path we always accept everything userspace gives us.
1021 static netdev_features_t
tun_net_fix_features(struct net_device
*dev
,
1022 netdev_features_t features
)
1024 struct tun_struct
*tun
= netdev_priv(dev
);
1026 return (features
& tun
->set_features
) | (features
& ~TUN_USER_FEATURES
);
1028 #ifdef CONFIG_NET_POLL_CONTROLLER
1029 static void tun_poll_controller(struct net_device
*dev
)
1032 * Tun only receives frames when:
1033 * 1) the char device endpoint gets data from user space
1034 * 2) the tun socket gets a sendmsg call from user space
1035 * If NAPI is not enabled, since both of those are synchronous
1036 * operations, we are guaranteed never to have pending data when we poll
1037 * for it so there is nothing to do here but return.
1038 * We need this though so netpoll recognizes us as an interface that
1039 * supports polling, which enables bridge devices in virt setups to
1040 * still use netconsole
1041 * If NAPI is enabled, however, we need to schedule polling for all
1042 * queues unless we are using napi_gro_frags(), which we call in
1043 * process context and not in NAPI context.
1045 struct tun_struct
*tun
= netdev_priv(dev
);
1047 if (tun
->flags
& IFF_NAPI
) {
1048 struct tun_file
*tfile
;
1052 for (i
= 0; i
< tun
->numqueues
; i
++) {
1053 tfile
= rcu_dereference(tun
->tfiles
[i
]);
1054 if (!tun_napi_frags_enabled(tfile
) &&
1055 tfile
->napi_enabled
)
1056 napi_schedule(&tfile
->napi
);
1064 static void tun_set_headroom(struct net_device
*dev
, int new_hr
)
1066 struct tun_struct
*tun
= netdev_priv(dev
);
1068 if (new_hr
< NET_SKB_PAD
)
1069 new_hr
= NET_SKB_PAD
;
1071 tun
->align
= new_hr
;
1075 tun_net_get_stats64(struct net_device
*dev
, struct rtnl_link_stats64
*stats
)
1077 u32 rx_dropped
= 0, tx_dropped
= 0, rx_frame_errors
= 0;
1078 struct tun_struct
*tun
= netdev_priv(dev
);
1079 struct tun_pcpu_stats
*p
;
1082 for_each_possible_cpu(i
) {
1083 u64 rxpackets
, rxbytes
, txpackets
, txbytes
;
1086 p
= per_cpu_ptr(tun
->pcpu_stats
, i
);
1088 start
= u64_stats_fetch_begin(&p
->syncp
);
1089 rxpackets
= p
->rx_packets
;
1090 rxbytes
= p
->rx_bytes
;
1091 txpackets
= p
->tx_packets
;
1092 txbytes
= p
->tx_bytes
;
1093 } while (u64_stats_fetch_retry(&p
->syncp
, start
));
1095 stats
->rx_packets
+= rxpackets
;
1096 stats
->rx_bytes
+= rxbytes
;
1097 stats
->tx_packets
+= txpackets
;
1098 stats
->tx_bytes
+= txbytes
;
1101 rx_dropped
+= p
->rx_dropped
;
1102 rx_frame_errors
+= p
->rx_frame_errors
;
1103 tx_dropped
+= p
->tx_dropped
;
1105 stats
->rx_dropped
= rx_dropped
;
1106 stats
->rx_frame_errors
= rx_frame_errors
;
1107 stats
->tx_dropped
= tx_dropped
;
1110 static int tun_xdp_set(struct net_device
*dev
, struct bpf_prog
*prog
,
1111 struct netlink_ext_ack
*extack
)
1113 struct tun_struct
*tun
= netdev_priv(dev
);
1114 struct bpf_prog
*old_prog
;
1116 old_prog
= rtnl_dereference(tun
->xdp_prog
);
1117 rcu_assign_pointer(tun
->xdp_prog
, prog
);
1119 bpf_prog_put(old_prog
);
1124 static u32
tun_xdp_query(struct net_device
*dev
)
1126 struct tun_struct
*tun
= netdev_priv(dev
);
1127 const struct bpf_prog
*xdp_prog
;
1129 xdp_prog
= rtnl_dereference(tun
->xdp_prog
);
1131 return xdp_prog
->aux
->id
;
1136 static int tun_xdp(struct net_device
*dev
, struct netdev_bpf
*xdp
)
1138 switch (xdp
->command
) {
1139 case XDP_SETUP_PROG
:
1140 return tun_xdp_set(dev
, xdp
->prog
, xdp
->extack
);
1141 case XDP_QUERY_PROG
:
1142 xdp
->prog_id
= tun_xdp_query(dev
);
1143 xdp
->prog_attached
= !!xdp
->prog_id
;
1150 static int tun_net_change_carrier(struct net_device
*dev
, bool new_carrier
)
1153 struct tun_struct
*tun
= netdev_priv(dev
);
1155 if (!tun
->numqueues
)
1158 netif_carrier_on(dev
);
1160 netif_carrier_off(dev
);
1165 static const struct net_device_ops tun_netdev_ops
= {
1166 .ndo_uninit
= tun_net_uninit
,
1167 .ndo_open
= tun_net_open
,
1168 .ndo_stop
= tun_net_close
,
1169 .ndo_start_xmit
= tun_net_xmit
,
1170 .ndo_fix_features
= tun_net_fix_features
,
1171 .ndo_select_queue
= tun_select_queue
,
1172 #ifdef CONFIG_NET_POLL_CONTROLLER
1173 .ndo_poll_controller
= tun_poll_controller
,
1175 .ndo_set_rx_headroom
= tun_set_headroom
,
1176 .ndo_get_stats64
= tun_net_get_stats64
,
1177 .ndo_change_carrier
= tun_net_change_carrier
,
1180 static const struct net_device_ops tap_netdev_ops
= {
1181 .ndo_uninit
= tun_net_uninit
,
1182 .ndo_open
= tun_net_open
,
1183 .ndo_stop
= tun_net_close
,
1184 .ndo_start_xmit
= tun_net_xmit
,
1185 .ndo_fix_features
= tun_net_fix_features
,
1186 .ndo_set_rx_mode
= tun_net_mclist
,
1187 .ndo_set_mac_address
= eth_mac_addr
,
1188 .ndo_validate_addr
= eth_validate_addr
,
1189 .ndo_select_queue
= tun_select_queue
,
1190 #ifdef CONFIG_NET_POLL_CONTROLLER
1191 .ndo_poll_controller
= tun_poll_controller
,
1193 .ndo_features_check
= passthru_features_check
,
1194 .ndo_set_rx_headroom
= tun_set_headroom
,
1195 .ndo_get_stats64
= tun_net_get_stats64
,
1197 .ndo_change_carrier
= tun_net_change_carrier
,
1200 static void tun_flow_init(struct tun_struct
*tun
)
1204 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++)
1205 INIT_HLIST_HEAD(&tun
->flows
[i
]);
1207 tun
->ageing_time
= TUN_FLOW_EXPIRE
;
1208 timer_setup(&tun
->flow_gc_timer
, tun_flow_cleanup
, 0);
1209 mod_timer(&tun
->flow_gc_timer
,
1210 round_jiffies_up(jiffies
+ tun
->ageing_time
));
1213 static void tun_flow_uninit(struct tun_struct
*tun
)
1215 del_timer_sync(&tun
->flow_gc_timer
);
1216 tun_flow_flush(tun
);
1220 #define MAX_MTU 65535
1222 /* Initialize net device. */
1223 static void tun_net_init(struct net_device
*dev
)
1225 struct tun_struct
*tun
= netdev_priv(dev
);
1227 switch (tun
->flags
& TUN_TYPE_MASK
) {
1229 dev
->netdev_ops
= &tun_netdev_ops
;
1231 /* Point-to-Point TUN Device */
1232 dev
->hard_header_len
= 0;
1236 /* Zero header length */
1237 dev
->type
= ARPHRD_NONE
;
1238 dev
->flags
= IFF_POINTOPOINT
| IFF_NOARP
| IFF_MULTICAST
;
1242 dev
->netdev_ops
= &tap_netdev_ops
;
1243 /* Ethernet TAP Device */
1245 dev
->priv_flags
&= ~IFF_TX_SKB_SHARING
;
1246 dev
->priv_flags
|= IFF_LIVE_ADDR_CHANGE
;
1248 eth_hw_addr_random(dev
);
1253 dev
->min_mtu
= MIN_MTU
;
1254 dev
->max_mtu
= MAX_MTU
- dev
->hard_header_len
;
1257 static bool tun_sock_writeable(struct tun_struct
*tun
, struct tun_file
*tfile
)
1259 struct sock
*sk
= tfile
->socket
.sk
;
1261 return (tun
->dev
->flags
& IFF_UP
) && sock_writeable(sk
);
1264 /* Character device part */
1267 static unsigned int tun_chr_poll(struct file
*file
, poll_table
*wait
)
1269 struct tun_file
*tfile
= file
->private_data
;
1270 struct tun_struct
*tun
= tun_get(tfile
);
1272 unsigned int mask
= 0;
1277 sk
= tfile
->socket
.sk
;
1279 tun_debug(KERN_INFO
, tun
, "tun_chr_poll\n");
1281 poll_wait(file
, sk_sleep(sk
), wait
);
1283 if (!skb_array_empty(&tfile
->tx_array
))
1284 mask
|= POLLIN
| POLLRDNORM
;
1286 /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1287 * guarantee EPOLLOUT to be raised by either here or
1288 * tun_sock_write_space(). Then process could get notification
1289 * after it writes to a down device and meets -EIO.
