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 <net/ip_tunnels.h>
66 #include <linux/seq_file.h>
67 #include <linux/uio.h>
68 #include <linux/skb_array.h>
69 #include <linux/bpf.h>
70 #include <linux/bpf_trace.h>
71 #include <linux/mutex.h>
72 #include <linux/ieee802154.h>
73 #include <linux/if_ltalk.h>
74 #include <uapi/linux/if_fddi.h>
75 #include <uapi/linux/if_hippi.h>
76 #include <uapi/linux/if_fc.h>
79 #include <net/6lowpan.h>
81 #include <linux/uaccess.h>
82 #include <linux/proc_fs.h>
84 static void tun_default_link_ksettings(struct net_device *dev,
85 struct ethtool_link_ksettings *cmd);
87 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
89 /* TUN device flags */
91 /* IFF_ATTACH_QUEUE is never stored in device flags,
92 * overload it to mean fasync when stored there.
94 #define TUN_FASYNC IFF_ATTACH_QUEUE
95 /* High bits in flags field are unused. */
96 #define TUN_VNET_LE 0x80000000
97 #define TUN_VNET_BE 0x40000000
99 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
100 IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
102 #define GOODCOPY_LEN 128
104 #define FLT_EXACT_COUNT 8
106 unsigned int count; /* Number of addrs. Zero means disabled */
107 u32 mask[2]; /* Mask of the hashed addrs */
108 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
111 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
112 * to max number of VCPUs in guest. */
113 #define MAX_TAP_QUEUES 256
114 #define MAX_TAP_FLOWS 4096
116 #define TUN_FLOW_EXPIRE (3 * HZ)
118 /* A tun_file connects an open character device to a tuntap netdevice. It
119 * also contains all socket related structures (except sock_fprog and tap_filter)
120 * to serve as one transmit queue for tuntap device. The sock_fprog and
121 * tap_filter were kept in tun_struct since they were used for filtering for the
122 * netdevice not for a specific queue (at least I didn't see the requirement for
126 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
127 * other can only be read while rcu_read_lock or rtnl_lock is held.
131 struct socket socket;
132 struct tun_struct __rcu *tun;
133 struct fasync_struct *fasync;
134 /* only used for fasnyc */
138 unsigned int ifindex;
140 struct napi_struct napi;
142 bool napi_frags_enabled;
143 struct mutex napi_mutex; /* Protects access to the above napi */
144 struct list_head next;
145 struct tun_struct *detached;
146 struct ptr_ring tx_ring;
147 struct xdp_rxq_info xdp_rxq;
155 struct tun_flow_entry {
156 struct hlist_node hash_link;
158 struct tun_struct *tun;
163 unsigned long updated ____cacheline_aligned_in_smp;
166 #define TUN_NUM_FLOW_ENTRIES 1024
167 #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
171 struct bpf_prog *prog;
174 /* Since the socket were moved to tun_file, to preserve the behavior of persist
175 * device, socket filter, sndbuf and vnet header size were restore when the
176 * file were attached to a persist device.
179 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
180 unsigned int numqueues;
185 struct net_device *dev;
186 netdev_features_t set_features;
187 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
193 struct tap_filter txflt;
194 struct sock_fprog fprog;
195 /* protected by rtnl lock */
196 bool filter_attached;
199 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
200 struct timer_list flow_gc_timer;
201 unsigned long ageing_time;
202 unsigned int numdisabled;
203 struct list_head disabled;
207 atomic_long_t rx_frame_errors;
208 struct bpf_prog __rcu *xdp_prog;
209 struct tun_prog __rcu *steering_prog;
210 struct tun_prog __rcu *filter_prog;
211 struct ethtool_link_ksettings link_ksettings;
222 static void tun_flow_init(struct tun_struct *tun);
223 static void tun_flow_uninit(struct tun_struct *tun);
225 static int tun_napi_receive(struct napi_struct *napi, int budget)
227 struct tun_file *tfile = container_of(napi, struct tun_file, napi);
228 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
229 struct sk_buff_head process_queue;
233 __skb_queue_head_init(&process_queue);
235 spin_lock(&queue->lock);
236 skb_queue_splice_tail_init(queue, &process_queue);
237 spin_unlock(&queue->lock);
239 while (received < budget && (skb = __skb_dequeue(&process_queue))) {
240 napi_gro_receive(napi, skb);
244 if (!skb_queue_empty(&process_queue)) {
245 spin_lock(&queue->lock);
246 skb_queue_splice(&process_queue, queue);
247 spin_unlock(&queue->lock);
253 static int tun_napi_poll(struct napi_struct *napi, int budget)
255 unsigned int received;
257 received = tun_napi_receive(napi, budget);
259 if (received < budget)
260 napi_complete_done(napi, received);
265 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
266 bool napi_en, bool napi_frags)
268 tfile->napi_enabled = napi_en;
269 tfile->napi_frags_enabled = napi_en && napi_frags;
271 netif_napi_add_tx(tun->dev, &tfile->napi, tun_napi_poll);
272 napi_enable(&tfile->napi);
276 static void tun_napi_enable(struct tun_file *tfile)
278 if (tfile->napi_enabled)
279 napi_enable(&tfile->napi);
282 static void tun_napi_disable(struct tun_file *tfile)
284 if (tfile->napi_enabled)
285 napi_disable(&tfile->napi);
288 static void tun_napi_del(struct tun_file *tfile)
290 if (tfile->napi_enabled)
291 netif_napi_del(&tfile->napi);
294 static bool tun_napi_frags_enabled(const struct tun_file *tfile)
296 return tfile->napi_frags_enabled;
299 #ifdef CONFIG_TUN_VNET_CROSS_LE
300 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
302 return tun->flags & TUN_VNET_BE ? false :
303 virtio_legacy_is_little_endian();
306 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
308 int be = !!(tun->flags & TUN_VNET_BE);
310 if (put_user(be, argp))
316 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
320 if (get_user(be, argp))
324 tun->flags |= TUN_VNET_BE;
326 tun->flags &= ~TUN_VNET_BE;
331 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
333 return virtio_legacy_is_little_endian();
336 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
341 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
345 #endif /* CONFIG_TUN_VNET_CROSS_LE */
347 static inline bool tun_is_little_endian(struct tun_struct *tun)
349 return tun->flags & TUN_VNET_LE ||
350 tun_legacy_is_little_endian(tun);
353 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
355 return __virtio16_to_cpu(tun_is_little_endian(tun), val);
358 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
360 return __cpu_to_virtio16(tun_is_little_endian(tun), val);
363 static inline u32 tun_hashfn(u32 rxhash)
365 return rxhash & TUN_MASK_FLOW_ENTRIES;
368 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
370 struct tun_flow_entry *e;
372 hlist_for_each_entry_rcu(e, head, hash_link) {
373 if (e->rxhash == rxhash)
379 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
380 struct hlist_head *head,
381 u32 rxhash, u16 queue_index)
383 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
386 netif_info(tun, tx_queued, tun->dev,
387 "create flow: hash %u index %u\n",
388 rxhash, queue_index);
389 e->updated = jiffies;
392 e->queue_index = queue_index;
394 hlist_add_head_rcu(&e->hash_link, head);
400 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
402 netif_info(tun, tx_queued, tun->dev, "delete flow: hash %u index %u\n",
403 e->rxhash, e->queue_index);
404 hlist_del_rcu(&e->hash_link);
409 static void tun_flow_flush(struct tun_struct *tun)
413 spin_lock_bh(&tun->lock);
414 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
415 struct tun_flow_entry *e;
416 struct hlist_node *n;
418 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
419 tun_flow_delete(tun, e);
421 spin_unlock_bh(&tun->lock);
424 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
428 spin_lock_bh(&tun->lock);
429 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
430 struct tun_flow_entry *e;
431 struct hlist_node *n;
433 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
434 if (e->queue_index == queue_index)
435 tun_flow_delete(tun, e);
438 spin_unlock_bh(&tun->lock);
441 static void tun_flow_cleanup(struct timer_list *t)
443 struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
444 unsigned long delay = tun->ageing_time;
445 unsigned long next_timer = jiffies + delay;
446 unsigned long count = 0;
449 spin_lock(&tun->lock);
450 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
451 struct tun_flow_entry *e;
452 struct hlist_node *n;
454 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
455 unsigned long this_timer;
457 this_timer = e->updated + delay;
458 if (time_before_eq(this_timer, jiffies)) {
459 tun_flow_delete(tun, e);
463 if (time_before(this_timer, next_timer))
464 next_timer = this_timer;
469 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
470 spin_unlock(&tun->lock);
473 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
474 struct tun_file *tfile)
476 struct hlist_head *head;
477 struct tun_flow_entry *e;
478 unsigned long delay = tun->ageing_time;
479 u16 queue_index = tfile->queue_index;
481 head = &tun->flows[tun_hashfn(rxhash)];
485 e = tun_flow_find(head, rxhash);
487 /* TODO: keep queueing to old queue until it's empty? */
488 if (READ_ONCE(e->queue_index) != queue_index)
489 WRITE_ONCE(e->queue_index, queue_index);
490 if (e->updated != jiffies)
491 e->updated = jiffies;
492 sock_rps_record_flow_hash(e->rps_rxhash);
494 spin_lock_bh(&tun->lock);
495 if (!tun_flow_find(head, rxhash) &&
496 tun->flow_count < MAX_TAP_FLOWS)
497 tun_flow_create(tun, head, rxhash, queue_index);
499 if (!timer_pending(&tun->flow_gc_timer))
500 mod_timer(&tun->flow_gc_timer,
501 round_jiffies_up(jiffies + delay));
502 spin_unlock_bh(&tun->lock);
508 /* Save the hash received in the stack receive path and update the
509 * flow_hash table accordingly.
