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 struct tun_struct
*tun
= netdev_priv(dev
);
922 netif_tx_start_all_queues(dev
);
924 for (i
= 0; i
< tun
->numqueues
; i
++) {
925 struct tun_file
*tfile
;
927 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
928 tfile
->socket
.sk
->sk_write_space(tfile
->socket
.sk
);
934 /* Net device close. */
935 static int tun_net_close(struct net_device
*dev
)
937 netif_tx_stop_all_queues(dev
);
941 /* Net device start xmit */
942 static netdev_tx_t
tun_net_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
944 struct tun_struct
*tun
= netdev_priv(dev
);
945 int txq
= skb
->queue_mapping
;
946 struct tun_file
*tfile
;
950 tfile
= rcu_dereference(tun
->tfiles
[txq
]);
951 numqueues
= READ_ONCE(tun
->numqueues
);
953 /* Drop packet if interface is not attached */
954 if (txq
>= numqueues
)
958 if (numqueues
== 1 && static_key_false(&rps_needed
)) {
959 /* Select queue was not called for the skbuff, so we extract the
960 * RPS hash and save it into the flow_table here.
964 rxhash
= __skb_get_hash_symmetric(skb
);
966 struct tun_flow_entry
*e
;
967 e
= tun_flow_find(&tun
->flows
[tun_hashfn(rxhash
)],
970 tun_flow_save_rps_rxhash(e
, rxhash
);
975 tun_debug(KERN_INFO
, tun
, "tun_net_xmit %d\n", skb
->len
);
979 /* Drop if the filter does not like it.
980 * This is a noop if the filter is disabled.
981 * Filter can be enabled only for the TAP devices. */
982 if (!check_filter(&tun
->txflt
, skb
))
985 if (tfile
->socket
.sk
->sk_filter
&&
986 sk_filter(tfile
->socket
.sk
, skb
))
989 if (unlikely(skb_orphan_frags_rx(skb
, GFP_ATOMIC
)))
992 skb_tx_timestamp(skb
);
994 /* Orphan the skb - required as we might hang on to it
995 * for indefinite time.
1001 if (skb_array_produce(&tfile
->tx_array
, skb
))
1004 /* Notify and wake up reader process */
1005 if (tfile
->flags
& TUN_FASYNC
)
1006 kill_fasync(&tfile
->fasync
, SIGIO
, POLL_IN
);
1007 tfile
->socket
.sk
->sk_data_ready(tfile
->socket
.sk
);
1010 return NETDEV_TX_OK
;
1013 this_cpu_inc(tun
->pcpu_stats
->tx_dropped
);
1017 return NET_XMIT_DROP
;
1020 static void tun_net_mclist(struct net_device
*dev
)
1023 * This callback is supposed to deal with mc filter in
1024 * _rx_ path and has nothing to do with the _tx_ path.
1025 * In rx path we always accept everything userspace gives us.
1029 static netdev_features_t
tun_net_fix_features(struct net_device
*dev
,
1030 netdev_features_t features
)
1032 struct tun_struct
*tun
= netdev_priv(dev
);
1034 return (features
& tun
->set_features
) | (features
& ~TUN_USER_FEATURES
);
1036 #ifdef CONFIG_NET_POLL_CONTROLLER
1037 static void tun_poll_controller(struct net_device
*dev
)
1040 * Tun only receives frames when:
1041 * 1) the char device endpoint gets data from user space
1042 * 2) the tun socket gets a sendmsg call from user space
1043 * If NAPI is not enabled, since both of those are synchronous
1044 * operations, we are guaranteed never to have pending data when we poll
1045 * for it so there is nothing to do here but return.
1046 * We need this though so netpoll recognizes us as an interface that
1047 * supports polling, which enables bridge devices in virt setups to
1048 * still use netconsole
1049 * If NAPI is enabled, however, we need to schedule polling for all
1050 * queues unless we are using napi_gro_frags(), which we call in
1051 * process context and not in NAPI context.
1053 struct tun_struct
*tun
= netdev_priv(dev
);
1055 if (tun
->flags
& IFF_NAPI
) {
1056 struct tun_file
*tfile
;
1060 for (i
= 0; i
< tun
->numqueues
; i
++) {
1061 tfile
= rcu_dereference(tun
->tfiles
[i
]);
1062 if (!tun_napi_frags_enabled(tfile
) &&
1063 tfile
->napi_enabled
)
1064 napi_schedule(&tfile
->napi
);
1072 static void tun_set_headroom(struct net_device
*dev
, int new_hr
)
1074 struct tun_struct
*tun
= netdev_priv(dev
);
1076 if (new_hr
< NET_SKB_PAD
)
1077 new_hr
= NET_SKB_PAD
;
1079 tun
->align
= new_hr
;
1083 tun_net_get_stats64(struct net_device
*dev
, struct rtnl_link_stats64
*stats
)
1085 u32 rx_dropped
= 0, tx_dropped
= 0, rx_frame_errors
= 0;
1086 struct tun_struct
*tun
= netdev_priv(dev
);
1087 struct tun_pcpu_stats
*p
;
1090 for_each_possible_cpu(i
) {
1091 u64 rxpackets
, rxbytes
, txpackets
, txbytes
;
1094 p
= per_cpu_ptr(tun
->pcpu_stats
, i
);
1096 start
= u64_stats_fetch_begin(&p
->syncp
);
1097 rxpackets
= p
->rx_packets
;
1098 rxbytes
= p
->rx_bytes
;
1099 txpackets
= p
->tx_packets
;
1100 txbytes
= p
->tx_bytes
;
1101 } while (u64_stats_fetch_retry(&p
->syncp
, start
));
1103 stats
->rx_packets
+= rxpackets
;
1104 stats
->rx_bytes
+= rxbytes
;
1105 stats
->tx_packets
+= txpackets
;
1106 stats
->tx_bytes
+= txbytes
;
1109 rx_dropped
+= p
->rx_dropped
;
1110 rx_frame_errors
+= p
->rx_frame_errors
;
1111 tx_dropped
+= p
->tx_dropped
;
1113 stats
->rx_dropped
= rx_dropped
;
1114 stats
->rx_frame_errors
= rx_frame_errors
;
1115 stats
->tx_dropped
= tx_dropped
;
1118 static int tun_xdp_set(struct net_device
*dev
, struct bpf_prog
*prog
,
1119 struct netlink_ext_ack
*extack
)
1121 struct tun_struct
*tun
= netdev_priv(dev
);
1122 struct bpf_prog
