1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * TUN - Universal TUN/TAP device driver.
4 * Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
6 * $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
12 * Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
13 * Add TUNSETLINK ioctl to set the link encapsulation
15 * Mark Smith <markzzzsmith@yahoo.com.au>
16 * Use eth_random_addr() for tap MAC address.
18 * Harald Roelle <harald.roelle@ifi.lmu.de> 2004/04/20
19 * Fixes in packet dropping, queue length setting and queue wakeup.
20 * Increased default tx queue length.
24 * Daniel Podlejski <underley@underley.eu.org>
25 * Modifications for 2.3.99-pre5 kernel.
28 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
30 #define DRV_NAME "tun"
31 #define DRV_VERSION "1.6"
32 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
33 #define DRV_COPYRIGHT "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
35 #include <linux/module.h>
36 #include <linux/errno.h>
37 #include <linux/kernel.h>
38 #include <linux/sched/signal.h>
39 #include <linux/major.h>
40 #include <linux/slab.h>
41 #include <linux/poll.h>
42 #include <linux/fcntl.h>
43 #include <linux/init.h>
44 #include <linux/skbuff.h>
45 #include <linux/netdevice.h>
46 #include <linux/etherdevice.h>
47 #include <linux/miscdevice.h>
48 #include <linux/ethtool.h>
49 #include <linux/rtnetlink.h>
50 #include <linux/compat.h>
52 #include <linux/if_arp.h>
53 #include <linux/if_ether.h>
54 #include <linux/if_tun.h>
55 #include <linux/if_vlan.h>
56 #include <linux/crc32.h>
57 #include <linux/nsproxy.h>
58 #include <linux/virtio_net.h>
59 #include <linux/rcupdate.h>
60 #include <net/net_namespace.h>
61 #include <net/netns/generic.h>
62 #include <net/rtnetlink.h>
65 #include <linux/seq_file.h>
66 #include <linux/uio.h>
67 #include <linux/skb_array.h>
68 #include <linux/bpf.h>
69 #include <linux/bpf_trace.h>
70 #include <linux/mutex.h>
72 #include <linux/uaccess.h>
73 #include <linux/proc_fs.h>
75 static void tun_default_link_ksettings(struct net_device *dev,
76 struct ethtool_link_ksettings *cmd);
78 /* Uncomment to enable debugging */
79 /* #define TUN_DEBUG 1 */
84 #define tun_debug(level, tun, fmt, args...) \
87 netdev_printk(level, tun->dev, fmt, ##args); \
89 #define DBG1(level, fmt, args...) \
92 printk(level fmt, ##args); \
95 #define tun_debug(level, tun, fmt, args...) \
98 netdev_printk(level, tun->dev, fmt, ##args); \
100 #define DBG1(level, fmt, args...) \
103 printk(level fmt, ##args); \
107 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
109 /* TUN device flags */
111 /* IFF_ATTACH_QUEUE is never stored in device flags,
112 * overload it to mean fasync when stored there.
114 #define TUN_FASYNC IFF_ATTACH_QUEUE
115 /* High bits in flags field are unused. */
116 #define TUN_VNET_LE 0x80000000
117 #define TUN_VNET_BE 0x40000000
119 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
120 IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
122 #define GOODCOPY_LEN 128
124 #define FLT_EXACT_COUNT 8
126 unsigned int count; /* Number of addrs. Zero means disabled */
127 u32 mask[2]; /* Mask of the hashed addrs */
128 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
131 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
132 * to max number of VCPUs in guest. */
133 #define MAX_TAP_QUEUES 256
134 #define MAX_TAP_FLOWS 4096
136 #define TUN_FLOW_EXPIRE (3 * HZ)
138 struct tun_pcpu_stats {
143 struct u64_stats_sync syncp;
149 /* A tun_file connects an open character device to a tuntap netdevice. It
150 * also contains all socket related structures (except sock_fprog and tap_filter)
151 * to serve as one transmit queue for tuntap device. The sock_fprog and
152 * tap_filter were kept in tun_struct since they were used for filtering for the
153 * netdevice not for a specific queue (at least I didn't see the requirement for
157 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
158 * other can only be read while rcu_read_lock or rtnl_lock is held.
162 struct socket socket;
163 struct tun_struct __rcu *tun;
164 struct fasync_struct *fasync;
165 /* only used for fasnyc */
169 unsigned int ifindex;
171 struct napi_struct napi;
173 bool napi_frags_enabled;
174 struct mutex napi_mutex; /* Protects access to the above napi */
175 struct list_head next;
176 struct tun_struct *detached;
177 struct ptr_ring tx_ring;
178 struct xdp_rxq_info xdp_rxq;
186 struct tun_flow_entry {
187 struct hlist_node hash_link;
189 struct tun_struct *tun;
194 unsigned long updated ____cacheline_aligned_in_smp;
197 #define TUN_NUM_FLOW_ENTRIES 1024
198 #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
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;
244 struct ethtool_link_ksettings link_ksettings;
252 bool tun_is_xdp_frame(void *ptr)
254 return (unsigned long)ptr & TUN_XDP_FLAG;
256 EXPORT_SYMBOL(tun_is_xdp_frame);
258 void *tun_xdp_to_ptr(void *ptr)
260 return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
262 EXPORT_SYMBOL(tun_xdp_to_ptr);
264 void *tun_ptr_to_xdp(void *ptr)
266 return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
268 EXPORT_SYMBOL(tun_ptr_to_xdp);
270 static int tun_napi_receive(struct napi_struct *napi, int budget)
272 struct tun_file *tfile = container_of(napi, struct tun_file, napi);
273 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
274 struct sk_buff_head process_queue;
278 __skb_queue_head_init(&process_queue);
280 spin_lock(&queue->lock);
281 skb_queue_splice_tail_init(queue, &process_queue);
282 spin_unlock(&queue->lock);
284 while (received < budget && (skb = __skb_dequeue(&process_queue))) {
285 napi_gro_receive(napi, skb);
289 if (!skb_queue_empty(&process_queue)) {
290 spin_lock(&queue->lock);
291 skb_queue_splice(&process_queue, queue);
292 spin_unlock(&queue->lock);
298 static int tun_napi_poll(struct napi_struct *napi, int budget)
300 unsigned int received;
302 received = tun_napi_receive(napi, budget);
304 if (received < budget)
305 napi_complete_done(napi, received);
310 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
311 bool napi_en, bool napi_frags)
313 tfile->napi_enabled = napi_en;
314 tfile->napi_frags_enabled = napi_en && napi_frags;
316 netif_tx_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
318 napi_enable(&tfile->napi);
322 static void tun_napi_disable(struct tun_file *tfile)
324 if (tfile->napi_enabled)
325 napi_disable(&tfile->napi);
328 static void tun_napi_del(struct tun_file *tfile)
330 if (tfile->napi_enabled)
331 netif_napi_del(&tfile->napi);
334 static bool tun_napi_frags_enabled(const struct tun_file *tfile)
336 return tfile->napi_frags_enabled;
339 #ifdef CONFIG_TUN_VNET_CROSS_LE
340 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
342 return tun->flags & TUN_VNET_BE ? false :
343 virtio_legacy_is_little_endian();
346 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
348 int be = !!(tun->flags & TUN_VNET_BE);
350 if (put_user(be, argp))
356 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
360 if (get_user(be, argp))
364 tun->flags |= TUN_VNET_BE;
366 tun->flags &= ~TUN_VNET_BE;
371 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
373 return virtio_legacy_is_little_endian();
376 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
381 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
385 #endif /* CONFIG_TUN_VNET_CROSS_LE */
387 static inline bool tun_is_little_endian(struct tun_struct *tun)
389 return tun->flags & TUN_VNET_LE ||
390 tun_legacy_is_little_endian(tun);
393 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
395 return __virtio16_to_cpu(tun_is_little_endian(tun), val);
398 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
400 return __cpu_to_virtio16(tun_is_little_endian(tun), val);
403 static inline u32 tun_hashfn(u32 rxhash)
405 return rxhash & TUN_MASK_FLOW_ENTRIES;
408 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
410 struct tun_flow_entry *e;
412 hlist_for_each_entry_rcu(e, head, hash_link) {
413 if (e->rxhash == rxhash)
419 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
420 struct hlist_head *head,
421 u32 rxhash, u16 queue_index)
423 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
426 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
427 rxhash, queue_index);
428 e->updated = jiffies;
431 e->queue_index = queue_index;
433 hlist_add_head_rcu(&e->hash_link, head);
439 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
441 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
442 e->rxhash, e->queue_index);
443 hlist_del_rcu(&e->hash_link);
448 static void tun_flow_flush(struct tun_struct *tun)
452 spin_lock_bh(&tun->lock);
453 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
454 struct tun_flow_entry *e;
455 struct hlist_node *n;
457 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
458 tun_flow_delete(tun, e);
460 spin_unlock_bh(&tun->lock);
463 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
467 spin_lock_bh(&tun->lock);
468 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
469 struct tun_flow_entry *e;
470 struct hlist_node *n;
472 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
473 if (e->queue_index == queue_index)
474 tun_flow_delete(tun, e);
477 spin_unlock_bh(&tun->lock);
480 static void tun_flow_cleanup(struct timer_list *t)
482 struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
483 unsigned long delay = tun->ageing_time;
484 unsigned long next_timer = jiffies + delay;
485 unsigned long count = 0;
488 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
490 spin_lock(&tun->lock);
491 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
492 struct tun_flow_entry *e;
493 struct hlist_node *n;
495 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
496 unsigned long this_timer;
498 this_timer = e->updated + delay;
499 if (time_before_eq(this_timer, jiffies)) {
500 tun_flow_delete(tun, e);
504 if (time_before(this_timer, next_timer))
505 next_timer = this_timer;
510 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
511 spin_unlock(&tun->lock);
514 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
515 struct tun_file *tfile)
517 struct hlist_head *head;
518 struct tun_flow_entry *e;
519 unsigned long delay = tun->ageing_time;
520 u16 queue_index = tfile->queue_index;
522 head = &tun->flows[tun_hashfn(rxhash)];
526 e = tun_flow_find(head, rxhash);
528 /* TODO: keep queueing to old queue until it's empty? */
529 if (READ_ONCE(e->queue_index) != queue_index)
530 WRITE_ONCE(e->queue_index, queue_index);
531 if (e->updated != jiffies)
532 e->updated = jiffies;
533 sock_rps_record_flow_hash(e->rps_rxhash);
535 spin_lock_bh(&tun->lock);
536 if (!tun_flow_find(head, rxhash) &&
537 tun->flow_count < MAX_TAP_FLOWS)
538 tun_flow_create(tun, head, rxhash, queue_index);
540 if (!timer_pending(&tun->flow_gc_timer))
541 mod_timer(&tun->flow_gc_timer,
542 round_jiffies_up(jiffies + delay));
543 spin_unlock_bh(&tun->lock);
550 * Save the hash received in the stack receive path and update the
551 * flow_hash table accordingly.
