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>
79 #include <linux/uaccess.h>
81 /* Uncomment to enable debugging */
82 /* #define TUN_DEBUG 1 */
87 #define tun_debug(level, tun, fmt, args...) \
90 netdev_printk(level, tun->dev, fmt, ##args); \
92 #define DBG1(level, fmt, args...) \
95 printk(level fmt, ##args); \
98 #define tun_debug(level, tun, fmt, args...) \
101 netdev_printk(level, tun->dev, fmt, ##args); \
103 #define DBG1(level, fmt, args...) \
106 printk(level fmt, ##args); \
110 #define TUN_HEADROOM 256
111 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
113 /* TUN device flags */
115 /* IFF_ATTACH_QUEUE is never stored in device flags,
116 * overload it to mean fasync when stored there.
118 #define TUN_FASYNC IFF_ATTACH_QUEUE
119 /* High bits in flags field are unused. */
120 #define TUN_VNET_LE 0x80000000
121 #define TUN_VNET_BE 0x40000000
123 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
125 #define GOODCOPY_LEN 128
127 #define FLT_EXACT_COUNT 8
129 unsigned int count
; /* Number of addrs. Zero means disabled */
130 u32 mask
[2]; /* Mask of the hashed addrs */
131 unsigned char addr
[FLT_EXACT_COUNT
][ETH_ALEN
];
134 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
135 * to max number of VCPUs in guest. */
136 #define MAX_TAP_QUEUES 256
137 #define MAX_TAP_FLOWS 4096
139 #define TUN_FLOW_EXPIRE (3 * HZ)
141 struct tun_pcpu_stats
{
146 struct u64_stats_sync syncp
;
152 /* A tun_file connects an open character device to a tuntap netdevice. It
153 * also contains all socket related structures (except sock_fprog and tap_filter)
154 * to serve as one transmit queue for tuntap device. The sock_fprog and
155 * tap_filter were kept in tun_struct since they were used for filtering for the
156 * netdevice not for a specific queue (at least I didn't see the requirement for
160 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
161 * other can only be read while rcu_read_lock or rtnl_lock is held.
165 struct socket socket
;
167 struct tun_struct __rcu
*tun
;
168 struct fasync_struct
*fasync
;
169 /* only used for fasnyc */
173 unsigned int ifindex
;
175 struct list_head next
;
176 struct tun_struct
*detached
;
177 struct skb_array tx_array
;
180 struct tun_flow_entry
{
181 struct hlist_node hash_link
;
183 struct tun_struct
*tun
;
188 unsigned long updated
;
191 #define TUN_NUM_FLOW_ENTRIES 1024
193 /* Since the socket were moved to tun_file, to preserve the behavior of persist
194 * device, socket filter, sndbuf and vnet header size were restore when the
195 * file were attached to a persist device.
198 struct tun_file __rcu
*tfiles
[MAX_TAP_QUEUES
];
199 unsigned int numqueues
;
204 struct net_device
*dev
;
205 netdev_features_t set_features
;
206 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
212 struct tap_filter txflt
;
213 struct sock_fprog fprog
;
214 /* protected by rtnl lock */
215 bool filter_attached
;
220 struct hlist_head flows
[TUN_NUM_FLOW_ENTRIES
];
221 struct timer_list flow_gc_timer
;
222 unsigned long ageing_time
;
223 unsigned int numdisabled
;
224 struct list_head disabled
;
228 struct tun_pcpu_stats __percpu
*pcpu_stats
;
229 struct bpf_prog __rcu
*xdp_prog
;
232 #ifdef CONFIG_TUN_VNET_CROSS_LE
233 static inline bool tun_legacy_is_little_endian(struct tun_struct
*tun
)
235 return tun
->flags
& TUN_VNET_BE
? false :
236 virtio_legacy_is_little_endian();
239 static long tun_get_vnet_be(struct tun_struct
*tun
, int __user
*argp
)
241 int be
= !!(tun
->flags
& TUN_VNET_BE
);
243 if (put_user(be
, argp
))
249 static long tun_set_vnet_be(struct tun_struct
*tun
, int __user
*argp
)
253 if (get_user(be
, argp
))
257 tun
->flags
|= TUN_VNET_BE
;
259 tun
->flags
&= ~TUN_VNET_BE
;
264 static inline bool tun_legacy_is_little_endian(struct tun_struct
*tun
)
266 return virtio_legacy_is_little_endian();
269 static long tun_get_vnet_be(struct tun_struct
*tun
, int __user
*argp
)
274 static long tun_set_vnet_be(struct tun_struct
*tun
, int __user
*argp
)
278 #endif /* CONFIG_TUN_VNET_CROSS_LE */
280 static inline bool tun_is_little_endian(struct tun_struct
*tun
)
282 return tun
->flags
& TUN_VNET_LE
||
283 tun_legacy_is_little_endian(tun
);
286 static inline u16
tun16_to_cpu(struct tun_struct
*tun
, __virtio16 val
)
288 return __virtio16_to_cpu(tun_is_little_endian(tun
), val
);
291 static inline __virtio16
cpu_to_tun16(struct tun_struct
*tun
, u16 val
)
293 return __cpu_to_virtio16(tun_is_little_endian(tun
), val
);
296 static inline u32
tun_hashfn(u32 rxhash
)
298 return rxhash
& 0x3ff;
301 static struct tun_flow_entry
*tun_flow_find(struct hlist_head
*head
, u32 rxhash
)
303 struct tun_flow_entry
*e
;
305 hlist_for_each_entry_rcu(e
, head
, hash_link
) {
306 if (e
->rxhash
== rxhash
)
312 static struct tun_flow_entry
*tun_flow_create(struct tun_struct
*tun
,
313 struct hlist_head
*head
,
314 u32 rxhash
, u16 queue_index
)
316 struct tun_flow_entry
*e
= kmalloc(sizeof(*e
), GFP_ATOMIC
);
319 tun_debug(KERN_INFO
, tun
, "create flow: hash %u index %u\n",
320 rxhash
, queue_index
);
321 e
->updated
= jiffies
;
324 e
->queue_index
= queue_index
;
326 hlist_add_head_rcu(&e
->hash_link
, head
);
332 static void tun_flow_delete(struct tun_struct
*tun
, struct tun_flow_entry
*e
)
334 tun_debug(KERN_INFO
, tun
, "delete flow: hash %u index %u\n",
335 e
->rxhash
, e
->queue_index
);
336 hlist_del_rcu(&e
->hash_link
);
341 static void tun_flow_flush(struct tun_struct
*tun
)
345 spin_lock_bh(&tun
->lock
);
346 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
347 struct tun_flow_entry
*e
;
348 struct hlist_node
*n
;
350 hlist_for_each_entry_safe(e
, n
, &tun
->flows
[i
], hash_link
)
351 tun_flow_delete(tun
, e
);
353 spin_unlock_bh(&tun
->lock
);
356 static void tun_flow_delete_by_queue(struct tun_struct
*tun
, u16 queue_index
)
360 spin_lock_bh(&tun
->lock
);
361 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
362 struct tun_flow_entry
*e
;
363 struct hlist_node
*n
;
365 hlist_for_each_entry_safe(e
, n
, &tun
->flows
[i
], hash_link
) {
366 if (e
->queue_index
== queue_index
)
367 tun_flow_delete(tun
, e
);
370 spin_unlock_bh(&tun
->lock
);
373 static void tun_flow_cleanup(unsigned long data
)
375 struct tun_struct
*tun
= (struct tun_struct
*)data
;
376 unsigned long delay
= tun
->ageing_time
;
377 unsigned long next_timer
= jiffies
+ delay
;
378 unsigned long count
= 0;
381 tun_debug(KERN_INFO
, tun
, "tun_flow_cleanup\n");
383 spin_lock_bh(&tun
->lock
);
384 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
385 struct tun_flow_entry
*e
;
386 struct hlist_node
*n
;
388 hlist_for_each_entry_safe(e
, n
, &tun
->flows
[i
], hash_link
) {
389 unsigned long this_timer
;
391 this_timer
= e
->updated
+ delay
;
392 if (time_before_eq(this_timer
, jiffies
))
393 tun_flow_delete(tun
, e
);
394 else if (time_before(this_timer
, next_timer
))
395 next_timer
= this_timer
;
400 mod_timer(&tun
->flow_gc_timer
, round_jiffies_up(next_timer
));
401 spin_unlock_bh(&tun
->lock
);
404 static void tun_flow_update(struct tun_struct
*tun
, u32 rxhash
,
405 struct tun_file
*tfile
)
407 struct hlist_head
*head
;
408 struct tun_flow_entry
*e
;
409 unsigned long delay
= tun
->ageing_time
;
410 u16 queue_index
= tfile
->queue_index
;
415 head
= &tun
->flows
[tun_hashfn(rxhash
)];
419 /* We may get a very small possibility of OOO during switching, not
420 * worth to optimize.*/
421 if (tun
->numqueues
== 1 || tfile
->detached
)
424 e
= tun_flow_find(head
, rxhash
);
426 /* TODO: keep queueing to old queue until it's empty? */
427 e
->queue_index
= queue_index
;
428 e
->updated
= jiffies
;
429 sock_rps_record_flow_hash(e
->rps_rxhash
);
431 spin_lock_bh(&tun
->lock
);
432 if (!tun_flow_find(head
, rxhash
) &&
433 tun
->flow_count
< MAX_TAP_FLOWS
)
434 tun_flow_create(tun
, head
, rxhash
, queue_index
);
436 if (!timer_pending(&tun
->flow_gc_timer
))
437 mod_timer(&tun
->flow_gc_timer
,
438 round_jiffies_up(jiffies
+ delay
));
439 spin_unlock_bh(&tun
->lock
);
447 * Save the hash received in the stack receive path and update the
448 * flow_hash table accordingly.
450 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry
*e
, u32 hash
)
452 if (unlikely(e
->rps_rxhash
!= hash
))
453 e
->rps_rxhash
= hash
;
456 /* We try to identify a flow through its rxhash first. The reason that
457 * we do not check rxq no. is because some cards(e.g 82599), chooses
458 * the rxq based on the txq where the last packet of the flow comes. As
459 * the userspace application move between processors, we may get a
460 * different rxq no. here. If we could not get rxhash, then we would
461 * hope the rxq no. may help here.
