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>
75 #include <linux/ieee802154.h>
76 #include <linux/if_ltalk.h>
77 #include <uapi/linux/if_fddi.h>
78 #include <uapi/linux/if_hippi.h>
79 #include <uapi/linux/if_fc.h>
82 #include <net/6lowpan.h>
84 #include <asm/uaccess.h>
86 /* Uncomment to enable debugging */
87 /* #define TUN_DEBUG 1 */
92 #define tun_debug(level, tun, fmt, args...) \
95 netdev_printk(level, tun->dev, fmt, ##args); \
97 #define DBG1(level, fmt, args...) \
100 printk(level fmt, ##args); \
103 #define tun_debug(level, tun, fmt, args...) \
106 netdev_printk(level, tun->dev, fmt, ##args); \
108 #define DBG1(level, fmt, args...) \
111 printk(level fmt, ##args); \
115 /* TUN device flags */
117 /* IFF_ATTACH_QUEUE is never stored in device flags,
118 * overload it to mean fasync when stored there.
120 #define TUN_FASYNC IFF_ATTACH_QUEUE
121 /* High bits in flags field are unused. */
122 #define TUN_VNET_LE 0x80000000
123 #define TUN_VNET_BE 0x40000000
125 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
127 #define GOODCOPY_LEN 128
129 #define FLT_EXACT_COUNT 8
131 unsigned int count; /* Number of addrs. Zero means disabled */
132 u32 mask[2]; /* Mask of the hashed addrs */
133 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
136 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
137 * to max number of VCPUs in guest. */
138 #define MAX_TAP_QUEUES 256
139 #define MAX_TAP_FLOWS 4096
141 #define TUN_FLOW_EXPIRE (3 * HZ)
143 struct tun_pcpu_stats {
148 struct u64_stats_sync syncp;
154 /* A tun_file connects an open character device to a tuntap netdevice. It
155 * also contains all socket related structures (except sock_fprog and tap_filter)
156 * to serve as one transmit queue for tuntap device. The sock_fprog and
157 * tap_filter were kept in tun_struct since they were used for filtering for the
158 * netdevice not for a specific queue (at least I didn't see the requirement for
162 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
163 * other can only be read while rcu_read_lock or rtnl_lock is held.
167 struct socket socket;
169 struct tun_struct __rcu *tun;
170 struct fasync_struct *fasync;
171 /* only used for fasnyc */
175 unsigned int ifindex;
177 struct list_head next;
178 struct tun_struct *detached;
179 struct skb_array tx_array;
182 struct tun_flow_entry {
183 struct hlist_node hash_link;
185 struct tun_struct *tun;
190 unsigned long updated;
193 #define TUN_NUM_FLOW_ENTRIES 1024
195 /* Since the socket were moved to tun_file, to preserve the behavior of persist
196 * device, socket filter, sndbuf and vnet header size were restore when the
197 * file were attached to a persist device.
200 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
201 unsigned int numqueues;
206 struct net_device *dev;
207 netdev_features_t set_features;
208 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
209 NETIF_F_TSO6|NETIF_F_UFO)
214 struct tap_filter txflt;
215 struct sock_fprog fprog;
216 /* protected by rtnl lock */
217 bool filter_attached;
222 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
223 struct timer_list flow_gc_timer;
224 unsigned long ageing_time;
225 unsigned int numdisabled;
226 struct list_head disabled;
229 struct tun_pcpu_stats __percpu *pcpu_stats;
232 #ifdef CONFIG_TUN_VNET_CROSS_LE
233 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
235 return tun->flags & TUN_VNET_BE ? false :
236 virtio_legacy_is_little_endian();
239 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
241 int be = !!(tun->flags & TUN_VNET_BE);
243 if (put_user(be, argp))
249 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
253 if (get_user(be, argp))
257 tun->flags |= TUN_VNET_BE;
259 tun->flags &= ~TUN_VNET_BE;
264 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
266 return virtio_legacy_is_little_endian();
269 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
274 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
278 #endif /* CONFIG_TUN_VNET_CROSS_LE */
280 static inline bool tun_is_little_endian(struct tun_struct *tun)
282 return tun->flags & TUN_VNET_LE ||
283 tun_legacy_is_little_endian(tun);
286 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
288 return __virtio16_to_cpu(tun_is_little_endian(tun), val);
291 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
293 return __cpu_to_virtio16(tun_is_little_endian(tun), val);
296 static inline u32 tun_hashfn(u32 rxhash)
298 return rxhash & 0x3ff;
301 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
303 struct tun_flow_entry *e;
305 hlist_for_each_entry_rcu(e, head, hash_link) {
306 if (e->rxhash == rxhash)
312 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
313 struct hlist_head *head,
314 u32 rxhash, u16 queue_index)
316 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
319 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
320 rxhash, queue_index);
321 e->updated = jiffies;
324 e->queue_index = queue_index;
326 hlist_add_head_rcu(&e->hash_link, head);
332 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
334 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
335 e->rxhash, e->queue_index);
336 hlist_del_rcu(&e->hash_link);
341 static void tun_flow_flush(struct tun_struct *tun)
345 spin_lock_bh(&tun->lock);
346 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
347 struct tun_flow_entry *e;
348 struct hlist_node *n;
350 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
351 tun_flow_delete(tun, e);
353 spin_unlock_bh(&tun->lock);
356 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
360 spin_lock_bh(&tun->lock);
361 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
362 struct tun_flow_entry *e;
363 struct hlist_node *n;
365 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
366 if (e->queue_index == queue_index)
367 tun_flow_delete(tun, e);
370 spin_unlock_bh(&tun->lock);
373 static void tun_flow_cleanup(unsigned long data)
375 struct tun_struct *tun = (struct tun_struct *)data;
376 unsigned long delay = tun->ageing_time;
377 unsigned long next_timer = jiffies + delay;
378 unsigned long count = 0;
381 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
383 spin_lock_bh(&tun->lock);
384 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
385 struct tun_flow_entry *e;
386 struct hlist_node *n;
388 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
389 unsigned long this_timer;
391 this_timer = e->updated + delay;
392 if (time_before_eq(this_timer, jiffies))
393 tun_flow_delete(tun, e);
394 else if (time_before(this_timer, next_timer))
395 next_timer = this_timer;
400 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
401 spin_unlock_bh(&tun->lock);
404 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
405 struct tun_file *tfile)
407 struct hlist_head *head;
408 struct tun_flow_entry *e;
409 unsigned long delay = tun->ageing_time;
410 u16 queue_index = tfile->queue_index;
415 head = &tun->flows[tun_hashfn(rxhash)];
419 /* We may get a very small possibility of OOO during switching, not
420 * worth to optimize.*/
421 if (tun->numqueues == 1 || tfile->detached)
424 e = tun_flow_find(head, rxhash);
426 /* TODO: keep queueing to old queue until it's empty? */
427 e->queue_index = queue_index;
428 e->updated = jiffies;
429 sock_rps_record_flow_hash(e->rps_rxhash);
431 spin_lock_bh(&tun->lock);
432 if (!tun_flow_find(head, rxhash) &&
433 tun->flow_count < MAX_TAP_FLOWS)
434 tun_flow_create(tun, head, rxhash, queue_index);
436 if (!timer_pending(&tun->flow_gc_timer))
437 mod_timer(&tun->flow_gc_timer,
438 round_jiffies_up(jiffies + delay));
439 spin_unlock_bh(&tun->lock);
447 * Save the hash received in the stack receive path and update the
448 * flow_hash table accordingly.
