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/major.h>
48 #include <linux/slab.h>
49 #include <linux/poll.h>
50 #include <linux/fcntl.h>
51 #include <linux/init.h>
52 #include <linux/skbuff.h>
53 #include <linux/netdevice.h>
54 #include <linux/etherdevice.h>
55 #include <linux/miscdevice.h>
56 #include <linux/ethtool.h>
57 #include <linux/rtnetlink.h>
58 #include <linux/compat.h>
60 #include <linux/if_arp.h>
61 #include <linux/if_ether.h>
62 #include <linux/if_tun.h>
63 #include <linux/if_vlan.h>
64 #include <linux/crc32.h>
65 #include <linux/nsproxy.h>
66 #include <linux/virtio_net.h>
67 #include <linux/rcupdate.h>
68 #include <net/net_namespace.h>
69 #include <net/netns/generic.h>
70 #include <net/rtnetlink.h>
72 #include <linux/seq_file.h>
73 #include <linux/uio.h>
75 #include <asm/uaccess.h>
77 /* Uncomment to enable debugging */
78 /* #define TUN_DEBUG 1 */
83 #define tun_debug(level, tun, fmt, args...) \
86 netdev_printk(level, tun->dev, fmt, ##args); \
88 #define DBG1(level, fmt, args...) \
91 printk(level fmt, ##args); \
94 #define tun_debug(level, tun, fmt, args...) \
97 netdev_printk(level, tun->dev, fmt, ##args); \
99 #define DBG1(level, fmt, args...) \
102 printk(level fmt, ##args); \
106 /* TUN device flags */
108 /* IFF_ATTACH_QUEUE is never stored in device flags,
109 * overload it to mean fasync when stored there.
111 #define TUN_FASYNC IFF_ATTACH_QUEUE
112 /* High bits in flags field are unused. */
113 #define TUN_VNET_LE 0x80000000
114 #define TUN_VNET_BE 0x40000000
116 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
118 #define GOODCOPY_LEN 128
120 #define FLT_EXACT_COUNT 8
122 unsigned int count; /* Number of addrs. Zero means disabled */
123 u32 mask[2]; /* Mask of the hashed addrs */
124 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
127 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
128 * to max number of VCPUs in guest. */
129 #define MAX_TAP_QUEUES 256
130 #define MAX_TAP_FLOWS 4096
132 #define TUN_FLOW_EXPIRE (3 * HZ)
134 /* A tun_file connects an open character device to a tuntap netdevice. It
135 * also contains all socket related structures (except sock_fprog and tap_filter)
136 * to serve as one transmit queue for tuntap device. The sock_fprog and
137 * tap_filter were kept in tun_struct since they were used for filtering for the
138 * netdevice not for a specific queue (at least I didn't see the requirement for
142 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
143 * other can only be read while rcu_read_lock or rtnl_lock is held.
147 struct socket socket;
149 struct tun_struct __rcu *tun;
150 struct fasync_struct *fasync;
151 /* only used for fasnyc */
155 unsigned int ifindex;
157 struct list_head next;
158 struct tun_struct *detached;
161 struct tun_flow_entry {
162 struct hlist_node hash_link;
164 struct tun_struct *tun;
169 unsigned long updated;
172 #define TUN_NUM_FLOW_ENTRIES 1024
174 /* Since the socket were moved to tun_file, to preserve the behavior of persist
175 * device, socket filter, sndbuf and vnet header size were restore when the
176 * file were attached to a persist device.
179 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
180 unsigned int numqueues;
185 struct net_device *dev;
186 netdev_features_t set_features;
187 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
188 NETIF_F_TSO6|NETIF_F_UFO)
192 struct tap_filter txflt;
193 struct sock_fprog fprog;
194 /* protected by rtnl lock */
195 bool filter_attached;
200 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
201 struct timer_list flow_gc_timer;
202 unsigned long ageing_time;
203 unsigned int numdisabled;
204 struct list_head disabled;
209 #ifdef CONFIG_TUN_VNET_CROSS_LE
210 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
212 return tun->flags & TUN_VNET_BE ? false :
213 virtio_legacy_is_little_endian();
216 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
218 int be = !!(tun->flags & TUN_VNET_BE);
220 if (put_user(be, argp))
226 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
230 if (get_user(be, argp))
234 tun->flags |= TUN_VNET_BE;
236 tun->flags &= ~TUN_VNET_BE;
241 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
243 return virtio_legacy_is_little_endian();
246 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
251 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
255 #endif /* CONFIG_TUN_VNET_CROSS_LE */
257 static inline bool tun_is_little_endian(struct tun_struct *tun)
259 return tun->flags & TUN_VNET_LE ||
260 tun_legacy_is_little_endian(tun);
263 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
265 return __virtio16_to_cpu(tun_is_little_endian(tun), val);
268 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
270 return __cpu_to_virtio16(tun_is_little_endian(tun), val);
273 static inline u32 tun_hashfn(u32 rxhash)
275 return rxhash & 0x3ff;
278 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
280 struct tun_flow_entry *e;
282 hlist_for_each_entry_rcu(e, head, hash_link) {
283 if (e->rxhash == rxhash)
289 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
290 struct hlist_head *head,
291 u32 rxhash, u16 queue_index)
293 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
296 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
297 rxhash, queue_index);
298 e->updated = jiffies;
301 e->queue_index = queue_index;
303 hlist_add_head_rcu(&e->hash_link, head);
309 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
311 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
312 e->rxhash, e->queue_index);
313 hlist_del_rcu(&e->hash_link);
318 static void tun_flow_flush(struct tun_struct *tun)
322 spin_lock_bh(&tun->lock);
323 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
324 struct tun_flow_entry *e;
325 struct hlist_node *n;
327 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
328 tun_flow_delete(tun, e);
330 spin_unlock_bh(&tun->lock);
333 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
337 spin_lock_bh(&tun->lock);
338 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
339 struct tun_flow_entry *e;
340 struct hlist_node *n;
342 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
343 if (e->queue_index == queue_index)
344 tun_flow_delete(tun, e);
347 spin_unlock_bh(&tun->lock);
350 static void tun_flow_cleanup(unsigned long data)
352 struct tun_struct *tun = (struct tun_struct *)data;
353 unsigned long delay = tun->ageing_time;
354 unsigned long next_timer = jiffies + delay;
355 unsigned long count = 0;
358 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
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 unsigned long this_timer;
368 this_timer = e->updated + delay;
369 if (time_before_eq(this_timer, jiffies))
370 tun_flow_delete(tun, e);
371 else if (time_before(this_timer, next_timer))
372 next_timer = this_timer;
377 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
378 spin_unlock_bh(&tun->lock);
381 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
382 struct tun_file *tfile)
384 struct hlist_head *head;
385 struct tun_flow_entry *e;
386 unsigned long delay = tun->ageing_time;
387 u16 queue_index = tfile->queue_index;
392 head = &tun->flows[tun_hashfn(rxhash)];
396 /* We may get a very small possibility of OOO during switching, not
397 * worth to optimize.*/
398 if (tun->numqueues == 1 || tfile->detached)
401 e = tun_flow_find(head, rxhash);
403 /* TODO: keep queueing to old queue until it's empty? */
404 e->queue_index = queue_index;
405 e->updated = jiffies;
406 sock_rps_record_flow_hash(e->rps_rxhash);
408 spin_lock_bh(&tun->lock);
409 if (!tun_flow_find(head, rxhash) &&
410 tun->flow_count < MAX_TAP_FLOWS)
411 tun_flow_create(tun, head, rxhash, queue_index);
413 if (!timer_pending(&tun->flow_gc_timer))
414 mod_timer(&tun->flow_gc_timer,
415 round_jiffies_up(jiffies + delay));
416 spin_unlock_bh(&tun->lock);
424 * Save the hash received in the stack receive path and update the
425 * flow_hash table accordingly.
