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 /* 1024 is probably a high enough limit: modern hypervisors seem to support on
113 * the order of 100-200 CPUs so this leaves us some breathing space if we want
114 * to match a queue per guest CPU.
116 #define MAX_TAP_QUEUES 1024
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;
190 static inline u32 tun_hashfn(u32 rxhash)
192 return rxhash & 0x3ff;
195 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
197 struct tun_flow_entry *e;
198 struct hlist_node *n;
200 hlist_for_each_entry_rcu(e, n, head, hash_link) {
201 if (e->rxhash == rxhash)
207 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
208 struct hlist_head *head,
209 u32 rxhash, u16 queue_index)
211 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
214 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
215 rxhash, queue_index);
216 e->updated = jiffies;
218 e->queue_index = queue_index;
220 hlist_add_head_rcu(&e->hash_link, head);
225 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
227 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
228 e->rxhash, e->queue_index);
229 hlist_del_rcu(&e->hash_link);
233 static void tun_flow_flush(struct tun_struct *tun)
237 spin_lock_bh(&tun->lock);
238 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
239 struct tun_flow_entry *e;
240 struct hlist_node *h, *n;
242 hlist_for_each_entry_safe(e, h, n, &tun->flows[i], hash_link)
243 tun_flow_delete(tun, e);
245 spin_unlock_bh(&tun->lock);
248 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
252 spin_lock_bh(&tun->lock);
253 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
254 struct tun_flow_entry *e;
255 struct hlist_node *h, *n;
257 hlist_for_each_entry_safe(e, h, n, &tun->flows[i], hash_link) {
258 if (e->queue_index == queue_index)
259 tun_flow_delete(tun, e);
262 spin_unlock_bh(&tun->lock);
265 static void tun_flow_cleanup(unsigned long data)
267 struct tun_struct *tun = (struct tun_struct *)data;
268 unsigned long delay = tun->ageing_time;
269 unsigned long next_timer = jiffies + delay;
270 unsigned long count = 0;
273 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
275 spin_lock_bh(&tun->lock);
276 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
277 struct tun_flow_entry *e;
278 struct hlist_node *h, *n;
280 hlist_for_each_entry_safe(e, h, n, &tun->flows[i], hash_link) {
281 unsigned long this_timer;
283 this_timer = e->updated + delay;
284 if (time_before_eq(this_timer, jiffies))
285 tun_flow_delete(tun, e);
286 else if (time_before(this_timer, next_timer))
287 next_timer = this_timer;
292 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
293 spin_unlock_bh(&tun->lock);
296 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
299 struct hlist_head *head;
300 struct tun_flow_entry *e;
301 unsigned long delay = tun->ageing_time;
306 head = &tun->flows[tun_hashfn(rxhash)];
310 if (tun->numqueues == 1)
313 e = tun_flow_find(head, rxhash);
315 /* TODO: keep queueing to old queue until it's empty? */
316 e->queue_index = queue_index;
317 e->updated = jiffies;
319 spin_lock_bh(&tun->lock);
320 if (!tun_flow_find(head, rxhash))
321 tun_flow_create(tun, head, rxhash, queue_index);
323 if (!timer_pending(&tun->flow_gc_timer))
324 mod_timer(&tun->flow_gc_timer,
325 round_jiffies_up(jiffies + delay));
326 spin_unlock_bh(&tun->lock);
333 /* We try to identify a flow through its rxhash first. The reason that
334 * we do not check rxq no. is becuase some cards(e.g 82599), chooses
335 * the rxq based on the txq where the last packet of the flow comes. As
336 * the userspace application move between processors, we may get a
337 * different rxq no. here. If we could not get rxhash, then we would
338 * hope the rxq no. may help here.