463 static u16
tun_select_queue(struct net_device
*dev
, struct sk_buff
*skb
,
464 void *accel_priv
, select_queue_fallback_t fallback
)
466 struct tun_struct
*tun
= netdev_priv(dev
);
467 struct tun_flow_entry
*e
;
472 numqueues
= ACCESS_ONCE(tun
->numqueues
);
474 txq
= __skb_get_hash_symmetric(skb
);
476 e
= tun_flow_find(&tun
->flows
[tun_hashfn(txq
)], txq
);
478 tun_flow_save_rps_rxhash(e
, txq
);
479 txq
= e
->queue_index
;
481 /* use multiply and shift instead of expensive divide */
482 txq
= ((u64
)txq
* numqueues
) >> 32;
483 } else if (likely(skb_rx_queue_recorded(skb
))) {
484 txq
= skb_get_rx_queue(skb
);
485 while (unlikely(txq
>= numqueues
))
493 static inline bool tun_not_capable(struct tun_struct
*tun
)
495 const struct cred
*cred
= current_cred();
496 struct net
*net
= dev_net(tun
->dev
);
498 return ((uid_valid(tun
->owner
) && !uid_eq(cred
->euid
, tun
->owner
)) ||
499 (gid_valid(tun
->group
) && !in_egroup_p(tun
->group
))) &&
500 !ns_capable(net
->user_ns
, CAP_NET_ADMIN
);
503 static void tun_set_real_num_queues(struct tun_struct
*tun
)
505 netif_set_real_num_tx_queues(tun
->dev
, tun
->numqueues
);
506 netif_set_real_num_rx_queues(tun
->dev
, tun
->numqueues
);
509 static void tun_disable_queue(struct tun_struct
*tun
, struct tun_file
*tfile
)
511 tfile
->detached
= tun
;
512 list_add_tail(&tfile
->next
, &tun
->disabled
);
516 static struct tun_struct
*tun_enable_queue(struct tun_file
*tfile
)
518 struct tun_struct
*tun
= tfile
->detached
;
520 tfile
->detached
= NULL
;
521 list_del_init(&tfile
->next
);
526 static void tun_queue_purge(struct tun_file
*tfile
)
530 while ((skb
= skb_array_consume(&tfile
->tx_array
)) != NULL
)
533 skb_queue_purge(&tfile
->sk
.sk_write_queue
);
534 skb_queue_purge(&tfile
->sk
.sk_error_queue
);
537 static void __tun_detach(struct tun_file
*tfile
, bool clean
)
539 struct tun_file
*ntfile
;
540 struct tun_struct
*tun
;
542 tun
= rtnl_dereference(tfile
->tun
);
544 if (tun
&& !tfile
->detached
) {
545 u16 index
= tfile
->queue_index
;
546 BUG_ON(index
>= tun
->numqueues
);
548 rcu_assign_pointer(tun
->tfiles
[index
],
549 tun
->tfiles
[tun
->numqueues
- 1]);
550 ntfile
= rtnl_dereference(tun
->tfiles
[index
]);
551 ntfile
->queue_index
= index
;
555 RCU_INIT_POINTER(tfile
->tun
, NULL
);
556 sock_put(&tfile
->sk
);
558 tun_disable_queue(tun
, tfile
);
561 tun_flow_delete_by_queue(tun
, tun
->numqueues
+ 1);
562 /* Drop read queue */
563 tun_queue_purge(tfile
);
564 tun_set_real_num_queues(tun
);
565 } else if (tfile
->detached
&& clean
) {
566 tun
= tun_enable_queue(tfile
);
567 sock_put(&tfile
->sk
);
571 if (tun
&& tun
->numqueues
== 0 && tun
->numdisabled
== 0) {
572 netif_carrier_off(tun
->dev
);
574 if (!(tun
->flags
& IFF_PERSIST
) &&
575 tun
->dev
->reg_state
== NETREG_REGISTERED
)
576 unregister_netdevice(tun
->dev
);
579 skb_array_cleanup(&tfile
->tx_array
);
580 sock_put(&tfile
->sk
);
584 static void tun_detach(struct tun_file
*tfile
, bool clean
)
587 __tun_detach(tfile
, clean
);
591 static void tun_detach_all(struct net_device
*dev
)
593 struct tun_struct
*tun
= netdev_priv(dev
);
594 struct bpf_prog
*xdp_prog
= rtnl_dereference(tun
->xdp_prog
);
595 struct tun_file
*tfile
, *tmp
;
596 int i
, n
= tun
->numqueues
;
598 for (i
= 0; i
< n
; i
++) {
599 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
601 tfile
->socket
.sk
->sk_shutdown
= RCV_SHUTDOWN
;
602 tfile
->socket
.sk
->sk_data_ready(tfile
->socket
.sk
);
603 RCU_INIT_POINTER(tfile
->tun
, NULL
);
606 list_for_each_entry(tfile
, &tun
->disabled
, next
) {
607 tfile
->socket
.sk
->sk_shutdown
= RCV_SHUTDOWN
;
608 tfile
->socket
.sk
->sk_data_ready(tfile
->socket
.sk
);
609 RCU_INIT_POINTER(tfile
->tun
, NULL
);
611 BUG_ON(tun
->numqueues
!= 0);
614 for (i
= 0; i
< n
; i
++) {
615 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
616 /* Drop read queue */
617 tun_queue_purge(tfile
);
618 sock_put(&tfile
->sk
);
620 list_for_each_entry_safe(tfile
, tmp
, &tun
->disabled
, next
) {
621 tun_enable_queue(tfile
);
622 tun_queue_purge(tfile
);
623 sock_put(&tfile
->sk
);
625 BUG_ON(tun
->numdisabled
!= 0);
628 bpf_prog_put(xdp_prog
);
630 if (tun
->flags
& IFF_PERSIST
)
631 module_put(THIS_MODULE
);
634 static int tun_attach(struct tun_struct
*tun
, struct file
*file
, bool skip_filter
)
636 struct tun_file
*tfile
= file
->private_data
;
637 struct net_device
*dev
= tun
->dev
;
640 err
= security_tun_dev_attach(tfile
->socket
.sk
, tun
->security
);
645 if (rtnl_dereference(tfile
->tun
) && !tfile
->detached
)
649 if (!(tun
->flags
& IFF_MULTI_QUEUE
) && tun
->numqueues
== 1)
653 if (!tfile
->detached
&&
654 tun
->numqueues
+ tun
->numdisabled
== MAX_TAP_QUEUES
)
659 /* Re-attach the filter to persist device */
660 if (!skip_filter
&& (tun
->filter_attached
== true)) {
661 lock_sock(tfile
->socket
.sk
);
662 err
= sk_attach_filter(&tun
->fprog
, tfile
->socket
.sk
);
663 release_sock(tfile
->socket
.sk
);
668 if (!tfile
->detached
&&
669 skb_array_init(&tfile
->tx_array
, dev
->tx_queue_len
, GFP_KERNEL
)) {
674 tfile
->queue_index
= tun
->numqueues
;
675 tfile
->socket
.sk
->sk_shutdown
&= ~RCV_SHUTDOWN
;
676 rcu_assign_pointer(tfile
->tun
, tun
);
677 rcu_assign_pointer(tun
->tfiles
[tun
->numqueues
], tfile
);
681 tun_enable_queue(tfile
);
683 sock_hold(&tfile
->sk
);
685 tun_set_real_num_queues(tun
);
687 /* device is allowed to go away first, so no need to hold extra
695 static struct tun_struct
*__tun_get(struct tun_file
*tfile
)
697 struct tun_struct
*tun
;
700 tun
= rcu_dereference(tfile
->tun
);
708 static struct tun_struct
*tun_get(struct file
*file
)
710 return __tun_get(file
->private_data
);
713 static void tun_put(struct tun_struct
*tun
)
719 static void addr_hash_set(u32
*mask
, const u8
*addr
)
721 int n
= ether_crc(ETH_ALEN
, addr
) >> 26;
722 mask
[n
>> 5] |= (1 << (n
& 31));
725 static unsigned int addr_hash_test(const u32
*mask
, const u8
*addr
)
727 int n
= ether_crc(ETH_ALEN
, addr
) >> 26;
728 return mask
[n
>> 5] & (1 << (n
& 31));
731 static int update_filter(struct tap_filter
*filter
, void __user
*arg
)
733 struct { u8 u
[ETH_ALEN
]; } *addr
;
734 struct tun_filter uf
;
735 int err
, alen
, n
, nexact
;
737 if (copy_from_user(&uf
, arg
, sizeof(uf
)))
746 alen
= ETH_ALEN
* uf
.count
;
747 addr
= memdup_user(arg
+ sizeof(uf
), alen
);
749 return PTR_ERR(addr
);
751 /* The filter is updated without holding any locks. Which is
752 * perfectly safe. We disable it first and in the worst
753 * case we'll accept a few undesired packets. */
757 /* Use first set of addresses as an exact filter */
758 for (n
= 0; n
< uf
.count
&& n
< FLT_EXACT_COUNT
; n
++)
759 memcpy(filter
->addr
[n
], addr
[n
].u
, ETH_ALEN
);
763 /* Remaining multicast addresses are hashed,
764 * unicast will leave the filter disabled. */
765 memset(filter
->mask
, 0, sizeof(filter
->mask
));
766 for (; n
< uf
.count
; n
++) {
767 if (!is_multicast_ether_addr(addr
[n
].u
)) {
768 err
= 0; /* no filter */
771 addr_hash_set(filter
->mask
, addr
[n
].u
);
774 /* For ALLMULTI just set the mask to all ones.
