2 * TUN - Universal TUN/TAP device driver.
3 * Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
21 * Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
22 * Add TUNSETLINK ioctl to set the link encapsulation
24 * Mark Smith <markzzzsmith@yahoo.com.au>
25 * Use eth_random_addr() for tap MAC address.
27 * Harald Roelle <harald.roelle@ifi.lmu.de> 2004/04/20
28 * Fixes in packet dropping, queue length setting and queue wakeup.
29 * Increased default tx queue length.
33 * Daniel Podlejski <underley@underley.eu.org>
34 * Modifications for 2.3.99-pre5 kernel.
37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
39 #define DRV_NAME "tun"
40 #define DRV_VERSION "1.6"
41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
42 #define DRV_COPYRIGHT "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
44 #include <linux/module.h>
45 #include <linux/errno.h>
46 #include <linux/kernel.h>
47 #include <linux/sched/signal.h>
48 #include <linux/major.h>
49 #include <linux/slab.h>
50 #include <linux/poll.h>
51 #include <linux/fcntl.h>
52 #include <linux/init.h>
53 #include <linux/skbuff.h>
54 #include <linux/netdevice.h>
55 #include <linux/etherdevice.h>
56 #include <linux/miscdevice.h>
57 #include <linux/ethtool.h>
58 #include <linux/rtnetlink.h>
59 #include <linux/compat.h>
61 #include <linux/if_arp.h>
62 #include <linux/if_ether.h>
63 #include <linux/if_tun.h>
64 #include <linux/if_vlan.h>
65 #include <linux/crc32.h>
66 #include <linux/nsproxy.h>
67 #include <linux/virtio_net.h>
68 #include <linux/rcupdate.h>
69 #include <net/net_namespace.h>
70 #include <net/netns/generic.h>
71 #include <net/rtnetlink.h>
73 #include <linux/seq_file.h>
74 #include <linux/uio.h>
75 #include <linux/skb_array.h>
76 #include <linux/bpf.h>
77 #include <linux/bpf_trace.h>
79 #include <linux/uaccess.h>
81 /* Uncomment to enable debugging */
82 /* #define TUN_DEBUG 1 */
87 #define tun_debug(level, tun, fmt, args...) \
90 netdev_printk(level, tun->dev, fmt, ##args); \
92 #define DBG1(level, fmt, args...) \
95 printk(level fmt, ##args); \
98 #define tun_debug(level, tun, fmt, args...) \
101 netdev_printk(level, tun->dev, fmt, ##args); \
103 #define DBG1(level, fmt, args...) \
106 printk(level fmt, ##args); \
110 #define TUN_HEADROOM 256
111 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
113 /* TUN device flags */
115 /* IFF_ATTACH_QUEUE is never stored in device flags,
116 * overload it to mean fasync when stored there.
118 #define TUN_FASYNC IFF_ATTACH_QUEUE
119 /* High bits in flags field are unused. */
120 #define TUN_VNET_LE 0x80000000
121 #define TUN_VNET_BE 0x40000000
123 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
125 #define GOODCOPY_LEN 128
127 #define FLT_EXACT_COUNT 8
129 unsigned int count; /* Number of addrs. Zero means disabled */
130 u32 mask[2]; /* Mask of the hashed addrs */
131 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
134 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
135 * to max number of VCPUs in guest. */
136 #define MAX_TAP_QUEUES 256
137 #define MAX_TAP_FLOWS 4096
139 #define TUN_FLOW_EXPIRE (3 * HZ)
141 struct tun_pcpu_stats {
146 struct u64_stats_sync syncp;
152 /* A tun_file connects an open character device to a tuntap netdevice. It
153 * also contains all socket related structures (except sock_fprog and tap_filter)
154 * to serve as one transmit queue for tuntap device. The sock_fprog and
155 * tap_filter were kept in tun_struct since they were used for filtering for the
156 * netdevice not for a specific queue (at least I didn't see the requirement for
160 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
161 * other can only be read while rcu_read_lock or rtnl_lock is held.
165 struct socket socket;
167 struct tun_struct __rcu *tun;
168 struct fasync_struct *fasync;
169 /* only used for fasnyc */
173 unsigned int ifindex;
175 struct list_head next;
176 struct tun_struct *detached;
177 struct skb_array tx_array;
180 struct tun_flow_entry {
181 struct hlist_node hash_link;
183 struct tun_struct *tun;
188 unsigned long updated;
191 #define TUN_NUM_FLOW_ENTRIES 1024
193 /* Since the socket were moved to tun_file, to preserve the behavior of persist
194 * device, socket filter, sndbuf and vnet header size were restore when the
195 * file were attached to a persist device.
198 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
199 unsigned int numqueues;
204 struct net_device *dev;
205 netdev_features_t set_features;
206 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
212 struct tap_filter txflt;
213 struct sock_fprog fprog;
214 /* protected by rtnl lock */
215 bool filter_attached;
220 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
221 struct timer_list flow_gc_timer;
222 unsigned long ageing_time;
223 unsigned int numdisabled;
224 struct list_head disabled;
228 struct tun_pcpu_stats __percpu *pcpu_stats;
229 struct bpf_prog __rcu *xdp_prog;
232 #ifdef CONFIG_TUN_VNET_CROSS_LE
233 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
235 return tun->flags & TUN_VNET_BE ? false :
236 virtio_legacy_is_little_endian();
239 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
241 int be = !!(tun->flags & TUN_VNET_BE);
243 if (put_user(be, argp))
249 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
253 if (get_user(be, argp))
257 tun->flags |= TUN_VNET_BE;
259 tun->flags &= ~TUN_VNET_BE;
264 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
266 return virtio_legacy_is_little_endian();
269 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
274 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
278 #endif /* CONFIG_TUN_VNET_CROSS_LE */
280 static inline bool tun_is_little_endian(struct tun_struct *tun)
282 return tun->flags & TUN_VNET_LE ||
283 tun_legacy_is_little_endian(tun);
286 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
288 return __virtio16_to_cpu(tun_is_little_endian(tun), val);
291 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
293 return __cpu_to_virtio16(tun_is_little_endian(tun), val);
296 static inline u32 tun_hashfn(u32 rxhash)
298 return rxhash & 0x3ff;
301 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
303 struct tun_flow_entry *e;
305 hlist_for_each_entry_rcu(e, head, hash_link) {
306 if (e->rxhash == rxhash)
312 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
313 struct hlist_head *head,
314 u32 rxhash, u16 queue_index)
316 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
319 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
320 rxhash, queue_index);
321 e->updated = jiffies;
324 e->queue_index = queue_index;
326 hlist_add_head_rcu(&e->hash_link, head);
332 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
334 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
335 e->rxhash, e->queue_index);
336 hlist_del_rcu(&e->hash_link);
341 static void tun_flow_flush(struct tun_struct *tun)
345 spin_lock_bh(&tun->lock);
346 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
347 struct tun_flow_entry *e;
348 struct hlist_node *n;
350 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
351 tun_flow_delete(tun, e);
353 spin_unlock_bh(&tun->lock);
356 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
360 spin_lock_bh(&tun->lock);
361 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
362 struct tun_flow_entry *e;
363 struct hlist_node *n;
365 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
366 if (e->queue_index == queue_index)
367 tun_flow_delete(tun, e);
370 spin_unlock_bh(&tun->lock);
373 static void tun_flow_cleanup(unsigned long data)
375 struct tun_struct *tun = (struct tun_struct *)data;
376 unsigned long delay = tun->ageing_time;
377 unsigned long next_timer = jiffies + delay;
378 unsigned long count = 0;
381 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
383 spin_lock_bh(&tun->lock);
384 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
385 struct tun_flow_entry *e;
386 struct hlist_node *n;
388 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
389 unsigned long this_timer;
391 this_timer = e->updated + delay;
392 if (time_before_eq(this_timer, jiffies))
393 tun_flow_delete(tun, e);
394 else if (time_before(this_timer, next_timer))
395 next_timer = this_timer;
400 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
401 spin_unlock_bh(&tun->lock);
404 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
405 struct tun_file *tfile)
407 struct hlist_head *head;
408 struct tun_flow_entry *e;
409 unsigned long delay = tun->ageing_time;
410 u16 queue_index = tfile->queue_index;
415 head = &tun->flows[tun_hashfn(rxhash)];
419 /* We may get a very small possibility of OOO during switching, not
420 * worth to optimize.*/
421 if (tun->numqueues == 1 || tfile->detached)
424 e = tun_flow_find(head, rxhash);
426 /* TODO: keep queueing to old queue until it's empty? */
427 e->queue_index = queue_index;
428 e->updated = jiffies;
429 sock_rps_record_flow_hash(e->rps_rxhash);
431 spin_lock_bh(&tun->lock);
432 if (!tun_flow_find(head, rxhash) &&
433 tun->flow_count < MAX_TAP_FLOWS)
434 tun_flow_create(tun, head, rxhash, queue_index);
436 if (!timer_pending(&tun->flow_gc_timer))
437 mod_timer(&tun->flow_gc_timer,
438 round_jiffies_up(jiffies + delay));
439 spin_unlock_bh(&tun->lock);
447 * Save the hash received in the stack receive path and update the
448 * flow_hash table accordingly.