511 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
513 if (unlikely(e->rps_rxhash != hash))
514 e->rps_rxhash = hash;
517 /* We try to identify a flow through its rxhash. The reason that
518 * we do not check rxq no. is because some cards(e.g 82599), chooses
519 * the rxq based on the txq where the last packet of the flow comes. As
520 * the userspace application move between processors, we may get a
521 * different rxq no. here.
523 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
525 struct tun_flow_entry *e;
529 numqueues = READ_ONCE(tun->numqueues);
531 txq = __skb_get_hash_symmetric(skb);
532 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
534 tun_flow_save_rps_rxhash(e, txq);
535 txq = e->queue_index;
537 /* use multiply and shift instead of expensive divide */
538 txq = ((u64)txq * numqueues) >> 32;
544 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
546 struct tun_prog *prog;
550 numqueues = READ_ONCE(tun->numqueues);
554 prog = rcu_dereference(tun->steering_prog);
556 ret = bpf_prog_run_clear_cb(prog->prog, skb);
558 return ret % numqueues;
561 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
562 struct net_device *sb_dev)
564 struct tun_struct *tun = netdev_priv(dev);
568 if (rcu_dereference(tun->steering_prog))
569 ret = tun_ebpf_select_queue(tun, skb);
571 ret = tun_automq_select_queue(tun, skb);
577 static inline bool tun_not_capable(struct tun_struct *tun)
579 const struct cred *cred = current_cred();
580 struct net *net = dev_net(tun->dev);
582 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
583 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
584 !ns_capable(net->user_ns, CAP_NET_ADMIN);
587 static void tun_set_real_num_queues(struct tun_struct *tun)
589 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
590 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
593 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
595 tfile->detached = tun;
596 list_add_tail(&tfile->next, &tun->disabled);
600 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
602 struct tun_struct *tun = tfile->detached;
604 tfile->detached = NULL;
605 list_del_init(&tfile->next);
610 void tun_ptr_free(void *ptr)
614 if (tun_is_xdp_frame(ptr)) {
615 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
617 xdp_return_frame(xdpf);
619 __skb_array_destroy_skb(ptr);
622 EXPORT_SYMBOL_GPL(tun_ptr_free);
624 static void tun_queue_purge(struct tun_file *tfile)
628 while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
631 skb_queue_purge(&tfile->sk.sk_write_queue);
632 skb_queue_purge(&tfile->sk.sk_error_queue);
635 static void __tun_detach(struct tun_file *tfile, bool clean)
637 struct tun_file *ntfile;
638 struct tun_struct *tun;
640 tun = rtnl_dereference(tfile->tun);
643 if (!tfile->detached)
644 tun_napi_disable(tfile);
648 if (tun && !tfile->detached) {
649 u16 index = tfile->queue_index;
650 BUG_ON(index >= tun->numqueues);
652 rcu_assign_pointer(tun->tfiles[index],
653 tun->tfiles[tun->numqueues - 1]);
654 ntfile = rtnl_dereference(tun->tfiles[index]);
655 ntfile->queue_index = index;
656 rcu_assign_pointer(tun->tfiles[tun->numqueues - 1],
661 RCU_INIT_POINTER(tfile->tun, NULL);
662 sock_put(&tfile->sk);
664 tun_disable_queue(tun, tfile);
665 tun_napi_disable(tfile);
669 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
670 /* Drop read queue */
671 tun_queue_purge(tfile);
672 tun_set_real_num_queues(tun);
673 } else if (tfile->detached && clean) {
674 tun = tun_enable_queue(tfile);
675 sock_put(&tfile->sk);
679 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
680 netif_carrier_off(tun->dev);
682 if (!(tun->flags & IFF_PERSIST) &&
683 tun->dev->reg_state == NETREG_REGISTERED)
684 unregister_netdevice(tun->dev);
687 xdp_rxq_info_unreg(&tfile->xdp_rxq);
688 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
689 sock_put(&tfile->sk);
693 static void tun_detach(struct tun_file *tfile, bool clean)
695 struct tun_struct *tun;
696 struct net_device *dev;
699 tun = rtnl_dereference(tfile->tun);
700 dev = tun ? tun->dev : NULL;
701 __tun_detach(tfile, clean);
703 netdev_state_change(dev);
707 static void tun_detach_all(struct net_device *dev)
709 struct tun_struct *tun = netdev_priv(dev);
710 struct tun_file *tfile, *tmp;
711 int i, n = tun->numqueues;
713 for (i = 0; i < n; i++) {
714 tfile = rtnl_dereference(tun->tfiles[i]);
716 tun_napi_disable(tfile);
717 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
718 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
719 RCU_INIT_POINTER(tfile->tun, NULL);
722 list_for_each_entry(tfile, &tun->disabled, next) {
723 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
724 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
725 RCU_INIT_POINTER(tfile->tun, NULL);
727 BUG_ON(tun->numqueues != 0);
730 for (i = 0; i < n; i++) {
731 tfile = rtnl_dereference(tun->tfiles[i]);
733 /* Drop read queue */
734 tun_queue_purge(tfile);
735 xdp_rxq_info_unreg(&tfile->xdp_rxq);
736 sock_put(&tfile->sk);
738 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
740 tun_enable_queue(tfile);
741 tun_queue_purge(tfile);
742 xdp_rxq_info_unreg(&tfile->xdp_rxq);
743 sock_put(&tfile->sk);
745 BUG_ON(tun->numdisabled != 0);
747 if (tun->flags & IFF_PERSIST)
748 module_put(THIS_MODULE);
751 static int tun_attach(struct tun_struct *tun, struct file *file,
752 bool skip_filter, bool napi, bool napi_frags,
755 struct tun_file *tfile = file->private_data;
756 struct net_device *dev = tun->dev;
759 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
764 if (rtnl_dereference(tfile->tun) && !tfile->detached)
768 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
772 if (!tfile->detached &&
773 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
778 /* Re-attach the filter to persist device */
779 if (!skip_filter && (tun->filter_attached == true)) {
780 lock_sock(tfile->socket.sk);
781 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
782 release_sock(tfile->socket.sk);
787 if (!tfile->detached &&
788 ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
789 GFP_KERNEL, tun_ptr_free)) {
794 tfile->queue_index = tun->numqueues;
795 tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
797 if (tfile->detached) {
798 /* Re-attach detached tfile, updating XDP queue_index */
799 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
801 if (tfile->xdp_rxq.queue_index != tfile->queue_index)
802 tfile->xdp_rxq.queue_index = tfile->queue_index;
804 /* Setup XDP RX-queue info, for new tfile getting attached */
805 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
806 tun->dev, tfile->queue_index, 0);
809 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
810 MEM_TYPE_PAGE_SHARED, NULL);
812 xdp_rxq_info_unreg(&tfile->xdp_rxq);
818 if (tfile->detached) {
819 tun_enable_queue(tfile);
820 tun_napi_enable(tfile);
822 sock_hold(&tfile->sk);
823 tun_napi_init(tun, tfile, napi, napi_frags);
826 if (rtnl_dereference(tun->xdp_prog))
827 sock_set_flag(&tfile->sk, SOCK_XDP);
829 /* device is allowed to go away first, so no need to hold extra
833 /* Publish tfile->tun and tun->tfiles only after we've fully
834 * initialized tfile; otherwise we risk using half-initialized
838 rcu_assign_pointer(tfile->tun, tun);
839 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
841 tun_set_real_num_queues(tun);
846 static struct tun_struct *tun_get(struct tun_file *tfile)
848 struct tun_struct *tun;
851 tun = rcu_dereference(tfile->tun);
859 static void tun_put(struct tun_struct *tun)
865 static void addr_hash_set(u32 *mask, const u8 *addr)
867 int n = ether_crc(ETH_ALEN, addr) >> 26;
868 mask[n >> 5] |= (1 << (n & 31));
871 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
873 int n = ether_crc(ETH_ALEN, addr) >> 26;
874 return mask[n >> 5] & (1 << (n & 31));
877 static int update_filter(struct tap_filter *filter, void __user *arg)
879 struct { u8 u[ETH_ALEN]; } *addr;
880 struct tun_filter uf;
881 int err, alen, n, nexact;
883 if (copy_from_user(&uf, arg, sizeof(uf)))
892 alen = ETH_ALEN * uf.count;
893 addr = memdup_user(arg + sizeof(uf), alen);
895 return PTR_ERR(addr);
897 /* The filter is updated without holding any locks. Which is
898 * perfectly safe. We disable it first and in the worst
899 * case we'll accept a few undesired packets. */
903 /* Use first set of addresses as an exact filter */
904 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
905 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
909 /* Remaining multicast addresses are hashed,
910 * unicast will leave the filter disabled. */
911 memset(filter->mask, 0, sizeof(filter->mask));
912 for (; n < uf.count; n++) {
913 if (!is_multicast_ether_addr(addr[n].u)) {
914 err = 0; /* no filter */
917 addr_hash_set(filter->mask, addr[n].u);
920 /* For ALLMULTI just set the mask to all ones.