1291 if (tun_sock_writeable(tun
, tfile
) ||
1292 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
) &&
1293 tun_sock_writeable(tun
, tfile
)))
1294 mask
|= POLLOUT
| POLLWRNORM
;
1296 if (tun
->dev
->reg_state
!= NETREG_REGISTERED
)
1303 static struct sk_buff
*tun_napi_alloc_frags(struct tun_file
*tfile
,
1305 const struct iov_iter
*it
)
1307 struct sk_buff
*skb
;
1312 if (it
->nr_segs
> MAX_SKB_FRAGS
+ 1)
1313 return ERR_PTR(-ENOMEM
);
1316 skb
= napi_get_frags(&tfile
->napi
);
1319 return ERR_PTR(-ENOMEM
);
1321 linear
= iov_iter_single_seg_count(it
);
1322 err
= __skb_grow(skb
, linear
);
1327 skb
->data_len
= len
- linear
;
1328 skb
->truesize
+= skb
->data_len
;
1330 for (i
= 1; i
< it
->nr_segs
; i
++) {
1331 struct page_frag
*pfrag
= ¤t
->task_frag
;
1332 size_t fragsz
= it
->iov
[i
].iov_len
;
1334 if (fragsz
== 0 || fragsz
> PAGE_SIZE
) {
1339 if (!skb_page_frag_refill(fragsz
, pfrag
, GFP_KERNEL
)) {
1344 skb_fill_page_desc(skb
, i
- 1, pfrag
->page
,
1345 pfrag
->offset
, fragsz
);
1346 page_ref_inc(pfrag
->page
);
1347 pfrag
->offset
+= fragsz
;
1352 /* frees skb and all frags allocated with napi_alloc_frag() */
1353 napi_free_frags(&tfile
->napi
);
1354 return ERR_PTR(err
);
1357 /* prepad is the amount to reserve at front. len is length after that.
1358 * linear is a hint as to how much to copy (usually headers). */
1359 static struct sk_buff
*tun_alloc_skb(struct tun_file
*tfile
,
1360 size_t prepad
, size_t len
,
1361 size_t linear
, int noblock
)
1363 struct sock
*sk
= tfile
->socket
.sk
;
1364 struct sk_buff
*skb
;
1367 /* Under a page? Don't bother with paged skb. */
1368 if (prepad
+ len
< PAGE_SIZE
|| !linear
)
1371 skb
= sock_alloc_send_pskb(sk
, prepad
+ linear
, len
- linear
, noblock
,
1374 return ERR_PTR(err
);
1376 skb_reserve(skb
, prepad
);
1377 skb_put(skb
, linear
);
1378 skb
->data_len
= len
- linear
;
1379 skb
->len
+= len
- linear
;
1384 static void tun_rx_batched(struct tun_struct
*tun
, struct tun_file
*tfile
,
1385 struct sk_buff
*skb
, int more
)
1387 struct sk_buff_head
*queue
= &tfile
->sk
.sk_write_queue
;
1388 struct sk_buff_head process_queue
;
1389 u32 rx_batched
= tun
->rx_batched
;
1392 if (!rx_batched
|| (!more
&& skb_queue_empty(queue
))) {
1394 skb_record_rx_queue(skb
, tfile
->queue_index
);
1395 netif_receive_skb(skb
);
1400 spin_lock(&queue
->lock
);
1401 if (!more
|| skb_queue_len(queue
) == rx_batched
) {
1402 __skb_queue_head_init(&process_queue
);
1403 skb_queue_splice_tail_init(queue
, &process_queue
);
1406 __skb_queue_tail(queue
, skb
);
1408 spin_unlock(&queue
->lock
);
1411 struct sk_buff
*nskb
;
1414 while ((nskb
= __skb_dequeue(&process_queue
))) {
1415 skb_record_rx_queue(nskb
, tfile
->queue_index
);
1416 netif_receive_skb(nskb
);
1418 skb_record_rx_queue(skb
, tfile
->queue_index
);
1419 netif_receive_skb(skb
);
1424 static bool tun_can_build_skb(struct tun_struct
*tun
, struct tun_file
*tfile
,
1425 int len
, int noblock
, bool zerocopy
)
1427 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
1430 if (tfile
->socket
.sk
->sk_sndbuf
!= INT_MAX
)
1439 if (SKB_DATA_ALIGN(len
+ TUN_RX_PAD
) +
1440 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
)) > PAGE_SIZE
)
1446 static struct sk_buff
*tun_build_skb(struct tun_struct
*tun
,
1447 struct tun_file
*tfile
,
1448 struct iov_iter
*from
,
1449 struct virtio_net_hdr
*hdr
,
1450 int len
, int *skb_xdp
)
1452 struct page_frag
*alloc_frag
= ¤t
->task_frag
;
1453 struct sk_buff
*skb
;
1454 struct bpf_prog
*xdp_prog
;
1455 int buflen
= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
1456 unsigned int delta
= 0;
1459 bool xdp_xmit
= false;
1460 int err
, pad
= TUN_RX_PAD
;
1463 xdp_prog
= rcu_dereference(tun
->xdp_prog
);
1465 pad
+= TUN_HEADROOM
;
1466 buflen
+= SKB_DATA_ALIGN(len
+ pad
);
1469 alloc_frag
->offset
= ALIGN((u64
)alloc_frag
->offset
, SMP_CACHE_BYTES
);
1470 if (unlikely(!skb_page_frag_refill(buflen
, alloc_frag
, GFP_KERNEL
)))
1471 return ERR_PTR(-ENOMEM
);
1473 buf
= (char *)page_address(alloc_frag
->page
) + alloc_frag
->offset
;
1474 copied
= copy_page_from_iter(alloc_frag
->page
,
1475 alloc_frag
->offset
+ pad
,
1478 return ERR_PTR(-EFAULT
);
1480 /* There's a small window that XDP may be set after the check
1481 * of xdp_prog above, this should be rare and for simplicity
1482 * we do XDP on skb in case the headroom is not enough.