450 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
452 if (unlikely(e->rps_rxhash != hash))
453 e->rps_rxhash = hash;
456 /* We try to identify a flow through its rxhash first. The reason that
457 * we do not check rxq no. is because some cards(e.g 82599), chooses
458 * the rxq based on the txq where the last packet of the flow comes. As
459 * the userspace application move between processors, we may get a
460 * different rxq no. here. If we could not get rxhash, then we would
461 * hope the rxq no. may help here.
463 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
464 void *accel_priv, select_queue_fallback_t fallback)
466 struct tun_struct *tun = netdev_priv(dev);
467 struct tun_flow_entry *e;
472 numqueues = ACCESS_ONCE(tun->numqueues);
474 txq = skb_get_hash(skb);
476 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
478 tun_flow_save_rps_rxhash(e, txq);
479 txq = e->queue_index;
481 /* use multiply and shift instead of expensive divide */
482 txq = ((u64)txq * numqueues) >> 32;
483 } else if (likely(skb_rx_queue_recorded(skb))) {
484 txq = skb_get_rx_queue(skb);
485 while (unlikely(txq >= numqueues))
493 static inline bool tun_not_capable(struct tun_struct *tun)
495 const struct cred *cred = current_cred();
496 struct net *net = dev_net(tun->dev);
498 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
499 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
500 !ns_capable(net->user_ns, CAP_NET_ADMIN);
503 static void tun_set_real_num_queues(struct tun_struct *tun)
505 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
506 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
509 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
511 tfile->detached = tun;
512 list_add_tail(&tfile->next, &tun->disabled);
516 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
518 struct tun_struct *tun = tfile->detached;
520 tfile->detached = NULL;
521 list_del_init(&tfile->next);
526 static void tun_queue_purge(struct tun_file *tfile)
530 while ((skb = skb_array_consume(&tfile->tx_array)) != NULL)
533 skb_queue_purge(&tfile->sk.sk_error_queue);
536 static void tun_cleanup_tx_array(struct tun_file *tfile)
538 if (tfile->tx_array.ring.queue) {
539 skb_array_cleanup(&tfile->tx_array);
540 memset(&tfile->tx_array, 0, sizeof(tfile->tx_array));
544 static void __tun_detach(struct tun_file *tfile, bool clean)
546 struct tun_file *ntfile;
547 struct tun_struct *tun;
549 tun = rtnl_dereference(tfile->tun);
551 if (tun && !tfile->detached) {
552 u16 index = tfile->queue_index;
553 BUG_ON(index >= tun->numqueues);
555 rcu_assign_pointer(tun->tfiles[index],
556 tun->tfiles[tun->numqueues - 1]);
557 ntfile = rtnl_dereference(tun->tfiles[index]);
558 ntfile->queue_index = index;
562 RCU_INIT_POINTER(tfile->tun, NULL);
563 sock_put(&tfile->sk);
565 tun_disable_queue(tun, tfile);
568 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
569 /* Drop read queue */
570 tun_queue_purge(tfile);
571 tun_set_real_num_queues(tun);
572 } else if (tfile->detached && clean) {
573 tun = tun_enable_queue(tfile);
574 sock_put(&tfile->sk);
578 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
579 netif_carrier_off(tun->dev);
581 if (!(tun->flags & IFF_PERSIST) &&
582 tun->dev->reg_state == NETREG_REGISTERED)
583 unregister_netdevice(tun->dev);
585 tun_cleanup_tx_array(tfile);
586 sock_put(&tfile->sk);
590 static void tun_detach(struct tun_file *tfile, bool clean)
593 __tun_detach(tfile, clean);
597 static void tun_detach_all(struct net_device *dev)
599 struct tun_struct *tun = netdev_priv(dev);
600 struct tun_file *tfile, *tmp;
601 int i, n = tun->numqueues;
603 for (i = 0; i < n; i++) {
604 tfile = rtnl_dereference(tun->tfiles[i]);
606 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
607 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
608 RCU_INIT_POINTER(tfile->tun, NULL);
611 list_for_each_entry(tfile, &tun->disabled, next) {
612 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
613 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
614 RCU_INIT_POINTER(tfile->tun, NULL);
616 BUG_ON(tun->numqueues != 0);
619 for (i = 0; i < n; i++) {
620 tfile = rtnl_dereference(tun->tfiles[i]);
621 /* Drop read queue */
622 tun_queue_purge(tfile);
623 sock_put(&tfile->sk);
624 tun_cleanup_tx_array(tfile);
626 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
627 tun_enable_queue(tfile);
628 tun_queue_purge(tfile);
629 sock_put(&tfile->sk);
630 tun_cleanup_tx_array(tfile);
632 BUG_ON(tun->numdisabled != 0);
634 if (tun->flags & IFF_PERSIST)
635 module_put(THIS_MODULE);
638 static int tun_attach(struct tun_struct *tun, struct file *file,
639 bool skip_filter, bool publish_tun)
641 struct tun_file *tfile = file->private_data;
642 struct net_device *dev = tun->dev;
645 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
650 if (rtnl_dereference(tfile->tun) && !tfile->detached)
654 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
658 if (!tfile->detached &&
659 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
664 /* Re-attach the filter to persist device */
665 if (!skip_filter && (tun->filter_attached == true)) {
666 lock_sock(tfile->socket.sk);
667 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
668 release_sock(tfile->socket.sk);
673 if (!tfile->detached &&
674 skb_array_init(&tfile->tx_array, dev->tx_queue_len, GFP_KERNEL)) {
679 tfile->queue_index = tun->numqueues;
680 tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
682 rcu_assign_pointer(tfile->tun, tun);
683 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
687 tun_enable_queue(tfile);
689 sock_hold(&tfile->sk);
691 tun_set_real_num_queues(tun);
693 /* device is allowed to go away first, so no need to hold extra
701 static struct tun_struct *__tun_get(struct tun_file *tfile)
703 struct tun_struct *tun;
706 tun = rcu_dereference(tfile->tun);
714 static struct tun_struct *tun_get(struct file *file)
716 return __tun_get(file->private_data);
719 static void tun_put(struct tun_struct *tun)
725 static void addr_hash_set(u32 *mask, const u8 *addr)
727 int n = ether_crc(ETH_ALEN, addr) >> 26;
728 mask[n >> 5] |= (1 << (n & 31));
731 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
733 int n = ether_crc(ETH_ALEN, addr) >> 26;
734 return mask[n >> 5] & (1 << (n & 31));
737 static int update_filter(struct tap_filter *filter, void __user *arg)
739 struct { u8 u[ETH_ALEN]; } *addr;
740 struct tun_filter uf;
741 int err, alen, n, nexact;
743 if (copy_from_user(&uf, arg, sizeof(uf)))
752 alen = ETH_ALEN * uf.count;
753 addr = memdup_user(arg + sizeof(uf), alen);
755 return PTR_ERR(addr);
757 /* The filter is updated without holding any locks. Which is
758 * perfectly safe. We disable it first and in the worst
759 * case we'll accept a few undesired packets. */
763 /* Use first set of addresses as an exact filter */
764 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
765 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
769 /* Remaining multicast addresses are hashed,
770 * unicast will leave the filter disabled. */
771 memset(filter->mask, 0, sizeof(filter->mask));
772 for (; n < uf.count; n++) {
773 if (!is_multicast_ether_addr(addr[n].u)) {
774 err = 0; /* no filter */
777 addr_hash_set(filter->mask, addr[n].u);
780 /* For ALLMULTI just set the mask to all ones.
