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/crc32.h>
64 #include <linux/nsproxy.h>
65 #include <linux/virtio_net.h>
66 #include <linux/rcupdate.h>
67 #include <net/net_namespace.h>
68 #include <net/netns/generic.h>
69 #include <net/rtnetlink.h>
72 #include <asm/uaccess.h>
74 /* Uncomment to enable debugging */
75 /* #define TUN_DEBUG 1 */
80 #define tun_debug(level, tun, fmt, args...) \
83 netdev_printk(level, tun->dev, fmt, ##args); \
85 #define DBG1(level, fmt, args...) \
88 printk(level fmt, ##args); \
91 #define tun_debug(level, tun, fmt, args...) \
94 netdev_printk(level, tun->dev, fmt, ##args); \
96 #define DBG1(level, fmt, args...) \
99 printk(level fmt, ##args); \
103 #define GOODCOPY_LEN 128
105 #define FLT_EXACT_COUNT 8
107 unsigned int count; /* Number of addrs. Zero means disabled */
108 u32 mask[2]; /* Mask of the hashed addrs */
109 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
112 /* DEFAULT_MAX_NUM_RSS_QUEUES were choosed to let the rx/tx queues allocated for
113 * the netdevice to be fit in one page. So we can make sure the success of
114 * memory allocation. TODO: increase the limit. */
115 #define MAX_TAP_QUEUES DEFAULT_MAX_NUM_RSS_QUEUES
116 #define MAX_TAP_FLOWS 4096
118 #define TUN_FLOW_EXPIRE (3 * HZ)
120 /* A tun_file connects an open character device to a tuntap netdevice. It
121 * also contains all socket related strctures (except sock_fprog and tap_filter)
122 * to serve as one transmit queue for tuntap device. The sock_fprog and
123 * tap_filter were kept in tun_struct since they were used for filtering for the
124 * netdevice not for a specific queue (at least I didn't see the requirement for
128 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
129 * other can only be read while rcu_read_lock or rtnl_lock is held.
133 struct socket socket;
135 struct tun_struct __rcu *tun;
137 struct fasync_struct *fasync;
138 /* only used for fasnyc */
141 struct list_head next;
142 struct tun_struct *detached;
145 struct tun_flow_entry {
146 struct hlist_node hash_link;
148 struct tun_struct *tun;
152 unsigned long updated;
155 #define TUN_NUM_FLOW_ENTRIES 1024
157 /* Since the socket were moved to tun_file, to preserve the behavior of persist
158 * device, socket filter, sndbuf and vnet header size were restore when the
159 * file were attached to a persist device.
162 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
163 unsigned int numqueues;
168 struct net_device *dev;
169 netdev_features_t set_features;
170 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
171 NETIF_F_TSO6|NETIF_F_UFO)
175 struct tap_filter txflt;
176 struct sock_fprog fprog;
177 /* protected by rtnl lock */
178 bool filter_attached;
183 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
184 struct timer_list flow_gc_timer;
185 unsigned long ageing_time;
186 unsigned int numdisabled;
187 struct list_head disabled;
192 static inline u32 tun_hashfn(u32 rxhash)
194 return rxhash & 0x3ff;
197 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
199 struct tun_flow_entry *e;
200 struct hlist_node *n;
202 hlist_for_each_entry_rcu(e, n, head, hash_link) {
203 if (e->rxhash == rxhash)
209 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
210 struct hlist_head *head,
211 u32 rxhash, u16 queue_index)
213 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
216 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
217 rxhash, queue_index);
218 e->updated = jiffies;
220 e->queue_index = queue_index;
222 hlist_add_head_rcu(&e->hash_link, head);
228 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
230 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
231 e->rxhash, e->queue_index);
232 hlist_del_rcu(&e->hash_link);
237 static void tun_flow_flush(struct tun_struct *tun)
241 spin_lock_bh(&tun->lock);
242 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
243 struct tun_flow_entry *e;
244 struct hlist_node *h, *n;
246 hlist_for_each_entry_safe(e, h, n, &tun->flows[i], hash_link)
247 tun_flow_delete(tun, e);
249 spin_unlock_bh(&tun->lock);
252 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
256 spin_lock_bh(&tun->lock);
257 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
258 struct tun_flow_entry *e;
259 struct hlist_node *h, *n;
261 hlist_for_each_entry_safe(e, h, n, &tun->flows[i], hash_link) {
262 if (e->queue_index == queue_index)
263 tun_flow_delete(tun, e);
266 spin_unlock_bh(&tun->lock);
269 static void tun_flow_cleanup(unsigned long data)
271 struct tun_struct *tun = (struct tun_struct *)data;
272 unsigned long delay = tun->ageing_time;
273 unsigned long next_timer = jiffies + delay;
274 unsigned long count = 0;
277 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
279 spin_lock_bh(&tun->lock);
280 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
281 struct tun_flow_entry *e;
282 struct hlist_node *h, *n;
284 hlist_for_each_entry_safe(e, h, n, &tun->flows[i], hash_link) {
285 unsigned long this_timer;
287 this_timer = e->updated + delay;
288 if (time_before_eq(this_timer, jiffies))
289 tun_flow_delete(tun, e);
290 else if (time_before(this_timer, next_timer))
291 next_timer = this_timer;
296 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
297 spin_unlock_bh(&tun->lock);
300 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
301 struct tun_file *tfile)
303 struct hlist_head *head;
304 struct tun_flow_entry *e;
305 unsigned long delay = tun->ageing_time;
306 u16 queue_index = tfile->queue_index;
311 head = &tun->flows[tun_hashfn(rxhash)];
315 /* We may get a very small possibility of OOO during switching, not
316 * worth to optimize.*/
317 if (tun->numqueues == 1 || tfile->detached)
320 e = tun_flow_find(head, rxhash);
322 /* TODO: keep queueing to old queue until it's empty? */
323 e->queue_index = queue_index;
324 e->updated = jiffies;
326 spin_lock_bh(&tun->lock);
327 if (!tun_flow_find(head, rxhash) &&
328 tun->flow_count < MAX_TAP_FLOWS)
329 tun_flow_create(tun, head, rxhash, queue_index);
331 if (!timer_pending(&tun->flow_gc_timer))
332 mod_timer(&tun->flow_gc_timer,
333 round_jiffies_up(jiffies + delay));
334 spin_unlock_bh(&tun->lock);
341 /* We try to identify a flow through its rxhash first. The reason that
342 * we do not check rxq no. is becuase some cards(e.g 82599), chooses
343 * the rxq based on the txq where the last packet of the flow comes. As
344 * the userspace application move between processors, we may get a
345 * different rxq no. here. If we could not get rxhash, then we would
346 * hope the rxq no. may help here.