1484 if (hdr
->gso_type
|| !xdp_prog
)
1491 xdp_prog
= rcu_dereference(tun
->xdp_prog
);
1492 if (xdp_prog
&& !*skb_xdp
) {
1493 struct xdp_buff xdp
;
1497 xdp
.data_hard_start
= buf
;
1498 xdp
.data
= buf
+ pad
;
1499 xdp_set_data_meta_invalid(&xdp
);
1500 xdp
.data_end
= xdp
.data
+ len
;
1501 orig_data
= xdp
.data
;
1502 act
= bpf_prog_run_xdp(xdp_prog
, &xdp
);
1506 get_page(alloc_frag
->page
);
1507 alloc_frag
->offset
+= buflen
;
1508 err
= xdp_do_redirect(tun
->dev
, &xdp
, xdp_prog
);
1519 delta
= orig_data
- xdp
.data
;
1522 bpf_warn_invalid_xdp_action(act
);
1525 trace_xdp_exception(tun
->dev
, xdp_prog
, act
);
1532 skb
= build_skb(buf
, buflen
);
1536 return ERR_PTR(-ENOMEM
);
1539 skb_reserve(skb
, pad
- delta
);
1540 skb_put(skb
, len
+ delta
);
1541 skb_set_owner_w(skb
, tfile
->socket
.sk
);
1542 get_page(alloc_frag
->page
);
1543 alloc_frag
->offset
+= buflen
;
1546 skb
->dev
= tun
->dev
;
1547 generic_xdp_tx(skb
, xdp_prog
);
1559 put_page(alloc_frag
->page
);
1563 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1567 /* Get packet from user space buffer */
1568 static ssize_t
tun_get_user(struct tun_struct
*tun
, struct tun_file
*tfile
,
1569 void *msg_control
, struct iov_iter
*from
,
1570 int noblock
, bool more
)
1572 struct tun_pi pi
= { 0, cpu_to_be16(ETH_P_IP
) };
1573 struct sk_buff
*skb
;
1574 size_t total_len
= iov_iter_count(from
);
1575 size_t len
= total_len
, align
= tun
->align
, linear
;
1576 struct virtio_net_hdr gso
= { 0 };
1577 struct tun_pcpu_stats
*stats
;
1580 bool zerocopy
= false;
1584 bool frags
= tun_napi_frags_enabled(tfile
);
1586 if (!(tun
->flags
& IFF_NO_PI
)) {
1587 if (len
< sizeof(pi
))
1591 if (!copy_from_iter_full(&pi
, sizeof(pi
), from
))
1595 if (tun
->flags
& IFF_VNET_HDR
) {
1596 int vnet_hdr_sz
= READ_ONCE(tun
->vnet_hdr_sz
);
1598 if (len
< vnet_hdr_sz
)
1602 if (!copy_from_iter_full(&gso
, sizeof(gso
), from
))
1605 if ((gso
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) &&
1606 tun16_to_cpu(tun
, gso
.csum_start
) + tun16_to_cpu(tun
, gso
.csum_offset
) + 2 > tun16_to_cpu(tun
, gso
.hdr_len
))
1607 gso
.hdr_len
= cpu_to_tun16(tun
, tun16_to_cpu(tun
, gso
.csum_start
) + tun16_to_cpu(tun
, gso
.csum_offset
) + 2);
1609 if (tun16_to_cpu(tun
, gso
.hdr_len
) > len
)
1611 iov_iter_advance(from
, vnet_hdr_sz
- sizeof(gso
));
1614 if ((tun
->flags
& TUN_TYPE_MASK
) == IFF_TAP
) {
1615 align
+= NET_IP_ALIGN
;
1616 if (unlikely(len
< ETH_HLEN
||
1617 (gso
.hdr_len
&& tun16_to_cpu(tun
, gso
.hdr_len
) < ETH_HLEN
)))
1621 good_linear
= SKB_MAX_HEAD(align
);
1624 struct iov_iter i
= *from
;
1626 /* There are 256 bytes to be copied in skb, so there is
1627 * enough room for skb expand head in case it is used.
1628 * The rest of the buffer is mapped from userspace.
1630 copylen
= gso
.hdr_len
? tun16_to_cpu(tun
, gso
.hdr_len
) : GOODCOPY_LEN
;
1631 if (copylen
> good_linear
)
1632 copylen
= good_linear
;
1634 iov_iter_advance(&i
, copylen
);
1635 if (iov_iter_npages(&i
, INT_MAX
) <= MAX_SKB_FRAGS
)
1639 if (!frags
&& tun_can_build_skb(tun
, tfile
, len
, noblock
, zerocopy
)) {
1640 /* For the packet that is not easy to be processed
1641 * (e.g gso or jumbo packet), we will do it at after
1642 * skb was created with generic XDP routine.
1644 skb
= tun_build_skb(tun
, tfile
, from
, &gso
, len
, &skb_xdp
);
1646 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1647 return PTR_ERR(skb
);
1654 if (tun16_to_cpu(tun
, gso
.hdr_len
) > good_linear
)
1655 linear
= good_linear
;
1657 linear
= tun16_to_cpu(tun
, gso
.hdr_len
);
1661 mutex_lock(&tfile
->napi_mutex
);
1662 skb
= tun_napi_alloc_frags(tfile
, copylen
, from
);
1663 /* tun_napi_alloc_frags() enforces a layout for the skb.
1664 * If zerocopy is enabled, then this layout will be
1665 * overwritten by zerocopy_sg_from_iter().
1669 skb
= tun_alloc_skb(tfile
, align
, copylen
, linear
,
1674 if (PTR_ERR(skb
) != -EAGAIN
)
1675 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1677 mutex_unlock(&tfile
->napi_mutex
);
1678 return PTR_ERR(skb
);
1682 err
= zerocopy_sg_from_iter(skb
, from
);
1684 err
= skb_copy_datagram_from_iter(skb
, 0, from
, len
);
1689 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1692 tfile
->napi
.skb
= NULL
;
1693 mutex_unlock(&tfile
->napi_mutex
);
1700 if (virtio_net_hdr_to_skb(skb
, &gso
, tun_is_little_endian(tun
))) {
1701 this_cpu_inc(tun
->pcpu_stats
->rx_frame_errors
);
1704 tfile
->napi
.skb
= NULL
;
1705 mutex_unlock(&tfile
->napi_mutex
);
1711 switch (tun
->flags
& TUN_TYPE_MASK
) {
1713 if (tun
->flags
& IFF_NO_PI
) {
1714 u8 ip_version
= skb
->len
? (skb
->data
[0] >> 4) : 0;
1716 switch (ip_version
) {
1718 pi
.proto
= htons(ETH_P_IP
);
1721 pi
.proto
= htons(ETH_P_IPV6
);
1724 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1730 skb_reset_mac_header(skb
);
1731 skb
->protocol
= pi
.proto
;
1732 skb
->dev
= tun
->dev
;
1736 skb
->protocol
= eth_type_trans(skb
, tun
->dev
);
1740 /* copy skb_ubuf_info for callback when skb has no error */
1742 skb_shinfo(skb
)->destructor_arg
= msg_control
;
1743 skb_shinfo(skb
)->tx_flags
|= SKBTX_DEV_ZEROCOPY
;
1744 skb_shinfo(skb
)->tx_flags
|= SKBTX_SHARED_FRAG
;
1745 } else if (msg_control
) {
1746 struct ubuf_info
*uarg
= msg_control
;
1747 uarg
->callback(uarg
, false);
1750 skb_reset_network_header(skb
);
1751 skb_probe_transport_header(skb
, 0);
1754 struct bpf_prog
*xdp_prog
;
1759 xdp_prog
= rcu_dereference(tun
->xdp_prog
);
1761 ret
= do_xdp_generic(xdp_prog
, skb
);
1762 if (ret
!= XDP_PASS
) {
1766 tfile
->napi
.skb
= NULL
;
1767 mutex_unlock(&tfile
->napi_mutex
);
1776 rxhash
= __skb_get_hash_symmetric(skb
);
1779 if (unlikely(!(tun
->dev
->flags
& IFF_UP
))) {
