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>
81 #include <linux/proc_fs.h>
83 /* Uncomment to enable debugging */
84 /* #define TUN_DEBUG 1 */
89 #define tun_debug(level, tun, fmt, args...) \
92 netdev_printk(level, tun->dev, fmt, ##args); \
94 #define DBG1(level, fmt, args...) \
97 printk(level fmt, ##args); \
100 #define tun_debug(level, tun, fmt, args...) \
103 netdev_printk(level, tun->dev, fmt, ##args); \
105 #define DBG1(level, fmt, args...) \
108 printk(level fmt, ##args); \
112 #define TUN_HEADROOM 256
113 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
115 /* TUN device flags */
117 /* IFF_ATTACH_QUEUE is never stored in device flags,
118 * overload it to mean fasync when stored there.
120 #define TUN_FASYNC IFF_ATTACH_QUEUE
121 /* High bits in flags field are unused. */
122 #define TUN_VNET_LE 0x80000000
123 #define TUN_VNET_BE 0x40000000
125 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
126 IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
128 #define GOODCOPY_LEN 128
130 #define FLT_EXACT_COUNT 8
132 unsigned int count
; /* Number of addrs. Zero means disabled */
133 u32 mask
[2]; /* Mask of the hashed addrs */
134 unsigned char addr
[FLT_EXACT_COUNT
][ETH_ALEN
];
137 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
138 * to max number of VCPUs in guest. */
139 #define MAX_TAP_QUEUES 256
140 #define MAX_TAP_FLOWS 4096
142 #define TUN_FLOW_EXPIRE (3 * HZ)
144 struct tun_pcpu_stats
{
149 struct u64_stats_sync syncp
;
155 /* A tun_file connects an open character device to a tuntap netdevice. It
156 * also contains all socket related structures (except sock_fprog and tap_filter)
157 * to serve as one transmit queue for tuntap device. The sock_fprog and
158 * tap_filter were kept in tun_struct since they were used for filtering for the
159 * netdevice not for a specific queue (at least I didn't see the requirement for
163 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
164 * other can only be read while rcu_read_lock or rtnl_lock is held.
168 struct socket socket
;
170 struct tun_struct __rcu
*tun
;
171 struct fasync_struct
*fasync
;
172 /* only used for fasnyc */
176 unsigned int ifindex
;
178 struct napi_struct napi
;
180 struct mutex napi_mutex
; /* Protects access to the above napi */
181 struct list_head next
;
182 struct tun_struct
*detached
;
183 struct ptr_ring tx_ring
;
184 struct xdp_rxq_info xdp_rxq
;
187 struct tun_flow_entry
{
188 struct hlist_node hash_link
;
190 struct tun_struct
*tun
;
195 unsigned long updated
;
198 #define TUN_NUM_FLOW_ENTRIES 1024
202 struct bpf_prog
*prog
;
205 /* Since the socket were moved to tun_file, to preserve the behavior of persist
206 * device, socket filter, sndbuf and vnet header size were restore when the
207 * file were attached to a persist device.
210 struct tun_file __rcu
*tfiles
[MAX_TAP_QUEUES
];
211 unsigned int numqueues
;
216 struct net_device
*dev
;
217 netdev_features_t set_features
;
218 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
224 struct tap_filter txflt
;
225 struct sock_fprog fprog
;
226 /* protected by rtnl lock */
227 bool filter_attached
;
232 struct hlist_head flows
[TUN_NUM_FLOW_ENTRIES
];
233 struct timer_list flow_gc_timer
;
234 unsigned long ageing_time
;
235 unsigned int numdisabled
;
236 struct list_head disabled
;
240 struct tun_pcpu_stats __percpu
*pcpu_stats
;
241 struct bpf_prog __rcu
*xdp_prog
;
242 struct tun_prog __rcu
*steering_prog
;
243 struct tun_prog __rcu
*filter_prog
;
251 bool tun_is_xdp_buff(void *ptr
)
253 return (unsigned long)ptr
& TUN_XDP_FLAG
;
255 EXPORT_SYMBOL(tun_is_xdp_buff
);
257 void *tun_xdp_to_ptr(void *ptr
)
259 return (void *)((unsigned long)ptr
| TUN_XDP_FLAG
);
261 EXPORT_SYMBOL(tun_xdp_to_ptr
);
263 void *tun_ptr_to_xdp(void *ptr
)
265 return (void *)((unsigned long)ptr
& ~TUN_XDP_FLAG
);
267 EXPORT_SYMBOL(tun_ptr_to_xdp
);
269 static int tun_napi_receive(struct napi_struct
*napi
, int budget
)
271 struct tun_file
*tfile
= container_of(napi
, struct tun_file
, napi
);
272 struct sk_buff_head
*queue
= &tfile
->sk
.sk_write_queue
;
273 struct sk_buff_head process_queue
;
277 __skb_queue_head_init(&process_queue
);
279 spin_lock(&queue
->lock
);
280 skb_queue_splice_tail_init(queue
, &process_queue
);
281 spin_unlock(&queue
->lock
);
283 while (received
< budget
&& (skb
= __skb_dequeue(&process_queue
))) {
284 napi_gro_receive(napi
, skb
);
288 if (!skb_queue_empty(&process_queue
)) {
289 spin_lock(&queue
->lock
);
290 skb_queue_splice(&process_queue
, queue
);
291 spin_unlock(&queue
->lock
);
297 static int tun_napi_poll(struct napi_struct
*napi
, int budget
)
299 unsigned int received
;
301 received
= tun_napi_receive(napi
, budget
);
303 if (received
< budget
)
304 napi_complete_done(napi
, received
);
309 static void tun_napi_init(struct tun_struct
*tun
, struct tun_file
*tfile
,
312 tfile
->napi_enabled
= napi_en
;
314 netif_napi_add(tun
->dev
, &tfile
->napi
, tun_napi_poll
,
316 napi_enable(&tfile
->napi
);
317 mutex_init(&tfile
->napi_mutex
);
321 static void tun_napi_disable(struct tun_struct
*tun
, struct tun_file
*tfile
)
323 if (tfile
->napi_enabled
)
324 napi_disable(&tfile
->napi
);
327 static void tun_napi_del(struct tun_struct
*tun
, struct tun_file
*tfile
)
329 if (tfile
->napi_enabled
)
330 netif_napi_del(&tfile
->napi
);
333 static bool tun_napi_frags_enabled(const struct tun_struct
*tun
)
335 return READ_ONCE(tun
->flags
) & IFF_NAPI_FRAGS
;
338 #ifdef CONFIG_TUN_VNET_CROSS_LE
339 static inline bool tun_legacy_is_little_endian(struct tun_struct
*tun
)
341 return tun
->flags
& TUN_VNET_BE
? false :
342 virtio_legacy_is_little_endian();
345 static long tun_get_vnet_be(struct tun_struct
*tun
, int __user
*argp
)
347 int be
= !!(tun
->flags
& TUN_VNET_BE
);
349 if (put_user(be
, argp
))
355 static long tun_set_vnet_be(struct tun_struct
*tun
, int __user
*argp
)
359 if (get_user(be
, argp
))
363 tun
->flags
|= TUN_VNET_BE
;
365 tun
->flags
&= ~TUN_VNET_BE
;
370 static inline bool tun_legacy_is_little_endian(struct tun_struct
*tun
)
372 return virtio_legacy_is_little_endian();
375 static long tun_get_vnet_be(struct tun_struct
*tun
, int __user
*argp
)
380 static long tun_set_vnet_be(struct tun_struct
*tun
, int __user
*argp
)
384 #endif /* CONFIG_TUN_VNET_CROSS_LE */
386 static inline bool tun_is_little_endian(struct tun_struct
*tun
)
388 return tun
->flags
& TUN_VNET_LE
||
389 tun_legacy_is_little_endian(tun
);
392 static inline u16
tun16_to_cpu(struct tun_struct
*tun
, __virtio16 val
)
394 return __virtio16_to_cpu(tun_is_little_endian(tun
), val
);
397 static inline __virtio16
cpu_to_tun16(struct tun_struct
*tun
, u16 val
)
399 return __cpu_to_virtio16(tun_is_little_endian(tun
), val
);
402 static inline u32
tun_hashfn(u32 rxhash
)
404 return rxhash
& 0x3ff;
407 static struct tun_flow_entry
*tun_flow_find(struct hlist_head
*head
, u32 rxhash
)
409 struct tun_flow_entry
*e
;
411 hlist_for_each_entry_rcu(e
, head
, hash_link
) {
412 if (e
->rxhash
== rxhash
)
418 static struct tun_flow_entry
*tun_flow_create(struct tun_struct
*tun
,
419 struct hlist_head
*head
,
420 u32 rxhash
, u16 queue_index
)
422 struct tun_flow_entry
*e
= kmalloc(sizeof(*e
), GFP_ATOMIC
);
425 tun_debug(KERN_INFO
, tun
, "create flow: hash %u index %u\n",
426 rxhash
, queue_index
);
427 e
->updated
= jiffies
;
430 e
->queue_index
= queue_index
;
432 hlist_add_head_rcu(&e
->hash_link
, head
);
438 static void tun_flow_delete(struct tun_struct
*tun
, struct tun_flow_entry
*e
)
440 tun_debug(KERN_INFO
, tun
, "delete flow: hash %u index %u\n",
441 e
->rxhash
, e
->queue_index
);
442 hlist_del_rcu(&e
->hash_link
);
447 static void tun_flow_flush(struct tun_struct
*tun
)
451 spin_lock_bh(&tun
->lock
);
452 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
453 struct tun_flow_entry
*e
;
454 struct hlist_node
*n
;
456 hlist_for_each_entry_safe(e
, n
, &tun
->flows
[i
], hash_link
)
457 tun_flow_delete(tun
, e
);
459 spin_unlock_bh(&tun
->lock
);
462 static void tun_flow_delete_by_queue(struct tun_struct
*tun
, u16 queue_index
)
466 spin_lock_bh(&tun
->lock
);
467 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
468 struct tun_flow_entry
*e
;
469 struct hlist_node
*n
;
471 hlist_for_each_entry_safe(e
, n
, &tun
->flows
[i
], hash_link
) {
472 if (e
->queue_index
== queue_index
)
473 tun_flow_delete(tun
, e
);
476 spin_unlock_bh(&tun
->lock
);
479 static void tun_flow_cleanup(struct timer_list
*t
)
481 struct tun_struct
*tun
= from_timer(tun
, t
, flow_gc_timer
);
482 unsigned long delay
= tun
->ageing_time
;
483 unsigned long next_timer
= jiffies
+ delay
;
484 unsigned long count
= 0;
487 tun_debug(KERN_INFO
, tun
, "tun_flow_cleanup\n");
489 spin_lock(&tun
->lock
);
490 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++) {
491 struct tun_flow_entry
*e
;
492 struct hlist_node
*n
;
494 hlist_for_each_entry_safe(e
, n
, &tun
->flows
[i
], hash_link
) {
495 unsigned long this_timer
;
497 this_timer
= e
->updated
+ delay
;
498 if (time_before_eq(this_timer
, jiffies
)) {
499 tun_flow_delete(tun
, e
);
503 if (time_before(this_timer
, next_timer
))
504 next_timer
= this_timer
;
509 mod_timer(&tun
->flow_gc_timer
, round_jiffies_up(next_timer
));
510 spin_unlock(&tun
->lock
);
513 static void tun_flow_update(struct tun_struct
*tun
, u32 rxhash
,
514 struct tun_file
*tfile
)
516 struct hlist_head
*head
;
517 struct tun_flow_entry
*e
;
518 unsigned long delay
= tun
->ageing_time
;
519 u16 queue_index
= tfile
->queue_index
;
524 head
= &tun
->flows
[tun_hashfn(rxhash
)];
528 /* We may get a very small possibility of OOO during switching, not
529 * worth to optimize.*/
530 if (tun
->numqueues
== 1 || tfile
->detached
)
533 e
= tun_flow_find(head
, rxhash
);
535 /* TODO: keep queueing to old queue until it's empty? */
536 e
->queue_index
= queue_index
;
537 e
->updated
= jiffies
;
538 sock_rps_record_flow_hash(e
->rps_rxhash
);
540 spin_lock_bh(&tun
->lock
);
541 if (!tun_flow_find(head
, rxhash
) &&
542 tun
->flow_count
< MAX_TAP_FLOWS
)
543 tun_flow_create(tun
, head
, rxhash
, queue_index
);
545 if (!timer_pending(&tun
->flow_gc_timer
))
546 mod_timer(&tun
->flow_gc_timer
,
547 round_jiffies_up(jiffies
+ delay
));
548 spin_unlock_bh(&tun
->lock
);
556 * Save the hash received in the stack receive path and update the
557 * flow_hash table accordingly.