553 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
555 if (unlikely(e->rps_rxhash != hash))
556 e->rps_rxhash = hash;
559 /* We try to identify a flow through its rxhash. The reason that
560 * we do not check rxq no. is because some cards(e.g 82599), chooses
561 * the rxq based on the txq where the last packet of the flow comes. As
562 * the userspace application move between processors, we may get a
563 * different rxq no. here.
565 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
567 struct tun_flow_entry *e;
571 numqueues = READ_ONCE(tun->numqueues);
573 txq = __skb_get_hash_symmetric(skb);
574 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
576 tun_flow_save_rps_rxhash(e, txq);
577 txq = e->queue_index;
579 /* use multiply and shift instead of expensive divide */
580 txq = ((u64)txq * numqueues) >> 32;
586 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
588 struct tun_prog *prog;
592 numqueues = READ_ONCE(tun->numqueues);
596 prog = rcu_dereference(tun->steering_prog);
598 ret = bpf_prog_run_clear_cb(prog->prog, skb);
600 return ret % numqueues;
603 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
604 struct net_device *sb_dev)
606 struct tun_struct *tun = netdev_priv(dev);
610 if (rcu_dereference(tun->steering_prog))
611 ret = tun_ebpf_select_queue(tun, skb);
613 ret = tun_automq_select_queue(tun, skb);
619 static inline bool tun_not_capable(struct tun_struct *tun)
621 const struct cred *cred = current_cred();
622 struct net *net = dev_net(tun->dev);
624 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
625 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
626 !ns_capable(net->user_ns, CAP_NET_ADMIN);
629 static void tun_set_real_num_queues(struct tun_struct *tun)
631 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
632 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
635 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
637 tfile->detached = tun;
638 list_add_tail(&tfile->next, &tun->disabled);
642 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
644 struct tun_struct *tun = tfile->detached;
646 tfile->detached = NULL;
647 list_del_init(&tfile->next);
652 void tun_ptr_free(void *ptr)
656 if (tun_is_xdp_frame(ptr)) {
657 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
659 xdp_return_frame(xdpf);
661 __skb_array_destroy_skb(ptr);
664 EXPORT_SYMBOL_GPL(tun_ptr_free);
666 static void tun_queue_purge(struct tun_file *tfile)
670 while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
673 skb_queue_purge(&tfile->sk.sk_write_queue);
674 skb_queue_purge(&tfile->sk.sk_error_queue);
677 static void __tun_detach(struct tun_file *tfile, bool clean)
679 struct tun_file *ntfile;
680 struct tun_struct *tun;
682 tun = rtnl_dereference(tfile->tun);
685 tun_napi_disable(tfile);
689 if (tun && !tfile->detached) {
690 u16 index = tfile->queue_index;
691 BUG_ON(index >= tun->numqueues);
693 rcu_assign_pointer(tun->tfiles[index],
694 tun->tfiles[tun->numqueues - 1]);
695 ntfile = rtnl_dereference(tun->tfiles[index]);
696 ntfile->queue_index = index;
697 rcu_assign_pointer(tun->tfiles[tun->numqueues - 1],
702 RCU_INIT_POINTER(tfile->tun, NULL);
703 sock_put(&tfile->sk);
705 tun_disable_queue(tun, tfile);
708 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
709 /* Drop read queue */
710 tun_queue_purge(tfile);
711 tun_set_real_num_queues(tun);
712 } else if (tfile->detached && clean) {
713 tun = tun_enable_queue(tfile);
714 sock_put(&tfile->sk);
718 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
719 netif_carrier_off(tun->dev);
721 if (!(tun->flags & IFF_PERSIST) &&
722 tun->dev->reg_state == NETREG_REGISTERED)
723 unregister_netdevice(tun->dev);
726 xdp_rxq_info_unreg(&tfile->xdp_rxq);
727 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
728 sock_put(&tfile->sk);
732 static void tun_detach(struct tun_file *tfile, bool clean)
734 struct tun_struct *tun;
735 struct net_device *dev;
738 tun = rtnl_dereference(tfile->tun);
739 dev = tun ? tun->dev : NULL;
740 __tun_detach(tfile, clean);
742 netdev_state_change(dev);
746 static void tun_detach_all(struct net_device *dev)
748 struct tun_struct *tun = netdev_priv(dev);
749 struct tun_file *tfile, *tmp;
750 int i, n = tun->numqueues;
752 for (i = 0; i < n; i++) {
753 tfile = rtnl_dereference(tun->tfiles[i]);
755 tun_napi_disable(tfile);
756 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
757 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
758 RCU_INIT_POINTER(tfile->tun, NULL);
761 list_for_each_entry(tfile, &tun->disabled, next) {
762 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
763 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
764 RCU_INIT_POINTER(tfile->tun, NULL);
766 BUG_ON(tun->numqueues != 0);
769 for (i = 0; i < n; i++) {
770 tfile = rtnl_dereference(tun->tfiles[i]);
772 /* Drop read queue */
773 tun_queue_purge(tfile);
774 xdp_rxq_info_unreg(&tfile->xdp_rxq);
775 sock_put(&tfile->sk);
777 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
778 tun_enable_queue(tfile);
779 tun_queue_purge(tfile);
780 xdp_rxq_info_unreg(&tfile->xdp_rxq);
781 sock_put(&tfile->sk);
783 BUG_ON(tun->numdisabled != 0);
785 if (tun->flags & IFF_PERSIST)
786 module_put(THIS_MODULE);
789 static int tun_attach(struct tun_struct *tun, struct file *file,
790 bool skip_filter, bool napi, bool napi_frags,
793 struct tun_file *tfile = file->private_data;
794 struct net_device *dev = tun->dev;
797 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
802 if (rtnl_dereference(tfile->tun) && !tfile->detached)
806 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
810 if (!tfile->detached &&
811 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
816 /* Re-attach the filter to persist device */
817 if (!skip_filter && (tun->filter_attached == true)) {
818 lock_sock(tfile->socket.sk);
819 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
820 release_sock(tfile->socket.sk);
825 if (!tfile->detached &&
826 ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
827 GFP_KERNEL, tun_ptr_free)) {
832 tfile->queue_index = tun->numqueues;
833 tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
835 if (tfile->detached) {
836 /* Re-attach detached tfile, updating XDP queue_index */
837 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
839 if (tfile->xdp_rxq.queue_index != tfile->queue_index)
840 tfile->xdp_rxq.queue_index = tfile->queue_index;
842 /* Setup XDP RX-queue info, for new tfile getting attached */
843 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
844 tun->dev, tfile->queue_index);
847 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
848 MEM_TYPE_PAGE_SHARED, NULL);
850 xdp_rxq_info_unreg(&tfile->xdp_rxq);
856 if (tfile->detached) {
857 tun_enable_queue(tfile);
859 sock_hold(&tfile->sk);
860 tun_napi_init(tun, tfile, napi, napi_frags);
863 if (rtnl_dereference(tun->xdp_prog))
864 sock_set_flag(&tfile->sk, SOCK_XDP);
866 /* device is allowed to go away first, so no need to hold extra
870 /* Publish tfile->tun and tun->tfiles only after we've fully
871 * initialized tfile; otherwise we risk using half-initialized
875 rcu_assign_pointer(tfile->tun, tun);
876 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
878 tun_set_real_num_queues(tun);
883 static struct tun_struct *tun_get(struct tun_file *tfile)
885 struct tun_struct *tun;
888 tun = rcu_dereference(tfile->tun);
896 static void tun_put(struct tun_struct *tun)
902 static void addr_hash_set(u32 *mask, const u8 *addr)
904 int n = ether_crc(ETH_ALEN, addr) >> 26;
905 mask[n >> 5] |= (1 << (n & 31));
908 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
910 int n = ether_crc(ETH_ALEN, addr) >> 26;
911 return mask[n >> 5] & (1 << (n & 31));
914 static int update_filter(struct tap_filter *filter, void __user *arg)
916 struct { u8 u[ETH_ALEN]; } *addr;
917 struct tun_filter uf;
918 int err, alen, n, nexact;
920 if (copy_from_user(&uf, arg, sizeof(uf)))
929 alen = ETH_ALEN * uf.count;
930 addr = memdup_user(arg + sizeof(uf), alen);
932 return PTR_ERR(addr);
934 /* The filter is updated without holding any locks. Which is
935 * perfectly safe. We disable it first and in the worst
936 * case we'll accept a few undesired packets. */
940 /* Use first set of addresses as an exact filter */
941 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
942 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
946 /* Remaining multicast addresses are hashed,
947 * unicast will leave the filter disabled. */
948 memset(filter->mask, 0, sizeof(filter->mask));
949 for (; n < uf.count; n++) {
950 if (!is_multicast_ether_addr(addr[n].u)) {
951 err = 0; /* no filter */
954 addr_hash_set(filter->mask, addr[n].u);
957 /* For ALLMULTI just set the mask to all ones.