781 * This overrides the mask populated above. */
782 if ((uf.flags & TUN_FLT_ALLMULTI))
783 memset(filter->mask, ~0, sizeof(filter->mask));
785 /* Now enable the filter */
787 filter->count = nexact;
789 /* Return the number of exact filters */
796 /* Returns: 0 - drop, !=0 - accept */
797 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
799 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
801 struct ethhdr *eh = (struct ethhdr *) skb->data;
805 for (i = 0; i < filter->count; i++)
806 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
809 /* Inexact match (multicast only) */
810 if (is_multicast_ether_addr(eh->h_dest))
811 return addr_hash_test(filter->mask, eh->h_dest);
817 * Checks whether the packet is accepted or not.
818 * Returns: 0 - drop, !=0 - accept
820 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
825 return run_filter(filter, skb);
828 /* Network device part of the driver */
830 static const struct ethtool_ops tun_ethtool_ops;
832 /* Net device detach from fd. */
833 static void tun_net_uninit(struct net_device *dev)
838 /* Net device open. */
839 static int tun_net_open(struct net_device *dev)
841 netif_tx_start_all_queues(dev);
846 /* Net device close. */
847 static int tun_net_close(struct net_device *dev)
849 netif_tx_stop_all_queues(dev);
853 /* Net device start xmit */
854 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
856 struct tun_struct *tun = netdev_priv(dev);
857 int txq = skb->queue_mapping;
858 struct netdev_queue *queue;
859 struct tun_file *tfile;
863 tfile = rcu_dereference(tun->tfiles[txq]);
864 numqueues = ACCESS_ONCE(tun->numqueues);
866 /* Drop packet if interface is not attached */
867 if (txq >= numqueues)
871 if (numqueues == 1 && static_key_false(&rps_needed)) {
872 /* Select queue was not called for the skbuff, so we extract the
873 * RPS hash and save it into the flow_table here.
877 rxhash = skb_get_hash(skb);
879 struct tun_flow_entry *e;
880 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
883 tun_flow_save_rps_rxhash(e, rxhash);
888 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
892 /* Drop if the filter does not like it.
893 * This is a noop if the filter is disabled.
894 * Filter can be enabled only for the TAP devices. */
895 if (!check_filter(&tun->txflt, skb))
898 if (tfile->socket.sk->sk_filter &&
899 sk_filter(tfile->socket.sk, skb))
902 /* Limit the number of packets queued by dividing txq length with the
905 if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
906 >= dev->tx_queue_len)
909 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
912 skb_tx_timestamp(skb);
914 /* Orphan the skb - required as we might hang on to it
915 * for indefinite time.
921 if (skb_array_produce(&tfile->tx_array, skb))
924 /* NETIF_F_LLTX requires to do our own update of trans_start */
925 queue = netdev_get_tx_queue(dev, txq);
926 queue->trans_start = jiffies;
928 /* Notify and wake up reader process */
929 if (tfile->flags & TUN_FASYNC)
930 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
931 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
937 this_cpu_inc(tun->pcpu_stats->tx_dropped);
941 return NET_XMIT_DROP;
944 static void tun_net_mclist(struct net_device *dev)
947 * This callback is supposed to deal with mc filter in
948 * _rx_ path and has nothing to do with the _tx_ path.
949 * In rx path we always accept everything userspace gives us.
954 #define MAX_MTU 65535
957 tun_net_change_mtu(struct net_device *dev, int new_mtu)
959 if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
965 static netdev_features_t tun_net_fix_features(struct net_device *dev,
966 netdev_features_t features)
968 struct tun_struct *tun = netdev_priv(dev);
970 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
972 #ifdef CONFIG_NET_POLL_CONTROLLER
973 static void tun_poll_controller(struct net_device *dev)
976 * Tun only receives frames when:
977 * 1) the char device endpoint gets data from user space
978 * 2) the tun socket gets a sendmsg call from user space
979 * Since both of those are synchronous operations, we are guaranteed
980 * never to have pending data when we poll for it
981 * so there is nothing to do here but return.