559 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry
*e
, u32 hash
)
561 if (unlikely(e
->rps_rxhash
!= hash
))
562 e
->rps_rxhash
= hash
;
565 /* We try to identify a flow through its rxhash first. The reason that
566 * we do not check rxq no. is because some cards(e.g 82599), chooses
567 * the rxq based on the txq where the last packet of the flow comes. As
568 * the userspace application move between processors, we may get a
569 * different rxq no. here. If we could not get rxhash, then we would
570 * hope the rxq no. may help here.
572 static u16
tun_automq_select_queue(struct tun_struct
*tun
, struct sk_buff
*skb
)
574 struct tun_flow_entry
*e
;
578 numqueues
= READ_ONCE(tun
->numqueues
);
580 txq
= __skb_get_hash_symmetric(skb
);
582 e
= tun_flow_find(&tun
->flows
[tun_hashfn(txq
)], txq
);
584 tun_flow_save_rps_rxhash(e
, txq
);
585 txq
= e
->queue_index
;
587 /* use multiply and shift instead of expensive divide */
588 txq
= ((u64
)txq
* numqueues
) >> 32;
589 } else if (likely(skb_rx_queue_recorded(skb
))) {
590 txq
= skb_get_rx_queue(skb
);
591 while (unlikely(txq
>= numqueues
))
598 static u16
tun_ebpf_select_queue(struct tun_struct
*tun
, struct sk_buff
*skb
)
600 struct tun_prog
*prog
;
603 prog
= rcu_dereference(tun
->steering_prog
);
605 ret
= bpf_prog_run_clear_cb(prog
->prog
, skb
);
607 return ret
% tun
->numqueues
;
610 static u16
tun_select_queue(struct net_device
*dev
, struct sk_buff
*skb
,
611 void *accel_priv
, select_queue_fallback_t fallback
)
613 struct tun_struct
*tun
= netdev_priv(dev
);
617 if (rcu_dereference(tun
->steering_prog
))
618 ret
= tun_ebpf_select_queue(tun
, skb
);
620 ret
= tun_automq_select_queue(tun
, skb
);
626 static inline bool tun_not_capable(struct tun_struct
*tun
)
628 const struct cred
*cred
= current_cred();
629 struct net
*net
= dev_net(tun
->dev
);
631 return ((uid_valid(tun
->owner
) && !uid_eq(cred
->euid
, tun
->owner
)) ||
632 (gid_valid(tun
->group
) && !in_egroup_p(tun
->group
))) &&
633 !ns_capable(net
->user_ns
, CAP_NET_ADMIN
);
636 static void tun_set_real_num_queues(struct tun_struct
*tun
)
638 netif_set_real_num_tx_queues(tun
->dev
, tun
->numqueues
);
639 netif_set_real_num_rx_queues(tun
->dev
, tun
->numqueues
);
642 static void tun_disable_queue(struct tun_struct
*tun
, struct tun_file
*tfile
)
644 tfile
->detached
= tun
;
645 list_add_tail(&tfile
->next
, &tun
->disabled
);
649 static struct tun_struct
*tun_enable_queue(struct tun_file
*tfile
)
651 struct tun_struct
*tun
= tfile
->detached
;
653 tfile
->detached
= NULL
;
654 list_del_init(&tfile
->next
);
659 static void tun_ptr_free(void *ptr
)
663 if (tun_is_xdp_buff(ptr
)) {
664 struct xdp_buff
*xdp
= tun_ptr_to_xdp(ptr
);
666 put_page(virt_to_head_page(xdp
->data
));
668 __skb_array_destroy_skb(ptr
);
672 static void tun_queue_purge(struct tun_file
*tfile
)
676 while ((ptr
= ptr_ring_consume(&tfile
->tx_ring
)) != NULL
)
679 skb_queue_purge(&tfile
->sk
.sk_write_queue
);
680 skb_queue_purge(&tfile
->sk
.sk_error_queue
);
683 static void tun_cleanup_tx_ring(struct tun_file
*tfile
)
685 if (tfile
->tx_ring
.queue
) {
686 ptr_ring_cleanup(&tfile
->tx_ring
, tun_ptr_free
);
687 xdp_rxq_info_unreg(&tfile
->xdp_rxq
);
688 memset(&tfile
->tx_ring
, 0, sizeof(tfile
->tx_ring
));
692 static void __tun_detach(struct tun_file
*tfile
, bool clean
)
694 struct tun_file
*ntfile
;
695 struct tun_struct
*tun
;
697 tun
= rtnl_dereference(tfile
->tun
);
700 tun_napi_disable(tun
, tfile
);
701 tun_napi_del(tun
, tfile
);
704 if (tun
&& !tfile
->detached
) {
705 u16 index
= tfile
->queue_index
;
706 BUG_ON(index
>= tun
->numqueues
);
708 rcu_assign_pointer(tun
->tfiles
[index
],
709 tun
->tfiles
[tun
->numqueues
- 1]);
710 ntfile
= rtnl_dereference(tun
->tfiles
[index
]);
711 ntfile
->queue_index
= index
;
715 RCU_INIT_POINTER(tfile
->tun
, NULL
);
716 sock_put(&tfile
->sk
);
718 tun_disable_queue(tun
, tfile
);
721 tun_flow_delete_by_queue(tun
, tun
->numqueues
+ 1);
722 /* Drop read queue */
723 tun_queue_purge(tfile
);
724 tun_set_real_num_queues(tun
);
725 } else if (tfile
->detached
&& clean
) {
726 tun
= tun_enable_queue(tfile
);
727 sock_put(&tfile
->sk
);
731 if (tun
&& tun
->numqueues
== 0 && tun
->numdisabled
== 0) {
732 netif_carrier_off(tun
->dev
);
734 if (!(tun
->flags
& IFF_PERSIST
) &&
735 tun
->dev
->reg_state
== NETREG_REGISTERED
)
736 unregister_netdevice(tun
->dev
);
738 tun_cleanup_tx_ring(tfile
);
739 sock_put(&tfile
->sk
);
743 static void tun_detach(struct tun_file
*tfile
, bool clean
)
746 __tun_detach(tfile
, clean
);
750 static void tun_detach_all(struct net_device
*dev
)
752 struct tun_struct
*tun
= netdev_priv(dev
);
753 struct tun_file
*tfile
, *tmp
;
754 int i
, n
= tun
->numqueues
;
756 for (i
= 0; i
< n
; i
++) {
757 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
759 tun_napi_disable(tun
, tfile
);
760 tfile
->socket
.sk
->sk_shutdown
= RCV_SHUTDOWN
;
761 tfile
->socket
.sk
->sk_data_ready(tfile
->socket
.sk
);
762 RCU_INIT_POINTER(tfile
->tun
, NULL
);
765 list_for_each_entry(tfile
, &tun
->disabled
, next
) {
766 tfile
->socket
.sk
->sk_shutdown
= RCV_SHUTDOWN
;
767 tfile
->socket
.sk
->sk_data_ready(tfile
->socket
.sk
);
768 RCU_INIT_POINTER(tfile
->tun
, NULL
);
770 BUG_ON(tun
->numqueues
!= 0);
773 for (i
= 0; i
< n
; i
++) {
774 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
775 tun_napi_del(tun
, tfile
);
776 /* Drop read queue */
777 tun_queue_purge(tfile
);
778 sock_put(&tfile
->sk
);
779 tun_cleanup_tx_ring(tfile
);
781 list_for_each_entry_safe(tfile
, tmp
, &tun
->disabled
, next
) {
782 tun_enable_queue(tfile
);
783 tun_queue_purge(tfile
);
784 sock_put(&tfile
->sk
);
785 tun_cleanup_tx_ring(tfile
);
787 BUG_ON(tun
->numdisabled
!= 0);
789 if (tun
->flags
& IFF_PERSIST
)
790 module_put(THIS_MODULE
);
793 static int tun_attach(struct tun_struct
*tun
, struct file
*file
,
794 bool skip_filter
, bool napi
)
796 struct tun_file
*tfile
= file
->private_data
;
797 struct net_device
*dev
= tun
->dev
;
800 err
= security_tun_dev_attach(tfile
->socket
.sk
, tun
->security
);
805 if (rtnl_dereference(tfile
->tun
) && !tfile
->detached
)
809 if (!(tun
->flags
& IFF_MULTI_QUEUE
) && tun
->numqueues
== 1)
813 if (!tfile
->detached
&&
814 tun
->numqueues
+ tun
->numdisabled
== MAX_TAP_QUEUES
)
819 /* Re-attach the filter to persist device */
820 if (!skip_filter
&& (tun
->filter_attached
== true)) {
821 lock_sock(tfile
->socket
.sk
);
822 err
= sk_attach_filter(&tun
->fprog
, tfile
->socket
.sk
);
823 release_sock(tfile
->socket
.sk
);
828 if (!tfile
->detached
&&
829 ptr_ring_init(&tfile
->tx_ring
, dev
->tx_queue_len
, GFP_KERNEL
)) {
834 tfile
->queue_index
= tun
->numqueues
;
835 tfile
->socket
.sk
->sk_shutdown
&= ~RCV_SHUTDOWN
;
837 if (tfile
->detached
) {
838 /* Re-attach detached tfile, updating XDP queue_index */
839 WARN_ON(!xdp_rxq_info_is_reg(&tfile
->xdp_rxq
));
841 if (tfile
->xdp_rxq
.queue_index
!= tfile
->queue_index
)
842 tfile
->xdp_rxq
.queue_index
= tfile
->queue_index
;
844 /* Setup XDP RX-queue info, for new tfile getting attached */
845 err
= xdp_rxq_info_reg(&tfile
->xdp_rxq
,
846 tun
->dev
, tfile
->queue_index
);
852 rcu_assign_pointer(tfile
->tun
, tun
);
853 rcu_assign_pointer(tun
->tfiles
[tun
->numqueues
], tfile
);
856 if (tfile
->detached
) {
857 tun_enable_queue(tfile
);
859 sock_hold(&tfile
->sk
);
860 tun_napi_init(tun
, tfile
, napi
);
863 tun_set_real_num_queues(tun
);
865 /* device is allowed to go away first, so no need to hold extra
873 static struct tun_struct
*tun_get(struct tun_file
*tfile
)
875 struct tun_struct
*tun
;
878 tun
= rcu_dereference(tfile
->tun
);
886 static void tun_put(struct tun_struct
*tun
)
892 static void addr_hash_set(u32
*mask
, const u8
*addr
)
894 int n
= ether_crc(ETH_ALEN
, addr
) >> 26;
895 mask
[n
>> 5] |= (1 << (n
& 31));
898 static unsigned int addr_hash_test(const u32
*mask
, const u8
*addr
)
900 int n
= ether_crc(ETH_ALEN
, addr
) >> 26;
901 return mask
[n
>> 5] & (1 << (n
& 31));
904 static int update_filter(struct tap_filter
*filter
, void __user
*arg
)
906 struct { u8 u
[ETH_ALEN
]; } *addr
;
907 struct tun_filter uf
;
908 int err
, alen
, n
, nexact
;
910 if (copy_from_user(&uf
, arg
, sizeof(uf
)))
919 alen
= ETH_ALEN
* uf
.count
;
920 addr
= memdup_user(arg
+ sizeof(uf
), alen
);
922 return PTR_ERR(addr
);
924 /* The filter is updated without holding any locks. Which is
925 * perfectly safe. We disable it first and in the worst
926 * case we'll accept a few undesired packets. */
930 /* Use first set of addresses as an exact filter */
931 for (n
= 0; n
< uf
.count
&& n
< FLT_EXACT_COUNT
; n
++)
932 memcpy(filter
->addr
[n
], addr
[n
].u
, ETH_ALEN
);
936 /* Remaining multicast addresses are hashed,
937 * unicast will leave the filter disabled. */
938 memset(filter
->mask
, 0, sizeof(filter
->mask
));
939 for (; n
< uf
.count
; n
++) {
940 if (!is_multicast_ether_addr(addr
[n
].u
)) {
941 err
= 0; /* no filter */
944 addr_hash_set(filter
->mask
, addr
[n
].u
);
947 /* For ALLMULTI just set the mask to all ones.