958 * This overrides the mask populated above. */
959 if ((uf.flags & TUN_FLT_ALLMULTI))
960 memset(filter->mask, ~0, sizeof(filter->mask));
962 /* Now enable the filter */
964 filter->count = nexact;
966 /* Return the number of exact filters */
973 /* Returns: 0 - drop, !=0 - accept */
974 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
976 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
978 struct ethhdr *eh = (struct ethhdr *) skb->data;
982 for (i = 0; i < filter->count; i++)
983 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
986 /* Inexact match (multicast only) */
987 if (is_multicast_ether_addr(eh->h_dest))
988 return addr_hash_test(filter->mask, eh->h_dest);
994 * Checks whether the packet is accepted or not.
995 * Returns: 0 - drop, !=0 - accept
997 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
1002 return run_filter(filter, skb);
1005 /* Network device part of the driver */
1007 static const struct ethtool_ops tun_ethtool_ops;
1009 /* Net device detach from fd. */
1010 static void tun_net_uninit(struct net_device *dev)
1012 tun_detach_all(dev);
1015 /* Net device open. */
1016 static int tun_net_open(struct net_device *dev)
1018 netif_tx_start_all_queues(dev);
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_branch_unlikely(&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.
1038 struct tun_flow_entry *e;
1041 rxhash = __skb_get_hash_symmetric(skb);
1042 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
1044 tun_flow_save_rps_rxhash(e, rxhash);
1049 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1050 struct sk_buff *skb,
1053 struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1056 len = bpf_prog_run_clear_cb(prog->prog, skb);
1061 /* Net device start xmit */
1062 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1064 struct tun_struct *tun = netdev_priv(dev);
1065 int txq = skb->queue_mapping;
1066 struct tun_file *tfile;
1070 tfile = rcu_dereference(tun->tfiles[txq]);
1072 /* Drop packet if interface is not attached */
1076 if (!rcu_dereference(tun->steering_prog))
1077 tun_automq_xmit(tun, skb);
1079 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
1083 /* Drop if the filter does not like it.
1084 * This is a noop if the filter is disabled.
1085 * Filter can be enabled only for the TAP devices. */
1086 if (!check_filter(&tun->txflt, skb))
1089 if (tfile->socket.sk->sk_filter &&
1090 sk_filter(tfile->socket.sk, skb))
1093 len = run_ebpf_filter(tun, skb, len);
1094 if (len == 0 || pskb_trim(skb, len))
1097 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1100 skb_tx_timestamp(skb);
1102 /* Orphan the skb - required as we might hang on to it
1103 * for indefinite time.
1109 if (ptr_ring_produce(&tfile->tx_ring, skb))
1112 /* Notify and wake up reader process */
1113 if (tfile->flags & TUN_FASYNC)
1114 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1115 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1118 return NETDEV_TX_OK;
1121 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1125 return NET_XMIT_DROP;
1128 static void tun_net_mclist(struct net_device *dev)
1131 * This callback is supposed to deal with mc filter in
1132 * _rx_ path and has nothing to do with the _tx_ path.
1133 * In rx path we always accept everything userspace gives us.
1137 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1138 netdev_features_t features)
1140 struct tun_struct *tun = netdev_priv(dev);
1142 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1145 static void tun_set_headroom(struct net_device *dev, int new_hr)
1147 struct tun_struct *tun = netdev_priv(dev);
1149 if (new_hr < NET_SKB_PAD)
1150 new_hr = NET_SKB_PAD;
1152 tun->align = new_hr;
1156 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1158 u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1159 struct tun_struct *tun = netdev_priv(dev);
1160 struct tun_pcpu_stats *p;
1163 for_each_possible_cpu(i) {
1164 u64 rxpackets, rxbytes, txpackets, txbytes;
1167 p = per_cpu_ptr(tun->pcpu_stats, i);
1169 start = u64_stats_fetch_begin(&p->syncp);
1170 rxpackets = p->rx_packets;
1171 rxbytes = p->rx_bytes;
1172 txpackets = p->tx_packets;
1173 txbytes = p->tx_bytes;
1174 } while (u64_stats_fetch_retry(&p->syncp, start));
1176 stats->rx_packets += rxpackets;
1177 stats->rx_bytes += rxbytes;
1178 stats->tx_packets += txpackets;
1179 stats->tx_bytes += txbytes;
1182 rx_dropped += p->rx_dropped;
1183 rx_frame_errors += p->rx_frame_errors;
1184 tx_dropped += p->tx_dropped;
1186 stats->rx_dropped = rx_dropped;
1187 stats->rx_frame_errors = rx_frame_errors;
1188 stats->tx_dropped = tx_dropped;
1191 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1192 struct netlink_ext_ack *extack)
1194 struct tun_struct *tun = netdev_priv(dev);
1195 struct tun_file *tfile;
1196 struct bpf_prog *old_prog;
1199 old_prog = rtnl_dereference(tun->xdp_prog);
1200 rcu_assign_pointer(tun->xdp_prog, prog);
1202 bpf_prog_put(old_prog);
1204 for (i = 0; i < tun->numqueues; i++) {
1205 tfile = rtnl_dereference(tun->tfiles[i]);
1207 sock_set_flag(&tfile->sk, SOCK_XDP);
1209 sock_reset_flag(&tfile->sk, SOCK_XDP);
1211 list_for_each_entry(tfile, &tun->disabled, next) {
1213 sock_set_flag(&tfile->sk, SOCK_XDP);
1215 sock_reset_flag(&tfile->sk, SOCK_XDP);
1221 static u32 tun_xdp_query(struct net_device *dev)
1223 struct tun_struct *tun = netdev_priv(dev);
1224 const struct bpf_prog *xdp_prog;
1226 xdp_prog = rtnl_dereference(tun->xdp_prog);
1228 return xdp_prog->aux->id;
1233 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1235 switch (xdp->command) {
1236 case XDP_SETUP_PROG:
1237 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1238 case XDP_QUERY_PROG:
1239 xdp->prog_id = tun_xdp_query(dev);
1246 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1249 struct tun_struct *tun = netdev_priv(dev);
1251 if (!tun->numqueues)
1254 netif_carrier_on(dev);
1256 netif_carrier_off(dev);
1261 static const struct net_device_ops tun_netdev_ops = {
1262 .ndo_uninit = tun_net_uninit,
1263 .ndo_open = tun_net_open,
1264 .ndo_stop = tun_net_close,
1265 .ndo_start_xmit = tun_net_xmit,
1266 .ndo_fix_features = tun_net_fix_features,
1267 .ndo_select_queue = tun_select_queue,
1268 .ndo_set_rx_headroom = tun_set_headroom,
1269 .ndo_get_stats64 = tun_net_get_stats64,
1270 .ndo_change_carrier = tun_net_change_carrier,
1273 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1275 /* Notify and wake up reader process */
1276 if (tfile->flags & TUN_FASYNC)
1277 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1278 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1281 static int tun_xdp_xmit(struct net_device *dev, int n,
1282 struct xdp_frame **frames, u32 flags)
1284 struct tun_struct *tun = netdev_priv(dev);
1285 struct tun_file *tfile;
1291 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1297 numqueues = READ_ONCE(tun->numqueues);
1300 return -ENXIO; /* Caller will free/return all frames */
1303 tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1305 if (unlikely(!tfile))
1308 spin_lock(&tfile->tx_ring.producer_lock);
1309 for (i = 0; i < n; i++) {
1310 struct xdp_frame *xdp = frames[i];
1311 /* Encode the XDP flag into lowest bit for consumer to differ
1312 * XDP buffer from sk_buff.