348 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb)
350 struct tun_struct *tun = netdev_priv(dev);
351 struct tun_flow_entry *e;
356 numqueues = tun->numqueues;
358 txq = skb_get_rxhash(skb);
360 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
362 txq = e->queue_index;
364 /* use multiply and shift instead of expensive divide */
365 txq = ((u64)txq * numqueues) >> 32;
366 } else if (likely(skb_rx_queue_recorded(skb))) {
367 txq = skb_get_rx_queue(skb);
368 while (unlikely(txq >= numqueues))
376 static inline bool tun_not_capable(struct tun_struct *tun)
378 const struct cred *cred = current_cred();
379 struct net *net = dev_net(tun->dev);
381 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
382 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
383 !ns_capable(net->user_ns, CAP_NET_ADMIN);
386 static void tun_set_real_num_queues(struct tun_struct *tun)
388 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
389 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
392 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
394 tfile->detached = tun;
395 list_add_tail(&tfile->next, &tun->disabled);
399 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
401 struct tun_struct *tun = tfile->detached;
403 tfile->detached = NULL;
404 list_del_init(&tfile->next);
409 static void __tun_detach(struct tun_file *tfile, bool clean)
411 struct tun_file *ntfile;
412 struct tun_struct *tun;
413 struct net_device *dev;
415 tun = rtnl_dereference(tfile->tun);
417 if (tun && !tfile->detached) {
418 u16 index = tfile->queue_index;
419 BUG_ON(index >= tun->numqueues);
422 rcu_assign_pointer(tun->tfiles[index],
423 tun->tfiles[tun->numqueues - 1]);
424 ntfile = rtnl_dereference(tun->tfiles[index]);
425 ntfile->queue_index = index;
429 rcu_assign_pointer(tfile->tun, NULL);
430 sock_put(&tfile->sk);
432 tun_disable_queue(tun, tfile);
435 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
436 /* Drop read queue */
437 skb_queue_purge(&tfile->sk.sk_receive_queue);
438 tun_set_real_num_queues(tun);
439 } else if (tfile->detached && clean) {
440 tun = tun_enable_queue(tfile);
441 sock_put(&tfile->sk);
445 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
446 netif_carrier_off(tun->dev);
448 if (!(tun->flags & TUN_PERSIST) &&
449 tun->dev->reg_state == NETREG_REGISTERED)
450 unregister_netdevice(tun->dev);
453 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
454 &tfile->socket.flags));
455 sk_release_kernel(&tfile->sk);
459 static void tun_detach(struct tun_file *tfile, bool clean)
462 __tun_detach(tfile, clean);
466 static void tun_detach_all(struct net_device *dev)
468 struct tun_struct *tun = netdev_priv(dev);
469 struct tun_file *tfile, *tmp;
470 int i, n = tun->numqueues;
472 for (i = 0; i < n; i++) {
473 tfile = rtnl_dereference(tun->tfiles[i]);
475 wake_up_all(&tfile->wq.wait);
476 rcu_assign_pointer(tfile->tun, NULL);
479 list_for_each_entry(tfile, &tun->disabled, next) {
480 wake_up_all(&tfile->wq.wait);
481 rcu_assign_pointer(tfile->tun, NULL);
483 BUG_ON(tun->numqueues != 0);
486 for (i = 0; i < n; i++) {
487 tfile = rtnl_dereference(tun->tfiles[i]);
488 /* Drop read queue */
489 skb_queue_purge(&tfile->sk.sk_receive_queue);
490 sock_put(&tfile->sk);
492 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
493 tun_enable_queue(tfile);
494 skb_queue_purge(&tfile->sk.sk_receive_queue);
495 sock_put(&tfile->sk);
497 BUG_ON(tun->numdisabled != 0);
499 if (tun->flags & TUN_PERSIST)
500 module_put(THIS_MODULE);
503 static int tun_attach(struct tun_struct *tun, struct file *file)
505 struct tun_file *tfile = file->private_data;
508 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
513 if (rtnl_dereference(tfile->tun) && !tfile->detached)
517 if (!(tun->flags & TUN_TAP_MQ) && tun->numqueues == 1)
521 if (!tfile->detached &&
522 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
527 /* Re-attach the filter to presist device */
528 if (tun->filter_attached == true) {
529 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
533 tfile->queue_index = tun->numqueues;
534 rcu_assign_pointer(tfile->tun, tun);
535 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
539 tun_enable_queue(tfile);
541 sock_hold(&tfile->sk);
543 tun_set_real_num_queues(tun);
545 /* device is allowed to go away first, so no need to hold extra
553 static struct tun_struct *__tun_get(struct tun_file *tfile)
555 struct tun_struct *tun;
558 tun = rcu_dereference(tfile->tun);
566 static struct tun_struct *tun_get(struct file *file)
568 return __tun_get(file->private_data);
571 static void tun_put(struct tun_struct *tun)
577 static void addr_hash_set(u32 *mask, const u8 *addr)
579 int n = ether_crc(ETH_ALEN, addr) >> 26;
580 mask[n >> 5] |= (1 << (n & 31));
583 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
585 int n = ether_crc(ETH_ALEN, addr) >> 26;
586 return mask[n >> 5] & (1 << (n & 31));
589 static int update_filter(struct tap_filter *filter, void __user *arg)
591 struct { u8 u[ETH_ALEN]; } *addr;
592 struct tun_filter uf;
593 int err, alen, n, nexact;
595 if (copy_from_user(&uf, arg, sizeof(uf)))
604 alen = ETH_ALEN * uf.count;
605 addr = kmalloc(alen, GFP_KERNEL);
609 if (copy_from_user(addr, arg + sizeof(uf), alen)) {
614 /* The filter is updated without holding any locks. Which is
615 * perfectly safe. We disable it first and in the worst
616 * case we'll accept a few undesired packets. */
620 /* Use first set of addresses as an exact filter */
621 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
622 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
626 /* Remaining multicast addresses are hashed,
627 * unicast will leave the filter disabled. */
628 memset(filter->mask, 0, sizeof(filter->mask));
629 for (; n < uf.count; n++) {
630 if (!is_multicast_ether_addr(addr[n].u)) {
631 err = 0; /* no filter */
634 addr_hash_set(filter->mask, addr[n].u);
637 /* For ALLMULTI just set the mask to all ones.
