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>
74 #include <linux/skb_array.h>
76 #include <linux/uaccess.h>
78 /* Uncomment to enable debugging */
79 /* #define TUN_DEBUG 1 */
84 #define tun_debug(level, tun, fmt, args...) \
87 netdev_printk(level, tun->dev, fmt, ##args); \
89 #define DBG1(level, fmt, args...) \
92 printk(level fmt, ##args); \
95 #define tun_debug(level, tun, fmt, args...) \
98 netdev_printk(level, tun->dev, fmt, ##args); \
100 #define DBG1(level, fmt, args...) \
103 printk(level fmt, ##args); \
107 /* TUN device flags */
109 /* IFF_ATTACH_QUEUE is never stored in device flags,
110 * overload it to mean fasync when stored there.
112 #define TUN_FASYNC IFF_ATTACH_QUEUE
113 /* High bits in flags field are unused. */
114 #define TUN_VNET_LE 0x80000000
115 #define TUN_VNET_BE 0x40000000
117 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
119 #define GOODCOPY_LEN 128
121 #define FLT_EXACT_COUNT 8
123 unsigned int count; /* Number of addrs. Zero means disabled */
124 u32 mask[2]; /* Mask of the hashed addrs */
125 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
128 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
129 * to max number of VCPUs in guest. */
130 #define MAX_TAP_QUEUES 256
131 #define MAX_TAP_FLOWS 4096
133 #define TUN_FLOW_EXPIRE (3 * HZ)
135 struct tun_pcpu_stats {
140 struct u64_stats_sync syncp;
146 /* A tun_file connects an open character device to a tuntap netdevice. It
147 * also contains all socket related structures (except sock_fprog and tap_filter)
148 * to serve as one transmit queue for tuntap device. The sock_fprog and
149 * tap_filter were kept in tun_struct since they were used for filtering for the
150 * netdevice not for a specific queue (at least I didn't see the requirement for
154 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
155 * other can only be read while rcu_read_lock or rtnl_lock is held.
159 struct socket socket;
161 struct tun_struct __rcu *tun;
162 struct fasync_struct *fasync;
163 /* only used for fasnyc */
167 unsigned int ifindex;
169 struct list_head next;
170 struct tun_struct *detached;
171 struct skb_array tx_array;
174 struct tun_flow_entry {
175 struct hlist_node hash_link;
177 struct tun_struct *tun;
182 unsigned long updated;
185 #define TUN_NUM_FLOW_ENTRIES 1024
187 /* Since the socket were moved to tun_file, to preserve the behavior of persist
188 * device, socket filter, sndbuf and vnet header size were restore when the
189 * file were attached to a persist device.
192 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
193 unsigned int numqueues;
198 struct net_device *dev;
199 netdev_features_t set_features;
200 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
201 NETIF_F_TSO6|NETIF_F_UFO)
206 struct tap_filter txflt;
207 struct sock_fprog fprog;
208 /* protected by rtnl lock */
209 bool filter_attached;
214 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
215 struct timer_list flow_gc_timer;
216 unsigned long ageing_time;
217 unsigned int numdisabled;
218 struct list_head disabled;
222 struct tun_pcpu_stats __percpu *pcpu_stats;
225 #ifdef CONFIG_TUN_VNET_CROSS_LE
226 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
228 return tun->flags & TUN_VNET_BE ? false :
229 virtio_legacy_is_little_endian();
232 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
234 int be = !!(tun->flags & TUN_VNET_BE);
236 if (put_user(be, argp))
242 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
246 if (get_user(be, argp))
250 tun->flags |= TUN_VNET_BE;
252 tun->flags &= ~TUN_VNET_BE;
257 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
259 return virtio_legacy_is_little_endian();
262 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
267 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
271 #endif /* CONFIG_TUN_VNET_CROSS_LE */
273 static inline bool tun_is_little_endian(struct tun_struct *tun)
275 return tun->flags & TUN_VNET_LE ||
276 tun_legacy_is_little_endian(tun);
279 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
281 return __virtio16_to_cpu(tun_is_little_endian(tun), val);
284 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
286 return __cpu_to_virtio16(tun_is_little_endian(tun), val);
289 static inline u32 tun_hashfn(u32 rxhash)
291 return rxhash & 0x3ff;
294 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
296 struct tun_flow_entry *e;
298 hlist_for_each_entry_rcu(e, head, hash_link) {
299 if (e->rxhash == rxhash)
305 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
306 struct hlist_head *head,
307 u32 rxhash, u16 queue_index)
309 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
312 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
313 rxhash, queue_index);
314 e->updated = jiffies;
317 e->queue_index = queue_index;
319 hlist_add_head_rcu(&e->hash_link, head);
325 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
327 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
328 e->rxhash, e->queue_index);
329 hlist_del_rcu(&e->hash_link);
334 static void tun_flow_flush(struct tun_struct *tun)
338 spin_lock_bh(&tun->lock);
339 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
340 struct tun_flow_entry *e;
341 struct hlist_node *n;
343 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
344 tun_flow_delete(tun, e);
346 spin_unlock_bh(&tun->lock);
349 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
353 spin_lock_bh(&tun->lock);
354 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
355 struct tun_flow_entry *e;
356 struct hlist_node *n;
358 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
359 if (e->queue_index == queue_index)
360 tun_flow_delete(tun, e);
363 spin_unlock_bh(&tun->lock);
366 static void tun_flow_cleanup(unsigned long data)
368 struct tun_struct *tun = (struct tun_struct *)data;
369 unsigned long delay = tun->ageing_time;
370 unsigned long next_timer = jiffies + delay;
371 unsigned long count = 0;
374 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
376 spin_lock_bh(&tun->lock);
377 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
378 struct tun_flow_entry *e;
379 struct hlist_node *n;
381 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
382 unsigned long this_timer;
384 this_timer = e->updated + delay;
385 if (time_before_eq(this_timer, jiffies))
386 tun_flow_delete(tun, e);
387 else if (time_before(this_timer, next_timer))
388 next_timer = this_timer;
393 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
394 spin_unlock_bh(&tun->lock);
397 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
398 struct tun_file *tfile)
400 struct hlist_head *head;
401 struct tun_flow_entry *e;
402 unsigned long delay = tun->ageing_time;
403 u16 queue_index = tfile->queue_index;
408 head = &tun->flows[tun_hashfn(rxhash)];
412 /* We may get a very small possibility of OOO during switching, not
413 * worth to optimize.*/
414 if (tun->numqueues == 1 || tfile->detached)
417 e = tun_flow_find(head, rxhash);
419 /* TODO: keep queueing to old queue until it's empty? */
420 e->queue_index = queue_index;
421 e->updated = jiffies;
422 sock_rps_record_flow_hash(e->rps_rxhash);
424 spin_lock_bh(&tun->lock);
425 if (!tun_flow_find(head, rxhash) &&
426 tun->flow_count < MAX_TAP_FLOWS)
427 tun_flow_create(tun, head, rxhash, queue_index);
429 if (!timer_pending(&tun->flow_gc_timer))
430 mod_timer(&tun->flow_gc_timer,
431 round_jiffies_up(jiffies + delay));
432 spin_unlock_bh(&tun->lock);
440 * Save the hash received in the stack receive path and update the
441 * flow_hash table accordingly.