921 * This overrides the mask populated above. */
922 if ((uf.flags & TUN_FLT_ALLMULTI))
923 memset(filter->mask, ~0, sizeof(filter->mask));
925 /* Now enable the filter */
927 filter->count = nexact;
929 /* Return the number of exact filters */
936 /* Returns: 0 - drop, !=0 - accept */
937 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
939 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
941 struct ethhdr *eh = (struct ethhdr *) skb->data;
945 for (i = 0; i < filter->count; i++)
946 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
949 /* Inexact match (multicast only) */
950 if (is_multicast_ether_addr(eh->h_dest))
951 return addr_hash_test(filter->mask, eh->h_dest);
957 * Checks whether the packet is accepted or not.
958 * Returns: 0 - drop, !=0 - accept
960 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
965 return run_filter(filter, skb);
968 /* Network device part of the driver */
970 static const struct ethtool_ops tun_ethtool_ops;
972 static int tun_net_init(struct net_device *dev)
974 struct tun_struct *tun = netdev_priv(dev);
975 struct ifreq *ifr = tun->ifr;
978 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
982 spin_lock_init(&tun->lock);
984 err = security_tun_dev_alloc_security(&tun->security);
986 free_percpu(dev->tstats);
992 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
993 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
994 NETIF_F_HW_VLAN_STAG_TX;
995 dev->features = dev->hw_features | NETIF_F_LLTX;
996 dev->vlan_features = dev->features &
997 ~(NETIF_F_HW_VLAN_CTAG_TX |
998 NETIF_F_HW_VLAN_STAG_TX);
1000 tun->flags = (tun->flags & ~TUN_FEATURES) |
1001 (ifr->ifr_flags & TUN_FEATURES);
1003 INIT_LIST_HEAD(&tun->disabled);
1004 err = tun_attach(tun, tun->file, false, ifr->ifr_flags & IFF_NAPI,
1005 ifr->ifr_flags & IFF_NAPI_FRAGS, false);
1007 tun_flow_uninit(tun);
1008 security_tun_dev_free_security(tun->security);
1009 free_percpu(dev->tstats);
1015 /* Net device detach from fd. */
1016 static void tun_net_uninit(struct net_device *dev)
1018 tun_detach_all(dev);
1021 /* Net device open. */
1022 static int tun_net_open(struct net_device *dev)
1024 netif_tx_start_all_queues(dev);
1029 /* Net device close. */
1030 static int tun_net_close(struct net_device *dev)
1032 netif_tx_stop_all_queues(dev);
1036 /* Net device start xmit */
1037 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1040 if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
1041 /* Select queue was not called for the skbuff, so we extract the
1042 * RPS hash and save it into the flow_table here.
1044 struct tun_flow_entry *e;
1047 rxhash = __skb_get_hash_symmetric(skb);
1048 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
1050 tun_flow_save_rps_rxhash(e, rxhash);
1055 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1056 struct sk_buff *skb,
1059 struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1062 len = bpf_prog_run_clear_cb(prog->prog, skb);
1067 /* Net device start xmit */
1068 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1070 struct tun_struct *tun = netdev_priv(dev);
1071 enum skb_drop_reason drop_reason;
1072 int txq = skb->queue_mapping;
1073 struct netdev_queue *queue;
1074 struct tun_file *tfile;
1078 tfile = rcu_dereference(tun->tfiles[txq]);
1080 /* Drop packet if interface is not attached */
1082 drop_reason = SKB_DROP_REASON_DEV_READY;
1086 if (!rcu_dereference(tun->steering_prog))
1087 tun_automq_xmit(tun, skb);
1089 netif_info(tun, tx_queued, tun->dev, "%s %d\n", __func__, skb->len);
1091 /* Drop if the filter does not like it.
1092 * This is a noop if the filter is disabled.
1093 * Filter can be enabled only for the TAP devices. */
1094 if (!check_filter(&tun->txflt, skb)) {
1095 drop_reason = SKB_DROP_REASON_TAP_TXFILTER;
1099 if (tfile->socket.sk->sk_filter &&
1100 sk_filter(tfile->socket.sk, skb)) {
1101 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
1105 len = run_ebpf_filter(tun, skb, len);
1107 drop_reason = SKB_DROP_REASON_TAP_FILTER;
1111 if (pskb_trim(skb, len)) {
1112 drop_reason = SKB_DROP_REASON_NOMEM;
1116 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) {
1117 drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
1121 skb_tx_timestamp(skb);
1123 /* Orphan the skb - required as we might hang on to it
1124 * for indefinite time.
1130 if (ptr_ring_produce(&tfile->tx_ring, skb)) {
1131 drop_reason = SKB_DROP_REASON_FULL_RING;
1135 /* NETIF_F_LLTX requires to do our own update of trans_start */
1136 queue = netdev_get_tx_queue(dev, txq);
1137 txq_trans_cond_update(queue);
1139 /* Notify and wake up reader process */
1140 if (tfile->flags & TUN_FASYNC)
1141 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1142 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1145 return NETDEV_TX_OK;
1148 dev_core_stats_tx_dropped_inc(dev);
1150 kfree_skb_reason(skb, drop_reason);
1152 return NET_XMIT_DROP;
1155 static void tun_net_mclist(struct net_device *dev)
1158 * This callback is supposed to deal with mc filter in
1159 * _rx_ path and has nothing to do with the _tx_ path.
1160 * In rx path we always accept everything userspace gives us.
1164 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1165 netdev_features_t features)
1167 struct tun_struct *tun = netdev_priv(dev);
1169 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1172 static void tun_set_headroom(struct net_device *dev, int new_hr)
1174 struct tun_struct *tun = netdev_priv(dev);
1176 if (new_hr < NET_SKB_PAD)
1177 new_hr = NET_SKB_PAD;
1179 tun->align = new_hr;
1183 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1185 struct tun_struct *tun = netdev_priv(dev);
1187 dev_get_tstats64(dev, stats);
1189 stats->rx_frame_errors +=
1190 (unsigned long)atomic_long_read(&tun->rx_frame_errors);
1193 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1194 struct netlink_ext_ack *extack)
1196 struct tun_struct *tun = netdev_priv(dev);
1197 struct tun_file *tfile;
1198 struct bpf_prog *old_prog;
1201 old_prog = rtnl_dereference(tun->xdp_prog);
1202 rcu_assign_pointer(tun->xdp_prog, prog);
1204 bpf_prog_put(old_prog);
1206 for (i = 0; i < tun->numqueues; i++) {
1207 tfile = rtnl_dereference(tun->tfiles[i]);
1209 sock_set_flag(&tfile->sk, SOCK_XDP);
1211 sock_reset_flag(&tfile->sk, SOCK_XDP);
1213 list_for_each_entry(tfile, &tun->disabled, next) {
1215 sock_set_flag(&tfile->sk, SOCK_XDP);
1217 sock_reset_flag(&tfile->sk, SOCK_XDP);
1223 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1225 switch (xdp->command) {
1226 case XDP_SETUP_PROG:
1227 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1233 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1236 struct tun_struct *tun = netdev_priv(dev);
1238 if (!tun->numqueues)
1241 netif_carrier_on(dev);
1243 netif_carrier_off(dev);
1248 static const struct net_device_ops tun_netdev_ops = {
1249 .ndo_init = tun_net_init,
1250 .ndo_uninit = tun_net_uninit,
1251 .ndo_open = tun_net_open,
1252 .ndo_stop = tun_net_close,
1253 .ndo_start_xmit = tun_net_xmit,
1254 .ndo_fix_features = tun_net_fix_features,
1255 .ndo_select_queue = tun_select_queue,
1256 .ndo_set_rx_headroom = tun_set_headroom,
1257 .ndo_get_stats64 = tun_net_get_stats64,
1258 .ndo_change_carrier = tun_net_change_carrier,
1261 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1263 /* Notify and wake up reader process */
1264 if (tfile->flags & TUN_FASYNC)
1265 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1266 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1269 static int tun_xdp_xmit(struct net_device *dev, int n,
1270 struct xdp_frame **frames, u32 flags)
1272 struct tun_struct *tun = netdev_priv(dev);
1273 struct tun_file *tfile;
1278 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1284 numqueues = READ_ONCE(tun->numqueues);
1287 return -ENXIO; /* Caller will free/return all frames */
1290 tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1292 if (unlikely(!tfile))
1295 spin_lock(&tfile->tx_ring.producer_lock);
1296 for (i = 0; i < n; i++) {
1297 struct xdp_frame *xdp = frames[i];
1298 /* Encode the XDP flag into lowest bit for consumer to differ
1299 * XDP buffer from sk_buff.