*old_prog
;
1124 old_prog
= rtnl_dereference(tun
->xdp_prog
);
1125 rcu_assign_pointer(tun
->xdp_prog
, prog
);
1127 bpf_prog_put(old_prog
);
1132 static u32
tun_xdp_query(struct net_device
*dev
)
1134 struct tun_struct
*tun
= netdev_priv(dev
);
1135 const struct bpf_prog
*xdp_prog
;
1137 xdp_prog
= rtnl_dereference(tun
->xdp_prog
);
1139 return xdp_prog
->aux
->id
;
1144 static int tun_xdp(struct net_device
*dev
, struct netdev_bpf
*xdp
)
1146 switch (xdp
->command
) {
1147 case XDP_SETUP_PROG
:
1148 return tun_xdp_set(dev
, xdp
->prog
, xdp
->extack
);
1149 case XDP_QUERY_PROG
:
1150 xdp
->prog_id
= tun_xdp_query(dev
);
1151 xdp
->prog_attached
= !!xdp
->prog_id
;
1158 static int tun_net_change_carrier(struct net_device
*dev
, bool new_carrier
)
1161 struct tun_struct
*tun
= netdev_priv(dev
);
1163 if (!tun
->numqueues
)
1166 netif_carrier_on(dev
);
1168 netif_carrier_off(dev
);
1173 static const struct net_device_ops tun_netdev_ops
= {
1174 .ndo_uninit
= tun_net_uninit
,
1175 .ndo_open
= tun_net_open
,
1176 .ndo_stop
= tun_net_close
,
1177 .ndo_start_xmit
= tun_net_xmit
,
1178 .ndo_fix_features
= tun_net_fix_features
,
1179 .ndo_select_queue
= tun_select_queue
,
1180 #ifdef CONFIG_NET_POLL_CONTROLLER
1181 .ndo_poll_controller
= tun_poll_controller
,
1183 .ndo_set_rx_headroom
= tun_set_headroom
,
1184 .ndo_get_stats64
= tun_net_get_stats64
,
1185 .ndo_change_carrier
= tun_net_change_carrier
,
1188 static const struct net_device_ops tap_netdev_ops
= {
1189 .ndo_uninit
= tun_net_uninit
,
1190 .ndo_open
= tun_net_open
,
1191 .ndo_stop
= tun_net_close
,
1192 .ndo_start_xmit
= tun_net_xmit
,
1193 .ndo_fix_features
= tun_net_fix_features
,
1194 .ndo_set_rx_mode
= tun_net_mclist
,
1195 .ndo_set_mac_address
= eth_mac_addr
,
1196 .ndo_validate_addr
= eth_validate_addr
,
1197 .ndo_select_queue
= tun_select_queue
,
1198 #ifdef CONFIG_NET_POLL_CONTROLLER
1199 .ndo_poll_controller
= tun_poll_controller
,
1201 .ndo_features_check
= passthru_features_check
,
1202 .ndo_set_rx_headroom
= tun_set_headroom
,
1203 .ndo_get_stats64
= tun_net_get_stats64
,
1205 .ndo_change_carrier
= tun_net_change_carrier
,
1208 static void tun_flow_init(struct tun_struct
*tun
)
1212 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++)
1213 INIT_HLIST_HEAD(&tun
->flows
[i
]);
1215 tun
->ageing_time
= TUN_FLOW_EXPIRE
;
1216 timer_setup(&tun
->flow_gc_timer
, tun_flow_cleanup
, 0);
1217 mod_timer(&tun
->flow_gc_timer
,
1218 round_jiffies_up(jiffies
+ tun
->ageing_time
));
1221 static void tun_flow_uninit(struct tun_struct
*tun
)
1223 del_timer_sync(&tun
->flow_gc_timer
);
1224 tun_flow_flush(tun
);
1228 #define MAX_MTU 65535
1230 /* Initialize net device. */
1231 static void tun_net_init(struct net_device
*dev
)
1233 struct tun_struct
*tun
= netdev_priv(dev
);
1235 switch (tun
->flags
& TUN_TYPE_MASK
) {
1237 dev
->netdev_ops
= &tun_netdev_ops
;
1239 /* Point-to-Point TUN Device */
1240 dev
->hard_header_len
= 0;
1244 /* Zero header length */
1245 dev
->type
= ARPHRD_NONE
;
1246 dev
->flags
= IFF_POINTOPOINT
| IFF_NOARP
| IFF_MULTICAST
;
1250 dev
->netdev_ops
= &tap_netdev_ops
;
1251 /* Ethernet TAP Device */
1253 dev
->priv_flags
&= ~IFF_TX_SKB_SHARING
;
1254 dev
->priv_flags
|= IFF_LIVE_ADDR_CHANGE
;
1256 eth_hw_addr_random(dev
);
1261 dev
->min_mtu
= MIN_MTU
;
1262 dev
->max_mtu
= MAX_MTU
- dev
->hard_header_len
;
1265 /* Character device part */
1268 static unsigned int tun_chr_poll(struct file
*file
, poll_table
*wait
)
1270 struct tun_file
*tfile
= file
->private_data
;
1271 struct tun_struct
*tun
= tun_get(tfile
);
1273 unsigned int mask
= 0;
1278 sk
= tfile
->socket
.sk
;
1280 tun_debug(KERN_INFO
, tun
, "tun_chr_poll\n");
1282 poll_wait(file
, sk_sleep(sk
), wait
);
1284 if (!skb_array_empty(&tfile
->tx_array
))
1285 mask
|= POLLIN
| POLLRDNORM
;
1287 if (tun
->dev
->flags
& IFF_UP
&&
1288 (sock_writeable(sk
) ||
1289 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
) &&
1290 sock_writeable(sk
))))
1291 mask
|= POLLOUT
| POLLWRNORM
;
1293 if (tun
->dev
->reg_state
!= NETREG_REGISTERED
)
1300 static struct sk_buff
*tun_napi_alloc_frags(struct tun_file
*tfile
,
1302 const struct iov_iter
*it
)
1304 struct sk_buff
*skb
;
1309 if (it
->nr_segs
> MAX_SKB_FRAGS
+ 1)
1310 return ERR_PTR(-ENOMEM
);
1313 skb
= napi_get_frags(&tfile
->napi
);
1316 return ERR_PTR(-ENOMEM
);
1318 linear
= iov_iter_single_seg_count(it
);
1319 err
= __skb_grow(skb
, linear
);
1324 skb
->data_len
= len
- linear
;
1325 skb
->truesize
+= skb
->data_len
;
1327 for (i
= 1; i
< it
->nr_segs
; i
++) {
1328 struct page_frag
*pfrag
= ¤t
->task_frag
;
1329 size_t fragsz
= it
->iov
[i
].iov_len
;
1331 if (fragsz
== 0 || fragsz
> PAGE_SIZE
) {
1336 if (!skb_page_frag_refill(fragsz
, pfrag
, GFP_KERNEL
)) {
1341 skb_fill_page_desc(skb
, i
- 1, pfrag
->page
,
1342 pfrag
->offset
, fragsz
);
1343 page_ref_inc(pfrag
->page
);
1344 pfrag
->offset
+= fragsz
;
1349 /* frees skb and all frags allocated with napi_alloc_frag() */
1350 napi_free_frags(&tfile
->napi
);
1351 return ERR_PTR(err
);
1354 /* prepad is the amount to reserve at front. len is length after that.