443 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
445 if (unlikely(e->rps_rxhash != hash))
446 e->rps_rxhash = hash;
449 /* We try to identify a flow through its rxhash first. The reason that
450 * we do not check rxq no. is because some cards(e.g 82599), chooses
451 * the rxq based on the txq where the last packet of the flow comes. As
452 * the userspace application move between processors, we may get a
453 * different rxq no. here. If we could not get rxhash, then we would
454 * hope the rxq no. may help here.
456 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
457 void *accel_priv, select_queue_fallback_t fallback)
459 struct tun_struct *tun = netdev_priv(dev);
460 struct tun_flow_entry *e;
465 numqueues = ACCESS_ONCE(tun->numqueues);
467 txq = skb_get_hash(skb);
469 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
471 tun_flow_save_rps_rxhash(e, txq);
472 txq = e->queue_index;
474 /* use multiply and shift instead of expensive divide */
475 txq = ((u64)txq * numqueues) >> 32;
476 } else if (likely(skb_rx_queue_recorded(skb))) {
477 txq = skb_get_rx_queue(skb);
478 while (unlikely(txq >= numqueues))
486 static inline bool tun_not_capable(struct tun_struct *tun)
488 const struct cred *cred = current_cred();
489 struct net *net = dev_net(tun->dev);
491 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
492 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
493 !ns_capable(net->user_ns, CAP_NET_ADMIN);
496 static void tun_set_real_num_queues(struct tun_struct *tun)
498 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
499 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
502 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
504 tfile->detached = tun;
505 list_add_tail(&tfile->next, &tun->disabled);
509 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
511 struct tun_struct *tun = tfile->detached;
513 tfile->detached = NULL;
514 list_del_init(&tfile->next);
519 static void tun_queue_purge(struct tun_file *tfile)
523 while ((skb = skb_array_consume(&tfile->tx_array)) != NULL)
526 skb_queue_purge(&tfile->sk.sk_write_queue);
527 skb_queue_purge(&tfile->sk.sk_error_queue);
530 static void __tun_detach(struct tun_file *tfile, bool clean)
532 struct tun_file *ntfile;
533 struct tun_struct *tun;
535 tun = rtnl_dereference(tfile->tun);
537 if (tun && !tfile->detached) {
538 u16 index = tfile->queue_index;
539 BUG_ON(index >= tun->numqueues);
541 rcu_assign_pointer(tun->tfiles[index],
542 tun->tfiles[tun->numqueues - 1]);
543 ntfile = rtnl_dereference(tun->tfiles[index]);
544 ntfile->queue_index = index;
548 RCU_INIT_POINTER(tfile->tun, NULL);
549 sock_put(&tfile->sk);
551 tun_disable_queue(tun, tfile);
554 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
555 /* Drop read queue */
556 tun_queue_purge(tfile);
557 tun_set_real_num_queues(tun);
558 } else if (tfile->detached && clean) {
559 tun = tun_enable_queue(tfile);
560 sock_put(&tfile->sk);
564 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
565 netif_carrier_off(tun->dev);
567 if (!(tun->flags & IFF_PERSIST) &&
568 tun->dev->reg_state == NETREG_REGISTERED)
569 unregister_netdevice(tun->dev);
572 skb_array_cleanup(&tfile->tx_array);
573 sock_put(&tfile->sk);
577 static void tun_detach(struct tun_file *tfile, bool clean)
580 __tun_detach(tfile, clean);
584 static void tun_detach_all(struct net_device *dev)
586 struct tun_struct *tun = netdev_priv(dev);
587 struct tun_file *tfile, *tmp;
588 int i, n = tun->numqueues;
590 for (i = 0; i < n; i++) {
591 tfile = rtnl_dereference(tun->tfiles[i]);
593 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
594 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
595 RCU_INIT_POINTER(tfile->tun, NULL);
598 list_for_each_entry(tfile, &tun->disabled, next) {
599 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
600 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
601 RCU_INIT_POINTER(tfile->tun, NULL);
603 BUG_ON(tun->numqueues != 0);
606 for (i = 0; i < n; i++) {
607 tfile = rtnl_dereference(tun->tfiles[i]);
608 /* Drop read queue */
609 tun_queue_purge(tfile);
610 sock_put(&tfile->sk);
612 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
613 tun_enable_queue(tfile);
614 tun_queue_purge(tfile);
615 sock_put(&tfile->sk);
617 BUG_ON(tun->numdisabled != 0);
619 if (tun->flags & IFF_PERSIST)
620 module_put(THIS_MODULE);
623 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
625 struct tun_file *tfile = file->private_data;
626 struct net_device *dev = tun->dev;
629 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
634 if (rtnl_dereference(tfile->tun) && !tfile->detached)
638 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
642 if (!tfile->detached &&
643 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
648 /* Re-attach the filter to persist device */
649 if (!skip_filter && (tun->filter_attached == true)) {
650 lock_sock(tfile->socket.sk);
651 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
652 release_sock(tfile->socket.sk);
657 if (!tfile->detached &&
658 skb_array_init(&tfile->tx_array, dev->tx_queue_len, GFP_KERNEL)) {
663 tfile->queue_index = tun->numqueues;
664 tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
665 rcu_assign_pointer(tfile->tun, tun);
666 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
670 tun_enable_queue(tfile);
672 sock_hold(&tfile->sk);
674 tun_set_real_num_queues(tun);
676 /* device is allowed to go away first, so no need to hold extra
684 static struct tun_struct *__tun_get(struct tun_file *tfile)
686 struct tun_struct *tun;
689 tun = rcu_dereference(tfile->tun);
697 static struct tun_struct *tun_get(struct file *file)
699 return __tun_get(file->private_data);
702 static void tun_put(struct tun_struct *tun)
708 static void addr_hash_set(u32 *mask, const u8 *addr)
710 int n = ether_crc(ETH_ALEN, addr) >> 26;
711 mask[n >> 5] |= (1 << (n & 31));
714 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
716 int n = ether_crc(ETH_ALEN, addr) >> 26;
717 return mask[n >> 5] & (1 << (n & 31));
720 static int update_filter(struct tap_filter *filter, void __user *arg)
722 struct { u8 u[ETH_ALEN]; } *addr;
723 struct tun_filter uf;
724 int err, alen, n, nexact;
726 if (copy_from_user(&uf, arg, sizeof(uf)))
735 alen = ETH_ALEN * uf.count;
736 addr = memdup_user(arg + sizeof(uf), alen);
738 return PTR_ERR(addr);
740 /* The filter is updated without holding any locks. Which is
741 * perfectly safe. We disable it first and in the worst
742 * case we'll accept a few undesired packets. */
746 /* Use first set of addresses as an exact filter */
747 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
748 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
752 /* Remaining multicast addresses are hashed,
753 * unicast will leave the filter disabled. */
754 memset(filter->mask, 0, sizeof(filter->mask));
755 for (; n < uf.count; n++) {
756 if (!is_multicast_ether_addr(addr[n].u)) {
757 err = 0; /* no filter */
760 addr_hash_set(filter->mask, addr[n].u);
763 /* For ALLMULTI just set the mask to all ones.