948 * This overrides the mask populated above. */
949 if ((uf
.flags
& TUN_FLT_ALLMULTI
))
950 memset(filter
->mask
, ~0, sizeof(filter
->mask
));
952 /* Now enable the filter */
954 filter
->count
= nexact
;
956 /* Return the number of exact filters */
963 /* Returns: 0 - drop, !=0 - accept */
964 static int run_filter(struct tap_filter
*filter
, const struct sk_buff
*skb
)
966 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
968 struct ethhdr
*eh
= (struct ethhdr
*) skb
->data
;
972 for (i
= 0; i
< filter
->count
; i
++)
973 if (ether_addr_equal(eh
->h_dest
, filter
->addr
[i
]))
976 /* Inexact match (multicast only) */
977 if (is_multicast_ether_addr(eh
->h_dest
))
978 return addr_hash_test(filter
->mask
, eh
->h_dest
);
984 * Checks whether the packet is accepted or not.
985 * Returns: 0 - drop, !=0 - accept
987 static int check_filter(struct tap_filter
*filter
, const struct sk_buff
*skb
)
992 return run_filter(filter
, skb
);
995 /* Network device part of the driver */
997 static const struct ethtool_ops tun_ethtool_ops
;
999 /* Net device detach from fd. */
1000 static void tun_net_uninit(struct net_device
*dev
)
1002 tun_detach_all(dev
);
1005 /* Net device open. */
1006 static int tun_net_open(struct net_device
*dev
)
1008 struct tun_struct
*tun
= netdev_priv(dev
);
1011 netif_tx_start_all_queues(dev
);
1013 for (i
= 0; i
< tun
->numqueues
; i
++) {
1014 struct tun_file
*tfile
;
1016 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
1017 tfile
->socket
.sk
->sk_write_space(tfile
->socket
.sk
);
1023 /* Net device close. */
1024 static int tun_net_close(struct net_device
*dev
)
1026 netif_tx_stop_all_queues(dev
);
1030 /* Net device start xmit */
1031 static void tun_automq_xmit(struct tun_struct
*tun
, struct sk_buff
*skb
)
1034 if (tun
->numqueues
== 1 && static_key_false(&rps_needed
)) {
1035 /* Select queue was not called for the skbuff, so we extract the
1036 * RPS hash and save it into the flow_table here.
1040 rxhash
= __skb_get_hash_symmetric(skb
);
1042 struct tun_flow_entry
*e
;
1043 e
= tun_flow_find(&tun
->flows
[tun_hashfn(rxhash
)],
1046 tun_flow_save_rps_rxhash(e
, rxhash
);
1052 static unsigned int run_ebpf_filter(struct tun_struct
*tun
,
1053 struct sk_buff
*skb
,
1056 struct tun_prog
*prog
= rcu_dereference(tun
->filter_prog
);
1059 len
= bpf_prog_run_clear_cb(prog
->prog
, skb
);
1064 /* Net device start xmit */
1065 static netdev_tx_t
tun_net_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
1067 struct tun_struct
*tun
= netdev_priv(dev
);
1068 int txq
= skb
->queue_mapping
;
1069 struct tun_file
*tfile
;
1073 tfile
= rcu_dereference(tun
->tfiles
[txq
]);
1075 /* Drop packet if interface is not attached */
1076 if (txq
>= tun
->numqueues
)
1079 if (!rcu_dereference(tun
->steering_prog
))
1080 tun_automq_xmit(tun
, skb
);
1082 tun_debug(KERN_INFO
, tun
, "tun_net_xmit %d\n", skb
->len
);
1086 /* Drop if the filter does not like it.
1087 * This is a noop if the filter is disabled.
1088 * Filter can be enabled only for the TAP devices. */
1089 if (!check_filter(&tun
->txflt
, skb
))
1092 if (tfile
->socket
.sk
->sk_filter
&&
1093 sk_filter(tfile
->socket
.sk
, skb
))
1096 len
= run_ebpf_filter(tun
, skb
, len
);
1098 /* Trim extra bytes since we may insert vlan proto & TCI
1099 * in tun_put_user().
1101 len
-= skb_vlan_tag_present(skb
) ? sizeof(struct veth
) : 0;
1102 if (len
<= 0 || pskb_trim(skb
, len
))
1105 if (unlikely(skb_orphan_frags_rx(skb
, GFP_ATOMIC
)))
1108 skb_tx_timestamp(skb
);
1110 /* Orphan the skb - required as we might hang on to it
1111 * for indefinite time.
1117 if (ptr_ring_produce(&tfile
->tx_ring
, skb
))
1120 /* Notify and wake up reader process */
1121 if (tfile
->flags
& TUN_FASYNC
)
1122 kill_fasync(&tfile
->fasync
, SIGIO
, POLL_IN
);
1123 tfile
->socket
.sk
->sk_data_ready(tfile
->socket
.sk
);
1126 return NETDEV_TX_OK
;
1129 this_cpu_inc(tun
->pcpu_stats
->tx_dropped
);
1133 return NET_XMIT_DROP
;
1136 static void tun_net_mclist(struct net_device
*dev
)
1139 * This callback is supposed to deal with mc filter in
1140 * _rx_ path and has nothing to do with the _tx_ path.
1141 * In rx path we always accept everything userspace gives us.
1145 static netdev_features_t
tun_net_fix_features(struct net_device
*dev
,
1146 netdev_features_t features
)
1148 struct tun_struct
*tun
= netdev_priv(dev
);
1150 return (features
& tun
->set_features
) | (features
& ~TUN_USER_FEATURES
);
1152 #ifdef CONFIG_NET_POLL_CONTROLLER
1153 static void tun_poll_controller(struct net_device
*dev
)
1156 * Tun only receives frames when:
1157 * 1) the char device endpoint gets data from user space
1158 * 2) the tun socket gets a sendmsg call from user space
1159 * If NAPI is not enabled, since both of those are synchronous
1160 * operations, we are guaranteed never to have pending data when we poll
1161 * for it so there is nothing to do here but return.
1162 * We need this though so netpoll recognizes us as an interface that
1163 * supports polling, which enables bridge devices in virt setups to
1164 * still use netconsole
1165 * If NAPI is enabled, however, we need to schedule polling for all
1166 * queues unless we are using napi_gro_frags(), which we call in
1167 * process context and not in NAPI context.
1169 struct tun_struct
*tun
= netdev_priv(dev
);
1171 if (tun
->flags
& IFF_NAPI
) {
1172 struct tun_file
*tfile
;
1175 if (tun_napi_frags_enabled(tun
))
1179 for (i
= 0; i
< tun
->numqueues
; i
++) {
1180 tfile
= rcu_dereference(tun
->tfiles
[i
]);
1181 if (tfile
->napi_enabled
)
1182 napi_schedule(&tfile
->napi
);
1190 static void tun_set_headroom(struct net_device
*dev
, int new_hr
)
1192 struct tun_struct
*tun
= netdev_priv(dev
);
1194 if (new_hr
< NET_SKB_PAD
)
1195 new_hr
= NET_SKB_PAD
;
1197 tun
->align
= new_hr
;
1201 tun_net_get_stats64(struct net_device
*dev
, struct rtnl_link_stats64
*stats
)
1203 u32 rx_dropped
= 0, tx_dropped
= 0, rx_frame_errors
= 0;
1204 struct tun_struct
*tun
= netdev_priv(dev
);
1205 struct tun_pcpu_stats
*p
;
1208 for_each_possible_cpu(i
) {
1209 u64 rxpackets
, rxbytes
, txpackets
, txbytes
;
1212 p
= per_cpu_ptr(tun
->pcpu_stats
, i
);
1214 start
= u64_stats_fetch_begin(&p
->syncp
);
1215 rxpackets
= p
->rx_packets
;
1216 rxbytes
= p
->rx_bytes
;
1217 txpackets
= p
->tx_packets
;
1218 txbytes
= p
->tx_bytes
;
1219 } while (u64_stats_fetch_retry(&p
->syncp
, start
));
1221 stats
->rx_packets
+= rxpackets
;
1222 stats
->rx_bytes
+= rxbytes
;
1223 stats
->tx_packets
+= txpackets
;
1224 stats
->tx_bytes
+= txbytes
;
1227 rx_dropped
+= p
->rx_dropped
;
1228 rx_frame_errors
+= p
->rx_frame_errors
;
1229 tx_dropped
+= p
->tx_dropped
;
1231 stats
->rx_dropped
= rx_dropped
;
1232 stats
->rx_frame_errors
= rx_frame_errors
;
1233 stats
->tx_dropped
= tx_dropped
;
1236 static int tun_xdp_set(struct net_device
*dev
, struct bpf_prog
*prog
,
1237 struct netlink_ext_ack
*extack
)
1239 struct tun_struct
*tun
= netdev_priv(dev
);
1240 struct bpf_prog
*old_prog
;
1242 old_prog
= rtnl_dereference(tun
->xdp_prog
);
1243 rcu_assign_pointer(tun
->xdp_prog
, prog
);
1245 bpf_prog_put(old_prog
);
1250 static u32
tun_xdp_query(struct net_device
*dev
)
1252 struct tun_struct
*tun
= netdev_priv(dev
);
1253 const struct bpf_prog
*xdp_prog
;
1255 xdp_prog
= rtnl_dereference(tun
->xdp_prog
);
1257 return xdp_prog
->aux
->id
;
1262 static int tun_xdp(struct net_device
*dev
, struct netdev_bpf
*xdp
)
1264 switch (xdp
->command
) {
1265 case XDP_SETUP_PROG
:
1266 return tun_xdp_set(dev
, xdp
->prog
, xdp
->extack
);
1267 case XDP_QUERY_PROG
:
1268 xdp
->prog_id
= tun_xdp_query(dev
);
1269 xdp
->prog_attached
= !!xdp
->prog_id
;
1276 static const struct net_device_ops tun_netdev_ops
= {
1277 .ndo_uninit
= tun_net_uninit
,
1278 .ndo_open
= tun_net_open
,
1279 .ndo_stop
= tun_net_close
,
1280 .ndo_start_xmit
= tun_net_xmit
,
1281 .ndo_fix_features
= tun_net_fix_features
,
1282 .ndo_select_queue
= tun_select_queue
,
1283 #ifdef CONFIG_NET_POLL_CONTROLLER
1284 .ndo_poll_controller
= tun_poll_controller
,
1286 .ndo_set_rx_headroom
= tun_set_headroom
,
1287 .ndo_get_stats64
= tun_net_get_stats64
,
1290 static int tun_xdp_xmit(struct net_device
*dev
, struct xdp_buff
*xdp
)
1292 struct tun_struct
*tun
= netdev_priv(dev
);
1293 struct xdp_buff
*buff
= xdp
->data_hard_start
;
1294 int headroom
= xdp
->data
- xdp
->data_hard_start
;
1295 struct tun_file
*tfile
;
1299 /* Assure headroom is available and buff is properly aligned */
1300 if (unlikely(headroom
< sizeof(*xdp
) || tun_is_xdp_buff(xdp
)))
1307 numqueues
= READ_ONCE(tun
->numqueues
);
1313 tfile
= rcu_dereference(tun
->tfiles
[smp_processor_id() %
1315 /* Encode the XDP flag into lowest bit for consumer to differ
1316 * XDP buffer from sk_buff.