638 * This overrides the mask populated above. */
639 if ((uf.flags & TUN_FLT_ALLMULTI))
640 memset(filter->mask, ~0, sizeof(filter->mask));
642 /* Now enable the filter */
644 filter->count = nexact;
646 /* Return the number of exact filters */
654 /* Returns: 0 - drop, !=0 - accept */
655 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
657 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
659 struct ethhdr *eh = (struct ethhdr *) skb->data;
663 for (i = 0; i < filter->count; i++)
664 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
667 /* Inexact match (multicast only) */
668 if (is_multicast_ether_addr(eh->h_dest))
669 return addr_hash_test(filter->mask, eh->h_dest);
675 * Checks whether the packet is accepted or not.
676 * Returns: 0 - drop, !=0 - accept
678 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
683 return run_filter(filter, skb);
686 /* Network device part of the driver */
688 static const struct ethtool_ops tun_ethtool_ops;
690 /* Net device detach from fd. */
691 static void tun_net_uninit(struct net_device *dev)
696 /* Net device open. */
697 static int tun_net_open(struct net_device *dev)
699 netif_tx_start_all_queues(dev);
703 /* Net device close. */
704 static int tun_net_close(struct net_device *dev)
706 netif_tx_stop_all_queues(dev);
710 /* Net device start xmit */
711 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
713 struct tun_struct *tun = netdev_priv(dev);
714 int txq = skb->queue_mapping;
715 struct tun_file *tfile;
718 tfile = rcu_dereference(tun->tfiles[txq]);
720 /* Drop packet if interface is not attached */
721 if (txq >= tun->numqueues)
724 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
728 /* Drop if the filter does not like it.
729 * This is a noop if the filter is disabled.
730 * Filter can be enabled only for the TAP devices. */
731 if (!check_filter(&tun->txflt, skb))
734 if (tfile->socket.sk->sk_filter &&
735 sk_filter(tfile->socket.sk, skb))
738 /* Limit the number of packets queued by dividing txq length with the
741 if (skb_queue_len(&tfile->socket.sk->sk_receive_queue)
742 >= dev->tx_queue_len / tun->numqueues)
745 /* Orphan the skb - required as we might hang on to it
746 * for indefinite time. */
747 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
752 skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
754 /* Notify and wake up reader process */
755 if (tfile->flags & TUN_FASYNC)
756 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
757 wake_up_interruptible_poll(&tfile->wq.wait, POLLIN |
758 POLLRDNORM | POLLRDBAND);
764 dev->stats.tx_dropped++;
771 static void tun_net_mclist(struct net_device *dev)
774 * This callback is supposed to deal with mc filter in
775 * _rx_ path and has nothing to do with the _tx_ path.
776 * In rx path we always accept everything userspace gives us.
781 #define MAX_MTU 65535
784 tun_net_change_mtu(struct net_device *dev, int new_mtu)
786 if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
792 static netdev_features_t tun_net_fix_features(struct net_device *dev,
793 netdev_features_t features)
795 struct tun_struct *tun = netdev_priv(dev);
797 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
799 #ifdef CONFIG_NET_POLL_CONTROLLER
800 static void tun_poll_controller(struct net_device *dev)
803 * Tun only receives frames when:
804 * 1) the char device endpoint gets data from user space
805 * 2) the tun socket gets a sendmsg call from user space
806 * Since both of those are syncronous operations, we are guaranteed
807 * never to have pending data when we poll for it
808 * so theres nothing to do here but return.