764 * This overrides the mask populated above. */
765 if ((uf.flags & TUN_FLT_ALLMULTI))
766 memset(filter->mask, ~0, sizeof(filter->mask));
768 /* Now enable the filter */
770 filter->count = nexact;
772 /* Return the number of exact filters */
779 /* Returns: 0 - drop, !=0 - accept */
780 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
782 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
784 struct ethhdr *eh = (struct ethhdr *) skb->data;
788 for (i = 0; i < filter->count; i++)
789 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
792 /* Inexact match (multicast only) */
793 if (is_multicast_ether_addr(eh->h_dest))
794 return addr_hash_test(filter->mask, eh->h_dest);
800 * Checks whether the packet is accepted or not.
801 * Returns: 0 - drop, !=0 - accept
803 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
808 return run_filter(filter, skb);
811 /* Network device part of the driver */
813 static const struct ethtool_ops tun_ethtool_ops;
815 /* Net device detach from fd. */
816 static void tun_net_uninit(struct net_device *dev)
821 /* Net device open. */
822 static int tun_net_open(struct net_device *dev)
824 netif_tx_start_all_queues(dev);
828 /* Net device close. */
829 static int tun_net_close(struct net_device *dev)
831 netif_tx_stop_all_queues(dev);
835 /* Net device start xmit */
836 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
838 struct tun_struct *tun = netdev_priv(dev);
839 int txq = skb->queue_mapping;
840 struct tun_file *tfile;
844 tfile = rcu_dereference(tun->tfiles[txq]);
845 numqueues = ACCESS_ONCE(tun->numqueues);
847 /* Drop packet if interface is not attached */
848 if (txq >= numqueues)
852 if (numqueues == 1 && static_key_false(&rps_needed)) {
853 /* Select queue was not called for the skbuff, so we extract the
854 * RPS hash and save it into the flow_table here.
858 rxhash = skb_get_hash(skb);
860 struct tun_flow_entry *e;
861 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
864 tun_flow_save_rps_rxhash(e, rxhash);
869 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
873 /* Drop if the filter does not like it.
874 * This is a noop if the filter is disabled.
875 * Filter can be enabled only for the TAP devices. */
876 if (!check_filter(&tun->txflt, skb))
879 if (tfile->socket.sk->sk_filter &&
880 sk_filter(tfile->socket.sk, skb))
883 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
886 skb_tx_timestamp(skb);
888 /* Orphan the skb - required as we might hang on to it
889 * for indefinite time.
895 if (skb_array_produce(&tfile->tx_array, skb))
898 /* Notify and wake up reader process */
899 if (tfile->flags & TUN_FASYNC)
900 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
901 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
907 this_cpu_inc(tun->pcpu_stats->tx_dropped);
911 return NET_XMIT_DROP;
914 static void tun_net_mclist(struct net_device *dev)
917 * This callback is supposed to deal with mc filter in
918 * _rx_ path and has nothing to do with the _tx_ path.
919 * In rx path we always accept everything userspace gives us.
923 static netdev_features_t tun_net_fix_features(struct net_device *dev,
924 netdev_features_t features)
926 struct tun_struct *tun = netdev_priv(dev);
928 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
930 #ifdef CONFIG_NET_POLL_CONTROLLER
931 static void tun_poll_controller(struct net_device *dev)
934 * Tun only receives frames when:
935 * 1) the char device endpoint gets data from user space
936 * 2) the tun socket gets a sendmsg call from user space
937 * Since both of those are synchronous operations, we are guaranteed
938 * never to have pending data when we poll for it
939 * so there is nothing to do here but return.