340 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb)
342 struct tun_struct *tun = netdev_priv(dev);
343 struct tun_flow_entry *e;
348 numqueues = tun->numqueues;
350 txq = skb_get_rxhash(skb);
352 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
354 txq = e->queue_index;
356 /* use multiply and shift instead of expensive divide */
357 txq = ((u64)txq * numqueues) >> 32;
358 } else if (likely(skb_rx_queue_recorded(skb))) {
359 txq = skb_get_rx_queue(skb);
360 while (unlikely(txq >= numqueues))
368 static inline bool tun_not_capable(struct tun_struct *tun)
370 const struct cred *cred = current_cred();
371 struct net *net = dev_net(tun->dev);
373 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
374 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
375 !ns_capable(net->user_ns, CAP_NET_ADMIN);
378 static void tun_set_real_num_queues(struct tun_struct *tun)
380 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
381 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
384 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
386 tfile->detached = tun;
387 list_add_tail(&tfile->next, &tun->disabled);
391 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
393 struct tun_struct *tun = tfile->detached;
395 tfile->detached = NULL;
396 list_del_init(&tfile->next);
401 static void __tun_detach(struct tun_file *tfile, bool clean)
403 struct tun_file *ntfile;
404 struct tun_struct *tun;
405 struct net_device *dev;
407 tun = rtnl_dereference(tfile->tun);
410 u16 index = tfile->queue_index;
411 BUG_ON(index >= tun->numqueues);
414 rcu_assign_pointer(tun->tfiles[index],
415 tun->tfiles[tun->numqueues - 1]);
416 rcu_assign_pointer(tfile->tun, NULL);
417 ntfile = rtnl_dereference(tun->tfiles[index]);
418 ntfile->queue_index = index;
422 sock_put(&tfile->sk);
424 tun_disable_queue(tun, tfile);
427 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
428 /* Drop read queue */
429 skb_queue_purge(&tfile->sk.sk_receive_queue);
430 tun_set_real_num_queues(tun);
431 } else if (tfile->detached && clean)
432 tun = tun_enable_queue(tfile);
435 if (tun && tun->numqueues == 0 && tun->numdisabled == 0 &&
436 !(tun->flags & TUN_PERSIST))
437 if (tun->dev->reg_state == NETREG_REGISTERED)
438 unregister_netdevice(tun->dev);
440 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
441 &tfile->socket.flags));
442 sk_release_kernel(&tfile->sk);
446 static void tun_detach(struct tun_file *tfile, bool clean)
449 __tun_detach(tfile, clean);
453 static void tun_detach_all(struct net_device *dev)
455 struct tun_struct *tun = netdev_priv(dev);
456 struct tun_file *tfile, *tmp;
457 int i, n = tun->numqueues;
459 for (i = 0; i < n; i++) {
460 tfile = rtnl_dereference(tun->tfiles[i]);
462 wake_up_all(&tfile->wq.wait);
463 rcu_assign_pointer(tfile->tun, NULL);
466 BUG_ON(tun->numqueues != 0);
469 for (i = 0; i < n; i++) {
470 tfile = rtnl_dereference(tun->tfiles[i]);
471 /* Drop read queue */
472 skb_queue_purge(&tfile->sk.sk_receive_queue);
473 sock_put(&tfile->sk);
475 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
476 tun_enable_queue(tfile);
477 skb_queue_purge(&tfile->sk.sk_receive_queue);
478 sock_put(&tfile->sk);
480 BUG_ON(tun->numdisabled != 0);
483 static int tun_attach(struct tun_struct *tun, struct file *file)
485 struct tun_file *tfile = file->private_data;
489 if (rtnl_dereference(tfile->tun))
491 if (tfile->detached && tun != tfile->detached)
495 if (!(tun->flags & TUN_TAP_MQ) && tun->numqueues == 1)
499 if (!tfile->detached &&
500 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
505 /* Re-attach the filter to presist device */
506 if (tun->filter_attached == true) {
507 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
511 tfile->queue_index = tun->numqueues;
512 rcu_assign_pointer(tfile->tun, tun);
513 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
517 tun_enable_queue(tfile);
519 sock_hold(&tfile->sk);
521 tun_set_real_num_queues(tun);
523 /* device is allowed to go away first, so no need to hold extra
531 static struct tun_struct *__tun_get(struct tun_file *tfile)
533 struct tun_struct *tun;
536 tun = rcu_dereference(tfile->tun);
544 static struct tun_struct *tun_get(struct file *file)
546 return __tun_get(file->private_data);
549 static void tun_put(struct tun_struct *tun)
555 static void addr_hash_set(u32 *mask, const u8 *addr)
557 int n = ether_crc(ETH_ALEN, addr) >> 26;
558 mask[n >> 5] |= (1 << (n & 31));
561 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
563 int n = ether_crc(ETH_ALEN, addr) >> 26;
564 return mask[n >> 5] & (1 << (n & 31));
567 static int update_filter(struct tap_filter *filter, void __user *arg)
569 struct { u8 u[ETH_ALEN]; } *addr;
570 struct tun_filter uf;
571 int err, alen, n, nexact;
573 if (copy_from_user(&uf, arg, sizeof(uf)))
582 alen = ETH_ALEN * uf.count;
583 addr = kmalloc(alen, GFP_KERNEL);
587 if (copy_from_user(addr, arg + sizeof(uf), alen)) {
592 /* The filter is updated without holding any locks. Which is
593 * perfectly safe. We disable it first and in the worst
594 * case we'll accept a few undesired packets. */
598 /* Use first set of addresses as an exact filter */
599 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
600 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
604 /* Remaining multicast addresses are hashed,
605 * unicast will leave the filter disabled. */
606 memset(filter->mask, 0, sizeof(filter->mask));
607 for (; n < uf.count; n++) {
608 if (!is_multicast_ether_addr(addr[n].u)) {
609 err = 0; /* no filter */
612 addr_hash_set(filter->mask, addr[n].u);
615 /* For ALLMULTI just set the mask to all ones.