450 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
452 if (unlikely(e->rps_rxhash != hash))
453 e->rps_rxhash = hash;
456 /* We try to identify a flow through its rxhash first. The reason that
457 * we do not check rxq no. is because some cards(e.g 82599), chooses
458 * the rxq based on the txq where the last packet of the flow comes. As
459 * the userspace application move between processors, we may get a
460 * different rxq no. here. If we could not get rxhash, then we would
461 * hope the rxq no. may help here.
463 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
464 void *accel_priv, select_queue_fallback_t fallback)
466 struct tun_struct *tun = netdev_priv(dev);
467 struct tun_flow_entry *e;
472 numqueues = ACCESS_ONCE(tun->numqueues);
474 txq = __skb_get_hash_symmetric(skb);
476 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
478 tun_flow_save_rps_rxhash(e, txq);
479 txq = e->queue_index;
481 /* use multiply and shift instead of expensive divide */
482 txq = ((u64)txq * numqueues) >> 32;
483 } else if (likely(skb_rx_queue_recorded(skb))) {
484 txq = skb_get_rx_queue(skb);
485 while (unlikely(txq >= numqueues))
493 static inline bool tun_not_capable(struct tun_struct *tun)
495 const struct cred *cred = current_cred();
496 struct net *net = dev_net(tun->dev);
498 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
499 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
500 !ns_capable(net->user_ns, CAP_NET_ADMIN);
503 static void tun_set_real_num_queues(struct tun_struct *tun)
505 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
506 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
509 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
511 tfile->detached = tun;
512 list_add_tail(&tfile->next, &tun->disabled);
516 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
518 struct tun_struct *tun = tfile->detached;
520 tfile->detached = NULL;
521 list_del_init(&tfile->next);
526 static void tun_queue_purge(struct tun_file *tfile)
530 while ((skb = skb_array_consume(&tfile->tx_array)) != NULL)
533 skb_queue_purge(&tfile->sk.sk_write_queue);
534 skb_queue_purge(&tfile->sk.sk_error_queue);
537 static void __tun_detach(struct tun_file *tfile, bool clean)
539 struct tun_file *ntfile;
540 struct tun_struct *tun;
542 tun = rtnl_dereference(tfile->tun);
544 if (tun && !tfile->detached) {
545 u16 index = tfile->queue_index;
546 BUG_ON(index >= tun->numqueues);
548 rcu_assign_pointer(tun->tfiles[index],
549 tun->tfiles[tun->numqueues - 1]);
550 ntfile = rtnl_dereference(tun->tfiles[index]);
551 ntfile->queue_index = index;
555 RCU_INIT_POINTER(tfile->tun, NULL);
556 sock_put(&tfile->sk);
558 tun_disable_queue(tun, tfile);
561 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
562 /* Drop read queue */
563 tun_queue_purge(tfile);
564 tun_set_real_num_queues(tun);
565 } else if (tfile->detached && clean) {
566 tun = tun_enable_queue(tfile);
567 sock_put(&tfile->sk);
571 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
572 netif_carrier_off(tun->dev);
574 if (!(tun->flags & IFF_PERSIST) &&
575 tun->dev->reg_state == NETREG_REGISTERED)
576 unregister_netdevice(tun->dev);
579 skb_array_cleanup(&tfile->tx_array);
580 sock_put(&tfile->sk);
584 static void tun_detach(struct tun_file *tfile, bool clean)
587 __tun_detach(tfile, clean);
591 static void tun_detach_all(struct net_device *dev)
593 struct tun_struct *tun = netdev_priv(dev);
594 struct bpf_prog *xdp_prog = rtnl_dereference(tun->xdp_prog);
595 struct tun_file *tfile, *tmp;
596 int i, n = tun->numqueues;
598 for (i = 0; i < n; i++) {
599 tfile = rtnl_dereference(tun->tfiles[i]);
601 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
602 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
603 RCU_INIT_POINTER(tfile->tun, NULL);
606 list_for_each_entry(tfile, &tun->disabled, next) {
607 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
608 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
609 RCU_INIT_POINTER(tfile->tun, NULL);
611 BUG_ON(tun->numqueues != 0);
614 for (i = 0; i < n; i++) {
615 tfile = rtnl_dereference(tun->tfiles[i]);
616 /* Drop read queue */
617 tun_queue_purge(tfile);
618 sock_put(&tfile->sk);
620 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
621 tun_enable_queue(tfile);
622 tun_queue_purge(tfile);
623 sock_put(&tfile->sk);
625 BUG_ON(tun->numdisabled != 0);
628 bpf_prog_put(xdp_prog);
630 if (tun->flags & IFF_PERSIST)
631 module_put(THIS_MODULE);
634 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
636 struct tun_file *tfile = file->private_data;
637 struct net_device *dev = tun->dev;
640 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
645 if (rtnl_dereference(tfile->tun) && !tfile->detached)
649 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
653 if (!tfile->detached &&
654 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
659 /* Re-attach the filter to persist device */
660 if (!skip_filter && (tun->filter_attached == true)) {
661 lock_sock(tfile->socket.sk);
662 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
663 release_sock(tfile->socket.sk);
668 if (!tfile->detached &&
669 skb_array_init(&tfile->tx_array, dev->tx_queue_len, GFP_KERNEL)) {
674 tfile->queue_index = tun->numqueues;
675 tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
676 rcu_assign_pointer(tfile->tun, tun);
677 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
681 tun_enable_queue(tfile);
683 sock_hold(&tfile->sk);
685 tun_set_real_num_queues(tun);
687 /* device is allowed to go away first, so no need to hold extra
695 static struct tun_struct *__tun_get(struct tun_file *tfile)
697 struct tun_struct *tun;
700 tun = rcu_dereference(tfile->tun);
708 static struct tun_struct *tun_get(struct file *file)
710 return __tun_get(file->private_data);
713 static void tun_put(struct tun_struct *tun)
719 static void addr_hash_set(u32 *mask, const u8 *addr)
721 int n = ether_crc(ETH_ALEN, addr) >> 26;
722 mask[n >> 5] |= (1 << (n & 31));
725 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
727 int n = ether_crc(ETH_ALEN, addr) >> 26;
728 return mask[n >> 5] & (1 << (n & 31));
731 static int update_filter(struct tap_filter *filter, void __user *arg)
733 struct { u8 u[ETH_ALEN]; } *addr;
734 struct tun_filter uf;
735 int err, alen, n, nexact;
737 if (copy_from_user(&uf, arg, sizeof(uf)))
746 alen = ETH_ALEN * uf.count;
747 addr = memdup_user(arg + sizeof(uf), alen);
749 return PTR_ERR(addr);
751 /* The filter is updated without holding any locks. Which is
752 * perfectly safe. We disable it first and in the worst
753 * case we'll accept a few undesired packets. */
757 /* Use first set of addresses as an exact filter */
758 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
759 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
763 /* Remaining multicast addresses are hashed,
764 * unicast will leave the filter disabled. */
765 memset(filter->mask, 0, sizeof(filter->mask));
766 for (; n < uf.count; n++) {
767 if (!is_multicast_ether_addr(addr[n].u)) {
768 err = 0; /* no filter */
771 addr_hash_set(filter->mask, addr[n].u);
774 /* For ALLMULTI just set the mask to all ones.