1301 void *frame = tun_xdp_to_ptr(xdp);
1303 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1304 dev_core_stats_tx_dropped_inc(dev);
1309 spin_unlock(&tfile->tx_ring.producer_lock);
1311 if (flags & XDP_XMIT_FLUSH)
1312 __tun_xdp_flush_tfile(tfile);
1318 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1320 struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
1323 if (unlikely(!frame))
1326 nxmit = tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1328 xdp_return_frame_rx_napi(frame);
1332 static const struct net_device_ops tap_netdev_ops = {
1333 .ndo_init = tun_net_init,
1334 .ndo_uninit = tun_net_uninit,
1335 .ndo_open = tun_net_open,
1336 .ndo_stop = tun_net_close,
1337 .ndo_start_xmit = tun_net_xmit,
1338 .ndo_fix_features = tun_net_fix_features,
1339 .ndo_set_rx_mode = tun_net_mclist,
1340 .ndo_set_mac_address = eth_mac_addr,
1341 .ndo_validate_addr = eth_validate_addr,
1342 .ndo_select_queue = tun_select_queue,
1343 .ndo_features_check = passthru_features_check,
1344 .ndo_set_rx_headroom = tun_set_headroom,
1345 .ndo_get_stats64 = dev_get_tstats64,
1347 .ndo_xdp_xmit = tun_xdp_xmit,
1348 .ndo_change_carrier = tun_net_change_carrier,
1351 static void tun_flow_init(struct tun_struct *tun)
1355 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1356 INIT_HLIST_HEAD(&tun->flows[i]);
1358 tun->ageing_time = TUN_FLOW_EXPIRE;
1359 timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1360 mod_timer(&tun->flow_gc_timer,
1361 round_jiffies_up(jiffies + tun->ageing_time));
1364 static void tun_flow_uninit(struct tun_struct *tun)
1366 del_timer_sync(&tun->flow_gc_timer);
1367 tun_flow_flush(tun);
1371 #define MAX_MTU 65535
1373 /* Initialize net device. */
1374 static void tun_net_initialize(struct net_device *dev)
1376 struct tun_struct *tun = netdev_priv(dev);
1378 switch (tun->flags & TUN_TYPE_MASK) {
1380 dev->netdev_ops = &tun_netdev_ops;
1381 dev->header_ops = &ip_tunnel_header_ops;
1383 /* Point-to-Point TUN Device */
1384 dev->hard_header_len = 0;
1388 /* Zero header length */
1389 dev->type = ARPHRD_NONE;
1390 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1394 dev->netdev_ops = &tap_netdev_ops;
1395 /* Ethernet TAP Device */
1397 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1398 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1400 eth_hw_addr_random(dev);
1405 dev->min_mtu = MIN_MTU;
1406 dev->max_mtu = MAX_MTU - dev->hard_header_len;
1409 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1411 struct sock *sk = tfile->socket.sk;
1413 return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1416 /* Character device part */
1419 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1421 struct tun_file *tfile = file->private_data;
1422 struct tun_struct *tun = tun_get(tfile);
1429 sk = tfile->socket.sk;
1431 poll_wait(file, sk_sleep(sk), wait);
1433 if (!ptr_ring_empty(&tfile->tx_ring))
1434 mask |= EPOLLIN | EPOLLRDNORM;
1436 /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1437 * guarantee EPOLLOUT to be raised by either here or
1438 * tun_sock_write_space(). Then process could get notification
1439 * after it writes to a down device and meets -EIO.
1441 if (tun_sock_writeable(tun, tfile) ||
1442 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1443 tun_sock_writeable(tun, tfile)))
1444 mask |= EPOLLOUT | EPOLLWRNORM;
1446 if (tun->dev->reg_state != NETREG_REGISTERED)
1453 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1455 const struct iov_iter *it)
1457 struct sk_buff *skb;
1462 if (it->nr_segs > MAX_SKB_FRAGS + 1 ||
1463 len > (ETH_MAX_MTU - NET_SKB_PAD - NET_IP_ALIGN))
1464 return ERR_PTR(-EMSGSIZE);
1467 skb = napi_get_frags(&tfile->napi);
1470 return ERR_PTR(-ENOMEM);
1472 linear = iov_iter_single_seg_count(it);
1473 err = __skb_grow(skb, linear);
1478 skb->data_len = len - linear;
1479 skb->truesize += skb->data_len;
1481 for (i = 1; i < it->nr_segs; i++) {
1482 size_t fragsz = it->iov[i].iov_len;
1486 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1490 frag = netdev_alloc_frag(fragsz);
1495 page = virt_to_head_page(frag);
1496 skb_fill_page_desc(skb, i - 1, page,
1497 frag - page_address(page), fragsz);
1502 /* frees skb and all frags allocated with napi_alloc_frag() */
1503 napi_free_frags(&tfile->napi);
1504 return ERR_PTR(err);
1507 /* prepad is the amount to reserve at front. len is length after that.
1508 * linear is a hint as to how much to copy (usually headers). */
1509 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1510 size_t prepad, size_t len,
1511 size_t linear, int noblock)
1513 struct sock *sk = tfile->socket.sk;
1514 struct sk_buff *skb;
1517 /* Under a page? Don't bother with paged skb. */
1518 if (prepad + len < PAGE_SIZE || !linear)
1521 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1524 return ERR_PTR(err);
1526 skb_reserve(skb, prepad);
1527 skb_put(skb, linear);
1528 skb->data_len = len - linear;
1529 skb->len += len - linear;
1534 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1535 struct sk_buff *skb, int more)
1537 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1538 struct sk_buff_head process_queue;
1539 u32 rx_batched = tun->rx_batched;
1542 if (!rx_batched || (!more && skb_queue_empty(queue))) {
1544 skb_record_rx_queue(skb, tfile->queue_index);
1545 netif_receive_skb(skb);
1550 spin_lock(&queue->lock);
1551 if (!more || skb_queue_len(queue) == rx_batched) {
1552 __skb_queue_head_init(&process_queue);
1553 skb_queue_splice_tail_init(queue, &process_queue);
1556 __skb_queue_tail(queue, skb);
1558 spin_unlock(&queue->lock);
1561 struct sk_buff *nskb;
1564 while ((nskb = __skb_dequeue(&process_queue))) {
1565 skb_record_rx_queue(nskb, tfile->queue_index);
1566 netif_receive_skb(nskb);
1568 skb_record_rx_queue(skb, tfile->queue_index);
1569 netif_receive_skb(skb);
1574 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1575 int len, int noblock, bool zerocopy)
1577 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1580 if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1589 if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1590 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1596 static struct sk_buff *__tun_build_skb(struct tun_file *tfile,
1597 struct page_frag *alloc_frag, char *buf,
1598 int buflen, int len, int pad)
1600 struct sk_buff *skb = build_skb(buf, buflen);
1603 return ERR_PTR(-ENOMEM);
1605 skb_reserve(skb, pad);
1607 skb_set_owner_w(skb, tfile->socket.sk);
1609 get_page(alloc_frag->page);
1610 alloc_frag->offset += buflen;
1615 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1616 struct xdp_buff *xdp, u32 act)
1622 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1627 err = tun_xdp_tx(tun->dev, xdp);
1634 bpf_warn_invalid_xdp_action(tun->dev, xdp_prog, act);
1637 trace_xdp_exception(tun->dev, xdp_prog, act);
1640 dev_core_stats_rx_dropped_inc(tun->dev);
1647 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1648 struct tun_file *tfile,
1649 struct iov_iter *from,
1650 struct virtio_net_hdr *hdr,
1651 int len, int *skb_xdp)
1653 struct page_frag *alloc_frag = ¤t->task_frag;
1654 struct bpf_prog *xdp_prog;
1655 int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1658 int pad = TUN_RX_PAD;
1662 xdp_prog = rcu_dereference(tun->xdp_prog);
1664 pad += XDP_PACKET_HEADROOM;
1665 buflen += SKB_DATA_ALIGN(len + pad);
1668 alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1669 if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1670 return ERR_PTR(-ENOMEM);
1672 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1673 copied = copy_page_from_iter(alloc_frag->page,
1674 alloc_frag->offset + pad,
1677 return ERR_PTR(-EFAULT);
1679 /* There's a small window that XDP may be set after the check
1680 * of xdp_prog above, this should be rare and for simplicity
1681 * we do XDP on skb in case the headroom is not enough.
1683 if (hdr->gso_type || !xdp_prog) {
1685 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len,
1693 xdp_prog = rcu_dereference(tun->xdp_prog);
1695 struct xdp_buff xdp;
1698 xdp_init_buff(&xdp, buflen, &tfile->xdp_rxq);
1699 xdp_prepare_buff(&xdp, buf, pad, len, false);
1701 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1702 if (act == XDP_REDIRECT || act == XDP_TX) {
1703 get_page(alloc_frag->page);
1704 alloc_frag->offset += buflen;
1706 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1708 if (act == XDP_REDIRECT || act == XDP_TX)
1709 put_page(alloc_frag->page);
1713 if (err == XDP_REDIRECT)
1715 if (err != XDP_PASS)
1718 pad = xdp.data - xdp.data_hard_start;
1719 len = xdp.data_end - xdp.data;
1724 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad);
1732 /* Get packet from user space buffer */
1733 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1734 void *msg_control, struct iov_iter *from,
1735 int noblock, bool more)
1737 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1738 struct sk_buff *skb;
1739 size_t total_len = iov_iter_count(from);
1740 size_t len = total_len, align = tun->align, linear;
1741 struct virtio_net_hdr gso = { 0 };
1744 bool zerocopy = false;
1748 bool frags = tun_napi_frags_enabled(tfile);
1749 enum skb_drop_reason drop_reason;
1751 if (!(tun->flags & IFF_NO_PI)) {
1752 if (len < sizeof(pi))
1756 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1760 if (tun->flags & IFF_VNET_HDR) {
1761 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1763 if (len < vnet_hdr_sz)
1767 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1770 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1771 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1772 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1774 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1776 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1779 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1780 align += NET_IP_ALIGN;
1781 if (unlikely(len < ETH_HLEN ||
1782 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1786 good_linear = SKB_MAX_HEAD(align);
1789 struct iov_iter i = *from;
1791 /* There are 256 bytes to be copied in skb, so there is
1792 * enough room for skb expand head in case it is used.