775 * This overrides the mask populated above. */
776 if ((uf
.flags
& TUN_FLT_ALLMULTI
))
777 memset(filter
->mask
, ~0, sizeof(filter
->mask
));
779 /* Now enable the filter */
781 filter
->count
= nexact
;
783 /* Return the number of exact filters */
790 /* Returns: 0 - drop, !=0 - accept */
791 static int run_filter(struct tap_filter
*filter
, const struct sk_buff
*skb
)
793 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
795 struct ethhdr
*eh
= (struct ethhdr
*) skb
->data
;
799 for (i
= 0; i
< filter
->count
; i
++)
800 if (ether_addr_equal(eh
->h_dest
, filter
->addr
[i
]))
803 /* Inexact match (multicast only) */
804 if (is_multicast_ether_addr(eh
->h_dest
))
805 return addr_hash_test(filter
->mask
, eh
->h_dest
);
811 * Checks whether the packet is accepted or not.
812 * Returns: 0 - drop, !=0 - accept
814 static int check_filter(struct tap_filter
*filter
, const struct sk_buff
*skb
)
819 return run_filter(filter
, skb
);
822 /* Network device part of the driver */
824 static const struct ethtool_ops tun_ethtool_ops
;
826 /* Net device detach from fd. */
827 static void tun_net_uninit(struct net_device
*dev
)
832 /* Net device open. */
833 static int tun_net_open(struct net_device
*dev
)
835 struct tun_struct
*tun
= netdev_priv(dev
);
838 netif_tx_start_all_queues(dev
);
840 for (i
= 0; i
< tun
->numqueues
; i
++) {
841 struct tun_file
*tfile
;
843 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
844 tfile
->socket
.sk
->sk_write_space(tfile
->socket
.sk
);
850 /* Net device close. */
851 static int tun_net_close(struct net_device
*dev
)
853 netif_tx_stop_all_queues(dev
);
857 /* Net device start xmit */
858 static netdev_tx_t
tun_net_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
860 struct tun_struct
*tun
= netdev_priv(dev
);
861 int txq
= skb
->queue_mapping
;
862 struct tun_file
*tfile
;
866 tfile
= rcu_dereference(tun
->tfiles
[txq
]);
867 numqueues
= ACCESS_ONCE(tun
->numqueues
);
869 /* Drop packet if interface is not attached */
870 if (txq
>= numqueues
)
874 if (numqueues
== 1 && static_key_false(&rps_needed
)) {
875 /* Select queue was not called for the skbuff, so we extract the
876 * RPS hash and save it into the flow_table here.
880 rxhash
= __skb_get_hash_symmetric(skb
);
882 struct tun_flow_entry
*e
;
883 e
= tun_flow_find(&tun
->flows
[tun_hashfn(rxhash
)],
886 tun_flow_save_rps_rxhash(e
, rxhash
);
891 tun_debug(KERN_INFO
, tun
, "tun_net_xmit %d\n", skb
->len
);
895 /* Drop if the filter does not like it.
896 * This is a noop if the filter is disabled.
897 * Filter can be enabled only for the TAP devices. */
898 if (!check_filter(&tun
->txflt
, skb
))
901 if (tfile
->socket
.sk
->sk_filter
&&
902 sk_filter(tfile
->socket
.sk
, skb
))
905 if (unlikely(skb_orphan_frags_rx(skb
, GFP_ATOMIC
)))
908 skb_tx_timestamp(skb
);
910 /* Orphan the skb - required as we might hang on to it
911 * for indefinite time.
917 if (skb_array_produce(&tfile
->tx_array
, skb
))
920 /* Notify and wake up reader process */
921 if (tfile
->flags
& TUN_FASYNC
)
922 kill_fasync(&tfile
->fasync
, SIGIO
, POLL_IN
);
923 tfile
->socket
.sk
->sk_data_ready(tfile
->socket
.sk
);
929 this_cpu_inc(tun
->pcpu_stats
->tx_dropped
);
933 return NET_XMIT_DROP
;
936 static void tun_net_mclist(struct net_device
*dev
)
939 * This callback is supposed to deal with mc filter in
940 * _rx_ path and has nothing to do with the _tx_ path.
941 * In rx path we always accept everything userspace gives us.
945 static netdev_features_t
tun_net_fix_features(struct net_device
*dev
,
946 netdev_features_t features
)
948 struct tun_struct
*tun
= netdev_priv(dev
);
950 return (features
& tun
->set_features
) | (features
& ~TUN_USER_FEATURES
);
952 #ifdef CONFIG_NET_POLL_CONTROLLER
953 static void tun_poll_controller(struct net_device
*dev
)
956 * Tun only receives frames when:
957 * 1) the char device endpoint gets data from user space
958 * 2) the tun socket gets a sendmsg call from user space
959 * Since both of those are synchronous operations, we are guaranteed
960 * never to have pending data when we poll for it
961 * so there is nothing to do here but return.
962 * We need this though so netpoll recognizes us as an interface that
963 * supports polling, which enables bridge devices in virt setups to
964 * still use netconsole
970 static void tun_set_headroom(struct net_device
*dev
, int new_hr
)
972 struct tun_struct
*tun
= netdev_priv(dev
);
974 if (new_hr
< NET_SKB_PAD
)
975 new_hr
= NET_SKB_PAD
;
981 tun_net_get_stats64(struct net_device
*dev
, struct rtnl_link_stats64
*stats
)
983 u32 rx_dropped
= 0, tx_dropped
= 0, rx_frame_errors
= 0;
984 struct tun_struct
*tun
= netdev_priv(dev
);
985 struct tun_pcpu_stats
*p
;
988 for_each_possible_cpu(i
) {
989 u64 rxpackets
, rxbytes
, txpackets
, txbytes
;
992 p
= per_cpu_ptr(tun
->pcpu_stats
, i
);
994 start
= u64_stats_fetch_begin(&p
->syncp
);
995 rxpackets
= p
->rx_packets
;
996 rxbytes
= p
->rx_bytes
;
997 txpackets
= p
->tx_packets
;
998 txbytes
= p
->tx_bytes
;
999 } while (u64_stats_fetch_retry(&p
->syncp
, start
));
1001 stats
->rx_packets
+= rxpackets
;
1002 stats
->rx_bytes
+= rxbytes
;
1003 stats
->tx_packets
+= txpackets
;
1004 stats
->tx_bytes
+= txbytes
;
1007 rx_dropped
+= p
->rx_dropped
;
1008 rx_frame_errors
+= p
->rx_frame_errors
;
1009 tx_dropped
+= p
->tx_dropped
;
1011 stats
->rx_dropped
= rx_dropped
;
1012 stats
->rx_frame_errors
= rx_frame_errors
;
1013 stats
->tx_dropped
= tx_dropped
;
1016 static int tun_xdp_set(struct net_device
*dev
, struct bpf_prog
*prog
,
1017 struct netlink_ext_ack
*extack
)
1019 struct tun_struct
*tun
= netdev_priv(dev
);
1020 struct bpf_prog
*old_prog
;
1022 old_prog
= rtnl_dereference(tun
->xdp_prog
);
1023 rcu_assign_pointer(tun
->xdp_prog
, prog
);
1025 bpf_prog_put(old_prog
);
1030 static u32
tun_xdp_query(struct net_device
*dev
)
1032 struct tun_struct
*tun
= netdev_priv(dev
);
1033 const struct bpf_prog
*xdp_prog
;
1035 xdp_prog
= rtnl_dereference(tun
->xdp_prog
);
1037 return xdp_prog
->aux
->id
;
1042 static int tun_xdp(struct net_device
*dev
, struct netdev_xdp
*xdp
)
1044 switch (xdp
->command
) {
1045 case XDP_SETUP_PROG
:
1046 return tun_xdp_set(dev
, xdp
->prog
, xdp
->extack
);
1047 case XDP_QUERY_PROG
:
1048 xdp
->prog_id
= tun_xdp_query(dev
);
1049 xdp
->prog_attached
= !!xdp
->prog_id
;
1056 static const struct net_device_ops tun_netdev_ops
= {
1057 .ndo_uninit
= tun_net_uninit
,
1058 .ndo_open
= tun_net_open
,
1059 .ndo_stop
= tun_net_close
,
1060 .ndo_start_xmit
= tun_net_xmit
,
1061 .ndo_fix_features
= tun_net_fix_features
,
1062 .ndo_select_queue
= tun_select_queue
,
1063 #ifdef CONFIG_NET_POLL_CONTROLLER
1064 .ndo_poll_controller
= tun_poll_controller
,
1066 .ndo_set_rx_headroom
= tun_set_headroom
,
1067 .ndo_get_stats64
= tun_net_get_stats64
,
1070 static const struct net_device_ops tap_netdev_ops
= {
1071 .ndo_uninit
= tun_net_uninit
,
1072 .ndo_open
= tun_net_open
,
1073 .ndo_stop
= tun_net_close
,
1074 .ndo_start_xmit
= tun_net_xmit
,
1075 .ndo_fix_features
= tun_net_fix_features
,
1076 .ndo_set_rx_mode
= tun_net_mclist
,
1077 .ndo_set_mac_address
= eth_mac_addr
,
1078 .ndo_validate_addr
= eth_validate_addr
,
1079 .ndo_select_queue
= tun_select_queue
,
1080 #ifdef CONFIG_NET_POLL_CONTROLLER
1081 .ndo_poll_controller
= tun_poll_controller
,
1083 .ndo_features_check
= passthru_features_check
,
1084 .ndo_set_rx_headroom
= tun_set_headroom
,
1085 .ndo_get_stats64
= tun_net_get_stats64
,
1089 static void tun_flow_init(struct tun_struct
*tun
)
1093 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++)
1094 INIT_HLIST_HEAD(&tun
->flows
[i
]);
1096 tun
->ageing_time
= TUN_FLOW_EXPIRE
;
1097 setup_timer(&tun
->flow_gc_timer
, tun_flow_cleanup
, (unsigned long)tun
);
1098 mod_timer(&tun
->flow_gc_timer
,
1099 round_jiffies_up(jiffies
+ tun
->ageing_time
));
1102 static void tun_flow_uninit(struct tun_struct
*tun
)
1104 del_timer_sync(&tun
->flow_gc_timer
);
1105 tun_flow_flush(tun
);
1109 #define MAX_MTU 65535
1111 /* Initialize net device. */
1112 static void tun_net_init(struct net_device
*dev
)
1114 struct tun_struct
*tun
= netdev_priv(dev
);
1116 switch (tun
->flags
& TUN_TYPE_MASK
) {
1118 dev
->netdev_ops
= &tun_netdev_ops
;
1120 /* Point-to-Point TUN Device */
1121 dev
->hard_header_len
= 0;
1125 /* Zero header length */
1126 dev
->type
= ARPHRD_NONE
;
1127 dev
->flags
= IFF_POINTOPOINT
| IFF_NOARP
| IFF_MULTICAST
;
1131 dev
->netdev_ops
= &tap_netdev_ops
;
1132 /* Ethernet TAP Device */
1134 dev
->priv_flags
&= ~IFF_TX_SKB_SHARING
;
1135 dev
->priv_flags
|= IFF_LIVE_ADDR_CHANGE
;
1137 eth_hw_addr_random(dev
);
1142 dev
->min_mtu
= MIN_MTU
;
1143 dev
->max_mtu
= MAX_MTU
- dev
->hard_header_len
;
1146 /* Character device part */
1149 static unsigned int tun_chr_poll(struct file
*file
, poll_table
*wait
)
1151 struct tun_file
*tfile
= file
->private_data
;
1152 struct tun_struct
*tun
= __tun_get(tfile
);
1154 unsigned int mask
= 0;
1159 sk
= tfile
->socket
.sk
;
1161 tun_debug(KERN_INFO
, tun
, "tun_chr_poll\n");
1163 poll_wait(file
, sk_sleep(sk
), wait
);
1165 if (!skb_array_empty(&tfile
->tx_array
))
1166 mask
|= POLLIN
| POLLRDNORM
;
1168 if (tun
->dev
->flags
& IFF_UP
&&
1169 (sock_writeable(sk
) ||
1170 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
) &&
1171 sock_writeable(sk
))))
1172 mask
|= POLLOUT
| POLLWRNORM
;
1174 if (tun
->dev
->reg_state
!= NETREG_REGISTERED
)
1181 /* prepad is the amount to reserve at front. len is length after that.