775 * This overrides the mask populated above. */
776 if ((uf.flags & TUN_FLT_ALLMULTI))
777 memset(filter->mask, ~0, sizeof(filter->mask));
779 /* Now enable the filter */
781 filter->count = nexact;
783 /* Return the number of exact filters */
790 /* Returns: 0 - drop, !=0 - accept */
791 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
793 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
795 struct ethhdr *eh = (struct ethhdr *) skb->data;
799 for (i = 0; i < filter->count; i++)
800 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
803 /* Inexact match (multicast only) */
804 if (is_multicast_ether_addr(eh->h_dest))
805 return addr_hash_test(filter->mask, eh->h_dest);
811 * Checks whether the packet is accepted or not.
812 * Returns: 0 - drop, !=0 - accept
814 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
819 return run_filter(filter, skb);
822 /* Network device part of the driver */
824 static const struct ethtool_ops tun_ethtool_ops;
826 /* Net device detach from fd. */
827 static void tun_net_uninit(struct net_device *dev)
832 /* Net device open. */
833 static int tun_net_open(struct net_device *dev)
835 struct tun_struct *tun = netdev_priv(dev);
838 netif_tx_start_all_queues(dev);
840 for (i = 0; i < tun->numqueues; i++) {
841 struct tun_file *tfile;
843 tfile = rtnl_dereference(tun->tfiles[i]);
844 tfile->socket.sk->sk_write_space(tfile->socket.sk);
850 /* Net device close. */
851 static int tun_net_close(struct net_device *dev)
853 netif_tx_stop_all_queues(dev);
857 /* Net device start xmit */
858 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
860 struct tun_struct *tun = netdev_priv(dev);
861 int txq = skb->queue_mapping;
862 struct tun_file *tfile;
866 tfile = rcu_dereference(tun->tfiles[txq]);
867 numqueues = ACCESS_ONCE(tun->numqueues);
869 /* Drop packet if interface is not attached */
870 if (txq >= numqueues)
874 if (numqueues == 1 && static_key_false(&rps_needed)) {
875 /* Select queue was not called for the skbuff, so we extract the
876 * RPS hash and save it into the flow_table here.
880 rxhash = __skb_get_hash_symmetric(skb);
882 struct tun_flow_entry *e;
883 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
886 tun_flow_save_rps_rxhash(e, rxhash);
891 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
895 /* Drop if the filter does not like it.
896 * This is a noop if the filter is disabled.
897 * Filter can be enabled only for the TAP devices. */
898 if (!check_filter(&tun->txflt, skb))
901 if (tfile->socket.sk->sk_filter &&
902 sk_filter(tfile->socket.sk, skb))
905 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
908 skb_tx_timestamp(skb);
910 /* Orphan the skb - required as we might hang on to it
911 * for indefinite time.
917 if (skb_array_produce(&tfile->tx_array, skb))
920 /* Notify and wake up reader process */
921 if (tfile->flags & TUN_FASYNC)
922 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
923 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
929 this_cpu_inc(tun->pcpu_stats->tx_dropped);
933 return NET_XMIT_DROP;
936 static void tun_net_mclist(struct net_device *dev)
939 * This callback is supposed to deal with mc filter in
940 * _rx_ path and has nothing to do with the _tx_ path.
941 * In rx path we always accept everything userspace gives us.
945 static netdev_features_t tun_net_fix_features(struct net_device *dev,
946 netdev_features_t features)
948 struct tun_struct *tun = netdev_priv(dev);
950 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
952 #ifdef CONFIG_NET_POLL_CONTROLLER
953 static void tun_poll_controller(struct net_device *dev)
956 * Tun only receives frames when:
957 * 1) the char device endpoint gets data from user space
958 * 2) the tun socket gets a sendmsg call from user space
959 * Since both of those are synchronous operations, we are guaranteed
960 * never to have pending data when we poll for it
961 * so there is nothing to do here but return.