1793 * The rest of the buffer is mapped from userspace.
1795 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1796 if (copylen > good_linear)
1797 copylen = good_linear;
1799 iov_iter_advance(&i, copylen);
1800 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1804 if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1805 /* For the packet that is not easy to be processed
1806 * (e.g gso or jumbo packet), we will do it at after
1807 * skb was created with generic XDP routine.
1809 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1811 dev_core_stats_rx_dropped_inc(tun->dev);
1812 return PTR_ERR(skb);
1819 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1820 linear = good_linear;
1822 linear = tun16_to_cpu(tun, gso.hdr_len);
1826 mutex_lock(&tfile->napi_mutex);
1827 skb = tun_napi_alloc_frags(tfile, copylen, from);
1828 /* tun_napi_alloc_frags() enforces a layout for the skb.
1829 * If zerocopy is enabled, then this layout will be
1830 * overwritten by zerocopy_sg_from_iter().
1834 skb = tun_alloc_skb(tfile, align, copylen, linear,
1839 if (PTR_ERR(skb) != -EAGAIN)
1840 dev_core_stats_rx_dropped_inc(tun->dev);
1842 mutex_unlock(&tfile->napi_mutex);
1843 return PTR_ERR(skb);
1847 err = zerocopy_sg_from_iter(skb, from);
1849 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1853 drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
1855 dev_core_stats_rx_dropped_inc(tun->dev);
1856 kfree_skb_reason(skb, drop_reason);
1858 tfile->napi.skb = NULL;
1859 mutex_unlock(&tfile->napi_mutex);
1866 if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1867 atomic_long_inc(&tun->rx_frame_errors);
1870 tfile->napi.skb = NULL;
1871 mutex_unlock(&tfile->napi_mutex);
1877 switch (tun->flags & TUN_TYPE_MASK) {
1879 if (tun->flags & IFF_NO_PI) {
1880 u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1882 switch (ip_version) {
1884 pi.proto = htons(ETH_P_IP);
1887 pi.proto = htons(ETH_P_IPV6);
1890 dev_core_stats_rx_dropped_inc(tun->dev);
1896 skb_reset_mac_header(skb);
1897 skb->protocol = pi.proto;
1898 skb->dev = tun->dev;
1901 if (frags && !pskb_may_pull(skb, ETH_HLEN)) {
1903 drop_reason = SKB_DROP_REASON_HDR_TRUNC;
1906 skb->protocol = eth_type_trans(skb, tun->dev);
1910 /* copy skb_ubuf_info for callback when skb has no error */
1912 skb_zcopy_init(skb, msg_control);
1913 } else if (msg_control) {
1914 struct ubuf_info *uarg = msg_control;
1915 uarg->callback(NULL, uarg, false);
1918 skb_reset_network_header(skb);
1919 skb_probe_transport_header(skb);
1920 skb_record_rx_queue(skb, tfile->queue_index);
1923 struct bpf_prog *xdp_prog;
1928 xdp_prog = rcu_dereference(tun->xdp_prog);
1930 ret = do_xdp_generic(xdp_prog, skb);
1931 if (ret != XDP_PASS) {
1935 tfile->napi.skb = NULL;
1936 mutex_unlock(&tfile->napi_mutex);
1945 /* Compute the costly rx hash only if needed for flow updates.
1946 * We may get a very small possibility of OOO during switching, not
1947 * worth to optimize.
1949 if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1951 rxhash = __skb_get_hash_symmetric(skb);
1954 if (unlikely(!(tun->dev->flags & IFF_UP))) {
1957 drop_reason = SKB_DROP_REASON_DEV_READY;
1964 /* Exercise flow dissector code path. */
1965 skb_push(skb, ETH_HLEN);
1966 headlen = eth_get_headlen(tun->dev, skb->data,
1969 if (unlikely(headlen > skb_headlen(skb))) {
1972 dev_core_stats_rx_dropped_inc(tun->dev);
1974 napi_free_frags(&tfile->napi);
1976 mutex_unlock(&tfile->napi_mutex);
1980 if (likely(napi_schedule_prep(&tfile->napi))) {
1982 napi_gro_frags(&tfile->napi);
1983 napi_complete(&tfile->napi);
1989 mutex_unlock(&tfile->napi_mutex);
1990 } else if (tfile->napi_enabled) {
1991 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1994 spin_lock_bh(&queue->lock);
1995 __skb_queue_tail(queue, skb);
1996 queue_len = skb_queue_len(queue);
1997 spin_unlock(&queue->lock);
1999 if (!more || queue_len > NAPI_POLL_WEIGHT)
2000 napi_schedule(&tfile->napi);
2003 } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
2004 tun_rx_batched(tun, tfile, skb, more);
2011 dev_sw_netstats_rx_add(tun->dev, len);
2015 tun_flow_update(tun, rxhash, tfile);
2020 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
2022 struct file *file = iocb->ki_filp;
2023 struct tun_file *tfile = file->private_data;
2024 struct tun_struct *tun = tun_get(tfile);
2031 if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2034 result = tun_get_user(tun, tfile, NULL, from, noblock, false);
2040 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2041 struct tun_file *tfile,
2042 struct xdp_frame *xdp_frame,
2043 struct iov_iter *iter)
2045 int vnet_hdr_sz = 0;
2046 size_t size = xdp_frame->len;
2049 if (tun->flags & IFF_VNET_HDR) {
2050 struct virtio_net_hdr gso = { 0 };
2052 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2053 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2055 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2058 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2061 ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2064 dev_sw_netstats_tx_add(tun->dev, 1, ret);
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 };
2078 int vlan_offset = 0;
2080 int vnet_hdr_sz = 0;
2082 if (skb_vlan_tag_present(skb))
2083 vlan_hlen = VLAN_HLEN;
2085 if (tun->flags & IFF_VNET_HDR)
2086 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2088 total = skb->len + vlan_hlen + vnet_hdr_sz;
2090 if (!(tun->flags & IFF_NO_PI)) {
2091 if (iov_iter_count(iter) < sizeof(pi))
2094 total += sizeof(pi);
2095 if (iov_iter_count(iter) < total) {
2096 /* Packet will be striped */
2097 pi.flags |= TUN_PKT_STRIP;
2100 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2105 struct virtio_net_hdr gso;
2107 if (iov_iter_count(iter) < vnet_hdr_sz)
2110 if (virtio_net_hdr_from_skb(skb, &gso,
2111 tun_is_little_endian(tun), true,
2113 struct skb_shared_info *sinfo = skb_shinfo(skb);
2114 pr_err("unexpected GSO type: "
2115 "0x%x, gso_size %d, hdr_len %d\n",
2116 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2117 tun16_to_cpu(tun, gso.hdr_len));
2118 print_hex_dump(KERN_ERR, "tun: ",
2121 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2126 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2129 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2136 veth.h_vlan_proto = skb->vlan_proto;
2137 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2139 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2141 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2142 if (ret || !iov_iter_count(iter))
2145 ret = copy_to_iter(&veth, sizeof(veth), iter);
2146 if (ret != sizeof(veth) || !iov_iter_count(iter))
2150 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2153 /* caller is in process context, */
2155 dev_sw_netstats_tx_add(tun->dev, 1, skb->len + vlan_hlen);
2161 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2163 DECLARE_WAITQUEUE(wait, current);
2167 ptr = ptr_ring_consume(&tfile->tx_ring);
2175 add_wait_queue(&tfile->socket.wq.wait, &wait);
2178 set_current_state(TASK_INTERRUPTIBLE);
2179 ptr = ptr_ring_consume(&tfile->tx_ring);
2182 if (signal_pending(current)) {
2183 error = -ERESTARTSYS;
2186 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2194 __set_current_state(TASK_RUNNING);
2195 remove_wait_queue(&tfile->socket.wq.wait, &wait);
2202 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2203 struct iov_iter *to,
2204 int noblock, void *ptr)
2209 if (!iov_iter_count(to)) {
2215 /* Read frames from ring */
2216 ptr = tun_ring_recv(tfile, noblock, &err);
2221 if (tun_is_xdp_frame(ptr)) {
2222 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2224 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2225 xdp_return_frame(xdpf);
2227 struct sk_buff *skb = ptr;
2229 ret = tun_put_user(tun, tfile, skb, to);
2230 if (unlikely(ret < 0))
2239 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2241 struct file *file = iocb->ki_filp;
2242 struct tun_file *tfile = file->private_data;
2243 struct tun_struct *tun = tun_get(tfile);
2244 ssize_t len = iov_iter_count(to), ret;