809 * We need this though so netpoll recognizes us as an interface that
810 * supports polling, which enables bridge devices in virt setups to
811 * still use netconsole
816 static const struct net_device_ops tun_netdev_ops = {
817 .ndo_uninit = tun_net_uninit,
818 .ndo_open = tun_net_open,
819 .ndo_stop = tun_net_close,
820 .ndo_start_xmit = tun_net_xmit,
821 .ndo_change_mtu = tun_net_change_mtu,
822 .ndo_fix_features = tun_net_fix_features,
823 .ndo_select_queue = tun_select_queue,
824 #ifdef CONFIG_NET_POLL_CONTROLLER
825 .ndo_poll_controller = tun_poll_controller,
829 static const struct net_device_ops tap_netdev_ops = {
830 .ndo_uninit = tun_net_uninit,
831 .ndo_open = tun_net_open,
832 .ndo_stop = tun_net_close,
833 .ndo_start_xmit = tun_net_xmit,
834 .ndo_change_mtu = tun_net_change_mtu,
835 .ndo_fix_features = tun_net_fix_features,
836 .ndo_set_rx_mode = tun_net_mclist,
837 .ndo_set_mac_address = eth_mac_addr,
838 .ndo_validate_addr = eth_validate_addr,
839 .ndo_select_queue = tun_select_queue,
840 #ifdef CONFIG_NET_POLL_CONTROLLER
841 .ndo_poll_controller = tun_poll_controller,
845 static int tun_flow_init(struct tun_struct *tun)
849 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
850 INIT_HLIST_HEAD(&tun->flows[i]);
852 tun->ageing_time = TUN_FLOW_EXPIRE;
853 setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
854 mod_timer(&tun->flow_gc_timer,
855 round_jiffies_up(jiffies + tun->ageing_time));
860 static void tun_flow_uninit(struct tun_struct *tun)
862 del_timer_sync(&tun->flow_gc_timer);
866 /* Initialize net device. */
867 static void tun_net_init(struct net_device *dev)
869 struct tun_struct *tun = netdev_priv(dev);
871 switch (tun->flags & TUN_TYPE_MASK) {
873 dev->netdev_ops = &tun_netdev_ops;
875 /* Point-to-Point TUN Device */
876 dev->hard_header_len = 0;
880 /* Zero header length */
881 dev->type = ARPHRD_NONE;
882 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
883 dev->tx_queue_len = TUN_READQ_SIZE; /* We prefer our own queue length */
887 dev->netdev_ops = &tap_netdev_ops;
888 /* Ethernet TAP Device */
890 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
891 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
893 eth_hw_addr_random(dev);
895 dev->tx_queue_len = TUN_READQ_SIZE; /* We prefer our own queue length */
900 /* Character device part */
903 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
905 struct tun_file *tfile = file->private_data;
906 struct tun_struct *tun = __tun_get(tfile);
908 unsigned int mask = 0;
913 sk = tfile->socket.sk;
915 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
917 poll_wait(file, &tfile->wq.wait, wait);
919 if (!skb_queue_empty(&sk->sk_receive_queue))
920 mask |= POLLIN | POLLRDNORM;
922 if (sock_writeable(sk) ||
923 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
925 mask |= POLLOUT | POLLWRNORM;
927 if (tun->dev->reg_state != NETREG_REGISTERED)
934 /* prepad is the amount to reserve at front. len is length after that.
935 * linear is a hint as to how much to copy (usually headers). */
936 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
937 size_t prepad, size_t len,
938 size_t linear, int noblock)
940 struct sock *sk = tfile->socket.sk;
944 /* Under a page? Don't bother with paged skb. */
945 if (prepad + len < PAGE_SIZE || !linear)
948 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
953 skb_reserve(skb, prepad);
954 skb_put(skb, linear);
955 skb->data_len = len - linear;
956 skb->len += len - linear;
961 /* set skb frags from iovec, this can move to core network code for reuse */
962 static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
963 int offset, size_t count)
965 int len = iov_length(from, count) - offset;
966 int copy = skb_headlen(skb);
967 int size, offset1 = 0;
970 /* Skip over from offset */
971 while (count && (offset >= from->iov_len)) {
972 offset -= from->iov_len;
977 /* copy up to skb headlen */
978 while (count && (copy > 0)) {
979 size = min_t(unsigned int, copy, from->iov_len - offset);
980 if (copy_from_user(skb->data + offset1, from->iov_base + offset,
997 struct page *page[MAX_SKB_FRAGS];
1000 unsigned long truesize;
1002 len = from->iov_len - offset;
1008 base = (unsigned long)from->iov_base + offset;
1009 size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
1010 if (i + size > MAX_SKB_FRAGS)
1012 num_pages = get_user_pages_fast(base, size, 0, &page[i]);
1013 if (num_pages != size) {
1014 for (i = 0; i < num_pages; i++)
1018 truesize = size * PAGE_SIZE;
1019 skb->data_len += len;
1021 skb->truesize += truesize;
1022 atomic_add(truesize, &skb->sk->sk_wmem_alloc);
1024 int off = base & ~PAGE_MASK;
1025 int size = min_t(int, len, PAGE_SIZE - off);
1026 __skb_fill_page_desc(skb, i, page[i], off, size);
1027 skb_shinfo(skb)->nr_frags++;
1028 /* increase sk_wmem_alloc */
1039 /* Get packet from user space buffer */
1040 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1041 void *msg_control, const struct iovec *iv,
1042 size_t total_len, size_t count, int noblock)
1044 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1045 struct sk_buff *skb;
1046 size_t len = total_len, align = NET_SKB_PAD;
1047 struct virtio_net_hdr gso = { 0 };
1050 bool zerocopy = false;
1054 if (!(tun->flags & TUN_NO_PI)) {
1055 if ((len -= sizeof(pi)) > total_len)
1058 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
1060 offset += sizeof(pi);
1063 if (tun->flags & TUN_VNET_HDR) {
1064 if ((len -= tun->vnet_hdr_sz) > total_len)
1067 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1070 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1071 gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1072 gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1074 if (gso.hdr_len > len)
1076 offset += tun->vnet_hdr_sz;
1079 if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1080 align += NET_IP_ALIGN;
1081 if (unlikely(len < ETH_HLEN ||
1082 (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1090 /* Userspace may produce vectors with count greater than
1091 * MAX_SKB_FRAGS, so we need to linearize parts of the skb
1092 * to let the rest of data to be fit in the frags.
1094 if (count > MAX_SKB_FRAGS) {
1095 copylen = iov_length(iv, count - MAX_SKB_FRAGS);
1096 if (copylen < offset)
1102 /* There are 256 bytes to be copied in skb, so there is enough
1103 * room for skb expand head in case it is used.
1104 * The rest of the buffer is mapped from userspace.