962 * We need this though so netpoll recognizes us as an interface that
963 * supports polling, which enables bridge devices in virt setups to
964 * still use netconsole
970 static void tun_set_headroom(struct net_device *dev, int new_hr)
972 struct tun_struct *tun = netdev_priv(dev);
974 if (new_hr < NET_SKB_PAD)
975 new_hr = NET_SKB_PAD;
981 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
983 u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
984 struct tun_struct *tun = netdev_priv(dev);
985 struct tun_pcpu_stats *p;
988 for_each_possible_cpu(i) {
989 u64 rxpackets, rxbytes, txpackets, txbytes;
992 p = per_cpu_ptr(tun->pcpu_stats, i);
994 start = u64_stats_fetch_begin(&p->syncp);
995 rxpackets = p->rx_packets;
996 rxbytes = p->rx_bytes;
997 txpackets = p->tx_packets;
998 txbytes = p->tx_bytes;
999 } while (u64_stats_fetch_retry(&p->syncp, start));
1001 stats->rx_packets += rxpackets;
1002 stats->rx_bytes += rxbytes;
1003 stats->tx_packets += txpackets;
1004 stats->tx_bytes += txbytes;
1007 rx_dropped += p->rx_dropped;
1008 rx_frame_errors += p->rx_frame_errors;
1009 tx_dropped += p->tx_dropped;
1011 stats->rx_dropped = rx_dropped;
1012 stats->rx_frame_errors = rx_frame_errors;
1013 stats->tx_dropped = tx_dropped;
1016 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1017 struct netlink_ext_ack *extack)
1019 struct tun_struct *tun = netdev_priv(dev);
1020 struct bpf_prog *old_prog;
1022 old_prog = rtnl_dereference(tun->xdp_prog);
1023 rcu_assign_pointer(tun->xdp_prog, prog);
1025 bpf_prog_put(old_prog);
1030 static u32 tun_xdp_query(struct net_device *dev)
1032 struct tun_struct *tun = netdev_priv(dev);
1033 const struct bpf_prog *xdp_prog;
1035 xdp_prog = rtnl_dereference(tun->xdp_prog);
1037 return xdp_prog->aux->id;
1042 static int tun_xdp(struct net_device *dev, struct netdev_xdp *xdp)
1044 switch (xdp->command) {
1045 case XDP_SETUP_PROG:
1046 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1047 case XDP_QUERY_PROG:
1048 xdp->prog_id = tun_xdp_query(dev);
1049 xdp->prog_attached = !!xdp->prog_id;
1056 static const struct net_device_ops tun_netdev_ops = {
1057 .ndo_uninit = tun_net_uninit,
1058 .ndo_open = tun_net_open,
1059 .ndo_stop = tun_net_close,
1060 .ndo_start_xmit = tun_net_xmit,
1061 .ndo_fix_features = tun_net_fix_features,
1062 .ndo_select_queue = tun_select_queue,
1063 #ifdef CONFIG_NET_POLL_CONTROLLER
1064 .ndo_poll_controller = tun_poll_controller,
1066 .ndo_set_rx_headroom = tun_set_headroom,
1067 .ndo_get_stats64 = tun_net_get_stats64,
1070 static const struct net_device_ops tap_netdev_ops = {
1071 .ndo_uninit = tun_net_uninit,
1072 .ndo_open = tun_net_open,
1073 .ndo_stop = tun_net_close,
1074 .ndo_start_xmit = tun_net_xmit,
1075 .ndo_fix_features = tun_net_fix_features,
1076 .ndo_set_rx_mode = tun_net_mclist,
1077 .ndo_set_mac_address = eth_mac_addr,
1078 .ndo_validate_addr = eth_validate_addr,
1079 .ndo_select_queue = tun_select_queue,
1080 #ifdef CONFIG_NET_POLL_CONTROLLER
1081 .ndo_poll_controller = tun_poll_controller,
1083 .ndo_features_check = passthru_features_check,
1084 .ndo_set_rx_headroom = tun_set_headroom,
1085 .ndo_get_stats64 = tun_net_get_stats64,
1089 static void tun_flow_init(struct tun_struct *tun)
1093 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1094 INIT_HLIST_HEAD(&tun->flows[i]);
1096 tun->ageing_time = TUN_FLOW_EXPIRE;
1097 setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
1098 mod_timer(&tun->flow_gc_timer,
1099 round_jiffies_up(jiffies + tun->ageing_time));
1102 static void tun_flow_uninit(struct tun_struct *tun)
1104 del_timer_sync(&tun->flow_gc_timer);
1105 tun_flow_flush(tun);
1109 #define MAX_MTU 65535
1111 /* Initialize net device. */
1112 static void tun_net_init(struct net_device *dev)
1114 struct tun_struct *tun = netdev_priv(dev);
1116 switch (tun->flags & TUN_TYPE_MASK) {
1118 dev->netdev_ops = &tun_netdev_ops;
1120 /* Point-to-Point TUN Device */
1121 dev->hard_header_len = 0;
1125 /* Zero header length */
1126 dev->type = ARPHRD_NONE;
1127 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1131 dev->netdev_ops = &tap_netdev_ops;
1132 /* Ethernet TAP Device */
1134 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1135 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1137 eth_hw_addr_random(dev);
1142 dev->min_mtu = MIN_MTU;
1143 dev->max_mtu = MAX_MTU - dev->hard_header_len;
1146 /* Character device part */
1149 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1151 struct tun_file *tfile = file->private_data;
1152 struct tun_struct *tun = __tun_get(tfile);
1154 unsigned int mask = 0;
1159 sk = tfile->socket.sk;
1161 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1163 poll_wait(file, sk_sleep(sk), wait);
1165 if (!skb_array_empty(&tfile->tx_array))
1166 mask |= POLLIN | POLLRDNORM;
1168 if (tun->dev->flags & IFF_UP &&
1169 (sock_writeable(sk) ||
1170 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1171 sock_writeable(sk))))
1172 mask |= POLLOUT | POLLWRNORM;
1174 if (tun->dev->reg_state != NETREG_REGISTERED)
1181 /* prepad is the amount to reserve at front. len is length after that.
1182 * linear is a hint as to how much to copy (usually headers). */
1183 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1184 size_t prepad, size_t len,
1185 size_t linear, int noblock)
1187 struct sock *sk = tfile->socket.sk;
1188 struct sk_buff *skb;
1191 /* Under a page? Don't bother with paged skb. */
1192 if (prepad + len < PAGE_SIZE || !linear)
1195 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1198 return ERR_PTR(err);
1200 skb_reserve(skb, prepad);
1201 skb_put(skb, linear);
1202 skb->data_len = len - linear;
1203 skb->len += len - linear;
1208 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1209 struct sk_buff *skb, int more)
1211 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1212 struct sk_buff_head process_queue;
1213 u32 rx_batched = tun->rx_batched;
1216 if (!rx_batched || (!more && skb_queue_empty(queue))) {
1218 netif_receive_skb(skb);
1223 spin_lock(&queue->lock);
1224 if (!more || skb_queue_len(queue) == rx_batched) {
1225 __skb_queue_head_init(&process_queue);
1226 skb_queue_splice_tail_init(queue, &process_queue);
1229 __skb_queue_tail(queue, skb);
1231 spin_unlock(&queue->lock);
1234 struct sk_buff *nskb;
1237 while ((nskb = __skb_dequeue(&process_queue)))
1238 netif_receive_skb(nskb);
1239 netif_receive_skb(skb);
1244 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1245 int len, int noblock, bool zerocopy)
1247 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1250 if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1259 if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1260 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1266 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1267 struct tun_file *tfile,
1268 struct iov_iter *from,
1269 struct virtio_net_hdr *hdr,
1270 int len, int *skb_xdp)
1272 struct page_frag *alloc_frag = ¤t->task_frag;
1273 struct sk_buff *skb;
1274 struct bpf_prog *xdp_prog;
1275 int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1276 unsigned int delta = 0;
1279 bool xdp_xmit = false;
1280 int err, pad = TUN_RX_PAD;
1283 xdp_prog = rcu_dereference(tun->xdp_prog);
1285 pad += TUN_HEADROOM;
1286 buflen += SKB_DATA_ALIGN(len + pad);
1289 if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1290 return ERR_PTR(-ENOMEM);
1292 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1293 copied = copy_page_from_iter(alloc_frag->page,
1294 alloc_frag->offset + pad,
1297 return ERR_PTR(-EFAULT);
1299 /* There's a small window that XDP may be set after the check
1300 * of xdp_prog above, this should be rare and for simplicity
1301 * we do XDP on skb in case the headroom is not enough.