940 * We need this though so netpoll recognizes us as an interface that
941 * supports polling, which enables bridge devices in virt setups to
942 * still use netconsole
948 static void tun_set_headroom(struct net_device *dev, int new_hr)
950 struct tun_struct *tun = netdev_priv(dev);
952 if (new_hr < NET_SKB_PAD)
953 new_hr = NET_SKB_PAD;
959 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
961 u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
962 struct tun_struct *tun = netdev_priv(dev);
963 struct tun_pcpu_stats *p;
966 for_each_possible_cpu(i) {
967 u64 rxpackets, rxbytes, txpackets, txbytes;
970 p = per_cpu_ptr(tun->pcpu_stats, i);
972 start = u64_stats_fetch_begin(&p->syncp);
973 rxpackets = p->rx_packets;
974 rxbytes = p->rx_bytes;
975 txpackets = p->tx_packets;
976 txbytes = p->tx_bytes;
977 } while (u64_stats_fetch_retry(&p->syncp, start));
979 stats->rx_packets += rxpackets;
980 stats->rx_bytes += rxbytes;
981 stats->tx_packets += txpackets;
982 stats->tx_bytes += txbytes;
985 rx_dropped += p->rx_dropped;
986 rx_frame_errors += p->rx_frame_errors;
987 tx_dropped += p->tx_dropped;
989 stats->rx_dropped = rx_dropped;
990 stats->rx_frame_errors = rx_frame_errors;
991 stats->tx_dropped = tx_dropped;
994 static const struct net_device_ops tun_netdev_ops = {
995 .ndo_uninit = tun_net_uninit,
996 .ndo_open = tun_net_open,
997 .ndo_stop = tun_net_close,
998 .ndo_start_xmit = tun_net_xmit,
999 .ndo_fix_features = tun_net_fix_features,
1000 .ndo_select_queue = tun_select_queue,
1001 #ifdef CONFIG_NET_POLL_CONTROLLER
1002 .ndo_poll_controller = tun_poll_controller,
1004 .ndo_set_rx_headroom = tun_set_headroom,
1005 .ndo_get_stats64 = tun_net_get_stats64,
1008 static const struct net_device_ops tap_netdev_ops = {
1009 .ndo_uninit = tun_net_uninit,
1010 .ndo_open = tun_net_open,
1011 .ndo_stop = tun_net_close,
1012 .ndo_start_xmit = tun_net_xmit,
1013 .ndo_fix_features = tun_net_fix_features,
1014 .ndo_set_rx_mode = tun_net_mclist,
1015 .ndo_set_mac_address = eth_mac_addr,
1016 .ndo_validate_addr = eth_validate_addr,
1017 .ndo_select_queue = tun_select_queue,
1018 #ifdef CONFIG_NET_POLL_CONTROLLER
1019 .ndo_poll_controller = tun_poll_controller,
1021 .ndo_features_check = passthru_features_check,
1022 .ndo_set_rx_headroom = tun_set_headroom,
1023 .ndo_get_stats64 = tun_net_get_stats64,
1026 static void tun_flow_init(struct tun_struct *tun)
1030 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1031 INIT_HLIST_HEAD(&tun->flows[i]);
1033 tun->ageing_time = TUN_FLOW_EXPIRE;
1034 setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
1035 mod_timer(&tun->flow_gc_timer,
1036 round_jiffies_up(jiffies + tun->ageing_time));
1039 static void tun_flow_uninit(struct tun_struct *tun)
1041 del_timer_sync(&tun->flow_gc_timer);
1042 tun_flow_flush(tun);
1046 #define MAX_MTU 65535
1048 /* Initialize net device. */
1049 static void tun_net_init(struct net_device *dev)
1051 struct tun_struct *tun = netdev_priv(dev);
1053 switch (tun->flags & TUN_TYPE_MASK) {
1055 dev->netdev_ops = &tun_netdev_ops;
1057 /* Point-to-Point TUN Device */
1058 dev->hard_header_len = 0;
1062 /* Zero header length */
1063 dev->type = ARPHRD_NONE;
1064 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1068 dev->netdev_ops = &tap_netdev_ops;
1069 /* Ethernet TAP Device */
1071 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1072 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1074 eth_hw_addr_random(dev);
1079 dev->min_mtu = MIN_MTU;
1080 dev->max_mtu = MAX_MTU - dev->hard_header_len;
1083 /* Character device part */
1086 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1088 struct tun_file *tfile = file->private_data;
1089 struct tun_struct *tun = __tun_get(tfile);
1091 unsigned int mask = 0;
1096 sk = tfile->socket.sk;
1098 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1100 poll_wait(file, sk_sleep(sk), wait);
1102 if (!skb_array_empty(&tfile->tx_array))
1103 mask |= POLLIN | POLLRDNORM;
1105 if (sock_writeable(sk) ||
1106 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1107 sock_writeable(sk)))
1108 mask |= POLLOUT | POLLWRNORM;
1110 if (tun->dev->reg_state != NETREG_REGISTERED)
1117 /* prepad is the amount to reserve at front. len is length after that.
1118 * linear is a hint as to how much to copy (usually headers). */
1119 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1120 size_t prepad, size_t len,
1121 size_t linear, int noblock)
1123 struct sock *sk = tfile->socket.sk;
1124 struct sk_buff *skb;
1127 /* Under a page? Don't bother with paged skb. */
1128 if (prepad + len < PAGE_SIZE || !linear)
1131 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1134 return ERR_PTR(err);
1136 skb_reserve(skb, prepad);
1137 skb_put(skb, linear);
1138 skb->data_len = len - linear;
1139 skb->len += len - linear;
1144 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1145 struct sk_buff *skb, int more)
1147 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1148 struct sk_buff_head process_queue;
1149 u32 rx_batched = tun->rx_batched;
1152 if (!rx_batched || (!more && skb_queue_empty(queue))) {
1154 netif_receive_skb(skb);
1159 spin_lock(&queue->lock);
1160 if (!more || skb_queue_len(queue) == rx_batched) {
1161 __skb_queue_head_init(&process_queue);
1162 skb_queue_splice_tail_init(queue, &process_queue);
1165 __skb_queue_tail(queue, skb);
1167 spin_unlock(&queue->lock);
1170 struct sk_buff *nskb;
1173 while ((nskb = __skb_dequeue(&process_queue)))
1174 netif_receive_skb(nskb);
1175 netif_receive_skb(skb);
1180 /* Get packet from user space buffer */
1181 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1182 void *msg_control, struct iov_iter *from,
1183 int noblock, bool more)
1185 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1186 struct sk_buff *skb;
1187 size_t total_len = iov_iter_count(from);
1188 size_t len = total_len, align = tun->align, linear;
1189 struct virtio_net_hdr gso = { 0 };
1190 struct tun_pcpu_stats *stats;
1193 bool zerocopy = false;
1197 if (!(tun->dev->flags & IFF_UP))
1200 if (!(tun->flags & IFF_NO_PI)) {
1201 if (len < sizeof(pi))
1205 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1209 if (tun->flags & IFF_VNET_HDR) {
1210 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1212 if (len < vnet_hdr_sz)
1216 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1219 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1220 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1221 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1223 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1225 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1228 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1229 align += NET_IP_ALIGN;
1230 if (unlikely(len < ETH_HLEN ||
1231 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1235 good_linear = SKB_MAX_HEAD(align);
1238 struct iov_iter i = *from;
1240 /* There are 256 bytes to be copied in skb, so there is
1241 * enough room for skb expand head in case it is used.
1242 * The rest of the buffer is mapped from userspace.