616 * This overrides the mask populated above. */
617 if ((uf.flags & TUN_FLT_ALLMULTI))
618 memset(filter->mask, ~0, sizeof(filter->mask));
620 /* Now enable the filter */
622 filter->count = nexact;
624 /* Return the number of exact filters */
632 /* Returns: 0 - drop, !=0 - accept */
633 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
635 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
637 struct ethhdr *eh = (struct ethhdr *) skb->data;
641 for (i = 0; i < filter->count; i++)
642 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
645 /* Inexact match (multicast only) */
646 if (is_multicast_ether_addr(eh->h_dest))
647 return addr_hash_test(filter->mask, eh->h_dest);
653 * Checks whether the packet is accepted or not.
654 * Returns: 0 - drop, !=0 - accept
656 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
661 return run_filter(filter, skb);
664 /* Network device part of the driver */
666 static const struct ethtool_ops tun_ethtool_ops;
668 /* Net device detach from fd. */
669 static void tun_net_uninit(struct net_device *dev)
674 /* Net device open. */
675 static int tun_net_open(struct net_device *dev)
677 netif_tx_start_all_queues(dev);
681 /* Net device close. */
682 static int tun_net_close(struct net_device *dev)
684 netif_tx_stop_all_queues(dev);
688 /* Net device start xmit */
689 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
691 struct tun_struct *tun = netdev_priv(dev);
692 int txq = skb->queue_mapping;
693 struct tun_file *tfile;
696 tfile = rcu_dereference(tun->tfiles[txq]);
698 /* Drop packet if interface is not attached */
699 if (txq >= tun->numqueues)
702 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
706 /* Drop if the filter does not like it.
707 * This is a noop if the filter is disabled.
708 * Filter can be enabled only for the TAP devices. */
709 if (!check_filter(&tun->txflt, skb))
712 if (tfile->socket.sk->sk_filter &&
713 sk_filter(tfile->socket.sk, skb))
716 /* Limit the number of packets queued by dividing txq length with the
719 if (skb_queue_len(&tfile->socket.sk->sk_receive_queue)
720 >= dev->tx_queue_len / tun->numqueues)
723 /* Orphan the skb - required as we might hang on to it
724 * for indefinite time. */
725 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
730 skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
732 /* Notify and wake up reader process */
733 if (tfile->flags & TUN_FASYNC)
734 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
735 wake_up_interruptible_poll(&tfile->wq.wait, POLLIN |
736 POLLRDNORM | POLLRDBAND);
742 dev->stats.tx_dropped++;
749 static void tun_net_mclist(struct net_device *dev)
752 * This callback is supposed to deal with mc filter in
753 * _rx_ path and has nothing to do with the _tx_ path.
754 * In rx path we always accept everything userspace gives us.
759 #define MAX_MTU 65535
762 tun_net_change_mtu(struct net_device *dev, int new_mtu)
764 if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
770 static netdev_features_t tun_net_fix_features(struct net_device *dev,
771 netdev_features_t features)
773 struct tun_struct *tun = netdev_priv(dev);
775 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
777 #ifdef CONFIG_NET_POLL_CONTROLLER
778 static void tun_poll_controller(struct net_device *dev)
781 * Tun only receives frames when:
782 * 1) the char device endpoint gets data from user space
783 * 2) the tun socket gets a sendmsg call from user space
784 * Since both of those are syncronous operations, we are guaranteed
785 * never to have pending data when we poll for it
786 * so theres nothing to do here but return.