1303 if (hdr->gso_type || !xdp_prog)
1309 xdp_prog = rcu_dereference(tun->xdp_prog);
1310 if (xdp_prog && !*skb_xdp) {
1311 struct xdp_buff xdp;
1315 xdp.data_hard_start = buf;
1316 xdp.data = buf + pad;
1317 xdp.data_end = xdp.data + len;
1318 orig_data = xdp.data;
1319 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1323 get_page(alloc_frag->page);
1324 alloc_frag->offset += buflen;
1325 err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1333 delta = orig_data - xdp.data;
1336 bpf_warn_invalid_xdp_action(act);
1339 trace_xdp_exception(tun->dev, xdp_prog, act);
1346 skb = build_skb(buf, buflen);
1349 return ERR_PTR(-ENOMEM);
1352 skb_reserve(skb, pad - delta);
1353 skb_put(skb, len + delta);
1354 get_page(alloc_frag->page);
1355 alloc_frag->offset += buflen;
1358 skb->dev = tun->dev;
1359 generic_xdp_tx(skb, xdp_prog);
1369 put_page(alloc_frag->page);
1372 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1376 /* Get packet from user space buffer */
1377 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1378 void *msg_control, struct iov_iter *from,
1379 int noblock, bool more)
1381 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1382 struct sk_buff *skb;
1383 size_t total_len = iov_iter_count(from);
1384 size_t len = total_len, align = tun->align, linear;
1385 struct virtio_net_hdr gso = { 0 };
1386 struct tun_pcpu_stats *stats;
1389 bool zerocopy = false;
1394 if (!(tun->dev->flags & IFF_UP))
1397 if (!(tun->flags & IFF_NO_PI)) {
1398 if (len < sizeof(pi))
1402 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1406 if (tun->flags & IFF_VNET_HDR) {
1407 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1409 if (len < vnet_hdr_sz)
1413 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1416 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1417 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1418 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1420 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1422 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1425 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1426 align += NET_IP_ALIGN;
1427 if (unlikely(len < ETH_HLEN ||
1428 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1432 good_linear = SKB_MAX_HEAD(align);
1435 struct iov_iter i = *from;
1437 /* There are 256 bytes to be copied in skb, so there is
1438 * enough room for skb expand head in case it is used.
1439 * The rest of the buffer is mapped from userspace.
1441 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1442 if (copylen > good_linear)
1443 copylen = good_linear;
1445 iov_iter_advance(&i, copylen);
1446 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1450 if (tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1451 /* For the packet that is not easy to be processed
1452 * (e.g gso or jumbo packet), we will do it at after
1453 * skb was created with generic XDP routine.
1455 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1457 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1458 return PTR_ERR(skb);
1465 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1466 linear = good_linear;
1468 linear = tun16_to_cpu(tun, gso.hdr_len);
1471 skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1473 if (PTR_ERR(skb) != -EAGAIN)
1474 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1475 return PTR_ERR(skb);
1479 err = zerocopy_sg_from_iter(skb, from);
1481 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1484 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1490 if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1491 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1496 switch (tun->flags & TUN_TYPE_MASK) {
1498 if (tun->flags & IFF_NO_PI) {
1499 u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1501 switch (ip_version) {
1503 pi.proto = htons(ETH_P_IP);
1506 pi.proto = htons(ETH_P_IPV6);
1509 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1515 skb_reset_mac_header(skb);
1516 skb->protocol = pi.proto;
1517 skb->dev = tun->dev;
1520 skb->protocol = eth_type_trans(skb, tun->dev);
1524 /* copy skb_ubuf_info for callback when skb has no error */
1526 skb_shinfo(skb)->destructor_arg = msg_control;
1527 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1528 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1529 } else if (msg_control) {
1530 struct ubuf_info *uarg = msg_control;
1531 uarg->callback(uarg, false);
1534 skb_reset_network_header(skb);
1535 skb_probe_transport_header(skb, 0);
1538 struct bpf_prog *xdp_prog;
1542 xdp_prog = rcu_dereference(tun->xdp_prog);
1544 ret = do_xdp_generic(xdp_prog, skb);
1545 if (ret != XDP_PASS) {
1553 rxhash = __skb_get_hash_symmetric(skb);
1554 #ifndef CONFIG_4KSTACKS
1555 tun_rx_batched(tun, tfile, skb, more);
1560 stats = get_cpu_ptr(tun->pcpu_stats);
1561 u64_stats_update_begin(&stats->syncp);
1562 stats->rx_packets++;
1563 stats->rx_bytes += len;
1564 u64_stats_update_end(&stats->syncp);
1567 tun_flow_update(tun, rxhash, tfile);
1571 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1573 struct file *file = iocb->ki_filp;
1574 struct tun_struct *tun = tun_get(file);
1575 struct tun_file *tfile = file->private_data;
1581 result = tun_get_user(tun, tfile, NULL, from,
1582 file->f_flags & O_NONBLOCK, false);
1588 /* Put packet to the user space buffer */
1589 static ssize_t tun_put_user(struct tun_struct *tun,
1590 struct tun_file *tfile,
1591 struct sk_buff *skb,
1592 struct iov_iter *iter)
1594 struct tun_pi pi = { 0, skb->protocol };
1595 struct tun_pcpu_stats *stats;
1597 int vlan_offset = 0;
1599 int vnet_hdr_sz = 0;
1601 if (skb_vlan_tag_present(skb))
1602 vlan_hlen = VLAN_HLEN;
1604 if (tun->flags & IFF_VNET_HDR)
1605 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1607 total = skb->len + vlan_hlen + vnet_hdr_sz;
1609 if (!(tun->flags & IFF_NO_PI)) {
1610 if (iov_iter_count(iter) < sizeof(pi))
1613 total += sizeof(pi);
1614 if (iov_iter_count(iter) < total) {
1615 /* Packet will be striped */
1616 pi.flags |= TUN_PKT_STRIP;
1619 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1624 struct virtio_net_hdr gso;
1626 if (iov_iter_count(iter) < vnet_hdr_sz)
1629 if (virtio_net_hdr_from_skb(skb, &gso,
1630 tun_is_little_endian(tun), true)) {
1631 struct skb_shared_info *sinfo = skb_shinfo(skb);
1632 pr_err("unexpected GSO type: "
1633 "0x%x, gso_size %d, hdr_len %d\n",
1634 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1635 tun16_to_cpu(tun, gso.hdr_len));
1636 print_hex_dump(KERN_ERR, "tun: ",
1639 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1644 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1647 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1653 __be16 h_vlan_proto;
1657 veth.h_vlan_proto = skb->vlan_proto;
1658 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1660 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1662 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1663 if (ret || !iov_iter_count(iter))
1666 ret = copy_to_iter(&veth, sizeof(veth), iter);
1667 if (ret != sizeof(veth) || !iov_iter_count(iter))
1671 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1674 /* caller is in process context, */
1675 stats = get_cpu_ptr(tun->pcpu_stats);
1676 u64_stats_update_begin(&stats->syncp);
1677 stats->tx_packets++;
1678 stats->tx_bytes += skb->len + vlan_hlen;
1679 u64_stats_update_end(&stats->syncp);
1680 put_cpu_ptr(tun->pcpu_stats);
1685 static struct sk_buff *tun_ring_recv(struct tun_file *tfile, int noblock,
1688 DECLARE_WAITQUEUE(wait, current);
1689 struct sk_buff *skb = NULL;
1692 skb = skb_array_consume(&tfile->tx_array);
1700 add_wait_queue(&tfile->wq.wait, &wait);
1701 current->state = TASK_INTERRUPTIBLE;
1704 skb = skb_array_consume(&tfile->tx_array);
1707 if (signal_pending(current)) {
1708 error = -ERESTARTSYS;
1711 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
1719 current->state = TASK_RUNNING;
1720 remove_wait_queue(&tfile->wq.wait, &wait);
1727 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1728 struct iov_iter *to,
1729 int noblock, struct sk_buff *skb)
1734 tun_debug(KERN_INFO, tun, "tun_do_read\n");
1736 if (!iov_iter_count(to))
1740 /* Read frames from ring */
1741 skb = tun_ring_recv(tfile, noblock, &err);
1746 ret = tun_put_user(tun, tfile, skb, to);
1747 if (unlikely(ret < 0))
1755 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1757 struct file *file = iocb->ki_filp;
1758 struct tun_file *tfile = file->private_data;
1759 struct tun_struct *tun = __tun_get(tfile);
1760 ssize_t len = iov_iter_count(to), ret;
1764 ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
1765 ret = min_t(ssize_t, ret, len);
1772 static void tun_free_netdev(struct net_device *dev)
1774 struct tun_struct *tun = netdev_priv(dev);
1776 BUG_ON(!(list_empty(&tun->disabled)));
1777 free_percpu(tun->pcpu_stats);
1778 tun_flow_uninit(tun);
1779 security_tun_dev_free_security(tun->security);
1782 static void tun_setup(struct net_device *dev)
1784 struct tun_struct *tun = netdev_priv(dev);
1786 tun->owner = INVALID_UID;
1787 tun->group = INVALID_GID;
1789 dev->ethtool_ops = &tun_ethtool_ops;
1790 dev->needs_free_netdev = true;
1791 dev->priv_destructor = tun_free_netdev;
1792 /* We prefer our own queue length */
1793 dev->tx_queue_len = TUN_READQ_SIZE;
1796 /* Trivial set of netlink ops to allow deleting tun or tap
1797 * device with netlink.