1244 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1245 if (copylen > good_linear)
1246 copylen = good_linear;
1248 iov_iter_advance(&i, copylen);
1249 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1255 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1256 linear = good_linear;
1258 linear = tun16_to_cpu(tun, gso.hdr_len);
1261 skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1263 if (PTR_ERR(skb) != -EAGAIN)
1264 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1265 return PTR_ERR(skb);
1269 err = zerocopy_sg_from_iter(skb, from);
1271 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1274 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1279 if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1280 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1285 switch (tun->flags & TUN_TYPE_MASK) {
1287 if (tun->flags & IFF_NO_PI) {
1288 switch (skb->data[0] & 0xf0) {
1290 pi.proto = htons(ETH_P_IP);
1293 pi.proto = htons(ETH_P_IPV6);
1296 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1302 skb_reset_mac_header(skb);
1303 skb->protocol = pi.proto;
1304 skb->dev = tun->dev;
1307 skb->protocol = eth_type_trans(skb, tun->dev);
1311 /* copy skb_ubuf_info for callback when skb has no error */
1313 skb_shinfo(skb)->destructor_arg = msg_control;
1314 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1315 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1316 } else if (msg_control) {
1317 struct ubuf_info *uarg = msg_control;
1318 uarg->callback(uarg, false);
1321 skb_reset_network_header(skb);
1322 skb_probe_transport_header(skb, 0);
1324 rxhash = skb_get_hash(skb);
1325 #ifndef CONFIG_4KSTACKS
1326 tun_rx_batched(tun, tfile, skb, more);
1331 stats = get_cpu_ptr(tun->pcpu_stats);
1332 u64_stats_update_begin(&stats->syncp);
1333 stats->rx_packets++;
1334 stats->rx_bytes += len;
1335 u64_stats_update_end(&stats->syncp);
1338 tun_flow_update(tun, rxhash, tfile);
1342 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1344 struct file *file = iocb->ki_filp;
1345 struct tun_struct *tun = tun_get(file);
1346 struct tun_file *tfile = file->private_data;
1352 result = tun_get_user(tun, tfile, NULL, from,
1353 file->f_flags & O_NONBLOCK, false);
1359 /* Put packet to the user space buffer */
1360 static ssize_t tun_put_user(struct tun_struct *tun,
1361 struct tun_file *tfile,
1362 struct sk_buff *skb,
1363 struct iov_iter *iter)
1365 struct tun_pi pi = { 0, skb->protocol };
1366 struct tun_pcpu_stats *stats;
1368 int vlan_offset = 0;
1370 int vnet_hdr_sz = 0;
1372 if (skb_vlan_tag_present(skb))
1373 vlan_hlen = VLAN_HLEN;
1375 if (tun->flags & IFF_VNET_HDR)
1376 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1378 total = skb->len + vlan_hlen + vnet_hdr_sz;
1380 if (!(tun->flags & IFF_NO_PI)) {
1381 if (iov_iter_count(iter) < sizeof(pi))
1384 total += sizeof(pi);
1385 if (iov_iter_count(iter) < total) {
1386 /* Packet will be striped */
1387 pi.flags |= TUN_PKT_STRIP;
1390 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1395 struct virtio_net_hdr gso;
1397 if (iov_iter_count(iter) < vnet_hdr_sz)
1400 if (virtio_net_hdr_from_skb(skb, &gso,
1401 tun_is_little_endian(tun), true)) {
1402 struct skb_shared_info *sinfo = skb_shinfo(skb);
1403 pr_err("unexpected GSO type: "
1404 "0x%x, gso_size %d, hdr_len %d\n",
1405 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1406 tun16_to_cpu(tun, gso.hdr_len));
1407 print_hex_dump(KERN_ERR, "tun: ",
1410 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1415 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1418 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1424 __be16 h_vlan_proto;
1428 veth.h_vlan_proto = skb->vlan_proto;
1429 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1431 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1433 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1434 if (ret || !iov_iter_count(iter))
1437 ret = copy_to_iter(&veth, sizeof(veth), iter);
1438 if (ret != sizeof(veth) || !iov_iter_count(iter))
1442 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1445 /* caller is in process context, */
1446 stats = get_cpu_ptr(tun->pcpu_stats);
1447 u64_stats_update_begin(&stats->syncp);
1448 stats->tx_packets++;
1449 stats->tx_bytes += skb->len + vlan_hlen;
1450 u64_stats_update_end(&stats->syncp);
1451 put_cpu_ptr(tun->pcpu_stats);
1456 static struct sk_buff *tun_ring_recv(struct tun_file *tfile, int noblock,
1459 DECLARE_WAITQUEUE(wait, current);
1460 struct sk_buff *skb = NULL;
1463 skb = skb_array_consume(&tfile->tx_array);
1471 add_wait_queue(&tfile->wq.wait, &wait);
1472 current->state = TASK_INTERRUPTIBLE;
1475 skb = skb_array_consume(&tfile->tx_array);
1478 if (signal_pending(current)) {
1479 error = -ERESTARTSYS;
1482 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
1490 current->state = TASK_RUNNING;
1491 remove_wait_queue(&tfile->wq.wait, &wait);
1498 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1499 struct iov_iter *to,
1502 struct sk_buff *skb;
1506 tun_debug(KERN_INFO, tun, "tun_do_read\n");
1508 if (!iov_iter_count(to))
1511 /* Read frames from ring */
1512 skb = tun_ring_recv(tfile, noblock, &err);
1516 ret = tun_put_user(tun, tfile, skb, to);
1517 if (unlikely(ret < 0))
1525 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1527 struct file *file = iocb->ki_filp;
1528 struct tun_file *tfile = file->private_data;
1529 struct tun_struct *tun = __tun_get(tfile);
1530 ssize_t len = iov_iter_count(to), ret;
1534 ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK);
1535 ret = min_t(ssize_t, ret, len);
1542 static void tun_free_netdev(struct net_device *dev)
1544 struct tun_struct *tun = netdev_priv(dev);
1546 BUG_ON(!(list_empty(&tun->disabled)));
1547 free_percpu(tun->pcpu_stats);
1548 tun_flow_uninit(tun);
1549 security_tun_dev_free_security(tun->security);
1553 static void tun_setup(struct net_device *dev)
1555 struct tun_struct *tun = netdev_priv(dev);
1557 tun->owner = INVALID_UID;
1558 tun->group = INVALID_GID;
1560 dev->ethtool_ops = &tun_ethtool_ops;
1561 dev->destructor = tun_free_netdev;
1562 /* We prefer our own queue length */
1563 dev->tx_queue_len = TUN_READQ_SIZE;
1566 /* Trivial set of netlink ops to allow deleting tun or tap
1567 * device with netlink.