1318 if (ptr_ring_produce(&tfile
->tx_ring
, tun_xdp_to_ptr(buff
))) {
1319 this_cpu_inc(tun
->pcpu_stats
->tx_dropped
);
1328 static void tun_xdp_flush(struct net_device
*dev
)
1330 struct tun_struct
*tun
= netdev_priv(dev
);
1331 struct tun_file
*tfile
;
1336 numqueues
= READ_ONCE(tun
->numqueues
);
1340 tfile
= rcu_dereference(tun
->tfiles
[smp_processor_id() %
1342 /* Notify and wake up reader process */
1343 if (tfile
->flags
& TUN_FASYNC
)
1344 kill_fasync(&tfile
->fasync
, SIGIO
, POLL_IN
);
1345 tfile
->socket
.sk
->sk_data_ready(tfile
->socket
.sk
);
1351 static const struct net_device_ops tap_netdev_ops
= {
1352 .ndo_uninit
= tun_net_uninit
,
1353 .ndo_open
= tun_net_open
,
1354 .ndo_stop
= tun_net_close
,
1355 .ndo_start_xmit
= tun_net_xmit
,
1356 .ndo_fix_features
= tun_net_fix_features
,
1357 .ndo_set_rx_mode
= tun_net_mclist
,
1358 .ndo_set_mac_address
= eth_mac_addr
,
1359 .ndo_validate_addr
= eth_validate_addr
,
1360 .ndo_select_queue
= tun_select_queue
,
1361 #ifdef CONFIG_NET_POLL_CONTROLLER
1362 .ndo_poll_controller
= tun_poll_controller
,
1364 .ndo_features_check
= passthru_features_check
,
1365 .ndo_set_rx_headroom
= tun_set_headroom
,
1366 .ndo_get_stats64
= tun_net_get_stats64
,
1368 .ndo_xdp_xmit
= tun_xdp_xmit
,
1369 .ndo_xdp_flush
= tun_xdp_flush
,
1372 static void tun_flow_init(struct tun_struct
*tun
)
1376 for (i
= 0; i
< TUN_NUM_FLOW_ENTRIES
; i
++)
1377 INIT_HLIST_HEAD(&tun
->flows
[i
]);
1379 tun
->ageing_time
= TUN_FLOW_EXPIRE
;
1380 timer_setup(&tun
->flow_gc_timer
, tun_flow_cleanup
, 0);
1381 mod_timer(&tun
->flow_gc_timer
,
1382 round_jiffies_up(jiffies
+ tun
->ageing_time
));
1385 static void tun_flow_uninit(struct tun_struct
*tun
)
1387 del_timer_sync(&tun
->flow_gc_timer
);
1388 tun_flow_flush(tun
);
1392 #define MAX_MTU 65535
1394 /* Initialize net device. */
1395 static void tun_net_init(struct net_device
*dev
)
1397 struct tun_struct
*tun
= netdev_priv(dev
);
1399 switch (tun
->flags
& TUN_TYPE_MASK
) {
1401 dev
->netdev_ops
= &tun_netdev_ops
;
1403 /* Point-to-Point TUN Device */
1404 dev
->hard_header_len
= 0;
1408 /* Zero header length */
1409 dev
->type
= ARPHRD_NONE
;
1410 dev
->flags
= IFF_POINTOPOINT
| IFF_NOARP
| IFF_MULTICAST
;
1414 dev
->netdev_ops
= &tap_netdev_ops
;
1415 /* Ethernet TAP Device */
1417 dev
->priv_flags
&= ~IFF_TX_SKB_SHARING
;
1418 dev
->priv_flags
|= IFF_LIVE_ADDR_CHANGE
;
1420 eth_hw_addr_random(dev
);
1425 dev
->min_mtu
= MIN_MTU
;
1426 dev
->max_mtu
= MAX_MTU
- dev
->hard_header_len
;
1429 /* Character device part */
1432 static __poll_t
tun_chr_poll(struct file
*file
, poll_table
*wait
)
1434 struct tun_file
*tfile
= file
->private_data
;
1435 struct tun_struct
*tun
= tun_get(tfile
);
1442 sk
= tfile
->socket
.sk
;
1444 tun_debug(KERN_INFO
, tun
, "tun_chr_poll\n");
1446 poll_wait(file
, sk_sleep(sk
), wait
);
1448 if (!ptr_ring_empty(&tfile
->tx_ring
))
1449 mask
|= EPOLLIN
| EPOLLRDNORM
;
1451 if (tun
->dev
->flags
& IFF_UP
&&
1452 (sock_writeable(sk
) ||
1453 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
) &&
1454 sock_writeable(sk
))))
1455 mask
|= EPOLLOUT
| EPOLLWRNORM
;
1457 if (tun
->dev
->reg_state
!= NETREG_REGISTERED
)
1464 static struct sk_buff
*tun_napi_alloc_frags(struct tun_file
*tfile
,
1466 const struct iov_iter
*it
)
1468 struct sk_buff
*skb
;
1473 if (it
->nr_segs
> MAX_SKB_FRAGS
+ 1)
1474 return ERR_PTR(-ENOMEM
);
1477 skb
= napi_get_frags(&tfile
->napi
);
1480 return ERR_PTR(-ENOMEM
);
1482 linear
= iov_iter_single_seg_count(it
);
1483 err
= __skb_grow(skb
, linear
);
1488 skb
->data_len
= len
- linear
;
1489 skb
->truesize
+= skb
->data_len
;
1491 for (i
= 1; i
< it
->nr_segs
; i
++) {
1492 struct page_frag
*pfrag
= ¤t
->task_frag
;
1493 size_t fragsz
= it
->iov
[i
].iov_len
;
1495 if (fragsz
== 0 || fragsz
> PAGE_SIZE
) {
1500 if (!skb_page_frag_refill(fragsz
, pfrag
, GFP_KERNEL
)) {
1505 skb_fill_page_desc(skb
, i
- 1, pfrag
->page
,
1506 pfrag
->offset
, fragsz
);
1507 page_ref_inc(pfrag
->page
);
1508 pfrag
->offset
+= fragsz
;
1513 /* frees skb and all frags allocated with napi_alloc_frag() */
1514 napi_free_frags(&tfile
->napi
);
1515 return ERR_PTR(err
);
1518 /* prepad is the amount to reserve at front. len is length after that.
1519 * linear is a hint as to how much to copy (usually headers). */
1520 static struct sk_buff
*tun_alloc_skb(struct tun_file
*tfile
,
1521 size_t prepad
, size_t len
,
1522 size_t linear
, int noblock
)
1524 struct sock
*sk
= tfile
->socket
.sk
;
1525 struct sk_buff
*skb
;
1528 /* Under a page? Don't bother with paged skb. */
1529 if (prepad
+ len
< PAGE_SIZE
|| !linear
)
1532 skb
= sock_alloc_send_pskb(sk
, prepad
+ linear
, len
- linear
, noblock
,
1535 return ERR_PTR(err
);
1537 skb_reserve(skb
, prepad
);
1538 skb_put(skb
, linear
);
1539 skb
->data_len
= len
- linear
;
1540 skb
->len
+= len
- linear
;
1545 static void tun_rx_batched(struct tun_struct
*tun
, struct tun_file
*tfile
,
1546 struct sk_buff
*skb
, int more
)
1548 struct sk_buff_head
*queue
= &tfile
->sk
.sk_write_queue
;
1549 struct sk_buff_head process_queue
;
1550 u32 rx_batched
= tun
->rx_batched
;
1553 if (!rx_batched
|| (!more
&& skb_queue_empty(queue
))) {
1555 netif_receive_skb(skb
);
1560 spin_lock(&queue
->lock
);
1561 if (!more
|| skb_queue_len(queue
) == rx_batched
) {
1562 __skb_queue_head_init(&process_queue
);
1563 skb_queue_splice_tail_init(queue
, &process_queue
);
1566 __skb_queue_tail(queue
, skb
);
1568 spin_unlock(&queue
->lock
);
1571 struct sk_buff
*nskb
;
1574 while ((nskb
= __skb_dequeue(&process_queue
)))
1575 netif_receive_skb(nskb
);
1576 netif_receive_skb(skb
);
1581 static bool tun_can_build_skb(struct tun_struct
*tun
, struct tun_file
*tfile
,
1582 int len
, int noblock
, bool zerocopy
)
1584 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
1587 if (tfile
->socket
.sk
->sk_sndbuf
!= INT_MAX
)
1596 if (SKB_DATA_ALIGN(len
+ TUN_RX_PAD
) +
1597 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
)) > PAGE_SIZE
)
1603 static struct sk_buff
*tun_build_skb(struct tun_struct
*tun
,
1604 struct tun_file
*tfile
,
1605 struct iov_iter
*from
,
1606 struct virtio_net_hdr
*hdr
,
1607 int len
, int *skb_xdp
)
1609 struct page_frag
*alloc_frag
= ¤t
->task_frag
;
1610 struct sk_buff
*skb
;
1611 struct bpf_prog
*xdp_prog
;
1612 int buflen
= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
1613 unsigned int delta
= 0;
1616 int err
, pad
= TUN_RX_PAD
;
1619 xdp_prog
= rcu_dereference(tun
->xdp_prog
);
1621 pad
+= TUN_HEADROOM
;
1622 buflen
+= SKB_DATA_ALIGN(len
+ pad
);
1625 alloc_frag
->offset
= ALIGN((u64
)alloc_frag
->offset
, SMP_CACHE_BYTES
);
1626 if (unlikely(!skb_page_frag_refill(buflen
, alloc_frag
, GFP_KERNEL
)))
1627 return ERR_PTR(-ENOMEM
);
1629 buf
= (char *)page_address(alloc_frag
->page
) + alloc_frag
->offset
;
1630 copied
= copy_page_from_iter(alloc_frag
->page
,
1631 alloc_frag
->offset
+ pad
,
1634 return ERR_PTR(-EFAULT
);
1636 /* There's a small window that XDP may be set after the check
1637 * of xdp_prog above, this should be rare and for simplicity
1638 * we do XDP on skb in case the headroom is not enough.
1640 if (hdr
->gso_type
|| !xdp_prog
)
1647 xdp_prog
= rcu_dereference(tun
->xdp_prog
);
1648 if (xdp_prog
&& !*skb_xdp
) {
1649 struct xdp_buff xdp
;
1653 xdp
.data_hard_start
= buf
;
1654 xdp
.data
= buf
+ pad
;
1655 xdp_set_data_meta_invalid(&xdp
);
1656 xdp
.data_end
= xdp
.data
+ len
;
1657 xdp
.rxq
= &tfile
->xdp_rxq
;
1658 orig_data
= xdp
.data
;
1659 act
= bpf_prog_run_xdp(xdp_prog
, &xdp
);
1663 get_page(alloc_frag
->page
);
1664 alloc_frag
->offset
+= buflen
;
1665 err
= xdp_do_redirect(tun
->dev
, &xdp
, xdp_prog
);
1673 get_page(alloc_frag
->page
);
1674 alloc_frag
->offset
+= buflen
;
1675 if (tun_xdp_xmit(tun
->dev
, &xdp
))
1677 tun_xdp_flush(tun
->dev
);
1682 delta
= orig_data
- xdp
.data
;
1685 bpf_warn_invalid_xdp_action(act
);
1688 trace_xdp_exception(tun
->dev
, xdp_prog
, act
);
1695 skb
= build_skb(buf
, buflen
);
1699 return ERR_PTR(-ENOMEM
);
1702 skb_reserve(skb
, pad
- delta
);
1703 skb_put(skb
, len
+ delta
);
1704 get_page(alloc_frag
->page
);
1705 alloc_frag
->offset
+= buflen
;
1713 put_page(alloc_frag
->page
);
1717 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1721 /* Get packet from user space buffer */
1722 static ssize_t
tun_get_user(struct tun_struct
*tun
, struct tun_file
*tfile
,
1723 void *msg_control
, struct iov_iter
*from
,
1724 int noblock
, bool more
)
1726 struct tun_pi pi
= { 0, cpu_to_be16(ETH_P_IP
) };
1727 struct sk_buff
*skb
;
1728 size_t total_len
= iov_iter_count(from
);
1729 size_t len
= total_len
, align
= tun
->align
, linear
;
1730 struct virtio_net_hdr gso
= { 0 };
1731 struct tun_pcpu_stats
*stats
;
1734 bool zerocopy
= false;
1738 bool frags
= tun_napi_frags_enabled(tun
);
1740 if (!(tun
->dev
->flags
& IFF_UP
))
1743 if (!(tun
->flags
& IFF_NO_PI
)) {
1744 if (len
< sizeof(pi
))
1748 if (!copy_from_iter_full(&pi
, sizeof(pi
), from
))
1752 if (tun
->flags
& IFF_VNET_HDR
) {
1753 int vnet_hdr_sz
= READ_ONCE(tun
->vnet_hdr_sz
);
1755 if (len
< vnet_hdr_sz
)
1759 if (!copy_from_iter_full(&gso
, sizeof(gso
), from
))
1762 if ((gso
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) &&
1763 tun16_to_cpu(tun
, gso
.csum_start
) + tun16_to_cpu(tun
, gso
.csum_offset
) + 2 > tun16_to_cpu(tun
, gso
.hdr_len
))
1764 gso
.hdr_len
= cpu_to_tun16(tun
, tun16_to_cpu(tun
, gso
.csum_start
) + tun16_to_cpu(tun
, gso
.csum_offset
) + 2);
1766 if (tun16_to_cpu(tun
, gso
.hdr_len
) > len
)
1768 iov_iter_advance(from
, vnet_hdr_sz
- sizeof(gso
));
1771 if ((tun
->flags
& TUN_TYPE_MASK
) == IFF_TAP
) {
1772 align
+= NET_IP_ALIGN
;
1773 if (unlikely(len
< ETH_HLEN
||
1774 (gso
.hdr_len
&& tun16_to_cpu(tun
, gso
.hdr_len
) < ETH_HLEN
)))
1778 good_linear
= SKB_MAX_HEAD(align
);
1781 struct iov_iter i
= *from
;
1783 /* There are 256 bytes to be copied in skb, so there is
1784 * enough room for skb expand head in case it is used.