2250 if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2253 ret = tun_do_read(tun, tfile, to, noblock, NULL);
2254 ret = min_t(ssize_t, ret, len);
2261 static void tun_prog_free(struct rcu_head *rcu)
2263 struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2265 bpf_prog_destroy(prog->prog);
2269 static int __tun_set_ebpf(struct tun_struct *tun,
2270 struct tun_prog __rcu **prog_p,
2271 struct bpf_prog *prog)
2273 struct tun_prog *old, *new = NULL;
2276 new = kmalloc(sizeof(*new), GFP_KERNEL);
2282 spin_lock_bh(&tun->lock);
2283 old = rcu_dereference_protected(*prog_p,
2284 lockdep_is_held(&tun->lock));
2285 rcu_assign_pointer(*prog_p, new);
2286 spin_unlock_bh(&tun->lock);
2289 call_rcu(&old->rcu, tun_prog_free);
2294 static void tun_free_netdev(struct net_device *dev)
2296 struct tun_struct *tun = netdev_priv(dev);
2298 BUG_ON(!(list_empty(&tun->disabled)));
2300 free_percpu(dev->tstats);
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 bpf_prog *xdp_prog;
2431 struct sk_buff *skb = NULL;
2432 struct sk_buff_head *queue;
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) {
2446 xdp_init_buff(xdp, buflen, &tfile->xdp_rxq);
2447 xdp_set_data_meta_invalid(xdp);
2449 act = bpf_prog_run_xdp(xdp_prog, xdp);
2450 ret = tun_xdp_act(tun, xdp_prog, xdp, act);
2452 put_page(virt_to_head_page(xdp->data));
2465 page = virt_to_head_page(xdp->data);
2466 if (tpage->page == page) {
2469 tun_put_page(tpage);
2478 skb = build_skb(xdp->data_hard_start, buflen);
2484 skb_reserve(skb, xdp->data - xdp->data_hard_start);
2485 skb_put(skb, xdp->data_end - xdp->data);
2487 if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2488 atomic_long_inc(&tun->rx_frame_errors);
2494 skb->protocol = eth_type_trans(skb, tun->dev);
2495 skb_reset_network_header(skb);
2496 skb_probe_transport_header(skb);
2497 skb_record_rx_queue(skb, tfile->queue_index);
2500 ret = do_xdp_generic(xdp_prog, skb);
2501 if (ret != XDP_PASS) {
2507 if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2509 rxhash = __skb_get_hash_symmetric(skb);
2511 if (tfile->napi_enabled) {
2512 queue = &tfile->sk.sk_write_queue;
2513 spin_lock(&queue->lock);
2514 __skb_queue_tail(queue, skb);
2515 spin_unlock(&queue->lock);
2518 netif_receive_skb(skb);
2522 /* No need to disable preemption here since this function is
2523 * always called with bh disabled
2525 dev_sw_netstats_rx_add(tun->dev, datasize);
2528 tun_flow_update(tun, rxhash, tfile);
2534 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2537 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2538 struct tun_struct *tun = tun_get(tfile);
2539 struct tun_msg_ctl *ctl = m->msg_control;
2540 struct xdp_buff *xdp;
2545 if (m->msg_controllen == sizeof(struct tun_msg_ctl) &&
2546 ctl && ctl->type == TUN_MSG_PTR) {
2547 struct tun_page tpage;
2549 int flush = 0, queued = 0;
2551 memset(&tpage, 0, sizeof(tpage));
2556 for (i = 0; i < n; i++) {
2557 xdp = &((struct xdp_buff *)ctl->ptr)[i];
2558 ret = tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2566 if (tfile->napi_enabled && queued > 0)
2567 napi_schedule(&tfile->napi);
2572 tun_put_page(&tpage);
2578 ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2579 m->msg_flags & MSG_DONTWAIT,
2580 m->msg_flags & MSG_MORE);
2586 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2589 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2590 struct tun_struct *tun = tun_get(tfile);
2591 void *ptr = m->msg_control;
2599 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2603 if (flags & MSG_ERRQUEUE) {
2604 ret = sock_recv_errqueue(sock->sk, m, total_len,
2605 SOL_PACKET, TUN_TX_TIMESTAMP);
2608 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2609 if (ret > (ssize_t)total_len) {
2610 m->msg_flags |= MSG_TRUNC;
2611 ret = flags & MSG_TRUNC ? ret : total_len;
2624 static int tun_ptr_peek_len(void *ptr)
2627 if (tun_is_xdp_frame(ptr)) {
2628 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2632 return __skb_array_len_with_tag(ptr);
2638 static int tun_peek_len(struct socket *sock)
2640 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2641 struct tun_struct *tun;
2644 tun = tun_get(tfile);
2648 ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2654 /* Ops structure to mimic raw sockets with tun */
2655 static const struct proto_ops tun_socket_ops = {
2656 .peek_len = tun_peek_len,
2657 .sendmsg = tun_sendmsg,
2658 .recvmsg = tun_recvmsg,
2661 static struct proto tun_proto = {
2663 .owner = THIS_MODULE,
2664 .obj_size = sizeof(struct tun_file),
2667 static int tun_flags(struct tun_struct *tun)
2669 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2672 static ssize_t tun_flags_show(struct device *dev, struct device_attribute *attr,
2675 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2676 return sysfs_emit(buf, "0x%x\n", tun_flags(tun));
2679 static ssize_t owner_show(struct device *dev, struct device_attribute *attr,
2682 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2683 return uid_valid(tun->owner)?
2684 sysfs_emit(buf, "%u\n",
2685 from_kuid_munged(current_user_ns(), tun->owner)) :
2686 sysfs_emit(buf, "-1\n");
2689 static ssize_t group_show(struct device *dev, struct device_attribute *attr,
2692 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2693 return gid_valid(tun->group) ?
2694 sysfs_emit(buf, "%u\n",
2695 from_kgid_munged(current_user_ns(), tun->group)) :
2696 sysfs_emit(buf, "-1\n");
2699 static DEVICE_ATTR_RO(tun_flags);
2700 static DEVICE_ATTR_RO(owner);
2701 static DEVICE_ATTR_RO(group);
2703 static struct attribute *tun_dev_attrs[] = {
2704 &dev_attr_tun_flags.attr,
2705 &dev_attr_owner.attr,
2706 &dev_attr_group.attr,
2710 static const struct attribute_group tun_attr_group = {
2711 .attrs = tun_dev_attrs
2714 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2716 struct tun_struct *tun;
2717 struct tun_file *tfile = file->private_data;
2718 struct net_device *dev;
2721 if (tfile->detached)
2724 if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2725 if (!capable(CAP_NET_ADMIN))
2728 if (!(ifr->ifr_flags & IFF_NAPI) ||
2729 (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2733 dev = __dev_get_by_name(net, ifr->ifr_name);
2735 if (ifr->ifr_flags & IFF_TUN_EXCL)
2737 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2738 tun = netdev_priv(dev);
2739 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2740 tun = netdev_priv(dev);
2744 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2745 !!(tun->flags & IFF_MULTI_QUEUE))
2748 if (tun_not_capable(tun))
2750 err = security_tun_dev_open(tun->security);
2754 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2755 ifr->ifr_flags & IFF_NAPI,
2756 ifr->ifr_flags & IFF_NAPI_FRAGS, true);
2760 if (tun->flags & IFF_MULTI_QUEUE &&
2761 (tun->numqueues + tun->numdisabled > 1)) {
2762 /* One or more queue has already been attached, no need
2763 * to initialize the device again.
2765 netdev_state_change(dev);
2769 tun->flags = (tun->flags & ~TUN_FEATURES) |
2770 (ifr->ifr_flags & TUN_FEATURES);
2772 netdev_state_change(dev);
2775 unsigned long flags = 0;
2776 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2779 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2781 err = security_tun_dev_create();
2786 if (ifr->ifr_flags & IFF_TUN) {
2790 } else if (ifr->ifr_flags & IFF_TAP) {
2798 name = ifr->ifr_name;
2800 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2801 NET_NAME_UNKNOWN, tun_setup, queues,
2807 dev_net_set(dev, net);
2808 dev->rtnl_link_ops = &tun_link_ops;
2809 dev->ifindex = tfile->ifindex;
2810 dev->sysfs_groups[0] = &tun_attr_group;
2812 tun = netdev_priv(dev);
2815 tun->txflt.count = 0;
2816 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2818 tun->align = NET_SKB_PAD;
2819 tun->filter_attached = false;
2820 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2821 tun->rx_batched = 0;
2822 RCU_INIT_POINTER(tun->steering_prog, NULL);
2827 tun_net_initialize(dev);
2829 err = register_netdevice(tun->dev);
2834 /* free_netdev() won't check refcnt, to avoid race
2835 * with dev_put() we need publish tun after registration.
