2 * TUN - Universal TUN/TAP device driver.
3 * Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
21 * Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
22 * Add TUNSETLINK ioctl to set the link encapsulation
24 * Mark Smith <markzzzsmith@yahoo.com.au>
25 * Use eth_random_addr() for tap MAC address.
27 * Harald Roelle <harald.roelle@ifi.lmu.de> 2004/04/20
28 * Fixes in packet dropping, queue length setting and queue wakeup.
29 * Increased default tx queue length.
33 * Daniel Podlejski <underley@underley.eu.org>
34 * Modifications for 2.3.99-pre5 kernel.
37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
39 #define DRV_NAME "tun"
40 #define DRV_VERSION "1.6"
41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
42 #define DRV_COPYRIGHT "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
44 #include <linux/module.h>
45 #include <linux/errno.h>
46 #include <linux/kernel.h>
47 #include <linux/sched/signal.h>
48 #include <linux/major.h>
49 #include <linux/slab.h>
50 #include <linux/poll.h>
51 #include <linux/fcntl.h>
52 #include <linux/init.h>
53 #include <linux/skbuff.h>
54 #include <linux/netdevice.h>
55 #include <linux/etherdevice.h>
56 #include <linux/miscdevice.h>
57 #include <linux/ethtool.h>
58 #include <linux/rtnetlink.h>
59 #include <linux/compat.h>
61 #include <linux/if_arp.h>
62 #include <linux/if_ether.h>
63 #include <linux/if_tun.h>
64 #include <linux/if_vlan.h>
65 #include <linux/crc32.h>
66 #include <linux/nsproxy.h>
67 #include <linux/virtio_net.h>
68 #include <linux/rcupdate.h>
69 #include <net/net_namespace.h>
70 #include <net/netns/generic.h>
71 #include <net/rtnetlink.h>
74 #include <linux/seq_file.h>
75 #include <linux/uio.h>
76 #include <linux/skb_array.h>
77 #include <linux/bpf.h>
78 #include <linux/bpf_trace.h>
79 #include <linux/mutex.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 /* Uncomment to enable debugging */
88 /* #define TUN_DEBUG 1 */
93 #define tun_debug(level, tun, fmt, args...) \
96 netdev_printk(level, tun->dev, fmt, ##args); \
98 #define DBG1(level, fmt, args...) \
101 printk(level fmt, ##args); \
104 #define tun_debug(level, tun, fmt, args...) \
107 netdev_printk(level, tun->dev, fmt, ##args); \
109 #define DBG1(level, fmt, args...) \
112 printk(level fmt, ##args); \
116 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
118 /* TUN device flags */
120 /* IFF_ATTACH_QUEUE is never stored in device flags,
121 * overload it to mean fasync when stored there.
123 #define TUN_FASYNC IFF_ATTACH_QUEUE
124 /* High bits in flags field are unused. */
125 #define TUN_VNET_LE 0x80000000
126 #define TUN_VNET_BE 0x40000000
128 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
129 IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
131 #define GOODCOPY_LEN 128
133 #define FLT_EXACT_COUNT 8
135 unsigned int count; /* Number of addrs. Zero means disabled */
136 u32 mask[2]; /* Mask of the hashed addrs */
137 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
140 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
141 * to max number of VCPUs in guest. */
142 #define MAX_TAP_QUEUES 256
143 #define MAX_TAP_FLOWS 4096
145 #define TUN_FLOW_EXPIRE (3 * HZ)
147 struct tun_pcpu_stats {
152 struct u64_stats_sync syncp;
158 /* A tun_file connects an open character device to a tuntap netdevice. It
159 * also contains all socket related structures (except sock_fprog and tap_filter)
160 * to serve as one transmit queue for tuntap device. The sock_fprog and
161 * tap_filter were kept in tun_struct since they were used for filtering for the
162 * netdevice not for a specific queue (at least I didn't see the requirement for
166 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
167 * other can only be read while rcu_read_lock or rtnl_lock is held.
171 struct socket socket;
173 struct tun_struct __rcu *tun;
174 struct fasync_struct *fasync;
175 /* only used for fasnyc */
179 unsigned int ifindex;
181 struct napi_struct napi;
183 bool napi_frags_enabled;
184 struct mutex napi_mutex; /* Protects access to the above napi */
185 struct list_head next;
186 struct tun_struct *detached;
187 struct ptr_ring tx_ring;
188 struct xdp_rxq_info xdp_rxq;
196 struct tun_flow_entry {
197 struct hlist_node hash_link;
199 struct tun_struct *tun;
204 unsigned long updated ____cacheline_aligned_in_smp;
207 #define TUN_NUM_FLOW_ENTRIES 1024
208 #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
212 struct bpf_prog *prog;
215 /* Since the socket were moved to tun_file, to preserve the behavior of persist
216 * device, socket filter, sndbuf and vnet header size were restore when the
217 * file were attached to a persist device.
220 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
221 unsigned int numqueues;
226 struct net_device *dev;
227 netdev_features_t set_features;
228 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
234 struct tap_filter txflt;
235 struct sock_fprog fprog;
236 /* protected by rtnl lock */
237 bool filter_attached;
242 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
243 struct timer_list flow_gc_timer;
244 unsigned long ageing_time;
245 unsigned int numdisabled;
246 struct list_head disabled;
250 struct tun_pcpu_stats __percpu *pcpu_stats;
251 struct bpf_prog __rcu *xdp_prog;
252 struct tun_prog __rcu *steering_prog;
253 struct tun_prog __rcu *filter_prog;
254 struct ethtool_link_ksettings link_ksettings;
262 bool tun_is_xdp_frame(void *ptr)
264 return (unsigned long)ptr & TUN_XDP_FLAG;
266 EXPORT_SYMBOL(tun_is_xdp_frame);
268 void *tun_xdp_to_ptr(void *ptr)
270 return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
272 EXPORT_SYMBOL(tun_xdp_to_ptr);
274 void *tun_ptr_to_xdp(void *ptr)
276 return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
278 EXPORT_SYMBOL(tun_ptr_to_xdp);
280 static int tun_napi_receive(struct napi_struct *napi, int budget)
282 struct tun_file *tfile = container_of(napi, struct tun_file, napi);
283 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
284 struct sk_buff_head process_queue;
288 __skb_queue_head_init(&process_queue);
290 spin_lock(&queue->lock);
291 skb_queue_splice_tail_init(queue, &process_queue);
292 spin_unlock(&queue->lock);
294 while (received < budget && (skb = __skb_dequeue(&process_queue))) {
295 napi_gro_receive(napi, skb);
299 if (!skb_queue_empty(&process_queue)) {
300 spin_lock(&queue->lock);
301 skb_queue_splice(&process_queue, queue);
302 spin_unlock(&queue->lock);
308 static int tun_napi_poll(struct napi_struct *napi, int budget)
310 unsigned int received;
312 received = tun_napi_receive(napi, budget);
314 if (received < budget)
315 napi_complete_done(napi, received);
320 static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
321 bool napi_en, bool napi_frags)
323 tfile->napi_enabled = napi_en;
324 tfile->napi_frags_enabled = napi_en && napi_frags;
326 netif_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
328 napi_enable(&tfile->napi);
332 static void tun_napi_disable(struct tun_file *tfile)
334 if (tfile->napi_enabled)
335 napi_disable(&tfile->napi);
338 static void tun_napi_del(struct tun_file *tfile)
340 if (tfile->napi_enabled)
341 netif_napi_del(&tfile->napi);
344 static bool tun_napi_frags_enabled(const struct tun_file *tfile)
346 return tfile->napi_frags_enabled;
349 #ifdef CONFIG_TUN_VNET_CROSS_LE
350 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
352 return tun->flags & TUN_VNET_BE ? false :
353 virtio_legacy_is_little_endian();
356 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
358 int be = !!(tun->flags & TUN_VNET_BE);
360 if (put_user(be, argp))
366 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
370 if (get_user(be, argp))
374 tun->flags |= TUN_VNET_BE;
376 tun->flags &= ~TUN_VNET_BE;
381 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
383 return virtio_legacy_is_little_endian();
386 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
391 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
395 #endif /* CONFIG_TUN_VNET_CROSS_LE */
397 static inline bool tun_is_little_endian(struct tun_struct *tun)
399 return tun->flags & TUN_VNET_LE ||
400 tun_legacy_is_little_endian(tun);
403 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
405 return __virtio16_to_cpu(tun_is_little_endian(tun), val);
408 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
410 return __cpu_to_virtio16(tun_is_little_endian(tun), val);
413 static inline u32 tun_hashfn(u32 rxhash)
415 return rxhash & TUN_MASK_FLOW_ENTRIES;
418 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
420 struct tun_flow_entry *e;
422 hlist_for_each_entry_rcu(e, head, hash_link) {
423 if (e->rxhash == rxhash)
429 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
430 struct hlist_head *head,
431 u32 rxhash, u16 queue_index)
433 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
436 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
437 rxhash, queue_index);
438 e->updated = jiffies;
441 e->queue_index = queue_index;
443 hlist_add_head_rcu(&e->hash_link, head);
449 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
451 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
452 e->rxhash, e->queue_index);
453 hlist_del_rcu(&e->hash_link);
458 static void tun_flow_flush(struct tun_struct *tun)
462 spin_lock_bh(&tun->lock);
463 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
464 struct tun_flow_entry *e;
465 struct hlist_node *n;
467 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
468 tun_flow_delete(tun, e);
470 spin_unlock_bh(&tun->lock);
473 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
477 spin_lock_bh(&tun->lock);
478 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
479 struct tun_flow_entry *e;
480 struct hlist_node *n;
482 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
483 if (e->queue_index == queue_index)
484 tun_flow_delete(tun, e);
487 spin_unlock_bh(&tun->lock);
490 static void tun_flow_cleanup(struct timer_list *t)
492 struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
493 unsigned long delay = tun->ageing_time;
494 unsigned long next_timer = jiffies + delay;
495 unsigned long count = 0;
498 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
500 spin_lock(&tun->lock);
501 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
502 struct tun_flow_entry *e;
503 struct hlist_node *n;
505 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
506 unsigned long this_timer;
508 this_timer = e->updated + delay;
509 if (time_before_eq(this_timer, jiffies)) {
510 tun_flow_delete(tun, e);
514 if (time_before(this_timer, next_timer))
515 next_timer = this_timer;
520 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
521 spin_unlock(&tun->lock);
524 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
525 struct tun_file *tfile)
527 struct hlist_head *head;
528 struct tun_flow_entry *e;
529 unsigned long delay = tun->ageing_time;
530 u16 queue_index = tfile->queue_index;
532 head = &tun->flows[tun_hashfn(rxhash)];
536 e = tun_flow_find(head, rxhash);
538 /* TODO: keep queueing to old queue until it's empty? */
539 if (e->queue_index != queue_index)
540 e->queue_index = queue_index;
541 if (e->updated != jiffies)
542 e->updated = jiffies;
543 sock_rps_record_flow_hash(e->rps_rxhash);
545 spin_lock_bh(&tun->lock);
546 if (!tun_flow_find(head, rxhash) &&
547 tun->flow_count < MAX_TAP_FLOWS)
548 tun_flow_create(tun, head, rxhash, queue_index);
550 if (!timer_pending(&tun->flow_gc_timer))
551 mod_timer(&tun->flow_gc_timer,
552 round_jiffies_up(jiffies + delay));
553 spin_unlock_bh(&tun->lock);
560 * Save the hash received in the stack receive path and update the
561 * flow_hash table accordingly.