1314 void *frame = tun_xdp_to_ptr(xdp);
1316 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1317 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1318 xdp_return_frame_rx_napi(xdp);
1322 spin_unlock(&tfile->tx_ring.producer_lock);
1324 if (flags & XDP_XMIT_FLUSH)
1325 __tun_xdp_flush_tfile(tfile);
1331 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1333 struct xdp_frame *frame = convert_to_xdp_frame(xdp);
1335 if (unlikely(!frame))
1338 return tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1341 static const struct net_device_ops tap_netdev_ops = {
1342 .ndo_uninit = tun_net_uninit,
1343 .ndo_open = tun_net_open,
1344 .ndo_stop = tun_net_close,
1345 .ndo_start_xmit = tun_net_xmit,
1346 .ndo_fix_features = tun_net_fix_features,
1347 .ndo_set_rx_mode = tun_net_mclist,
1348 .ndo_set_mac_address = eth_mac_addr,
1349 .ndo_validate_addr = eth_validate_addr,
1350 .ndo_select_queue = tun_select_queue,
1351 .ndo_features_check = passthru_features_check,
1352 .ndo_set_rx_headroom = tun_set_headroom,
1353 .ndo_get_stats64 = tun_net_get_stats64,
1355 .ndo_xdp_xmit = tun_xdp_xmit,
1356 .ndo_change_carrier = tun_net_change_carrier,
1359 static void tun_flow_init(struct tun_struct *tun)
1363 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1364 INIT_HLIST_HEAD(&tun->flows[i]);
1366 tun->ageing_time = TUN_FLOW_EXPIRE;
1367 timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1368 mod_timer(&tun->flow_gc_timer,
1369 round_jiffies_up(jiffies + tun->ageing_time));
1372 static void tun_flow_uninit(struct tun_struct *tun)
1374 del_timer_sync(&tun->flow_gc_timer);
1375 tun_flow_flush(tun);
1379 #define MAX_MTU 65535
1381 /* Initialize net device. */
1382 static void tun_net_init(struct net_device *dev)
1384 struct tun_struct *tun = netdev_priv(dev);
1386 switch (tun->flags & TUN_TYPE_MASK) {
1388 dev->netdev_ops = &tun_netdev_ops;
1390 /* Point-to-Point TUN Device */
1391 dev->hard_header_len = 0;
1395 /* Zero header length */
1396 dev->type = ARPHRD_NONE;
1397 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1401 dev->netdev_ops = &tap_netdev_ops;
1402 /* Ethernet TAP Device */
1404 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1405 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1407 eth_hw_addr_random(dev);
1412 dev->min_mtu = MIN_MTU;
1413 dev->max_mtu = MAX_MTU - dev->hard_header_len;
1416 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1418 struct sock *sk = tfile->socket.sk;
1420 return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1423 /* Character device part */
1426 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1428 struct tun_file *tfile = file->private_data;
1429 struct tun_struct *tun = tun_get(tfile);
1436 sk = tfile->socket.sk;
1438 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1440 poll_wait(file, sk_sleep(sk), wait);
1442 if (!ptr_ring_empty(&tfile->tx_ring))
1443 mask |= EPOLLIN | EPOLLRDNORM;
1445 /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1446 * guarantee EPOLLOUT to be raised by either here or
1447 * tun_sock_write_space(). Then process could get notification
1448 * after it writes to a down device and meets -EIO.
1450 if (tun_sock_writeable(tun, tfile) ||
1451 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1452 tun_sock_writeable(tun, tfile)))
1453 mask |= EPOLLOUT | EPOLLWRNORM;
1455 if (tun->dev->reg_state != NETREG_REGISTERED)
1462 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1464 const struct iov_iter *it)
1466 struct sk_buff *skb;
1471 if (it->nr_segs > MAX_SKB_FRAGS + 1)
1472 return ERR_PTR(-ENOMEM);
1475 skb = napi_get_frags(&tfile->napi);
1478 return ERR_PTR(-ENOMEM);
1480 linear = iov_iter_single_seg_count(it);
1481 err = __skb_grow(skb, linear);
1486 skb->data_len = len - linear;
1487 skb->truesize += skb->data_len;
1489 for (i = 1; i < it->nr_segs; i++) {
1490 size_t fragsz = it->iov[i].iov_len;
1494 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1498 frag = netdev_alloc_frag(fragsz);
1503 page = virt_to_head_page(frag);
1504 skb_fill_page_desc(skb, i - 1, page,
1505 frag - page_address(page), fragsz);
1510 /* frees skb and all frags allocated with napi_alloc_frag() */
1511 napi_free_frags(&tfile->napi);
1512 return ERR_PTR(err);
1515 /* prepad is the amount to reserve at front. len is length after that.
1516 * linear is a hint as to how much to copy (usually headers). */
1517 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1518 size_t prepad, size_t len,
1519 size_t linear, int noblock)
1521 struct sock *sk = tfile->socket.sk;
1522 struct sk_buff *skb;
1525 /* Under a page? Don't bother with paged skb. */
1526 if (prepad + len < PAGE_SIZE || !linear)
1529 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1532 return ERR_PTR(err);
1534 skb_reserve(skb, prepad);
1535 skb_put(skb, linear);
1536 skb->data_len = len - linear;
1537 skb->len += len - linear;
1542 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1543 struct sk_buff *skb, int more)
1545 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1546 struct sk_buff_head process_queue;
1547 u32 rx_batched = tun->rx_batched;
1550 if (!rx_batched || (!more && skb_queue_empty(queue))) {
1552 skb_record_rx_queue(skb, tfile->queue_index);
1553 netif_receive_skb(skb);
1558 spin_lock(&queue->lock);
1559 if (!more || skb_queue_len(queue) == rx_batched) {
1560 __skb_queue_head_init(&process_queue);
1561 skb_queue_splice_tail_init(queue, &process_queue);
1564 __skb_queue_tail(queue, skb);
1566 spin_unlock(&queue->lock);
1569 struct sk_buff *nskb;
1572 while ((nskb = __skb_dequeue(&process_queue))) {
1573 skb_record_rx_queue(nskb, tfile->queue_index);
1574 netif_receive_skb(nskb);
1576 skb_record_rx_queue(skb, tfile->queue_index);
1577 netif_receive_skb(skb);
1582 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1583 int len, int noblock, bool zerocopy)
1585 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1588 if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1597 if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1598 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1604 static struct sk_buff *__tun_build_skb(struct tun_file *tfile,
1605 struct page_frag *alloc_frag, char *buf,
1606 int buflen, int len, int pad)
1608 struct sk_buff *skb = build_skb(buf, buflen);
1611 return ERR_PTR(-ENOMEM);
1613 skb_reserve(skb, pad);
1615 skb_set_owner_w(skb, tfile->socket.sk);
1617 get_page(alloc_frag->page);
1618 alloc_frag->offset += buflen;
1623 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1624 struct xdp_buff *xdp, u32 act)
1630 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1635 err = tun_xdp_tx(tun->dev, xdp);
1642 bpf_warn_invalid_xdp_action(act);
1645 trace_xdp_exception(tun->dev, xdp_prog, act);
1648 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1655 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1656 struct tun_file *tfile,
1657 struct iov_iter *from,
1658 struct virtio_net_hdr *hdr,
1659 int len, int *skb_xdp)
1661 struct page_frag *alloc_frag = ¤t->task_frag;
1662 struct bpf_prog *xdp_prog;
1663 int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1666 int pad = TUN_RX_PAD;
1670 xdp_prog = rcu_dereference(tun->xdp_prog);
1672 pad += XDP_PACKET_HEADROOM;
1673 buflen += SKB_DATA_ALIGN(len + pad);
1676 alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1677 if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1678 return ERR_PTR(-ENOMEM);
1680 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1681 copied = copy_page_from_iter(alloc_frag->page,
1682 alloc_frag->offset + pad,
1685 return ERR_PTR(-EFAULT);
1687 /* There's a small window that XDP may be set after the check
1688 * of xdp_prog above, this should be rare and for simplicity
1689 * we do XDP on skb in case the headroom is not enough.
1691 if (hdr->gso_type || !xdp_prog) {
1693 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len,
1701 xdp_prog = rcu_dereference(tun->xdp_prog);
1703 struct xdp_buff xdp;
1706 xdp.data_hard_start = buf;
1707 xdp.data = buf + pad;
1708 xdp_set_data_meta_invalid(&xdp);
1709 xdp.data_end = xdp.data + len;
1710 xdp.rxq = &tfile->xdp_rxq;
1712 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1713 if (act == XDP_REDIRECT || act == XDP_TX) {
1714 get_page(alloc_frag->page);
1715 alloc_frag->offset += buflen;
1717 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1719 if (act == XDP_REDIRECT || act == XDP_TX)
1720 put_page(alloc_frag->page);
1724 if (err == XDP_REDIRECT)
1726 if (err != XDP_PASS)
1729 pad = xdp.data - xdp.data_hard_start;
1730 len = xdp.data_end - xdp.data;
1735 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad);
1743 /* Get packet from user space buffer */
1744 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1745 void *msg_control, struct iov_iter *from,
1746 int noblock, bool more)
1748 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1749 struct sk_buff *skb;
1750 size_t total_len = iov_iter_count(from);
1751 size_t len = total_len, align = tun->align, linear;
1752 struct virtio_net_hdr gso = { 0 };
1753 struct tun_pcpu_stats *stats;
1756 bool zerocopy = false;
1760 bool frags = tun_napi_frags_enabled(tfile);
1762 if (!(tun->flags & IFF_NO_PI)) {
1763 if (len < sizeof(pi))
1767 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1771 if (tun->flags & IFF_VNET_HDR) {
1772 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1774 if (len < vnet_hdr_sz)
1778 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1781 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1782 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1783 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1785 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1787 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1790 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1791 align += NET_IP_ALIGN;
1792 if (unlikely(len < ETH_HLEN ||
1793 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1797 good_linear = SKB_MAX_HEAD(align);
1800 struct iov_iter i = *from;
1802 /* There are 256 bytes to be copied in skb, so there is
1803 * enough room for skb expand head in case it is used.
1804 * The rest of the buffer is mapped from userspace.
1806 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1807 if (copylen > good_linear)
1808 copylen = good_linear;
1810 iov_iter_advance(&i, copylen);
1811 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1815 if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1816 /* For the packet that is not easy to be processed
1817 * (e.g gso or jumbo packet), we will do it at after
1818 * skb was created with generic XDP routine.
1820 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1822 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1823 return PTR_ERR(skb);
1830 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1831 linear = good_linear;
1833 linear = tun16_to_cpu(tun, gso.hdr_len);
1837 mutex_lock(&tfile->napi_mutex);
1838 skb = tun_napi_alloc_frags(tfile, copylen, from);
1839 /* tun_napi_alloc_frags() enforces a layout for the skb.