1355 * linear is a hint as to how much to copy (usually headers). */
1356 static struct sk_buff
*tun_alloc_skb(struct tun_file
*tfile
,
1357 size_t prepad
, size_t len
,
1358 size_t linear
, int noblock
)
1360 struct sock
*sk
= tfile
->socket
.sk
;
1361 struct sk_buff
*skb
;
1364 /* Under a page? Don't bother with paged skb. */
1365 if (prepad
+ len
< PAGE_SIZE
|| !linear
)
1368 skb
= sock_alloc_send_pskb(sk
, prepad
+ linear
, len
- linear
, noblock
,
1371 return ERR_PTR(err
);
1373 skb_reserve(skb
, prepad
);
1374 skb_put(skb
, linear
);
1375 skb
->data_len
= len
- linear
;
1376 skb
->len
+= len
- linear
;
1381 static void tun_rx_batched(struct tun_struct
*tun
, struct tun_file
*tfile
,
1382 struct sk_buff
*skb
, int more
)
1384 struct sk_buff_head
*queue
= &tfile
->sk
.sk_write_queue
;
1385 struct sk_buff_head process_queue
;
1386 u32 rx_batched
= tun
->rx_batched
;
1389 if (!rx_batched
|| (!more
&& skb_queue_empty(queue
))) {
1391 skb_record_rx_queue(skb
, tfile
->queue_index
);
1392 netif_receive_skb(skb
);
1397 spin_lock(&queue
->lock
);
1398 if (!more
|| skb_queue_len(queue
) == rx_batched
) {
1399 __skb_queue_head_init(&process_queue
);
1400 skb_queue_splice_tail_init(queue
, &process_queue
);
1403 __skb_queue_tail(queue
, skb
);
1405 spin_unlock(&queue
->lock
);
1408 struct sk_buff
*nskb
;
1411 while ((nskb
= __skb_dequeue(&process_queue
))) {
1412 skb_record_rx_queue(nskb
, tfile
->queue_index
);
1413 netif_receive_skb(nskb
);
1415 skb_record_rx_queue(skb
, tfile
->queue_index
);
1416 netif_receive_skb(skb
);
1421 static bool tun_can_build_skb(struct tun_struct
*tun
, struct tun_file
*tfile
,
1422 int len
, int noblock
, bool zerocopy
)
1424 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
1427 if (tfile
->socket
.sk
->sk_sndbuf
!= INT_MAX
)
1436 if (SKB_DATA_ALIGN(len
+ TUN_RX_PAD
) +
1437 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
)) > PAGE_SIZE
)
1443 static struct sk_buff
*tun_build_skb(struct tun_struct
*tun
,
1444 struct tun_file
*tfile
,
1445 struct iov_iter
*from
,
1446 struct virtio_net_hdr
*hdr
,
1447 int len
, int *skb_xdp
)
1449 struct page_frag
*alloc_frag
= ¤t
->task_frag
;
1450 struct sk_buff
*skb
;
1451 struct bpf_prog
*xdp_prog
;
1452 int buflen
= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
1453 unsigned int delta
= 0;
1456 bool xdp_xmit
= false;
1457 int err
, pad
= TUN_RX_PAD
;
1460 xdp_prog
= rcu_dereference(tun
->xdp_prog
);
1462 pad
+= TUN_HEADROOM
;
1463 buflen
+= SKB_DATA_ALIGN(len
+ pad
);
1466 alloc_frag
->offset
= ALIGN((u64
)alloc_frag
->offset
, SMP_CACHE_BYTES
);
1467 if (unlikely(!skb_page_frag_refill(buflen
, alloc_frag
, GFP_KERNEL
)))
1468 return ERR_PTR(-ENOMEM
);
1470 buf
= (char *)page_address(alloc_frag
->page
) + alloc_frag
->offset
;
1471 copied
= copy_page_from_iter(alloc_frag
->page
,
1472 alloc_frag
->offset
+ pad
,
1475 return ERR_PTR(-EFAULT
);
1477 /* There's a small window that XDP may be set after the check
1478 * of xdp_prog above, this should be rare and for simplicity
1479 * we do XDP on skb in case the headroom is not enough.
1481 if (hdr
->gso_type
|| !xdp_prog
)
1488 xdp_prog
= rcu_dereference(tun
->xdp_prog
);
1489 if (xdp_prog
&& !*skb_xdp
) {
1490 struct xdp_buff xdp
;
1494 xdp
.data_hard_start
= buf
;
1495 xdp
.data
= buf
+ pad
;
1496 xdp_set_data_meta_invalid(&xdp
);
1497 xdp
.data_end
= xdp
.data
+ len
;
1498 orig_data
= xdp
.data
;
1499 act
= bpf_prog_run_xdp(xdp_prog
, &xdp
);
1503 get_page(alloc_frag
->page
);
1504 alloc_frag
->offset
+= buflen
;
1505 err
= xdp_do_redirect(tun
->dev
, &xdp
, xdp_prog
);
1516 delta
= orig_data
- xdp
.data
;
1519 bpf_warn_invalid_xdp_action(act
);
1522 trace_xdp_exception(tun
->dev
, xdp_prog
, act
);
1529 skb
= build_skb(buf
, buflen
);
1533 return ERR_PTR(-ENOMEM
);
1536 skb_reserve(skb
, pad
- delta
);
1537 skb_put(skb
, len
+ delta
);
1538 get_page(alloc_frag
->page
);
1539 alloc_frag
->offset
+= buflen
;
1542 skb
->dev
= tun
->dev
;
1543 generic_xdp_tx(skb
, xdp_prog
);
1555 put_page(alloc_frag
->page
);
1559 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1563 /* Get packet from user space buffer */
1564 static ssize_t
tun_get_user(struct tun_struct
*tun
, struct tun_file
*tfile
,
1565 void *msg_control
, struct iov_iter
*from
,
1566 int noblock
, bool more
)
1568 struct tun_pi pi
= { 0, cpu_to_be16(ETH_P_IP
) };
1569 struct sk_buff
*skb
;
1570 size_t total_len
= iov_iter_count(from
);
1571 size_t len
= total_len
, align
= tun
->align
, linear
;
1572 struct virtio_net_hdr gso
= { 0 };
1573 struct tun_pcpu_stats
*stats
;
1576 bool zerocopy
= false;
1580 bool frags
= tun_napi_frags_enabled(tfile
);
1582 if (!(tun
->flags
& IFF_NO_PI
)) {
1583 if (len
< sizeof(pi
))
1587 if (!copy_from_iter_full(&pi
, sizeof(pi
), from
))
1591 if (tun
->flags
& IFF_VNET_HDR
) {
1592 int vnet_hdr_sz
= READ_ONCE(tun
->vnet_hdr_sz
);
1594 if (len
< vnet_hdr_sz
)
1598 if (!copy_from_iter_full(&gso
, sizeof(gso
), from
))
1601 if ((gso
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) &&
1602 tun16_to_cpu(tun
, gso
.csum_start
) + tun16_to_cpu(tun
, gso
.csum_offset
) + 2 > tun16_to_cpu(tun
, gso
.hdr_len
))
1603 gso
.hdr_len
= cpu_to_tun16(tun
, tun16_to_cpu(tun
, gso
.csum_start
) + tun16_to_cpu(tun
, gso
.csum_offset
) + 2);
1605 if (tun16_to_cpu(tun
, gso
.hdr_len
) > len
)
1607 iov_iter_advance(from
, vnet_hdr_sz
- sizeof(gso
));
1610 if ((tun
->flags
& TUN_TYPE_MASK
) == IFF_TAP
) {
1611 align
+= NET_IP_ALIGN
;
1612 if (unlikely(len
< ETH_HLEN
||
1613 (gso
.hdr_len
&& tun16_to_cpu(tun
, gso
.hdr_len
) < ETH_HLEN
)))
1617 good_linear
= SKB_MAX_HEAD(align
);
1620 struct iov_iter i
= *from
;
1622 /* There are 256 bytes to be copied in skb, so there is
1623 * enough room for skb expand head in case it is used.
1624 * The rest of the buffer is mapped from userspace.
1626 copylen
= gso
.hdr_len
? tun16_to_cpu(tun
, gso
.hdr_len
) : GOODCOPY_LEN
;
1627 if (copylen
> good_linear
)
1628 copylen
= good_linear
;
1630 iov_iter_advance(&i
, copylen
);
1631 if (iov_iter_npages(&i
, INT_MAX
) <= MAX_SKB_FRAGS
)
1635 if (!frags
&& tun_can_build_skb(tun
, tfile
, len
, noblock
, zerocopy
)) {
1636 /* For the packet that is not easy to be processed
1637 * (e.g gso or jumbo packet), we will do it at after
1638 * skb was created with generic XDP routine.
1640 skb
= tun_build_skb(tun
, tfile
, from
, &gso
, len
, &skb_xdp
);
1642 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1643 return PTR_ERR(skb
);
1650 if (tun16_to_cpu(tun
, gso
.hdr_len
) > good_linear
)
1651 linear
= good_linear
;
1653 linear
= tun16_to_cpu(tun
, gso
.hdr_len
);
1657 mutex_lock(&tfile
->napi_mutex
);
1658 skb
= tun_napi_alloc_frags(tfile
, copylen
, from
);
1659 /* tun_napi_alloc_frags() enforces a layout for the skb.
1660 * If zerocopy is enabled, then this layout will be
1661 * overwritten by zerocopy_sg_from_iter().