1106 if (copylen < gso.hdr_len)
1107 copylen = gso.hdr_len;
1109 copylen = GOODCOPY_LEN;
1113 skb = tun_alloc_skb(tfile, align, copylen, gso.hdr_len, noblock);
1115 if (PTR_ERR(skb) != -EAGAIN)
1116 tun->dev->stats.rx_dropped++;
1117 return PTR_ERR(skb);
1121 err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1123 err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1126 tun->dev->stats.rx_dropped++;
1131 if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1132 if (!skb_partial_csum_set(skb, gso.csum_start,
1134 tun->dev->stats.rx_frame_errors++;
1140 switch (tun->flags & TUN_TYPE_MASK) {
1142 if (tun->flags & TUN_NO_PI) {
1143 switch (skb->data[0] & 0xf0) {
1145 pi.proto = htons(ETH_P_IP);
1148 pi.proto = htons(ETH_P_IPV6);
1151 tun->dev->stats.rx_dropped++;
1157 skb_reset_mac_header(skb);
1158 skb->protocol = pi.proto;
1159 skb->dev = tun->dev;
1162 skb->protocol = eth_type_trans(skb, tun->dev);
1166 if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1168 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1169 case VIRTIO_NET_HDR_GSO_TCPV4:
1170 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1172 case VIRTIO_NET_HDR_GSO_TCPV6:
1173 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1175 case VIRTIO_NET_HDR_GSO_UDP:
1176 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1179 tun->dev->stats.rx_frame_errors++;
1184 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1185 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1187 skb_shinfo(skb)->gso_size = gso.gso_size;
1188 if (skb_shinfo(skb)->gso_size == 0) {
1189 tun->dev->stats.rx_frame_errors++;
1194 /* Header must be checked, and gso_segs computed. */
1195 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1196 skb_shinfo(skb)->gso_segs = 0;
1199 /* copy skb_ubuf_info for callback when skb has no error */
1201 skb_shinfo(skb)->destructor_arg = msg_control;
1202 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1203 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1206 skb_reset_network_header(skb);
1207 rxhash = skb_get_rxhash(skb);
1210 tun->dev->stats.rx_packets++;
1211 tun->dev->stats.rx_bytes += len;
1213 tun_flow_update(tun, rxhash, tfile);
1217 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1218 unsigned long count, loff_t pos)
1220 struct file *file = iocb->ki_filp;
1221 struct tun_struct *tun = tun_get(file);
1222 struct tun_file *tfile = file->private_data;
1228 tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1230 result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1231 count, file->f_flags & O_NONBLOCK);
1237 /* Put packet to the user space buffer */
1238 static ssize_t tun_put_user(struct tun_struct *tun,
1239 struct tun_file *tfile,
1240 struct sk_buff *skb,
1241 const struct iovec *iv, int len)
1243 struct tun_pi pi = { 0, skb->protocol };
1246 if (!(tun->flags & TUN_NO_PI)) {
1247 if ((len -= sizeof(pi)) < 0)
1250 if (len < skb->len) {
1251 /* Packet will be striped */
1252 pi.flags |= TUN_PKT_STRIP;
1255 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1257 total += sizeof(pi);
1260 if (tun->flags & TUN_VNET_HDR) {
1261 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1262 if ((len -= tun->vnet_hdr_sz) < 0)
1265 if (skb_is_gso(skb)) {
1266 struct skb_shared_info *sinfo = skb_shinfo(skb);
1268 /* This is a hint as to how much should be linear. */
1269 gso.hdr_len = skb_headlen(skb);
1270 gso.gso_size = sinfo->gso_size;
1271 if (sinfo->gso_type & SKB_GSO_TCPV4)
1272 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1273 else if (sinfo->gso_type & SKB_GSO_TCPV6)
1274 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1275 else if (sinfo->gso_type & SKB_GSO_UDP)
1276 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1278 pr_err("unexpected GSO type: "
1279 "0x%x, gso_size %d, hdr_len %d\n",
1280 sinfo->gso_type, gso.gso_size,
1282 print_hex_dump(KERN_ERR, "tun: ",
1285 min((int)gso.hdr_len, 64), true);
1289 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1290 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1292 gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1294 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1295 gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1296 gso.csum_start = skb_checksum_start_offset(skb);
1297 gso.csum_offset = skb->csum_offset;
1298 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1299 gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1300 } /* else everything is zero */
1302 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1305 total += tun->vnet_hdr_sz;
1308 len = min_t(int, skb->len, len);
1310 skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
1313 tun->dev->stats.tx_packets++;
1314 tun->dev->stats.tx_bytes += len;
1319 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1320 struct kiocb *iocb, const struct iovec *iv,
1321 ssize_t len, int noblock)
1323 DECLARE_WAITQUEUE(wait, current);
1324 struct sk_buff *skb;
1327 tun_debug(KERN_INFO, tun, "tun_do_read\n");
1329 if (unlikely(!noblock))
1330 add_wait_queue(&tfile->wq.wait, &wait);
1332 current->state = TASK_INTERRUPTIBLE;
1334 /* Read frames from the queue */
1335 if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) {
1340 if (signal_pending(current)) {
1344 if (tun->dev->reg_state != NETREG_REGISTERED) {
1349 /* Nothing to read, let's sleep */
1354 ret = tun_put_user(tun, tfile, skb, iv, len);
1359 current->state = TASK_RUNNING;
1360 if (unlikely(!noblock))
1361 remove_wait_queue(&tfile->wq.wait, &wait);
1366 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1367 unsigned long count, loff_t pos)
1369 struct file *file = iocb->ki_filp;
1370 struct tun_file *tfile = file->private_data;
1371 struct tun_struct *tun = __tun_get(tfile);
1376 len = iov_length(iv, count);
1382 ret = tun_do_read(tun, tfile, iocb, iv, len,
1383 file->f_flags & O_NONBLOCK);
1384 ret = min_t(ssize_t, ret, len);
1390 static void tun_free_netdev(struct net_device *dev)
1392 struct tun_struct *tun = netdev_priv(dev);
1394 BUG_ON(!(list_empty(&tun->disabled)));
1395 tun_flow_uninit(tun);
1396 security_tun_dev_free_security(tun->security);
1400 static void tun_setup(struct net_device *dev)
1402 struct tun_struct *tun = netdev_priv(dev);
1404 tun->owner = INVALID_UID;
1405 tun->group = INVALID_GID;
1407 dev->ethtool_ops = &tun_ethtool_ops;
1408 dev->destructor = tun_free_netdev;
1411 /* Trivial set of netlink ops to allow deleting tun or tap
1412 * device with netlink.