1785 * The rest of the buffer is mapped from userspace.
1787 copylen
= gso
.hdr_len
? tun16_to_cpu(tun
, gso
.hdr_len
) : GOODCOPY_LEN
;
1788 if (copylen
> good_linear
)
1789 copylen
= good_linear
;
1791 iov_iter_advance(&i
, copylen
);
1792 if (iov_iter_npages(&i
, INT_MAX
) <= MAX_SKB_FRAGS
)
1796 if (!frags
&& tun_can_build_skb(tun
, tfile
, len
, noblock
, zerocopy
)) {
1797 /* For the packet that is not easy to be processed
1798 * (e.g gso or jumbo packet), we will do it at after
1799 * skb was created with generic XDP routine.
1801 skb
= tun_build_skb(tun
, tfile
, from
, &gso
, len
, &skb_xdp
);
1803 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1804 return PTR_ERR(skb
);
1811 if (tun16_to_cpu(tun
, gso
.hdr_len
) > good_linear
)
1812 linear
= good_linear
;
1814 linear
= tun16_to_cpu(tun
, gso
.hdr_len
);
1818 mutex_lock(&tfile
->napi_mutex
);
1819 skb
= tun_napi_alloc_frags(tfile
, copylen
, from
);
1820 /* tun_napi_alloc_frags() enforces a layout for the skb.
1821 * If zerocopy is enabled, then this layout will be
1822 * overwritten by zerocopy_sg_from_iter().
1826 skb
= tun_alloc_skb(tfile
, align
, copylen
, linear
,
1831 if (PTR_ERR(skb
) != -EAGAIN
)
1832 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1834 mutex_unlock(&tfile
->napi_mutex
);
1835 return PTR_ERR(skb
);
1839 err
= zerocopy_sg_from_iter(skb
, from
);
1841 err
= skb_copy_datagram_from_iter(skb
, 0, from
, len
);
1844 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1847 tfile
->napi
.skb
= NULL
;
1848 mutex_unlock(&tfile
->napi_mutex
);
1855 if (virtio_net_hdr_to_skb(skb
, &gso
, tun_is_little_endian(tun
))) {
1856 this_cpu_inc(tun
->pcpu_stats
->rx_frame_errors
);
1859 tfile
->napi
.skb
= NULL
;
1860 mutex_unlock(&tfile
->napi_mutex
);
1866 switch (tun
->flags
& TUN_TYPE_MASK
) {
1868 if (tun
->flags
& IFF_NO_PI
) {
1869 u8 ip_version
= skb
->len
? (skb
->data
[0] >> 4) : 0;
1871 switch (ip_version
) {
1873 pi
.proto
= htons(ETH_P_IP
);
1876 pi
.proto
= htons(ETH_P_IPV6
);
1879 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1885 skb_reset_mac_header(skb
);
1886 skb
->protocol
= pi
.proto
;
1887 skb
->dev
= tun
->dev
;
1891 skb
->protocol
= eth_type_trans(skb
, tun
->dev
);
1895 /* copy skb_ubuf_info for callback when skb has no error */
1897 skb_shinfo(skb
)->destructor_arg
= msg_control
;
1898 skb_shinfo(skb
)->tx_flags
|= SKBTX_DEV_ZEROCOPY
;
1899 skb_shinfo(skb
)->tx_flags
|= SKBTX_SHARED_FRAG
;
1900 } else if (msg_control
) {
1901 struct ubuf_info
*uarg
= msg_control
;
1902 uarg
->callback(uarg
, false);
1905 skb_reset_network_header(skb
);
1906 skb_probe_transport_header(skb
, 0);
1909 struct bpf_prog
*xdp_prog
;
1913 xdp_prog
= rcu_dereference(tun
->xdp_prog
);
1915 ret
= do_xdp_generic(xdp_prog
, skb
);
1916 if (ret
!= XDP_PASS
) {
1925 if (!rcu_dereference(tun
->steering_prog
))
1926 rxhash
= __skb_get_hash_symmetric(skb
);
1930 /* Exercise flow dissector code path. */
1931 u32 headlen
= eth_get_headlen(skb
->data
, skb_headlen(skb
));
1933 if (unlikely(headlen
> skb_headlen(skb
))) {
1934 this_cpu_inc(tun
->pcpu_stats
->rx_dropped
);
1935 napi_free_frags(&tfile
->napi
);
1936 mutex_unlock(&tfile
->napi_mutex
);
1942 napi_gro_frags(&tfile
->napi
);
1944 mutex_unlock(&tfile
->napi_mutex
);
1945 } else if (tfile
->napi_enabled
) {
1946 struct sk_buff_head
*queue
= &tfile
->sk
.sk_write_queue
;
1949 spin_lock_bh(&queue
->lock
);
1950 __skb_queue_tail(queue
, skb
);
1951 queue_len
= skb_queue_len(queue
);
1952 spin_unlock(&queue
->lock
);
1954 if (!more
|| queue_len
> NAPI_POLL_WEIGHT
)
1955 napi_schedule(&tfile
->napi
);
1958 } else if (!IS_ENABLED(CONFIG_4KSTACKS
)) {
1959 tun_rx_batched(tun
, tfile
, skb
, more
);
1964 stats
= get_cpu_ptr(tun
->pcpu_stats
);
1965 u64_stats_update_begin(&stats
->syncp
);
1966 stats
->rx_packets
++;
1967 stats
->rx_bytes
+= len
;
1968 u64_stats_update_end(&stats
->syncp
);
1972 tun_flow_update(tun
, rxhash
, tfile
);
1977 static ssize_t
tun_chr_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
1979 struct file
*file
= iocb
->ki_filp
;
1980 struct tun_file
*tfile
= file
->private_data
;
1981 struct tun_struct
*tun
= tun_get(tfile
);
1987 result
= tun_get_user(tun
, tfile
, NULL
, from
,
1988 file
->f_flags
& O_NONBLOCK
, false);
1994 static ssize_t
tun_put_user_xdp(struct tun_struct
*tun
,
1995 struct tun_file
*tfile
,
1996 struct xdp_buff
*xdp
,
1997 struct iov_iter
*iter
)
1999 int vnet_hdr_sz
= 0;
2000 size_t size
= xdp
->data_end
- xdp
->data
;
2001 struct tun_pcpu_stats
*stats
;
2004 if (tun
->flags
& IFF_VNET_HDR
) {
2005 struct virtio_net_hdr gso
= { 0 };
2007 vnet_hdr_sz
= READ_ONCE(tun
->vnet_hdr_sz
);
2008 if (unlikely(iov_iter_count(iter
) < vnet_hdr_sz
))
2010 if (unlikely(copy_to_iter(&gso
, sizeof(gso
), iter
) !=
2013 iov_iter_advance(iter
, vnet_hdr_sz
- sizeof(gso
));
2016 ret
= copy_to_iter(xdp
->data
, size
, iter
) + vnet_hdr_sz
;
2018 stats
= get_cpu_ptr(tun
->pcpu_stats
);
2019 u64_stats_update_begin(&stats
->syncp
);
2020 stats
->tx_packets
++;
2021 stats
->tx_bytes
+= ret
;
2022 u64_stats_update_end(&stats
->syncp
);
2023 put_cpu_ptr(tun
->pcpu_stats
);
2028 /* Put packet to the user space buffer */
2029 static ssize_t
tun_put_user(struct tun_struct
*tun
,
2030 struct tun_file
*tfile
,
2031 struct sk_buff
*skb
,
2032 struct iov_iter
*iter
)
2034 struct tun_pi pi
= { 0, skb
->protocol
};
2035 struct tun_pcpu_stats
*stats
;
2037 int vlan_offset
= 0;
2039 int vnet_hdr_sz
= 0;
2041 if (skb_vlan_tag_present(skb
))
2042 vlan_hlen
= VLAN_HLEN
;
2044 if (tun
->flags
& IFF_VNET_HDR
)
2045 vnet_hdr_sz
= READ_ONCE(tun
->vnet_hdr_sz
);
2047 total
= skb
->len
+ vlan_hlen
+ vnet_hdr_sz
;
2049 if (!(tun
->flags
& IFF_NO_PI
)) {
2050 if (iov_iter_count(iter
) < sizeof(pi
))
2053 total
+= sizeof(pi
);
2054 if (iov_iter_count(iter
) < total
) {
2055 /* Packet will be striped */
2056 pi
.flags
|= TUN_PKT_STRIP
;
2059 if (copy_to_iter(&pi
, sizeof(pi
), iter
) != sizeof(pi
))
2064 struct virtio_net_hdr gso
;
2066 if (iov_iter_count(iter
) < vnet_hdr_sz
)
2069 if (virtio_net_hdr_from_skb(skb
, &gso
,
2070 tun_is_little_endian(tun
), true)) {
2071 struct skb_shared_info
*sinfo
= skb_shinfo(skb
);
2072 pr_err("unexpected GSO type: "
2073 "0x%x, gso_size %d, hdr_len %d\n",
2074 sinfo
->gso_type
, tun16_to_cpu(tun
, gso
.gso_size
),
2075 tun16_to_cpu(tun
, gso
.hdr_len
));
2076 print_hex_dump(KERN_ERR
, "tun: ",
2079 min((int)tun16_to_cpu(tun
, gso
.hdr_len
), 64), true);
2084 if (copy_to_iter(&gso
, sizeof(gso
), iter
) != sizeof(gso
))
2087 iov_iter_advance(iter
, vnet_hdr_sz
- sizeof(gso
));
2094 veth
.h_vlan_proto
= skb
->vlan_proto
;
2095 veth
.h_vlan_TCI
= htons(skb_vlan_tag_get(skb
));
2097 vlan_offset
= offsetof(struct vlan_ethhdr
, h_vlan_proto
);
2099 ret
= skb_copy_datagram_iter(skb
, 0, iter
, vlan_offset
);
2100 if (ret
|| !iov_iter_count(iter
))
2103 ret
= copy_to_iter(&veth
, sizeof(veth
), iter
);
2104 if (ret
!= sizeof(veth
) || !iov_iter_count(iter
))
2108 skb_copy_datagram_iter(skb
, vlan_offset
, iter
, skb
->len
- vlan_offset
);
2111 /* caller is in process context, */
2112 stats
= get_cpu_ptr(tun
->pcpu_stats
);
2113 u64_stats_update_begin(&stats
->syncp
);
2114 stats
->tx_packets
++;
2115 stats
->tx_bytes
+= skb
->len
+ vlan_hlen
;
2116 u64_stats_update_end(&stats
->syncp
);
2117 put_cpu_ptr(tun
->pcpu_stats
);
2122 static void *tun_ring_recv(struct tun_file
*tfile
, int noblock
, int *err
)
2124 DECLARE_WAITQUEUE(wait
, current
);
2128 ptr
= ptr_ring_consume(&tfile
->tx_ring
);
2136 add_wait_queue(&tfile
->wq
.wait
, &wait
);
2137 current
->state
= TASK_INTERRUPTIBLE
;
2140 ptr
= ptr_ring_consume(&tfile
->tx_ring
);
2143 if (signal_pending(current
)) {
2144 error
= -ERESTARTSYS
;
2147 if (tfile
->socket
.sk
->sk_shutdown
& RCV_SHUTDOWN
) {
2155 current
->state
= TASK_RUNNING
;
2156 remove_wait_queue(&tfile
->wq
.wait
, &wait
);
2163 static ssize_t
tun_do_read(struct tun_struct
*tun
, struct tun_file
*tfile
,
2164 struct iov_iter
*to
,
2165 int noblock
, void *ptr
)
2170 tun_debug(KERN_INFO
, tun
, "tun_do_read\n");
2172 if (!iov_iter_count(to
)) {
2178 /* Read frames from ring */
2179 ptr
= tun_ring_recv(tfile
, noblock
, &err
);
2184 if (tun_is_xdp_buff(ptr
)) {
2185 struct xdp_buff
*xdp
= tun_ptr_to_xdp(ptr
);
2187 ret
= tun_put_user_xdp(tun
, tfile
, xdp
, to
);
2188 put_page(virt_to_head_page(xdp
->data
));
2190 struct sk_buff
*skb
= ptr
;
2192 ret
= tun_put_user(tun
, tfile
, skb
, to
);
2193 if (unlikely(ret
< 0))
2202 static ssize_t
tun_chr_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
2204 struct file
*file
= iocb
->ki_filp
;
2205 struct tun_file
*tfile
= file
->private_data
;
2206 struct tun_struct
*tun
= tun_get(tfile
);
2207 ssize_t len
= iov_iter_count(to
), ret
;
2211 ret
= tun_do_read(tun
, tfile
, to
, file
->f_flags
& O_NONBLOCK
, NULL
);
2212 ret
= min_t(ssize_t
, ret
, len
);
2219 static void tun_prog_free(struct rcu_head
*rcu
)
2221 struct tun_prog
*prog
= container_of(rcu
, struct tun_prog
, rcu
);
2223 bpf_prog_destroy(prog
->prog
);
2227 static int __tun_set_ebpf(struct tun_struct
*tun
,
2228 struct tun_prog __rcu
**prog_p
,
2229 struct bpf_prog
*prog
)
2231 struct tun_prog
*old
, *new = NULL
;
2234 new = kmalloc(sizeof(*new), GFP_KERNEL
);
2240 spin_lock_bh(&tun
->lock
);
2241 old
= rcu_dereference_protected(*prog_p
,
2242 lockdep_is_held(&tun
->lock
));
2243 rcu_assign_pointer(*prog_p
, new);
2244 spin_unlock_bh(&tun
->lock
);
2247 call_rcu(&old
->rcu
, tun_prog_free
);
2252 static void tun_free_netdev(struct net_device
*dev
)
2254 struct tun_struct
*tun
= netdev_priv(dev
);
2256 BUG_ON(!(list_empty(&tun
->disabled
)));
2257 free_percpu(tun
->pcpu_stats
);
2258 tun_flow_uninit(tun
);
2259 security_tun_dev_free_security(tun
->security
);
2260 __tun_set_ebpf(tun
, &tun
->steering_prog
, NULL
);
2261 __tun_set_ebpf(tun
, &tun
->filter_prog
, NULL
);
2264 static void tun_setup(struct net_device
*dev
)
2266 struct tun_struct
*tun
= netdev_priv(dev
);
2268 tun
->owner
= INVALID_UID
;
2269 tun
->group
= INVALID_GID
;
2271 dev
->ethtool_ops
= &tun_ethtool_ops
;
2272 dev
->needs_free_netdev
= true;
2273 dev
->priv_destructor
= tun_free_netdev
;
2274 /* We prefer our own queue length */
2275 dev
->tx_queue_len
= TUN_READQ_SIZE
;
2278 /* Trivial set of netlink ops to allow deleting tun or tap
2279 * device with netlink.