427 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
429 if (unlikely(e->rps_rxhash != hash))
430 e->rps_rxhash = hash;
433 /* We try to identify a flow through its rxhash first. The reason that
434 * we do not check rxq no. is because some cards(e.g 82599), chooses
435 * the rxq based on the txq where the last packet of the flow comes. As
436 * the userspace application move between processors, we may get a
437 * different rxq no. here. If we could not get rxhash, then we would
438 * hope the rxq no. may help here.
440 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
441 void *accel_priv, select_queue_fallback_t fallback)
443 struct tun_struct *tun = netdev_priv(dev);
444 struct tun_flow_entry *e;
449 numqueues = ACCESS_ONCE(tun->numqueues);
451 txq = skb_get_hash(skb);
453 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
455 tun_flow_save_rps_rxhash(e, txq);
456 txq = e->queue_index;
458 /* use multiply and shift instead of expensive divide */
459 txq = ((u64)txq * numqueues) >> 32;
460 } else if (likely(skb_rx_queue_recorded(skb))) {
461 txq = skb_get_rx_queue(skb);
462 while (unlikely(txq >= numqueues))
470 static inline bool tun_not_capable(struct tun_struct *tun)
472 const struct cred *cred = current_cred();
473 struct net *net = dev_net(tun->dev);
475 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
476 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
477 !ns_capable(net->user_ns, CAP_NET_ADMIN);
480 static void tun_set_real_num_queues(struct tun_struct *tun)
482 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
483 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
486 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
488 tfile->detached = tun;
489 list_add_tail(&tfile->next, &tun->disabled);
493 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
495 struct tun_struct *tun = tfile->detached;
497 tfile->detached = NULL;
498 list_del_init(&tfile->next);
503 static void tun_queue_purge(struct tun_file *tfile)
505 skb_queue_purge(&tfile->sk.sk_receive_queue);
506 skb_queue_purge(&tfile->sk.sk_error_queue);
509 static void __tun_detach(struct tun_file *tfile, bool clean)
511 struct tun_file *ntfile;
512 struct tun_struct *tun;
514 tun = rtnl_dereference(tfile->tun);
516 if (tun && !tfile->detached) {
517 u16 index = tfile->queue_index;
518 BUG_ON(index >= tun->numqueues);
520 rcu_assign_pointer(tun->tfiles[index],
521 tun->tfiles[tun->numqueues - 1]);
522 ntfile = rtnl_dereference(tun->tfiles[index]);
523 ntfile->queue_index = index;
527 RCU_INIT_POINTER(tfile->tun, NULL);
528 sock_put(&tfile->sk);
530 tun_disable_queue(tun, tfile);
533 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
534 /* Drop read queue */
535 tun_queue_purge(tfile);
536 tun_set_real_num_queues(tun);
537 } else if (tfile->detached && clean) {
538 tun = tun_enable_queue(tfile);
539 sock_put(&tfile->sk);
543 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
544 netif_carrier_off(tun->dev);
546 if (!(tun->flags & IFF_PERSIST) &&
547 tun->dev->reg_state == NETREG_REGISTERED)
548 unregister_netdevice(tun->dev);
550 sock_put(&tfile->sk);
554 static void tun_detach(struct tun_file *tfile, bool clean)
557 __tun_detach(tfile, clean);
561 static void tun_detach_all(struct net_device *dev)
563 struct tun_struct *tun = netdev_priv(dev);
564 struct tun_file *tfile, *tmp;
565 int i, n = tun->numqueues;
567 for (i = 0; i < n; i++) {
568 tfile = rtnl_dereference(tun->tfiles[i]);
570 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
571 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
572 RCU_INIT_POINTER(tfile->tun, NULL);
575 list_for_each_entry(tfile, &tun->disabled, next) {
576 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
577 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
578 RCU_INIT_POINTER(tfile->tun, NULL);
580 BUG_ON(tun->numqueues != 0);
583 for (i = 0; i < n; i++) {
584 tfile = rtnl_dereference(tun->tfiles[i]);
585 /* Drop read queue */
586 tun_queue_purge(tfile);
587 sock_put(&tfile->sk);
589 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
590 tun_enable_queue(tfile);
591 tun_queue_purge(tfile);
592 sock_put(&tfile->sk);
594 BUG_ON(tun->numdisabled != 0);
596 if (tun->flags & IFF_PERSIST)
597 module_put(THIS_MODULE);
600 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
602 struct tun_file *tfile = file->private_data;
605 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
610 if (rtnl_dereference(tfile->tun) && !tfile->detached)
614 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
618 if (!tfile->detached &&
619 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
624 /* Re-attach the filter to persist device */
625 if (!skip_filter && (tun->filter_attached == true)) {
626 err = __sk_attach_filter(&tun->fprog, tfile->socket.sk,
627 lockdep_rtnl_is_held());
631 tfile->queue_index = tun->numqueues;
632 tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
633 rcu_assign_pointer(tfile->tun, tun);
634 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
638 tun_enable_queue(tfile);
640 sock_hold(&tfile->sk);
642 tun_set_real_num_queues(tun);
644 /* device is allowed to go away first, so no need to hold extra
652 static struct tun_struct *__tun_get(struct tun_file *tfile)
654 struct tun_struct *tun;
657 tun = rcu_dereference(tfile->tun);
665 static struct tun_struct *tun_get(struct file *file)