1786 /* Exercise flow dissector code path. */
1787 u32 headlen
= eth_get_headlen(skb
->data
, skb_headlen(skb
));
1789 if (unlikely(headlen
> skb_headlen(skb
))) {
1790 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1791 napi_free_frags(&tfile
->napi
);
1793 mutex_unlock(&tfile
->napi_mutex
);
1799 napi_gro_frags(&tfile
->napi
);
1801 mutex_unlock(&tfile
->napi_mutex
);
1802 } else if (tfile
->napi_enabled
) {
1803 struct sk_buff_head
*queue
= &tfile
->sk
.sk_write_queue
;
1806 spin_lock_bh(&queue
->lock
);
1807 __skb_queue_tail(queue
, skb
);
1808 queue_len
= skb_queue_len(queue
);
1809 spin_unlock(&queue
->lock
);
1811 if (!more
|| queue_len
> NAPI_POLL_WEIGHT
)
1812 napi_schedule(&tfile
->napi
);
1815 } else if (!IS_ENABLED(CONFIG_4KSTACKS
)) {
1816 tun_rx_batched(tun
, tfile
, skb
, more
);
1822 stats
= get_cpu_ptr(tun
->pcpu_stats
);
1823 u64_stats_update_begin(&stats
->syncp
);
1824 stats
->rx_packets
++;
1825 stats
->rx_bytes
+= len
;
1826 u64_stats_update_end(&stats
->syncp
);
1829 tun_flow_update(tun
, rxhash
, tfile
);
1833 static ssize_t
tun_chr_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
1835 struct file
*file
= iocb
->ki_filp
;
1836 struct tun_file
*tfile
= file
->private_data
;
1837 struct tun_struct
*tun
= tun_get(tfile
);
1843 result
= tun_get_user(tun
, tfile
, NULL
, from
,
1844 file
->f_flags
& O_NONBLOCK
, false);
1850 /* Put packet to the user space buffer */
1851 static ssize_t
tun_put_user(struct tun_struct
*tun
,
1852 struct tun_file
*tfile
,
1853 struct sk_buff
*skb
,
1854 struct iov_iter
*iter
)
1856 struct tun_pi pi
= { 0, skb
->protocol
};
1857 struct tun_pcpu_stats
*stats
;
1859 int vlan_offset
= 0;
1861 int vnet_hdr_sz
= 0;
1863 if (skb_vlan_tag_present(skb
))
1864 vlan_hlen
= VLAN_HLEN
;
1866 if (tun
->flags
& IFF_VNET_HDR
)
1867 vnet_hdr_sz
= READ_ONCE(tun
->vnet_hdr_sz
);
1869 total
= skb
->len
+ vlan_hlen
+ vnet_hdr_sz
;
1871 if (!(tun
->flags
& IFF_NO_PI
)) {
1872 if (iov_iter_count(iter
) < sizeof(pi
))
1875 total
+= sizeof(pi
);
1876 if (iov_iter_count(iter
) < total
) {
1877 /* Packet will be striped */
1878 pi
.flags
|= TUN_PKT_STRIP
;
1881 if (copy_to_iter(&pi
, sizeof(pi
), iter
) != sizeof(pi
))
1886 struct virtio_net_hdr gso
;
1888 if (iov_iter_count(iter
) < vnet_hdr_sz
)
1891 if (virtio_net_hdr_from_skb(skb
, &gso
,
1892 tun_is_little_endian(tun
), true,
1894 struct skb_shared_info
*sinfo
= skb_shinfo(skb
);
1895 pr_err("unexpected GSO type: "
1896 "0x%x, gso_size %d, hdr_len %d\n",
1897 sinfo
->gso_type
, tun16_to_cpu(tun
, gso
.gso_size
),
1898 tun16_to_cpu(tun
, gso
.hdr_len
));
1899 print_hex_dump(KERN_ERR
, "tun: ",
1902 min((int)tun16_to_cpu(tun
, gso
.hdr_len
), 64), true);
1907 if (copy_to_iter(&gso
, sizeof(gso
), iter
) != sizeof(gso
))
1910 iov_iter_advance(iter
, vnet_hdr_sz
- sizeof(gso
));
1916 __be16 h_vlan_proto
;
1920 veth
.h_vlan_proto
= skb
->vlan_proto
;
1921 veth
.h_vlan_TCI
= htons(skb_vlan_tag_get(skb
));
1923 vlan_offset
= offsetof(struct vlan_ethhdr
, h_vlan_proto
);
1925 ret
= skb_copy_datagram_iter(skb
, 0, iter
, vlan_offset
);
1926 if (ret
|| !iov_iter_count(iter
))
1929 ret
= copy_to_iter(&veth
, sizeof(veth
), iter
);
1930 if (ret
!= sizeof(veth
) || !iov_iter_count(iter
))
1934 skb_copy_datagram_iter(skb
, vlan_offset
, iter
, skb
->len
- vlan_offset
);
1937 /* caller is in process context, */
1938 stats
= get_cpu_ptr(tun
->pcpu_stats
);
1939 u64_stats_update_begin(&stats
->syncp
);
1940 stats
->tx_packets
++;
1941 stats
->tx_bytes
+= skb
->len
+ vlan_hlen
;
1942 u64_stats_update_end(&stats
->syncp
);
1943 put_cpu_ptr(tun
->pcpu_stats
);
1948 static struct sk_buff
*tun_ring_recv(struct tun_file
*tfile
, int noblock
,
1951 DECLARE_WAITQUEUE(wait
, current
);
1952 struct sk_buff
*skb
= NULL
;
1955 skb
= skb_array_consume(&tfile
->tx_array
);
1963 add_wait_queue(&tfile
->wq
.wait
, &wait
);
1966 set_current_state(TASK_INTERRUPTIBLE
);
1967 skb
= skb_array_consume(&tfile
->tx_array
);
1970 if (signal_pending(current
)) {
1971 error
= -ERESTARTSYS
;
1974 if (tfile
->socket
.sk
->sk_shutdown
& RCV_SHUTDOWN
) {
1982 __set_current_state(TASK_RUNNING
);
1983 remove_wait_queue(&tfile
->wq
.wait
, &wait
);
1990 static ssize_t
tun_do_read(struct tun_struct
*tun
, struct tun_file
*tfile
,
1991 struct iov_iter
*to
,
1992 int noblock
, struct sk_buff
*skb
)
1997 tun_debug(KERN_INFO
, tun
, "tun_do_read\n");
1999 if (!iov_iter_count(to
)) {
2006 /* Read frames from ring */
2007 skb
= tun_ring_recv(tfile
, noblock
, &err
);
2012 ret
= tun_put_user(tun
, tfile
, skb
, to
);
2013 if (unlikely(ret
< 0))
2021 static ssize_t
tun_chr_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
2023 struct file
*file
= iocb
->ki_filp
;
2024 struct tun_file
*tfile
= file
->private_data
;
2025 struct tun_struct
*tun
= tun_get(tfile
);
2026 ssize_t len
= iov_iter_count(to
), ret
;
2030 ret
= tun_do_read(tun
, tfile
, to
, file
->f_flags
& O_NONBLOCK
, NULL
);
2031 ret
= min_t(ssize_t
, ret
, len
);
2038 static void tun_free_netdev(struct net_device
*dev
)
2040 struct tun_struct
*tun
= netdev_priv(dev
);
2042 BUG_ON(!(list_empty(&tun
->disabled
)));
2043 free_percpu(tun
->pcpu_stats
);
2044 tun_flow_uninit(tun
);
2045 security_tun_dev_free_security(tun
->security
);
2048 static void tun_setup(struct net_device
*dev
)
2050 struct tun_struct
*tun
= netdev_priv(dev
);
2052 tun
->owner
= INVALID_UID
;
2053 tun
->group
= INVALID_GID
;
2055 dev
->ethtool_ops
= &tun_ethtool_ops
;
2056 dev
->needs_free_netdev
= true;
2057 dev
->priv_destructor
= tun_free_netdev
;
2058 /* We prefer our own queue length */
2059 dev
->tx_queue_len
= TUN_READQ_SIZE
;
2062 /* Trivial set of netlink ops to allow deleting tun or tap
2063 * device with netlink.