787 * We need this though so netpoll recognizes us as an interface that
788 * supports polling, which enables bridge devices in virt setups to
789 * still use netconsole
794 static const struct net_device_ops tun_netdev_ops = {
795 .ndo_uninit = tun_net_uninit,
796 .ndo_open = tun_net_open,
797 .ndo_stop = tun_net_close,
798 .ndo_start_xmit = tun_net_xmit,
799 .ndo_change_mtu = tun_net_change_mtu,
800 .ndo_fix_features = tun_net_fix_features,
801 .ndo_select_queue = tun_select_queue,
802 #ifdef CONFIG_NET_POLL_CONTROLLER
803 .ndo_poll_controller = tun_poll_controller,
807 static const struct net_device_ops tap_netdev_ops = {
808 .ndo_uninit = tun_net_uninit,
809 .ndo_open = tun_net_open,
810 .ndo_stop = tun_net_close,
811 .ndo_start_xmit = tun_net_xmit,
812 .ndo_change_mtu = tun_net_change_mtu,
813 .ndo_fix_features = tun_net_fix_features,
814 .ndo_set_rx_mode = tun_net_mclist,
815 .ndo_set_mac_address = eth_mac_addr,
816 .ndo_validate_addr = eth_validate_addr,
817 .ndo_select_queue = tun_select_queue,
818 #ifdef CONFIG_NET_POLL_CONTROLLER
819 .ndo_poll_controller = tun_poll_controller,
823 static int tun_flow_init(struct tun_struct *tun)
827 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
828 INIT_HLIST_HEAD(&tun->flows[i]);
830 tun->ageing_time = TUN_FLOW_EXPIRE;
831 setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
832 mod_timer(&tun->flow_gc_timer,
833 round_jiffies_up(jiffies + tun->ageing_time));
838 static void tun_flow_uninit(struct tun_struct *tun)
840 del_timer_sync(&tun->flow_gc_timer);
844 /* Initialize net device. */
845 static void tun_net_init(struct net_device *dev)
847 struct tun_struct *tun = netdev_priv(dev);
849 switch (tun->flags & TUN_TYPE_MASK) {
851 dev->netdev_ops = &tun_netdev_ops;
853 /* Point-to-Point TUN Device */
854 dev->hard_header_len = 0;
858 /* Zero header length */
859 dev->type = ARPHRD_NONE;
860 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
861 dev->tx_queue_len = TUN_READQ_SIZE; /* We prefer our own queue length */
865 dev->netdev_ops = &tap_netdev_ops;
866 /* Ethernet TAP Device */
868 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
869 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
871 eth_hw_addr_random(dev);
873 dev->tx_queue_len = TUN_READQ_SIZE; /* We prefer our own queue length */
878 /* Character device part */
881 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
883 struct tun_file *tfile = file->private_data;
884 struct tun_struct *tun = __tun_get(tfile);
886 unsigned int mask = 0;
891 sk = tfile->socket.sk;
893 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
895 poll_wait(file, &tfile->wq.wait, wait);
897 if (!skb_queue_empty(&sk->sk_receive_queue))
898 mask |= POLLIN | POLLRDNORM;
900 if (sock_writeable(sk) ||
901 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
903 mask |= POLLOUT | POLLWRNORM;
905 if (tun->dev->reg_state != NETREG_REGISTERED)
912 /* prepad is the amount to reserve at front. len is length after that.