667 return __tun_get(file->private_data);
670 static void tun_put(struct tun_struct *tun)
676 static void addr_hash_set(u32 *mask, const u8 *addr)
678 int n = ether_crc(ETH_ALEN, addr) >> 26;
679 mask[n >> 5] |= (1 << (n & 31));
682 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
684 int n = ether_crc(ETH_ALEN, addr) >> 26;
685 return mask[n >> 5] & (1 << (n & 31));
688 static int update_filter(struct tap_filter *filter, void __user *arg)
690 struct { u8 u[ETH_ALEN]; } *addr;
691 struct tun_filter uf;
692 int err, alen, n, nexact;
694 if (copy_from_user(&uf, arg, sizeof(uf)))
703 alen = ETH_ALEN * uf.count;
704 addr = kmalloc(alen, GFP_KERNEL);
708 if (copy_from_user(addr, arg + sizeof(uf), alen)) {
713 /* The filter is updated without holding any locks. Which is
714 * perfectly safe. We disable it first and in the worst
715 * case we'll accept a few undesired packets. */
719 /* Use first set of addresses as an exact filter */
720 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
721 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
725 /* Remaining multicast addresses are hashed,
726 * unicast will leave the filter disabled. */
727 memset(filter->mask, 0, sizeof(filter->mask));
728 for (; n < uf.count; n++) {
729 if (!is_multicast_ether_addr(addr[n].u)) {
730 err = 0; /* no filter */
733 addr_hash_set(filter->mask, addr[n].u);
736 /* For ALLMULTI just set the mask to all ones.
737 * This overrides the mask populated above. */
738 if ((uf.flags & TUN_FLT_ALLMULTI))
739 memset(filter->mask, ~0, sizeof(filter->mask));
741 /* Now enable the filter */
743 filter->count = nexact;
745 /* Return the number of exact filters */
753 /* Returns: 0 - drop, !=0 - accept */
754 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
756 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
758 struct ethhdr *eh = (struct ethhdr *) skb->data;
762 for (i = 0; i < filter->count; i++)
763 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
766 /* Inexact match (multicast only) */
767 if (is_multicast_ether_addr(eh->h_dest))
768 return addr_hash_test(filter->mask, eh->h_dest);
774 * Checks whether the packet is accepted or not.
775 * Returns: 0 - drop, !=0 - accept
777 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
782 return run_filter(filter, skb);
785 /* Network device part of the driver */
787 static const struct ethtool_ops tun_ethtool_ops;
789 /* Net device detach from fd. */
790 static void tun_net_uninit(struct net_device *dev)
795 /* Net device open. */
796 static int tun_net_open(struct net_device *dev)
798 netif_tx_start_all_queues(dev);
802 /* Net device close. */
803 static int tun_net_close(struct net_device *dev)
805 netif_tx_stop_all_queues(dev);
809 /* Net device start xmit */
810 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
812 struct tun_struct *tun = netdev_priv(dev);
813 int txq = skb->queue_mapping;
814 struct tun_file *tfile;
818 tfile = rcu_dereference(tun->tfiles[txq]);
819 numqueues = ACCESS_ONCE(tun->numqueues);
821 /* Drop packet if interface is not attached */
822 if (txq >= numqueues)
825 if (numqueues == 1) {
826 /* Select queue was not called for the skbuff, so we extract the
827 * RPS hash and save it into the flow_table here.
831 rxhash = skb_get_hash(skb);
833 struct tun_flow_entry *e;
834 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
837 tun_flow_save_rps_rxhash(e, rxhash);
841 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
845 /* Drop if the filter does not like it.
846 * This is a noop if the filter is disabled.
847 * Filter can be enabled only for the TAP devices. */
848 if (!check_filter(&tun->txflt, skb))
851 if (tfile->socket.sk->sk_filter &&
852 sk_filter(tfile->socket.sk, skb))
855 /* Limit the number of packets queued by dividing txq length with the
858 if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
859 >= dev->tx_queue_len)
862 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
865 skb_tx_timestamp(skb);
867 /* Orphan the skb - required as we might hang on to it
868 * for indefinite time.
875 skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
877 /* Notify and wake up reader process */
878 if (tfile->flags & TUN_FASYNC)
879 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
880 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
886 dev->stats.tx_dropped++;
890 return NET_XMIT_DROP;
893 static void tun_net_mclist(struct net_device *dev)
896 * This callback is supposed to deal with mc filter in
897 * _rx_ path and has nothing to do with the _tx_ path.
898 * In rx path we always accept everything userspace gives us.
903 #define MAX_MTU 65535
906 tun_net_change_mtu(struct net_device *dev, int new_mtu)
908 if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
914 static netdev_features_t tun_net_fix_features(struct net_device *dev,
915 netdev_features_t features)
917 struct tun_struct *tun = netdev_priv(dev);
919 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
921 #ifdef CONFIG_NET_POLL_CONTROLLER
922 static void tun_poll_controller(struct net_device *dev)
925 * Tun only receives frames when:
926 * 1) the char device endpoint gets data from user space
927 * 2) the tun socket gets a sendmsg call from user space
928 * Since both of those are synchronous operations, we are guaranteed
929 * never to have pending data when we poll for it
930 * so there is nothing to do here but return.