563 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
565 if (unlikely(e->rps_rxhash != hash))
566 e->rps_rxhash = hash;
569 /* We try to identify a flow through its rxhash. The reason that
570 * we do not check rxq no. is because some cards(e.g 82599), chooses
571 * the rxq based on the txq where the last packet of the flow comes. As
572 * the userspace application move between processors, we may get a
573 * different rxq no. here.
575 static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
577 struct tun_flow_entry *e;
581 numqueues = READ_ONCE(tun->numqueues);
583 txq = __skb_get_hash_symmetric(skb);
584 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
586 tun_flow_save_rps_rxhash(e, txq);
587 txq = e->queue_index;
589 /* use multiply and shift instead of expensive divide */
590 txq = ((u64)txq * numqueues) >> 32;
596 static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
598 struct tun_prog *prog;
601 prog = rcu_dereference(tun->steering_prog);
603 ret = bpf_prog_run_clear_cb(prog->prog, skb);
605 return ret % tun->numqueues;
608 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
609 struct net_device *sb_dev,
610 select_queue_fallback_t fallback)
612 struct tun_struct *tun = netdev_priv(dev);
616 if (rcu_dereference(tun->steering_prog))
617 ret = tun_ebpf_select_queue(tun, skb);
619 ret = tun_automq_select_queue(tun, skb);
625 static inline bool tun_not_capable(struct tun_struct *tun)
627 const struct cred *cred = current_cred();
628 struct net *net = dev_net(tun->dev);
630 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
631 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
632 !ns_capable(net->user_ns, CAP_NET_ADMIN);
635 static void tun_set_real_num_queues(struct tun_struct *tun)
637 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
638 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
641 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
643 tfile->detached = tun;
644 list_add_tail(&tfile->next, &tun->disabled);
648 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
650 struct tun_struct *tun = tfile->detached;
652 tfile->detached = NULL;
653 list_del_init(&tfile->next);
658 void tun_ptr_free(void *ptr)
662 if (tun_is_xdp_frame(ptr)) {
663 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
665 xdp_return_frame(xdpf);
667 __skb_array_destroy_skb(ptr);
670 EXPORT_SYMBOL_GPL(tun_ptr_free);
672 static void tun_queue_purge(struct tun_file *tfile)
676 while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
679 skb_queue_purge(&tfile->sk.sk_write_queue);
680 skb_queue_purge(&tfile->sk.sk_error_queue);
683 static void __tun_detach(struct tun_file *tfile, bool clean)
685 struct tun_file *ntfile;
686 struct tun_struct *tun;
688 tun = rtnl_dereference(tfile->tun);
691 tun_napi_disable(tfile);
695 if (tun && !tfile->detached) {
696 u16 index = tfile->queue_index;
697 BUG_ON(index >= tun->numqueues);
699 rcu_assign_pointer(tun->tfiles[index],
700 tun->tfiles[tun->numqueues - 1]);
701 ntfile = rtnl_dereference(tun->tfiles[index]);
702 ntfile->queue_index = index;
706 RCU_INIT_POINTER(tfile->tun, NULL);
707 sock_put(&tfile->sk);
709 tun_disable_queue(tun, tfile);
712 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
713 /* Drop read queue */
714 tun_queue_purge(tfile);
715 tun_set_real_num_queues(tun);
716 } else if (tfile->detached && clean) {
717 tun = tun_enable_queue(tfile);
718 sock_put(&tfile->sk);
722 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
723 netif_carrier_off(tun->dev);
725 if (!(tun->flags & IFF_PERSIST) &&
726 tun->dev->reg_state == NETREG_REGISTERED)
727 unregister_netdevice(tun->dev);
730 xdp_rxq_info_unreg(&tfile->xdp_rxq);
731 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
732 sock_put(&tfile->sk);
736 static void tun_detach(struct tun_file *tfile, bool clean)
738 struct tun_struct *tun;
739 struct net_device *dev;
742 tun = rtnl_dereference(tfile->tun);
743 dev = tun ? tun->dev : NULL;
744 __tun_detach(tfile, clean);
746 netdev_state_change(dev);
750 static void tun_detach_all(struct net_device *dev)
752 struct tun_struct *tun = netdev_priv(dev);
753 struct tun_file *tfile, *tmp;
754 int i, n = tun->numqueues;
756 for (i = 0; i < n; i++) {
757 tfile = rtnl_dereference(tun->tfiles[i]);
759 tun_napi_disable(tfile);
760 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
761 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
762 RCU_INIT_POINTER(tfile->tun, NULL);
765 list_for_each_entry(tfile, &tun->disabled, next) {
766 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
767 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
768 RCU_INIT_POINTER(tfile->tun, NULL);
770 BUG_ON(tun->numqueues != 0);
773 for (i = 0; i < n; i++) {
774 tfile = rtnl_dereference(tun->tfiles[i]);
776 /* Drop read queue */
777 tun_queue_purge(tfile);
778 xdp_rxq_info_unreg(&tfile->xdp_rxq);
779 sock_put(&tfile->sk);
781 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
782 tun_enable_queue(tfile);
783 tun_queue_purge(tfile);
784 xdp_rxq_info_unreg(&tfile->xdp_rxq);
785 sock_put(&tfile->sk);
787 BUG_ON(tun->numdisabled != 0);
789 if (tun->flags & IFF_PERSIST)
790 module_put(THIS_MODULE);
793 static int tun_attach(struct tun_struct *tun, struct file *file,
794 bool skip_filter, bool napi, bool napi_frags)
796 struct tun_file *tfile = file->private_data;
797 struct net_device *dev = tun->dev;
800 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
805 if (rtnl_dereference(tfile->tun) && !tfile->detached)
809 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
813 if (!tfile->detached &&
814 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
819 /* Re-attach the filter to persist device */
820 if (!skip_filter && (tun->filter_attached == true)) {
821 lock_sock(tfile->socket.sk);
822 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
823 release_sock(tfile->socket.sk);
828 if (!tfile->detached &&
829 ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
830 GFP_KERNEL, tun_ptr_free)) {
835 tfile->queue_index = tun->numqueues;
836 tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
838 if (tfile->detached) {
839 /* Re-attach detached tfile, updating XDP queue_index */
840 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
842 if (tfile->xdp_rxq.queue_index != tfile->queue_index)
843 tfile->xdp_rxq.queue_index = tfile->queue_index;
845 /* Setup XDP RX-queue info, for new tfile getting attached */
846 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
847 tun->dev, tfile->queue_index);
850 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
851 MEM_TYPE_PAGE_SHARED, NULL);
853 xdp_rxq_info_unreg(&tfile->xdp_rxq);
859 if (tfile->detached) {
860 tun_enable_queue(tfile);
862 sock_hold(&tfile->sk);
863 tun_napi_init(tun, tfile, napi, napi_frags);
866 if (rtnl_dereference(tun->xdp_prog))
867 sock_set_flag(&tfile->sk, SOCK_XDP);
869 /* device is allowed to go away first, so no need to hold extra
873 /* Publish tfile->tun and tun->tfiles only after we've fully
874 * initialized tfile; otherwise we risk using half-initialized
877 rcu_assign_pointer(tfile->tun, tun);
878 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
880 tun_set_real_num_queues(tun);
885 static struct tun_struct *tun_get(struct tun_file *tfile)
887 struct tun_struct *tun;
890 tun = rcu_dereference(tfile->tun);
898 static void tun_put(struct tun_struct *tun)
904 static void addr_hash_set(u32 *mask, const u8 *addr)
906 int n = ether_crc(ETH_ALEN, addr) >> 26;
907 mask[n >> 5] |= (1 << (n & 31));
910 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
912 int n = ether_crc(ETH_ALEN, addr) >> 26;
913 return mask[n >> 5] & (1 << (n & 31));
916 static int update_filter(struct tap_filter *filter, void __user *arg)
918 struct { u8 u[ETH_ALEN]; } *addr;
919 struct tun_filter uf;
920 int err, alen, n, nexact;
922 if (copy_from_user(&uf, arg, sizeof(uf)))
931 alen = ETH_ALEN * uf.count;
932 addr = memdup_user(arg + sizeof(uf), alen);
934 return PTR_ERR(addr);
936 /* The filter is updated without holding any locks. Which is
937 * perfectly safe. We disable it first and in the worst
938 * case we'll accept a few undesired packets. */
942 /* Use first set of addresses as an exact filter */
943 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
944 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
948 /* Remaining multicast addresses are hashed,
949 * unicast will leave the filter disabled. */
950 memset(filter->mask, 0, sizeof(filter->mask));
951 for (; n < uf.count; n++) {
952 if (!is_multicast_ether_addr(addr[n].u)) {
953 err = 0; /* no filter */
956 addr_hash_set(filter->mask, addr[n].u);
959 /* For ALLMULTI just set the mask to all ones.