913 * linear is a hint as to how much to copy (usually headers). */
914 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
915 size_t prepad, size_t len,
916 size_t linear, int noblock)
918 struct sock *sk = tfile->socket.sk;
922 /* Under a page? Don't bother with paged skb. */
923 if (prepad + len < PAGE_SIZE || !linear)
926 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
931 skb_reserve(skb, prepad);
932 skb_put(skb, linear);
933 skb->data_len = len - linear;
934 skb->len += len - linear;
939 /* set skb frags from iovec, this can move to core network code for reuse */
940 static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
941 int offset, size_t count)
943 int len = iov_length(from, count) - offset;
944 int copy = skb_headlen(skb);
945 int size, offset1 = 0;
948 /* Skip over from offset */
949 while (count && (offset >= from->iov_len)) {
950 offset -= from->iov_len;
955 /* copy up to skb headlen */
956 while (count && (copy > 0)) {
957 size = min_t(unsigned int, copy, from->iov_len - offset);
958 if (copy_from_user(skb->data + offset1, from->iov_base + offset,
975 struct page *page[MAX_SKB_FRAGS];
978 unsigned long truesize;
980 len = from->iov_len - offset;
986 base = (unsigned long)from->iov_base + offset;
987 size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
988 if (i + size > MAX_SKB_FRAGS)
990 num_pages = get_user_pages_fast(base, size, 0, &page[i]);
991 if (num_pages != size) {
992 for (i = 0; i < num_pages; i++)
996 truesize = size * PAGE_SIZE;
997 skb->data_len += len;
999 skb->truesize += truesize;
1000 atomic_add(truesize, &skb->sk->sk_wmem_alloc);
1002 int off = base & ~PAGE_MASK;
1003 int size = min_t(int, len, PAGE_SIZE - off);
1004 __skb_fill_page_desc(skb, i, page[i], off, size);
1005 skb_shinfo(skb)->nr_frags++;
1006 /* increase sk_wmem_alloc */
1017 /* Get packet from user space buffer */
1018 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1019 void *msg_control, const struct iovec *iv,
1020 size_t total_len, size_t count, int noblock)
1022 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1023 struct sk_buff *skb;
1024 size_t len = total_len, align = NET_SKB_PAD;
1025 struct virtio_net_hdr gso = { 0 };
1028 bool zerocopy = false;
1032 if (!(tun->flags & TUN_NO_PI)) {
1033 if ((len -= sizeof(pi)) > total_len)
1036 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
1038 offset += sizeof(pi);
1041 if (tun->flags & TUN_VNET_HDR) {
1042 if ((len -= tun->vnet_hdr_sz) > total_len)
1045 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1048 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1049 gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1050 gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1052 if (gso.hdr_len > len)
1054 offset += tun->vnet_hdr_sz;
1057 if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1058 align += NET_IP_ALIGN;
1059 if (unlikely(len < ETH_HLEN ||
1060 (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1068 /* Userspace may produce vectors with count greater than
1069 * MAX_SKB_FRAGS, so we need to linearize parts of the skb
1070 * to let the rest of data to be fit in the frags.
1072 if (count > MAX_SKB_FRAGS) {
1073 copylen = iov_length(iv, count - MAX_SKB_FRAGS);
1074 if (copylen < offset)
1080 /* There are 256 bytes to be copied in skb, so there is enough
1081 * room for skb expand head in case it is used.
1082 * The rest of the buffer is mapped from userspace.
1084 if (copylen < gso.hdr_len)
1085 copylen = gso.hdr_len;
1087 copylen = GOODCOPY_LEN;
1091 skb = tun_alloc_skb(tfile, align, copylen, gso.hdr_len, noblock);
1093 if (PTR_ERR(skb) != -EAGAIN)
1094 tun->dev->stats.rx_dropped++;
1095 return PTR_ERR(skb);
1099 err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1101 err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1104 tun->dev->stats.rx_dropped++;
1109 if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1110 if (!skb_partial_csum_set(skb, gso.csum_start,
1112 tun->dev->stats.rx_frame_errors++;
1118 switch (tun->flags & TUN_TYPE_MASK) {
1120 if (tun->flags & TUN_NO_PI) {
1121 switch (skb->data[0] & 0xf0) {
1123 pi.proto = htons(ETH_P_IP);
1126 pi.proto = htons(ETH_P_IPV6);
1129 tun->dev->stats.rx_dropped++;
1135 skb_reset_mac_header(skb);
1136 skb->protocol = pi.proto;
1137 skb->dev = tun->dev;
1140 skb->protocol = eth_type_trans(skb, tun->dev);
1144 if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1146 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1147 case VIRTIO_NET_HDR_GSO_TCPV4:
1148 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1150 case VIRTIO_NET_HDR_GSO_TCPV6:
1151 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1153 case VIRTIO_NET_HDR_GSO_UDP:
1154 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1157 tun->dev->stats.rx_frame_errors++;
1162 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1163 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1165 skb_shinfo(skb)->gso_size = gso.gso_size;
1166 if (skb_shinfo(skb)->gso_size == 0) {