1840 * If zerocopy is enabled, then this layout will be
1841 * overwritten by zerocopy_sg_from_iter().
1845 skb = tun_alloc_skb(tfile, align, copylen, linear,
1850 if (PTR_ERR(skb) != -EAGAIN)
1851 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1853 mutex_unlock(&tfile->napi_mutex);
1854 return PTR_ERR(skb);
1858 err = zerocopy_sg_from_iter(skb, from);
1860 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1865 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1868 tfile->napi.skb = NULL;
1869 mutex_unlock(&tfile->napi_mutex);
1876 if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1877 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1880 tfile->napi.skb = NULL;
1881 mutex_unlock(&tfile->napi_mutex);
1887 switch (tun->flags & TUN_TYPE_MASK) {
1889 if (tun->flags & IFF_NO_PI) {
1890 u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1892 switch (ip_version) {
1894 pi.proto = htons(ETH_P_IP);
1897 pi.proto = htons(ETH_P_IPV6);
1900 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1906 skb_reset_mac_header(skb);
1907 skb->protocol = pi.proto;
1908 skb->dev = tun->dev;
1912 skb->protocol = eth_type_trans(skb, tun->dev);
1916 /* copy skb_ubuf_info for callback when skb has no error */
1918 skb_shinfo(skb)->destructor_arg = msg_control;
1919 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1920 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1921 } else if (msg_control) {
1922 struct ubuf_info *uarg = msg_control;
1923 uarg->callback(uarg, false);
1926 skb_reset_network_header(skb);
1927 skb_probe_transport_header(skb);
1928 skb_record_rx_queue(skb, tfile->queue_index);
1931 struct bpf_prog *xdp_prog;
1936 xdp_prog = rcu_dereference(tun->xdp_prog);
1938 ret = do_xdp_generic(xdp_prog, skb);
1939 if (ret != XDP_PASS) {
1943 tfile->napi.skb = NULL;
1944 mutex_unlock(&tfile->napi_mutex);
1953 /* Compute the costly rx hash only if needed for flow updates.
1954 * We may get a very small possibility of OOO during switching, not
1955 * worth to optimize.
1957 if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1959 rxhash = __skb_get_hash_symmetric(skb);
1962 if (unlikely(!(tun->dev->flags & IFF_UP))) {
1969 /* Exercise flow dissector code path. */
1970 u32 headlen = eth_get_headlen(tun->dev, skb->data,
1973 if (unlikely(headlen > skb_headlen(skb))) {
1974 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1975 napi_free_frags(&tfile->napi);
1977 mutex_unlock(&tfile->napi_mutex);
1983 napi_gro_frags(&tfile->napi);
1985 mutex_unlock(&tfile->napi_mutex);
1986 } else if (tfile->napi_enabled) {
1987 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1990 spin_lock_bh(&queue->lock);
1991 __skb_queue_tail(queue, skb);
1992 queue_len = skb_queue_len(queue);
1993 spin_unlock(&queue->lock);
1995 if (!more || queue_len > NAPI_POLL_WEIGHT)
1996 napi_schedule(&tfile->napi);
1999 } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
2000 tun_rx_batched(tun, tfile, skb, more);
2006 stats = get_cpu_ptr(tun->pcpu_stats);
2007 u64_stats_update_begin(&stats->syncp);
2008 stats->rx_packets++;
2009 stats->rx_bytes += len;
2010 u64_stats_update_end(&stats->syncp);
2014 tun_flow_update(tun, rxhash, tfile);
2019 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
2021 struct file *file = iocb->ki_filp;
2022 struct tun_file *tfile = file->private_data;
2023 struct tun_struct *tun = tun_get(tfile);
2029 result = tun_get_user(tun, tfile, NULL, from,
2030 file->f_flags & O_NONBLOCK, false);
2036 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2037 struct tun_file *tfile,
2038 struct xdp_frame *xdp_frame,
2039 struct iov_iter *iter)
2041 int vnet_hdr_sz = 0;
2042 size_t size = xdp_frame->len;
2043 struct tun_pcpu_stats *stats;
2046 if (tun->flags & IFF_VNET_HDR) {
2047 struct virtio_net_hdr gso = { 0 };
2049 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2050 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2052 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2055 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2058 ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2060 stats = get_cpu_ptr(tun->pcpu_stats);
2061 u64_stats_update_begin(&stats->syncp);
2062 stats->tx_packets++;
2063 stats->tx_bytes += ret;
2064 u64_stats_update_end(&stats->syncp);
2065 put_cpu_ptr(tun->pcpu_stats);
2070 /* Put packet to the user space buffer */
2071 static ssize_t tun_put_user(struct tun_struct *tun,
2072 struct tun_file *tfile,
2073 struct sk_buff *skb,
2074 struct iov_iter *iter)
2076 struct tun_pi pi = { 0, skb->protocol };
2077 struct tun_pcpu_stats *stats;
2079 int vlan_offset = 0;
2081 int vnet_hdr_sz = 0;
2083 if (skb_vlan_tag_present(skb))
2084 vlan_hlen = VLAN_HLEN;
2086 if (tun->flags & IFF_VNET_HDR)
2087 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2089 total = skb->len + vlan_hlen + vnet_hdr_sz;
2091 if (!(tun->flags & IFF_NO_PI)) {
2092 if (iov_iter_count(iter) < sizeof(pi))
2095 total += sizeof(pi);
2096 if (iov_iter_count(iter) < total) {
2097 /* Packet will be striped */
2098 pi.flags |= TUN_PKT_STRIP;
2101 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2106 struct virtio_net_hdr gso;
2108 if (iov_iter_count(iter) < vnet_hdr_sz)
2111 if (virtio_net_hdr_from_skb(skb, &gso,
2112 tun_is_little_endian(tun), true,
2114 struct skb_shared_info *sinfo = skb_shinfo(skb);
2115 pr_err("unexpected GSO type: "
2116 "0x%x, gso_size %d, hdr_len %d\n",
2117 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2118 tun16_to_cpu(tun, gso.hdr_len));
2119 print_hex_dump(KERN_ERR, "tun: ",
2122 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2127 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2130 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2137 veth.h_vlan_proto = skb->vlan_proto;
2138 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2140 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2142 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2143 if (ret || !iov_iter_count(iter))
2146 ret = copy_to_iter(&veth, sizeof(veth), iter);
2147 if (ret != sizeof(veth) || !iov_iter_count(iter))
2151 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2154 /* caller is in process context, */
2155 stats = get_cpu_ptr(tun->pcpu_stats);
2156 u64_stats_update_begin(&stats->syncp);
2157 stats->tx_packets++;
2158 stats->tx_bytes += skb->len + vlan_hlen;
2159 u64_stats_update_end(&stats->syncp);
2160 put_cpu_ptr(tun->pcpu_stats);
2165 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2167 DECLARE_WAITQUEUE(wait, current);
2171 ptr = ptr_ring_consume(&tfile->tx_ring);
2179 add_wait_queue(&tfile->socket.wq.wait, &wait);
2182 set_current_state(TASK_INTERRUPTIBLE);
2183 ptr = ptr_ring_consume(&tfile->tx_ring);
2186 if (signal_pending(current)) {
2187 error = -ERESTARTSYS;
2190 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2198 __set_current_state(TASK_RUNNING);
2199 remove_wait_queue(&tfile->socket.wq.wait, &wait);
2206 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2207 struct iov_iter *to,
2208 int noblock, void *ptr)
2213 tun_debug(KERN_INFO, tun, "tun_do_read\n");
2215 if (!iov_iter_count(to)) {
2221 /* Read frames from ring */
2222 ptr = tun_ring_recv(tfile, noblock, &err);
2227 if (tun_is_xdp_frame(ptr)) {
2228 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2230 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2231 xdp_return_frame(xdpf);
2233 struct sk_buff *skb = ptr;
2235 ret = tun_put_user(tun, tfile, skb, to);
2236 if (unlikely(ret < 0))
2245 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2247 struct file *file = iocb->ki_filp;
2248 struct tun_file *tfile = file->private_data;
2249 struct tun_struct *tun = tun_get(tfile);
2250 ssize_t len = iov_iter_count(to), ret;
2254 ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2255 ret = min_t(ssize_t, ret, len);
2262 static void tun_prog_free(struct rcu_head *rcu)
2264 struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2266 bpf_prog_destroy(prog->prog);
2270 static int __tun_set_ebpf(struct tun_struct *tun,
2271 struct tun_prog __rcu **prog_p,
2272 struct bpf_prog *prog)
2274 struct tun_prog *old, *new = NULL;
2277 new = kmalloc(sizeof(*new), GFP_KERNEL);
2283 spin_lock_bh(&tun->lock);
2284 old = rcu_dereference_protected(*prog_p,
2285 lockdep_is_held(&tun->lock));
2286 rcu_assign_pointer(*prog_p, new);
2287 spin_unlock_bh(&tun->lock);
2290 call_rcu(&old->rcu, tun_prog_free);
2295 static void tun_free_netdev(struct net_device *dev)
2297 struct tun_struct *tun = netdev_priv(dev);
2299 BUG_ON(!(list_empty(&tun->disabled)));
2300 free_percpu(tun->pcpu_stats);
2301 tun_flow_uninit(tun);
2302 security_tun_dev_free_security(tun->security);
2303 __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2304 __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2307 static void tun_setup(struct net_device *dev)
2309 struct tun_struct *tun = netdev_priv(dev);
2311 tun->owner = INVALID_UID;
2312 tun->group = INVALID_GID;
2313 tun_default_link_ksettings(dev, &tun->link_ksettings);
2315 dev->ethtool_ops = &tun_ethtool_ops;
2316 dev->needs_free_netdev = true;
2317 dev->priv_destructor = tun_free_netdev;
2318 /* We prefer our own queue length */
2319 dev->tx_queue_len = TUN_READQ_SIZE;
2322 /* Trivial set of netlink ops to allow deleting tun or tap
2323 * device with netlink.