982 * We need this though so netpoll recognizes us as an interface that
983 * supports polling, which enables bridge devices in virt setups to
984 * still use netconsole
990 static void tun_set_headroom(struct net_device *dev, int new_hr)
992 struct tun_struct *tun = netdev_priv(dev);
994 if (new_hr < NET_SKB_PAD)
995 new_hr = NET_SKB_PAD;
1000 static struct rtnl_link_stats64 *
1001 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1003 u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1004 struct tun_struct *tun = netdev_priv(dev);
1005 struct tun_pcpu_stats *p;
1008 for_each_possible_cpu(i) {
1009 u64 rxpackets, rxbytes, txpackets, txbytes;
1012 p = per_cpu_ptr(tun->pcpu_stats, i);
1014 start = u64_stats_fetch_begin(&p->syncp);
1015 rxpackets = p->rx_packets;
1016 rxbytes = p->rx_bytes;
1017 txpackets = p->tx_packets;
1018 txbytes = p->tx_bytes;
1019 } while (u64_stats_fetch_retry(&p->syncp, start));
1021 stats->rx_packets += rxpackets;
1022 stats->rx_bytes += rxbytes;
1023 stats->tx_packets += txpackets;
1024 stats->tx_bytes += txbytes;
1027 rx_dropped += p->rx_dropped;
1028 rx_frame_errors += p->rx_frame_errors;
1029 tx_dropped += p->tx_dropped;
1031 stats->rx_dropped = rx_dropped;
1032 stats->rx_frame_errors = rx_frame_errors;
1033 stats->tx_dropped = tx_dropped;
1037 static const struct net_device_ops tun_netdev_ops = {
1038 .ndo_uninit = tun_net_uninit,
1039 .ndo_open = tun_net_open,
1040 .ndo_stop = tun_net_close,
1041 .ndo_start_xmit = tun_net_xmit,
1042 .ndo_change_mtu = tun_net_change_mtu,
1043 .ndo_fix_features = tun_net_fix_features,
1044 .ndo_select_queue = tun_select_queue,
1045 #ifdef CONFIG_NET_POLL_CONTROLLER
1046 .ndo_poll_controller = tun_poll_controller,
1048 .ndo_set_rx_headroom = tun_set_headroom,
1049 .ndo_get_stats64 = tun_net_get_stats64,
1052 static const struct net_device_ops tap_netdev_ops = {
1053 .ndo_uninit = tun_net_uninit,
1054 .ndo_open = tun_net_open,
1055 .ndo_stop = tun_net_close,
1056 .ndo_start_xmit = tun_net_xmit,
1057 .ndo_change_mtu = tun_net_change_mtu,
1058 .ndo_fix_features = tun_net_fix_features,
1059 .ndo_set_rx_mode = tun_net_mclist,
1060 .ndo_set_mac_address = eth_mac_addr,
1061 .ndo_validate_addr = eth_validate_addr,
1062 .ndo_select_queue = tun_select_queue,
1063 #ifdef CONFIG_NET_POLL_CONTROLLER
1064 .ndo_poll_controller = tun_poll_controller,
1066 .ndo_features_check = passthru_features_check,
1067 .ndo_set_rx_headroom = tun_set_headroom,
1068 .ndo_get_stats64 = tun_net_get_stats64,
1071 static void tun_flow_init(struct tun_struct *tun)
1075 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1076 INIT_HLIST_HEAD(&tun->flows[i]);
1078 tun->ageing_time = TUN_FLOW_EXPIRE;
1079 setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
1080 mod_timer(&tun->flow_gc_timer,
1081 round_jiffies_up(jiffies + tun->ageing_time));
1084 static void tun_flow_uninit(struct tun_struct *tun)
1086 del_timer_sync(&tun->flow_gc_timer);
1087 tun_flow_flush(tun);
1090 /* Initialize net device. */
1091 static void tun_net_init(struct net_device *dev)
1093 struct tun_struct *tun = netdev_priv(dev);
1095 switch (tun->flags & TUN_TYPE_MASK) {
1097 dev->netdev_ops = &tun_netdev_ops;
1099 /* Point-to-Point TUN Device */
1100 dev->hard_header_len = 0;
1104 /* Zero header length */
1105 dev->type = ARPHRD_NONE;
1106 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1110 dev->netdev_ops = &tap_netdev_ops;
1111 /* Ethernet TAP Device */
1113 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1114 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1116 eth_hw_addr_random(dev);
1122 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1124 struct sock *sk = tfile->socket.sk;
1126 return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1129 /* Character device part */
1132 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1134 struct tun_file *tfile = file->private_data;
1135 struct tun_struct *tun = __tun_get(tfile);
1137 unsigned int mask = 0;
1142 sk = tfile->socket.sk;
1144 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1146 poll_wait(file, sk_sleep(sk), wait);
1148 if (!skb_array_empty(&tfile->tx_array))
1149 mask |= POLLIN | POLLRDNORM;
1151 /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1152 * guarantee EPOLLOUT to be raised by either here or
1153 * tun_sock_write_space(). Then process could get notification
1154 * after it writes to a down device and meets -EIO.
1156 if (tun_sock_writeable(tun, tfile) ||
1157 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1158 tun_sock_writeable(tun, tfile)))
1159 mask |= POLLOUT | POLLWRNORM;
1161 if (tun->dev->reg_state != NETREG_REGISTERED)
1168 /* prepad is the amount to reserve at front. len is length after that.
1169 * linear is a hint as to how much to copy (usually headers). */
1170 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1171 size_t prepad, size_t len,
1172 size_t linear, int noblock)
1174 struct sock *sk = tfile->socket.sk;
1175 struct sk_buff *skb;
1178 /* Under a page? Don't bother with paged skb. */
1179 if (prepad + len < PAGE_SIZE || !linear)
1182 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1185 return ERR_PTR(err);
1187 skb_reserve(skb, prepad);
1188 skb_put(skb, linear);
1189 skb->data_len = len - linear;
1190 skb->len += len - linear;
1195 /* Get packet from user space buffer */
1196 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1197 void *msg_control, struct iov_iter *from,
1200 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1201 struct sk_buff *skb;
1202 size_t total_len = iov_iter_count(from);
1203 size_t len = total_len, align = tun->align, linear;
1204 struct virtio_net_hdr gso = { 0 };
1205 struct tun_pcpu_stats *stats;
1208 bool zerocopy = false;
1213 if (!(tun->flags & IFF_NO_PI)) {
1214 if (len < sizeof(pi))
1218 n = copy_from_iter(&pi, sizeof(pi), from);
1219 if (n != sizeof(pi))
1223 if (tun->flags & IFF_VNET_HDR) {
1224 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1226 if (len < vnet_hdr_sz)
1230 n = copy_from_iter(&gso, sizeof(gso), from);
1231 if (n != sizeof(gso))
1234 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1235 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1236 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1238 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1240 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1243 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1244 align += NET_IP_ALIGN;
1245 if (unlikely(len < ETH_HLEN ||
1246 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1250 good_linear = SKB_MAX_HEAD(align);
1253 struct iov_iter i = *from;
1255 /* There are 256 bytes to be copied in skb, so there is
1256 * enough room for skb expand head in case it is used.
1257 * The rest of the buffer is mapped from userspace.