960 * This overrides the mask populated above. */
961 if ((uf.flags & TUN_FLT_ALLMULTI))
962 memset(filter->mask, ~0, sizeof(filter->mask));
964 /* Now enable the filter */
966 filter->count = nexact;
968 /* Return the number of exact filters */
975 /* Returns: 0 - drop, !=0 - accept */
976 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
978 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
980 struct ethhdr *eh = (struct ethhdr *) skb->data;
984 for (i = 0; i < filter->count; i++)
985 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
988 /* Inexact match (multicast only) */
989 if (is_multicast_ether_addr(eh->h_dest))
990 return addr_hash_test(filter->mask, eh->h_dest);
996 * Checks whether the packet is accepted or not.
997 * Returns: 0 - drop, !=0 - accept
999 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
1004 return run_filter(filter, skb);
1007 /* Network device part of the driver */
1009 static const struct ethtool_ops tun_ethtool_ops;
1011 /* Net device detach from fd. */
1012 static void tun_net_uninit(struct net_device *dev)
1014 tun_detach_all(dev);
1017 /* Net device open. */
1018 static int tun_net_open(struct net_device *dev)
1020 struct tun_struct *tun = netdev_priv(dev);
1023 netif_tx_start_all_queues(dev);
1025 for (i = 0; i < tun->numqueues; i++) {
1026 struct tun_file *tfile;
1028 tfile = rtnl_dereference(tun->tfiles[i]);
1029 tfile->socket.sk->sk_write_space(tfile->socket.sk);
1035 /* Net device close. */
1036 static int tun_net_close(struct net_device *dev)
1038 netif_tx_stop_all_queues(dev);
1042 /* Net device start xmit */
1043 static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
1046 if (tun->numqueues == 1 && static_key_false(&rps_needed)) {
1047 /* Select queue was not called for the skbuff, so we extract the
1048 * RPS hash and save it into the flow_table here.
1050 struct tun_flow_entry *e;
1053 rxhash = __skb_get_hash_symmetric(skb);
1054 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
1056 tun_flow_save_rps_rxhash(e, rxhash);
1061 static unsigned int run_ebpf_filter(struct tun_struct *tun,
1062 struct sk_buff *skb,
1065 struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1068 len = bpf_prog_run_clear_cb(prog->prog, skb);
1073 /* Net device start xmit */
1074 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1076 struct tun_struct *tun = netdev_priv(dev);
1077 int txq = skb->queue_mapping;
1078 struct tun_file *tfile;
1082 tfile = rcu_dereference(tun->tfiles[txq]);
1084 /* Drop packet if interface is not attached */
1085 if (txq >= tun->numqueues)
1088 if (!rcu_dereference(tun->steering_prog))
1089 tun_automq_xmit(tun, skb);
1091 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
1095 /* Drop if the filter does not like it.
1096 * This is a noop if the filter is disabled.
1097 * Filter can be enabled only for the TAP devices. */
1098 if (!check_filter(&tun->txflt, skb))
1101 if (tfile->socket.sk->sk_filter &&
1102 sk_filter(tfile->socket.sk, skb))
1105 len = run_ebpf_filter(tun, skb, len);
1106 if (len == 0 || pskb_trim(skb, len))
1109 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1112 skb_tx_timestamp(skb);
1114 /* Orphan the skb - required as we might hang on to it
1115 * for indefinite time.
1121 if (ptr_ring_produce(&tfile->tx_ring, skb))
1124 /* Notify and wake up reader process */
1125 if (tfile->flags & TUN_FASYNC)
1126 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1127 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1130 return NETDEV_TX_OK;
1133 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1137 return NET_XMIT_DROP;
1140 static void tun_net_mclist(struct net_device *dev)
1143 * This callback is supposed to deal with mc filter in
1144 * _rx_ path and has nothing to do with the _tx_ path.
1145 * In rx path we always accept everything userspace gives us.
1149 static netdev_features_t tun_net_fix_features(struct net_device *dev,
1150 netdev_features_t features)
1152 struct tun_struct *tun = netdev_priv(dev);
1154 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1157 static void tun_set_headroom(struct net_device *dev, int new_hr)
1159 struct tun_struct *tun = netdev_priv(dev);
1161 if (new_hr < NET_SKB_PAD)
1162 new_hr = NET_SKB_PAD;
1164 tun->align = new_hr;
1168 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1170 u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1171 struct tun_struct *tun = netdev_priv(dev);
1172 struct tun_pcpu_stats *p;
1175 for_each_possible_cpu(i) {
1176 u64 rxpackets, rxbytes, txpackets, txbytes;
1179 p = per_cpu_ptr(tun->pcpu_stats, i);
1181 start = u64_stats_fetch_begin(&p->syncp);
1182 rxpackets = p->rx_packets;
1183 rxbytes = p->rx_bytes;
1184 txpackets = p->tx_packets;
1185 txbytes = p->tx_bytes;
1186 } while (u64_stats_fetch_retry(&p->syncp, start));
1188 stats->rx_packets += rxpackets;
1189 stats->rx_bytes += rxbytes;
1190 stats->tx_packets += txpackets;
1191 stats->tx_bytes += txbytes;
1194 rx_dropped += p->rx_dropped;
1195 rx_frame_errors += p->rx_frame_errors;
1196 tx_dropped += p->tx_dropped;
1198 stats->rx_dropped = rx_dropped;
1199 stats->rx_frame_errors = rx_frame_errors;
1200 stats->tx_dropped = tx_dropped;
1203 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1204 struct netlink_ext_ack *extack)
1206 struct tun_struct *tun = netdev_priv(dev);
1207 struct tun_file *tfile;
1208 struct bpf_prog *old_prog;
1211 old_prog = rtnl_dereference(tun->xdp_prog);
1212 rcu_assign_pointer(tun->xdp_prog, prog);
1214 bpf_prog_put(old_prog);
1216 for (i = 0; i < tun->numqueues; i++) {
1217 tfile = rtnl_dereference(tun->tfiles[i]);
1219 sock_set_flag(&tfile->sk, SOCK_XDP);
1221 sock_reset_flag(&tfile->sk, SOCK_XDP);
1223 list_for_each_entry(tfile, &tun->disabled, next) {
1225 sock_set_flag(&tfile->sk, SOCK_XDP);
1227 sock_reset_flag(&tfile->sk, SOCK_XDP);
1233 static u32 tun_xdp_query(struct net_device *dev)
1235 struct tun_struct *tun = netdev_priv(dev);
1236 const struct bpf_prog *xdp_prog;
1238 xdp_prog = rtnl_dereference(tun->xdp_prog);
1240 return xdp_prog->aux->id;
1245 static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1247 switch (xdp->command) {
1248 case XDP_SETUP_PROG:
1249 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1250 case XDP_QUERY_PROG:
1251 xdp->prog_id = tun_xdp_query(dev);
1258 static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1261 struct tun_struct *tun = netdev_priv(dev);
1263 if (!tun->numqueues)
1266 netif_carrier_on(dev);
1268 netif_carrier_off(dev);
1273 static const struct net_device_ops tun_netdev_ops = {
1274 .ndo_uninit = tun_net_uninit,
1275 .ndo_open = tun_net_open,
1276 .ndo_stop = tun_net_close,
1277 .ndo_start_xmit = tun_net_xmit,
1278 .ndo_fix_features = tun_net_fix_features,
1279 .ndo_select_queue = tun_select_queue,
1280 .ndo_set_rx_headroom = tun_set_headroom,
1281 .ndo_get_stats64 = tun_net_get_stats64,
1282 .ndo_change_carrier = tun_net_change_carrier,
1285 static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1287 /* Notify and wake up reader process */
1288 if (tfile->flags & TUN_FASYNC)
1289 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1290 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1293 static int tun_xdp_xmit(struct net_device *dev, int n,
1294 struct xdp_frame **frames, u32 flags)
1296 struct tun_struct *tun = netdev_priv(dev);
1297 struct tun_file *tfile;
1303 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
1308 numqueues = READ_ONCE(tun->numqueues);
1311 return -ENXIO; /* Caller will free/return all frames */
1314 tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1317 spin_lock(&tfile->tx_ring.producer_lock);
1318 for (i = 0; i < n; i++) {
1319 struct xdp_frame *xdp = frames[i];
1320 /* Encode the XDP flag into lowest bit for consumer to differ
1321 * XDP buffer from sk_buff.