1182 * linear is a hint as to how much to copy (usually headers). */
1183 static struct sk_buff
*tun_alloc_skb(struct tun_file
*tfile
,
1184 size_t prepad
, size_t len
,
1185 size_t linear
, int noblock
)
1187 struct sock
*sk
= tfile
->socket
.sk
;
1188 struct sk_buff
*skb
;
1191 /* Under a page? Don't bother with paged skb. */
1192 if (prepad
+ len
< PAGE_SIZE
|| !linear
)
1195 skb
= sock_alloc_send_pskb(sk
, prepad
+ linear
, len
- linear
, noblock
,
1198 return ERR_PTR(err
);
1200 skb_reserve(skb
, prepad
);
1201 skb_put(skb
, linear
);
1202 skb
->data_len
= len
- linear
;
1203 skb
->len
+= len
- linear
;
1208 static void tun_rx_batched(struct tun_struct
*tun
, struct tun_file
*tfile
,
1209 struct sk_buff
*skb
, int more
)
1211 struct sk_buff_head
*queue
= &tfile
->sk
.sk_write_queue
;
1212 struct sk_buff_head process_queue
;
1213 u32 rx_batched
= tun
->rx_batched
;
1216 if (!rx_batched
|| (!more
&& skb_queue_empty(queue
))) {
1218 netif_receive_skb(skb
);
1223 spin_lock(&queue
->lock
);
1224 if (!more
|| skb_queue_len(queue
) == rx_batched
) {
1225 __skb_queue_head_init(&process_queue
);
1226 skb_queue_splice_tail_init(queue
, &process_queue
);
1229 __skb_queue_tail(queue
, skb
);
1231 spin_unlock(&queue
->lock
);
1234 struct sk_buff
*nskb
;
1237 while ((nskb
= __skb_dequeue(&process_queue
)))
1238 netif_receive_skb(nskb
);
1239 netif_receive_skb(skb
);
1244 static bool tun_can_build_skb(struct tun_struct
*tun
, struct tun_file
*tfile
,
1245 int len
, int noblock
, bool zerocopy
)
1247 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
1250 if (tfile
->socket
.sk
->sk_sndbuf
!= INT_MAX
)
1259 if (SKB_DATA_ALIGN(len
+ TUN_RX_PAD
) +
1260 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
)) > PAGE_SIZE
)
1266 static struct sk_buff
*tun_build_skb(struct tun_struct
*tun
,
1267 struct tun_file
*tfile
,
1268 struct iov_iter
*from
,
1269 struct virtio_net_hdr
*hdr
,
1270 int len
, int *skb_xdp
)
1272 struct page_frag
*alloc_frag
= ¤t
->task_frag
;
1273 struct sk_buff
*skb
;
1274 struct bpf_prog
*xdp_prog
;
1275 int buflen
= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
1276 unsigned int delta
= 0;
1279 bool xdp_xmit
= false;
1280 int err
, pad
= TUN_RX_PAD
;
1283 xdp_prog
= rcu_dereference(tun
->xdp_prog
);
1285 pad
+= TUN_HEADROOM
;
1286 buflen
+= SKB_DATA_ALIGN(len
+ pad
);
1289 alloc_frag
->offset
= ALIGN((u64
)alloc_frag
->offset
, SMP_CACHE_BYTES
);
1290 if (unlikely(!skb_page_frag_refill(buflen
, alloc_frag
, GFP_KERNEL
)))
1291 return ERR_PTR(-ENOMEM
);
1293 buf
= (char *)page_address(alloc_frag
->page
) + alloc_frag
->offset
;
1294 copied
= copy_page_from_iter(alloc_frag
->page
,
1295 alloc_frag
->offset
+ pad
,
1298 return ERR_PTR(-EFAULT
);
1300 /* There's a small window that XDP may be set after the check
1301 * of xdp_prog above, this should be rare and for simplicity
1302 * we do XDP on skb in case the headroom is not enough.
1304 if (hdr
->gso_type
|| !xdp_prog
)
1310 xdp_prog
= rcu_dereference(tun
->xdp_prog
);
1311 if (xdp_prog
&& !*skb_xdp
) {
1312 struct xdp_buff xdp
;
1316 xdp
.data_hard_start
= buf
;
1317 xdp
.data
= buf
+ pad
;
1318 xdp
.data_end
= xdp
.data
+ len
;
1319 orig_data
= xdp
.data
;
1320 act
= bpf_prog_run_xdp(xdp_prog
, &xdp
);
1324 get_page(alloc_frag
->page
);
1325 alloc_frag
->offset
+= buflen
;
1326 err
= xdp_do_redirect(tun
->dev
, &xdp
, xdp_prog
);
1335 delta
= orig_data
- xdp
.data
;
1338 bpf_warn_invalid_xdp_action(act
);
1341 trace_xdp_exception(tun
->dev
, xdp_prog
, act
);
1348 skb
= build_skb(buf
, buflen
);
1351 return ERR_PTR(-ENOMEM
);
1354 skb_reserve(skb
, pad
- delta
);
1355 skb_put(skb
, len
+ delta
);
1356 get_page(alloc_frag
->page
);
1357 alloc_frag
->offset
+= buflen
;
1360 skb
->dev
= tun
->dev
;
1361 generic_xdp_tx(skb
, xdp_prog
);
1371 put_page(alloc_frag
->page
);
1374 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1378 /* Get packet from user space buffer */
1379 static ssize_t
tun_get_user(struct tun_struct
*tun
, struct tun_file
*tfile
,
1380 void *msg_control
, struct iov_iter
*from
,
1381 int noblock
, bool more
)
1383 struct tun_pi pi
= { 0, cpu_to_be16(ETH_P_IP
) };
1384 struct sk_buff
*skb
;
1385 size_t total_len
= iov_iter_count(from
);
1386 size_t len
= total_len
, align
= tun
->align
, linear
;
1387 struct virtio_net_hdr gso
= { 0 };
1388 struct tun_pcpu_stats
*stats
;
1391 bool zerocopy
= false;
1396 if (!(tun
->dev
->flags
& IFF_UP
))
1399 if (!(tun
->flags
& IFF_NO_PI
)) {
1400 if (len
< sizeof(pi
))
1404 if (!copy_from_iter_full(&pi
, sizeof(pi
), from
))
1408 if (tun
->flags
& IFF_VNET_HDR
) {
1409 int vnet_hdr_sz
= READ_ONCE(tun
->vnet_hdr_sz
);
1411 if (len
< vnet_hdr_sz
)
1415 if (!copy_from_iter_full(&gso
, sizeof(gso
), from
))
1418 if ((gso
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) &&
1419 tun16_to_cpu(tun
, gso
.csum_start
) + tun16_to_cpu(tun
, gso
.csum_offset
) + 2 > tun16_to_cpu(tun
, gso
.hdr_len
))
1420 gso
.hdr_len
= cpu_to_tun16(tun
, tun16_to_cpu(tun
, gso
.csum_start
) + tun16_to_cpu(tun
, gso
.csum_offset
) + 2);
1422 if (tun16_to_cpu(tun
, gso
.hdr_len
) > len
)
1424 iov_iter_advance(from
, vnet_hdr_sz
- sizeof(gso
));
1427 if ((tun
->flags
& TUN_TYPE_MASK
) == IFF_TAP
) {
1428 align
+= NET_IP_ALIGN
;
1429 if (unlikely(len
< ETH_HLEN
||
1430 (gso
.hdr_len
&& tun16_to_cpu(tun
, gso
.hdr_len
) < ETH_HLEN
)))
1434 good_linear
= SKB_MAX_HEAD(align
);
1437 struct iov_iter i
= *from
;
1439 /* There are 256 bytes to be copied in skb, so there is
1440 * enough room for skb expand head in case it is used.
1441 * The rest of the buffer is mapped from userspace.