1259 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1260 if (copylen > good_linear)
1261 copylen = good_linear;
1263 iov_iter_advance(&i, copylen);
1264 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1270 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1271 linear = good_linear;
1273 linear = tun16_to_cpu(tun, gso.hdr_len);
1276 skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1278 if (PTR_ERR(skb) != -EAGAIN)
1279 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1280 return PTR_ERR(skb);
1284 err = zerocopy_sg_from_iter(skb, from);
1286 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1291 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1296 err = virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun));
1298 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1303 switch (tun->flags & TUN_TYPE_MASK) {
1305 if (tun->flags & IFF_NO_PI) {
1306 u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1308 switch (ip_version) {
1310 pi.proto = htons(ETH_P_IP);
1313 pi.proto = htons(ETH_P_IPV6);
1316 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1322 skb_reset_mac_header(skb);
1323 skb->protocol = pi.proto;
1324 skb->dev = tun->dev;
1327 skb->protocol = eth_type_trans(skb, tun->dev);
1331 /* copy skb_ubuf_info for callback when skb has no error */
1333 skb_shinfo(skb)->destructor_arg = msg_control;
1334 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1335 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1336 } else if (msg_control) {
1337 struct ubuf_info *uarg = msg_control;
1338 uarg->callback(uarg, false);
1341 skb_reset_network_header(skb);
1342 skb_probe_transport_header(skb, 0);
1344 rxhash = skb_get_hash(skb);
1347 if (unlikely(!(tun->dev->flags & IFF_UP))) {
1356 stats = get_cpu_ptr(tun->pcpu_stats);
1357 u64_stats_update_begin(&stats->syncp);
1358 stats->rx_packets++;
1359 stats->rx_bytes += len;
1360 u64_stats_update_end(&stats->syncp);
1363 tun_flow_update(tun, rxhash, tfile);
1367 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1369 struct file *file = iocb->ki_filp;
1370 struct tun_struct *tun = tun_get(file);
1371 struct tun_file *tfile = file->private_data;
1377 result = tun_get_user(tun, tfile, NULL, from, file->f_flags & O_NONBLOCK);
1383 /* Put packet to the user space buffer */
1384 static ssize_t tun_put_user(struct tun_struct *tun,
1385 struct tun_file *tfile,
1386 struct sk_buff *skb,
1387 struct iov_iter *iter)
1389 struct tun_pi pi = { 0, skb->protocol };
1390 struct tun_pcpu_stats *stats;
1392 int vlan_offset = 0;
1394 int vnet_hdr_sz = 0;
1396 if (skb_vlan_tag_present(skb))
1397 vlan_hlen = VLAN_HLEN;
1399 if (tun->flags & IFF_VNET_HDR)
1400 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1402 total = skb->len + vlan_hlen + vnet_hdr_sz;
1404 if (!(tun->flags & IFF_NO_PI)) {
1405 if (iov_iter_count(iter) < sizeof(pi))
1408 total += sizeof(pi);
1409 if (iov_iter_count(iter) < total) {
1410 /* Packet will be striped */
1411 pi.flags |= TUN_PKT_STRIP;
1414 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1419 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1422 if (iov_iter_count(iter) < vnet_hdr_sz)
1425 ret = virtio_net_hdr_from_skb(skb, &gso,
1426 tun_is_little_endian(tun), true);
1428 struct skb_shared_info *sinfo = skb_shinfo(skb);
1429 pr_err("unexpected GSO type: "
1430 "0x%x, gso_size %d, hdr_len %d\n",
1431 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1432 tun16_to_cpu(tun, gso.hdr_len));
1433 print_hex_dump(KERN_ERR, "tun: ",
1436 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1441 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1444 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1450 __be16 h_vlan_proto;
1454 veth.h_vlan_proto = skb->vlan_proto;
1455 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1457 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1459 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1460 if (ret || !iov_iter_count(iter))
1463 ret = copy_to_iter(&veth, sizeof(veth), iter);
1464 if (ret != sizeof(veth) || !iov_iter_count(iter))
1468 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1471 /* caller is in process context, */
1472 stats = get_cpu_ptr(tun->pcpu_stats);
1473 u64_stats_update_begin(&stats->syncp);
1474 stats->tx_packets++;
1475 stats->tx_bytes += skb->len + vlan_hlen;
1476 u64_stats_update_end(&stats->syncp);
1477 put_cpu_ptr(tun->pcpu_stats);
1482 static struct sk_buff *tun_ring_recv(struct tun_file *tfile, int noblock,
1485 DECLARE_WAITQUEUE(wait, current);
1486 struct sk_buff *skb = NULL;
1489 skb = skb_array_consume(&tfile->tx_array);
1497 add_wait_queue(&tfile->wq.wait, &wait);
1500 set_current_state(TASK_INTERRUPTIBLE);
1501 skb = skb_array_consume(&tfile->tx_array);
1504 if (signal_pending(current)) {
1505 error = -ERESTARTSYS;
1508 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
1516 __set_current_state(TASK_RUNNING);
1517 remove_wait_queue(&tfile->wq.wait, &wait);
1524 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1525 struct iov_iter *to,
1528 struct sk_buff *skb;
1532 tun_debug(KERN_INFO, tun, "tun_do_read\n");
1534 if (!iov_iter_count(to))
1537 /* Read frames from ring */
1538 skb = tun_ring_recv(tfile, noblock, &err);
1542 ret = tun_put_user(tun, tfile, skb, to);
1543 if (unlikely(ret < 0))
1551 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1553 struct file *file = iocb->ki_filp;
1554 struct tun_file *tfile = file->private_data;
1555 struct tun_struct *tun = __tun_get(tfile);
1556 ssize_t len = iov_iter_count(to), ret;
1560 ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK);
1561 ret = min_t(ssize_t, ret, len);
1568 static void tun_free_netdev(struct net_device *dev)
1570 struct tun_struct *tun = netdev_priv(dev);
1572 BUG_ON(!(list_empty(&tun->disabled)));
1573 free_percpu(tun->pcpu_stats);
1574 tun_flow_uninit(tun);
1575 security_tun_dev_free_security(tun->security);
1579 static void tun_setup(struct net_device *dev)
1581 struct tun_struct *tun = netdev_priv(dev);
1583 tun->owner = INVALID_UID;
1584 tun->group = INVALID_GID;
1586 dev->ethtool_ops = &tun_ethtool_ops;
1587 dev->destructor = tun_free_netdev;
1588 /* We prefer our own queue length */
1589 dev->tx_queue_len = TUN_READQ_SIZE;
1592 /* Trivial set of netlink ops to allow deleting tun or tap
1593 * device with netlink.