1323 void *frame = tun_xdp_to_ptr(xdp);
1325 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
1326 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1327 xdp_return_frame_rx_napi(xdp);
1331 spin_unlock(&tfile->tx_ring.producer_lock);
1333 if (flags & XDP_XMIT_FLUSH)
1334 __tun_xdp_flush_tfile(tfile);
1340 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1342 struct xdp_frame *frame = convert_to_xdp_frame(xdp);
1344 if (unlikely(!frame))
1347 return tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1350 static const struct net_device_ops tap_netdev_ops = {
1351 .ndo_uninit = tun_net_uninit,
1352 .ndo_open = tun_net_open,
1353 .ndo_stop = tun_net_close,
1354 .ndo_start_xmit = tun_net_xmit,
1355 .ndo_fix_features = tun_net_fix_features,
1356 .ndo_set_rx_mode = tun_net_mclist,
1357 .ndo_set_mac_address = eth_mac_addr,
1358 .ndo_validate_addr = eth_validate_addr,
1359 .ndo_select_queue = tun_select_queue,
1360 .ndo_features_check = passthru_features_check,
1361 .ndo_set_rx_headroom = tun_set_headroom,
1362 .ndo_get_stats64 = tun_net_get_stats64,
1364 .ndo_xdp_xmit = tun_xdp_xmit,
1365 .ndo_change_carrier = tun_net_change_carrier,
1368 static void tun_flow_init(struct tun_struct *tun)
1372 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1373 INIT_HLIST_HEAD(&tun->flows[i]);
1375 tun->ageing_time = TUN_FLOW_EXPIRE;
1376 timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1377 mod_timer(&tun->flow_gc_timer,
1378 round_jiffies_up(jiffies + tun->ageing_time));
1381 static void tun_flow_uninit(struct tun_struct *tun)
1383 del_timer_sync(&tun->flow_gc_timer);
1384 tun_flow_flush(tun);
1388 #define MAX_MTU 65535
1390 /* Initialize net device. */
1391 static void tun_net_init(struct net_device *dev)
1393 struct tun_struct *tun = netdev_priv(dev);
1395 switch (tun->flags & TUN_TYPE_MASK) {
1397 dev->netdev_ops = &tun_netdev_ops;
1399 /* Point-to-Point TUN Device */
1400 dev->hard_header_len = 0;
1404 /* Zero header length */
1405 dev->type = ARPHRD_NONE;
1406 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1410 dev->netdev_ops = &tap_netdev_ops;
1411 /* Ethernet TAP Device */
1413 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1414 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1416 eth_hw_addr_random(dev);
1421 dev->min_mtu = MIN_MTU;
1422 dev->max_mtu = MAX_MTU - dev->hard_header_len;
1425 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1427 struct sock *sk = tfile->socket.sk;
1429 return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1432 /* Character device part */
1435 static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
1437 struct tun_file *tfile = file->private_data;
1438 struct tun_struct *tun = tun_get(tfile);
1445 sk = tfile->socket.sk;
1447 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1449 poll_wait(file, sk_sleep(sk), wait);
1451 if (!ptr_ring_empty(&tfile->tx_ring))
1452 mask |= EPOLLIN | EPOLLRDNORM;
1454 /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1455 * guarantee EPOLLOUT to be raised by either here or
1456 * tun_sock_write_space(). Then process could get notification
1457 * after it writes to a down device and meets -EIO.
1459 if (tun_sock_writeable(tun, tfile) ||
1460 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1461 tun_sock_writeable(tun, tfile)))
1462 mask |= EPOLLOUT | EPOLLWRNORM;
1464 if (tun->dev->reg_state != NETREG_REGISTERED)
1471 static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1473 const struct iov_iter *it)
1475 struct sk_buff *skb;
1480 if (it->nr_segs > MAX_SKB_FRAGS + 1)
1481 return ERR_PTR(-ENOMEM);
1484 skb = napi_get_frags(&tfile->napi);
1487 return ERR_PTR(-ENOMEM);
1489 linear = iov_iter_single_seg_count(it);
1490 err = __skb_grow(skb, linear);
1495 skb->data_len = len - linear;
1496 skb->truesize += skb->data_len;
1498 for (i = 1; i < it->nr_segs; i++) {
1499 size_t fragsz = it->iov[i].iov_len;
1503 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1507 frag = netdev_alloc_frag(fragsz);
1512 page = virt_to_head_page(frag);
1513 skb_fill_page_desc(skb, i - 1, page,
1514 frag - page_address(page), fragsz);
1519 /* frees skb and all frags allocated with napi_alloc_frag() */
1520 napi_free_frags(&tfile->napi);
1521 return ERR_PTR(err);
1524 /* prepad is the amount to reserve at front. len is length after that.
1525 * linear is a hint as to how much to copy (usually headers). */
1526 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1527 size_t prepad, size_t len,
1528 size_t linear, int noblock)
1530 struct sock *sk = tfile->socket.sk;
1531 struct sk_buff *skb;
1534 /* Under a page? Don't bother with paged skb. */
1535 if (prepad + len < PAGE_SIZE || !linear)
1538 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1541 return ERR_PTR(err);
1543 skb_reserve(skb, prepad);
1544 skb_put(skb, linear);
1545 skb->data_len = len - linear;
1546 skb->len += len - linear;
1551 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1552 struct sk_buff *skb, int more)
1554 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1555 struct sk_buff_head process_queue;
1556 u32 rx_batched = tun->rx_batched;
1559 if (!rx_batched || (!more && skb_queue_empty(queue))) {
1561 skb_record_rx_queue(skb, tfile->queue_index);
1562 netif_receive_skb(skb);
1567 spin_lock(&queue->lock);
1568 if (!more || skb_queue_len(queue) == rx_batched) {
1569 __skb_queue_head_init(&process_queue);
1570 skb_queue_splice_tail_init(queue, &process_queue);
1573 __skb_queue_tail(queue, skb);
1575 spin_unlock(&queue->lock);
1578 struct sk_buff *nskb;
1581 while ((nskb = __skb_dequeue(&process_queue))) {
1582 skb_record_rx_queue(nskb, tfile->queue_index);
1583 netif_receive_skb(nskb);
1585 skb_record_rx_queue(skb, tfile->queue_index);
1586 netif_receive_skb(skb);
1591 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1592 int len, int noblock, bool zerocopy)
1594 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1597 if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1606 if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1607 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1613 static struct sk_buff *__tun_build_skb(struct page_frag *alloc_frag, char *buf,
1614 int buflen, int len, int pad)
1616 struct sk_buff *skb = build_skb(buf, buflen);
1619 return ERR_PTR(-ENOMEM);
1621 skb_reserve(skb, pad);
1624 get_page(alloc_frag->page);
1625 alloc_frag->offset += buflen;
1630 static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1631 struct xdp_buff *xdp, u32 act)
1637 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
1642 err = tun_xdp_tx(tun->dev, xdp);
1649 bpf_warn_invalid_xdp_action(act);
1652 trace_xdp_exception(tun->dev, xdp_prog, act);
1655 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1662 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1663 struct tun_file *tfile,
1664 struct iov_iter *from,
1665 struct virtio_net_hdr *hdr,
1666 int len, int *skb_xdp)
1668 struct page_frag *alloc_frag = ¤t->task_frag;
1669 struct bpf_prog *xdp_prog;
1670 int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1673 int pad = TUN_RX_PAD;
1677 xdp_prog = rcu_dereference(tun->xdp_prog);
1679 pad += XDP_PACKET_HEADROOM;
1680 buflen += SKB_DATA_ALIGN(len + pad);
1683 alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1684 if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1685 return ERR_PTR(-ENOMEM);
1687 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1688 copied = copy_page_from_iter(alloc_frag->page,
1689 alloc_frag->offset + pad,
1692 return ERR_PTR(-EFAULT);
1694 /* There's a small window that XDP may be set after the check
1695 * of xdp_prog above, this should be rare and for simplicity
1696 * we do XDP on skb in case the headroom is not enough.
1698 if (hdr->gso_type || !xdp_prog) {
1700 return __tun_build_skb(alloc_frag, buf, buflen, len, pad);
1707 xdp_prog = rcu_dereference(tun->xdp_prog);
1709 struct xdp_buff xdp;
1712 xdp.data_hard_start = buf;
1713 xdp.data = buf + pad;
1714 xdp_set_data_meta_invalid(&xdp);
1715 xdp.data_end = xdp.data + len;
1716 xdp.rxq = &tfile->xdp_rxq;
1718 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1719 if (act == XDP_REDIRECT || act == XDP_TX) {
1720 get_page(alloc_frag->page);
1721 alloc_frag->offset += buflen;
1723 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
1726 if (err == XDP_REDIRECT)
1728 if (err != XDP_PASS)
1731 pad = xdp.data - xdp.data_hard_start;
1732 len = xdp.data_end - xdp.data;
1737 return __tun_build_skb(alloc_frag, buf, buflen, len, pad);
1740 put_page(alloc_frag->page);
1747 /* Get packet from user space buffer */
1748 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1749 void *msg_control, struct iov_iter *from,
1750 int noblock, bool more)
1752 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1753 struct sk_buff *skb;
1754 size_t total_len = iov_iter_count(from);
1755 size_t len = total_len, align = tun->align, linear;
1756 struct virtio_net_hdr gso = { 0 };
1757 struct tun_pcpu_stats *stats;
1760 bool zerocopy = false;
1764 bool frags = tun_napi_frags_enabled(tfile);
1766 if (!(tun->flags & IFF_NO_PI)) {
1767 if (len < sizeof(pi))
1771 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1775 if (tun->flags & IFF_VNET_HDR) {
1776 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1778 if (len < vnet_hdr_sz)
1782 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1785 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1786 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1787 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1789 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1791 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1794 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1795 align += NET_IP_ALIGN;
1796 if (unlikely(len < ETH_HLEN ||
1797 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1801 good_linear = SKB_MAX_HEAD(align);
1804 struct iov_iter i = *from;
1806 /* There are 256 bytes to be copied in skb, so there is
1807 * enough room for skb expand head in case it is used.