1569 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1574 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1576 .priv_size = sizeof(struct tun_struct),
1578 .validate = tun_validate,
1581 static void tun_sock_write_space(struct sock *sk)
1583 struct tun_file *tfile;
1584 wait_queue_head_t *wqueue;
1586 if (!sock_writeable(sk))
1589 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
1592 wqueue = sk_sleep(sk);
1593 if (wqueue && waitqueue_active(wqueue))
1594 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1595 POLLWRNORM | POLLWRBAND);
1597 tfile = container_of(sk, struct tun_file, sk);
1598 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1601 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1604 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1605 struct tun_struct *tun = __tun_get(tfile);
1610 ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1611 m->msg_flags & MSG_DONTWAIT,
1612 m->msg_flags & MSG_MORE);
1617 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1620 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1621 struct tun_struct *tun = __tun_get(tfile);
1627 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1631 if (flags & MSG_ERRQUEUE) {
1632 ret = sock_recv_errqueue(sock->sk, m, total_len,
1633 SOL_PACKET, TUN_TX_TIMESTAMP);
1636 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT);
1637 if (ret > (ssize_t)total_len) {
1638 m->msg_flags |= MSG_TRUNC;
1639 ret = flags & MSG_TRUNC ? ret : total_len;
1646 static int tun_peek_len(struct socket *sock)
1648 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1649 struct tun_struct *tun;
1652 tun = __tun_get(tfile);
1656 ret = skb_array_peek_len(&tfile->tx_array);
1662 /* Ops structure to mimic raw sockets with tun */
1663 static const struct proto_ops tun_socket_ops = {
1664 .peek_len = tun_peek_len,
1665 .sendmsg = tun_sendmsg,
1666 .recvmsg = tun_recvmsg,
1669 static struct proto tun_proto = {
1671 .owner = THIS_MODULE,
1672 .obj_size = sizeof(struct tun_file),
1675 static int tun_flags(struct tun_struct *tun)
1677 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1680 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1683 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1684 return sprintf(buf, "0x%x\n", tun_flags(tun));
1687 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1690 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1691 return uid_valid(tun->owner)?
1692 sprintf(buf, "%u\n",
1693 from_kuid_munged(current_user_ns(), tun->owner)):
1694 sprintf(buf, "-1\n");
1697 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1700 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1701 return gid_valid(tun->group) ?
1702 sprintf(buf, "%u\n",
1703 from_kgid_munged(current_user_ns(), tun->group)):
1704 sprintf(buf, "-1\n");
1707 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1708 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1709 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1711 static struct attribute *tun_dev_attrs[] = {
1712 &dev_attr_tun_flags.attr,
1713 &dev_attr_owner.attr,
1714 &dev_attr_group.attr,
1718 static const struct attribute_group tun_attr_group = {
1719 .attrs = tun_dev_attrs
1722 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1724 struct tun_struct *tun;
1725 struct tun_file *tfile = file->private_data;
1726 struct net_device *dev;
1729 if (tfile->detached)
1732 dev = __dev_get_by_name(net, ifr->ifr_name);
1734 if (ifr->ifr_flags & IFF_TUN_EXCL)
1736 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1737 tun = netdev_priv(dev);
1738 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1739 tun = netdev_priv(dev);
1743 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1744 !!(tun->flags & IFF_MULTI_QUEUE))
1747 if (tun_not_capable(tun))
1749 err = security_tun_dev_open(tun->security);
1753 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1757 if (tun->flags & IFF_MULTI_QUEUE &&
1758 (tun->numqueues + tun->numdisabled > 1)) {
1759 /* One or more queue has already been attached, no need
1760 * to initialize the device again.
1767 unsigned long flags = 0;
1768 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1771 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1773 err = security_tun_dev_create();
1778 if (ifr->ifr_flags & IFF_TUN) {
1782 } else if (ifr->ifr_flags & IFF_TAP) {
1790 name = ifr->ifr_name;
1792 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1793 NET_NAME_UNKNOWN, tun_setup, queues,
1799 dev_net_set(dev, net);
1800 dev->rtnl_link_ops = &tun_link_ops;
1801 dev->ifindex = tfile->ifindex;
1802 dev->sysfs_groups[0] = &tun_attr_group;
1804 tun = netdev_priv(dev);
1807 tun->txflt.count = 0;
1808 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1810 tun->align = NET_SKB_PAD;
1811 tun->filter_attached = false;
1812 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1813 tun->rx_batched = 0;
1815 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
1816 if (!tun->pcpu_stats) {
1821 spin_lock_init(&tun->lock);
1823 err = security_tun_dev_alloc_security(&tun->security);
1830 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1831 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1832 NETIF_F_HW_VLAN_STAG_TX;
1833 dev->features = dev->hw_features | NETIF_F_LLTX;
1834 dev->vlan_features = dev->features &
1835 ~(NETIF_F_HW_VLAN_CTAG_TX |
1836 NETIF_F_HW_VLAN_STAG_TX);
1838 INIT_LIST_HEAD(&tun->disabled);
1839 err = tun_attach(tun, file, false);
1843 err = register_netdevice(tun->dev);
1848 netif_carrier_on(tun->dev);
1850 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1852 tun->flags = (tun->flags & ~TUN_FEATURES) |
1853 (ifr->ifr_flags & TUN_FEATURES);
1855 /* Make sure persistent devices do not get stuck in
1858 if (netif_running(tun->dev))
1859 netif_tx_wake_all_queues(tun->dev);
1861 strcpy(ifr->ifr_name, tun->dev->name);
1865 tun_detach_all(dev);
1867 tun_flow_uninit(tun);
1868 security_tun_dev_free_security(tun->security);
1870 free_percpu(tun->pcpu_stats);
1876 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1879 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1881 strcpy(ifr->ifr_name, tun->dev->name);
1883 ifr->ifr_flags = tun_flags(tun);
1887 /* This is like a cut-down ethtool ops, except done via tun fd so no
1888 * privs required. */
1889 static int set_offload(struct tun_struct *tun, unsigned long arg)
1891 netdev_features_t features = 0;
1893 if (arg & TUN_F_CSUM) {
1894 features |= NETIF_F_HW_CSUM;
1897 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1898 if (arg & TUN_F_TSO_ECN) {
1899 features |= NETIF_F_TSO_ECN;
1900 arg &= ~TUN_F_TSO_ECN;
1902 if (arg & TUN_F_TSO4)
1903 features |= NETIF_F_TSO;
1904 if (arg & TUN_F_TSO6)
1905 features |= NETIF_F_TSO6;
1906 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1909 if (arg & TUN_F_UFO) {
1910 features |= NETIF_F_UFO;
1915 /* This gives the user a way to test for new features in future by
1916 * trying to set them. */
1920 tun->set_features = features;
1921 netdev_update_features(tun->dev);
1926 static void tun_detach_filter(struct tun_struct *tun, int n)
1929 struct tun_file *tfile;
1931 for (i = 0; i < n; i++) {
1932 tfile = rtnl_dereference(tun->tfiles[i]);
1933 lock_sock(tfile->socket.sk);
1934 sk_detach_filter(tfile->socket.sk);
1935 release_sock(tfile->socket.sk);
1938 tun->filter_attached = false;
1941 static int tun_attach_filter(struct tun_struct *tun)
1944 struct tun_file *tfile;
1946 for (i = 0; i < tun->numqueues; i++) {
1947 tfile = rtnl_dereference(tun->tfiles[i]);
1948 lock_sock(tfile->socket.sk);
1949 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1950 release_sock(tfile->socket.sk);
1952 tun_detach_filter(tun, i);
1957 tun->filter_attached = true;
1961 static void tun_set_sndbuf(struct tun_struct *tun)
1963 struct tun_file *tfile;
1966 for (i = 0; i < tun->numqueues; i++) {
1967 tfile = rtnl_dereference(tun->tfiles[i]);
1968 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1972 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1974 struct tun_file *tfile = file->private_data;
1975 struct tun_struct *tun;
1980 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1981 tun = tfile->detached;
1986 ret = security_tun_dev_attach_queue(tun->security);
1989 ret = tun_attach(tun, file, false);
1990 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1991 tun = rtnl_dereference(tfile->tun);
1992 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
1995 __tun_detach(tfile, false);
2004 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2005 unsigned long arg, int ifreq_len)
2007 struct tun_file *tfile = file->private_data;
2008 struct tun_struct *tun;
2009 void __user* argp = (void __user*)arg;
2015 unsigned int ifindex;
2019 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == SOCK_IOC_TYPE) {
2020 if (copy_from_user(&ifr, argp, ifreq_len))
2023 memset(&ifr, 0, sizeof(ifr));
2025 if (cmd == TUNGETFEATURES) {
2026 /* Currently this just means: "what IFF flags are valid?".