2837 rcu_assign_pointer(tfile->tun, tun);
2840 if (ifr->ifr_flags & IFF_NO_CARRIER)
2841 netif_carrier_off(tun->dev);
2843 netif_carrier_on(tun->dev);
2845 /* Make sure persistent devices do not get stuck in
2848 if (netif_running(tun->dev))
2849 netif_tx_wake_all_queues(tun->dev);
2851 strcpy(ifr->ifr_name, tun->dev->name);
2855 static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
2857 strcpy(ifr->ifr_name, tun->dev->name);
2859 ifr->ifr_flags = tun_flags(tun);
2863 /* This is like a cut-down ethtool ops, except done via tun fd so no
2864 * privs required. */
2865 static int set_offload(struct tun_struct *tun, unsigned long arg)
2867 netdev_features_t features = 0;
2869 if (arg & TUN_F_CSUM) {
2870 features |= NETIF_F_HW_CSUM;
2873 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2874 if (arg & TUN_F_TSO_ECN) {
2875 features |= NETIF_F_TSO_ECN;
2876 arg &= ~TUN_F_TSO_ECN;
2878 if (arg & TUN_F_TSO4)
2879 features |= NETIF_F_TSO;
2880 if (arg & TUN_F_TSO6)
2881 features |= NETIF_F_TSO6;
2882 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2888 /* This gives the user a way to test for new features in future by
2889 * trying to set them. */
2893 tun->set_features = features;
2894 tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2895 tun->dev->wanted_features |= features;
2896 netdev_update_features(tun->dev);
2901 static void tun_detach_filter(struct tun_struct *tun, int n)
2904 struct tun_file *tfile;
2906 for (i = 0; i < n; i++) {
2907 tfile = rtnl_dereference(tun->tfiles[i]);
2908 lock_sock(tfile->socket.sk);
2909 sk_detach_filter(tfile->socket.sk);
2910 release_sock(tfile->socket.sk);
2913 tun->filter_attached = false;
2916 static int tun_attach_filter(struct tun_struct *tun)
2919 struct tun_file *tfile;
2921 for (i = 0; i < tun->numqueues; i++) {
2922 tfile = rtnl_dereference(tun->tfiles[i]);
2923 lock_sock(tfile->socket.sk);
2924 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2925 release_sock(tfile->socket.sk);
2927 tun_detach_filter(tun, i);
2932 tun->filter_attached = true;
2936 static void tun_set_sndbuf(struct tun_struct *tun)
2938 struct tun_file *tfile;
2941 for (i = 0; i < tun->numqueues; i++) {
2942 tfile = rtnl_dereference(tun->tfiles[i]);
2943 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2947 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2949 struct tun_file *tfile = file->private_data;
2950 struct tun_struct *tun;
2955 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2956 tun = tfile->detached;
2961 ret = security_tun_dev_attach_queue(tun->security);
2964 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2965 tun->flags & IFF_NAPI_FRAGS, true);
2966 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2967 tun = rtnl_dereference(tfile->tun);
2968 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2971 __tun_detach(tfile, false);
2976 netdev_state_change(tun->dev);
2983 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog __rcu **prog_p,
2986 struct bpf_prog *prog;
2989 if (copy_from_user(&fd, data, sizeof(fd)))
2995 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2997 return PTR_ERR(prog);
3000 return __tun_set_ebpf(tun, prog_p, prog);
3003 /* Return correct value for tun->dev->addr_len based on tun->dev->type. */
3004 static unsigned char tun_get_addr_len(unsigned short type)
3008 case ARPHRD_TUNNEL6:
3009 return sizeof(struct in6_addr);
3016 case ARPHRD_IEEE802154:
3017 case ARPHRD_IEEE802154_MONITOR:
3018 return IEEE802154_EXTENDED_ADDR_LEN;
3019 case ARPHRD_PHONET_PIPE:
3023 case ARPHRD_6LOWPAN:
3024 return EUI64_ADDR_LEN;
3029 case ARPHRD_IEEE802:
3032 return ROSE_ADDR_LEN;
3034 return AX25_ADDR_LEN;
3035 case ARPHRD_LOCALTLK:
3042 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
3043 unsigned long arg, int ifreq_len)
3045 struct tun_file *tfile = file->private_data;
3046 struct net *net = sock_net(&tfile->sk);
3047 struct tun_struct *tun;
3048 void __user* argp = (void __user*)arg;
3049 unsigned int ifindex, carrier;
3057 bool do_notify = false;
3059 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
3060 (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
3061 if (copy_from_user(&ifr, argp, ifreq_len))
3064 memset(&ifr, 0, sizeof(ifr));
3066 if (cmd == TUNGETFEATURES) {
3067 /* Currently this just means: "what IFF flags are valid?".
3068 * This is needed because we never checked for invalid flags on
3071 return put_user(IFF_TUN | IFF_TAP | IFF_NO_CARRIER |
3072 TUN_FEATURES, (unsigned int __user*)argp);
3073 } else if (cmd == TUNSETQUEUE) {
3074 return tun_set_queue(file, &ifr);
3075 } else if (cmd == SIOCGSKNS) {
3076 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3078 return open_related_ns(&net->ns, get_net_ns);
3083 tun = tun_get(tfile);
3084 if (cmd == TUNSETIFF) {
3089 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3091 ret = tun_set_iff(net, file, &ifr);
3096 if (copy_to_user(argp, &ifr, ifreq_len))
3100 if (cmd == TUNSETIFINDEX) {
3106 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3110 tfile->ifindex = ifindex;
3118 netif_info(tun, drv, tun->dev, "tun_chr_ioctl cmd %u\n", cmd);
3120 net = dev_net(tun->dev);
3124 tun_get_iff(tun, &ifr);
3126 if (tfile->detached)
3127 ifr.ifr_flags |= IFF_DETACH_QUEUE;
3128 if (!tfile->socket.sk->sk_filter)
3129 ifr.ifr_flags |= IFF_NOFILTER;
3131 if (copy_to_user(argp, &ifr, ifreq_len))
3136 /* Disable/Enable checksum */
3138 /* [unimplemented] */
3139 netif_info(tun, drv, tun->dev, "ignored: set checksum %s\n",
3140 arg ? "disabled" : "enabled");
3144 /* Disable/Enable persist mode. Keep an extra reference to the
3145 * module to prevent the module being unprobed.
3147 if (arg && !(tun->flags & IFF_PERSIST)) {
3148 tun->flags |= IFF_PERSIST;
3149 __module_get(THIS_MODULE);
3152 if (!arg && (tun->flags & IFF_PERSIST)) {
3153 tun->flags &= ~IFF_PERSIST;
3154 module_put(THIS_MODULE);
3158 netif_info(tun, drv, tun->dev, "persist %s\n",
3159 arg ? "enabled" : "disabled");
3163 /* Set owner of the device */
3164 owner = make_kuid(current_user_ns(), arg);
3165 if (!uid_valid(owner)) {
3171 netif_info(tun, drv, tun->dev, "owner set to %u\n",
3172 from_kuid(&init_user_ns, tun->owner));
3176 /* Set group of the device */
3177 group = make_kgid(current_user_ns(), arg);
3178 if (!gid_valid(group)) {
3184 netif_info(tun, drv, tun->dev, "group set to %u\n",
3185 from_kgid(&init_user_ns, tun->group));
3189 /* Only allow setting the type when the interface is down */
3190 if (tun->dev->flags & IFF_UP) {
3191 netif_info(tun, drv, tun->dev,
3192 "Linktype set failed because interface is up\n");
3195 ret = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
3197 ret = notifier_to_errno(ret);
3199 netif_info(tun, drv, tun->dev,
3200 "Refused to change device type\n");
3203 tun->dev->type = (int) arg;
3204 tun->dev->addr_len = tun_get_addr_len(tun->dev->type);
3205 netif_info(tun, drv, tun->dev, "linktype set to %d\n",
3207 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
3213 tun->msg_enable = (u32)arg;
3217 ret = set_offload(tun, arg);
3220 case TUNSETTXFILTER:
3221 /* Can be set only for TAPs */
3223 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3225 ret = update_filter(&tun->txflt, (void __user *)arg);
3229 /* Get hw address */
3230 dev_get_mac_address(&ifr.ifr_hwaddr, net, tun->dev->name);
3231 if (copy_to_user(argp, &ifr, ifreq_len))
3236 /* Set hw address */
3237 ret = dev_set_mac_address_user(tun->dev, &ifr.ifr_hwaddr, NULL);
3241 sndbuf = tfile->socket.sk->sk_sndbuf;
3242 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3247 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3256 tun->sndbuf = sndbuf;
3257 tun_set_sndbuf(tun);
3260 case TUNGETVNETHDRSZ:
3261 vnet_hdr_sz = tun->vnet_hdr_sz;
3262 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3266 case TUNSETVNETHDRSZ:
3267 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3271 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3276 tun->vnet_hdr_sz = vnet_hdr_sz;
3280 le = !!(tun->flags & TUN_VNET_LE);
3281 if (put_user(le, (int __user *)argp))
3286 if (get_user(le, (int __user *)argp)) {
3291 tun->flags |= TUN_VNET_LE;
3293 tun->flags &= ~TUN_VNET_LE;
3297 ret = tun_get_vnet_be(tun, argp);
3301 ret = tun_set_vnet_be(tun, argp);
3304 case TUNATTACHFILTER:
3305 /* Can be set only for TAPs */
3307 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3310 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3313 ret = tun_attach_filter(tun);
3316 case TUNDETACHFILTER:
3317 /* Can be set only for TAPs */
3319 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3322 tun_detach_filter(tun, tun->numqueues);
3327 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3330 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3335 case TUNSETSTEERINGEBPF:
3336 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3339 case TUNSETFILTEREBPF:
3340 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3345 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3348 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3351 case TUNGETDEVNETNS:
3353 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3355 ret = open_related_ns(&net->ns, get_net_ns);
3364 netdev_state_change(tun->dev);
3373 static long tun_chr_ioctl(struct file *file,
3374 unsigned int cmd, unsigned long arg)
3376 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3379 #ifdef CONFIG_COMPAT
3380 static long tun_chr_compat_ioctl(struct file *file,
3381 unsigned int cmd, unsigned long arg)
3386 case TUNSETTXFILTER:
3391 arg = (unsigned long)compat_ptr(arg);
3394 arg = (compat_ulong_t)arg;
3399 * compat_ifreq is shorter than ifreq, so we must not access beyond
3400 * the end of that structure. All fields that are used in this
3401 * driver are compatible though, we don't need to convert the
3404 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3406 #endif /* CONFIG_COMPAT */
3408 static int tun_chr_fasync(int fd, struct file *file, int on)
3410 struct tun_file *tfile = file->private_data;
3413 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3417 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3418 tfile->flags |= TUN_FASYNC;
3420 tfile->flags &= ~TUN_FASYNC;
3426 static int tun_chr_open(struct inode *inode, struct file * file)
3428 struct net *net = current->nsproxy->net_ns;
3429 struct tun_file *tfile;
3431 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3435 if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3436 sk_free(&tfile->sk);
3440 mutex_init(&tfile->napi_mutex);
3441 RCU_INIT_POINTER(tfile->tun, NULL);
3445 init_waitqueue_head(&tfile->socket.wq.wait);
3447 tfile->socket.file = file;
3448 tfile->socket.ops = &tun_socket_ops;
3450 sock_init_data(&tfile->socket, &tfile->sk);
3452 tfile->sk.sk_write_space = tun_sock_write_space;
3453 tfile->sk.sk_sndbuf = INT_MAX;
3455 file->private_data = tfile;
3456 INIT_LIST_HEAD(&tfile->next);
3458 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3463 static int tun_chr_close(struct inode *inode, struct file *file)
3465 struct tun_file *tfile = file->private_data;
3467 tun_detach(tfile, true);
3472 #ifdef CONFIG_PROC_FS
3473 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3475 struct tun_file *tfile = file->private_data;
3476 struct tun_struct *tun;
3479 memset(&ifr, 0, sizeof(ifr));
3482 tun = tun_get(tfile);
3484 tun_get_iff(tun, &ifr);
3490 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3494 static const struct file_operations tun_fops = {
3495 .owner = THIS_MODULE,
3496 .llseek = no_llseek,
3497 .read_iter = tun_chr_read_iter,
3498 .write_iter = tun_chr_write_iter,
3499 .poll = tun_chr_poll,
3500 .unlocked_ioctl = tun_chr_ioctl,
3501 #ifdef CONFIG_COMPAT
3502 .compat_ioctl = tun_chr_compat_ioctl,
3504 .open = tun_chr_open,
3505 .release = tun_chr_close,
3506 .fasync = tun_chr_fasync,
3507 #ifdef CONFIG_PROC_FS
3508 .show_fdinfo = tun_chr_show_fdinfo,
3512 static struct miscdevice tun_miscdev = {
3515 .nodename = "net/tun",
3519 /* ethtool interface */
3521 static void tun_default_link_ksettings(struct net_device *dev,
3522 struct ethtool_link_ksettings *cmd)
3524 ethtool_link_ksettings_zero_link_mode(cmd, supported);
3525 ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3526 cmd->base.speed = SPEED_10;
3527 cmd->base.duplex = DUPLEX_FULL;
3528 cmd->base.port = PORT_TP;
3529 cmd->base.phy_address = 0;
3530 cmd->base.autoneg = AUTONEG_DISABLE;
3533 static int tun_get_link_ksettings(struct net_device *dev,
3534 struct ethtool_link_ksettings *cmd)
3536 struct tun_struct *tun = netdev_priv(dev);
3538 memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3542 static int tun_set_link_ksettings(struct net_device *dev,
3543 const struct ethtool_link_ksettings *cmd)
3545 struct tun_struct *tun = netdev_priv(dev);
3547 memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3551 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3553 struct tun_struct *tun = netdev_priv(dev);
3555 strscpy(info->driver, DRV_NAME, sizeof(info->driver));
3556 strscpy(info->version, DRV_VERSION, sizeof(info->version));
3558 switch (tun->flags & TUN_TYPE_MASK) {
3560 strscpy(info->bus_info, "tun", sizeof(info->bus_info));
3563 strscpy(info->bus_info, "tap", sizeof(info->bus_info));
3568 static u32 tun_get_msglevel(struct net_device *dev)
3570 struct tun_struct *tun = netdev_priv(dev);
3572 return tun->msg_enable;
3575 static void tun_set_msglevel(struct net_device *dev, u32 value)
3577 struct tun_struct *tun = netdev_priv(dev);
3579 tun->msg_enable = value;
3582 static int tun_get_coalesce(struct net_device *dev,
3583 struct ethtool_coalesce *ec,
3584 struct kernel_ethtool_coalesce *kernel_coal,
3585 struct netlink_ext_ack *extack)
3587 struct tun_struct *tun = netdev_priv(dev);
3589 ec->rx_max_coalesced_frames = tun->rx_batched;
3594 static int tun_set_coalesce(struct net_device *dev,
3595 struct ethtool_coalesce *ec,
3596 struct kernel_ethtool_coalesce *kernel_coal,
3597 struct netlink_ext_ack *extack)
3599 struct tun_struct *tun = netdev_priv(dev);
3601 if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3602 tun->rx_batched = NAPI_POLL_WEIGHT;
3604 tun->rx_batched = ec->rx_max_coalesced_frames;
3609 static const struct ethtool_ops tun_ethtool_ops = {
3610 .supported_coalesce_params = ETHTOOL_COALESCE_RX_MAX_FRAMES,
3611 .get_drvinfo = tun_get_drvinfo,
3612 .get_msglevel = tun_get_msglevel,
3613 .set_msglevel = tun_set_msglevel,
3614 .get_link = ethtool_op_get_link,
3615 .get_ts_info = ethtool_op_get_ts_info,
3616 .get_coalesce = tun_get_coalesce,
3617 .set_coalesce = tun_set_coalesce,
3618 .get_link_ksettings = tun_get_link_ksettings,
3619 .set_link_ksettings = tun_set_link_ksettings,
3622 static int tun_queue_resize(struct tun_struct *tun)
3624 struct net_device *dev = tun->dev;
3625 struct tun_file *tfile;
3626 struct ptr_ring **rings;
3627 int n = tun->numqueues + tun->numdisabled;
3630 rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3634 for (i = 0; i < tun->numqueues; i++) {
3635 tfile = rtnl_dereference(tun->tfiles[i]);
3636 rings[i] = &tfile->tx_ring;
3638 list_for_each_entry(tfile, &tun->disabled, next)
3639 rings[i++] = &tfile->tx_ring;
3641 ret = ptr_ring_resize_multiple(rings, n,
3642 dev->tx_queue_len, GFP_KERNEL,
3649 static int tun_device_event(struct notifier_block *unused,
3650 unsigned long event, void *ptr)
3652 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3653 struct tun_struct *tun = netdev_priv(dev);
3656 if (dev->rtnl_link_ops != &tun_link_ops)
3660 case NETDEV_CHANGE_TX_QUEUE_LEN:
3661 if (tun_queue_resize(tun))
3665 for (i = 0; i < tun->numqueues; i++) {
3666 struct tun_file *tfile;
3668 tfile = rtnl_dereference(tun->tfiles[i]);
3669 tfile->socket.sk->sk_write_space(tfile->socket.sk);
3679 static struct notifier_block tun_notifier_block __read_mostly = {
3680 .notifier_call = tun_device_event,
3683 static int __init tun_init(void)
3687 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3689 ret = rtnl_link_register(&tun_link_ops);
3691 pr_err("Can't register link_ops\n");
3695 ret = misc_register(&tun_miscdev);
3697 pr_err("Can't register misc device %d\n", TUN_MINOR);
3701 ret = register_netdevice_notifier(&tun_notifier_block);
3703 pr_err("Can't register netdevice notifier\n");
3710 misc_deregister(&tun_miscdev);
3712 rtnl_link_unregister(&tun_link_ops);
3717 static void tun_cleanup(void)
3719 misc_deregister(&tun_miscdev);
3720 rtnl_link_unregister(&tun_link_ops);
3721 unregister_netdevice_notifier(&tun_notifier_block);
3724 /* Get an underlying socket object from tun file. Returns error unless file is
3725 * attached to a device. The returned object works like a packet socket, it
3726 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
3727 * holding a reference to the file for as long as the socket is in use. */
3728 struct socket *tun_get_socket(struct file *file)
3730 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->socket;
3738 EXPORT_SYMBOL_GPL(tun_get_socket);
3740 struct ptr_ring *tun_get_tx_ring(struct file *file)
3742 struct tun_file *tfile;
3744 if (file->f_op != &tun_fops)
3745 return ERR_PTR(-EINVAL);
3746 tfile = file->private_data;
3748 return ERR_PTR(-EBADFD);
3749 return &tfile->tx_ring;
3751 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3753 module_init(tun_init);
3754 module_exit(tun_cleanup);
3755 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3756 MODULE_AUTHOR(DRV_COPYRIGHT);
3757 MODULE_LICENSE("GPL");
3758 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3759 MODULE_ALIAS("devname:net/tun");