2281 static int tun_validate(struct nlattr
*tb
[], struct nlattr
*data
[],
2282 struct netlink_ext_ack
*extack
)
2287 static size_t tun_get_size(const struct net_device
*dev
)
2289 BUILD_BUG_ON(sizeof(u32
) != sizeof(uid_t
));
2290 BUILD_BUG_ON(sizeof(u32
) != sizeof(gid_t
));
2292 return nla_total_size(sizeof(uid_t
)) + /* OWNER */
2293 nla_total_size(sizeof(gid_t
)) + /* GROUP */
2294 nla_total_size(sizeof(u8
)) + /* TYPE */
2295 nla_total_size(sizeof(u8
)) + /* PI */
2296 nla_total_size(sizeof(u8
)) + /* VNET_HDR */
2297 nla_total_size(sizeof(u8
)) + /* PERSIST */
2298 nla_total_size(sizeof(u8
)) + /* MULTI_QUEUE */
2299 nla_total_size(sizeof(u32
)) + /* NUM_QUEUES */
2300 nla_total_size(sizeof(u32
)) + /* NUM_DISABLED_QUEUES */
2304 static int tun_fill_info(struct sk_buff
*skb
, const struct net_device
*dev
)
2306 struct tun_struct
*tun
= netdev_priv(dev
);
2308 if (nla_put_u8(skb
, IFLA_TUN_TYPE
, tun
->flags
& TUN_TYPE_MASK
))
2309 goto nla_put_failure
;
2310 if (uid_valid(tun
->owner
) &&
2311 nla_put_u32(skb
, IFLA_TUN_OWNER
,
2312 from_kuid_munged(current_user_ns(), tun
->owner
)))
2313 goto nla_put_failure
;
2314 if (gid_valid(tun
->group
) &&
2315 nla_put_u32(skb
, IFLA_TUN_GROUP
,
2316 from_kgid_munged(current_user_ns(), tun
->group
)))
2317 goto nla_put_failure
;
2318 if (nla_put_u8(skb
, IFLA_TUN_PI
, !(tun
->flags
& IFF_NO_PI
)))
2319 goto nla_put_failure
;
2320 if (nla_put_u8(skb
, IFLA_TUN_VNET_HDR
, !!(tun
->flags
& IFF_VNET_HDR
)))
2321 goto nla_put_failure
;
2322 if (nla_put_u8(skb
, IFLA_TUN_PERSIST
, !!(tun
->flags
& IFF_PERSIST
)))
2323 goto nla_put_failure
;
2324 if (nla_put_u8(skb
, IFLA_TUN_MULTI_QUEUE
,
2325 !!(tun
->flags
& IFF_MULTI_QUEUE
)))
2326 goto nla_put_failure
;
2327 if (tun
->flags
& IFF_MULTI_QUEUE
) {
2328 if (nla_put_u32(skb
, IFLA_TUN_NUM_QUEUES
, tun
->numqueues
))
2329 goto nla_put_failure
;
2330 if (nla_put_u32(skb
, IFLA_TUN_NUM_DISABLED_QUEUES
,
2332 goto nla_put_failure
;
2341 static struct rtnl_link_ops tun_link_ops __read_mostly
= {
2343 .priv_size
= sizeof(struct tun_struct
),
2345 .validate
= tun_validate
,
2346 .get_size
= tun_get_size
,
2347 .fill_info
= tun_fill_info
,
2350 static void tun_sock_write_space(struct sock
*sk
)
2352 struct tun_file
*tfile
;
2353 wait_queue_head_t
*wqueue
;
2355 if (!sock_writeable(sk
))
2358 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
))
2361 wqueue
= sk_sleep(sk
);
2362 if (wqueue
&& waitqueue_active(wqueue
))
2363 wake_up_interruptible_sync_poll(wqueue
, EPOLLOUT
|
2364 EPOLLWRNORM
| EPOLLWRBAND
);
2366 tfile
= container_of(sk
, struct tun_file
, sk
);
2367 kill_fasync(&tfile
->fasync
, SIGIO
, POLL_OUT
);
2370 static int tun_sendmsg(struct socket
*sock
, struct msghdr
*m
, size_t total_len
)
2373 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
2374 struct tun_struct
*tun
= tun_get(tfile
);
2379 ret
= tun_get_user(tun
, tfile
, m
->msg_control
, &m
->msg_iter
,
2380 m
->msg_flags
& MSG_DONTWAIT
,
2381 m
->msg_flags
& MSG_MORE
);
2386 static int tun_recvmsg(struct socket
*sock
, struct msghdr
*m
, size_t total_len
,
2389 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
2390 struct tun_struct
*tun
= tun_get(tfile
);
2391 void *ptr
= m
->msg_control
;
2399 if (flags
& ~(MSG_DONTWAIT
|MSG_TRUNC
|MSG_ERRQUEUE
)) {
2403 if (flags
& MSG_ERRQUEUE
) {
2404 ret
= sock_recv_errqueue(sock
->sk
, m
, total_len
,
2405 SOL_PACKET
, TUN_TX_TIMESTAMP
);
2408 ret
= tun_do_read(tun
, tfile
, &m
->msg_iter
, flags
& MSG_DONTWAIT
, ptr
);
2409 if (ret
> (ssize_t
)total_len
) {
2410 m
->msg_flags
|= MSG_TRUNC
;
2411 ret
= flags
& MSG_TRUNC
? ret
: total_len
;
2424 static int tun_ptr_peek_len(void *ptr
)
2427 if (tun_is_xdp_buff(ptr
)) {
2428 struct xdp_buff
*xdp
= tun_ptr_to_xdp(ptr
);
2430 return xdp
->data_end
- xdp
->data
;
2432 return __skb_array_len_with_tag(ptr
);
2438 static int tun_peek_len(struct socket
*sock
)
2440 struct tun_file
*tfile
= container_of(sock
, struct tun_file
, socket
);
2441 struct tun_struct
*tun
;
2444 tun
= tun_get(tfile
);
2448 ret
= PTR_RING_PEEK_CALL(&tfile
->tx_ring
, tun_ptr_peek_len
);
2454 /* Ops structure to mimic raw sockets with tun */
2455 static const struct proto_ops tun_socket_ops
= {
2456 .peek_len
= tun_peek_len
,
2457 .sendmsg
= tun_sendmsg
,
2458 .recvmsg
= tun_recvmsg
,
2461 static struct proto tun_proto
= {
2463 .owner
= THIS_MODULE
,
2464 .obj_size
= sizeof(struct tun_file
),
2467 static int tun_flags(struct tun_struct
*tun
)
2469 return tun
->flags
& (TUN_FEATURES
| IFF_PERSIST
| IFF_TUN
| IFF_TAP
);
2472 static ssize_t
tun_show_flags(struct device
*dev
, struct device_attribute
*attr
,
2475 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
2476 return sprintf(buf
, "0x%x\n", tun_flags(tun
));
2479 static ssize_t
tun_show_owner(struct device
*dev
, struct device_attribute
*attr
,
2482 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
2483 return uid_valid(tun
->owner
)?
2484 sprintf(buf
, "%u\n",
2485 from_kuid_munged(current_user_ns(), tun
->owner
)):
2486 sprintf(buf
, "-1\n");
2489 static ssize_t
tun_show_group(struct device
*dev
, struct device_attribute
*attr
,
2492 struct tun_struct
*tun
= netdev_priv(to_net_dev(dev
));
2493 return gid_valid(tun
->group
) ?
2494 sprintf(buf
, "%u\n",
2495 from_kgid_munged(current_user_ns(), tun
->group
)):
2496 sprintf(buf
, "-1\n");
2499 static DEVICE_ATTR(tun_flags
, 0444, tun_show_flags
, NULL
);
2500 static DEVICE_ATTR(owner
, 0444, tun_show_owner
, NULL
);
2501 static DEVICE_ATTR(group
, 0444, tun_show_group
, NULL
);
2503 static struct attribute
*tun_dev_attrs
[] = {
2504 &dev_attr_tun_flags
.attr
,
2505 &dev_attr_owner
.attr
,
2506 &dev_attr_group
.attr
,
2510 static const struct attribute_group tun_attr_group
= {
2511 .attrs
= tun_dev_attrs
2514 static int tun_set_iff(struct net
*net
, struct file
*file
, struct ifreq
*ifr
)
2516 struct tun_struct
*tun
;
2517 struct tun_file
*tfile
= file
->private_data
;
2518 struct net_device
*dev
;
2521 if (tfile
->detached
)
2524 if ((ifr
->ifr_flags
& IFF_NAPI_FRAGS
)) {
2525 if (!capable(CAP_NET_ADMIN
))
2528 if (!(ifr
->ifr_flags
& IFF_NAPI
) ||
2529 (ifr
->ifr_flags
& TUN_TYPE_MASK
) != IFF_TAP
)
2533 dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
2535 if (ifr
->ifr_flags
& IFF_TUN_EXCL
)
2537 if ((ifr
->ifr_flags
& IFF_TUN
) && dev
->netdev_ops
== &tun_netdev_ops
)
2538 tun
= netdev_priv(dev
);
2539 else if ((ifr
->ifr_flags
& IFF_TAP
) && dev
->netdev_ops
== &tap_netdev_ops
)
2540 tun
= netdev_priv(dev
);
2544 if (!!(ifr
->ifr_flags
& IFF_MULTI_QUEUE
) !=
2545 !!(tun
->flags
& IFF_MULTI_QUEUE
))
2548 if (tun_not_capable(tun
))
2550 err
= security_tun_dev_open(tun
->security
);
2554 err
= tun_attach(tun
, file
, ifr
->ifr_flags
& IFF_NOFILTER
,
2555 ifr
->ifr_flags
& IFF_NAPI
);
2559 if (tun
->flags
& IFF_MULTI_QUEUE
&&
2560 (tun
->numqueues
+ tun
->numdisabled
> 1)) {
2561 /* One or more queue has already been attached, no need
2562 * to initialize the device again.
2569 unsigned long flags
= 0;
2570 int queues
= ifr
->ifr_flags
& IFF_MULTI_QUEUE
?