1665 skb
= tun_alloc_skb(tfile
, align
, copylen
, linear
,
1670 if (PTR_ERR(skb
) != -EAGAIN
)
1671 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1673 mutex_unlock(&tfile
->napi_mutex
);
1674 return PTR_ERR(skb
);
1678 err
= zerocopy_sg_from_iter(skb
, from
);
1680 err
= skb_copy_datagram_from_iter(skb
, 0, from
, len
);
1685 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1688 tfile
->napi
.skb
= NULL
;
1689 mutex_unlock(&tfile
->napi_mutex
);
1696 if (virtio_net_hdr_to_skb(skb
, &gso
, tun_is_little_endian(tun
))) {
1697 this_cpu_inc(tun
->pcpu_stats
->rx_frame_errors
);
1700 tfile
->napi
.skb
= NULL
;
1701 mutex_unlock(&tfile
->napi_mutex
);
1707 switch (tun
->flags
& TUN_TYPE_MASK
) {
1709 if (tun
->flags
& IFF_NO_PI
) {
1710 u8 ip_version
= skb
->len
? (skb
->data
[0] >> 4) : 0;
1712 switch (ip_version
) {
1714 pi
.proto
= htons(ETH_P_IP
);
1717 pi
.proto
= htons(ETH_P_IPV6
);
1720 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1726 skb_reset_mac_header(skb
);
1727 skb
->protocol
= pi
.proto
;
1728 skb
->dev
= tun
->dev
;
1732 skb
->protocol
= eth_type_trans(skb
, tun
->dev
);
1736 /* copy skb_ubuf_info for callback when skb has no error */
1738 skb_shinfo(skb
)->destructor_arg
= msg_control
;
1739 skb_shinfo(skb
)->tx_flags
|= SKBTX_DEV_ZEROCOPY
;
1740 skb_shinfo(skb
)->tx_flags
|= SKBTX_SHARED_FRAG
;
1741 } else if (msg_control
) {
1742 struct ubuf_info
*uarg
= msg_control
;
1743 uarg
->callback(uarg
, false);
1746 skb_reset_network_header(skb
);
1747 skb_probe_transport_header(skb
, 0);
1750 struct bpf_prog
*xdp_prog
;
1755 xdp_prog
= rcu_dereference(tun
->xdp_prog
);
1757 ret
= do_xdp_generic(xdp_prog
, skb
);
1758 if (ret
!= XDP_PASS
) {
1768 rxhash
= __skb_get_hash_symmetric(skb
);
1771 if (unlikely(!(tun
->dev
->flags
& IFF_UP
))) {
1778 /* Exercise flow dissector code path. */
1779 u32 headlen
= eth_get_headlen(skb
->data
, skb_headlen(skb
));
1781 if (unlikely(headlen
> skb_headlen(skb
))) {
1782 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1783 napi_free_frags(&tfile
->napi
);
1785 mutex_unlock(&tfile
->napi_mutex
);
1791 napi_gro_frags(&tfile
->napi
);
1793 mutex_unlock(&tfile
->napi_mutex
);
1794 } else if (tfile
->napi_enabled
) {
1795 struct sk_buff_head
*queue
= &tfile
->sk
.sk_write_queue
;
1798 spin_lock_bh(&queue
->lock
);
1799 __skb_queue_tail(queue
, skb
);
1800 queue_len
= skb_queue_len(queue
);
1801 spin_unlock(&queue
->lock
);
1803 if (!more
|| queue_len
> NAPI_POLL_WEIGHT
)
1804 napi_schedule(&tfile
->napi
);
1807 } else if (!IS_ENABLED(CONFIG_4KSTACKS
)) {
1808 tun_rx_batched(tun
, tfile
, skb
, more
);
1814 stats
= get_cpu_ptr(tun
->pcpu_stats
);
1815 u64_stats_update_begin(&stats
->syncp
);
1816 stats
->rx_packets
++;
1817 stats
->rx_bytes
+= len
;
1818 u64_stats_update_end(&stats
->syncp
);
1821 tun_flow_update(tun
, rxhash
, tfile
);
1825 static ssize_t
tun_chr_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
1827 struct file
*file
= iocb
->ki_filp
;
1828 struct tun_file
*tfile
= file
->private_data
;
1829 struct tun_struct
*tun
= tun_get(tfile
);
1835 result
= tun_get_user(tun
, tfile
, NULL
, from
,
1836 file
->f_flags
& O_NONBLOCK
, false);
1842 /* Put packet to the user space buffer */
1843 static ssize_t
tun_put_user(struct tun_struct
*tun
,
1844 struct tun_file
*tfile
,
1845 struct sk_buff
*skb
,
1846 struct iov_iter
*iter
)
1848 struct tun_pi pi
= { 0, skb
->protocol
};
1849 struct tun_pcpu_stats
*stats
;
1851 int vlan_offset
= 0;
1853 int vnet_hdr_sz
= 0;
1855 if (skb_vlan_tag_present(skb
))
1856 vlan_hlen
= VLAN_HLEN
;
1858 if (tun
->flags
& IFF_VNET_HDR
)
1859 vnet_hdr_sz
= READ_ONCE(tun
->vnet_hdr_sz
);
1861 total
= skb
->len
+ vlan_hlen
+ vnet_hdr_sz
;
1863 if (!(tun
->flags
& IFF_NO_PI
)) {
1864 if (iov_iter_count(iter
) < sizeof(pi
))
1867 total
+= sizeof(pi
);
1868 if (iov_iter_count(iter
) < total
) {
1869 /* Packet will be striped */
1870 pi
.flags
|= TUN_PKT_STRIP
;
1873 if (copy_to_iter(&pi
, sizeof(pi
), iter
) != sizeof(pi
))
1878 struct virtio_net_hdr gso
;
1880 if (iov_iter_count(iter
) < vnet_hdr_sz
)
1883 if (virtio_net_hdr_from_skb(skb
, &gso
,
1884 tun_is_little_endian(tun
), true,
1886 struct skb_shared_info
*sinfo
= skb_shinfo(skb
);
1887 pr_err("unexpected GSO type: "
1888 "0x%x, gso_size %d, hdr_len %d\n",
1889 sinfo
->gso_type
, tun16_to_cpu(tun
, gso
.gso_size
),
1890 tun16_to_cpu(tun
, gso
.hdr_len
));
1891 print_hex_dump(KERN_ERR
, "tun: ",
1894 min((int)tun16_to_cpu(tun
, gso
.hdr_len
), 64), true);
1899 if (copy_to_iter(&gso
, sizeof(gso
), iter
) != sizeof(gso
))
1902 iov_iter_advance(iter
, vnet_hdr_sz
- sizeof(gso
));
1908 __be16 h_vlan_proto
;
1912 veth
.h_vlan_proto
= skb
->vlan_proto
;
1913 veth
.h_vlan_TCI
= htons(skb_vlan_tag_get(skb
));
1915 vlan_offset
= offsetof(struct vlan_ethhdr
, h_vlan_proto
);
1917 ret
= skb_copy_datagram_iter(skb
, 0, iter
, vlan_offset
);
1918 if (ret
|| !iov_iter_count(iter
))
1921 ret
= copy_to_iter(&veth
, sizeof(veth
), iter
);
1922 if (ret
!= sizeof(veth
) || !iov_iter_count(iter
))
1926 skb_copy_datagram_iter(skb
, vlan_offset
, iter
, skb
->len
- vlan_offset
);
1929 /* caller is in process context, */
1930 stats
= get_cpu_ptr(tun
->pcpu_stats
);
1931 u64_stats_update_begin(&stats
->syncp
);
1932 stats
->tx_packets
++;
1933 stats
->tx_bytes
+= skb
->len
+ vlan_hlen
;
1934 u64_stats_update_end(&stats
->syncp
);
1935 put_cpu_ptr(tun
->pcpu_stats
);
1940 static struct sk_buff
*tun_ring_recv(struct tun_file
*tfile
, int noblock
,
1943 DECLARE_WAITQUEUE(wait
, current
);
1944 struct sk_buff
*skb
= NULL
;
1947 skb
= skb_array_consume(&tfile
->tx_array
);
1955 add_wait_queue(&tfile
->wq
.wait
, &wait
);
1958 set_current_state(TASK_INTERRUPTIBLE
);
1959 skb
= skb_array_consume(&tfile
->tx_array
);
1962 if (signal_pending(current
)) {
1963 error
= -ERESTARTSYS
;
1966 if (tfile
->socket
.sk
->sk_shutdown
& RCV_SHUTDOWN
) {
1974 __set_current_state(TASK_RUNNING
);
1975 remove_wait_queue(&tfile
->wq
.wait
, &wait
);
1982 static ssize_t
tun_do_read(struct tun_struct
*tun
, struct tun_file
*tfile
,
1983 struct iov_iter
*to
,
1984 int noblock
, struct sk_buff
*skb
)
1989 tun_debug(KERN_INFO
, tun
, "tun_do_read\n");
1991 if (!iov_iter_count(to
)) {
1998 /* Read frames from ring */