1808 * The rest of the buffer is mapped from userspace.
1810 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1811 if (copylen > good_linear)
1812 copylen = good_linear;
1814 iov_iter_advance(&i, copylen);
1815 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1819 if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1820 /* For the packet that is not easy to be processed
1821 * (e.g gso or jumbo packet), we will do it at after
1822 * skb was created with generic XDP routine.
1824 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1826 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1827 return PTR_ERR(skb);
1834 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1835 linear = good_linear;
1837 linear = tun16_to_cpu(tun, gso.hdr_len);
1841 mutex_lock(&tfile->napi_mutex);
1842 skb = tun_napi_alloc_frags(tfile, copylen, from);
1843 /* tun_napi_alloc_frags() enforces a layout for the skb.
1844 * If zerocopy is enabled, then this layout will be
1845 * overwritten by zerocopy_sg_from_iter().
1849 skb = tun_alloc_skb(tfile, align, copylen, linear,
1854 if (PTR_ERR(skb) != -EAGAIN)
1855 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1857 mutex_unlock(&tfile->napi_mutex);
1858 return PTR_ERR(skb);
1862 err = zerocopy_sg_from_iter(skb, from);
1864 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1869 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1872 tfile->napi.skb = NULL;
1873 mutex_unlock(&tfile->napi_mutex);
1880 if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1881 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1884 tfile->napi.skb = NULL;
1885 mutex_unlock(&tfile->napi_mutex);
1891 switch (tun->flags & TUN_TYPE_MASK) {
1893 if (tun->flags & IFF_NO_PI) {
1894 u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1896 switch (ip_version) {
1898 pi.proto = htons(ETH_P_IP);
1901 pi.proto = htons(ETH_P_IPV6);
1904 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1910 skb_reset_mac_header(skb);
1911 skb->protocol = pi.proto;
1912 skb->dev = tun->dev;
1916 skb->protocol = eth_type_trans(skb, tun->dev);
1920 /* copy skb_ubuf_info for callback when skb has no error */
1922 skb_shinfo(skb)->destructor_arg = msg_control;
1923 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1924 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1925 } else if (msg_control) {
1926 struct ubuf_info *uarg = msg_control;
1927 uarg->callback(uarg, false);
1930 skb_reset_network_header(skb);
1931 skb_probe_transport_header(skb);
1934 struct bpf_prog *xdp_prog;
1939 xdp_prog = rcu_dereference(tun->xdp_prog);
1941 ret = do_xdp_generic(xdp_prog, skb);
1942 if (ret != XDP_PASS) {
1952 /* Compute the costly rx hash only if needed for flow updates.
1953 * We may get a very small possibility of OOO during switching, not
1954 * worth to optimize.
1956 if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1958 rxhash = __skb_get_hash_symmetric(skb);
1961 if (unlikely(!(tun->dev->flags & IFF_UP))) {
1968 /* Exercise flow dissector code path. */
1969 u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1971 if (unlikely(headlen > skb_headlen(skb))) {
1972 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1973 napi_free_frags(&tfile->napi);
1975 mutex_unlock(&tfile->napi_mutex);
1981 napi_gro_frags(&tfile->napi);
1983 mutex_unlock(&tfile->napi_mutex);
1984 } else if (tfile->napi_enabled) {
1985 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1988 spin_lock_bh(&queue->lock);
1989 __skb_queue_tail(queue, skb);
1990 queue_len = skb_queue_len(queue);
1991 spin_unlock(&queue->lock);
1993 if (!more || queue_len > NAPI_POLL_WEIGHT)
1994 napi_schedule(&tfile->napi);
1997 } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1998 tun_rx_batched(tun, tfile, skb, more);
2004 stats = get_cpu_ptr(tun->pcpu_stats);
2005 u64_stats_update_begin(&stats->syncp);
2006 stats->rx_packets++;
2007 stats->rx_bytes += len;
2008 u64_stats_update_end(&stats->syncp);
2012 tun_flow_update(tun, rxhash, tfile);
2017 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
2019 struct file *file = iocb->ki_filp;
2020 struct tun_file *tfile = file->private_data;
2021 struct tun_struct *tun = tun_get(tfile);
2027 result = tun_get_user(tun, tfile, NULL, from,
2028 file->f_flags & O_NONBLOCK, false);
2034 static ssize_t tun_put_user_xdp(struct tun_struct *tun,
2035 struct tun_file *tfile,
2036 struct xdp_frame *xdp_frame,
2037 struct iov_iter *iter)
2039 int vnet_hdr_sz = 0;
2040 size_t size = xdp_frame->len;
2041 struct tun_pcpu_stats *stats;
2044 if (tun->flags & IFF_VNET_HDR) {
2045 struct virtio_net_hdr gso = { 0 };
2047 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2048 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2050 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2053 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2056 ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
2058 stats = get_cpu_ptr(tun->pcpu_stats);
2059 u64_stats_update_begin(&stats->syncp);
2060 stats->tx_packets++;
2061 stats->tx_bytes += ret;
2062 u64_stats_update_end(&stats->syncp);
2063 put_cpu_ptr(tun->pcpu_stats);
2068 /* Put packet to the user space buffer */
2069 static ssize_t tun_put_user(struct tun_struct *tun,
2070 struct tun_file *tfile,
2071 struct sk_buff *skb,
2072 struct iov_iter *iter)
2074 struct tun_pi pi = { 0, skb->protocol };
2075 struct tun_pcpu_stats *stats;
2077 int vlan_offset = 0;
2079 int vnet_hdr_sz = 0;
2081 if (skb_vlan_tag_present(skb))
2082 vlan_hlen = VLAN_HLEN;
2084 if (tun->flags & IFF_VNET_HDR)
2085 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2087 total = skb->len + vlan_hlen + vnet_hdr_sz;
2089 if (!(tun->flags & IFF_NO_PI)) {
2090 if (iov_iter_count(iter) < sizeof(pi))
2093 total += sizeof(pi);
2094 if (iov_iter_count(iter) < total) {
2095 /* Packet will be striped */
2096 pi.flags |= TUN_PKT_STRIP;
2099 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
2104 struct virtio_net_hdr gso;
2106 if (iov_iter_count(iter) < vnet_hdr_sz)
2109 if (virtio_net_hdr_from_skb(skb, &gso,
2110 tun_is_little_endian(tun), true,
2112 struct skb_shared_info *sinfo = skb_shinfo(skb);
2113 pr_err("unexpected GSO type: "
2114 "0x%x, gso_size %d, hdr_len %d\n",
2115 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2116 tun16_to_cpu(tun, gso.hdr_len));
2117 print_hex_dump(KERN_ERR, "tun: ",
2120 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2125 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
2128 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2135 veth.h_vlan_proto = skb->vlan_proto;
2136 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
2138 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
2140 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2141 if (ret || !iov_iter_count(iter))
2144 ret = copy_to_iter(&veth, sizeof(veth), iter);
2145 if (ret != sizeof(veth) || !iov_iter_count(iter))
2149 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
2152 /* caller is in process context, */
2153 stats = get_cpu_ptr(tun->pcpu_stats);
2154 u64_stats_update_begin(&stats->syncp);
2155 stats->tx_packets++;
2156 stats->tx_bytes += skb->len + vlan_hlen;
2157 u64_stats_update_end(&stats->syncp);
2158 put_cpu_ptr(tun->pcpu_stats);
2163 static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
2165 DECLARE_WAITQUEUE(wait, current);
2169 ptr = ptr_ring_consume(&tfile->tx_ring);
2177 add_wait_queue(&tfile->wq.wait, &wait);
2180 set_current_state(TASK_INTERRUPTIBLE);
2181 ptr = ptr_ring_consume(&tfile->tx_ring);
2184 if (signal_pending(current)) {
2185 error = -ERESTARTSYS;
2188 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
2196 __set_current_state(TASK_RUNNING);
2197 remove_wait_queue(&tfile->wq.wait, &wait);
2204 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
2205 struct iov_iter *to,
2206 int noblock, void *ptr)
2211 tun_debug(KERN_INFO, tun, "tun_do_read\n");
2213 if (!iov_iter_count(to)) {
2219 /* Read frames from ring */
2220 ptr = tun_ring_recv(tfile, noblock, &err);
2225 if (tun_is_xdp_frame(ptr)) {
2226 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2228 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
2229 xdp_return_frame(xdpf);
2231 struct sk_buff *skb = ptr;
2233 ret = tun_put_user(tun, tfile, skb, to);
2234 if (unlikely(ret < 0))
2243 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
2245 struct file *file = iocb->ki_filp;
2246 struct tun_file *tfile = file->private_data;
2247 struct tun_struct *tun = tun_get(tfile);
2248 ssize_t len = iov_iter_count(to), ret;
2252 ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
2253 ret = min_t(ssize_t, ret, len);
2260 static void tun_prog_free(struct rcu_head *rcu)
2262 struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
2264 bpf_prog_destroy(prog->prog);
2268 static int __tun_set_ebpf(struct tun_struct *tun,
2269 struct tun_prog __rcu **prog_p,
2270 struct bpf_prog *prog)
2272 struct tun_prog *old, *new = NULL;
2275 new = kmalloc(sizeof(*new), GFP_KERNEL);
2281 spin_lock_bh(&tun->lock);
2282 old = rcu_dereference_protected(*prog_p,
2283 lockdep_is_held(&tun->lock));
2284 rcu_assign_pointer(*prog_p, new);
2285 spin_unlock_bh(&tun->lock);
2288 call_rcu(&old->rcu, tun_prog_free);
2293 static void tun_free_netdev(struct net_device *dev)
2295 struct tun_struct *tun = netdev_priv(dev);
2297 BUG_ON(!(list_empty(&tun->disabled)));
2298 free_percpu(tun->pcpu_stats);
2299 tun_flow_uninit(tun);
2300 security_tun_dev_free_security(tun->security);
2301 __tun_set_ebpf(tun, &tun->steering_prog, NULL);
2302 __tun_set_ebpf(tun, &tun->filter_prog, NULL);
2305 static void tun_setup(struct net_device *dev)
2307 struct tun_struct *tun = netdev_priv(dev);
2309 tun->owner = INVALID_UID;
2310 tun->group = INVALID_GID;
2311 tun_default_link_ksettings(dev, &tun->link_ksettings);
2313 dev->ethtool_ops = &tun_ethtool_ops;
2314 dev->needs_free_netdev = true;
2315 dev->priv_destructor = tun_free_netdev;
2316 /* We prefer our own queue length */
2317 dev->tx_queue_len = TUN_READQ_SIZE;
2320 /* Trivial set of netlink ops to allow deleting tun or tap
2321 * device with netlink.