931 * We need this though so netpoll recognizes us as an interface that
932 * supports polling, which enables bridge devices in virt setups to
933 * still use netconsole
938 static const struct net_device_ops tun_netdev_ops = {
939 .ndo_uninit = tun_net_uninit,
940 .ndo_open = tun_net_open,
941 .ndo_stop = tun_net_close,
942 .ndo_start_xmit = tun_net_xmit,
943 .ndo_change_mtu = tun_net_change_mtu,
944 .ndo_fix_features = tun_net_fix_features,
945 .ndo_select_queue = tun_select_queue,
946 #ifdef CONFIG_NET_POLL_CONTROLLER
947 .ndo_poll_controller = tun_poll_controller,
951 static const struct net_device_ops tap_netdev_ops = {
952 .ndo_uninit = tun_net_uninit,
953 .ndo_open = tun_net_open,
954 .ndo_stop = tun_net_close,
955 .ndo_start_xmit = tun_net_xmit,
956 .ndo_change_mtu = tun_net_change_mtu,
957 .ndo_fix_features = tun_net_fix_features,
958 .ndo_set_rx_mode = tun_net_mclist,
959 .ndo_set_mac_address = eth_mac_addr,
960 .ndo_validate_addr = eth_validate_addr,
961 .ndo_select_queue = tun_select_queue,
962 #ifdef CONFIG_NET_POLL_CONTROLLER
963 .ndo_poll_controller = tun_poll_controller,
965 .ndo_features_check = passthru_features_check,
968 static void tun_flow_init(struct tun_struct *tun)
972 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
973 INIT_HLIST_HEAD(&tun->flows[i]);
975 tun->ageing_time = TUN_FLOW_EXPIRE;
976 setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
977 mod_timer(&tun->flow_gc_timer,
978 round_jiffies_up(jiffies + tun->ageing_time));
981 static void tun_flow_uninit(struct tun_struct *tun)
983 del_timer_sync(&tun->flow_gc_timer);
987 /* Initialize net device. */
988 static void tun_net_init(struct net_device *dev)
990 struct tun_struct *tun = netdev_priv(dev);
992 switch (tun->flags & TUN_TYPE_MASK) {
994 dev->netdev_ops = &tun_netdev_ops;
996 /* Point-to-Point TUN Device */
997 dev->hard_header_len = 0;
1001 /* Zero header length */
1002 dev->type = ARPHRD_NONE;
1003 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1007 dev->netdev_ops = &tap_netdev_ops;
1008 /* Ethernet TAP Device */
1010 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1011 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1013 eth_hw_addr_random(dev);
1019 /* Character device part */
1022 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1024 struct tun_file *tfile = file->private_data;
1025 struct tun_struct *tun = __tun_get(tfile);
1027 unsigned int mask = 0;
1032 sk = tfile->socket.sk;
1034 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1036 poll_wait(file, sk_sleep(sk), wait);
1038 if (!skb_queue_empty(&sk->sk_receive_queue))
1039 mask |= POLLIN | POLLRDNORM;
1041 if (sock_writeable(sk) ||
1042 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1043 sock_writeable(sk)))
1044 mask |= POLLOUT | POLLWRNORM;
1046 if (tun->dev->reg_state != NETREG_REGISTERED)
1053 /* prepad is the amount to reserve at front. len is length after that.
1054 * linear is a hint as to how much to copy (usually headers). */
1055 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1056 size_t prepad, size_t len,
1057 size_t linear, int noblock)
1059 struct sock *sk = tfile->socket.sk;
1060 struct sk_buff *skb;
1063 /* Under a page? Don't bother with paged skb. */
1064 if (prepad + len < PAGE_SIZE || !linear)
1067 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1070 return ERR_PTR(err);
1072 skb_reserve(skb, prepad);
1073 skb_put(skb, linear);
1074 skb->data_len = len - linear;
1075 skb->len += len - linear;
1080 /* Get packet from user space buffer */
1081 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1082 void *msg_control, struct iov_iter *from,
1085 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1086 struct sk_buff *skb;
1087 size_t total_len = iov_iter_count(from);
1088 size_t len = total_len, align = NET_SKB_PAD, linear;
1089 struct virtio_net_hdr gso = { 0 };
1092 bool zerocopy = false;
1097 if (!(tun->flags & IFF_NO_PI)) {
1098 if (len < sizeof(pi))
1102 n = copy_from_iter(&pi, sizeof(pi), from);
1103 if (n != sizeof(pi))
1107 if (tun->flags & IFF_VNET_HDR) {
1108 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1110 if (len < vnet_hdr_sz)
1114 n = copy_from_iter(&gso, sizeof(gso), from);
1115 if (n != sizeof(gso))
1118 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1119 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1120 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1122 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1124 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1127 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1128 align += NET_IP_ALIGN;
1129 if (unlikely(len < ETH_HLEN ||
1130 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1134 good_linear = SKB_MAX_HEAD(align);
1137 struct iov_iter i = *from;
1139 /* There are 256 bytes to be copied in skb, so there is
1140 * enough room for skb expand head in case it is used.
1141 * The rest of the buffer is mapped from userspace.