1799 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
1800 struct netlink_ext_ack *extack)
1805 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1807 .priv_size = sizeof(struct tun_struct),
1809 .validate = tun_validate,
1812 static void tun_sock_write_space(struct sock *sk)
1814 struct tun_file *tfile;
1815 wait_queue_head_t *wqueue;
1817 if (!sock_writeable(sk))
1820 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
1823 wqueue = sk_sleep(sk);
1824 if (wqueue && waitqueue_active(wqueue))
1825 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1826 POLLWRNORM | POLLWRBAND);
1828 tfile = container_of(sk, struct tun_file, sk);
1829 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1832 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1835 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1836 struct tun_struct *tun = __tun_get(tfile);
1841 ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1842 m->msg_flags & MSG_DONTWAIT,
1843 m->msg_flags & MSG_MORE);
1848 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1851 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1852 struct tun_struct *tun = __tun_get(tfile);
1858 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1862 if (flags & MSG_ERRQUEUE) {
1863 ret = sock_recv_errqueue(sock->sk, m, total_len,
1864 SOL_PACKET, TUN_TX_TIMESTAMP);
1867 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT,
1869 if (ret > (ssize_t)total_len) {
1870 m->msg_flags |= MSG_TRUNC;
1871 ret = flags & MSG_TRUNC ? ret : total_len;
1878 static int tun_peek_len(struct socket *sock)
1880 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1881 struct tun_struct *tun;
1884 tun = __tun_get(tfile);
1888 ret = skb_array_peek_len(&tfile->tx_array);
1894 /* Ops structure to mimic raw sockets with tun */
1895 static const struct proto_ops tun_socket_ops = {
1896 .peek_len = tun_peek_len,
1897 .sendmsg = tun_sendmsg,
1898 .recvmsg = tun_recvmsg,
1901 static struct proto tun_proto = {
1903 .owner = THIS_MODULE,
1904 .obj_size = sizeof(struct tun_file),
1907 static int tun_flags(struct tun_struct *tun)
1909 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1912 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1915 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1916 return sprintf(buf, "0x%x\n", tun_flags(tun));
1919 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1922 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1923 return uid_valid(tun->owner)?
1924 sprintf(buf, "%u\n",
1925 from_kuid_munged(current_user_ns(), tun->owner)):
1926 sprintf(buf, "-1\n");
1929 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1932 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1933 return gid_valid(tun->group) ?
1934 sprintf(buf, "%u\n",
1935 from_kgid_munged(current_user_ns(), tun->group)):
1936 sprintf(buf, "-1\n");
1939 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1940 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1941 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1943 static struct attribute *tun_dev_attrs[] = {
1944 &dev_attr_tun_flags.attr,
1945 &dev_attr_owner.attr,
1946 &dev_attr_group.attr,
1950 static const struct attribute_group tun_attr_group = {
1951 .attrs = tun_dev_attrs
1954 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1956 struct tun_struct *tun;
1957 struct tun_file *tfile = file->private_data;
1958 struct net_device *dev;
1961 if (tfile->detached)
1964 dev = __dev_get_by_name(net, ifr->ifr_name);
1966 if (ifr->ifr_flags & IFF_TUN_EXCL)
1968 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1969 tun = netdev_priv(dev);
1970 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1971 tun = netdev_priv(dev);
1975 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1976 !!(tun->flags & IFF_MULTI_QUEUE))
1979 if (tun_not_capable(tun))
1981 err = security_tun_dev_open(tun->security);
1985 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1989 if (tun->flags & IFF_MULTI_QUEUE &&
1990 (tun->numqueues + tun->numdisabled > 1)) {
1991 /* One or more queue has already been attached, no need
1992 * to initialize the device again.
1999 unsigned long flags = 0;
2000 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2003 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2005 err = security_tun_dev_create();
2010 if (ifr->ifr_flags & IFF_TUN) {
2014 } else if (ifr->ifr_flags & IFF_TAP) {
2022 name = ifr->ifr_name;
2024 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2025 NET_NAME_UNKNOWN, tun_setup, queues,
2030 err = dev_get_valid_name(net, dev, name);
2034 dev_net_set(dev, net);
2035 dev->rtnl_link_ops = &tun_link_ops;
2036 dev->ifindex = tfile->ifindex;
2037 dev->sysfs_groups[0] = &tun_attr_group;
2039 tun = netdev_priv(dev);
2042 tun->txflt.count = 0;
2043 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2045 tun->align = NET_SKB_PAD;
2046 tun->filter_attached = false;
2047 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2048 tun->rx_batched = 0;
2050 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2051 if (!tun->pcpu_stats) {
2056 spin_lock_init(&tun->lock);
2058 err = security_tun_dev_alloc_security(&tun->security);
2065 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2066 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2067 NETIF_F_HW_VLAN_STAG_TX;
2068 dev->features = dev->hw_features | NETIF_F_LLTX;
2069 dev->vlan_features = dev->features &
2070 ~(NETIF_F_HW_VLAN_CTAG_TX |
2071 NETIF_F_HW_VLAN_STAG_TX);
2073 INIT_LIST_HEAD(&tun->disabled);
2074 err = tun_attach(tun, file, false);
2078 err = register_netdevice(tun->dev);
2083 netif_carrier_on(tun->dev);
2085 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2087 tun->flags = (tun->flags & ~TUN_FEATURES) |
2088 (ifr->ifr_flags & TUN_FEATURES);
2090 /* Make sure persistent devices do not get stuck in
2093 if (netif_running(tun->dev))
2094 netif_tx_wake_all_queues(tun->dev);
2096 strcpy(ifr->ifr_name, tun->dev->name);
2100 tun_detach_all(dev);
2101 /* register_netdevice() already called tun_free_netdev() */
2105 tun_flow_uninit(tun);
2106 security_tun_dev_free_security(tun->security);
2108 free_percpu(tun->pcpu_stats);
2114 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2117 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2119 strcpy(ifr->ifr_name, tun->dev->name);
2121 ifr->ifr_flags = tun_flags(tun);
2125 /* This is like a cut-down ethtool ops, except done via tun fd so no
2126 * privs required. */
2127 static int set_offload(struct tun_struct *tun, unsigned long arg)
2129 netdev_features_t features = 0;
2131 if (arg & TUN_F_CSUM) {
2132 features |= NETIF_F_HW_CSUM;
2135 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2136 if (arg & TUN_F_TSO_ECN) {
2137 features |= NETIF_F_TSO_ECN;
2138 arg &= ~TUN_F_TSO_ECN;
2140 if (arg & TUN_F_TSO4)
2141 features |= NETIF_F_TSO;
2142 if (arg & TUN_F_TSO6)