2323 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2324 struct netlink_ext_ack *extack)
2326 NL_SET_ERR_MSG(extack,
2327 "tun/tap creation via rtnetlink is not supported.");
2331 static size_t tun_get_size(const struct net_device *dev)
2333 BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2334 BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2336 return nla_total_size(sizeof(uid_t)) + /* OWNER */
2337 nla_total_size(sizeof(gid_t)) + /* GROUP */
2338 nla_total_size(sizeof(u8)) + /* TYPE */
2339 nla_total_size(sizeof(u8)) + /* PI */
2340 nla_total_size(sizeof(u8)) + /* VNET_HDR */
2341 nla_total_size(sizeof(u8)) + /* PERSIST */
2342 nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2343 nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2344 nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2348 static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2350 struct tun_struct *tun = netdev_priv(dev);
2352 if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2353 goto nla_put_failure;
2354 if (uid_valid(tun->owner) &&
2355 nla_put_u32(skb, IFLA_TUN_OWNER,
2356 from_kuid_munged(current_user_ns(), tun->owner)))
2357 goto nla_put_failure;
2358 if (gid_valid(tun->group) &&
2359 nla_put_u32(skb, IFLA_TUN_GROUP,
2360 from_kgid_munged(current_user_ns(), tun->group)))
2361 goto nla_put_failure;
2362 if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2363 goto nla_put_failure;
2364 if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2365 goto nla_put_failure;
2366 if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2367 goto nla_put_failure;
2368 if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2369 !!(tun->flags & IFF_MULTI_QUEUE)))
2370 goto nla_put_failure;
2371 if (tun->flags & IFF_MULTI_QUEUE) {
2372 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2373 goto nla_put_failure;
2374 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2376 goto nla_put_failure;
2385 static struct rtnl_link_ops tun_link_ops __read_mostly = {
2387 .priv_size = sizeof(struct tun_struct),
2389 .validate = tun_validate,
2390 .get_size = tun_get_size,
2391 .fill_info = tun_fill_info,
2394 static void tun_sock_write_space(struct sock *sk)
2396 struct tun_file *tfile;
2397 wait_queue_head_t *wqueue;
2399 if (!sock_writeable(sk))
2402 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
2405 wqueue = sk_sleep(sk);
2406 if (wqueue && waitqueue_active(wqueue))
2407 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2408 EPOLLWRNORM | EPOLLWRBAND);
2410 tfile = container_of(sk, struct tun_file, sk);
2411 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
2414 static void tun_put_page(struct tun_page *tpage)
2417 __page_frag_cache_drain(tpage->page, tpage->count);
2420 static int tun_xdp_one(struct tun_struct *tun,
2421 struct tun_file *tfile,
2422 struct xdp_buff *xdp, int *flush,
2423 struct tun_page *tpage)
2425 unsigned int datasize = xdp->data_end - xdp->data;
2426 struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2427 struct virtio_net_hdr *gso = &hdr->gso;
2428 struct tun_pcpu_stats *stats;
2429 struct bpf_prog *xdp_prog;
2430 struct sk_buff *skb = NULL;
2431 u32 rxhash = 0, act;
2432 int buflen = hdr->buflen;
2434 bool skb_xdp = false;
2437 xdp_prog = rcu_dereference(tun->xdp_prog);
2439 if (gso->gso_type) {
2443 xdp_set_data_meta_invalid(xdp);
2444 xdp->rxq = &tfile->xdp_rxq;
2446 act = bpf_prog_run_xdp(xdp_prog, xdp);
2447 err = tun_xdp_act(tun, xdp_prog, xdp, act);
2449 put_page(virt_to_head_page(xdp->data));
2462 page = virt_to_head_page(xdp->data);
2463 if (tpage->page == page) {
2466 tun_put_page(tpage);
2475 skb = build_skb(xdp->data_hard_start, buflen);
2481 skb_reserve(skb, xdp->data - xdp->data_hard_start);
2482 skb_put(skb, xdp->data_end - xdp->data);
2484 if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
2485 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
2491 skb->protocol = eth_type_trans(skb, tun->dev);
2492 skb_reset_network_header(skb);
2493 skb_probe_transport_header(skb);
2496 err = do_xdp_generic(xdp_prog, skb);
2497 if (err != XDP_PASS)
2501 if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2503 rxhash = __skb_get_hash_symmetric(skb);
2505 skb_record_rx_queue(skb, tfile->queue_index);
2506 netif_receive_skb(skb);
2508 /* No need for get_cpu_ptr() here since this function is
2509 * always called with bh disabled
2511 stats = this_cpu_ptr(tun->pcpu_stats);
2512 u64_stats_update_begin(&stats->syncp);
2513 stats->rx_packets++;
2514 stats->rx_bytes += datasize;
2515 u64_stats_update_end(&stats->syncp);
2518 tun_flow_update(tun, rxhash, tfile);
2524 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
2527 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2528 struct tun_struct *tun = tun_get(tfile);
2529 struct tun_msg_ctl *ctl = m->msg_control;
2530 struct xdp_buff *xdp;
2535 if (ctl && (ctl->type == TUN_MSG_PTR)) {
2536 struct tun_page tpage;
2540 memset(&tpage, 0, sizeof(tpage));
2545 for (i = 0; i < n; i++) {
2546 xdp = &((struct xdp_buff *)ctl->ptr)[i];
2547 tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
2556 tun_put_page(&tpage);
2562 ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
2563 m->msg_flags & MSG_DONTWAIT,
2564 m->msg_flags & MSG_MORE);
2570 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
2573 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2574 struct tun_struct *tun = tun_get(tfile);
2575 void *ptr = m->msg_control;
2583 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
2587 if (flags & MSG_ERRQUEUE) {
2588 ret = sock_recv_errqueue(sock->sk, m, total_len,
2589 SOL_PACKET, TUN_TX_TIMESTAMP);
2592 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
2593 if (ret > (ssize_t)total_len) {
2594 m->msg_flags |= MSG_TRUNC;
2595 ret = flags & MSG_TRUNC ? ret : total_len;
2608 static int tun_ptr_peek_len(void *ptr)
2611 if (tun_is_xdp_frame(ptr)) {
2612 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
2616 return __skb_array_len_with_tag(ptr);
2622 static int tun_peek_len(struct socket *sock)
2624 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2625 struct tun_struct *tun;
2628 tun = tun_get(tfile);
2632 ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
2638 /* Ops structure to mimic raw sockets with tun */
2639 static const struct proto_ops tun_socket_ops = {
2640 .peek_len = tun_peek_len,
2641 .sendmsg = tun_sendmsg,
2642 .recvmsg = tun_recvmsg,
2645 static struct proto tun_proto = {
2647 .owner = THIS_MODULE,
2648 .obj_size = sizeof(struct tun_file),
2651 static int tun_flags(struct tun_struct *tun)
2653 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
2656 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2659 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2660 return sprintf(buf, "0x%x\n", tun_flags(tun));
2663 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2666 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2667 return uid_valid(tun->owner)?
2668 sprintf(buf, "%u\n",
2669 from_kuid_munged(current_user_ns(), tun->owner)):
2670 sprintf(buf, "-1\n");
2673 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2676 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2677 return gid_valid(tun->group) ?
2678 sprintf(buf, "%u\n",
2679 from_kgid_munged(current_user_ns(), tun->group)):
2680 sprintf(buf, "-1\n");
2683 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2684 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2685 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2687 static struct attribute *tun_dev_attrs[] = {
2688 &dev_attr_tun_flags.attr,
2689 &dev_attr_owner.attr,
2690 &dev_attr_group.attr,
2694 static const struct attribute_group tun_attr_group = {
2695 .attrs = tun_dev_attrs
2698 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2700 struct tun_struct *tun;
2701 struct tun_file *tfile = file->private_data;
2702 struct net_device *dev;
2705 if (tfile->detached)
2708 if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2709 if (!capable(CAP_NET_ADMIN))
2712 if (!(ifr->ifr_flags & IFF_NAPI) ||
2713 (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2717 dev = __dev_get_by_name(net, ifr->ifr_name);
2719 if (ifr->ifr_flags & IFF_TUN_EXCL)
2721 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2722 tun = netdev_priv(dev);
2723 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2724 tun = netdev_priv(dev);
2728 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2729 !!(tun->flags & IFF_MULTI_QUEUE))
2732 if (tun_not_capable(tun))
2734 err = security_tun_dev_open(tun->security);
2738 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2739 ifr->ifr_flags & IFF_NAPI,
2740 ifr->ifr_flags & IFF_NAPI_FRAGS);
2744 if (tun->flags & IFF_MULTI_QUEUE &&
2745 (tun->numqueues + tun->numdisabled > 1)) {
2746 /* One or more queue has already been attached, no need
2747 * to initialize the device again.