2325 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2326 struct netlink_ext_ack *extack)
2328 NL_SET_ERR_MSG(extack,
2329 "tun/tap creation via rtnetlink is not supported.");
2333 static size_t tun_get_size(const struct net_device *dev)
2335 BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2336 BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2338 return nla_total_size(sizeof(uid_t)) + /* OWNER */
2339 nla_total_size(sizeof(gid_t)) + /* GROUP */
2340 nla_total_size(sizeof(u8)) + /* TYPE */
2341 nla_total_size(sizeof(u8)) + /* PI */
2342 nla_total_size(sizeof(u8)) + /* VNET_HDR */
2343 nla_total_size(sizeof(u8)) + /* PERSIST */
2344 nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2345 nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2346 nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2350 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2352 struct tun_struct *tun = netdev_priv(dev);
2354 if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2355 goto nla_put_failure;
2356 if (uid_valid(tun->owner) &&
2357 nla_put_u32(skb, IFLA_TUN_OWNER,
2358 from_kuid_munged(current_user_ns(), tun->owner)))
2359 goto nla_put_failure;
2360 if (gid_valid(tun->group) &&
2361 nla_put_u32(skb, IFLA_TUN_GROUP,
2362 from_kgid_munged(current_user_ns(), tun->group)))
2363 goto nla_put_failure;
2364 if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2365 goto nla_put_failure;
2366 if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2367 goto nla_put_failure;
2368 if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2369 goto nla_put_failure;
2370 if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2371 !!(tun->flags & IFF_MULTI_QUEUE)))
2372 goto nla_put_failure;
2373 if (tun->flags & IFF_MULTI_QUEUE) {
2374 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2375 goto nla_put_failure;
2376 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2378 goto nla_put_failure;
2387 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2389 .priv_size = sizeof(struct tun_struct),
2391 .validate = tun_validate,
2392 .get_size = tun_get_size,
2393 .fill_info = tun_fill_info,
2396 static void tun_sock_write_space(struct sock *sk)
2398 struct tun_file *tfile;
2399 wait_queue_head_t *wqueue;
2401 if (!sock_writeable(sk))
2404 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2407 wqueue = sk_sleep(sk);
2408 if (wqueue && waitqueue_active(wqueue))
2409 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2410 EPOLLWRNORM | EPOLLWRBAND);
2412 tfile = container_of(sk, struct tun_file, sk);
2413 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2416 static void tun_put_page(struct tun_page *tpage)
2419 __page_frag_cache_drain(tpage->page, tpage->count);
2422 static int tun_xdp_one(struct tun_struct *tun,
2423 struct tun_file *tfile,
2424 struct xdp_buff *xdp, int *flush,
2425 struct tun_page *tpage)
2427 unsigned int datasize = xdp->data_end - xdp->data;
2428 struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2429 struct virtio_net_hdr *gso = &hdr->gso;
2430 struct tun_pcpu_stats *stats;
2431 struct bpf_prog *xdp_prog;
2432 struct sk_buff *skb = NULL;
2433 u32 rxhash = 0, act;
2434 int buflen = hdr->buflen;
2436 bool skb_xdp = false;
2439 xdp_prog = rcu_dereference(tun->xdp_prog);
2441 if (gso->gso_type) {
2445 xdp_set_data_meta_invalid(xdp);
2446 xdp->rxq = &tfile->xdp_rxq;
2448 act = bpf_prog_run_xdp(xdp_prog, xdp);
2449 err = tun_xdp_act(tun, xdp_prog, xdp, act);
2451 put_page(virt_to_head_page(xdp->data));
2464 page = virt_to_head_page(xdp->data);
2465 if (tpage->page == page) {
2468 tun_put_page(tpage);
2477 skb = build_skb(xdp->data_hard_start, buflen);
2483 skb_reserve(skb, xdp->data - xdp->data_hard_start);
2484 skb_put(skb, xdp->data_end - xdp->data);
2486 if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2487 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
2493 skb->protocol = eth_type_trans(skb, tun->dev);
2494 skb_reset_network_header(skb);
2495 skb_probe_transport_header(skb);
2496 skb_record_rx_queue(skb, tfile->queue_index);
2499 err = do_xdp_generic(xdp_prog, skb);
2500 if (err != XDP_PASS)
2504 if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2506 rxhash = __skb_get_hash_symmetric(skb);
2508 netif_receive_skb(skb);
2510 /* No need for get_cpu_ptr() here since this function is
2511 * always called with bh disabled
2513 stats = this_cpu_ptr(tun->pcpu_stats);
2514 u64_stats_update_begin(&stats->syncp);
2515 stats->rx_packets++;
2516 stats->rx_bytes += datasize;
2517 u64_stats_update_end(&stats->syncp);
2520 tun_flow_update(tun, rxhash, tfile);
2526 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2529 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2530 struct tun_struct *tun = tun_get(tfile);
2531 struct tun_msg_ctl *ctl = m->msg_control;
2532 struct xdp_buff *xdp;
2537 if (ctl && (ctl->type == TUN_MSG_PTR)) {
2538 struct tun_page tpage;
2542 memset(&tpage, 0, sizeof(tpage));
2547 for (i = 0; i < n; i++) {
2548 xdp = &((struct xdp_buff *)ctl->ptr)[i];
2549 tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2558 tun_put_page(&tpage);
2564 ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2565 m->msg_flags & MSG_DONTWAIT,
2566 m->msg_flags & MSG_MORE);
2572 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2575 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2576 struct tun_struct *tun = tun_get(tfile);
2577 void *ptr = m->msg_control;
2585 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2589 if (flags & MSG_ERRQUEUE) {
2590 ret = sock_recv_errqueue(sock->sk, m, total_len,
2591 SOL_PACKET, TUN_TX_TIMESTAMP);
2594 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2595 if (ret > (ssize_t)total_len) {
2596 m->msg_flags |= MSG_TRUNC;
2597 ret = flags & MSG_TRUNC ? ret : total_len;
2610 static int tun_ptr_peek_len(void *ptr)
2613 if (tun_is_xdp_frame(ptr)) {
2614 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2618 return __skb_array_len_with_tag(ptr);
2624 static int tun_peek_len(struct socket *sock)
2626 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2627 struct tun_struct *tun;
2630 tun = tun_get(tfile);
2634 ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2640 /* Ops structure to mimic raw sockets with tun */
2641 static const struct proto_ops tun_socket_ops = {
2642 .peek_len = tun_peek_len,
2643 .sendmsg = tun_sendmsg,
2644 .recvmsg = tun_recvmsg,
2647 static struct proto tun_proto = {
2649 .owner = THIS_MODULE,
2650 .obj_size = sizeof(struct tun_file),
2653 static int tun_flags(struct tun_struct *tun)
2655 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2658 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2661 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2662 return sprintf(buf, "0x%x\n", tun_flags(tun));
2665 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2668 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2669 return uid_valid(tun->owner)?
2670 sprintf(buf, "%u\n",
2671 from_kuid_munged(current_user_ns(), tun->owner)):
2672 sprintf(buf, "-1\n");
2675 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2678 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2679 return gid_valid(tun->group) ?
2680 sprintf(buf, "%u\n",
2681 from_kgid_munged(current_user_ns(), tun->group)):
2682 sprintf(buf, "-1\n");
2685 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2686 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2687 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2689 static struct attribute *tun_dev_attrs[] = {
2690 &dev_attr_tun_flags.attr,
2691 &dev_attr_owner.attr,
2692 &dev_attr_group.attr,
2696 static const struct attribute_group tun_attr_group = {
2697 .attrs = tun_dev_attrs
2700 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2702 struct tun_struct *tun;
2703 struct tun_file *tfile = file->private_data;
2704 struct net_device *dev;
2707 if (tfile->detached)
2710 if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2711 if (!capable(CAP_NET_ADMIN))
2714 if (!(ifr->ifr_flags & IFF_NAPI) ||
2715 (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2719 dev = __dev_get_by_name(net, ifr->ifr_name);
2721 if (ifr->ifr_flags & IFF_TUN_EXCL)
2723 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2724 tun = netdev_priv(dev);
2725 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2726 tun = netdev_priv(dev);
2730 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2731 !!(tun->flags & IFF_MULTI_QUEUE))
2734 if (tun_not_capable(tun))
2736 err = security_tun_dev_open(tun->security);
2740 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2741 ifr->ifr_flags & IFF_NAPI,
2742 ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2746 if (tun->flags & IFF_MULTI_QUEUE &&
2747 (tun->numqueues + tun->numdisabled > 1)) {
2748 /* One or more queue has already been attached, no need
2749 * to initialize the device again.