2143 features |= NETIF_F_TSO6;
2144 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2148 /* This gives the user a way to test for new features in future by
2149 * trying to set them. */
2153 tun->set_features = features;
2154 tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2155 tun->dev->wanted_features |= features;
2156 netdev_update_features(tun->dev);
2161 static void tun_detach_filter(struct tun_struct *tun, int n)
2164 struct tun_file *tfile;
2166 for (i = 0; i < n; i++) {
2167 tfile = rtnl_dereference(tun->tfiles[i]);
2168 lock_sock(tfile->socket.sk);
2169 sk_detach_filter(tfile->socket.sk);
2170 release_sock(tfile->socket.sk);
2173 tun->filter_attached = false;
2176 static int tun_attach_filter(struct tun_struct *tun)
2179 struct tun_file *tfile;
2181 for (i = 0; i < tun->numqueues; i++) {
2182 tfile = rtnl_dereference(tun->tfiles[i]);
2183 lock_sock(tfile->socket.sk);
2184 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2185 release_sock(tfile->socket.sk);
2187 tun_detach_filter(tun, i);
2192 tun->filter_attached = true;
2196 static void tun_set_sndbuf(struct tun_struct *tun)
2198 struct tun_file *tfile;
2201 for (i = 0; i < tun->numqueues; i++) {
2202 tfile = rtnl_dereference(tun->tfiles[i]);
2203 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2207 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2209 struct tun_file *tfile = file->private_data;
2210 struct tun_struct *tun;
2215 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2216 tun = tfile->detached;
2221 ret = security_tun_dev_attach_queue(tun->security);
2224 ret = tun_attach(tun, file, false);
2225 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2226 tun = rtnl_dereference(tfile->tun);
2227 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2230 __tun_detach(tfile, false);
2239 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2240 unsigned long arg, int ifreq_len)
2242 struct tun_file *tfile = file->private_data;
2243 struct tun_struct *tun;
2244 void __user* argp = (void __user*)arg;
2250 unsigned int ifindex;
2254 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == SOCK_IOC_TYPE) {
2255 if (copy_from_user(&ifr, argp, ifreq_len))
2258 memset(&ifr, 0, sizeof(ifr));
2260 if (cmd == TUNGETFEATURES) {
2261 /* Currently this just means: "what IFF flags are valid?".
2262 * This is needed because we never checked for invalid flags on
2265 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2266 (unsigned int __user*)argp);
2267 } else if (cmd == TUNSETQUEUE)
2268 return tun_set_queue(file, &ifr);
2273 tun = __tun_get(tfile);
2274 if (cmd == TUNSETIFF) {
2279 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2281 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
2286 if (copy_to_user(argp, &ifr, ifreq_len))
2290 if (cmd == TUNSETIFINDEX) {
2296 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2300 tfile->ifindex = ifindex;
2308 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2313 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2315 if (tfile->detached)
2316 ifr.ifr_flags |= IFF_DETACH_QUEUE;
2317 if (!tfile->socket.sk->sk_filter)
2318 ifr.ifr_flags |= IFF_NOFILTER;
2320 if (copy_to_user(argp, &ifr, ifreq_len))
2325 /* Disable/Enable checksum */
2327 /* [unimplemented] */
2328 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2329 arg ? "disabled" : "enabled");
2333 /* Disable/Enable persist mode. Keep an extra reference to the
2334 * module to prevent the module being unprobed.
2336 if (arg && !(tun->flags & IFF_PERSIST)) {
2337 tun->flags |= IFF_PERSIST;
2338 __module_get(THIS_MODULE);
2340 if (!arg && (tun->flags & IFF_PERSIST)) {
2341 tun->flags &= ~IFF_PERSIST;
2342 module_put(THIS_MODULE);
2345 tun_debug(KERN_INFO, tun, "persist %s\n",
2346 arg ? "enabled" : "disabled");
2350 /* Set owner of the device */
2351 owner = make_kuid(current_user_ns(), arg);
2352 if (!uid_valid(owner)) {
2357 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2358 from_kuid(&init_user_ns, tun->owner));
2362 /* Set group of the device */
2363 group = make_kgid(current_user_ns(), arg);
2364 if (!gid_valid(group)) {
2369 tun_debug(KERN_INFO, tun, "group set to %u\n",
2370 from_kgid(&init_user_ns, tun->group));
2374 /* Only allow setting the type when the interface is down */
2375 if (tun->dev->flags & IFF_UP) {
2376 tun_debug(KERN_INFO, tun,
2377 "Linktype set failed because interface is up\n");
2380 tun->dev->type = (int) arg;
2381 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2393 ret = set_offload(tun, arg);
2396 case TUNSETTXFILTER:
2397 /* Can be set only for TAPs */
2399 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2401 ret = update_filter(&tun->txflt, (void __user *)arg);
2405 /* Get hw address */
2406 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2407 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2408 if (copy_to_user(argp, &ifr, ifreq_len))
2413 /* Set hw address */
2414 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2415 ifr.ifr_hwaddr.sa_data);
2417 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2421 sndbuf = tfile->socket.sk->sk_sndbuf;
2422 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2427 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2432 tun->sndbuf = sndbuf;
2433 tun_set_sndbuf(tun);
2436 case TUNGETVNETHDRSZ:
2437 vnet_hdr_sz = tun->vnet_hdr_sz;
2438 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2442 case TUNSETVNETHDRSZ:
2443 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2447 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2452 tun->vnet_hdr_sz = vnet_hdr_sz;
2456 le = !!(tun->flags & TUN_VNET_LE);
2457 if (put_user(le, (int __user *)argp))
2462 if (get_user(le, (int __user *)argp)) {
2467 tun->flags |= TUN_VNET_LE;
2469 tun->flags &= ~TUN_VNET_LE;
2473 ret = tun_get_vnet_be(tun, argp);
2477 ret = tun_set_vnet_be(tun, argp);
2480 case TUNATTACHFILTER:
2481 /* Can be set only for TAPs */
2483 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2486 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2489 ret = tun_attach_filter(tun);
2492 case TUNDETACHFILTER:
2493 /* Can be set only for TAPs */
2495 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2498 tun_detach_filter(tun, tun->numqueues);
2503 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2506 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2523 static long tun_chr_ioctl(struct file *file,
2524 unsigned int cmd, unsigned long arg)
2526 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2529 #ifdef CONFIG_COMPAT
2530 static long tun_chr_compat_ioctl(struct file *file,
2531 unsigned int cmd, unsigned long arg)
2536 case TUNSETTXFILTER:
2541 arg = (unsigned long)compat_ptr(arg);
2544 arg = (compat_ulong_t)arg;
2549 * compat_ifreq is shorter than ifreq, so we must not access beyond
2550 * the end of that structure. All fields that are used in this