1999 skb
= tun_ring_recv(tfile
, noblock
, &err
);
2004 ret
= tun_put_user(tun
, tfile
, skb
, to
);
2005 if (unlikely(ret
< 0))
2013 static ssize_t
tun_chr_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
2015 struct file
*file
= iocb
->ki_filp
;
2016 struct tun_file
*tfile
= file
->private_data
;
2017 struct tun_struct
*tun
= tun_get(tfile
);
2018 ssize_t len
= iov_iter_count(to
), ret
;
2022 ret
= tun_do_read(tun
, tfile
, to
, file
->f_flags
& O_NONBLOCK
, NULL
);
2023 ret
= min_t(ssize_t
, ret
, len
);
2030 static void tun_free_netdev(struct net_device
*dev
)
2032 struct tun_struct
*tun
= netdev_priv(dev
);
2034 BUG_ON(!(list_empty(&tun
->disabled
)));
2035 free_percpu(tun
->pcpu_stats
);
2036 tun_flow_uninit(tun
);
2037 security_tun_dev_free_security(tun
->security
);
2040 static void tun_setup(struct net_device
*dev
)
2042 struct tun_struct
*tun
= netdev_priv(dev
);
2044 tun
->owner
= INVALID_UID
;
2045 tun
->group
= INVALID_GID
;
2047 dev
->ethtool_ops
= &tun_ethtool_ops
;
2048 dev
->needs_free_netdev
= true;
2049 dev
->priv_destructor
= tun_free_netdev
;
2050 /* We prefer our own queue length */
2051 dev
->tx_queue_len
= TUN_READQ_SIZE
;
2054 /* Trivial set of netlink ops to allow deleting tun or tap
2055 * device with netlink.
2057 static int tun_validate(struct nlattr
*tb
[], struct nlattr
*data
[],
2058 struct netlink_ext_ack
*extack
)
2060 NL_SET_ERR_MSG(extack
,
2061 "tun/tap creation via rtnetlink is not supported.");
2065 static struct rtnl_link_ops tun_link_ops __read_mostly
= {
2067 .priv_size
= sizeof(struct tun_struct
),
2069 .validate
= tun_validate
,
2072 static void tun_sock_write_space(struct sock
*sk
)
2074 struct tun_file
*tfile
;
2075 wait_queue_head_t
*wqueue
;
2077 if (!sock_writeable(sk
))
2080 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
))
2083 wqueue
= sk_sleep(sk
);
2084 if (wqueue
&& waitqueue_active(wqueue
))
2085 wake_up_interruptible_sync_poll(wqueue
, POLLOUT
|
2086 POLLWRNORM
| POLLWRBAND
);
2088 tfile
= container_of(sk
, struct tun_file
, sk
);
2089 kill_fasync(&tfile
->fasync
, SIGIO
, POLL_OUT
);
2092 static int tun_sendmsg(struct socket
*sock
, struct msghdr
*m
, size_t total_len
)
2095 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
2096 struct tun_struct
*tun
= tun_get(tfile
);
2101 ret
= tun_get_user(tun
, tfile
, m
->msg_control
, &m
->msg_iter
,
2102 m
->msg_flags
& MSG_DONTWAIT
,
2103 m
->msg_flags
& MSG_MORE
);
2108 static int tun_recvmsg(struct socket
*sock
, struct msghdr
*m
, size_t total_len
,
2111 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
2112 struct tun_struct
*tun
= tun_get(tfile
);
2113 struct sk_buff
*skb
= m
->msg_control
;
2121 if (flags
& ~(MSG_DONTWAIT
|MSG_TRUNC
|MSG_ERRQUEUE
)) {
2125 if (flags
& MSG_ERRQUEUE
) {
2126 ret
= sock_recv_errqueue(sock
->sk
, m
, total_len
,
2127 SOL_PACKET
, TUN_TX_TIMESTAMP
);
2130 ret
= tun_do_read(tun
, tfile
, &m
->msg_iter
, flags
& MSG_DONTWAIT
, skb
);
2131 if (ret
> (ssize_t
)total_len
) {
2132 m
->msg_flags
|= MSG_TRUNC
;
2133 ret
= flags
& MSG_TRUNC
? ret
: total_len
;
2147 static int tun_peek_len(struct socket
*sock
)
2149 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
2150 struct tun_struct
*tun
;
2153 tun
= tun_get(tfile
);
2157 ret
= skb_array_peek_len(&tfile
->tx_array
);
2163 /* Ops structure to mimic raw sockets with tun */
2164 static const struct proto_ops tun_socket_ops
= {
2165 .peek_len
= tun_peek_len
,
2166 .sendmsg
= tun_sendmsg
,
2167 .recvmsg
= tun_recvmsg
,
2170 static struct proto tun_proto
= {
2172 .owner
= THIS_MODULE
,
2173 .obj_size
= sizeof(struct tun_file
),
2176 static int tun_flags(struct tun_struct
*tun
)
2178 return tun
->flags
& (TUN_FEATURES
| IFF_PERSIST
| IFF_TUN
| IFF_TAP
);
2181 static ssize_t
tun_show_flags(struct device
*dev
, struct device_attribute
*attr
,
2184 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
2185 return sprintf(buf
, "0x%x\n", tun_flags(tun
));
2188 static ssize_t
tun_show_owner(struct device
*dev
, struct device_attribute
*attr
,
2191 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
2192 return uid_valid(tun
->owner
)?
2193 sprintf(buf
, "%u\n",
2194 from_kuid_munged(current_user_ns(), tun
->owner
)):
2195 sprintf(buf
, "-1\n");
2198 static ssize_t
tun_show_group(struct device
*dev
, struct device_attribute
*attr
,
2201 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
2202 return gid_valid(tun
->group
) ?
2203 sprintf(buf
, "%u\n",
2204 from_kgid_munged(current_user_ns(), tun
->group
)):
2205 sprintf(buf
, "-1\n");
2208 static DEVICE_ATTR(tun_flags
, 0444, tun_show_flags
, NULL
);
2209 static DEVICE_ATTR(owner
, 0444, tun_show_owner
, NULL
);
2210 static DEVICE_ATTR(group
, 0444, tun_show_group
, NULL
);
2212 static struct attribute
*tun_dev_attrs
[] = {
2213 &dev_attr_tun_flags
.attr
,
2214 &dev_attr_owner
.attr
,
2215 &dev_attr_group
.attr
,
2219 static const struct attribute_group tun_attr_group
= {
2220 .attrs
= tun_dev_attrs
2223 static int tun_set_iff(struct net
*net
, struct file
*file
, struct ifreq
*ifr
)
2225 struct tun_struct
*tun
;
2226 struct tun_file
*tfile
= file
->private_data
;
2227 struct net_device
*dev
;
2230 if (tfile
->detached
)
2233 if ((ifr
->ifr_flags
& IFF_NAPI_FRAGS
)) {
2234 if (!capable(CAP_NET_ADMIN
))
2237 if (!(ifr
->ifr_flags
& IFF_NAPI
) ||
2238 (ifr
->ifr_flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2242 dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
2244 if (ifr
->ifr_flags
& IFF_TUN_EXCL
)
2246 if ((ifr
->ifr_flags
& IFF_TUN
) && dev
->netdev_ops
== &tun_netdev_ops
)
2247 tun
= netdev_priv(dev
);
2248 else if ((ifr
->ifr_flags
& IFF_TAP
) && dev
->netdev_ops
== &tap_netdev_ops
)
2249 tun
= netdev_priv(dev
);
2253 if (!!(ifr
->ifr_flags
& IFF_MULTI_QUEUE
) !=
2254 !!(tun
->flags
& IFF_MULTI_QUEUE
))
2257 if (tun_not_capable(tun
))
2259 err
= security_tun_dev_open(tun
->security
);
2263 err
= tun_attach(tun
, file
, ifr
->ifr_flags
& IFF_NOFILTER
,
2264 ifr
->ifr_flags
& IFF_NAPI
,
2265 ifr
->ifr_flags
& IFF_NAPI_FRAGS
);
2269 if (tun
->flags
& IFF_MULTI_QUEUE
&&
2270 (tun
->numqueues
+ tun
->numdisabled
> 1)) {
2271 /* One or more queue has already been attached, no need
2272 * to initialize the device again.