1443 copylen
= gso
.hdr_len
? tun16_to_cpu(tun
, gso
.hdr_len
) : GOODCOPY_LEN
;
1444 if (copylen
> good_linear
)
1445 copylen
= good_linear
;
1447 iov_iter_advance(&i
, copylen
);
1448 if (iov_iter_npages(&i
, INT_MAX
) <= MAX_SKB_FRAGS
)
1452 if (tun_can_build_skb(tun
, tfile
, len
, noblock
, zerocopy
)) {
1453 /* For the packet that is not easy to be processed
1454 * (e.g gso or jumbo packet), we will do it at after
1455 * skb was created with generic XDP routine.
1457 skb
= tun_build_skb(tun
, tfile
, from
, &gso
, len
, &skb_xdp
);
1459 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1460 return PTR_ERR(skb
);
1467 if (tun16_to_cpu(tun
, gso
.hdr_len
) > good_linear
)
1468 linear
= good_linear
;
1470 linear
= tun16_to_cpu(tun
, gso
.hdr_len
);
1473 skb
= tun_alloc_skb(tfile
, align
, copylen
, linear
, noblock
);
1475 if (PTR_ERR(skb
) != -EAGAIN
)
1476 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1477 return PTR_ERR(skb
);
1481 err
= zerocopy_sg_from_iter(skb
, from
);
1483 err
= skb_copy_datagram_from_iter(skb
, 0, from
, len
);
1486 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1492 if (virtio_net_hdr_to_skb(skb
, &gso
, tun_is_little_endian(tun
))) {
1493 this_cpu_inc(tun
->pcpu_stats
->rx_frame_errors
);
1498 switch (tun
->flags
& TUN_TYPE_MASK
) {
1500 if (tun
->flags
& IFF_NO_PI
) {
1501 u8 ip_version
= skb
->len
? (skb
->data
[0] >> 4) : 0;
1503 switch (ip_version
) {
1505 pi
.proto
= htons(ETH_P_IP
);
1508 pi
.proto
= htons(ETH_P_IPV6
);
1511 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1517 skb_reset_mac_header(skb
);
1518 skb
->protocol
= pi
.proto
;
1519 skb
->dev
= tun
->dev
;
1522 skb
->protocol
= eth_type_trans(skb
, tun
->dev
);
1526 /* copy skb_ubuf_info for callback when skb has no error */
1528 skb_shinfo(skb
)->destructor_arg
= msg_control
;
1529 skb_shinfo(skb
)->tx_flags
|= SKBTX_DEV_ZEROCOPY
;
1530 skb_shinfo(skb
)->tx_flags
|= SKBTX_SHARED_FRAG
;
1531 } else if (msg_control
) {
1532 struct ubuf_info
*uarg
= msg_control
;
1533 uarg
->callback(uarg
, false);
1536 skb_reset_network_header(skb
);
1537 skb_probe_transport_header(skb
, 0);
1540 struct bpf_prog
*xdp_prog
;
1544 xdp_prog
= rcu_dereference(tun
->xdp_prog
);
1546 ret
= do_xdp_generic(xdp_prog
, skb
);
1547 if (ret
!= XDP_PASS
) {
1555 rxhash
= __skb_get_hash_symmetric(skb
);
1556 #ifndef CONFIG_4KSTACKS
1557 tun_rx_batched(tun
, tfile
, skb
, more
);
1562 stats
= get_cpu_ptr(tun
->pcpu_stats
);
1563 u64_stats_update_begin(&stats
->syncp
);
1564 stats
->rx_packets
++;
1565 stats
->rx_bytes
+= len
;
1566 u64_stats_update_end(&stats
->syncp
);
1569 tun_flow_update(tun
, rxhash
, tfile
);
1573 static ssize_t
tun_chr_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
1575 struct file
*file
= iocb
->ki_filp
;
1576 struct tun_struct
*tun
= tun_get(file
);
1577 struct tun_file
*tfile
= file
->private_data
;
1583 result
= tun_get_user(tun
, tfile
, NULL
, from
,
1584 file
->f_flags
& O_NONBLOCK
, false);
1590 /* Put packet to the user space buffer */
1591 static ssize_t
tun_put_user(struct tun_struct
*tun
,
1592 struct tun_file
*tfile
,
1593 struct sk_buff
*skb
,
1594 struct iov_iter
*iter
)
1596 struct tun_pi pi
= { 0, skb
->protocol
};
1597 struct tun_pcpu_stats
*stats
;
1599 int vlan_offset
= 0;
1601 int vnet_hdr_sz
= 0;
1603 if (skb_vlan_tag_present(skb
))
1604 vlan_hlen
= VLAN_HLEN
;
1606 if (tun
->flags
& IFF_VNET_HDR
)
1607 vnet_hdr_sz
= READ_ONCE(tun
->vnet_hdr_sz
);
1609 total
= skb
->len
+ vlan_hlen
+ vnet_hdr_sz
;
1611 if (!(tun
->flags
& IFF_NO_PI
)) {
1612 if (iov_iter_count(iter
) < sizeof(pi
))
1615 total
+= sizeof(pi
);
1616 if (iov_iter_count(iter
) < total
) {
1617 /* Packet will be striped */
1618 pi
.flags
|= TUN_PKT_STRIP
;
1621 if (copy_to_iter(&pi
, sizeof(pi
), iter
) != sizeof(pi
))
1626 struct virtio_net_hdr gso
;
1628 if (iov_iter_count(iter
) < vnet_hdr_sz
)
1631 if (virtio_net_hdr_from_skb(skb
, &gso
,
1632 tun_is_little_endian(tun
), true)) {
1633 struct skb_shared_info
*sinfo
= skb_shinfo(skb
);
1634 pr_err("unexpected GSO type: "
1635 "0x%x, gso_size %d, hdr_len %d\n",
1636 sinfo
->gso_type
, tun16_to_cpu(tun
, gso
.gso_size
),
1637 tun16_to_cpu(tun
, gso
.hdr_len
));
1638 print_hex_dump(KERN_ERR
, "tun: ",
1641 min((int)tun16_to_cpu(tun
, gso
.hdr_len
), 64), true);
1646 if (copy_to_iter(&gso
, sizeof(gso
), iter
) != sizeof(gso
))
1649 iov_iter_advance(iter
, vnet_hdr_sz
- sizeof(gso
));
1655 __be16 h_vlan_proto
;
1659 veth
.h_vlan_proto
= skb
->vlan_proto
;
1660 veth
.h_vlan_TCI
= htons(skb_vlan_tag_get(skb
));
1662 vlan_offset
= offsetof(struct vlan_ethhdr
, h_vlan_proto
);
1664 ret
= skb_copy_datagram_iter(skb
, 0, iter
, vlan_offset
);
1665 if (ret
|| !iov_iter_count(iter
))
1668 ret
= copy_to_iter(&veth
, sizeof(veth
), iter
);
1669 if (ret
!= sizeof(veth
) || !iov_iter_count(iter
))
1673 skb_copy_datagram_iter(skb
, vlan_offset
, iter
, skb
->len
- vlan_offset
);
1676 /* caller is in process context, */
1677 stats
= get_cpu_ptr(tun
->pcpu_stats
);
1678 u64_stats_update_begin(&stats
->syncp
);
1679 stats
->tx_packets
++;
1680 stats
->tx_bytes
+= skb
->len
+ vlan_hlen
;
1681 u64_stats_update_end(&stats
->syncp
);
1682 put_cpu_ptr(tun
->pcpu_stats
);
1687 static struct sk_buff
*tun_ring_recv(struct tun_file
*tfile
, int noblock
,
1690 DECLARE_WAITQUEUE(wait
, current
);
1691 struct sk_buff
*skb
= NULL
;
1694 skb
= skb_array_consume(&tfile
->tx_array
);
1702 add_wait_queue(&tfile
->wq
.wait
, &wait
);
1703 current
->state
= TASK_INTERRUPTIBLE
;
1706 skb
= skb_array_consume(&tfile
->tx_array
);
1709 if (signal_pending(current
)) {
1710 error
= -ERESTARTSYS
;
1713 if (tfile
->socket
.sk
->sk_shutdown
& RCV_SHUTDOWN
) {
1721 current
->state
= TASK_RUNNING
;
1722 remove_wait_queue(&tfile
->wq
.wait
, &wait
);
1729 static ssize_t
tun_do_read(struct tun_struct
*tun
, struct tun_file
*tfile
,
1730 struct iov_iter
*to
,
1731 int noblock
, struct sk_buff
*skb
)
1736 tun_debug(KERN_INFO
, tun
, "tun_do_read\n");
1738 if (!iov_iter_count(to
)) {
1745 /* Read frames from ring */
1746 skb
= tun_ring_recv(tfile
, noblock
, &err
);
1751 ret
= tun_put_user(tun
, tfile
, skb
, to
);
1752 if (unlikely(ret
< 0))
1760 static ssize_t
tun_chr_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
1762 struct file
*file
= iocb
->ki_filp
;
1763 struct tun_file
*tfile
= file
->private_data
;
1764 struct tun_struct
*tun
= __tun_get(tfile
);
1765 ssize_t len
= iov_iter_count(to
), ret
;
1769 ret
= tun_do_read(tun
, tfile
, to
, file
->f_flags
& O_NONBLOCK
, NULL
);
1770 ret
= min_t(ssize_t
, ret
, len
);
1777 static void tun_free_netdev(struct net_device
*dev
)
1779 struct tun_struct
*tun
= netdev_priv(dev
);
1781 BUG_ON(!(list_empty(&tun
->disabled
)));
1782 free_percpu(tun
->pcpu_stats
);
1783 tun_flow_uninit(tun
);
1784 security_tun_dev_free_security(tun
->security
);
1787 static void tun_setup(struct net_device
*dev
)
1789 struct tun_struct
*tun
= netdev_priv(dev
);
1791 tun
->owner
= INVALID_UID
;
1792 tun
->group
= INVALID_GID
;
1794 dev
->ethtool_ops
= &tun_ethtool_ops
;
1795 dev
->needs_free_netdev
= true;
1796 dev
->priv_destructor
= tun_free_netdev
;
1797 /* We prefer our own queue length */
1798 dev
->tx_queue_len
= TUN_READQ_SIZE
;
1801 /* Trivial set of netlink ops to allow deleting tun or tap
1802 * device with netlink.