2065 static int tun_validate(struct nlattr
*tb
[], struct nlattr
*data
[],
2066 struct netlink_ext_ack
*extack
)
2068 NL_SET_ERR_MSG(extack
,
2069 "tun/tap creation via rtnetlink is not supported.");
2073 static struct rtnl_link_ops tun_link_ops __read_mostly
= {
2075 .priv_size
= sizeof(struct tun_struct
),
2077 .validate
= tun_validate
,
2080 static void tun_sock_write_space(struct sock
*sk
)
2082 struct tun_file
*tfile
;
2083 wait_queue_head_t
*wqueue
;
2085 if (!sock_writeable(sk
))
2088 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
))
2091 wqueue
= sk_sleep(sk
);
2092 if (wqueue
&& waitqueue_active(wqueue
))
2093 wake_up_interruptible_sync_poll(wqueue
, POLLOUT
|
2094 POLLWRNORM
| POLLWRBAND
);
2096 tfile
= container_of(sk
, struct tun_file
, sk
);
2097 kill_fasync(&tfile
->fasync
, SIGIO
, POLL_OUT
);
2100 static int tun_sendmsg(struct socket
*sock
, struct msghdr
*m
, size_t total_len
)
2103 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
2104 struct tun_struct
*tun
= tun_get(tfile
);
2109 ret
= tun_get_user(tun
, tfile
, m
->msg_control
, &m
->msg_iter
,
2110 m
->msg_flags
& MSG_DONTWAIT
,
2111 m
->msg_flags
& MSG_MORE
);
2116 static int tun_recvmsg(struct socket
*sock
, struct msghdr
*m
, size_t total_len
,
2119 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
2120 struct tun_struct
*tun
= tun_get(tfile
);
2121 struct sk_buff
*skb
= m
->msg_control
;
2129 if (flags
& ~(MSG_DONTWAIT
|MSG_TRUNC
|MSG_ERRQUEUE
)) {
2133 if (flags
& MSG_ERRQUEUE
) {
2134 ret
= sock_recv_errqueue(sock
->sk
, m
, total_len
,
2135 SOL_PACKET
, TUN_TX_TIMESTAMP
);
2138 ret
= tun_do_read(tun
, tfile
, &m
->msg_iter
, flags
& MSG_DONTWAIT
, skb
);
2139 if (ret
> (ssize_t
)total_len
) {
2140 m
->msg_flags
|= MSG_TRUNC
;
2141 ret
= flags
& MSG_TRUNC
? ret
: total_len
;
2155 static int tun_peek_len(struct socket
*sock
)
2157 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
2158 struct tun_struct
*tun
;
2161 tun
= tun_get(tfile
);
2165 ret
= skb_array_peek_len(&tfile
->tx_array
);
2171 /* Ops structure to mimic raw sockets with tun */
2172 static const struct proto_ops tun_socket_ops
= {
2173 .peek_len
= tun_peek_len
,
2174 .sendmsg
= tun_sendmsg
,
2175 .recvmsg
= tun_recvmsg
,
2178 static struct proto tun_proto
= {
2180 .owner
= THIS_MODULE
,
2181 .obj_size
= sizeof(struct tun_file
),
2184 static int tun_flags(struct tun_struct
*tun
)
2186 return tun
->flags
& (TUN_FEATURES
| IFF_PERSIST
| IFF_TUN
| IFF_TAP
);
2189 static ssize_t
tun_show_flags(struct device
*dev
, struct device_attribute
*attr
,
2192 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
2193 return sprintf(buf
, "0x%x\n", tun_flags(tun
));
2196 static ssize_t
tun_show_owner(struct device
*dev
, struct device_attribute
*attr
,
2199 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
2200 return uid_valid(tun
->owner
)?
2201 sprintf(buf
, "%u\n",
2202 from_kuid_munged(current_user_ns(), tun
->owner
)):
2203 sprintf(buf
, "-1\n");
2206 static ssize_t
tun_show_group(struct device
*dev
, struct device_attribute
*attr
,
2209 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
2210 return gid_valid(tun
->group
) ?
2211 sprintf(buf
, "%u\n",
2212 from_kgid_munged(current_user_ns(), tun
->group
)):
2213 sprintf(buf
, "-1\n");
2216 static DEVICE_ATTR(tun_flags
, 0444, tun_show_flags
, NULL
);
2217 static DEVICE_ATTR(owner
, 0444, tun_show_owner
, NULL
);
2218 static DEVICE_ATTR(group
, 0444, tun_show_group
, NULL
);
2220 static struct attribute
*tun_dev_attrs
[] = {
2221 &dev_attr_tun_flags
.attr
,
2222 &dev_attr_owner
.attr
,
2223 &dev_attr_group
.attr
,
2227 static const struct attribute_group tun_attr_group
= {
2228 .attrs
= tun_dev_attrs
2231 static int tun_set_iff(struct net
*net
, struct file
*file
, struct ifreq
*ifr
)
2233 struct tun_struct
*tun
;
2234 struct tun_file
*tfile
= file
->private_data
;
2235 struct net_device
*dev
;
2238 if (tfile
->detached
)
2241 if ((ifr
->ifr_flags
& IFF_NAPI_FRAGS
)) {
2242 if (!capable(CAP_NET_ADMIN
))
2245 if (!(ifr
->ifr_flags
& IFF_NAPI
) ||
2246 (ifr
->ifr_flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2250 dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
2252 if (ifr
->ifr_flags
& IFF_TUN_EXCL
)
2254 if ((ifr
->ifr_flags
& IFF_TUN
) && dev
->netdev_ops
== &tun_netdev_ops
)
2255 tun
= netdev_priv(dev
);
2256 else if ((ifr
->ifr_flags
& IFF_TAP
) && dev
->netdev_ops
== &tap_netdev_ops
)
2257 tun
= netdev_priv(dev
);
2261 if (!!(ifr
->ifr_flags
& IFF_MULTI_QUEUE
) !=
2262 !!(tun
->flags
& IFF_MULTI_QUEUE
))
2265 if (tun_not_capable(tun
))
2267 err
= security_tun_dev_open(tun
->security
);
2271 err
= tun_attach(tun
, file
, ifr
->ifr_flags
& IFF_NOFILTER
,
2272 ifr
->ifr_flags
& IFF_NAPI
,
2273 ifr
->ifr_flags
& IFF_NAPI_FRAGS
, true);
2277 if (tun
->flags
& IFF_MULTI_QUEUE
&&
2278 (tun
->numqueues
+ tun
->numdisabled
> 1)) {
2279 /* One or more queue has already been attached, no need
2280 * to initialize the device again.
2287 unsigned long flags
= 0;
2288 int queues
= ifr
->ifr_flags
& IFF_MULTI_QUEUE
?
2291 if (!ns_capable(net
->user_ns
, CAP_NET_ADMIN
))
2293 err
= security_tun_dev_create();
2298 if (ifr
->ifr_flags
& IFF_TUN
) {
2302 } else if (ifr
->ifr_flags
& IFF_TAP
) {
2310 name
= ifr
->ifr_name
;
2312 dev
= alloc_netdev_mqs(sizeof(struct tun_struct
), name
,
2313 NET_NAME_UNKNOWN
, tun_setup
, queues
,
2318 err
= dev_get_valid_name(net
, dev
, name
);
2322 dev_net_set(dev
, net
);
2323 dev
->rtnl_link_ops
= &tun_link_ops
;
2324 dev
->ifindex
= tfile
->ifindex
;
2325 dev
->sysfs_groups
[0] = &tun_attr_group
;
2327 tun
= netdev_priv(dev
);
2330 tun
->txflt
.count
= 0;
2331 tun
->vnet_hdr_sz
= sizeof(struct virtio_net_hdr
);
2333 tun
->align
= NET_SKB_PAD
;
2334 tun
->filter_attached
= false;
2335 tun
->sndbuf
= tfile
->socket
.sk
->sk_sndbuf
;
2336 tun
->rx_batched
= 0;
2338 tun
->pcpu_stats
= netdev_alloc_pcpu_stats(struct tun_pcpu_stats
);
2339 if (!tun
->pcpu_stats
) {
2344 spin_lock_init(&tun
->lock
);
2346 err
= security_tun_dev_alloc_security(&tun
->security
);
2353 dev
->hw_features
= NETIF_F_SG
| NETIF_F_FRAGLIST
|
2354 TUN_USER_FEATURES
| NETIF_F_HW_VLAN_CTAG_TX
|
2355 NETIF_F_HW_VLAN_STAG_TX
;
2356 dev
->features
= dev
->hw_features
| NETIF_F_LLTX
;
2357 dev
->vlan_features
= dev
->features
&
2358 ~(NETIF_F_HW_VLAN_CTAG_TX
|
2359 NETIF_F_HW_VLAN_STAG_TX
);
2361 INIT_LIST_HEAD(&tun
->disabled
);
2362 err
= tun_attach(tun
, file
, false, ifr
->ifr_flags
& IFF_NAPI
,
2363 ifr
->ifr_flags
& IFF_NAPI_FRAGS
, false);
2367 err
= register_netdevice(tun
->dev
);
2370 /* free_netdev() won't check refcnt, to aovid race
2371 * with dev_put() we need publish tun after registration.