1595 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1597 /* NL_SET_ERR_MSG(extack,
1598 "tun/tap creation via rtnetlink is not supported."); */
1602 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1604 .priv_size = sizeof(struct tun_struct),
1606 .validate = tun_validate,
1609 static void tun_sock_write_space(struct sock *sk)
1611 struct tun_file *tfile;
1612 wait_queue_head_t *wqueue;
1614 if (!sock_writeable(sk))
1617 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
1620 wqueue = sk_sleep(sk);
1621 if (wqueue && waitqueue_active(wqueue))
1622 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1623 POLLWRNORM | POLLWRBAND);
1625 tfile = container_of(sk, struct tun_file, sk);
1626 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1629 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1632 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1633 struct tun_struct *tun = __tun_get(tfile);
1638 ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1639 m->msg_flags & MSG_DONTWAIT);
1644 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1647 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1648 struct tun_struct *tun = __tun_get(tfile);
1654 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1658 if (flags & MSG_ERRQUEUE) {
1659 ret = sock_recv_errqueue(sock->sk, m, total_len,
1660 SOL_PACKET, TUN_TX_TIMESTAMP);
1663 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT);
1664 if (ret > (ssize_t)total_len) {
1665 m->msg_flags |= MSG_TRUNC;
1666 ret = flags & MSG_TRUNC ? ret : total_len;
1673 static int tun_peek_len(struct socket *sock)
1675 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1676 struct tun_struct *tun;
1679 tun = __tun_get(tfile);
1683 ret = skb_array_peek_len(&tfile->tx_array);
1689 /* Ops structure to mimic raw sockets with tun */
1690 static const struct proto_ops tun_socket_ops = {
1691 .peek_len = tun_peek_len,
1692 .sendmsg = tun_sendmsg,
1693 .recvmsg = tun_recvmsg,
1696 static struct proto tun_proto = {
1698 .owner = THIS_MODULE,
1699 .obj_size = sizeof(struct tun_file),
1702 static int tun_flags(struct tun_struct *tun)
1704 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1707 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1710 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1711 return sprintf(buf, "0x%x\n", tun_flags(tun));
1714 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1717 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1718 return uid_valid(tun->owner)?
1719 sprintf(buf, "%u\n",
1720 from_kuid_munged(current_user_ns(), tun->owner)):
1721 sprintf(buf, "-1\n");
1724 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1727 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1728 return gid_valid(tun->group) ?
1729 sprintf(buf, "%u\n",
1730 from_kgid_munged(current_user_ns(), tun->group)):
1731 sprintf(buf, "-1\n");
1734 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1735 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1736 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1738 static struct attribute *tun_dev_attrs[] = {
1739 &dev_attr_tun_flags.attr,
1740 &dev_attr_owner.attr,
1741 &dev_attr_group.attr,
1745 static const struct attribute_group tun_attr_group = {
1746 .attrs = tun_dev_attrs
1749 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1751 struct tun_struct *tun;
1752 struct tun_file *tfile = file->private_data;
1753 struct net_device *dev;
1756 if (tfile->detached)
1759 dev = __dev_get_by_name(net, ifr->ifr_name);
1761 if (ifr->ifr_flags & IFF_TUN_EXCL)
1763 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1764 tun = netdev_priv(dev);
1765 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1766 tun = netdev_priv(dev);
1770 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1771 !!(tun->flags & IFF_MULTI_QUEUE))
1774 if (tun_not_capable(tun))
1776 err = security_tun_dev_open(tun->security);
1780 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER, true);
1784 if (tun->flags & IFF_MULTI_QUEUE &&
1785 (tun->numqueues + tun->numdisabled > 1)) {
1786 /* One or more queue has already been attached, no need
1787 * to initialize the device again.
1794 unsigned long flags = 0;
1795 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1798 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1800 err = security_tun_dev_create();
1805 if (ifr->ifr_flags & IFF_TUN) {
1809 } else if (ifr->ifr_flags & IFF_TAP) {
1817 name = ifr->ifr_name;
1819 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1820 NET_NAME_UNKNOWN, tun_setup, queues,
1825 err = dev_get_valid_name(net, dev, name);
1829 dev_net_set(dev, net);
1830 dev->rtnl_link_ops = &tun_link_ops;
1831 dev->ifindex = tfile->ifindex;
1832 dev->sysfs_groups[0] = &tun_attr_group;
1834 tun = netdev_priv(dev);
1837 tun->txflt.count = 0;
1838 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1840 tun->align = NET_SKB_PAD;
1841 tun->filter_attached = false;
1842 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1844 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
1845 if (!tun->pcpu_stats) {
1850 spin_lock_init(&tun->lock);
1852 err = security_tun_dev_alloc_security(&tun->security);
1859 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1860 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1861 NETIF_F_HW_VLAN_STAG_TX;
1862 dev->features = dev->hw_features | NETIF_F_LLTX;
1863 dev->vlan_features = dev->features &
1864 ~(NETIF_F_HW_VLAN_CTAG_TX |
1865 NETIF_F_HW_VLAN_STAG_TX);
1867 INIT_LIST_HEAD(&tun->disabled);
1868 err = tun_attach(tun, file, false, false);
1872 err = register_netdevice(tun->dev);
1875 /* free_netdev() won't check refcnt, to aovid race
1876 * with dev_put() we need publish tun after registration.
1878 rcu_assign_pointer(tfile->tun, tun);
1881 netif_carrier_on(tun->dev);
1883 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1885 tun->flags = (tun->flags & ~TUN_FEATURES) |
1886 (ifr->ifr_flags & TUN_FEATURES);
1888 /* Make sure persistent devices do not get stuck in
1891 if (netif_running(tun->dev))
1892 netif_tx_wake_all_queues(tun->dev);
1894 strcpy(ifr->ifr_name, tun->dev->name);
1898 tun_detach_all(dev);
1900 tun_flow_uninit(tun);
1901 security_tun_dev_free_security(tun->security);
1903 free_percpu(tun->pcpu_stats);
1909 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1912 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1914 strcpy(ifr->ifr_name, tun->dev->name);
1916 ifr->ifr_flags = tun_flags(tun);
1920 /* This is like a cut-down ethtool ops, except done via tun fd so no
1921 * privs required. */
1922 static int set_offload(struct tun_struct *tun, unsigned long arg)
1924 netdev_features_t features = 0;
1926 if (arg & TUN_F_CSUM) {
1927 features |= NETIF_F_HW_CSUM;
1930 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1931 if (arg & TUN_F_TSO_ECN) {