2551 * driver are compatible though, we don't need to convert the
2554 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2556 #endif /* CONFIG_COMPAT */
2558 static int tun_chr_fasync(int fd, struct file *file, int on)
2560 struct tun_file *tfile = file->private_data;
2563 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2567 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2568 tfile->flags |= TUN_FASYNC;
2570 tfile->flags &= ~TUN_FASYNC;
2576 static int tun_chr_open(struct inode *inode, struct file * file)
2578 struct net *net = current->nsproxy->net_ns;
2579 struct tun_file *tfile;
2581 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2583 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2587 RCU_INIT_POINTER(tfile->tun, NULL);
2591 init_waitqueue_head(&tfile->wq.wait);
2592 RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2594 tfile->socket.file = file;
2595 tfile->socket.ops = &tun_socket_ops;
2597 sock_init_data(&tfile->socket, &tfile->sk);
2599 tfile->sk.sk_write_space = tun_sock_write_space;
2600 tfile->sk.sk_sndbuf = INT_MAX;
2602 file->private_data = tfile;
2603 INIT_LIST_HEAD(&tfile->next);
2605 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2610 static int tun_chr_close(struct inode *inode, struct file *file)
2612 struct tun_file *tfile = file->private_data;
2614 tun_detach(tfile, true);
2619 #ifdef CONFIG_PROC_FS
2620 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2622 struct tun_struct *tun;
2625 memset(&ifr, 0, sizeof(ifr));
2630 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2636 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2640 static const struct file_operations tun_fops = {
2641 .owner = THIS_MODULE,
2642 .llseek = no_llseek,
2643 .read_iter = tun_chr_read_iter,
2644 .write_iter = tun_chr_write_iter,
2645 .poll = tun_chr_poll,
2646 .unlocked_ioctl = tun_chr_ioctl,
2647 #ifdef CONFIG_COMPAT
2648 .compat_ioctl = tun_chr_compat_ioctl,
2650 .open = tun_chr_open,
2651 .release = tun_chr_close,
2652 .fasync = tun_chr_fasync,
2653 #ifdef CONFIG_PROC_FS
2654 .show_fdinfo = tun_chr_show_fdinfo,
2658 static struct miscdevice tun_miscdev = {
2661 .nodename = "net/tun",
2665 /* ethtool interface */
2667 static int tun_get_link_ksettings(struct net_device *dev,
2668 struct ethtool_link_ksettings *cmd)
2670 ethtool_link_ksettings_zero_link_mode(cmd, supported);
2671 ethtool_link_ksettings_zero_link_mode(cmd, advertising);
2672 cmd->base.speed = SPEED_10;
2673 cmd->base.duplex = DUPLEX_FULL;
2674 cmd->base.port = PORT_TP;
2675 cmd->base.phy_address = 0;
2676 cmd->base.autoneg = AUTONEG_DISABLE;
2680 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2682 struct tun_struct *tun = netdev_priv(dev);
2684 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2685 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2687 switch (tun->flags & TUN_TYPE_MASK) {
2689 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2692 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2697 static u32 tun_get_msglevel(struct net_device *dev)
2700 struct tun_struct *tun = netdev_priv(dev);
2707 static void tun_set_msglevel(struct net_device *dev, u32 value)
2710 struct tun_struct *tun = netdev_priv(dev);
2715 static int tun_get_coalesce(struct net_device *dev,
2716 struct ethtool_coalesce *ec)
2718 struct tun_struct *tun = netdev_priv(dev);
2720 ec->rx_max_coalesced_frames = tun->rx_batched;
2725 static int tun_set_coalesce(struct net_device *dev,
2726 struct ethtool_coalesce *ec)
2728 struct tun_struct *tun = netdev_priv(dev);
2730 if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
2731 tun->rx_batched = NAPI_POLL_WEIGHT;
2733 tun->rx_batched = ec->rx_max_coalesced_frames;
2738 static const struct ethtool_ops tun_ethtool_ops = {
2739 .get_drvinfo = tun_get_drvinfo,
2740 .get_msglevel = tun_get_msglevel,
2741 .set_msglevel = tun_set_msglevel,
2742 .get_link = ethtool_op_get_link,
2743 .get_ts_info = ethtool_op_get_ts_info,
2744 .get_coalesce = tun_get_coalesce,
2745 .set_coalesce = tun_set_coalesce,
2746 .get_link_ksettings = tun_get_link_ksettings,
2749 static int tun_queue_resize(struct tun_struct *tun)
2751 struct net_device *dev = tun->dev;
2752 struct tun_file *tfile;
2753 struct skb_array **arrays;
2754 int n = tun->numqueues + tun->numdisabled;
2757 arrays = kmalloc_array(n, sizeof(*arrays), GFP_KERNEL);
2761 for (i = 0; i < tun->numqueues; i++) {
2762 tfile = rtnl_dereference(tun->tfiles[i]);
2763 arrays[i] = &tfile->tx_array;
2765 list_for_each_entry(tfile, &tun->disabled, next)
2766 arrays[i++] = &tfile->tx_array;
2768 ret = skb_array_resize_multiple(arrays, n,
2769 dev->tx_queue_len, GFP_KERNEL);
2775 static int tun_device_event(struct notifier_block *unused,
2776 unsigned long event, void *ptr)
2778 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2779 struct tun_struct *tun = netdev_priv(dev);
2781 if (dev->rtnl_link_ops != &tun_link_ops)
2785 case NETDEV_CHANGE_TX_QUEUE_LEN:
2786 if (tun_queue_resize(tun))
2796 static struct notifier_block tun_notifier_block __read_mostly = {
2797 .notifier_call = tun_device_event,
2800 static int __init tun_init(void)
2804 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2806 ret = rtnl_link_register(&tun_link_ops);
2808 pr_err("Can't register link_ops\n");
2812 ret = misc_register(&tun_miscdev);
2814 pr_err("Can't register misc device %d\n", TUN_MINOR);
2818 ret = register_netdevice_notifier(&tun_notifier_block);
2820 pr_err("Can't register netdevice notifier\n");
2827 misc_deregister(&tun_miscdev);
2829 rtnl_link_unregister(&tun_link_ops);
2834 static void tun_cleanup(void)
2836 misc_deregister(&tun_miscdev);
2837 rtnl_link_unregister(&tun_link_ops);
2838 unregister_netdevice_notifier(&tun_notifier_block);
2841 /* Get an underlying socket object from tun file. Returns error unless file is
2842 * attached to a device. The returned object works like a packet socket, it
2843 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
2844 * holding a reference to the file for as long as the socket is in use. */
2845 struct socket *tun_get_socket(struct file *file)
2847 struct tun_file *tfile;
2848 if (file->f_op != &tun_fops)
2849 return ERR_PTR(-EINVAL);
2850 tfile = file->private_data;
2852 return ERR_PTR(-EBADFD);
2853 return &tfile->socket;
2855 EXPORT_SYMBOL_GPL(tun_get_socket);
2857 struct skb_array *tun_get_skb_array(struct file *file)
2859 struct tun_file *tfile;
2861 if (file->f_op != &tun_fops)
2862 return ERR_PTR(-EINVAL);
2863 tfile = file->private_data;
2865 return ERR_PTR(-EBADFD);
2866 return &tfile->tx_array;
2868 EXPORT_SYMBOL_GPL(tun_get_skb_array);
2870 module_init(tun_init);
2871 module_exit(tun_cleanup);
2872 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2873 MODULE_AUTHOR(DRV_COPYRIGHT);
2874 MODULE_LICENSE("GPL");
2875 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2876 MODULE_ALIAS("devname:net/tun");