2749 netdev_state_change(dev);
2753 tun->flags = (tun->flags & ~TUN_FEATURES) |
2754 (ifr->ifr_flags & TUN_FEATURES);
2756 netdev_state_change(dev);
2759 unsigned long flags = 0;
2760 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2763 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2765 err = security_tun_dev_create();
2770 if (ifr->ifr_flags & IFF_TUN) {
2774 } else if (ifr->ifr_flags & IFF_TAP) {
2782 name = ifr->ifr_name;
2784 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2785 NET_NAME_UNKNOWN, tun_setup, queues,
2790 err = dev_get_valid_name(net, dev, name);
2794 dev_net_set(dev, net);
2795 dev->rtnl_link_ops = &tun_link_ops;
2796 dev->ifindex = tfile->ifindex;
2797 dev->sysfs_groups[0] = &tun_attr_group;
2799 tun = netdev_priv(dev);
2802 tun->txflt.count = 0;
2803 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2805 tun->align = NET_SKB_PAD;
2806 tun->filter_attached = false;
2807 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2808 tun->rx_batched = 0;
2809 RCU_INIT_POINTER(tun->steering_prog, NULL);
2811 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2812 if (!tun->pcpu_stats) {
2817 spin_lock_init(&tun->lock);
2819 err = security_tun_dev_alloc_security(&tun->security);
2826 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2827 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2828 NETIF_F_HW_VLAN_STAG_TX;
2829 dev->features = dev->hw_features | NETIF_F_LLTX;
2830 dev->vlan_features = dev->features &
2831 ~(NETIF_F_HW_VLAN_CTAG_TX |
2832 NETIF_F_HW_VLAN_STAG_TX);
2834 tun->flags = (tun->flags & ~TUN_FEATURES) |
2835 (ifr->ifr_flags & TUN_FEATURES);
2837 INIT_LIST_HEAD(&tun->disabled);
2838 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI,
2839 ifr->ifr_flags & IFF_NAPI_FRAGS);
2843 err = register_netdevice(tun->dev);
2848 netif_carrier_on(tun->dev);
2850 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2852 /* Make sure persistent devices do not get stuck in
2855 if (netif_running(tun->dev))
2856 netif_tx_wake_all_queues(tun->dev);
2858 strcpy(ifr->ifr_name, tun->dev->name);
2862 tun_detach_all(dev);
2863 /* register_netdevice() already called tun_free_netdev() */
2867 tun_flow_uninit(tun);
2868 security_tun_dev_free_security(tun->security);
2870 free_percpu(tun->pcpu_stats);
2876 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2879 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2881 strcpy(ifr->ifr_name, tun->dev->name);
2883 ifr->ifr_flags = tun_flags(tun);
2887 /* This is like a cut-down ethtool ops, except done via tun fd so no
2888 * privs required. */
2889 static int set_offload(struct tun_struct *tun, unsigned long arg)
2891 netdev_features_t features = 0;
2893 if (arg & TUN_F_CSUM) {
2894 features |= NETIF_F_HW_CSUM;
2897 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2898 if (arg & TUN_F_TSO_ECN) {
2899 features |= NETIF_F_TSO_ECN;
2900 arg &= ~TUN_F_TSO_ECN;
2902 if (arg & TUN_F_TSO4)
2903 features |= NETIF_F_TSO;
2904 if (arg & TUN_F_TSO6)
2905 features |= NETIF_F_TSO6;
2906 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2912 /* This gives the user a way to test for new features in future by
2913 * trying to set them. */
2917 tun->set_features = features;
2918 tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2919 tun->dev->wanted_features |= features;
2920 netdev_update_features(tun->dev);
2925 static void tun_detach_filter(struct tun_struct *tun, int n)
2928 struct tun_file *tfile;
2930 for (i = 0; i < n; i++) {
2931 tfile = rtnl_dereference(tun->tfiles[i]);
2932 lock_sock(tfile->socket.sk);
2933 sk_detach_filter(tfile->socket.sk);
2934 release_sock(tfile->socket.sk);
2937 tun->filter_attached = false;
2940 static int tun_attach_filter(struct tun_struct *tun)
2943 struct tun_file *tfile;
2945 for (i = 0; i < tun->numqueues; i++) {
2946 tfile = rtnl_dereference(tun->tfiles[i]);
2947 lock_sock(tfile->socket.sk);
2948 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2949 release_sock(tfile->socket.sk);
2951 tun_detach_filter(tun, i);
2956 tun->filter_attached = true;
2960 static void tun_set_sndbuf(struct tun_struct *tun)
2962 struct tun_file *tfile;
2965 for (i = 0; i < tun->numqueues; i++) {
2966 tfile = rtnl_dereference(tun->tfiles[i]);
2967 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2971 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2973 struct tun_file *tfile = file->private_data;
2974 struct tun_struct *tun;
2979 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2980 tun = tfile->detached;
2985 ret = security_tun_dev_attach_queue(tun->security);
2988 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
2989 tun->flags & IFF_NAPI_FRAGS);
2990 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2991 tun = rtnl_dereference(tfile->tun);
2992 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2995 __tun_detach(tfile, false);
3000 netdev_state_change(tun->dev);
3007 static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
3010 struct bpf_prog *prog;
3013 if (copy_from_user(&fd, data, sizeof(fd)))
3019 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
3021 return PTR_ERR(prog);
3024 return __tun_set_ebpf(tun, prog_p, prog);
3027 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
3028 unsigned long arg, int ifreq_len)
3030 struct tun_file *tfile = file->private_data;
3031 struct net *net = sock_net(&tfile->sk);
3032 struct tun_struct *tun;
3033 void __user* argp = (void __user*)arg;
3034 unsigned int ifindex, carrier;
3042 bool do_notify = false;
3044 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
3045 (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
3046 if (copy_from_user(&ifr, argp, ifreq_len))
3049 memset(&ifr, 0, sizeof(ifr));
3051 if (cmd == TUNGETFEATURES) {
3052 /* Currently this just means: "what IFF flags are valid?".
3053 * This is needed because we never checked for invalid flags on
3056 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
3057 (unsigned int __user*)argp);
3058 } else if (cmd == TUNSETQUEUE) {
3059 return tun_set_queue(file, &ifr);
3060 } else if (cmd == SIOCGSKNS) {
3061 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3063 return open_related_ns(&net->ns, get_net_ns);
3069 tun = tun_get(tfile);
3070 if (cmd == TUNSETIFF) {
3075 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3077 ret = tun_set_iff(net, file, &ifr);
3082 if (copy_to_user(argp, &ifr, ifreq_len))
3086 if (cmd == TUNSETIFINDEX) {
3092 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3096 tfile->ifindex = ifindex;
3104 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
3109 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
3111 if (tfile->detached)
3112 ifr.ifr_flags |= IFF_DETACH_QUEUE;
3113 if (!tfile->socket.sk->sk_filter)
3114 ifr.ifr_flags |= IFF_NOFILTER;
3116 if (copy_to_user(argp, &ifr, ifreq_len))
3121 /* Disable/Enable checksum */
3123 /* [unimplemented] */
3124 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
3125 arg ? "disabled" : "enabled");
3129 /* Disable/Enable persist mode. Keep an extra reference to the
3130 * module to prevent the module being unprobed.