1932 features |= NETIF_F_TSO_ECN;
1933 arg &= ~TUN_F_TSO_ECN;
1935 if (arg & TUN_F_TSO4)
1936 features |= NETIF_F_TSO;
1937 if (arg & TUN_F_TSO6)
1938 features |= NETIF_F_TSO6;
1939 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1942 if (arg & TUN_F_UFO) {
1943 features |= NETIF_F_UFO;
1948 /* This gives the user a way to test for new features in future by
1949 * trying to set them. */
1953 tun->set_features = features;
1954 netdev_update_features(tun->dev);
1959 static void tun_detach_filter(struct tun_struct *tun, int n)
1962 struct tun_file *tfile;
1964 for (i = 0; i < n; i++) {
1965 tfile = rtnl_dereference(tun->tfiles[i]);
1966 lock_sock(tfile->socket.sk);
1967 sk_detach_filter(tfile->socket.sk);
1968 release_sock(tfile->socket.sk);
1971 tun->filter_attached = false;
1974 static int tun_attach_filter(struct tun_struct *tun)
1977 struct tun_file *tfile;
1979 for (i = 0; i < tun->numqueues; i++) {
1980 tfile = rtnl_dereference(tun->tfiles[i]);
1981 lock_sock(tfile->socket.sk);
1982 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1983 release_sock(tfile->socket.sk);
1985 tun_detach_filter(tun, i);
1990 tun->filter_attached = true;
1994 static void tun_set_sndbuf(struct tun_struct *tun)
1996 struct tun_file *tfile;
1999 for (i = 0; i < tun->numqueues; i++) {
2000 tfile = rtnl_dereference(tun->tfiles[i]);
2001 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2005 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2007 struct tun_file *tfile = file->private_data;
2008 struct tun_struct *tun;
2013 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2014 tun = tfile->detached;
2019 ret = security_tun_dev_attach_queue(tun->security);
2022 ret = tun_attach(tun, file, false, true);
2023 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2024 tun = rtnl_dereference(tfile->tun);
2025 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2028 __tun_detach(tfile, false);
2037 /* Return correct value for tun->dev->addr_len based on tun->dev->type. */
2038 static unsigned char tun_get_addr_len(unsigned short type)
2042 case ARPHRD_TUNNEL6:
2043 return sizeof(struct in6_addr);
2050 case ARPHRD_IEEE802154:
2051 case ARPHRD_IEEE802154_MONITOR:
2052 return IEEE802154_EXTENDED_ADDR_LEN;
2053 case ARPHRD_PHONET_PIPE:
2057 case ARPHRD_6LOWPAN:
2058 return EUI64_ADDR_LEN;
2063 case ARPHRD_IEEE802:
2066 return ROSE_ADDR_LEN;
2068 return AX25_ADDR_LEN;
2069 case ARPHRD_LOCALTLK:
2076 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2077 unsigned long arg, int ifreq_len)
2079 struct tun_file *tfile = file->private_data;
2080 struct tun_struct *tun;
2081 void __user* argp = (void __user*)arg;
2087 unsigned int ifindex;
2091 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
2092 if (copy_from_user(&ifr, argp, ifreq_len))
2095 memset(&ifr, 0, sizeof(ifr));
2097 if (cmd == TUNGETFEATURES) {
2098 /* Currently this just means: "what IFF flags are valid?".
2099 * This is needed because we never checked for invalid flags on
2102 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2103 (unsigned int __user*)argp);
2104 } else if (cmd == TUNSETQUEUE)
2105 return tun_set_queue(file, &ifr);
2110 tun = __tun_get(tfile);
2111 if (cmd == TUNSETIFF && !tun) {
2112 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2114 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
2119 if (copy_to_user(argp, &ifr, ifreq_len))
2123 if (cmd == TUNSETIFINDEX) {
2129 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2133 tfile->ifindex = ifindex;
2141 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2146 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2148 if (tfile->detached)
2149 ifr.ifr_flags |= IFF_DETACH_QUEUE;
2150 if (!tfile->socket.sk->sk_filter)
2151 ifr.ifr_flags |= IFF_NOFILTER;
2153 if (copy_to_user(argp, &ifr, ifreq_len))
2158 /* Disable/Enable checksum */
2160 /* [unimplemented] */
2161 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2162 arg ? "disabled" : "enabled");
2166 /* Disable/Enable persist mode. Keep an extra reference to the
2167 * module to prevent the module being unprobed.
2169 if (arg && !(tun->flags & IFF_PERSIST)) {
2170 tun->flags |= IFF_PERSIST;
2171 __module_get(THIS_MODULE);
2173 if (!arg && (tun->flags & IFF_PERSIST)) {
2174 tun->flags &= ~IFF_PERSIST;
2175 module_put(THIS_MODULE);
2178 tun_debug(KERN_INFO, tun, "persist %s\n",
2179 arg ? "enabled" : "disabled");
2183 /* Set owner of the device */
2184 owner = make_kuid(current_user_ns(), arg);
2185 if (!uid_valid(owner)) {
2190 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2191 from_kuid(&init_user_ns, tun->owner));
2195 /* Set group of the device */
2196 group = make_kgid(current_user_ns(), arg);
2197 if (!gid_valid(group)) {
2202 tun_debug(KERN_INFO, tun, "group set to %u\n",
2203 from_kgid(&init_user_ns, tun->group));
2207 /* Only allow setting the type when the interface is down */
2208 if (tun->dev->flags & IFF_UP) {
2209 tun_debug(KERN_INFO, tun,
2210 "Linktype set failed because interface is up\n");
2213 tun->dev->type = (int) arg;
2214 tun->dev->addr_len = tun_get_addr_len(tun->dev->type);
2215 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2227 ret = set_offload(tun, arg);
2230 case TUNSETTXFILTER:
2231 /* Can be set only for TAPs */
2233 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2235 ret = update_filter(&tun->txflt, (void __user *)arg);
2239 /* Get hw address */
2240 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2241 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2242 if (copy_to_user(argp, &ifr, ifreq_len))
2247 /* Set hw address */
2248 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2249 ifr.ifr_hwaddr.sa_data);
2251 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2255 sndbuf = tfile->socket.sk->sk_sndbuf;
2256 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2261 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2270 tun->sndbuf = sndbuf;
2271 tun_set_sndbuf(tun);
2274 case TUNGETVNETHDRSZ:
2275 vnet_hdr_sz = tun->vnet_hdr_sz;
2276 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2280 case TUNSETVNETHDRSZ:
2281 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2285 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2290 tun->vnet_hdr_sz = vnet_hdr_sz;
2294 le = !!(tun->flags & TUN_VNET_LE);
2295 if (put_user(le, (int __user *)argp))
2300 if (get_user(le, (int __user *)argp)) {
2305 tun->flags |= TUN_VNET_LE;
2307 tun->flags &= ~TUN_VNET_LE;
2311 ret = tun_get_vnet_be(tun, argp);
2315 ret = tun_set_vnet_be(tun, argp);
2318 case TUNATTACHFILTER:
2319 /* Can be set only for TAPs */
2321 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2324 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2327 ret = tun_attach_filter(tun);
2330 case TUNDETACHFILTER:
2331 /* Can be set only for TAPs */