2279 unsigned long flags
= 0;
2280 int queues
= ifr
->ifr_flags
& IFF_MULTI_QUEUE
?
2283 if (!ns_capable(net
->user_ns
, CAP_NET_ADMIN
))
2285 err
= security_tun_dev_create();
2290 if (ifr
->ifr_flags
& IFF_TUN
) {
2294 } else if (ifr
->ifr_flags
& IFF_TAP
) {
2302 name
= ifr
->ifr_name
;
2304 dev
= alloc_netdev_mqs(sizeof(struct tun_struct
), name
,
2305 NET_NAME_UNKNOWN
, tun_setup
, queues
,
2310 err
= dev_get_valid_name(net
, dev
, name
);
2314 dev_net_set(dev
, net
);
2315 dev
->rtnl_link_ops
= &tun_link_ops
;
2316 dev
->ifindex
= tfile
->ifindex
;
2317 dev
->sysfs_groups
[0] = &tun_attr_group
;
2319 tun
= netdev_priv(dev
);
2322 tun
->txflt
.count
= 0;
2323 tun
->vnet_hdr_sz
= sizeof(struct virtio_net_hdr
);
2325 tun
->align
= NET_SKB_PAD
;
2326 tun
->filter_attached
= false;
2327 tun
->sndbuf
= tfile
->socket
.sk
->sk_sndbuf
;
2328 tun
->rx_batched
= 0;
2330 tun
->pcpu_stats
= netdev_alloc_pcpu_stats(struct tun_pcpu_stats
);
2331 if (!tun
->pcpu_stats
) {
2336 spin_lock_init(&tun
->lock
);
2338 err
= security_tun_dev_alloc_security(&tun
->security
);
2345 dev
->hw_features
= NETIF_F_SG
| NETIF_F_FRAGLIST
|
2346 TUN_USER_FEATURES
| NETIF_F_HW_VLAN_CTAG_TX
|
2347 NETIF_F_HW_VLAN_STAG_TX
;
2348 dev
->features
= dev
->hw_features
| NETIF_F_LLTX
;
2349 dev
->vlan_features
= dev
->features
&
2350 ~(NETIF_F_HW_VLAN_CTAG_TX
|
2351 NETIF_F_HW_VLAN_STAG_TX
);
2353 INIT_LIST_HEAD(&tun
->disabled
);
2354 err
= tun_attach(tun
, file
, false, ifr
->ifr_flags
& IFF_NAPI
,
2355 ifr
->ifr_flags
& IFF_NAPI_FRAGS
);
2359 err
= register_netdevice(tun
->dev
);
2364 netif_carrier_on(tun
->dev
);
2366 tun_debug(KERN_INFO
, tun
, "tun_set_iff\n");
2368 tun
->flags
= (tun
->flags
& ~TUN_FEATURES
) |
2369 (ifr
->ifr_flags
& TUN_FEATURES
);
2371 /* Make sure persistent devices do not get stuck in
2374 if (netif_running(tun
->dev
))
2375 netif_tx_wake_all_queues(tun
->dev
);
2377 strcpy(ifr
->ifr_name
, tun
->dev
->name
);
2381 tun_detach_all(dev
);
2382 /* register_netdevice() already called tun_free_netdev() */
2386 tun_flow_uninit(tun
);
2387 security_tun_dev_free_security(tun
->security
);
2389 free_percpu(tun
->pcpu_stats
);
2395 static void tun_get_iff(struct net
*net
, struct tun_struct
*tun
,
2398 tun_debug(KERN_INFO
, tun
, "tun_get_iff\n");
2400 strcpy(ifr
->ifr_name
, tun
->dev
->name
);
2402 ifr
->ifr_flags
= tun_flags(tun
);
2406 /* This is like a cut-down ethtool ops, except done via tun fd so no
2407 * privs required. */
2408 static int set_offload(struct tun_struct
*tun
, unsigned long arg
)
2410 netdev_features_t features
= 0;
2412 if (arg
& TUN_F_CSUM
) {
2413 features
|= NETIF_F_HW_CSUM
;
2416 if (arg
& (TUN_F_TSO4
|TUN_F_TSO6
)) {
2417 if (arg
& TUN_F_TSO_ECN
) {
2418 features
|= NETIF_F_TSO_ECN
;
2419 arg
&= ~TUN_F_TSO_ECN
;
2421 if (arg
& TUN_F_TSO4
)
2422 features
|= NETIF_F_TSO
;
2423 if (arg
& TUN_F_TSO6
)
2424 features
|= NETIF_F_TSO6
;
2425 arg
&= ~(TUN_F_TSO4
|TUN_F_TSO6
);
2431 /* This gives the user a way to test for new features in future by
2432 * trying to set them. */
2436 tun
->set_features
= features
;
2437 tun
->dev
->wanted_features
&= ~TUN_USER_FEATURES
;
2438 tun
->dev
->wanted_features
|= features
;
2439 netdev_update_features(tun
->dev
);
2444 static void tun_detach_filter(struct tun_struct
*tun
, int n
)
2447 struct tun_file
*tfile
;
2449 for (i
= 0; i
< n
; i
++) {
2450 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
2451 lock_sock(tfile
->socket
.sk
);
2452 sk_detach_filter(tfile
->socket
.sk
);
2453 release_sock(tfile
->socket
.sk
);
2456 tun
->filter_attached
= false;
2459 static int tun_attach_filter(struct tun_struct
*tun
)
2462 struct tun_file
*tfile
;
2464 for (i
= 0; i
< tun
->numqueues
; i
++) {
2465 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
2466 lock_sock(tfile
->socket
.sk
);
2467 ret
= sk_attach_filter(&tun
->fprog
, tfile
->socket
.sk
);
2468 release_sock(tfile
->socket
.sk
);
2470 tun_detach_filter(tun
, i
);
2475 tun
->filter_attached
= true;
2479 static void tun_set_sndbuf(struct tun_struct
*tun
)
2481 struct tun_file
*tfile
;
2484 for (i
= 0; i
< tun
->numqueues
; i
++) {
2485 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
2486 tfile
->socket
.sk
->sk_sndbuf
= tun
->sndbuf
;
2490 static int tun_set_queue(struct file
*file
, struct ifreq
*ifr
)
2492 struct tun_file
*tfile
= file
->private_data
;
2493 struct tun_struct
*tun
;
2498 if (ifr
->ifr_flags
& IFF_ATTACH_QUEUE
) {
2499 tun
= tfile
->detached
;
2504 ret
= security_tun_dev_attach_queue(tun
->security
);
2507 ret
= tun_attach(tun
, file
, false, tun
->flags
& IFF_NAPI
,
2508 tun
->flags
& IFF_NAPI_FRAGS
);
2509 } else if (ifr
->ifr_flags
& IFF_DETACH_QUEUE
) {
2510 tun
= rtnl_dereference(tfile
->tun
);
2511 if (!tun
|| !(tun
->flags
& IFF_MULTI_QUEUE
) || tfile
->detached
)
2514 __tun_detach(tfile
, false);
2523 static long __tun_chr_ioctl(struct file
*file
, unsigned int cmd
,
2524 unsigned long arg
, int ifreq_len
)
2526 struct tun_file
*tfile
= file
->private_data
;
2527 struct tun_struct
*tun
;
2528 void __user
* argp
= (void __user
*)arg
;
2529 unsigned int ifindex
, carrier
;
2538 if (cmd
== TUNSETIFF
|| cmd
== TUNSETQUEUE
|| _IOC_TYPE(cmd
) == SOCK_IOC_TYPE
) {
2539 if (copy_from_user(&ifr
, argp
, ifreq_len
))
2542 memset(&ifr
, 0, sizeof(ifr
));
2544 if (cmd
== TUNGETFEATURES
) {
2545 /* Currently this just means: "what IFF flags are valid?".