1143 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1144 if (copylen > good_linear)
1145 copylen = good_linear;
1147 iov_iter_advance(&i, copylen);
1148 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1154 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1155 linear = good_linear;
1157 linear = tun16_to_cpu(tun, gso.hdr_len);
1160 skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1162 if (PTR_ERR(skb) != -EAGAIN)
1163 tun->dev->stats.rx_dropped++;
1164 return PTR_ERR(skb);
1168 err = zerocopy_sg_from_iter(skb, from);
1170 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1171 if (!err && msg_control) {
1172 struct ubuf_info *uarg = msg_control;
1173 uarg->callback(uarg, false);
1178 tun->dev->stats.rx_dropped++;
1183 if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1184 if (!skb_partial_csum_set(skb, tun16_to_cpu(tun, gso.csum_start),
1185 tun16_to_cpu(tun, gso.csum_offset))) {
1186 tun->dev->stats.rx_frame_errors++;
1192 switch (tun->flags & TUN_TYPE_MASK) {
1194 if (tun->flags & IFF_NO_PI) {
1195 switch (skb->data[0] & 0xf0) {
1197 pi.proto = htons(ETH_P_IP);
1200 pi.proto = htons(ETH_P_IPV6);
1203 tun->dev->stats.rx_dropped++;
1209 skb_reset_mac_header(skb);
1210 skb->protocol = pi.proto;
1211 skb->dev = tun->dev;
1214 skb->protocol = eth_type_trans(skb, tun->dev);
1218 if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1220 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1221 case VIRTIO_NET_HDR_GSO_TCPV4:
1222 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1224 case VIRTIO_NET_HDR_GSO_TCPV6:
1225 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1227 case VIRTIO_NET_HDR_GSO_UDP:
1228 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1231 tun->dev->stats.rx_frame_errors++;
1236 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1237 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1239 skb_shinfo(skb)->gso_size = tun16_to_cpu(tun, gso.gso_size);
1240 if (skb_shinfo(skb)->gso_size == 0) {
1241 tun->dev->stats.rx_frame_errors++;
1246 /* Header must be checked, and gso_segs computed. */
1247 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1248 skb_shinfo(skb)->gso_segs = 0;
1251 /* copy skb_ubuf_info for callback when skb has no error */
1253 skb_shinfo(skb)->destructor_arg = msg_control;
1254 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1255 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1258 skb_reset_network_header(skb);
1259 skb_probe_transport_header(skb, 0);
1261 rxhash = skb_get_hash(skb);
1264 tun->dev->stats.rx_packets++;
1265 tun->dev->stats.rx_bytes += len;
1267 tun_flow_update(tun, rxhash, tfile);
1271 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1273 struct file *file = iocb->ki_filp;
1274 struct tun_struct *tun = tun_get(file);
1275 struct tun_file *tfile = file->private_data;
1281 result = tun_get_user(tun, tfile, NULL, from, file->f_flags & O_NONBLOCK);
1287 /* Put packet to the user space buffer */
1288 static ssize_t tun_put_user(struct tun_struct *tun,
1289 struct tun_file *tfile,
1290 struct sk_buff *skb,
1291 struct iov_iter *iter)
1293 struct tun_pi pi = { 0, skb->protocol };
1295 int vlan_offset = 0;
1297 int vnet_hdr_sz = 0;
1299 if (skb_vlan_tag_present(skb))
1300 vlan_hlen = VLAN_HLEN;
1302 if (tun->flags & IFF_VNET_HDR)
1303 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1305 total = skb->len + vlan_hlen + vnet_hdr_sz;
1307 if (!(tun->flags & IFF_NO_PI)) {
1308 if (iov_iter_count(iter) < sizeof(pi))
1311 total += sizeof(pi);
1312 if (iov_iter_count(iter) < total) {
1313 /* Packet will be striped */
1314 pi.flags |= TUN_PKT_STRIP;
1317 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1322 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1323 if (iov_iter_count(iter) < vnet_hdr_sz)
1326 if (skb_is_gso(skb)) {
1327 struct skb_shared_info *sinfo = skb_shinfo(skb);
1329 /* This is a hint as to how much should be linear. */
1330 gso.hdr_len = cpu_to_tun16(tun, skb_headlen(skb));
1331 gso.gso_size = cpu_to_tun16(tun, sinfo->gso_size);
1332 if (sinfo->gso_type & SKB_GSO_TCPV4)
1333 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1334 else if (sinfo->gso_type & SKB_GSO_TCPV6)
1335 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1336 else if (sinfo->gso_type & SKB_GSO_UDP)
1337 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1339 pr_err("unexpected GSO type: "
1340 "0x%x, gso_size %d, hdr_len %d\n",
1341 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1342 tun16_to_cpu(tun, gso.hdr_len));
1343 print_hex_dump(KERN_ERR, "tun: ",
1346 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1350 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1351 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1353 gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1355 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1356 gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1357 gso.csum_start = cpu_to_tun16(tun, skb_checksum_start_offset(skb) +
1359 gso.csum_offset = cpu_to_tun16(tun, skb->csum_offset);
1360 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1361 gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1362 } /* else everything is zero */
1364 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1367 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1373 __be16 h_vlan_proto;
1377 veth.h_vlan_proto = skb->vlan_proto;
1378 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1380 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1382 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1383 if (ret || !iov_iter_count(iter))
1386 ret = copy_to_iter(&veth, sizeof(veth), iter);
1387 if (ret != sizeof(veth) || !iov_iter_count(iter))
1391 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1394 tun->dev->stats.tx_packets++;
1395 tun->dev->stats.tx_bytes += skb->len + vlan_hlen;
1400 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1401 struct iov_iter *to,
1404 struct sk_buff *skb;
1406 int peeked, err, off = 0;
1408 tun_debug(KERN_INFO, tun, "tun_do_read\n");
1410 if (!iov_iter_count(to))
1413 /* Read frames from queue */
1414 skb = __skb_recv_datagram(tfile->socket.sk, noblock ? MSG_DONTWAIT : 0,
1415 &peeked, &off, &err);
1419 ret = tun_put_user(tun, tfile, skb, to);
1420 if (unlikely(ret < 0))
1428 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1430 struct file *file = iocb->ki_filp;
1431 struct tun_file *tfile = file->private_data;
1432 struct tun_struct *tun = __tun_get(tfile);
1433 ssize_t len = iov_iter_count(to), ret;
1437 ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK);
1438 ret = min_t(ssize_t, ret, len);
1445 static void tun_free_netdev(struct net_device *dev)
1447 struct tun_struct *tun = netdev_priv(dev);
1449 BUG_ON(!(list_empty(&tun->disabled)));
1450 tun_flow_uninit(tun);
1451 security_tun_dev_free_security(tun->security);
1455 static void tun_setup(struct net_device *dev)
1457 struct tun_struct *tun = netdev_priv(dev);
1459 tun->owner = INVALID_UID;
1460 tun->group = INVALID_GID;
1462 dev->ethtool_ops = &tun_ethtool_ops;
1463 dev->destructor = tun_free_netdev;
1464 /* We prefer our own queue length */
1465 dev->tx_queue_len = TUN_READQ_SIZE;
1468 /* Trivial set of netlink ops to allow deleting tun or tap
1469 * device with netlink.