2573 if (!ns_capable(net
->user_ns
, CAP_NET_ADMIN
))
2575 err
= security_tun_dev_create();
2580 if (ifr
->ifr_flags
& IFF_TUN
) {
2584 } else if (ifr
->ifr_flags
& IFF_TAP
) {
2592 name
= ifr
->ifr_name
;
2594 dev
= alloc_netdev_mqs(sizeof(struct tun_struct
), name
,
2595 NET_NAME_UNKNOWN
, tun_setup
, queues
,
2600 err
= dev_get_valid_name(net
, dev
, name
);
2604 dev_net_set(dev
, net
);
2605 dev
->rtnl_link_ops
= &tun_link_ops
;
2606 dev
->ifindex
= tfile
->ifindex
;
2607 dev
->sysfs_groups
[0] = &tun_attr_group
;
2609 tun
= netdev_priv(dev
);
2612 tun
->txflt
.count
= 0;
2613 tun
->vnet_hdr_sz
= sizeof(struct virtio_net_hdr
);
2615 tun
->align
= NET_SKB_PAD
;
2616 tun
->filter_attached
= false;
2617 tun
->sndbuf
= tfile
->socket
.sk
->sk_sndbuf
;
2618 tun
->rx_batched
= 0;
2619 RCU_INIT_POINTER(tun
->steering_prog
, NULL
);
2621 tun
->pcpu_stats
= netdev_alloc_pcpu_stats(struct tun_pcpu_stats
);
2622 if (!tun
->pcpu_stats
) {
2627 spin_lock_init(&tun
->lock
);
2629 err
= security_tun_dev_alloc_security(&tun
->security
);
2636 dev
->hw_features
= NETIF_F_SG
| NETIF_F_FRAGLIST
|
2637 TUN_USER_FEATURES
| NETIF_F_HW_VLAN_CTAG_TX
|
2638 NETIF_F_HW_VLAN_STAG_TX
;
2639 dev
->features
= dev
->hw_features
| NETIF_F_LLTX
;
2640 dev
->vlan_features
= dev
->features
&
2641 ~(NETIF_F_HW_VLAN_CTAG_TX
|
2642 NETIF_F_HW_VLAN_STAG_TX
);
2644 INIT_LIST_HEAD(&tun
->disabled
);
2645 err
= tun_attach(tun
, file
, false, ifr
->ifr_flags
& IFF_NAPI
);
2649 err
= register_netdevice(tun
->dev
);
2654 netif_carrier_on(tun
->dev
);
2656 tun_debug(KERN_INFO
, tun
, "tun_set_iff\n");
2658 tun
->flags
= (tun
->flags
& ~TUN_FEATURES
) |
2659 (ifr
->ifr_flags
& TUN_FEATURES
);
2661 /* Make sure persistent devices do not get stuck in
2664 if (netif_running(tun
->dev
))
2665 netif_tx_wake_all_queues(tun
->dev
);
2667 strcpy(ifr
->ifr_name
, tun
->dev
->name
);
2671 tun_detach_all(dev
);
2672 /* register_netdevice() already called tun_free_netdev() */
2676 tun_flow_uninit(tun
);
2677 security_tun_dev_free_security(tun
->security
);
2679 free_percpu(tun
->pcpu_stats
);
2685 static void tun_get_iff(struct net
*net
, struct tun_struct
*tun
,
2688 tun_debug(KERN_INFO
, tun
, "tun_get_iff\n");
2690 strcpy(ifr
->ifr_name
, tun
->dev
->name
);
2692 ifr
->ifr_flags
= tun_flags(tun
);
2696 /* This is like a cut-down ethtool ops, except done via tun fd so no
2697 * privs required. */
2698 static int set_offload(struct tun_struct
*tun
, unsigned long arg
)
2700 netdev_features_t features
= 0;
2702 if (arg
& TUN_F_CSUM
) {
2703 features
|= NETIF_F_HW_CSUM
;
2706 if (arg
& (TUN_F_TSO4
|TUN_F_TSO6
)) {
2707 if (arg
& TUN_F_TSO_ECN
) {
2708 features
|= NETIF_F_TSO_ECN
;
2709 arg
&= ~TUN_F_TSO_ECN
;
2711 if (arg
& TUN_F_TSO4
)
2712 features
|= NETIF_F_TSO
;
2713 if (arg
& TUN_F_TSO6
)
2714 features
|= NETIF_F_TSO6
;
2715 arg
&= ~(TUN_F_TSO4
|TUN_F_TSO6
);
2721 /* This gives the user a way to test for new features in future by
2722 * trying to set them. */
2726 tun
->set_features
= features
;
2727 tun
->dev
->wanted_features
&= ~TUN_USER_FEATURES
;
2728 tun
->dev
->wanted_features
|= features
;
2729 netdev_update_features(tun
->dev
);
2734 static void tun_detach_filter(struct tun_struct
*tun
, int n
)
2737 struct tun_file
*tfile
;
2739 for (i
= 0; i
< n
; i
++) {
2740 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
2741 lock_sock(tfile
->socket
.sk
);
2742 sk_detach_filter(tfile
->socket
.sk
);
2743 release_sock(tfile
->socket
.sk
);
2746 tun
->filter_attached
= false;
2749 static int tun_attach_filter(struct tun_struct
*tun
)
2752 struct tun_file
*tfile
;
2754 for (i
= 0; i
< tun
->numqueues
; i
++) {
2755 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
2756 lock_sock(tfile
->socket
.sk
);
2757 ret
= sk_attach_filter(&tun
->fprog
, tfile
->socket
.sk
);
2758 release_sock(tfile
->socket
.sk
);
2760 tun_detach_filter(tun
, i
);
2765 tun
->filter_attached
= true;
2769 static void tun_set_sndbuf(struct tun_struct
*tun
)
2771 struct tun_file
*tfile
;
2774 for (i
= 0; i
< tun
->numqueues
; i
++) {
2775 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
2776 tfile
->socket
.sk
->sk_sndbuf
= tun
->sndbuf
;
2780 static int tun_set_queue(struct file
*file
, struct ifreq
*ifr
)
2782 struct tun_file
*tfile
= file
->private_data
;
2783 struct tun_struct
*tun
;
2788 if (ifr
->ifr_flags
& IFF_ATTACH_QUEUE
) {
2789 tun
= tfile
->detached
;
2794 ret
= security_tun_dev_attach_queue(tun
->security
);
2797 ret
= tun_attach(tun
, file
, false, tun
->flags
& IFF_NAPI
);
2798 } else if (ifr
->ifr_flags
& IFF_DETACH_QUEUE
) {
2799 tun
= rtnl_dereference(tfile
->tun
);
2800 if (!tun
|| !(tun
->flags
& IFF_MULTI_QUEUE
) || tfile
->detached
)
2803 __tun_detach(tfile
, false);
2812 static int tun_set_ebpf(struct tun_struct
*tun
, struct tun_prog
**prog_p
,
2815 struct bpf_prog
*prog
;
2818 if (copy_from_user(&fd
, data
, sizeof(fd
)))
2824 prog
= bpf_prog_get_type(fd
, BPF_PROG_TYPE_SOCKET_FILTER
);
2826 return PTR_ERR(prog
);
2829 return __tun_set_ebpf(tun
, prog_p
, prog
);
2832 static long __tun_chr_ioctl(struct file
*file
, unsigned int cmd
,
2833 unsigned long arg
, int ifreq_len
)
2835 struct tun_file
*tfile
= file
->private_data
;
2836 struct tun_struct
*tun
;
2837 void __user
* argp
= (void __user
*)arg
;
2844 unsigned int ifindex
;
2848 if (cmd
== TUNSETIFF
|| cmd
== TUNSETQUEUE
||
2849 (_IOC_TYPE(cmd
) == SOCK_IOC_TYPE
&& cmd
!= SIOCGSKNS
)) {
2850 if (copy_from_user(&ifr
, argp
, ifreq_len
))
2853 memset(&ifr
, 0, sizeof(ifr
));
2855 if (cmd
== TUNGETFEATURES
) {
2856 /* Currently this just means: "what IFF flags are valid?".
2857 * This is needed because we never checked for invalid flags on
2860 return put_user(IFF_TUN
| IFF_TAP
| TUN_FEATURES
,
2861 (unsigned int __user
*)argp
);
2862 } else if (cmd
== TUNSETQUEUE
)
2863 return tun_set_queue(file
, &ifr
);
2868 tun
= tun_get(tfile
);
2869 net
= sock_net(&tfile
->sk
);
2870 if (cmd
== TUNSETIFF
) {
2875 ifr
.ifr_name
[IFNAMSIZ
-1] = '\0';
2877 ret
= tun_set_iff(net
, file
, &ifr
);
2882 if (copy_to_user(argp
, &ifr
, ifreq_len
))
2886 if (cmd
== TUNSETIFINDEX
) {
2892 if (copy_from_user(&ifindex
, argp
, sizeof(ifindex
)))
2896 tfile
->ifindex
= ifindex
;
2899 if (cmd
== SIOCGSKNS
) {
2901 if (!ns_capable(net
->user_ns
, CAP_NET_ADMIN
))
2904 ret
= open_related_ns(&net
->ns
, get_net_ns
);
2912 tun_debug(KERN_INFO
, tun
, "tun_chr_ioctl cmd %u\n", cmd
);
2917 tun_get_iff(current
->nsproxy
->net_ns
, tun
, &ifr
);
2919 if (tfile
->detached
)
2920 ifr
.ifr_flags
|= IFF_DETACH_QUEUE
;
2921 if (!tfile
->socket
.sk
->sk_filter
)
2922 ifr
.ifr_flags
|= IFF_NOFILTER
;
2924 if (copy_to_user(argp
, &ifr
, ifreq_len
))
2929 /* Disable/Enable checksum */
2931 /* [unimplemented] */
2932 tun_debug(KERN_INFO
, tun
, "ignored: set checksum %s\n",
2933 arg
? "disabled" : "enabled");
2937 /* Disable/Enable persist mode. Keep an extra reference to the
2938 * module to prevent the module being unprobed.