2373 rcu_assign_pointer(tfile
->tun
, tun
);
2376 netif_carrier_on(tun
->dev
);
2378 tun_debug(KERN_INFO
, tun
, "tun_set_iff\n");
2380 tun
->flags
= (tun
->flags
& ~TUN_FEATURES
) |
2381 (ifr
->ifr_flags
& TUN_FEATURES
);
2383 /* Make sure persistent devices do not get stuck in
2386 if (netif_running(tun
->dev
))
2387 netif_tx_wake_all_queues(tun
->dev
);
2389 strcpy(ifr
->ifr_name
, tun
->dev
->name
);
2393 tun_detach_all(dev
);
2394 /* register_netdevice() already called tun_free_netdev() */
2398 tun_flow_uninit(tun
);
2399 security_tun_dev_free_security(tun
->security
);
2401 free_percpu(tun
->pcpu_stats
);
2407 static void tun_get_iff(struct net
*net
, struct tun_struct
*tun
,
2410 tun_debug(KERN_INFO
, tun
, "tun_get_iff\n");
2412 strcpy(ifr
->ifr_name
, tun
->dev
->name
);
2414 ifr
->ifr_flags
= tun_flags(tun
);
2418 /* This is like a cut-down ethtool ops, except done via tun fd so no
2419 * privs required. */
2420 static int set_offload(struct tun_struct
*tun
, unsigned long arg
)
2422 netdev_features_t features
= 0;
2424 if (arg
& TUN_F_CSUM
) {
2425 features
|= NETIF_F_HW_CSUM
;
2428 if (arg
& (TUN_F_TSO4
|TUN_F_TSO6
)) {
2429 if (arg
& TUN_F_TSO_ECN
) {
2430 features
|= NETIF_F_TSO_ECN
;
2431 arg
&= ~TUN_F_TSO_ECN
;
2433 if (arg
& TUN_F_TSO4
)
2434 features
|= NETIF_F_TSO
;
2435 if (arg
& TUN_F_TSO6
)
2436 features
|= NETIF_F_TSO6
;
2437 arg
&= ~(TUN_F_TSO4
|TUN_F_TSO6
);
2443 /* This gives the user a way to test for new features in future by
2444 * trying to set them. */
2448 tun
->set_features
= features
;
2449 tun
->dev
->wanted_features
&= ~TUN_USER_FEATURES
;
2450 tun
->dev
->wanted_features
|= features
;
2451 netdev_update_features(tun
->dev
);
2456 static void tun_detach_filter(struct tun_struct
*tun
, int n
)
2459 struct tun_file
*tfile
;
2461 for (i
= 0; i
< n
; i
++) {
2462 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
2463 lock_sock(tfile
->socket
.sk
);
2464 sk_detach_filter(tfile
->socket
.sk
);
2465 release_sock(tfile
->socket
.sk
);
2468 tun
->filter_attached
= false;
2471 static int tun_attach_filter(struct tun_struct
*tun
)
2474 struct tun_file
*tfile
;
2476 for (i
= 0; i
< tun
->numqueues
; i
++) {
2477 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
2478 lock_sock(tfile
->socket
.sk
);
2479 ret
= sk_attach_filter(&tun
->fprog
, tfile
->socket
.sk
);
2480 release_sock(tfile
->socket
.sk
);
2482 tun_detach_filter(tun
, i
);
2487 tun
->filter_attached
= true;
2491 static void tun_set_sndbuf(struct tun_struct
*tun
)
2493 struct tun_file
*tfile
;
2496 for (i
= 0; i
< tun
->numqueues
; i
++) {
2497 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
2498 tfile
->socket
.sk
->sk_sndbuf
= tun
->sndbuf
;
2502 static int tun_set_queue(struct file
*file
, struct ifreq
*ifr
)
2504 struct tun_file
*tfile
= file
->private_data
;
2505 struct tun_struct
*tun
;
2510 if (ifr
->ifr_flags
& IFF_ATTACH_QUEUE
) {
2511 tun
= tfile
->detached
;
2516 ret
= security_tun_dev_attach_queue(tun
->security
);
2519 ret
= tun_attach(tun
, file
, false, tun
->flags
& IFF_NAPI
,
2520 tun
->flags
& IFF_NAPI_FRAGS
, true);
2521 } else if (ifr
->ifr_flags
& IFF_DETACH_QUEUE
) {
2522 tun
= rtnl_dereference(tfile
->tun
);
2523 if (!tun
|| !(tun
->flags
& IFF_MULTI_QUEUE
) || tfile
->detached
)
2526 __tun_detach(tfile
, false);
2535 static long __tun_chr_ioctl(struct file
*file
, unsigned int cmd
,
2536 unsigned long arg
, int ifreq_len
)
2538 struct tun_file
*tfile
= file
->private_data
;
2539 struct tun_struct
*tun
;
2540 void __user
* argp
= (void __user
*)arg
;
2541 unsigned int ifindex
, carrier
;
2550 if (cmd
== TUNSETIFF
|| cmd
== TUNSETQUEUE
|| _IOC_TYPE(cmd
) == SOCK_IOC_TYPE
) {
2551 if (copy_from_user(&ifr
, argp
, ifreq_len
))
2554 memset(&ifr
, 0, sizeof(ifr
));
2556 if (cmd
== TUNGETFEATURES
) {
2557 /* Currently this just means: "what IFF flags are valid?".
2558 * This is needed because we never checked for invalid flags on
2561 return put_user(IFF_TUN
| IFF_TAP
| TUN_FEATURES
,
2562 (unsigned int __user
*)argp
);
2563 } else if (cmd
== TUNSETQUEUE
)
2564 return tun_set_queue(file
, &ifr
);
2569 tun
= tun_get(tfile
);
2570 if (cmd
== TUNSETIFF
) {
2575 ifr
.ifr_name
[IFNAMSIZ
-1] = '\0';
2577 ret
= tun_set_iff(sock_net(&tfile
->sk
), file
, &ifr
);
2582 if (copy_to_user(argp
, &ifr
, ifreq_len
))
2586 if (cmd
== TUNSETIFINDEX
) {
2592 if (copy_from_user(&ifindex
, argp
, sizeof(ifindex
)))
2596 tfile
->ifindex
= ifindex
;
2604 tun_debug(KERN_INFO
, tun
, "tun_chr_ioctl cmd %u\n", cmd
);
2609 tun_get_iff(current
->nsproxy
->net_ns
, tun
, &ifr
);
2611 if (tfile
->detached
)
2612 ifr
.ifr_flags
|= IFF_DETACH_QUEUE
;
2613 if (!tfile
->socket
.sk
->sk_filter
)
2614 ifr
.ifr_flags
|= IFF_NOFILTER
;
2616 if (copy_to_user(argp
, &ifr
, ifreq_len
))
2621 /* Disable/Enable checksum */
2623 /* [unimplemented] */
2624 tun_debug(KERN_INFO
, tun
, "ignored: set checksum %s\n",
2625 arg
? "disabled" : "enabled");
2629 /* Disable/Enable persist mode. Keep an extra reference to the
2630 * module to prevent the module being unprobed.