1471 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1476 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1478 .priv_size = sizeof(struct tun_struct),
1480 .validate = tun_validate,
1483 static void tun_sock_write_space(struct sock *sk)
1485 struct tun_file *tfile;
1486 wait_queue_head_t *wqueue;
1488 if (!sock_writeable(sk))
1491 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
1494 wqueue = sk_sleep(sk);
1495 if (wqueue && waitqueue_active(wqueue))
1496 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1497 POLLWRNORM | POLLWRBAND);
1499 tfile = container_of(sk, struct tun_file, sk);
1500 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1503 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1506 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1507 struct tun_struct *tun = __tun_get(tfile);
1512 ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1513 m->msg_flags & MSG_DONTWAIT);
1518 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1521 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1522 struct tun_struct *tun = __tun_get(tfile);
1528 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1532 if (flags & MSG_ERRQUEUE) {
1533 ret = sock_recv_errqueue(sock->sk, m, total_len,
1534 SOL_PACKET, TUN_TX_TIMESTAMP);
1537 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT);
1538 if (ret > (ssize_t)total_len) {
1539 m->msg_flags |= MSG_TRUNC;
1540 ret = flags & MSG_TRUNC ? ret : total_len;
1547 /* Ops structure to mimic raw sockets with tun */
1548 static const struct proto_ops tun_socket_ops = {
1549 .sendmsg = tun_sendmsg,
1550 .recvmsg = tun_recvmsg,
1553 static struct proto tun_proto = {
1555 .owner = THIS_MODULE,
1556 .obj_size = sizeof(struct tun_file),
1559 static int tun_flags(struct tun_struct *tun)
1561 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1564 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1567 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1568 return sprintf(buf, "0x%x\n", tun_flags(tun));
1571 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1574 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1575 return uid_valid(tun->owner)?
1576 sprintf(buf, "%u\n",
1577 from_kuid_munged(current_user_ns(), tun->owner)):
1578 sprintf(buf, "-1\n");
1581 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1584 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1585 return gid_valid(tun->group) ?
1586 sprintf(buf, "%u\n",
1587 from_kgid_munged(current_user_ns(), tun->group)):
1588 sprintf(buf, "-1\n");
1591 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1592 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1593 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1595 static struct attribute *tun_dev_attrs[] = {
1596 &dev_attr_tun_flags.attr,
1597 &dev_attr_owner.attr,
1598 &dev_attr_group.attr,
1602 static const struct attribute_group tun_attr_group = {
1603 .attrs = tun_dev_attrs
1606 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1608 struct tun_struct *tun;
1609 struct tun_file *tfile = file->private_data;
1610 struct net_device *dev;
1613 if (tfile->detached)
1616 dev = __dev_get_by_name(net, ifr->ifr_name);
1618 if (ifr->ifr_flags & IFF_TUN_EXCL)
1620 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1621 tun = netdev_priv(dev);
1622 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1623 tun = netdev_priv(dev);
1627 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1628 !!(tun->flags & IFF_MULTI_QUEUE))
1631 if (tun_not_capable(tun))
1633 err = security_tun_dev_open(tun->security);
1637 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1641 if (tun->flags & IFF_MULTI_QUEUE &&
1642 (tun->numqueues + tun->numdisabled > 1)) {
1643 /* One or more queue has already been attached, no need
1644 * to initialize the device again.
1651 unsigned long flags = 0;
1652 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1655 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1657 err = security_tun_dev_create();
1662 if (ifr->ifr_flags & IFF_TUN) {
1666 } else if (ifr->ifr_flags & IFF_TAP) {
1674 name = ifr->ifr_name;
1676 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1677 NET_NAME_UNKNOWN, tun_setup, queues,
1683 dev_net_set(dev, net);
1684 dev->rtnl_link_ops = &tun_link_ops;
1685 dev->ifindex = tfile->ifindex;
1686 dev->sysfs_groups[0] = &tun_attr_group;
1688 tun = netdev_priv(dev);
1691 tun->txflt.count = 0;
1692 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1694 tun->filter_attached = false;
1695 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1697 spin_lock_init(&tun->lock);
1699 err = security_tun_dev_alloc_security(&tun->security);
1706 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1707 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1708 NETIF_F_HW_VLAN_STAG_TX;
1709 dev->features = dev->hw_features;
1710 dev->vlan_features = dev->features &
1711 ~(NETIF_F_HW_VLAN_CTAG_TX |
1712 NETIF_F_HW_VLAN_STAG_TX);
1714 INIT_LIST_HEAD(&tun->disabled);
1715 err = tun_attach(tun, file, false);
1719 err = register_netdevice(tun->dev);
1724 netif_carrier_on(tun->dev);
1726 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1728 tun->flags = (tun->flags & ~TUN_FEATURES) |
1729 (ifr->ifr_flags & TUN_FEATURES);
1731 /* Make sure persistent devices do not get stuck in
1734 if (netif_running(tun->dev))
1735 netif_tx_wake_all_queues(tun->dev);
1737 strcpy(ifr->ifr_name, tun->dev->name);
1741 tun_detach_all(dev);
1743 tun_flow_uninit(tun);
1744 security_tun_dev_free_security(tun->security);
1750 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1753 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1755 strcpy(ifr->ifr_name, tun->dev->name);
1757 ifr->ifr_flags = tun_flags(tun);
1761 /* This is like a cut-down ethtool ops, except done via tun fd so no
1762 * privs required. */
1763 static int set_offload(struct tun_struct *tun, unsigned long arg)
1765 netdev_features_t features = 0;