3132 if (arg && !(tun->flags & IFF_PERSIST)) {
3133 tun->flags |= IFF_PERSIST;
3134 __module_get(THIS_MODULE);
3137 if (!arg && (tun->flags & IFF_PERSIST)) {
3138 tun->flags &= ~IFF_PERSIST;
3139 module_put(THIS_MODULE);
3143 tun_debug(KERN_INFO, tun, "persist %s\n",
3144 arg ? "enabled" : "disabled");
3148 /* Set owner of the device */
3149 owner = make_kuid(current_user_ns(), arg);
3150 if (!uid_valid(owner)) {
3156 tun_debug(KERN_INFO, tun, "owner set to %u\n",
3157 from_kuid(&init_user_ns, tun->owner));
3161 /* Set group of the device */
3162 group = make_kgid(current_user_ns(), arg);
3163 if (!gid_valid(group)) {
3169 tun_debug(KERN_INFO, tun, "group set to %u\n",
3170 from_kgid(&init_user_ns, tun->group));
3174 /* Only allow setting the type when the interface is down */
3175 if (tun->dev->flags & IFF_UP) {
3176 tun_debug(KERN_INFO, tun,
3177 "Linktype set failed because interface is up\n");
3180 tun->dev->type = (int) arg;
3181 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
3193 ret = set_offload(tun, arg);
3196 case TUNSETTXFILTER:
3197 /* Can be set only for TAPs */
3199 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3201 ret = update_filter(&tun->txflt, (void __user *)arg);
3205 /* Get hw address */
3206 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
3207 ifr.ifr_hwaddr.sa_family = tun->dev->type;
3208 if (copy_to_user(argp, &ifr, ifreq_len))
3213 /* Set hw address */
3214 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
3215 ifr.ifr_hwaddr.sa_data);
3217 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr, NULL);
3221 sndbuf = tfile->socket.sk->sk_sndbuf;
3222 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3227 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3236 tun->sndbuf = sndbuf;
3237 tun_set_sndbuf(tun);
3240 case TUNGETVNETHDRSZ:
3241 vnet_hdr_sz = tun->vnet_hdr_sz;
3242 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3246 case TUNSETVNETHDRSZ:
3247 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3251 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3256 tun->vnet_hdr_sz = vnet_hdr_sz;
3260 le = !!(tun->flags & TUN_VNET_LE);
3261 if (put_user(le, (int __user *)argp))
3266 if (get_user(le, (int __user *)argp)) {
3271 tun->flags |= TUN_VNET_LE;
3273 tun->flags &= ~TUN_VNET_LE;
3277 ret = tun_get_vnet_be(tun, argp);
3281 ret = tun_set_vnet_be(tun, argp);
3284 case TUNATTACHFILTER:
3285 /* Can be set only for TAPs */
3287 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3290 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
3293 ret = tun_attach_filter(tun);
3296 case TUNDETACHFILTER:
3297 /* Can be set only for TAPs */
3299 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3302 tun_detach_filter(tun, tun->numqueues);
3307 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
3310 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3315 case TUNSETSTEERINGEBPF:
3316 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
3319 case TUNSETFILTEREBPF:
3320 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3325 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3328 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3337 netdev_state_change(tun->dev);
3346 static long tun_chr_ioctl(struct file *file,
3347 unsigned int cmd, unsigned long arg)
3349 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3352 #ifdef CONFIG_COMPAT
3353 static long tun_chr_compat_ioctl(struct file *file,
3354 unsigned int cmd, unsigned long arg)
3359 case TUNSETTXFILTER:
3364 arg = (unsigned long)compat_ptr(arg);
3367 arg = (compat_ulong_t)arg;
3372 * compat_ifreq is shorter than ifreq, so we must not access beyond
3373 * the end of that structure. All fields that are used in this
3374 * driver are compatible though, we don't need to convert the
3377 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3379 #endif /* CONFIG_COMPAT */
3381 static int tun_chr_fasync(int fd, struct file *file, int on)
3383 struct tun_file *tfile = file->private_data;
3386 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
3390 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
3391 tfile->flags |= TUN_FASYNC;
3393 tfile->flags &= ~TUN_FASYNC;
3399 static int tun_chr_open(struct inode *inode, struct file * file)
3401 struct net *net = current->nsproxy->net_ns;
3402 struct tun_file *tfile;
3404 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
3406 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
3410 if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3411 sk_free(&tfile->sk);
3415 mutex_init(&tfile->napi_mutex);
3416 RCU_INIT_POINTER(tfile->tun, NULL);
3420 init_waitqueue_head(&tfile->wq.wait);
3421 RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
3423 tfile->socket.file = file;
3424 tfile->socket.ops = &tun_socket_ops;
3426 sock_init_data(&tfile->socket, &tfile->sk);
3428 tfile->sk.sk_write_space = tun_sock_write_space;
3429 tfile->sk.sk_sndbuf = INT_MAX;
3431 file->private_data = tfile;
3432 INIT_LIST_HEAD(&tfile->next);
3434 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3439 static int tun_chr_close(struct inode *inode, struct file *file)
3441 struct tun_file *tfile = file->private_data;
3443 tun_detach(tfile, true);
3448 #ifdef CONFIG_PROC_FS
3449 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
3451 struct tun_file *tfile = file->private_data;
3452 struct tun_struct *tun;
3455 memset(&ifr, 0, sizeof(ifr));
3458 tun = tun_get(tfile);
3460 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
3466 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
3470 static const struct file_operations tun_fops = {
3471 .owner = THIS_MODULE,
3472 .llseek = no_llseek,
3473 .read_iter = tun_chr_read_iter,
3474 .write_iter = tun_chr_write_iter,
3475 .poll = tun_chr_poll,
3476 .unlocked_ioctl = tun_chr_ioctl,
3477 #ifdef CONFIG_COMPAT
3478 .compat_ioctl = tun_chr_compat_ioctl,
3480 .open = tun_chr_open,
3481 .release = tun_chr_close,
3482 .fasync = tun_chr_fasync,
3483 #ifdef CONFIG_PROC_FS
3484 .show_fdinfo = tun_chr_show_fdinfo,
3488 static struct miscdevice tun_miscdev = {
3491 .nodename = "net/tun",
3495 /* ethtool interface */
3497 static void tun_default_link_ksettings(struct net_device *dev,
3498 struct ethtool_link_ksettings *cmd)
3500 ethtool_link_ksettings_zero_link_mode(cmd, supported);
3501 ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3502 cmd->base.speed = SPEED_10;
3503 cmd->base.duplex = DUPLEX_FULL;
3504 cmd->base.port = PORT_TP;
3505 cmd->base.phy_address = 0;
3506 cmd->base.autoneg = AUTONEG_DISABLE;
3509 static int tun_get_link_ksettings(struct net_device *dev,
3510 struct ethtool_link_ksettings *cmd)
3512 struct tun_struct *tun = netdev_priv(dev);
3514 memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3518 static int tun_set_link_ksettings(struct net_device *dev,
3519 const struct ethtool_link_ksettings *cmd)
3521 struct tun_struct *tun = netdev_priv(dev);
3523 memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
3527 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3529 struct tun_struct *tun = netdev_priv(dev);
3531 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3532 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
3534 switch (tun->flags & TUN_TYPE_MASK) {
3536 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
3539 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
3544 static u32 tun_get_msglevel(struct net_device *dev)
3547 struct tun_struct *tun = netdev_priv(dev);
3554 static void tun_set_msglevel(struct net_device *dev, u32 value)
3557 struct tun_struct *tun = netdev_priv(dev);
3562 static int tun_get_coalesce(struct net_device *dev,
3563 struct ethtool_coalesce *ec)
3565 struct tun_struct *tun = netdev_priv(dev);
3567 ec->rx_max_coalesced_frames = tun->rx_batched;
3572 static int tun_set_coalesce(struct net_device *dev,
3573 struct ethtool_coalesce *ec)
3575 struct tun_struct *tun = netdev_priv(dev);
3577 if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3578 tun->rx_batched = NAPI_POLL_WEIGHT;
3580 tun->rx_batched = ec->rx_max_coalesced_frames;
3585 static const struct ethtool_ops tun_ethtool_ops = {
3586 .get_drvinfo = tun_get_drvinfo,
3587 .get_msglevel = tun_get_msglevel,
3588 .set_msglevel = tun_set_msglevel,
3589 .get_link = ethtool_op_get_link,
3590 .get_ts_info = ethtool_op_get_ts_info,
3591 .get_coalesce = tun_get_coalesce,
3592 .set_coalesce = tun_set_coalesce,
3593 .get_link_ksettings = tun_get_link_ksettings,
3594 .set_link_ksettings = tun_set_link_ksettings,
3597 static int tun_queue_resize(struct tun_struct *tun)
3599 struct net_device *dev = tun->dev;
3600 struct tun_file *tfile;
3601 struct ptr_ring **rings;
3602 int n = tun->numqueues + tun->numdisabled;
3605 rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3609 for (i = 0; i < tun->numqueues; i++) {
3610 tfile = rtnl_dereference(tun->tfiles[i]);
3611 rings[i] = &tfile->tx_ring;
3613 list_for_each_entry(tfile, &tun->disabled, next)
3614 rings[i++] = &tfile->tx_ring;
3616 ret = ptr_ring_resize_multiple(rings, n,
3617 dev->tx_queue_len, GFP_KERNEL,
3624 static int tun_device_event(struct notifier_block *unused,
3625 unsigned long event, void *ptr)
3627 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3628 struct tun_struct *tun = netdev_priv(dev);
3630 if (dev->rtnl_link_ops != &tun_link_ops)
3634 case NETDEV_CHANGE_TX_QUEUE_LEN:
3635 if (tun_queue_resize(tun))
3645 static struct notifier_block tun_notifier_block __read_mostly = {
3646 .notifier_call = tun_device_event,
3649 static int __init tun_init(void)
3653 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
3655 ret = rtnl_link_register(&tun_link_ops);
3657 pr_err("Can't register link_ops\n");
3661 ret = misc_register(&tun_miscdev);
3663 pr_err("Can't register misc device %d\n", TUN_MINOR);
3667 ret = register_netdevice_notifier(&tun_notifier_block);
3669 pr_err("Can't register netdevice notifier\n");
3676 misc_deregister(&tun_miscdev);
3678 rtnl_link_unregister(&tun_link_ops);
3683 static void tun_cleanup(void)
3685 misc_deregister(&tun_miscdev);
3686 rtnl_link_unregister(&tun_link_ops);
3687 unregister_netdevice_notifier(&tun_notifier_block);
3690 /* Get an underlying socket object from tun file. Returns error unless file is
3691 * attached to a device. The returned object works like a packet socket, it
3692 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
3693 * holding a reference to the file for as long as the socket is in use. */
3694 struct socket *tun_get_socket(struct file *file)
3696 struct tun_file *tfile;
3697 if (file->f_op != &tun_fops)
3698 return ERR_PTR(-EINVAL);
3699 tfile = file->private_data;
3701 return ERR_PTR(-EBADFD);
3702 return &tfile->socket;
3704 EXPORT_SYMBOL_GPL(tun_get_socket);
3706 struct ptr_ring *tun_get_tx_ring(struct file *file)
3708 struct tun_file *tfile;
3710 if (file->f_op != &tun_fops)
3711 return ERR_PTR(-EINVAL);
3712 tfile = file->private_data;
3714 return ERR_PTR(-EBADFD);
3715 return &tfile->tx_ring;
3717 EXPORT_SYMBOL_GPL(tun_get_tx_ring);
3719 module_init(tun_init);
3720 module_exit(tun_cleanup);
3721 MODULE_DESCRIPTION(DRV_DESCRIPTION);
3722 MODULE_AUTHOR(DRV_COPYRIGHT);
3723 MODULE_LICENSE("GPL");
3724 MODULE_ALIAS_MISCDEV(TUN_MINOR);
3725 MODULE_ALIAS("devname:net/tun");