2333 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2336 tun_detach_filter(tun, tun->numqueues);
2341 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2344 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2361 static long tun_chr_ioctl(struct file *file,
2362 unsigned int cmd, unsigned long arg)
2364 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2367 #ifdef CONFIG_COMPAT
2368 static long tun_chr_compat_ioctl(struct file *file,
2369 unsigned int cmd, unsigned long arg)
2374 case TUNSETTXFILTER:
2379 arg = (unsigned long)compat_ptr(arg);
2382 arg = (compat_ulong_t)arg;
2387 * compat_ifreq is shorter than ifreq, so we must not access beyond
2388 * the end of that structure. All fields that are used in this
2389 * driver are compatible though, we don't need to convert the
2392 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2394 #endif /* CONFIG_COMPAT */
2396 static int tun_chr_fasync(int fd, struct file *file, int on)
2398 struct tun_file *tfile = file->private_data;
2401 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2405 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2406 tfile->flags |= TUN_FASYNC;
2408 tfile->flags &= ~TUN_FASYNC;
2414 static int tun_chr_open(struct inode *inode, struct file * file)
2416 struct net *net = current->nsproxy->net_ns;
2417 struct tun_file *tfile;
2419 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2421 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2425 RCU_INIT_POINTER(tfile->tun, NULL);
2429 init_waitqueue_head(&tfile->wq.wait);
2430 RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2432 tfile->socket.file = file;
2433 tfile->socket.ops = &tun_socket_ops;
2435 sock_init_data(&tfile->socket, &tfile->sk);
2437 tfile->sk.sk_write_space = tun_sock_write_space;
2438 tfile->sk.sk_sndbuf = INT_MAX;
2440 file->private_data = tfile;
2441 INIT_LIST_HEAD(&tfile->next);
2443 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2445 memset(&tfile->tx_array, 0, sizeof(tfile->tx_array));
2450 static int tun_chr_close(struct inode *inode, struct file *file)
2452 struct tun_file *tfile = file->private_data;
2454 tun_detach(tfile, true);
2459 #ifdef CONFIG_PROC_FS
2460 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2462 struct tun_struct *tun;
2465 memset(&ifr, 0, sizeof(ifr));
2470 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2476 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2480 static const struct file_operations tun_fops = {
2481 .owner = THIS_MODULE,
2482 .llseek = no_llseek,
2483 .read_iter = tun_chr_read_iter,
2484 .write_iter = tun_chr_write_iter,
2485 .poll = tun_chr_poll,
2486 .unlocked_ioctl = tun_chr_ioctl,
2487 #ifdef CONFIG_COMPAT
2488 .compat_ioctl = tun_chr_compat_ioctl,
2490 .open = tun_chr_open,
2491 .release = tun_chr_close,
2492 .fasync = tun_chr_fasync,
2493 #ifdef CONFIG_PROC_FS
2494 .show_fdinfo = tun_chr_show_fdinfo,
2498 static struct miscdevice tun_miscdev = {
2501 .nodename = "net/tun",
2505 /* ethtool interface */
2507 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2510 cmd->advertising = 0;
2511 ethtool_cmd_speed_set(cmd, SPEED_10);
2512 cmd->duplex = DUPLEX_FULL;
2513 cmd->port = PORT_TP;
2514 cmd->phy_address = 0;
2515 cmd->transceiver = XCVR_INTERNAL;
2516 cmd->autoneg = AUTONEG_DISABLE;
2522 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2524 struct tun_struct *tun = netdev_priv(dev);
2526 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2527 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2529 switch (tun->flags & TUN_TYPE_MASK) {
2531 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2534 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2539 static u32 tun_get_msglevel(struct net_device *dev)
2542 struct tun_struct *tun = netdev_priv(dev);
2549 static void tun_set_msglevel(struct net_device *dev, u32 value)
2552 struct tun_struct *tun = netdev_priv(dev);
2557 static const struct ethtool_ops tun_ethtool_ops = {
2558 .get_settings = tun_get_settings,
2559 .get_drvinfo = tun_get_drvinfo,
2560 .get_msglevel = tun_get_msglevel,
2561 .set_msglevel = tun_set_msglevel,
2562 .get_link = ethtool_op_get_link,
2563 .get_ts_info = ethtool_op_get_ts_info,
2566 static int tun_queue_resize(struct tun_struct *tun)
2568 struct net_device *dev = tun->dev;
2569 struct tun_file *tfile;
2570 struct skb_array **arrays;
2571 int n = tun->numqueues + tun->numdisabled;
2574 arrays = kmalloc(sizeof *arrays * n, GFP_KERNEL);
2578 for (i = 0; i < tun->numqueues; i++) {
2579 tfile = rtnl_dereference(tun->tfiles[i]);
2580 arrays[i] = &tfile->tx_array;
2582 list_for_each_entry(tfile, &tun->disabled, next)
2583 arrays[i++] = &tfile->tx_array;
2585 ret = skb_array_resize_multiple(arrays, n,
2586 dev->tx_queue_len, GFP_KERNEL);
2592 static int tun_device_event(struct notifier_block *unused,
2593 unsigned long event, void *ptr)
2595 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2596 struct tun_struct *tun = netdev_priv(dev);
2599 if (dev->rtnl_link_ops != &tun_link_ops)
2603 case NETDEV_CHANGE_TX_QUEUE_LEN:
2604 if (tun_queue_resize(tun))
2608 for (i = 0; i < tun->numqueues; i++) {
2609 struct tun_file *tfile;
2611 tfile = rtnl_dereference(tun->tfiles[i]);
2612 tfile->socket.sk->sk_write_space(tfile->socket.sk);
2622 static struct notifier_block tun_notifier_block __read_mostly = {
2623 .notifier_call = tun_device_event,
2626 static int __init tun_init(void)
2630 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2631 pr_info("%s\n", DRV_COPYRIGHT);
2633 ret = rtnl_link_register(&tun_link_ops);
2635 pr_err("Can't register link_ops\n");
2639 ret = misc_register(&tun_miscdev);
2641 pr_err("Can't register misc device %d\n", TUN_MINOR);
2645 register_netdevice_notifier(&tun_notifier_block);
2648 rtnl_link_unregister(&tun_link_ops);
2653 static void tun_cleanup(void)
2655 misc_deregister(&tun_miscdev);
2656 rtnl_link_unregister(&tun_link_ops);
2657 unregister_netdevice_notifier(&tun_notifier_block);
2660 /* Get an underlying socket object from tun file. Returns error unless file is
2661 * attached to a device. The returned object works like a packet socket, it
2662 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
2663 * holding a reference to the file for as long as the socket is in use. */
2664 struct socket *tun_get_socket(struct file *file)
2666 struct tun_file *tfile;
2667 if (file->f_op != &tun_fops)
2668 return ERR_PTR(-EINVAL);
2669 tfile = file->private_data;
2671 return ERR_PTR(-EBADFD);
2672 return &tfile->socket;
2674 EXPORT_SYMBOL_GPL(tun_get_socket);
2676 module_init(tun_init);
2677 module_exit(tun_cleanup);
2678 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2679 MODULE_AUTHOR(DRV_COPYRIGHT);
2680 MODULE_LICENSE("GPL");
2681 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2682 MODULE_ALIAS("devname:net/tun");