2027 * This is needed because we never checked for invalid flags on
2030 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2031 (unsigned int __user*)argp);
2032 } else if (cmd == TUNSETQUEUE)
2033 return tun_set_queue(file, &ifr);
2038 tun = __tun_get(tfile);
2039 if (cmd == TUNSETIFF) {
2044 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2046 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
2051 if (copy_to_user(argp, &ifr, ifreq_len))
2055 if (cmd == TUNSETIFINDEX) {
2061 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2065 tfile->ifindex = ifindex;
2073 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2078 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2080 if (tfile->detached)
2081 ifr.ifr_flags |= IFF_DETACH_QUEUE;
2082 if (!tfile->socket.sk->sk_filter)
2083 ifr.ifr_flags |= IFF_NOFILTER;
2085 if (copy_to_user(argp, &ifr, ifreq_len))
2090 /* Disable/Enable checksum */
2092 /* [unimplemented] */
2093 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2094 arg ? "disabled" : "enabled");
2098 /* Disable/Enable persist mode. Keep an extra reference to the
2099 * module to prevent the module being unprobed.
2101 if (arg && !(tun->flags & IFF_PERSIST)) {
2102 tun->flags |= IFF_PERSIST;
2103 __module_get(THIS_MODULE);
2105 if (!arg && (tun->flags & IFF_PERSIST)) {
2106 tun->flags &= ~IFF_PERSIST;
2107 module_put(THIS_MODULE);
2110 tun_debug(KERN_INFO, tun, "persist %s\n",
2111 arg ? "enabled" : "disabled");
2115 /* Set owner of the device */
2116 owner = make_kuid(current_user_ns(), arg);
2117 if (!uid_valid(owner)) {
2122 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2123 from_kuid(&init_user_ns, tun->owner));
2127 /* Set group of the device */
2128 group = make_kgid(current_user_ns(), arg);
2129 if (!gid_valid(group)) {
2134 tun_debug(KERN_INFO, tun, "group set to %u\n",
2135 from_kgid(&init_user_ns, tun->group));
2139 /* Only allow setting the type when the interface is down */
2140 if (tun->dev->flags & IFF_UP) {
2141 tun_debug(KERN_INFO, tun,
2142 "Linktype set failed because interface is up\n");
2145 tun->dev->type = (int) arg;
2146 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2158 ret = set_offload(tun, arg);
2161 case TUNSETTXFILTER:
2162 /* Can be set only for TAPs */
2164 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2166 ret = update_filter(&tun->txflt, (void __user *)arg);
2170 /* Get hw address */
2171 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2172 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2173 if (copy_to_user(argp, &ifr, ifreq_len))
2178 /* Set hw address */
2179 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2180 ifr.ifr_hwaddr.sa_data);
2182 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2186 sndbuf = tfile->socket.sk->sk_sndbuf;
2187 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2192 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2197 tun->sndbuf = sndbuf;
2198 tun_set_sndbuf(tun);
2201 case TUNGETVNETHDRSZ:
2202 vnet_hdr_sz = tun->vnet_hdr_sz;
2203 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2207 case TUNSETVNETHDRSZ:
2208 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2212 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2217 tun->vnet_hdr_sz = vnet_hdr_sz;
2221 le = !!(tun->flags & TUN_VNET_LE);
2222 if (put_user(le, (int __user *)argp))
2227 if (get_user(le, (int __user *)argp)) {
2232 tun->flags |= TUN_VNET_LE;
2234 tun->flags &= ~TUN_VNET_LE;
2238 ret = tun_get_vnet_be(tun, argp);
2242 ret = tun_set_vnet_be(tun, argp);
2245 case TUNATTACHFILTER:
2246 /* Can be set only for TAPs */
2248 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2251 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2254 ret = tun_attach_filter(tun);
2257 case TUNDETACHFILTER:
2258 /* Can be set only for TAPs */
2260 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2263 tun_detach_filter(tun, tun->numqueues);
2268 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2271 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2288 static long tun_chr_ioctl(struct file *file,
2289 unsigned int cmd, unsigned long arg)
2291 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2294 #ifdef CONFIG_COMPAT
2295 static long tun_chr_compat_ioctl(struct file *file,
2296 unsigned int cmd, unsigned long arg)
2301 case TUNSETTXFILTER:
2306 arg = (unsigned long)compat_ptr(arg);
2309 arg = (compat_ulong_t)arg;
2314 * compat_ifreq is shorter than ifreq, so we must not access beyond
2315 * the end of that structure. All fields that are used in this
2316 * driver are compatible though, we don't need to convert the
2319 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2321 #endif /* CONFIG_COMPAT */
2323 static int tun_chr_fasync(int fd, struct file *file, int on)
2325 struct tun_file *tfile = file->private_data;
2328 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2332 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2333 tfile->flags |= TUN_FASYNC;
2335 tfile->flags &= ~TUN_FASYNC;
2341 static int tun_chr_open(struct inode *inode, struct file * file)
2343 struct net *net = current->nsproxy->net_ns;
2344 struct tun_file *tfile;
2346 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2348 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2352 RCU_INIT_POINTER(tfile->tun, NULL);
2356 init_waitqueue_head(&tfile->wq.wait);
2357 RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2359 tfile->socket.file = file;
2360 tfile->socket.ops = &tun_socket_ops;