1767 if (arg & TUN_F_CSUM) {
1768 features |= NETIF_F_HW_CSUM;
1771 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1772 if (arg & TUN_F_TSO_ECN) {
1773 features |= NETIF_F_TSO_ECN;
1774 arg &= ~TUN_F_TSO_ECN;
1776 if (arg & TUN_F_TSO4)
1777 features |= NETIF_F_TSO;
1778 if (arg & TUN_F_TSO6)
1779 features |= NETIF_F_TSO6;
1780 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1783 if (arg & TUN_F_UFO) {
1784 features |= NETIF_F_UFO;
1789 /* This gives the user a way to test for new features in future by
1790 * trying to set them. */
1794 tun->set_features = features;
1795 netdev_update_features(tun->dev);
1800 static void tun_detach_filter(struct tun_struct *tun, int n)
1803 struct tun_file *tfile;
1805 for (i = 0; i < n; i++) {
1806 tfile = rtnl_dereference(tun->tfiles[i]);
1807 __sk_detach_filter(tfile->socket.sk, lockdep_rtnl_is_held());
1810 tun->filter_attached = false;
1813 static int tun_attach_filter(struct tun_struct *tun)
1816 struct tun_file *tfile;
1818 for (i = 0; i < tun->numqueues; i++) {
1819 tfile = rtnl_dereference(tun->tfiles[i]);
1820 ret = __sk_attach_filter(&tun->fprog, tfile->socket.sk,
1821 lockdep_rtnl_is_held());
1823 tun_detach_filter(tun, i);
1828 tun->filter_attached = true;
1832 static void tun_set_sndbuf(struct tun_struct *tun)
1834 struct tun_file *tfile;
1837 for (i = 0; i < tun->numqueues; i++) {
1838 tfile = rtnl_dereference(tun->tfiles[i]);
1839 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1843 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1845 struct tun_file *tfile = file->private_data;
1846 struct tun_struct *tun;
1851 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1852 tun = tfile->detached;
1857 ret = security_tun_dev_attach_queue(tun->security);
1860 ret = tun_attach(tun, file, false);
1861 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1862 tun = rtnl_dereference(tfile->tun);
1863 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
1866 __tun_detach(tfile, false);
1875 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1876 unsigned long arg, int ifreq_len)
1878 struct tun_file *tfile = file->private_data;
1879 struct tun_struct *tun;
1880 void __user* argp = (void __user*)arg;
1886 unsigned int ifindex;
1890 #ifdef CONFIG_ANDROID_PARANOID_NETWORK
1891 if (cmd != TUNGETIFF && !capable(CAP_NET_ADMIN)) {
1896 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1897 if (copy_from_user(&ifr, argp, ifreq_len))
1900 memset(&ifr, 0, sizeof(ifr));
1902 if (cmd == TUNGETFEATURES) {
1903 /* Currently this just means: "what IFF flags are valid?".
1904 * This is needed because we never checked for invalid flags on
1907 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
1908 (unsigned int __user*)argp);
1909 } else if (cmd == TUNSETQUEUE)
1910 return tun_set_queue(file, &ifr);
1915 tun = __tun_get(tfile);
1916 if (cmd == TUNSETIFF && !tun) {
1917 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1919 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
1924 if (copy_to_user(argp, &ifr, ifreq_len))
1928 if (cmd == TUNSETIFINDEX) {
1934 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
1938 tfile->ifindex = ifindex;
1946 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1951 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1953 if (tfile->detached)
1954 ifr.ifr_flags |= IFF_DETACH_QUEUE;
1955 if (!tfile->socket.sk->sk_filter)
1956 ifr.ifr_flags |= IFF_NOFILTER;
1958 if (copy_to_user(argp, &ifr, ifreq_len))
1963 /* Disable/Enable checksum */
1965 /* [unimplemented] */
1966 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1967 arg ? "disabled" : "enabled");
1971 /* Disable/Enable persist mode. Keep an extra reference to the
1972 * module to prevent the module being unprobed.
1974 if (arg && !(tun->flags & IFF_PERSIST)) {
1975 tun->flags |= IFF_PERSIST;
1976 __module_get(THIS_MODULE);
1978 if (!arg && (tun->flags & IFF_PERSIST)) {
1979 tun->flags &= ~IFF_PERSIST;
1980 module_put(THIS_MODULE);
1983 tun_debug(KERN_INFO, tun, "persist %s\n",
1984 arg ? "enabled" : "disabled");
1988 /* Set owner of the device */
1989 owner = make_kuid(current_user_ns(), arg);
1990 if (!uid_valid(owner)) {
1995 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1996 from_kuid(&init_user_ns, tun->owner));
2000 /* Set group of the device */
2001 group = make_kgid(current_user_ns(), arg);
2002 if (!gid_valid(group)) {
2007 tun_debug(KERN_INFO, tun, "group set to %u\n",
2008 from_kgid(&init_user_ns, tun->group));
2012 /* Only allow setting the type when the interface is down */
2013 if (tun->dev->flags & IFF_UP) {
2014 tun_debug(KERN_INFO, tun,
2015 "Linktype set failed because interface is up\n");
2018 tun->dev->type = (int) arg;
2019 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2031 ret = set_offload(tun, arg);
2034 case TUNSETTXFILTER:
2035 /* Can be set only for TAPs */
2037 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2039 ret = update_filter(&tun->txflt, (void __user *)arg);
2043 /* Get hw address */
2044 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2045 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2046 if (copy_to_user(argp, &ifr, ifreq_len))
2051 /* Set hw address */
2052 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2053 ifr.ifr_hwaddr.sa_data);
2055 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2059 sndbuf = tfile->socket.sk->sk_sndbuf;
2060 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2065 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2070 tun->sndbuf = sndbuf;
2071 tun_set_sndbuf(tun);
2074 case TUNGETVNETHDRSZ:
2075 vnet_hdr_sz = tun->vnet_hdr_sz;
2076 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2080 case TUNSETVNETHDRSZ:
2081 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2085 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2090 tun->vnet_hdr_sz = vnet_hdr_sz;
2094 le = !!(tun->flags & TUN_VNET_LE);
2095 if (put_user(le, (int __user *)argp))