2751 netdev_state_change(dev);
2755 tun->flags = (tun->flags & ~TUN_FEATURES) |
2756 (ifr->ifr_flags & TUN_FEATURES);
2758 netdev_state_change(dev);
2761 unsigned long flags = 0;
2762 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2765 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2767 err = security_tun_dev_create();
2772 if (ifr->ifr_flags & IFF_TUN) {
2776 } else if (ifr->ifr_flags & IFF_TAP) {
2784 name = ifr->ifr_name;
2786 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2787 NET_NAME_UNKNOWN, tun_setup, queues,
2792 err = dev_get_valid_name(net, dev, name);
2796 dev_net_set(dev, net);
2797 dev->rtnl_link_ops = &tun_link_ops;
2798 dev->ifindex = tfile->ifindex;
2799 dev->sysfs_groups[0] = &tun_attr_group;
2801 tun = netdev_priv(dev);
2804 tun->txflt.count = 0;
2805 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2807 tun->align = NET_SKB_PAD;
2808 tun->filter_attached = false;
2809 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2810 tun->rx_batched = 0;
2811 RCU_INIT_POINTER(tun->steering_prog, NULL);
2813 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2814 if (!tun->pcpu_stats) {
2819 spin_lock_init(&tun->lock);
2821 err = security_tun_dev_alloc_security(&tun->security);
2828 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2829 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2830 NETIF_F_HW_VLAN_STAG_TX;
2831 dev->features = dev->hw_features | NETIF_F_LLTX;
2832 dev->vlan_features = dev->features &
2833 ~(NETIF_F_HW_VLAN_CTAG_TX |
2834 NETIF_F_HW_VLAN_STAG_TX);
2836 tun->flags = (tun->flags & ~TUN_FEATURES) |
2837 (ifr->ifr_flags & TUN_FEATURES);
2839 INIT_LIST_HEAD(&tun->disabled);
2840 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI,
2841 ifr->ifr_flags & IFF_NAPI_FRAGS, false);
2845 err = register_netdevice(tun->dev);
2848 /* free_netdev() won't check refcnt, to aovid race
2849 * with dev_put() we need publish tun after registration.
2851 rcu_assign_pointer(tfile->tun, tun);
2854 netif_carrier_on(tun->dev);
2856 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2858 /* Make sure persistent devices do not get stuck in
2861 if (netif_running(tun->dev))
2862 netif_tx_wake_all_queues(tun->dev);
2864 strcpy(ifr->ifr_name, tun->dev->name);
2868 tun_detach_all(dev);
2869 /* register_netdevice() already called tun_free_netdev() */
2873 tun_flow_uninit(tun);
2874 security_tun_dev_free_security(tun->security);
2876 free_percpu(tun->pcpu_stats);
2882 static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2884 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2886 strcpy(ifr->ifr_name, tun->dev->name);
2888 ifr->ifr_flags = tun_flags(tun);
2892 /* This is like a cut-down ethtool ops, except done via tun fd so no
2893 * privs required. */
2894 static int set_offload(struct tun_struct *tun, unsigned long arg)
2896 netdev_features_t features = 0;
2898 if (arg & TUN_F_CSUM) {
2899 features |= NETIF_F_HW_CSUM;
2902 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2903 if (arg & TUN_F_TSO_ECN) {
2904 features |= NETIF_F_TSO_ECN;
2905 arg &= ~TUN_F_TSO_ECN;
2907 if (arg & TUN_F_TSO4)
2908 features |= NETIF_F_TSO;
2909 if (arg & TUN_F_TSO6)
2910 features |= NETIF_F_TSO6;
2911 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2917 /* This gives the user a way to test for new features in future by
2918 * trying to set them. */
2922 tun->set_features = features;
2923 tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2924 tun->dev->wanted_features |= features;
2925 netdev_update_features(tun->dev);
2930 static void tun_detach_filter(struct tun_struct *tun, int n)
2933 struct tun_file *tfile;
2935 for (i = 0; i < n; i++) {
2936 tfile = rtnl_dereference(tun->tfiles[i]);
2937 lock_sock(tfile->socket.sk);
2938 sk_detach_filter(tfile->socket.sk);
2939 release_sock(tfile->socket.sk);
2942 tun->filter_attached = false;
2945 static int tun_attach_filter(struct tun_struct *tun)
2948 struct tun_file *tfile;
2950 for (i = 0; i < tun->numqueues; i++) {
2951 tfile = rtnl_dereference(tun->tfiles[i]);
2952 lock_sock(tfile->socket.sk);
2953 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2954 release_sock(tfile->socket.sk);
2956 tun_detach_filter(tun, i);
2961 tun->filter_attached = true;
2965 static void tun_set_sndbuf(struct tun_struct *tun)
2967 struct tun_file *tfile;
2970 for (i = 0; i < tun->numqueues; i++) {
2971 tfile = rtnl_dereference(tun->tfiles[i]);
2972 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2976 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2978 struct tun_file *tfile = file->private_data;
2979 struct tun_struct *tun;
2984 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2985 tun = tfile->detached;
2990 ret = security_tun_dev_attach_queue(tun->security);
2993 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2994 tun->flags & IFF_NAPI_FRAGS, true);
2995 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2996 tun = rtnl_dereference(tfile->tun);
2997 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
3000 __tun_detach(tfile, false);
3005 netdev_state_change(tun->dev);
3012 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
3015 struct bpf_prog *prog;
3018 if (copy_from_user(&fd, data, sizeof(fd)))
3024 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
3026 return PTR_ERR(prog);
3029 return __tun_set_ebpf(tun, prog_p, prog);
3032 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
3033 unsigned long arg, int ifreq_len)
3035 struct tun_file *tfile = file->private_data;
3036 struct net *net = sock_net(&tfile->sk);
3037 struct tun_struct *tun;
3038 void __user* argp = (void __user*)arg;
3039 unsigned int ifindex, carrier;
3047 bool do_notify = false;
3049 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
3050 (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
3051 if (copy_from_user(&ifr, argp, ifreq_len))
3054 memset(&ifr, 0, sizeof(ifr));
3056 if (cmd == TUNGETFEATURES) {
3057 /* Currently this just means: "what IFF flags are valid?".
3058 * This is needed because we never checked for invalid flags on
3061 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
3062 (unsigned int __user*)argp);
3063 } else if (cmd == TUNSETQUEUE) {
3064 return tun_set_queue(file, &ifr);
3065 } else if (cmd == SIOCGSKNS) {
3066 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3068 return open_related_ns(&net->ns, get_net_ns);
3074 tun = tun_get(tfile);
3075 if (cmd == TUNSETIFF) {
3080 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3082 ret = tun_set_iff(net, file, &ifr);
3087 if (copy_to_user(argp, &ifr, ifreq_len))
3091 if (cmd == TUNSETIFINDEX) {
3097 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3101 tfile->ifindex = ifindex;
3109 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
3111 net = dev_net(tun->dev);
3115 tun_get_iff(tun, &ifr);
3117 if (tfile->detached)
3118 ifr.ifr_flags |= IFF_DETACH_QUEUE;
3119 if (!tfile->socket.sk->sk_filter)
3120 ifr.ifr_flags |= IFF_NOFILTER;
3122 if (copy_to_user(argp, &ifr, ifreq_len))
3127 /* Disable/Enable checksum */
3129 /* [unimplemented] */
3130 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
3131 arg ? "disabled" : "enabled");
3135 /* Disable/Enable persist mode. Keep an extra reference to the
3136 * module to prevent the module being unprobed.