1804 static int tun_validate(struct nlattr
*tb
[], struct nlattr
*data
[],
1805 struct netlink_ext_ack
*extack
)
1810 static struct rtnl_link_ops tun_link_ops __read_mostly
= {
1812 .priv_size
= sizeof(struct tun_struct
),
1814 .validate
= tun_validate
,
1817 static void tun_sock_write_space(struct sock
*sk
)
1819 struct tun_file
*tfile
;
1820 wait_queue_head_t
*wqueue
;
1822 if (!sock_writeable(sk
))
1825 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
))
1828 wqueue
= sk_sleep(sk
);
1829 if (wqueue
&& waitqueue_active(wqueue
))
1830 wake_up_interruptible_sync_poll(wqueue
, POLLOUT
|
1831 POLLWRNORM
| POLLWRBAND
);
1833 tfile
= container_of(sk
, struct tun_file
, sk
);
1834 kill_fasync(&tfile
->fasync
, SIGIO
, POLL_OUT
);
1837 static int tun_sendmsg(struct socket
*sock
, struct msghdr
*m
, size_t total_len
)
1840 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
1841 struct tun_struct
*tun
= __tun_get(tfile
);
1846 ret
= tun_get_user(tun
, tfile
, m
->msg_control
, &m
->msg_iter
,
1847 m
->msg_flags
& MSG_DONTWAIT
,
1848 m
->msg_flags
& MSG_MORE
);
1853 static int tun_recvmsg(struct socket
*sock
, struct msghdr
*m
, size_t total_len
,
1856 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
1857 struct tun_struct
*tun
= __tun_get(tfile
);
1858 struct sk_buff
*skb
= m
->msg_control
;
1866 if (flags
& ~(MSG_DONTWAIT
|MSG_TRUNC
|MSG_ERRQUEUE
)) {
1870 if (flags
& MSG_ERRQUEUE
) {
1871 ret
= sock_recv_errqueue(sock
->sk
, m
, total_len
,
1872 SOL_PACKET
, TUN_TX_TIMESTAMP
);
1875 ret
= tun_do_read(tun
, tfile
, &m
->msg_iter
, flags
& MSG_DONTWAIT
, skb
);
1876 if (ret
> (ssize_t
)total_len
) {
1877 m
->msg_flags
|= MSG_TRUNC
;
1878 ret
= flags
& MSG_TRUNC
? ret
: total_len
;
1892 static int tun_peek_len(struct socket
*sock
)
1894 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
1895 struct tun_struct
*tun
;
1898 tun
= __tun_get(tfile
);
1902 ret
= skb_array_peek_len(&tfile
->tx_array
);
1908 /* Ops structure to mimic raw sockets with tun */
1909 static const struct proto_ops tun_socket_ops
= {
1910 .peek_len
= tun_peek_len
,
1911 .sendmsg
= tun_sendmsg
,
1912 .recvmsg
= tun_recvmsg
,
1915 static struct proto tun_proto
= {
1917 .owner
= THIS_MODULE
,
1918 .obj_size
= sizeof(struct tun_file
),
1921 static int tun_flags(struct tun_struct
*tun
)
1923 return tun
->flags
& (TUN_FEATURES
| IFF_PERSIST
| IFF_TUN
| IFF_TAP
);
1926 static ssize_t
tun_show_flags(struct device
*dev
, struct device_attribute
*attr
,
1929 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
1930 return sprintf(buf
, "0x%x\n", tun_flags(tun
));
1933 static ssize_t
tun_show_owner(struct device
*dev
, struct device_attribute
*attr
,
1936 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
1937 return uid_valid(tun
->owner
)?
1938 sprintf(buf
, "%u\n",
1939 from_kuid_munged(current_user_ns(), tun
->owner
)):
1940 sprintf(buf
, "-1\n");
1943 static ssize_t
tun_show_group(struct device
*dev
, struct device_attribute
*attr
,
1946 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
1947 return gid_valid(tun
->group
) ?
1948 sprintf(buf
, "%u\n",
1949 from_kgid_munged(current_user_ns(), tun
->group
)):
1950 sprintf(buf
, "-1\n");
1953 static DEVICE_ATTR(tun_flags
, 0444, tun_show_flags
, NULL
);
1954 static DEVICE_ATTR(owner
, 0444, tun_show_owner
, NULL
);
1955 static DEVICE_ATTR(group
, 0444, tun_show_group
, NULL
);
1957 static struct attribute
*tun_dev_attrs
[] = {
1958 &dev_attr_tun_flags
.attr
,
1959 &dev_attr_owner
.attr
,
1960 &dev_attr_group
.attr
,
1964 static const struct attribute_group tun_attr_group
= {
1965 .attrs
= tun_dev_attrs
1968 static int tun_set_iff(struct net
*net
, struct file
*file
, struct ifreq
*ifr
)
1970 struct tun_struct
*tun
;
1971 struct tun_file
*tfile
= file
->private_data
;
1972 struct net_device
*dev
;
1975 if (tfile
->detached
)
1978 dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
1980 if (ifr
->ifr_flags
& IFF_TUN_EXCL
)
1982 if ((ifr
->ifr_flags
& IFF_TUN
) && dev
->netdev_ops
== &tun_netdev_ops
)
1983 tun
= netdev_priv(dev
);
1984 else if ((ifr
->ifr_flags
& IFF_TAP
) && dev
->netdev_ops
== &tap_netdev_ops
)
1985 tun
= netdev_priv(dev
);
1989 if (!!(ifr
->ifr_flags
& IFF_MULTI_QUEUE
) !=
1990 !!(tun
->flags
& IFF_MULTI_QUEUE
))
1993 if (tun_not_capable(tun
))
1995 err
= security_tun_dev_open(tun
->security
);
1999 err
= tun_attach(tun
, file
, ifr
->ifr_flags
& IFF_NOFILTER
);
2003 if (tun
->flags
& IFF_MULTI_QUEUE
&&
2004 (tun
->numqueues
+ tun
->numdisabled
> 1)) {
2005 /* One or more queue has already been attached, no need
2006 * to initialize the device again.
2013 unsigned long flags
= 0;
2014 int queues
= ifr
->ifr_flags
& IFF_MULTI_QUEUE
?
2017 if (!ns_capable(net
->user_ns
, CAP_NET_ADMIN
))
2019 err
= security_tun_dev_create();
2024 if (ifr
->ifr_flags
& IFF_TUN
) {
2028 } else if (ifr
->ifr_flags
& IFF_TAP
) {
2036 name
= ifr
->ifr_name
;
2038 dev
= alloc_netdev_mqs(sizeof(struct tun_struct
), name
,
2039 NET_NAME_UNKNOWN
, tun_setup
, queues
,
2044 err
= dev_get_valid_name(net
, dev
, name
);
2048 dev_net_set(dev
, net
);
2049 dev
->rtnl_link_ops
= &tun_link_ops
;
2050 dev
->ifindex
= tfile
->ifindex
;
2051 dev
->sysfs_groups
[0] = &tun_attr_group
;
2053 tun
= netdev_priv(dev
);
2056 tun
->txflt
.count
= 0;
2057 tun
->vnet_hdr_sz
= sizeof(struct virtio_net_hdr
);
2059 tun
->align
= NET_SKB_PAD
;
2060 tun
->filter_attached
= false;
2061 tun
->sndbuf
= tfile
->socket
.sk
->sk_sndbuf
;
2062 tun
->rx_batched
= 0;
2064 tun
->pcpu_stats
= netdev_alloc_pcpu_stats(struct tun_pcpu_stats
);
2065 if (!tun
->pcpu_stats
) {
2070 spin_lock_init(&tun
->lock
);
2072 err
= security_tun_dev_alloc_security(&tun
->security
);
2079 dev
->hw_features
= NETIF_F_SG
| NETIF_F_FRAGLIST
|
2080 TUN_USER_FEATURES
| NETIF_F_HW_VLAN_CTAG_TX
|
2081 NETIF_F_HW_VLAN_STAG_TX
;
2082 dev
->features
= dev
->hw_features
| NETIF_F_LLTX
;
2083 dev
->vlan_features
= dev
->features
&
2084 ~(NETIF_F_HW_VLAN_CTAG_TX
|
2085 NETIF_F_HW_VLAN_STAG_TX
);
2087 INIT_LIST_HEAD(&tun
->disabled
);
2088 err
= tun_attach(tun
, file
, false);
2092 err
= register_netdevice(tun
->dev
);
2097 netif_carrier_on(tun
->dev
);
2099 tun_debug(KERN_INFO
, tun
, "tun_set_iff\n");
2101 tun
->flags
= (tun
->flags
& ~TUN_FEATURES
) |
2102 (ifr
->ifr_flags
& TUN_FEATURES
);
2104 /* Make sure persistent devices do not get stuck in
2107 if (netif_running(tun
->dev
))
2108 netif_tx_wake_all_queues(tun
->dev
);
2110 strcpy(ifr
->ifr_name
, tun
->dev
->name
);
2114 tun_detach_all(dev
);
2115 /* register_netdevice() already called tun_free_netdev() */
2119 tun_flow_uninit(tun
);
2120 security_tun_dev_free_security(tun
->security
);
2122 free_percpu(tun
->pcpu_stats
);
2128 static void tun_get_iff(struct net
*net
, struct tun_struct
*tun
,
2131 tun_debug(KERN_INFO
, tun
, "tun_get_iff\n");
2133 strcpy(ifr
->ifr_name
, tun
->dev
->name
);
2135 ifr
->ifr_flags
= tun_flags(tun
);
2139 /* This is like a cut-down ethtool ops, except done via tun fd so no
2140 * privs required. */
2141 static int set_offload(struct tun_struct
*tun
, unsigned long arg
)
2143 netdev_features_t features
= 0;
2145 if (arg
& TUN_F_CSUM
) {
2146 features
|= NETIF_F_HW_CSUM
;
2149 if (arg
& (TUN_F_TSO4
|TUN_F_TSO6
)) {
2150 if (arg
& TUN_F_TSO_ECN
) {