2546 * This is needed because we never checked for invalid flags on
2549 return put_user(IFF_TUN
| IFF_TAP
| TUN_FEATURES
,
2550 (unsigned int __user
*)argp
);
2551 } else if (cmd
== TUNSETQUEUE
)
2552 return tun_set_queue(file
, &ifr
);
2557 tun
= tun_get(tfile
);
2558 if (cmd
== TUNSETIFF
) {
2563 ifr
.ifr_name
[IFNAMSIZ
-1] = '\0';
2565 ret
= tun_set_iff(sock_net(&tfile
->sk
), file
, &ifr
);
2570 if (copy_to_user(argp
, &ifr
, ifreq_len
))
2574 if (cmd
== TUNSETIFINDEX
) {
2580 if (copy_from_user(&ifindex
, argp
, sizeof(ifindex
)))
2584 tfile
->ifindex
= ifindex
;
2592 tun_debug(KERN_INFO
, tun
, "tun_chr_ioctl cmd %u\n", cmd
);
2597 tun_get_iff(current
->nsproxy
->net_ns
, tun
, &ifr
);
2599 if (tfile
->detached
)
2600 ifr
.ifr_flags
|= IFF_DETACH_QUEUE
;
2601 if (!tfile
->socket
.sk
->sk_filter
)
2602 ifr
.ifr_flags
|= IFF_NOFILTER
;
2604 if (copy_to_user(argp
, &ifr
, ifreq_len
))
2609 /* Disable/Enable checksum */
2611 /* [unimplemented] */
2612 tun_debug(KERN_INFO
, tun
, "ignored: set checksum %s\n",
2613 arg
? "disabled" : "enabled");
2617 /* Disable/Enable persist mode. Keep an extra reference to the
2618 * module to prevent the module being unprobed.
2620 if (arg
&& !(tun
->flags
& IFF_PERSIST
)) {
2621 tun
->flags
|= IFF_PERSIST
;
2622 __module_get(THIS_MODULE
);
2624 if (!arg
&& (tun
->flags
& IFF_PERSIST
)) {
2625 tun
->flags
&= ~IFF_PERSIST
;
2626 module_put(THIS_MODULE
);
2629 tun_debug(KERN_INFO
, tun
, "persist %s\n",
2630 arg
? "enabled" : "disabled");
2634 /* Set owner of the device */
2635 owner
= make_kuid(current_user_ns(), arg
);
2636 if (!uid_valid(owner
)) {
2641 tun_debug(KERN_INFO
, tun
, "owner set to %u\n",
2642 from_kuid(&init_user_ns
, tun
->owner
));
2646 /* Set group of the device */
2647 group
= make_kgid(current_user_ns(), arg
);
2648 if (!gid_valid(group
)) {
2653 tun_debug(KERN_INFO
, tun
, "group set to %u\n",
2654 from_kgid(&init_user_ns
, tun
->group
));
2658 /* Only allow setting the type when the interface is down */
2659 if (tun
->dev
->flags
& IFF_UP
) {
2660 tun_debug(KERN_INFO
, tun
,
2661 "Linktype set failed because interface is up\n");
2664 tun
->dev
->type
= (int) arg
;
2665 tun_debug(KERN_INFO
, tun
, "linktype set to %d\n",
2677 ret
= set_offload(tun
, arg
);
2680 case TUNSETTXFILTER
:
2681 /* Can be set only for TAPs */
2683 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2685 ret
= update_filter(&tun
->txflt
, (void __user
*)arg
);
2689 /* Get hw address */
2690 memcpy(ifr
.ifr_hwaddr
.sa_data
, tun
->dev
->dev_addr
, ETH_ALEN
);
2691 ifr
.ifr_hwaddr
.sa_family
= tun
->dev
->type
;
2692 if (copy_to_user(argp
, &ifr
, ifreq_len
))
2697 /* Set hw address */
2698 tun_debug(KERN_DEBUG
, tun
, "set hw address: %pM\n",
2699 ifr
.ifr_hwaddr
.sa_data
);
2701 ret
= dev_set_mac_address(tun
->dev
, &ifr
.ifr_hwaddr
);
2705 sndbuf
= tfile
->socket
.sk
->sk_sndbuf
;
2706 if (copy_to_user(argp
, &sndbuf
, sizeof(sndbuf
)))
2711 if (copy_from_user(&sndbuf
, argp
, sizeof(sndbuf
))) {
2720 tun
->sndbuf
= sndbuf
;
2721 tun_set_sndbuf(tun
);
2724 case TUNGETVNETHDRSZ
:
2725 vnet_hdr_sz
= tun
->vnet_hdr_sz
;
2726 if (copy_to_user(argp
, &vnet_hdr_sz
, sizeof(vnet_hdr_sz
)))
2730 case TUNSETVNETHDRSZ
:
2731 if (copy_from_user(&vnet_hdr_sz
, argp
, sizeof(vnet_hdr_sz
))) {
2735 if (vnet_hdr_sz
< (int)sizeof(struct virtio_net_hdr
)) {
2740 tun
->vnet_hdr_sz
= vnet_hdr_sz
;
2744 le
= !!(tun
->flags
& TUN_VNET_LE
);
2745 if (put_user(le
, (int __user
*)argp
))
2750 if (get_user(le
, (int __user
*)argp
)) {
2755 tun
->flags
|= TUN_VNET_LE
;
2757 tun
->flags
&= ~TUN_VNET_LE
;
2761 ret
= tun_get_vnet_be(tun
, argp
);
2765 ret
= tun_set_vnet_be(tun
, argp
);
2768 case TUNATTACHFILTER
:
2769 /* Can be set only for TAPs */
2771 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2774 if (copy_from_user(&tun
->fprog
, argp
, sizeof(tun
->fprog
)))
2777 ret
= tun_attach_filter(tun
);
2780 case TUNDETACHFILTER
:
2781 /* Can be set only for TAPs */
2783 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2786 tun_detach_filter(tun
, tun
->numqueues
);
2791 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2794 if (copy_to_user(argp
, &tun
->fprog
, sizeof(tun
->fprog
)))
2801 if (copy_from_user(&carrier
, argp
, sizeof(carrier
)))
2804 ret
= tun_net_change_carrier(tun
->dev
, (bool)carrier
);
2819 static long tun_chr_ioctl(struct file
*file
,
2820 unsigned int cmd
, unsigned long arg
)
2822 return __tun_chr_ioctl(file
, cmd
, arg
, sizeof (struct ifreq
));
2825 #ifdef CONFIG_COMPAT
2826 static long tun_chr_compat_ioctl(struct file
*file
,
2827 unsigned int cmd
, unsigned long arg
)
2832 case TUNSETTXFILTER
:
2837 arg
= (unsigned long)compat_ptr(arg
);
2840 arg
= (compat_ulong_t
)arg
;
2845 * compat_ifreq is shorter than ifreq, so we must not access beyond
2846 * the end of that structure. All fields that are used in this
2847 * driver are compatible though, we don't need to convert the
2850 return __tun_chr_ioctl(file
, cmd
, arg
, sizeof(struct compat_ifreq
));
2852 #endif /* CONFIG_COMPAT */
2854 static int tun_chr_fasync(int fd
, struct file
*file
, int on
)
2856 struct tun_file
*tfile
= file
->private_data
;
2859 if ((ret
= fasync_helper(fd