3138 if (arg && !(tun->flags & IFF_PERSIST)) {
3139 tun->flags |= IFF_PERSIST;
3140 __module_get(THIS_MODULE);
3143 if (!arg && (tun->flags & IFF_PERSIST)) {
3144 tun->flags &= ~IFF_PERSIST;
3145 module_put(THIS_MODULE);
3149 tun_debug(KERN_INFO, tun, "persist %s\n",
3150 arg ? "enabled" : "disabled");
3154 /* Set owner of the device */
3155 owner = make_kuid(current_user_ns(), arg);
3156 if (!uid_valid(owner)) {
3162 tun_debug(KERN_INFO, tun, "owner set to %u\n",
3163 from_kuid(&init_user_ns, tun->owner));
3167 /* Set group of the device */
3168 group = make_kgid(current_user_ns(), arg);
3169 if (!gid_valid(group)) {
3175 tun_debug(KERN_INFO, tun, "group set to %u\n",
3176 from_kgid(&init_user_ns, tun->group));
3180 /* Only allow setting the type when the interface is down */
3181 if (tun->dev->flags & IFF_UP) {
3182 tun_debug(KERN_INFO, tun,
3183 "Linktype set failed because interface is up\n");
3186 tun->dev->type = (int) arg;
3187 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
3199 ret = set_offload(tun, arg);
3202 case TUNSETTXFILTER:
3203 /* Can be set only for TAPs */
3205 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3207 ret = update_filter(&tun->txflt, (void __user *)arg);
3211 /* Get hw address */
3212 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3213 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3214 if (copy_to_user(argp, &ifr, ifreq_len))
3219 /* Set hw address */
3220 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
3221 ifr.ifr_hwaddr.sa_data);
3223 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr, NULL);
3227 sndbuf = tfile->socket.sk->sk_sndbuf;
3228 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3233 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3242 tun->sndbuf = sndbuf;
3243 tun_set_sndbuf(tun);
3246 case TUNGETVNETHDRSZ:
3247 vnet_hdr_sz = tun->vnet_hdr_sz;
3248 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3252 case TUNSETVNETHDRSZ:
3253 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3257 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3262 tun->vnet_hdr_sz = vnet_hdr_sz;
3266 le = !!(tun->flags & TUN_VNET_LE);
3267 if (put_user(le, (int __user *)argp))
3272 if (get_user(le, (int __user *)argp)) {
3277 tun->flags |= TUN_VNET_LE;
3279 tun->flags &= ~TUN_VNET_LE;
3283 ret = tun_get_vnet_be(tun, argp);
3287 ret = tun_set_vnet_be(tun, argp);
3290 case TUNATTACHFILTER:
3291 /* Can be set only for TAPs */
3293 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3296 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3299 ret = tun_attach_filter(tun);
3302 case TUNDETACHFILTER:
3303 /* Can be set only for TAPs */
3305 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3308 tun_detach_filter(tun, tun->numqueues);
3313 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3316 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3321 case TUNSETSTEERINGEBPF:
3322 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3325 case TUNSETFILTEREBPF:
3326 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3331 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3334 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3337 case TUNGETDEVNETNS:
3339 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3341 ret = open_related_ns(&net->ns, get_net_ns);
3350 netdev_state_change(tun->dev);
3359 static long tun_chr_ioctl(struct file *file,
3360 unsigned int cmd, unsigned long arg)
3362 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3365 #ifdef CONFIG_COMPAT
3366 static long tun_chr_compat_ioctl(struct file *file,
3367 unsigned int cmd, unsigned long arg)
3372 case TUNSETTXFILTER:
3377 arg = (unsigned long)compat_ptr(arg);
3380 arg = (compat_ulong_t)arg;
3385 * compat_ifreq is shorter than ifreq, so we must not access beyond
3386 * the end of that structure. All fields that are used in this
3387 * driver are compatible though, we don't need to convert the
3390 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3392 #endif /* CONFIG_COMPAT */
3394 static int tun_chr_fasync(int fd, struct file *file, int on)
3396 struct tun_file *tfile = file->private_data;
3399 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3403 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3404 tfile->flags |= TUN_FASYNC;
3406 tfile->flags &= ~TUN_FASYNC;
3412 static int tun_chr_open(struct inode *inode, struct file * file)
3414 struct net *net = current->nsproxy->net_ns;
3415 struct tun_file *tfile;
3417 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3419 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3423 if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3424 sk_free(&tfile->sk);
3428 mutex_init(&tfile->napi_mutex);
3429 RCU_INIT_POINTER(tfile->tun, NULL);
3433 init_waitqueue_head(&tfile->socket.wq.wait);
3435 tfile->socket.file = file;
3436 tfile->socket.ops = &tun_socket_ops;
3438 sock_init_data(&tfile->socket, &tfile->sk);
3440 tfile->sk.sk_write_space = tun_sock_write_space;
3441 tfile->sk.sk_sndbuf = INT_MAX;
3443 file->private_data = tfile;
3444 INIT_LIST_HEAD(&tfile->next);
3446 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3451 static int tun_chr_close(struct inode *inode, struct file *file)
3453 struct tun_file *tfile = file->private_data;
3455 tun_detach(tfile, true);
3460 #ifdef CONFIG_PROC_FS
3461 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3463 struct tun_file *tfile = file->private_data;
3464 struct tun_struct *tun;
3467 memset(&ifr, 0, sizeof(ifr));
3470 tun = tun_get(tfile);
3472 tun_get_iff(tun, &ifr);
3478 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3482 static const struct file_operations tun_fops = {
3483 .owner = THIS_MODULE,
3484 .llseek = no_llseek,
3485 .read_iter = tun_chr_read_iter,
3486 .write_iter = tun_chr_write_iter,
3487 .poll = tun_chr_poll,
3488 .unlocked_ioctl = tun_chr_ioctl,
3489 #ifdef CONFIG_COMPAT
3490 .compat_ioctl = tun_chr_compat_ioctl,
3492 .open = tun_chr_open,
3493 .release = tun_chr_close,
3494 .fasync = tun_chr_fasync,
3495 #ifdef CONFIG_PROC_FS
3496 .show_fdinfo = tun_chr_show_fdinfo,
3500 static struct miscdevice tun_miscdev = {
3503 .nodename = "net/tun",
3507 /* ethtool interface */
3509 static void tun_default_link_ksettings(struct net_device *dev,
3510 struct ethtool_link_ksettings *cmd)
3512 ethtool_link_ksettings_zero_link_mode(cmd, supported);
3513 ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3514 cmd->base.speed = SPEED_10;
3515 cmd->base.duplex = DUPLEX_FULL;
3516 cmd->base.port = PORT_TP;
3517 cmd->base.phy_address = 0;
3518 cmd->base.autoneg = AUTONEG_DISABLE;
3521 static int tun_get_link_ksettings(struct net_device *dev,
3522 struct ethtool_link_ksettings *cmd)
3524 struct tun_struct *tun = netdev_priv(dev);
3526 memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3530 static int tun_set_link_ksettings(struct net_device *dev,
3531 const struct ethtool_link_ksettings *cmd)
3533 struct tun_struct *tun = netdev_priv(dev);
3535 memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3539 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3541 struct tun_struct *tun = netdev_priv(dev);
3543 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3544 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3546 switch (tun->flags & TUN_TYPE_MASK) {
3548 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3551 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3556 static u32 tun_get_msglevel(struct net_device *dev)
3559 struct tun_struct *tun = netdev_priv(dev);
3566 static void tun_set_msglevel(struct net_device *dev, u32 value)
3569 struct tun_struct *tun = netdev_priv(dev);
3574 static int tun_get_coalesce(struct net_device *dev,
3575 struct ethtool_coalesce *ec)
3577 struct tun_struct *tun = netdev_priv(dev);
3579 ec->rx_max_coalesced_frames = tun->rx_batched;
3584 static int tun_set_coalesce(struct net_device *dev,
3585 struct ethtool_coalesce *ec)
3587 struct tun_struct *tun = netdev_priv(dev);
3589 if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3590 tun->rx_batched = NAPI_POLL_WEIGHT;
3592 tun->rx_batched = ec->rx_max_coalesced_frames;
3597 static const struct ethtool_ops tun_ethtool_ops = {
3598 .get_drvinfo = tun_get_drvinfo,
3599 .get_msglevel = tun_get_msglevel,
3600 .set_msglevel = tun_set_msglevel,
3601 .get_link = ethtool_op_get_link,
3602 .get_ts_info = ethtool_op_get_ts_info,
3603 .get_coalesce = tun_get_coalesce,
3604 .set_coalesce = tun_set_coalesce,
3605 .get_link_ksettings = tun_get_link_ksettings,
3606 .set_link_ksettings = tun_set_link_ksettings,
3609 static int tun_queue_resize(struct tun_struct *tun)
3611 struct net_device *dev = tun->dev;
3612 struct tun_file *tfile;
3613 struct ptr_ring **rings;
3614 int n = tun->numqueues + tun->numdisabled;
3617 rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3621 for (i = 0; i < tun->numqueues; i++) {
3622 tfile = rtnl_dereference(tun->tfiles[i]);
3623 rings[i] = &tfile->tx_ring;
3625 list_for_each_entry(tfile, &tun->disabled, next)
3626 rings[i++] = &tfile->tx_ring;
3628 ret = ptr_ring_resize_multiple(rings, n,
3629 dev->tx_queue_len, GFP_KERNEL,
3636 static int tun_device_event(struct notifier_block *unused,
3637 unsigned long event, void *ptr)
3639 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3640 struct tun_struct *tun = netdev_priv(dev);
3643 if (dev->rtnl_link_ops != &tun_link_ops)
3647 case NETDEV_CHANGE_TX_QUEUE_LEN:
3648 if (tun_queue_resize(tun))
3652 for (i = 0; i < tun->numqueues; i++) {
3653 struct tun_file *tfile;
3655 tfile = rtnl_dereference(tun->tfiles[i]);
3656 tfile->socket.sk->sk_write_space(tfile->socket.sk);
3666 static struct notifier_block tun_notifier_block __read_mostly = {
3667 .notifier_call = tun_device_event,
3670 static int __init tun_init(void)
3674 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3676 ret = rtnl_link_register(&tun_link_ops);
3678 pr_err("Can't register link_ops\n");
3682 ret = misc_register(&tun_miscdev);
3684 pr_err("Can't register misc device %d\n", TUN_MINOR);
3688 ret = register_netdevice_notifier(&tun_notifier_block);
3690 pr_err("Can't register netdevice notifier\n");
3697 misc_deregister(&tun_miscdev);
3699 rtnl_link_unregister(&tun_link_ops);
3704 static void tun_cleanup(void)
3706 misc_deregister(&tun_miscdev);
3707 rtnl_link_unregister(&tun_link_ops);
3708 unregister_netdevice_notifier(&tun_notifier_block);
3711 /* Get an underlying socket object from tun file. Returns error unless file is
3712 * attached to a device. The returned object works like a packet socket, it
3713 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
3714 * holding a reference to the file for as long as the socket is in use. */
3715 struct socket *tun_get_socket(struct file *file)
3717 struct tun_file *tfile;
3718 if (file->f_op != &tun_fops)
3719 return ERR_PTR(-EINVAL);
3720 tfile = file->private_data;
3722 return ERR_PTR(-EBADFD);
3723 return &tfile->socket;
3725 EXPORT_SYMBOL_GPL(tun_get_socket);
3727 struct ptr_ring *tun_get_tx_ring(struct file *file)
3729 struct tun_file *tfile;
3731 if (file->f_op != &tun_fops)
3732 return ERR_PTR(-EINVAL);
3733 tfile = file->private_data;
3735 return ERR_PTR(-EBADFD);
3736 return &tfile->tx_ring;
3738 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3740 module_init(tun_init);
3741 module_exit(tun_cleanup);
3742 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3743 MODULE_AUTHOR(DRV_COPYRIGHT);
3744 MODULE_LICENSE("GPL");
3745 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3746 MODULE_ALIAS("devname:net/tun");