2100 if (get_user(le, (int __user *)argp)) {
2105 tun->flags |= TUN_VNET_LE;
2107 tun->flags &= ~TUN_VNET_LE;
2111 ret = tun_get_vnet_be(tun, argp);
2115 ret = tun_set_vnet_be(tun, argp);
2118 case TUNATTACHFILTER:
2119 /* Can be set only for TAPs */
2121 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2124 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2127 ret = tun_attach_filter(tun);
2130 case TUNDETACHFILTER:
2131 /* Can be set only for TAPs */
2133 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2136 tun_detach_filter(tun, tun->numqueues);
2141 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2144 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2161 static long tun_chr_ioctl(struct file *file,
2162 unsigned int cmd, unsigned long arg)
2164 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2167 #ifdef CONFIG_COMPAT
2168 static long tun_chr_compat_ioctl(struct file *file,
2169 unsigned int cmd, unsigned long arg)
2174 case TUNSETTXFILTER:
2179 arg = (unsigned long)compat_ptr(arg);
2182 arg = (compat_ulong_t)arg;
2187 * compat_ifreq is shorter than ifreq, so we must not access beyond
2188 * the end of that structure. All fields that are used in this
2189 * driver are compatible though, we don't need to convert the
2192 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2194 #endif /* CONFIG_COMPAT */
2196 static int tun_chr_fasync(int fd, struct file *file, int on)
2198 struct tun_file *tfile = file->private_data;
2201 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2205 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2206 tfile->flags |= TUN_FASYNC;
2208 tfile->flags &= ~TUN_FASYNC;
2214 static int tun_chr_open(struct inode *inode, struct file * file)
2216 struct net *net = current->nsproxy->net_ns;
2217 struct tun_file *tfile;
2219 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2221 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2225 RCU_INIT_POINTER(tfile->tun, NULL);
2229 init_waitqueue_head(&tfile->wq.wait);
2230 RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2232 tfile->socket.file = file;
2233 tfile->socket.ops = &tun_socket_ops;
2235 sock_init_data(&tfile->socket, &tfile->sk);
2237 tfile->sk.sk_write_space = tun_sock_write_space;
2238 tfile->sk.sk_sndbuf = INT_MAX;
2240 file->private_data = tfile;
2241 INIT_LIST_HEAD(&tfile->next);
2243 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2248 static int tun_chr_close(struct inode *inode, struct file *file)
2250 struct tun_file *tfile = file->private_data;
2252 tun_detach(tfile, true);
2257 #ifdef CONFIG_PROC_FS
2258 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2260 struct tun_struct *tun;
2263 memset(&ifr, 0, sizeof(ifr));
2268 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2274 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2278 static const struct file_operations tun_fops = {
2279 .owner = THIS_MODULE,
2280 .llseek = no_llseek,
2281 .read_iter = tun_chr_read_iter,
2282 .write_iter = tun_chr_write_iter,
2283 .poll = tun_chr_poll,
2284 .unlocked_ioctl = tun_chr_ioctl,
2285 #ifdef CONFIG_COMPAT
2286 .compat_ioctl = tun_chr_compat_ioctl,
2288 .open = tun_chr_open,
2289 .release = tun_chr_close,
2290 .fasync = tun_chr_fasync,
2291 #ifdef CONFIG_PROC_FS
2292 .show_fdinfo = tun_chr_show_fdinfo,
2296 static struct miscdevice tun_miscdev = {
2299 .nodename = "net/tun",
2303 /* ethtool interface */
2305 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2308 cmd->advertising = 0;
2309 ethtool_cmd_speed_set(cmd, SPEED_10);
2310 cmd->duplex = DUPLEX_FULL;
2311 cmd->port = PORT_TP;
2312 cmd->phy_address = 0;
2313 cmd->transceiver = XCVR_INTERNAL;
2314 cmd->autoneg = AUTONEG_DISABLE;
2320 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2322 struct tun_struct *tun = netdev_priv(dev);
2324 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2325 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2327 switch (tun->flags & TUN_TYPE_MASK) {
2329 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2332 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2337 static u32 tun_get_msglevel(struct net_device *dev)
2340 struct tun_struct *tun = netdev_priv(dev);
2347 static void tun_set_msglevel(struct net_device *dev, u32 value)
2350 struct tun_struct *tun = netdev_priv(dev);
2355 static const struct ethtool_ops tun_ethtool_ops = {
2356 .get_settings = tun_get_settings,
2357 .get_drvinfo = tun_get_drvinfo,
2358 .get_msglevel = tun_get_msglevel,
2359 .set_msglevel = tun_set_msglevel,
2360 .get_link = ethtool_op_get_link,
2361 .get_ts_info = ethtool_op_get_ts_info,
2365 static int __init tun_init(void)
2369 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2370 pr_info("%s\n", DRV_COPYRIGHT);
2372 ret = rtnl_link_register(&tun_link_ops);
2374 pr_err("Can't register link_ops\n");
2378 ret = misc_register(&tun_miscdev);
2380 pr_err("Can't register misc device %d\n", TUN_MINOR);
2385 rtnl_link_unregister(&tun_link_ops);
2390 static void tun_cleanup(void)
2392 misc_deregister(&tun_miscdev);
2393 rtnl_link_unregister(&tun_link_ops);
2396 /* Get an underlying socket object from tun file. Returns error unless file is
2397 * attached to a device. The returned object works like a packet socket, it
2398 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
2399 * holding a reference to the file for as long as the socket is in use. */
2400 struct socket *tun_get_socket(struct file *file)
2402 struct tun_file *tfile;
2403 if (file->f_op != &tun_fops)
2404 return ERR_PTR(-EINVAL);
2405 tfile = file->private_data;
2407 return ERR_PTR(-EBADFD);
2408 return &tfile->socket;
2410 EXPORT_SYMBOL_GPL(tun_get_socket);
2412 module_init(tun_init);
2413 module_exit(tun_cleanup);
2414 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2415 MODULE_AUTHOR(DRV_COPYRIGHT);
2416 MODULE_LICENSE("GPL");
2417 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2418 MODULE_ALIAS("devname:net/tun");