1414 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1419 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1421 .priv_size = sizeof(struct tun_struct),
1423 .validate = tun_validate,
1426 static void tun_sock_write_space(struct sock *sk)
1428 struct tun_file *tfile;
1429 wait_queue_head_t *wqueue;
1431 if (!sock_writeable(sk))
1434 if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1437 wqueue = sk_sleep(sk);
1438 if (wqueue && waitqueue_active(wqueue))
1439 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1440 POLLWRNORM | POLLWRBAND);
1442 tfile = container_of(sk, struct tun_file, sk);
1443 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1446 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1447 struct msghdr *m, size_t total_len)
1450 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1451 struct tun_struct *tun = __tun_get(tfile);
1455 ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1456 m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1462 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1463 struct msghdr *m, size_t total_len,
1466 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1467 struct tun_struct *tun = __tun_get(tfile);
1473 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1475 ret = tun_do_read(tun, tfile, iocb, m->msg_iov, total_len,
1476 flags & MSG_DONTWAIT);
1477 if (ret > total_len) {
1478 m->msg_flags |= MSG_TRUNC;
1479 ret = flags & MSG_TRUNC ? ret : total_len;
1485 static int tun_release(struct socket *sock)
1492 /* Ops structure to mimic raw sockets with tun */
1493 static const struct proto_ops tun_socket_ops = {
1494 .sendmsg = tun_sendmsg,
1495 .recvmsg = tun_recvmsg,
1496 .release = tun_release,
1499 static struct proto tun_proto = {
1501 .owner = THIS_MODULE,
1502 .obj_size = sizeof(struct tun_file),
1505 static int tun_flags(struct tun_struct *tun)
1509 if (tun->flags & TUN_TUN_DEV)
1514 if (tun->flags & TUN_NO_PI)
1517 /* This flag has no real effect. We track the value for backwards
1520 if (tun->flags & TUN_ONE_QUEUE)
1521 flags |= IFF_ONE_QUEUE;
1523 if (tun->flags & TUN_VNET_HDR)
1524 flags |= IFF_VNET_HDR;
1526 if (tun->flags & TUN_TAP_MQ)
1527 flags |= IFF_MULTI_QUEUE;
1532 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1535 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1536 return sprintf(buf, "0x%x\n", tun_flags(tun));
1539 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1542 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1543 return uid_valid(tun->owner)?
1544 sprintf(buf, "%u\n",
1545 from_kuid_munged(current_user_ns(), tun->owner)):
1546 sprintf(buf, "-1\n");
1549 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1552 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1553 return gid_valid(tun->group) ?
1554 sprintf(buf, "%u\n",
1555 from_kgid_munged(current_user_ns(), tun->group)):
1556 sprintf(buf, "-1\n");
1559 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1560 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1561 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1563 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1565 struct tun_struct *tun;
1566 struct tun_file *tfile = file->private_data;
1567 struct net_device *dev;
1570 if (tfile->detached)
1573 dev = __dev_get_by_name(net, ifr->ifr_name);
1575 if (ifr->ifr_flags & IFF_TUN_EXCL)
1577 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1578 tun = netdev_priv(dev);
1579 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1580 tun = netdev_priv(dev);
1584 if (tun_not_capable(tun))
1586 err = security_tun_dev_open(tun->security);
1590 err = tun_attach(tun, file);
1594 if (tun->flags & TUN_TAP_MQ &&
1595 (tun->numqueues + tun->numdisabled > 1))
1600 unsigned long flags = 0;
1601 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1604 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1606 err = security_tun_dev_create();
1611 if (ifr->ifr_flags & IFF_TUN) {
1613 flags |= TUN_TUN_DEV;
1615 } else if (ifr->ifr_flags & IFF_TAP) {
1617 flags |= TUN_TAP_DEV;
1623 name = ifr->ifr_name;
1625 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1626 tun_setup, queues, queues);
1631 dev_net_set(dev, net);
1632 dev->rtnl_link_ops = &tun_link_ops;
1634 tun = netdev_priv(dev);
1637 tun->txflt.count = 0;
1638 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1640 tun->filter_attached = false;
1641 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1643 spin_lock_init(&tun->lock);
1645 err = security_tun_dev_alloc_security(&tun->security);
1651 err = tun_flow_init(tun);
1655 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1657 dev->features = dev->hw_features;
1659 INIT_LIST_HEAD(&tun->disabled);
1660 err = tun_attach(tun, file);
1664 err = register_netdevice(tun->dev);
1668 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1669 device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1670 device_create_file(&tun->dev->dev, &dev_attr_group))
1671 pr_err("Failed to create tun sysfs files\n");
1674 netif_carrier_on(tun->dev);
1676 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1678 if (ifr->ifr_flags & IFF_NO_PI)
1679 tun->flags |= TUN_NO_PI;
1681 tun->flags &= ~TUN_NO_PI;
1683 /* This flag has no real effect. We track the value for backwards
1686 if (ifr->ifr_flags & IFF_ONE_QUEUE)
1687 tun->flags |= TUN_ONE_QUEUE;
1689 tun->flags &= ~TUN_ONE_QUEUE;
1691 if (ifr->ifr_flags & IFF_VNET_HDR)
1692 tun->flags |= TUN_VNET_HDR;
1694 tun->flags &= ~TUN_VNET_HDR;
1696 if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1697 tun->flags |= TUN_TAP_MQ;
1699 tun->flags &= ~TUN_TAP_MQ;
1701 /* Make sure persistent devices do not get stuck in