, file
, on
, &tfile
->fasync
)) < 0)
2863 __f_setown(file
, task_pid(current
), PIDTYPE_PID
, 0);
2864 tfile
->flags
|= TUN_FASYNC
;
2866 tfile
->flags
&= ~TUN_FASYNC
;
2872 static int tun_chr_open(struct inode
*inode
, struct file
* file
)
2874 struct net
*net
= current
->nsproxy
->net_ns
;
2875 struct tun_file
*tfile
;
2877 DBG1(KERN_INFO
, "tunX: tun_chr_open\n");
2879 tfile
= (struct tun_file
*)sk_alloc(net
, AF_UNSPEC
, GFP_KERNEL
,
2883 if (skb_array_init(&tfile
->tx_array
, 0, GFP_KERNEL
)) {
2884 sk_free(&tfile
->sk
);
2888 mutex_init(&tfile
->napi_mutex
);
2889 RCU_INIT_POINTER(tfile
->tun
, NULL
);
2893 init_waitqueue_head(&tfile
->wq
.wait
);
2894 RCU_INIT_POINTER(tfile
->socket
.wq
, &tfile
->wq
);
2896 tfile
->socket
.file
= file
;
2897 tfile
->socket
.ops
= &tun_socket_ops
;
2899 sock_init_data(&tfile
->socket
, &tfile
->sk
);
2901 tfile
->sk
.sk_write_space
= tun_sock_write_space
;
2902 tfile
->sk
.sk_sndbuf
= INT_MAX
;
2904 file
->private_data
= tfile
;
2905 INIT_LIST_HEAD(&tfile
->next
);
2907 sock_set_flag(&tfile
->sk
, SOCK_ZEROCOPY
);
2912 static int tun_chr_close(struct inode
*inode
, struct file
*file
)
2914 struct tun_file
*tfile
= file
->private_data
;
2916 tun_detach(tfile
, true);
2921 #ifdef CONFIG_PROC_FS
2922 static void tun_chr_show_fdinfo(struct seq_file
*m
, struct file
*file
)
2924 struct tun_file
*tfile
= file
->private_data
;
2925 struct tun_struct
*tun
;
2928 memset(&ifr
, 0, sizeof(ifr
));
2931 tun
= tun_get(tfile
);
2933 tun_get_iff(current
->nsproxy
->net_ns
, tun
, &ifr
);
2939 seq_printf(m
, "iff:\t%s\n", ifr
.ifr_name
);
2943 static const struct file_operations tun_fops
= {
2944 .owner
= THIS_MODULE
,
2945 .llseek
= no_llseek
,
2946 .read_iter
= tun_chr_read_iter
,
2947 .write_iter
= tun_chr_write_iter
,
2948 .poll
= tun_chr_poll
,
2949 .unlocked_ioctl
= tun_chr_ioctl
,
2950 #ifdef CONFIG_COMPAT
2951 .compat_ioctl
= tun_chr_compat_ioctl
,
2953 .open
= tun_chr_open
,
2954 .release
= tun_chr_close
,
2955 .fasync
= tun_chr_fasync
,
2956 #ifdef CONFIG_PROC_FS
2957 .show_fdinfo
= tun_chr_show_fdinfo
,
2961 static struct miscdevice tun_miscdev
= {
2964 .nodename
= "net/tun",
2968 /* ethtool interface */
2970 static int tun_get_link_ksettings(struct net_device
*dev
,
2971 struct ethtool_link_ksettings
*cmd
)
2973 ethtool_link_ksettings_zero_link_mode(cmd
, supported
);
2974 ethtool_link_ksettings_zero_link_mode(cmd
, advertising
);
2975 cmd
->base
.speed
= SPEED_10
;
2976 cmd
->base
.duplex
= DUPLEX_FULL
;
2977 cmd
->base
.port
= PORT_TP
;
2978 cmd
->base
.phy_address
= 0;
2979 cmd
->base
.autoneg
= AUTONEG_DISABLE
;
2983 static void tun_get_drvinfo(struct net_device
*dev
, struct ethtool_drvinfo
*info
)
2985 struct tun_struct
*tun
= netdev_priv(dev
);
2987 strlcpy(info
->driver
, DRV_NAME
, sizeof(info
->driver
));
2988 strlcpy(info
->version
, DRV_VERSION
, sizeof(info
->version
));
2990 switch (tun
->flags
& TUN_TYPE_MASK
) {
2992 strlcpy(info
->bus_info
, "tun", sizeof(info
->bus_info
));
2995 strlcpy(info
->bus_info
, "tap", sizeof(info
->bus_info
));
3000 static u32
tun_get_msglevel(struct net_device
*dev
)
3003 struct tun_struct
*tun
= netdev_priv(dev
);
3010 static void tun_set_msglevel(struct net_device
*dev
, u32 value
)
3013 struct tun_struct
*tun
= netdev_priv(dev
);
3018 static int tun_get_coalesce(struct net_device
*dev
,
3019 struct ethtool_coalesce
*ec
)
3021 struct tun_struct
*tun
= netdev_priv(dev
);
3023 ec
->rx_max_coalesced_frames
= tun
->rx_batched
;
3028 static int tun_set_coalesce(struct net_device
*dev
,
3029 struct ethtool_coalesce
*ec
)
3031 struct tun_struct
*tun
= netdev_priv(dev
);
3033 if (ec
->rx_max_coalesced_frames
> NAPI_POLL_WEIGHT
)
3034 tun
->rx_batched
= NAPI_POLL_WEIGHT
;
3036 tun
->rx_batched
= ec
->rx_max_coalesced_frames
;
3041 static const struct ethtool_ops tun_ethtool_ops
= {
3042 .get_drvinfo
= tun_get_drvinfo
,
3043 .get_msglevel
= tun_get_msglevel
,
3044 .set_msglevel
= tun_set_msglevel
,
3045 .get_link
= ethtool_op_get_link
,
3046 .get_ts_info
= ethtool_op_get_ts_info
,
3047 .get_coalesce
= tun_get_coalesce
,
3048 .set_coalesce
= tun_set_coalesce
,
3049 .get_link_ksettings
= tun_get_link_ksettings
,
3052 static int tun_queue_resize(struct tun_struct
*tun
)
3054 struct net_device
*dev
= tun
->dev
;
3055 struct tun_file
*tfile
;
3056 struct skb_array
**arrays
;
3057 int n
= tun
->numqueues
+ tun
->numdisabled
;
3060 arrays
= kmalloc_array(n
, sizeof(*arrays
), GFP_KERNEL
);
3064 for (i
= 0; i
< tun
->numqueues
; i
++) {
3065 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
3066 arrays
[i
] = &tfile
->tx_array
;
3068 list_for_each_entry(tfile
, &tun
->disabled
, next
)
3069 arrays
[i
++] = &tfile
->tx_array
;
3071 ret
= skb_array_resize_multiple(arrays
, n
,
3072 dev
->tx_queue_len
, GFP_KERNEL
);
3078 static int tun_device_event(struct notifier_block
*unused
,
3079 unsigned long event
, void *ptr
)
3081 struct net_device
*dev
= netdev_notifier_info_to_dev(ptr
);
3082 struct tun_struct
*tun
= netdev_priv(dev
);
3084 if (dev
->rtnl_link_ops
!= &tun_link_ops
)
3088 case NETDEV_CHANGE_TX_QUEUE_LEN
:
3089 if (tun_queue_resize(tun
))
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");