2151 features
|= NETIF_F_TSO_ECN
;
2152 arg
&= ~TUN_F_TSO_ECN
;
2154 if (arg
& TUN_F_TSO4
)
2155 features
|= NETIF_F_TSO
;
2156 if (arg
& TUN_F_TSO6
)
2157 features
|= NETIF_F_TSO6
;
2158 arg
&= ~(TUN_F_TSO4
|TUN_F_TSO6
);
2162 /* This gives the user a way to test for new features in future by
2163 * trying to set them. */
2167 tun
->set_features
= features
;
2168 tun
->dev
->wanted_features
&= ~TUN_USER_FEATURES
;
2169 tun
->dev
->wanted_features
|= features
;
2170 netdev_update_features(tun
->dev
);
2175 static void tun_detach_filter(struct tun_struct
*tun
, int n
)
2178 struct tun_file
*tfile
;
2180 for (i
= 0; i
< n
; i
++) {
2181 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
2182 lock_sock(tfile
->socket
.sk
);
2183 sk_detach_filter(tfile
->socket
.sk
);
2184 release_sock(tfile
->socket
.sk
);
2187 tun
->filter_attached
= false;
2190 static int tun_attach_filter(struct tun_struct
*tun
)
2193 struct tun_file
*tfile
;
2195 for (i
= 0; i
< tun
->numqueues
; i
++) {
2196 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
2197 lock_sock(tfile
->socket
.sk
);
2198 ret
= sk_attach_filter(&tun
->fprog
, tfile
->socket
.sk
);
2199 release_sock(tfile
->socket
.sk
);
2201 tun_detach_filter(tun
, i
);
2206 tun
->filter_attached
= true;
2210 static void tun_set_sndbuf(struct tun_struct
*tun
)
2212 struct tun_file
*tfile
;
2215 for (i
= 0; i
< tun
->numqueues
; i
++) {
2216 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
2217 tfile
->socket
.sk
->sk_sndbuf
= tun
->sndbuf
;
2221 static int tun_set_queue(struct file
*file
, struct ifreq
*ifr
)
2223 struct tun_file
*tfile
= file
->private_data
;
2224 struct tun_struct
*tun
;
2229 if (ifr
->ifr_flags
& IFF_ATTACH_QUEUE
) {
2230 tun
= tfile
->detached
;
2235 ret
= security_tun_dev_attach_queue(tun
->security
);
2238 ret
= tun_attach(tun
, file
, false);
2239 } else if (ifr
->ifr_flags
& IFF_DETACH_QUEUE
) {
2240 tun
= rtnl_dereference(tfile
->tun
);
2241 if (!tun
|| !(tun
->flags
& IFF_MULTI_QUEUE
) || tfile
->detached
)
2244 __tun_detach(tfile
, false);
2253 static long __tun_chr_ioctl(struct file
*file
, unsigned int cmd
,
2254 unsigned long arg
, int ifreq_len
)
2256 struct tun_file
*tfile
= file
->private_data
;
2257 struct tun_struct
*tun
;
2258 void __user
* argp
= (void __user
*)arg
;
2264 unsigned int ifindex
;
2268 if (cmd
== TUNSETIFF
|| cmd
== TUNSETQUEUE
|| _IOC_TYPE(cmd
) == SOCK_IOC_TYPE
) {
2269 if (copy_from_user(&ifr
, argp
, ifreq_len
))
2272 memset(&ifr
, 0, sizeof(ifr
));
2274 if (cmd
== TUNGETFEATURES
) {
2275 /* Currently this just means: "what IFF flags are valid?".
2276 * This is needed because we never checked for invalid flags on
2279 return put_user(IFF_TUN
| IFF_TAP
| TUN_FEATURES
,
2280 (unsigned int __user
*)argp
);
2281 } else if (cmd
== TUNSETQUEUE
)
2282 return tun_set_queue(file
, &ifr
);
2287 tun
= __tun_get(tfile
);
2288 if (cmd
== TUNSETIFF
) {
2293 ifr
.ifr_name
[IFNAMSIZ
-1] = '\0';
2295 ret
= tun_set_iff(sock_net(&tfile
->sk
), file
, &ifr
);
2300 if (copy_to_user(argp
, &ifr
, ifreq_len
))
2304 if (cmd
== TUNSETIFINDEX
) {
2310 if (copy_from_user(&ifindex
, argp
, sizeof(ifindex
)))
2314 tfile
->ifindex
= ifindex
;
2322 tun_debug(KERN_INFO
, tun
, "tun_chr_ioctl cmd %u\n", cmd
);
2327 tun_get_iff(current
->nsproxy
->net_ns
, tun
, &ifr
);
2329 if (tfile
->detached
)
2330 ifr
.ifr_flags
|= IFF_DETACH_QUEUE
;
2331 if (!tfile
->socket
.sk
->sk_filter
)
2332 ifr
.ifr_flags
|= IFF_NOFILTER
;
2334 if (copy_to_user(argp
, &ifr
, ifreq_len
))
2339 /* Disable/Enable checksum */
2341 /* [unimplemented] */
2342 tun_debug(KERN_INFO
, tun
, "ignored: set checksum %s\n",
2343 arg
? "disabled" : "enabled");
2347 /* Disable/Enable persist mode. Keep an extra reference to the
2348 * module to prevent the module being unprobed.
2350 if (arg
&& !(tun
->flags
& IFF_PERSIST
)) {
2351 tun
->flags
|= IFF_PERSIST
;
2352 __module_get(THIS_MODULE
);
2354 if (!arg
&& (tun
->flags
& IFF_PERSIST
)) {
2355 tun
->flags
&= ~IFF_PERSIST
;
2356 module_put(THIS_MODULE
);
2359 tun_debug(KERN_INFO
, tun
, "persist %s\n",
2360 arg
? "enabled" : "disabled");
2364 /* Set owner of the device */
2365 owner
= make_kuid(current_user_ns(), arg
);
2366 if (!uid_valid(owner
)) {
2371 tun_debug(KERN_INFO
, tun
, "owner set to %u\n",
2372 from_kuid(&init_user_ns
, tun
->owner
));
2376 /* Set group of the device */
2377 group
= make_kgid(current_user_ns(), arg
);
2378 if (!gid_valid(group
)) {
2383 tun_debug(KERN_INFO
, tun
, "group set to %u\n",
2384 from_kgid(&init_user_ns
, tun
->group
));
2388 /* Only allow setting the type when the interface is down */
2389 if (tun
->dev
->flags
& IFF_UP
) {
2390 tun_debug(KERN_INFO
, tun
,
2391 "Linktype set failed because interface is up\n");
2394 tun
->dev
->type
= (int) arg
;
2395 tun_debug(KERN_INFO
, tun
, "linktype set to %d\n",
2407 ret
= set_offload(tun
, arg
);
2410 case TUNSETTXFILTER
:
2411 /* Can be set only for TAPs */
2413 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2415 ret
= update_filter(&tun
->txflt
, (void __user
*)arg
);
2419 /* Get hw address */
2420 memcpy(ifr
.ifr_hwaddr
.sa_data
, tun
->dev
->dev_addr
, ETH_ALEN
);
2421 ifr
.ifr_hwaddr
.sa_family
= tun
->dev
->type
;
2422 if (copy_to_user(argp
, &ifr
, ifreq_len
))
2427 /* Set hw address */
2428 tun_debug(KERN_DEBUG
, tun
, "set hw address: %pM\n",
2429 ifr
.ifr_hwaddr
.sa_data
);
2431 ret
= dev_set_mac_address(tun
->dev
, &ifr
.ifr_hwaddr
);
2435 sndbuf
= tfile
->socket
.sk
->sk_sndbuf
;
2436 if (copy_to_user(argp
, &sndbuf
, sizeof(sndbuf
)))
2441 if (copy_from_user(&sndbuf
, argp
, sizeof(sndbuf
))) {
2450 tun
->sndbuf
= sndbuf
;
2451 tun_set_sndbuf(tun
);
2454 case TUNGETVNETHDRSZ
:
2455 vnet_hdr_sz
= tun
->vnet_hdr_sz
;
2456 if (copy_to_user(argp
, &vnet_hdr_sz
, sizeof(vnet_hdr_sz
)))
2460 case TUNSETVNETHDRSZ
:
2461 if (copy_from_user(&vnet_hdr_sz
, argp
, sizeof(vnet_hdr_sz
))) {
2465 if (vnet_hdr_sz
< (int)sizeof(struct virtio_net_hdr
)) {
2470 tun
->vnet_hdr_sz
= vnet_hdr_sz
;
2474 le
= !!(tun
->flags
& TUN_VNET_LE
);
2475 if (put_user(le
, (int __user
*)argp
))
2480 if (get_user(le
, (int __user
*)argp
)) {
2485 tun
->flags
|= TUN_VNET_LE
;
2487 tun
->flags
&= ~TUN_VNET_LE
;
2491 ret
= tun_get_vnet_be(tun
, argp
);
2495 ret
= tun_set_vnet_be(tun
, argp
);
2498 case TUNATTACHFILTER
:
2499 /* Can be set only for TAPs */
2501 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2504 if (copy_from_user(&tun
->fprog
, argp
, sizeof(tun
->fprog
)))
2507 ret
= tun_attach_filter(tun
);
2510 case TUNDETACHFILTER
:
2511 /* Can be set only for TAPs */
2513 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2516 tun_detach_filter(tun
, tun
->numqueues
);
2521 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2524 if (copy_to_user(argp
, &tun
->fprog
, sizeof(tun
->fprog
)))
2541 static long tun_chr_ioctl(struct file
*file
,
2542 unsigned int cmd
, unsigned long arg
)
2544 return __tun_chr_ioctl(file
, cmd
, arg
, sizeof (struct ifreq
));
2547 #ifdef CONFIG_COMPAT
2548 static long tun_chr_compat_ioctl(struct file
*file
,
2549 unsigned int cmd
, unsigned long arg
)
2554 case TUNSETTXFILTER
:
2559 arg
= (unsigned long)compat_ptr(arg
);
2562 arg
= (compat_ulong_t
)arg
;
2567 * compat_ifreq is shorter than ifreq, so we must not access beyond