2632 if (arg
&& !(tun
->flags
& IFF_PERSIST
)) {
2633 tun
->flags
|= IFF_PERSIST
;
2634 __module_get(THIS_MODULE
);
2636 if (!arg
&& (tun
->flags
& IFF_PERSIST
)) {
2637 tun
->flags
&= ~IFF_PERSIST
;
2638 module_put(THIS_MODULE
);
2641 tun_debug(KERN_INFO
, tun
, "persist %s\n",
2642 arg
? "enabled" : "disabled");
2646 /* Set owner of the device */
2647 owner
= make_kuid(current_user_ns(), arg
);
2648 if (!uid_valid(owner
)) {
2653 tun_debug(KERN_INFO
, tun
, "owner set to %u\n",
2654 from_kuid(&init_user_ns
, tun
->owner
));
2658 /* Set group of the device */
2659 group
= make_kgid(current_user_ns(), arg
);
2660 if (!gid_valid(group
)) {
2665 tun_debug(KERN_INFO
, tun
, "group set to %u\n",
2666 from_kgid(&init_user_ns
, tun
->group
));
2670 /* Only allow setting the type when the interface is down */
2671 if (tun
->dev
->flags
& IFF_UP
) {
2672 tun_debug(KERN_INFO
, tun
,
2673 "Linktype set failed because interface is up\n");
2676 tun
->dev
->type
= (int) arg
;
2677 tun_debug(KERN_INFO
, tun
, "linktype set to %d\n",
2689 ret
= set_offload(tun
, arg
);
2692 case TUNSETTXFILTER
:
2693 /* Can be set only for TAPs */
2695 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2697 ret
= update_filter(&tun
->txflt
, (void __user
*)arg
);
2701 /* Get hw address */
2702 memcpy(ifr
.ifr_hwaddr
.sa_data
, tun
->dev
->dev_addr
, ETH_ALEN
);
2703 ifr
.ifr_hwaddr
.sa_family
= tun
->dev
->type
;
2704 if (copy_to_user(argp
, &ifr
, ifreq_len
))
2709 /* Set hw address */
2710 tun_debug(KERN_DEBUG
, tun
, "set hw address: %pM\n",
2711 ifr
.ifr_hwaddr
.sa_data
);
2713 ret
= dev_set_mac_address(tun
->dev
, &ifr
.ifr_hwaddr
);
2717 sndbuf
= tfile
->socket
.sk
->sk_sndbuf
;
2718 if (copy_to_user(argp
, &sndbuf
, sizeof(sndbuf
)))
2723 if (copy_from_user(&sndbuf
, argp
, sizeof(sndbuf
))) {
2732 tun
->sndbuf
= sndbuf
;
2733 tun_set_sndbuf(tun
);
2736 case TUNGETVNETHDRSZ
:
2737 vnet_hdr_sz
= tun
->vnet_hdr_sz
;
2738 if (copy_to_user(argp
, &vnet_hdr_sz
, sizeof(vnet_hdr_sz
)))
2742 case TUNSETVNETHDRSZ
:
2743 if (copy_from_user(&vnet_hdr_sz
, argp
, sizeof(vnet_hdr_sz
))) {
2747 if (vnet_hdr_sz
< (int)sizeof(struct virtio_net_hdr
)) {
2752 tun
->vnet_hdr_sz
= vnet_hdr_sz
;
2756 le
= !!(tun
->flags
& TUN_VNET_LE
);
2757 if (put_user(le
, (int __user
*)argp
))
2762 if (get_user(le
, (int __user
*)argp
)) {
2767 tun
->flags
|= TUN_VNET_LE
;
2769 tun
->flags
&= ~TUN_VNET_LE
;
2773 ret
= tun_get_vnet_be(tun
, argp
);
2777 ret
= tun_set_vnet_be(tun
, argp
);
2780 case TUNATTACHFILTER
:
2781 /* Can be set only for TAPs */
2783 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2786 if (copy_from_user(&tun
->fprog
, argp
, sizeof(tun
->fprog
)))
2789 ret
= tun_attach_filter(tun
);
2792 case TUNDETACHFILTER
:
2793 /* Can be set only for TAPs */
2795 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2798 tun_detach_filter(tun
, tun
->numqueues
);
2803 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2806 if (copy_to_user(argp
, &tun
->fprog
, sizeof(tun
->fprog
)))
2813 if (copy_from_user(&carrier
, argp
, sizeof(carrier
)))
2816 ret
= tun_net_change_carrier(tun
->dev
, (bool)carrier
);
2831 static long tun_chr_ioctl(struct file
*file
,
2832 unsigned int cmd
, unsigned long arg
)
2834 return __tun_chr_ioctl(file
, cmd
, arg
, sizeof (struct ifreq
));
2837 #ifdef CONFIG_COMPAT
2838 static long tun_chr_compat_ioctl(struct file
*file
,
2839 unsigned int cmd
, unsigned long arg
)
2844 case TUNSETTXFILTER
:
2849 arg
= (unsigned long)compat_ptr(arg
);
2852 arg
= (compat_ulong_t
)arg
;
2857 * compat_ifreq is shorter than ifreq, so we must not access beyond
2858 * the end of that structure. All fields that are used in this
2859 * driver are compatible though, we don't need to convert the
2862 return __tun_chr_ioctl(file
, cmd
, arg
, sizeof(struct compat_ifreq
));
2864 #endif /* CONFIG_COMPAT */
2866 static int tun_chr_fasync(int fd
, struct file
*file
, int on
)
2868 struct tun_file
*tfile
= file
->private_data
;
2871 if ((ret
= fasync_helper(fd
, file
, on
, &tfile
->fasync
)) < 0)
2875 __f_setown(file
, task_pid(current
), PIDTYPE_PID
, 0);
2876 tfile
->flags
|= TUN_FASYNC
;
2878 tfile
->flags
&= ~TUN_FASYNC
;
2884 static int tun_chr_open(struct inode
*inode
, struct file
* file
)
2886 struct net
*net
= current
->nsproxy
->net_ns
;
2887 struct tun_file
*tfile
;
2889 DBG1(KERN_INFO
, "tunX: tun_chr_open\n");
2891 tfile
= (struct tun_file
*)sk_alloc(net
, AF_UNSPEC
, GFP_KERNEL
,
2895 if (skb_array_init(&tfile
->tx_array
, 0, GFP_KERNEL
)) {
2896 sk_free(&tfile
->sk
);
2900 mutex_init(&tfile
->napi_mutex
);
2901 RCU_INIT_POINTER(tfile
->tun
, NULL
);
2905 init_waitqueue_head(&tfile
->wq
.wait
);
2906 RCU_INIT_POINTER(tfile
->socket
.wq
, &tfile
->wq
);
2908 tfile
->socket
.file
= file
;
2909 tfile
->socket
.ops
= &tun_socket_ops
;
2911 sock_init_data(&tfile
->socket
, &tfile
->sk
);
2913 tfile
->sk
.sk_write_space
= tun_sock_write_space
;
2914 tfile
->sk
.sk_sndbuf
= INT_MAX
;
2916 file
->private_data
= tfile
;
2917 INIT_LIST_HEAD(&tfile
->next
);
2919 sock_set_flag(&tfile
->sk
, SOCK_ZEROCOPY
);
2924 static int tun_chr_close(struct inode
*inode
, struct file
*file
)
2926 struct tun_file
*tfile
= file
->private_data
;
2928 tun_detach(tfile
, true);
2933 #ifdef CONFIG_PROC_FS
2934 static void tun_chr_show_fdinfo(struct seq_file
*m
, struct file
*file
)
2936 struct tun_file
*tfile
= file
->private_data
;
2937 struct tun_struct
*tun
;
2940 memset(&ifr
, 0, sizeof(ifr
));
2943 tun
= tun_get(tfile
);
2945 tun_get_iff(current
->nsproxy
->net_ns
, tun
, &ifr
);
2951 seq_printf(m
, "iff:\t%s\n", ifr
.ifr_name
);