2362 sock_init_data(&tfile->socket, &tfile->sk);
2364 tfile->sk.sk_write_space = tun_sock_write_space;
2365 tfile->sk.sk_sndbuf = INT_MAX;
2367 file->private_data = tfile;
2368 INIT_LIST_HEAD(&tfile->next);
2370 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2375 static int tun_chr_close(struct inode *inode, struct file *file)
2377 struct tun_file *tfile = file->private_data;
2379 tun_detach(tfile, true);
2384 #ifdef CONFIG_PROC_FS
2385 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2387 struct tun_struct *tun;
2390 memset(&ifr, 0, sizeof(ifr));
2395 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2401 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2405 static const struct file_operations tun_fops = {
2406 .owner = THIS_MODULE,
2407 .llseek = no_llseek,
2408 .read_iter = tun_chr_read_iter,
2409 .write_iter = tun_chr_write_iter,
2410 .poll = tun_chr_poll,
2411 .unlocked_ioctl = tun_chr_ioctl,
2412 #ifdef CONFIG_COMPAT
2413 .compat_ioctl = tun_chr_compat_ioctl,
2415 .open = tun_chr_open,
2416 .release = tun_chr_close,
2417 .fasync = tun_chr_fasync,
2418 #ifdef CONFIG_PROC_FS
2419 .show_fdinfo = tun_chr_show_fdinfo,
2423 static struct miscdevice tun_miscdev = {
2426 .nodename = "net/tun",
2430 /* ethtool interface */
2432 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2435 cmd->advertising = 0;
2436 ethtool_cmd_speed_set(cmd, SPEED_10);
2437 cmd->duplex = DUPLEX_FULL;
2438 cmd->port = PORT_TP;
2439 cmd->phy_address = 0;
2440 cmd->transceiver = XCVR_INTERNAL;
2441 cmd->autoneg = AUTONEG_DISABLE;
2447 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2449 struct tun_struct *tun = netdev_priv(dev);
2451 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2452 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2454 switch (tun->flags & TUN_TYPE_MASK) {
2456 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2459 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2464 static u32 tun_get_msglevel(struct net_device *dev)
2467 struct tun_struct *tun = netdev_priv(dev);
2474 static void tun_set_msglevel(struct net_device *dev, u32 value)
2477 struct tun_struct *tun = netdev_priv(dev);
2482 static int tun_get_coalesce(struct net_device *dev,
2483 struct ethtool_coalesce *ec)
2485 struct tun_struct *tun = netdev_priv(dev);
2487 ec->rx_max_coalesced_frames = tun->rx_batched;
2492 static int tun_set_coalesce(struct net_device *dev,
2493 struct ethtool_coalesce *ec)
2495 struct tun_struct *tun = netdev_priv(dev);
2497 if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
2498 tun->rx_batched = NAPI_POLL_WEIGHT;
2500 tun->rx_batched = ec->rx_max_coalesced_frames;
2505 static const struct ethtool_ops tun_ethtool_ops = {
2506 .get_settings = tun_get_settings,
2507 .get_drvinfo = tun_get_drvinfo,
2508 .get_msglevel = tun_get_msglevel,
2509 .set_msglevel = tun_set_msglevel,
2510 .get_link = ethtool_op_get_link,
2511 .get_ts_info = ethtool_op_get_ts_info,
2512 .get_coalesce = tun_get_coalesce,
2513 .set_coalesce = tun_set_coalesce,
2516 static int tun_queue_resize(struct tun_struct *tun)
2518 struct net_device *dev = tun->dev;
2519 struct tun_file *tfile;
2520 struct skb_array **arrays;
2521 int n = tun->numqueues + tun->numdisabled;
2524 arrays = kmalloc(sizeof *arrays * n, GFP_KERNEL);
2528 for (i = 0; i < tun->numqueues; i++) {
2529 tfile = rtnl_dereference(tun->tfiles[i]);
2530 arrays[i] = &tfile->tx_array;
2532 list_for_each_entry(tfile, &tun->disabled, next)
2533 arrays[i++] = &tfile->tx_array;
2535 ret = skb_array_resize_multiple(arrays, n,
2536 dev->tx_queue_len, GFP_KERNEL);
2542 static int tun_device_event(struct notifier_block *unused,
2543 unsigned long event, void *ptr)
2545 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2546 struct tun_struct *tun = netdev_priv(dev);
2548 if (dev->rtnl_link_ops != &tun_link_ops)
2552 case NETDEV_CHANGE_TX_QUEUE_LEN:
2553 if (tun_queue_resize(tun))
2563 static struct notifier_block tun_notifier_block __read_mostly = {
2564 .notifier_call = tun_device_event,
2567 static int __init tun_init(void)
2571 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2572 pr_info("%s\n", DRV_COPYRIGHT);
2574 ret = rtnl_link_register(&tun_link_ops);
2576 pr_err("Can't register link_ops\n");
2580 ret = misc_register(&tun_miscdev);
2582 pr_err("Can't register misc device %d\n", TUN_MINOR);
2586 register_netdevice_notifier(&tun_notifier_block);
2589 rtnl_link_unregister(&tun_link_ops);
2594 static void tun_cleanup(void)
2596 misc_deregister(&tun_miscdev);
2597 rtnl_link_unregister(&tun_link_ops);
2598 unregister_netdevice_notifier(&tun_notifier_block);
2601 /* Get an underlying socket object from tun file. Returns error unless file is
2602 * attached to a device. The returned object works like a packet socket, it
2603 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
2604 * holding a reference to the file for as long as the socket is in use. */
2605 struct socket *tun_get_socket(struct file *file)
2607 struct tun_file *tfile;
2608 if (file->f_op != &tun_fops)
2609 return ERR_PTR(-EINVAL);
2610 tfile = file->private_data;
2612 return ERR_PTR(-EBADFD);
2613 return &tfile->socket;
2615 EXPORT_SYMBOL_GPL(tun_get_socket);
2617 module_init(tun_init);
2618 module_exit(tun_cleanup);
2619 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2620 MODULE_AUTHOR(DRV_COPYRIGHT);
2621 MODULE_LICENSE("GPL");
2622 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2623 MODULE_ALIAS("devname:net/tun");