2568 * the end of that structure. All fields that are used in this
2569 * driver are compatible though, we don't need to convert the
2572 return __tun_chr_ioctl(file
, cmd
, arg
, sizeof(struct compat_ifreq
));
2574 #endif /* CONFIG_COMPAT */
2576 static int tun_chr_fasync(int fd
, struct file
*file
, int on
)
2578 struct tun_file
*tfile
= file
->private_data
;
2581 if ((ret
= fasync_helper(fd
, file
, on
, &tfile
->fasync
)) < 0)
2585 __f_setown(file
, task_pid(current
), PIDTYPE_PID
, 0);
2586 tfile
->flags
|= TUN_FASYNC
;
2588 tfile
->flags
&= ~TUN_FASYNC
;
2594 static int tun_chr_open(struct inode
*inode
, struct file
* file
)
2596 struct net
*net
= current
->nsproxy
->net_ns
;
2597 struct tun_file
*tfile
;
2599 DBG1(KERN_INFO
, "tunX: tun_chr_open\n");
2601 tfile
= (struct tun_file
*)sk_alloc(net
, AF_UNSPEC
, GFP_KERNEL
,
2605 RCU_INIT_POINTER(tfile
->tun
, NULL
);
2609 init_waitqueue_head(&tfile
->wq
.wait
);
2610 RCU_INIT_POINTER(tfile
->socket
.wq
, &tfile
->wq
);
2612 tfile
->socket
.file
= file
;
2613 tfile
->socket
.ops
= &tun_socket_ops
;
2615 sock_init_data(&tfile
->socket
, &tfile
->sk
);
2617 tfile
->sk
.sk_write_space
= tun_sock_write_space
;
2618 tfile
->sk
.sk_sndbuf
= INT_MAX
;
2620 file
->private_data
= tfile
;
2621 INIT_LIST_HEAD(&tfile
->next
);
2623 sock_set_flag(&tfile
->sk
, SOCK_ZEROCOPY
);
2628 static int tun_chr_close(struct inode
*inode
, struct file
*file
)
2630 struct tun_file
*tfile
= file
->private_data
;
2632 tun_detach(tfile
, true);
2637 #ifdef CONFIG_PROC_FS
2638 static void tun_chr_show_fdinfo(struct seq_file
*m
, struct file
*f
)
2640 struct tun_struct
*tun
;
2643 memset(&ifr
, 0, sizeof(ifr
));
2648 tun_get_iff(current
->nsproxy
->net_ns
, tun
, &ifr
);
2654 seq_printf(m
, "iff:\t%s\n", ifr
.ifr_name
);
2658 static const struct file_operations tun_fops
= {
2659 .owner
= THIS_MODULE
,
2660 .llseek
= no_llseek
,
2661 .read_iter
= tun_chr_read_iter
,
2662 .write_iter
= tun_chr_write_iter
,
2663 .poll
= tun_chr_poll
,
2664 .unlocked_ioctl
= tun_chr_ioctl
,
2665 #ifdef CONFIG_COMPAT
2666 .compat_ioctl
= tun_chr_compat_ioctl
,
2668 .open
= tun_chr_open
,
2669 .release
= tun_chr_close
,
2670 .fasync
= tun_chr_fasync
,
2671 #ifdef CONFIG_PROC_FS
2672 .show_fdinfo
= tun_chr_show_fdinfo
,
2676 static struct miscdevice tun_miscdev
= {
2679 .nodename
= "net/tun",
2683 /* ethtool interface */
2685 static int tun_get_link_ksettings(struct net_device
*dev
,
2686 struct ethtool_link_ksettings
*cmd
)
2688 ethtool_link_ksettings_zero_link_mode(cmd
, supported
);
2689 ethtool_link_ksettings_zero_link_mode(cmd
, advertising
);
2690 cmd
->base
.speed
= SPEED_10
;
2691 cmd
->base
.duplex
= DUPLEX_FULL
;
2692 cmd
->base
.port
= PORT_TP
;
2693 cmd
->base
.phy_address
= 0;
2694 cmd
->base
.autoneg
= AUTONEG_DISABLE
;
2698 static void tun_get_drvinfo(struct net_device
*dev
, struct ethtool_drvinfo
*info
)
2700 struct tun_struct
*tun
= netdev_priv(dev
);
2702 strlcpy(info
->driver
, DRV_NAME
, sizeof(info
->driver
));
2703 strlcpy(info
->version
, DRV_VERSION
, sizeof(info
->version
));
2705 switch (tun
->flags
& TUN_TYPE_MASK
) {
2707 strlcpy(info
->bus_info
, "tun", sizeof(info
->bus_info
));
2710 strlcpy(info
->bus_info
, "tap", sizeof(info
->bus_info
));
2715 static u32
tun_get_msglevel(struct net_device
*dev
)
2718 struct tun_struct
*tun
= netdev_priv(dev
);
2725 static void tun_set_msglevel(struct net_device
*dev
, u32 value
)
2728 struct tun_struct
*tun
= netdev_priv(dev
);
2733 static int tun_get_coalesce(struct net_device
*dev
,
2734 struct ethtool_coalesce
*ec
)
2736 struct tun_struct
*tun
= netdev_priv(dev
);
2738 ec
->rx_max_coalesced_frames
= tun
->rx_batched
;
2743 static int tun_set_coalesce(struct net_device
*dev
,
2744 struct ethtool_coalesce
*ec
)
2746 struct tun_struct
*tun
= netdev_priv(dev
);
2748 if (ec
->rx_max_coalesced_frames
> NAPI_POLL_WEIGHT
)
2749 tun
->rx_batched
= NAPI_POLL_WEIGHT
;
2751 tun
->rx_batched
= ec
->rx_max_coalesced_frames
;
2756 static const struct ethtool_ops tun_ethtool_ops
= {
2757 .get_drvinfo
= tun_get_drvinfo
,
2758 .get_msglevel
= tun_get_msglevel
,
2759 .set_msglevel
= tun_set_msglevel
,
2760 .get_link
= ethtool_op_get_link
,
2761 .get_ts_info
= ethtool_op_get_ts_info
,
2762 .get_coalesce
= tun_get_coalesce
,
2763 .set_coalesce
= tun_set_coalesce
,
2764 .get_link_ksettings
= tun_get_link_ksettings
,
2767 static int tun_queue_resize(struct tun_struct
*tun
)
2769 struct net_device
*dev
= tun
->dev
;
2770 struct tun_file
*tfile
;
2771 struct skb_array
**arrays
;
2772 int n
= tun
->numqueues
+ tun
->numdisabled
;
2775 arrays
= kmalloc_array(n
, sizeof(*arrays
), GFP_KERNEL
);
2779 for (i
= 0; i
< tun
->numqueues
; i
++) {
2780 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
2781 arrays
[i
] = &tfile
->tx_array
;
2783 list_for_each_entry(tfile
, &tun
->disabled
, next
)
2784 arrays
[i
++] = &tfile
->tx_array
;
2786 ret
= skb_array_resize_multiple(arrays
, n
,
2787 dev
->tx_queue_len
, GFP_KERNEL
);
2793 static int tun_device_event(struct notifier_block
*unused
,
2794 unsigned long event
, void *ptr
)
2796 struct net_device
*dev
= netdev_notifier_info_to_dev(ptr
);
2797 struct tun_struct
*tun
= netdev_priv(dev
);
2799 if (dev
->rtnl_link_ops
!= &tun_link_ops
)
2803 case NETDEV_CHANGE_TX_QUEUE_LEN
:
2804 if (tun_queue_resize(tun
))
2814 static struct notifier_block tun_notifier_block __read_mostly
= {
2815 .notifier_call
= tun_device_event
,
2818 static int __init
tun_init(void)
2822 pr_info("%s, %s\n", DRV_DESCRIPTION
, DRV_VERSION
);
2824 ret
= rtnl_link_register(&tun_link_ops
);
2826 pr_err("Can't register link_ops\n");
2830 ret
= misc_register(&tun_miscdev
);
2832 pr_err("Can't register misc device %d\n", TUN_MINOR
);
2836 ret
= register_netdevice_notifier(&tun_notifier_block
);
2838 pr_err("Can't register netdevice notifier\n");
2845 misc_deregister(&tun_miscdev
);
2847 rtnl_link_unregister(&tun_link_ops
);
2852 static void tun_cleanup(void)
2854 misc_deregister(&tun_miscdev
);
2855 rtnl_link_unregister(&tun_link_ops
);
2856 unregister_netdevice_notifier(&tun_notifier_block
);
2859 /* Get an underlying socket object from tun file. Returns error unless file is
2860 * attached to a device. The returned object works like a packet socket, it
2861 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
2862 * holding a reference to the file for as long as the socket is in use. */
2863 struct socket
*tun_get_socket(struct file
*file
)
2865 struct tun_file
*tfile
;
2866 if (file
->f_op
!= &tun_fops
)
2867 return ERR_PTR(-EINVAL
);
2868 tfile
= file
->private_data
;
2870 return ERR_PTR(-EBADFD
);
2871 return &tfile
->socket
;
2873 EXPORT_SYMBOL_GPL(tun_get_socket
);
2875 struct skb_array
*tun_get_skb_array(struct file
*file
)
2877 struct tun_file
*tfile
;
2879 if (file
->f_op
!= &tun_fops
)
2880 return ERR_PTR(-EINVAL
);
2881 tfile
= file
->private_data
;
2883 return ERR_PTR(-EBADFD
);
2884 return &tfile
->tx_array
;
2886 EXPORT_SYMBOL_GPL(tun_get_skb_array
);
2888 module_init(tun_init
);
2889 module_exit(tun_cleanup
);
2890 MODULE_DESCRIPTION(DRV_DESCRIPTION
);
2891 MODULE_AUTHOR(DRV_COPYRIGHT
);
2892 MODULE_LICENSE("GPL");
2893 MODULE_ALIAS_MISCDEV(TUN_MINOR
);
2894 MODULE_ALIAS("devname:net/tun");