2940 if (arg
&& !(tun
->flags
& IFF_PERSIST
)) {
2941 tun
->flags
|= IFF_PERSIST
;
2942 __module_get(THIS_MODULE
);
2944 if (!arg
&& (tun
->flags
& IFF_PERSIST
)) {
2945 tun
->flags
&= ~IFF_PERSIST
;
2946 module_put(THIS_MODULE
);
2949 tun_debug(KERN_INFO
, tun
, "persist %s\n",
2950 arg
? "enabled" : "disabled");
2954 /* Set owner of the device */
2955 owner
= make_kuid(current_user_ns(), arg
);
2956 if (!uid_valid(owner
)) {
2961 tun_debug(KERN_INFO
, tun
, "owner set to %u\n",
2962 from_kuid(&init_user_ns
, tun
->owner
));
2966 /* Set group of the device */
2967 group
= make_kgid(current_user_ns(), arg
);
2968 if (!gid_valid(group
)) {
2973 tun_debug(KERN_INFO
, tun
, "group set to %u\n",
2974 from_kgid(&init_user_ns
, tun
->group
));
2978 /* Only allow setting the type when the interface is down */
2979 if (tun
->dev
->flags
& IFF_UP
) {
2980 tun_debug(KERN_INFO
, tun
,
2981 "Linktype set failed because interface is up\n");
2984 tun
->dev
->type
= (int) arg
;
2985 tun_debug(KERN_INFO
, tun
, "linktype set to %d\n",
2997 ret
= set_offload(tun
, arg
);
3000 case TUNSETTXFILTER
:
3001 /* Can be set only for TAPs */
3003 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
3005 ret
= update_filter(&tun
->txflt
, (void __user
*)arg
);
3009 /* Get hw address */
3010 memcpy(ifr
.ifr_hwaddr
.sa_data
, tun
->dev
->dev_addr
, ETH_ALEN
);
3011 ifr
.ifr_hwaddr
.sa_family
= tun
->dev
->type
;
3012 if (copy_to_user(argp
, &ifr
, ifreq_len
))
3017 /* Set hw address */
3018 tun_debug(KERN_DEBUG
, tun
, "set hw address: %pM\n",
3019 ifr
.ifr_hwaddr
.sa_data
);
3021 ret
= dev_set_mac_address(tun
->dev
, &ifr
.ifr_hwaddr
);
3025 sndbuf
= tfile
->socket
.sk
->sk_sndbuf
;
3026 if (copy_to_user(argp
, &sndbuf
, sizeof(sndbuf
)))
3031 if (copy_from_user(&sndbuf
, argp
, sizeof(sndbuf
))) {
3040 tun
->sndbuf
= sndbuf
;
3041 tun_set_sndbuf(tun
);
3044 case TUNGETVNETHDRSZ
:
3045 vnet_hdr_sz
= tun
->vnet_hdr_sz
;
3046 if (copy_to_user(argp
, &vnet_hdr_sz
, sizeof(vnet_hdr_sz
)))
3050 case TUNSETVNETHDRSZ
:
3051 if (copy_from_user(&vnet_hdr_sz
, argp
, sizeof(vnet_hdr_sz
))) {
3055 if (vnet_hdr_sz
< (int)sizeof(struct virtio_net_hdr
)) {
3060 tun
->vnet_hdr_sz
= vnet_hdr_sz
;
3064 le
= !!(tun
->flags
& TUN_VNET_LE
);
3065 if (put_user(le
, (int __user
*)argp
))
3070 if (get_user(le
, (int __user
*)argp
)) {
3075 tun
->flags
|= TUN_VNET_LE
;
3077 tun
->flags
&= ~TUN_VNET_LE
;
3081 ret
= tun_get_vnet_be(tun
, argp
);
3085 ret
= tun_set_vnet_be(tun
, argp
);
3088 case TUNATTACHFILTER
:
3089 /* Can be set only for TAPs */
3091 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
3094 if (copy_from_user(&tun
->fprog
, argp
, sizeof(tun
->fprog
)))
3097 ret
= tun_attach_filter(tun
);
3100 case TUNDETACHFILTER
:
3101 /* Can be set only for TAPs */
3103 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
3106 tun_detach_filter(tun
, tun
->numqueues
);
3111 if ((tun
->flags
& TUN_TYPE_MASK
) != IFF_TAP
)
3114 if (copy_to_user(argp
, &tun
->fprog
, sizeof(tun
->fprog
)))
3119 case TUNSETSTEERINGEBPF
:
3120 ret
= tun_set_ebpf(tun
, &tun
->steering_prog
, argp
);
3123 case TUNSETFILTEREBPF
:
3124 ret
= tun_set_ebpf(tun
, &tun
->filter_prog
, argp
);
3139 static long tun_chr_ioctl(struct file
*file
,
3140 unsigned int cmd
, unsigned long arg
)
3142 return __tun_chr_ioctl(file
, cmd
, arg
, sizeof (struct ifreq
));
3145 #ifdef CONFIG_COMPAT
3146 static long tun_chr_compat_ioctl(struct file
*file
,
3147 unsigned int cmd
, unsigned long arg
)
3152 case TUNSETTXFILTER
:
3157 arg
= (unsigned long)compat_ptr(arg
);
3160 arg
= (compat_ulong_t
)arg
;
3165 * compat_ifreq is shorter than ifreq, so we must not access beyond
3166 * the end of that structure. All fields that are used in this
3167 * driver are compatible though, we don't need to convert the
3170 return __tun_chr_ioctl(file
, cmd
, arg
, sizeof(struct compat_ifreq
));
3172 #endif /* CONFIG_COMPAT */
3174 static int tun_chr_fasync(int fd
, struct file
*file
, int on
)
3176 struct tun_file
*tfile
= file
->private_data
;
3179 if ((ret
= fasync_helper(fd
, file
, on
, &tfile
->fasync
)) < 0)
3183 __f_setown(file
, task_pid(current
), PIDTYPE_PID
, 0);
3184 tfile
->flags
|= TUN_FASYNC
;
3186 tfile
->flags
&= ~TUN_FASYNC
;
3192 static int tun_chr_open(struct inode
*inode
, struct file
* file
)
3194 struct net
*net
= current
->nsproxy
->net_ns
;
3195 struct tun_file
*tfile
;
3197 DBG1(KERN_INFO
, "tunX: tun_chr_open\n");
3199 tfile
= (struct tun_file
*)sk_alloc(net
, AF_UNSPEC
, GFP_KERNEL
,
3203 RCU_INIT_POINTER(tfile
->tun
, NULL
);
3207 init_waitqueue_head(&tfile
->wq
.wait
);
3208 RCU_INIT_POINTER(tfile
->socket
.wq
, &tfile
->wq
);
3210 tfile
->socket
.file
= file
;
3211 tfile
->socket
.ops
= &tun_socket_ops
;
3213 sock_init_data(&tfile
->socket
, &tfile
->sk
);
3215 tfile
->sk
.sk_write_space
= tun_sock_write_space
;
3216 tfile
->sk
.sk_sndbuf
= INT_MAX
;
3218 file
->private_data
= tfile
;
3219 INIT_LIST_HEAD(&tfile
->next
);
3221 sock_set_flag(&tfile
->sk
, SOCK_ZEROCOPY
);
3223 memset(&tfile
->tx_ring
, 0, sizeof(tfile
->tx_ring
));
3228 static int tun_chr_close(struct inode
*inode
, struct file
*file
)
3230 struct tun_file
*tfile
= file
->private_data
;
3232 tun_detach(tfile
, true);
3237 #ifdef CONFIG_PROC_FS
3238 static void tun_chr_show_fdinfo(struct seq_file
*m
, struct file
*file
)
3240 struct tun_file
*tfile
= file
->private_data
;
3241 struct tun_struct
*tun
;
3244 memset(&ifr
, 0, sizeof(ifr
));
3247 tun
= tun_get(tfile
);
3249 tun_get_iff(current
->nsproxy
->net_ns
, tun
, &ifr
);
3255 seq_printf(m
, "iff:\t%s\n", ifr
.ifr_name
);
3259 static const struct file_operations tun_fops
= {
3260 .owner
= THIS_MODULE
,
3261 .llseek
= no_llseek
,
3262 .read_iter
= tun_chr_read_iter
,
3263 .write_iter
= tun_chr_write_iter
,
3264 .poll
= tun_chr_poll
,
3265 .unlocked_ioctl
= tun_chr_ioctl
,
3266 #ifdef CONFIG_COMPAT
3267 .compat_ioctl
= tun_chr_compat_ioctl
,
3269 .open
= tun_chr_open
,
3270 .release
= tun_chr_close
,
3271 .fasync
= tun_chr_fasync
,
3272 #ifdef CONFIG_PROC_FS
3273 .show_fdinfo
= tun_chr_show_fdinfo
,
3277 static struct miscdevice tun_miscdev
= {
3280 .nodename
= "net/tun",
3284 /* ethtool interface */
3286 static int tun_get_link_ksettings(struct net_device
*dev
,
3287 struct ethtool_link_ksettings
*cmd
)
3289 ethtool_link_ksettings_zero_link_mode(cmd
, supported
);
3290 ethtool_link_ksettings_zero_link_mode(cmd
, advertising
);
3291 cmd
->base
.speed
= SPEED_10
;
3292 cmd
->base
.duplex
= DUPLEX_FULL
;
3293 cmd
->base
.port
= PORT_TP
;
3294 cmd
->base
.phy_address
= 0;
3295 cmd
->base
.autoneg
= AUTONEG_DISABLE
;
3299 static void tun_get_drvinfo(struct net_device
*dev
, struct ethtool_drvinfo
*info
)
3301 struct tun_struct
*tun
= netdev_priv(dev
);
3303 strlcpy(info
->driver
, DRV_NAME
, sizeof(info
->driver
));
3304 strlcpy(info
->version
, DRV_VERSION
, sizeof(info
->version
));
3306 switch (tun
->flags
& TUN_TYPE_MASK
) {
3308 strlcpy(info
->bus_info
, "tun", sizeof(info
->bus_info
));
3311 strlcpy(info
->bus_info
, "tap", sizeof(info
->bus_info
));
3316 static u32
tun_get_msglevel(struct net_device
*dev
)
3319 struct tun_struct
*tun
= netdev_priv(dev
);
3326 static void tun_set_msglevel(struct net_device
*dev
, u32 value
)
3329 struct tun_struct
*tun
= netdev_priv(dev
);
3334 static int tun_get_coalesce(struct net_device
*dev
,
3335 struct ethtool_coalesce
*ec
)
3337 struct tun_struct
*tun
= netdev_priv(dev
);
3339 ec
->rx_max_coalesced_frames
= tun
->rx_batched
;
3344 static int tun_set_coalesce(struct net_device
*dev
,
3345 struct ethtool_coalesce
*ec
)
3347 struct tun_struct
*tun
= netdev_priv(dev
);
3349 if (ec
->rx_max_coalesced_frames
> NAPI_POLL_WEIGHT
)
3350 tun
->rx_batched
= NAPI_POLL_WEIGHT
;
3352 tun
->rx_batched
= ec
->rx_max_coalesced_frames
;
3357 static const struct ethtool_ops tun_ethtool_ops
= {
3358 .get_drvinfo
= tun_get_drvinfo
,
3359 .get_msglevel
= tun_get_msglevel
,
3360 .set_msglevel
= tun_set_msglevel
,
3361 .get_link
= ethtool_op_get_link
,
3362 .get_ts_info
= ethtool_op_get_ts_info
,
3363 .get_coalesce
= tun_get_coalesce
,
3364 .set_coalesce
= tun_set_coalesce
,
3365 .get_link_ksettings
= tun_get_link_ksettings
,
3368 static int tun_queue_resize(struct tun_struct
*tun
)
3370 struct net_device
*dev
= tun
->dev
;
3371 struct tun_file
*tfile
;
3372 struct ptr_ring
**rings
;
3373 int n
= tun
->numqueues
+ tun
->numdisabled
;
3376 rings
= kmalloc_array(n
, sizeof(*rings
), GFP_KERNEL
);
3380 for (i
= 0; i
< tun
->numqueues
; i
++) {
3381 tfile
= rtnl_dereference(tun
->tfiles
[i
]);
3382 rings
[i
] = &tfile
->tx_ring
;
3384 list_for_each_entry(tfile
, &tun
->disabled
, next
)
3385 rings
[i
++] = &tfile
->tx_ring
;
3387 ret
= ptr_ring_resize_multiple(rings
, n
,
3388 dev
->tx_queue_len
, GFP_KERNEL
,
3395 static int tun_device_event(struct notifier_block
*unused
,
3396 unsigned long event
, void *ptr
)
3398 struct net_device
*dev
= netdev_notifier_info_to_dev(ptr
);
3399 struct tun_struct
*tun
= netdev_priv(dev
);
3401 if (dev
->rtnl_link_ops
!= &tun_link_ops
)
3405 case NETDEV_CHANGE_TX_QUEUE_LEN
:
3406 if (tun_queue_resize(tun
))
3416 static struct notifier_block tun_notifier_block __read_mostly
= {
3417 .notifier_call
= tun_device_event
,
3420 static int __init
tun_init(void)
3424 pr_info("%s, %s\n", DRV_DESCRIPTION
, DRV_VERSION
);
3426 ret
= rtnl_link_register(&tun_link_ops
);
3428 pr_err("Can't register link_ops\n");
3432 ret
= misc_register(&tun_miscdev
);
3434 pr_err("Can't register misc device %d\n", TUN_MINOR
);
3438 ret
= register_netdevice_notifier(&tun_notifier_block
);
3440 pr_err("Can't register netdevice notifier\n");
3447 misc_deregister(&tun_miscdev
);
3449 rtnl_link_unregister(&tun_link_ops
);
3454 static void tun_cleanup(void)
3456 misc_deregister(&tun_miscdev
);
3457 rtnl_link_unregister(&tun_link_ops
);
3458 unregister_netdevice_notifier(&tun_notifier_block
);
3461 /* Get an underlying socket object from tun file. Returns error unless file is
3462 * attached to a device. The returned object works like a packet socket, it
3463 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
3464 * holding a reference to the file for as long as the socket is in use. */
3465 struct socket
*tun_get_socket(struct file
*file
)
3467 struct tun_file
*tfile
;
3468 if (file
->f_op
!= &tun_fops
)
3469 return ERR_PTR(-EINVAL
);
3470 tfile
= file
->private_data
;
3472 return ERR_PTR(-EBADFD
);
3473 return &tfile
->socket
;
3475 EXPORT_SYMBOL_GPL(tun_get_socket
);
3477 struct ptr_ring
*tun_get_tx_ring(struct file
*file
)
3479 struct tun_file
*tfile
;
3481 if (file
->f_op
!= &tun_fops
)
3482 return ERR_PTR(-EINVAL
);
3483 tfile
= file
->private_data
;
3485 return ERR_PTR(-EBADFD
);
3486 return &tfile
->tx_ring
;
3488 EXPORT_SYMBOL_GPL(tun_get_tx_ring
);
3490 module_init(tun_init
);
3491 module_exit(tun_cleanup
);
3492 MODULE_DESCRIPTION(DRV_DESCRIPTION
);
3493 MODULE_AUTHOR(DRV_COPYRIGHT
);
3494 MODULE_LICENSE("GPL");
3495 MODULE_ALIAS_MISCDEV(TUN_MINOR
);
3496 MODULE_ALIAS("devname:net/tun");