1704 if (netif_running(tun->dev))
1705 netif_tx_wake_all_queues(tun->dev);
1707 strcpy(ifr->ifr_name, tun->dev->name);
1715 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1718 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1720 strcpy(ifr->ifr_name, tun->dev->name);
1722 ifr->ifr_flags = tun_flags(tun);
1726 /* This is like a cut-down ethtool ops, except done via tun fd so no
1727 * privs required. */
1728 static int set_offload(struct tun_struct *tun, unsigned long arg)
1730 netdev_features_t features = 0;
1732 if (arg & TUN_F_CSUM) {
1733 features |= NETIF_F_HW_CSUM;
1736 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1737 if (arg & TUN_F_TSO_ECN) {
1738 features |= NETIF_F_TSO_ECN;
1739 arg &= ~TUN_F_TSO_ECN;
1741 if (arg & TUN_F_TSO4)
1742 features |= NETIF_F_TSO;
1743 if (arg & TUN_F_TSO6)
1744 features |= NETIF_F_TSO6;
1745 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1748 if (arg & TUN_F_UFO) {
1749 features |= NETIF_F_UFO;
1754 /* This gives the user a way to test for new features in future by
1755 * trying to set them. */
1759 tun->set_features = features;
1760 netdev_update_features(tun->dev);
1765 static void tun_detach_filter(struct tun_struct *tun, int n)
1768 struct tun_file *tfile;
1770 for (i = 0; i < n; i++) {
1771 tfile = rtnl_dereference(tun->tfiles[i]);
1772 sk_detach_filter(tfile->socket.sk);
1775 tun->filter_attached = false;
1778 static int tun_attach_filter(struct tun_struct *tun)
1781 struct tun_file *tfile;
1783 for (i = 0; i < tun->numqueues; i++) {
1784 tfile = rtnl_dereference(tun->tfiles[i]);
1785 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1787 tun_detach_filter(tun, i);
1792 tun->filter_attached = true;
1796 static void tun_set_sndbuf(struct tun_struct *tun)
1798 struct tun_file *tfile;
1801 for (i = 0; i < tun->numqueues; i++) {
1802 tfile = rtnl_dereference(tun->tfiles[i]);
1803 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1807 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1809 struct tun_file *tfile = file->private_data;
1810 struct tun_struct *tun;
1815 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1816 tun = tfile->detached;
1821 ret = security_tun_dev_attach_queue(tun->security);
1824 ret = tun_attach(tun, file);
1825 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1826 tun = rtnl_dereference(tfile->tun);
1827 if (!tun || !(tun->flags & TUN_TAP_MQ) || tfile->detached)
1830 __tun_detach(tfile, false);
1839 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1840 unsigned long arg, int ifreq_len)
1842 struct tun_file *tfile = file->private_data;
1843 struct tun_struct *tun;
1844 void __user* argp = (void __user*)arg;
1852 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1853 if (copy_from_user(&ifr, argp, ifreq_len))
1856 memset(&ifr, 0, sizeof(ifr));
1858 if (cmd == TUNGETFEATURES) {
1859 /* Currently this just means: "what IFF flags are valid?".
1860 * This is needed because we never checked for invalid flags on
1862 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1863 IFF_VNET_HDR | IFF_MULTI_QUEUE,
1864 (unsigned int __user*)argp);
1865 } else if (cmd == TUNSETQUEUE)
1866 return tun_set_queue(file, &ifr);
1871 tun = __tun_get(tfile);
1872 if (cmd == TUNSETIFF && !tun) {
1873 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1875 ret = tun_set_iff(tfile->net, file, &ifr);
1880 if (copy_to_user(argp, &ifr, ifreq_len))
1889 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1894 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1896 if (copy_to_user(argp, &ifr, ifreq_len))
1901 /* Disable/Enable checksum */
1903 /* [unimplemented] */
1904 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1905 arg ? "disabled" : "enabled");
1909 /* Disable/Enable persist mode. Keep an extra reference to the
1910 * module to prevent the module being unprobed.
1912 if (arg && !(tun->flags & TUN_PERSIST)) {
1913 tun->flags |= TUN_PERSIST;
1914 __module_get(THIS_MODULE);
1916 if (!arg && (tun->flags & TUN_PERSIST)) {
1917 tun->flags &= ~TUN_PERSIST;
1918 module_put(THIS_MODULE);
1921 tun_debug(KERN_INFO, tun, "persist %s\n",
1922 arg ? "enabled" : "disabled");
1926 /* Set owner of the device */
1927 owner = make_kuid(current_user_ns(), arg);
1928 if (!uid_valid(owner)) {
1933 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1934 from_kuid(&init_user_ns, tun->owner));
1938 /* Set group of the device */
1939 group = make_kgid(current_user_ns(), arg);
1940 if (!gid_valid(group)) {
1945 tun_debug(KERN_INFO, tun, "group set to %u\n",
1946 from_kgid(&init_user_ns, tun->group));
1950 /* Only allow setting the type when the interface is down */
1951 if (tun->dev->flags & IFF_UP) {
1952 tun_debug(KERN_INFO, tun,
1953 "Linktype set failed because interface is up\n");
1956 tun->dev->type = (int) arg;
1957 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1969 ret = set_offload(tun, arg);
1972 case TUNSETTXFILTER:
1973 /* Can be set only for TAPs */
1975 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1977 ret = update_filter(&tun->txflt, (void __user *)arg);
1981 /* Get hw address */
1982 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1983 ifr.ifr_hwaddr.sa_family = tun->dev->type;
1984 if (copy_to_user(argp, &ifr, ifreq_len))
1989 /* Set hw address */
1990 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1991 ifr.ifr_hwaddr.sa_data);
1993 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1997 sndbuf = tfile->socket.sk->sk_sndbuf;
1998 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2003 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2008 tun->sndbuf = sndbuf;
2009 tun_set_sndbuf(tun);
2012 case TUNGETVNETHDRSZ:
2013 vnet_hdr_sz = tun->vnet_hdr_sz;
2014 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2018 case TUNSETVNETHDRSZ:
2019 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2023 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2028 tun->vnet_hdr_sz = vnet_hdr_sz;
2031 case TUNATTACHFILTER:
2032 /* Can be set only for TAPs */
2034 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2037 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2040 ret = tun_attach_filter(tun);
2043 case TUNDETACHFILTER:
2044 /* Can be set only for TAPs */
2046 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2049 tun_detach_filter(tun, tun->numqueues);
2064 static long tun_chr_ioctl(struct file *file,
2065 unsigned int cmd, unsigned long arg)
2067 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2070 #ifdef CONFIG_COMPAT
2071 static long tun_chr_compat_ioctl(struct file *file,
2072 unsigned int cmd, unsigned long arg)
2077 case TUNSETTXFILTER:
2082 arg = (unsigned long)compat_ptr(arg);
2085 arg = (compat_ulong_t)arg;
2090 * compat_ifreq is shorter than ifreq, so we must not access beyond
2091 * the end of that structure. All fields that are used in this
2092 * driver are compatible though, we don't need to convert the
2095 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2097 #endif /* CONFIG_COMPAT */
2099 static int tun_chr_fasync(int fd, struct file *file, int on)
2101 struct tun_file *tfile = file->private_data;
2104 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2108 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2111 tfile->flags |= TUN_FASYNC;
2113 tfile->flags &= ~TUN_FASYNC;
2119 static int tun_chr_open(struct inode *inode, struct file * file)
2121 struct tun_file *tfile;
2123 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2125 tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2129 rcu_assign_pointer(tfile->tun, NULL);
2130 tfile->net = get_net(current->nsproxy->net_ns);
2133 rcu_assign_pointer(tfile->socket.wq, &tfile->wq);
2134 init_waitqueue_head(&tfile->wq.wait);
2136 tfile->socket.file = file;
2137 tfile->socket.ops = &tun_socket_ops;
2139 sock_init_data(&tfile->socket, &tfile->sk);
2140 sk_change_net(&tfile->sk, tfile->net);
2142 tfile->sk.sk_write_space = tun_sock_write_space;
2143 tfile->sk.sk_sndbuf = INT_MAX;
2145 file->private_data = tfile;
2146 set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2147 INIT_LIST_HEAD(&tfile->next);
2152 static int tun_chr_close(struct inode *inode, struct file *file)
2154 struct tun_file *tfile = file->private_data;
2155 struct net *net = tfile->net;
2157 tun_detach(tfile, true);
2163 static const struct file_operations tun_fops = {
2164 .owner = THIS_MODULE,
2165 .llseek = no_llseek,
2166 .read = do_sync_read,
2167 .aio_read = tun_chr_aio_read,
2168 .write = do_sync_write,
2169 .aio_write = tun_chr_aio_write,
2170 .poll = tun_chr_poll,
2171 .unlocked_ioctl = tun_chr_ioctl,
2172 #ifdef CONFIG_COMPAT
2173 .compat_ioctl = tun_chr_compat_ioctl,
2175 .open = tun_chr_open,
2176 .release = tun_chr_close,
2177 .fasync = tun_chr_fasync
2180 static struct miscdevice tun_miscdev = {
2183 .nodename = "net/tun",
2187 /* ethtool interface */
2189 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2192 cmd->advertising = 0;
2193 ethtool_cmd_speed_set(cmd, SPEED_10);
2194 cmd->duplex = DUPLEX_FULL;
2195 cmd->port = PORT_TP;
2196 cmd->phy_address = 0;
2197 cmd->transceiver = XCVR_INTERNAL;
2198 cmd->autoneg = AUTONEG_DISABLE;
2204 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2206 struct tun_struct *tun = netdev_priv(dev);
2208 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2209 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2211 switch (tun->flags & TUN_TYPE_MASK) {
2213 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2216 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2221 static u32 tun_get_msglevel(struct net_device *dev)
2224 struct tun_struct *tun = netdev_priv(dev);
2231 static void tun_set_msglevel(struct net_device *dev, u32 value)
2234 struct tun_struct *tun = netdev_priv(dev);
2239 static const struct ethtool_ops tun_ethtool_ops = {
2240 .get_settings = tun_get_settings,
2241 .get_drvinfo = tun_get_drvinfo,
2242 .get_msglevel = tun_get_msglevel,
2243 .set_msglevel = tun_set_msglevel,
2244 .get_link = ethtool_op_get_link,
2248 static int __init tun_init(void)
2252 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2253 pr_info("%s\n", DRV_COPYRIGHT);
2255 ret = rtnl_link_register(&tun_link_ops);
2257 pr_err("Can't register link_ops\n");
2261 ret = misc_register(&tun_miscdev);
2263 pr_err("Can't register misc device %d\n", TUN_MINOR);
2268 rtnl_link_unregister(&tun_link_ops);
2273 static void tun_cleanup(void)
2275 misc_deregister(&tun_miscdev);
2276 rtnl_link_unregister(&tun_link_ops);
2279 /* Get an underlying socket object from tun file. Returns error unless file is
2280 * attached to a device. The returned object works like a packet socket, it
2281 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
2282 * holding a reference to the file for as long as the socket is in use. */
2283 struct socket *tun_get_socket(struct file *file)
2285 struct tun_file *tfile;
2286 if (file->f_op != &tun_fops)
2287 return ERR_PTR(-EINVAL);
2288 tfile = file->private_data;
2290 return ERR_PTR(-EBADFD);
2291 return &tfile->socket;
2293 EXPORT_SYMBOL_GPL(tun_get_socket);
2295 module_init(tun_init);
2296 module_exit(tun_cleanup);
2297 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2298 MODULE_AUTHOR(DRV_COPYRIGHT);
2299 MODULE_LICENSE("GPL");
2300 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2301 MODULE_ALIAS("devname:net/tun");