2955 static const struct file_operations tun_fops
= {
2956 .owner
= THIS_MODULE
,
2957 .llseek
= no_llseek
,
2958 .read_iter
= tun_chr_read_iter
,
2959 .write_iter
= tun_chr_write_iter
,
2960 .poll
= tun_chr_poll
,
2961 .unlocked_ioctl
= tun_chr_ioctl
,
2962 #ifdef CONFIG_COMPAT
2963 .compat_ioctl
= tun_chr_compat_ioctl
,
2965 .open
= tun_chr_open
,
2966 .release
= tun_chr_close
,
2967 .fasync
= tun_chr_fasync
,
2968 #ifdef CONFIG_PROC_FS
2969 .show_fdinfo
= tun_chr_show_fdinfo
,
2973 static struct miscdevice tun_miscdev
= {
2976 .nodename
= "net/tun",
2980 /* ethtool interface */
2982 static int tun_get_link_ksettings(struct net_device
*dev
,
2983 struct ethtool_link_ksettings
*cmd
)
2985 ethtool_link_ksettings_zero_link_mode(cmd
, supported
);
2986 ethtool_link_ksettings_zero_link_mode(cmd
, advertising
);
2987 cmd
->base
.speed
= SPEED_10
;
2988 cmd
->base
.duplex
= DUPLEX_FULL
;
2989 cmd
->base
.port
= PORT_TP
;
2990 cmd
->base
.phy_address
= 0;
2991 cmd
->base
.autoneg
= AUTONEG_DISABLE
;
2995 static void tun_get_drvinfo(struct net_device
*dev
, struct ethtool_drvinfo
*info
)
2997 struct tun_struct
*tun
= netdev_priv(dev
);
2999 strlcpy(info
->driver
, DRV_NAME
, sizeof(info
->driver
));
3000 strlcpy(info
->version
, DRV_VERSION
, sizeof(info
->version
));
3002 switch (tun
->flags
& TUN_TYPE_MASK
) {
3004 strlcpy(info
->bus_info
, "tun", sizeof(info
->bus_info
));
3007 strlcpy(info
->bus_info
, "tap", sizeof(info
->bus_info
));
3012 static u32
tun_get_msglevel(struct net_device
*dev
)
3015 struct tun_struct
*tun
= netdev_priv(dev
);
3022 static void tun_set_msglevel(struct net_device
*dev
, u32 value
)
3025 struct tun_struct
*tun
= netdev_priv(dev
);
3030 static int tun_get_coalesce(struct net_device
*dev
,
3031 struct ethtool_coalesce
*ec
)
3033 struct tun_struct
*tun
= netdev_priv(dev
);
3035 ec
->rx_max_coalesced_frames
= tun
->rx_batched
;
3040 static int tun_set_coalesce(struct net_device
*dev
,
3041 struct ethtool_coalesce
*ec
)
3043 struct tun_struct
*tun
= netdev_priv(dev
);
3045 if (ec
->rx_max_coalesced_frames
> NAPI_POLL_WEIGHT
)
3046 tun
->rx_batched
= NAPI_POLL_WEIGHT
;
3048 tun
->rx_batched
= ec
->rx_max_coalesced_frames
;
3053 static const struct ethtool_ops tun_ethtool_ops
= {
3054 .get_drvinfo
= tun_get_drvinfo
,
3055 .get_msglevel
= tun_get_msglevel
,
3056 .set_msglevel
= tun_set_msglevel
,
3057 .get_link
= ethtool_op_get_link
,
3058 .get_ts_info
= ethtool_op_get_ts_info
,
3059 .get_coalesce
= tun_get_coalesce
,
3060 .set_coalesce
= tun_set_coalesce
,
3061 .get_link_ksettings
= tun_get_link_ksettings
,
3064 static int tun_queue_resize(struct tun_struct
*tun
)
3066 struct net_device
*dev
= tun
->dev
;
3067 struct tun_file
*tfile
;
3068 struct skb_array
**arrays
;
3069 int n
= tun
->numqueues
+ tun
->numdisabled
;
3072 arrays
= kmalloc_array(n
, sizeof(*arrays
), GFP_KERNEL
);
3076 for (i
= 0; i
< tun
->numqueues
; i
++) {
3077 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
3078 arrays
[i
] = &tfile
->tx_array
;
3080 list_for_each_entry(tfile
, &tun
->disabled
, next
)
3081 arrays
[i
++] = &tfile
->tx_array
;
3083 ret
= skb_array_resize_multiple(arrays
, n
,
3084 dev
->tx_queue_len
, GFP_KERNEL
);
3090 static int tun_device_event(struct notifier_block
*unused
,
3091 unsigned long event
, void *ptr
)
3093 struct net_device
*dev
= netdev_notifier_info_to_dev(ptr
);
3094 struct tun_struct
*tun
= netdev_priv(dev
);
3097 if (dev
->rtnl_link_ops
!= &tun_link_ops
)
3101 case NETDEV_CHANGE_TX_QUEUE_LEN
:
3102 if (tun_queue_resize(tun
))
3106 for (i
= 0; i
< tun
->numqueues
; i
++) {
3107 struct tun_file
*tfile
;
3109 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
3110 tfile
->socket
.sk
->sk_write_space(tfile
->socket
.sk
);
3120 static struct notifier_block tun_notifier_block __read_mostly
= {
3121 .notifier_call
= tun_device_event
,
3124 static int __init
tun_init(void)
3128 pr_info("%s, %s\n", DRV_DESCRIPTION
, DRV_VERSION
);
3130 ret
= rtnl_link_register(&tun_link_ops
);
3132 pr_err("Can't register link_ops\n");
3136 ret
= misc_register(&tun_miscdev
);
3138 pr_err("Can't register misc device %d\n", TUN_MINOR
);
3142 ret
= register_netdevice_notifier(&tun_notifier_block
);
3144 pr_err("Can't register netdevice notifier\n");
3151 misc_deregister(&tun_miscdev
);
3153 rtnl_link_unregister(&tun_link_ops
);
3158 static void tun_cleanup(void)
3160 misc_deregister(&tun_miscdev
);
3161 rtnl_link_unregister(&tun_link_ops
);
3162 unregister_netdevice_notifier(&tun_notifier_block
);
3165 /* Get an underlying socket object from tun file. Returns error unless file is
3166 * attached to a device. The returned object works like a packet socket, it
3167 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
3168 * holding a reference to the file for as long as the socket is in use. */
3169 struct socket
*tun_get_socket(struct file
*file
)
3171 struct tun_file
*tfile
;
3172 if (file
->f_op
!= &tun_fops
)
3173 return ERR_PTR(-EINVAL
);
3174 tfile
= file
->private_data
;
3176 return ERR_PTR(-EBADFD
);
3177 return &tfile
->socket
;
3179 EXPORT_SYMBOL_GPL(tun_get_socket
);
3181 struct skb_array
*tun_get_skb_array(struct file
*file
)
3183 struct tun_file
*tfile
;
3185 if (file
->f_op
!= &tun_fops
)
3186 return ERR_PTR(-EINVAL
);
3187 tfile
= file
->private_data
;
3189 return ERR_PTR(-EBADFD
);
3190 return &tfile
->tx_array
;
3192 EXPORT_SYMBOL_GPL(tun_get_skb_array
);
3194 module_init(tun_init
);
3195 module_exit(tun_cleanup
);
3196 MODULE_DESCRIPTION(DRV_DESCRIPTION
);
3197 MODULE_AUTHOR(DRV_COPYRIGHT
);
3198 MODULE_LICENSE("GPL");
3199 MODULE_ALIAS_MISCDEV(TUN_MINOR
);
3200 MODULE_ALIAS("devname:net/tun");