1167 tun->dev->stats.rx_frame_errors++;
1172 /* Header must be checked, and gso_segs computed. */
1173 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1174 skb_shinfo(skb)->gso_segs = 0;
1177 /* copy skb_ubuf_info for callback when skb has no error */
1179 skb_shinfo(skb)->destructor_arg = msg_control;
1180 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1183 skb_reset_network_header(skb);
1184 rxhash = skb_get_rxhash(skb);
1187 tun->dev->stats.rx_packets++;
1188 tun->dev->stats.rx_bytes += len;
1190 tun_flow_update(tun, rxhash, tfile->queue_index);
1194 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1195 unsigned long count, loff_t pos)
1197 struct file *file = iocb->ki_filp;
1198 struct tun_struct *tun = tun_get(file);
1199 struct tun_file *tfile = file->private_data;
1205 tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1207 result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1208 count, file->f_flags & O_NONBLOCK);
1214 /* Put packet to the user space buffer */
1215 static ssize_t tun_put_user(struct tun_struct *tun,
1216 struct tun_file *tfile,
1217 struct sk_buff *skb,
1218 const struct iovec *iv, int len)
1220 struct tun_pi pi = { 0, skb->protocol };
1223 if (!(tun->flags & TUN_NO_PI)) {
1224 if ((len -= sizeof(pi)) < 0)
1227 if (len < skb->len) {
1228 /* Packet will be striped */
1229 pi.flags |= TUN_PKT_STRIP;
1232 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1234 total += sizeof(pi);
1237 if (tun->flags & TUN_VNET_HDR) {
1238 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1239 if ((len -= tun->vnet_hdr_sz) < 0)
1242 if (skb_is_gso(skb)) {
1243 struct skb_shared_info *sinfo = skb_shinfo(skb);
1245 /* This is a hint as to how much should be linear. */
1246 gso.hdr_len = skb_headlen(skb);
1247 gso.gso_size = sinfo->gso_size;
1248 if (sinfo->gso_type & SKB_GSO_TCPV4)
1249 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1250 else if (sinfo->gso_type & SKB_GSO_TCPV6)
1251 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1252 else if (sinfo->gso_type & SKB_GSO_UDP)
1253 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1255 pr_err("unexpected GSO type: "
1256 "0x%x, gso_size %d, hdr_len %d\n",
1257 sinfo->gso_type, gso.gso_size,
1259 print_hex_dump(KERN_ERR, "tun: ",
1262 min((int)gso.hdr_len, 64), true);
1266 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1267 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1269 gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1271 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1272 gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1273 gso.csum_start = skb_checksum_start_offset(skb);
1274 gso.csum_offset = skb->csum_offset;
1275 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1276 gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1277 } /* else everything is zero */
1279 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1282 total += tun->vnet_hdr_sz;
1285 len = min_t(int, skb->len, len);
1287 skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
1290 tun->dev->stats.tx_packets++;
1291 tun->dev->stats.tx_bytes += len;
1296 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1297 struct kiocb *iocb, const struct iovec *iv,
1298 ssize_t len, int noblock)
1300 DECLARE_WAITQUEUE(wait, current);
1301 struct sk_buff *skb;
1304 tun_debug(KERN_INFO, tun, "tun_do_read\n");
1306 if (unlikely(!noblock))
1307 add_wait_queue(&tfile->wq.wait, &wait);
1309 current->state = TASK_INTERRUPTIBLE;
1311 /* Read frames from the queue */
1312 if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) {
1317 if (signal_pending(current)) {
1321 if (tun->dev->reg_state != NETREG_REGISTERED) {
1326 /* Nothing to read, let's sleep */
1331 ret = tun_put_user(tun, tfile, skb, iv, len);
1336 current->state = TASK_RUNNING;
1337 if (unlikely(!noblock))
1338 remove_wait_queue(&tfile->wq.wait, &wait);
1343 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1344 unsigned long count, loff_t pos)
1346 struct file *file = iocb->ki_filp;
1347 struct tun_file *tfile = file->private_data;
1348 struct tun_struct *tun = __tun_get(tfile);
1353 len = iov_length(iv, count);
1359 ret = tun_do_read(tun, tfile, iocb, iv, len,
1360 file->f_flags & O_NONBLOCK);
1361 ret = min_t(ssize_t, ret, len);
1367 static void tun_free_netdev(struct net_device *dev)
1369 struct tun_struct *tun = netdev_priv(dev);
1371 BUG_ON(!(list_empty(&tun->disabled)));
1372 tun_flow_uninit(tun);
1376 static void tun_setup(struct net_device *dev)
1378 struct tun_struct *tun = netdev_priv(dev);
1380 tun->owner = INVALID_UID;
1381 tun->group = INVALID_GID;
1383 dev->ethtool_ops = &tun_ethtool_ops;
1384 dev->destructor = tun_free_netdev;
1387 /* Trivial set of netlink ops to allow deleting tun or tap
1388 * device with netlink.
1390 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1395 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1397 .priv_size = sizeof(struct tun_struct),
1399 .validate = tun_validate,
1402 static void tun_sock_write_space(struct sock *sk)
1404 struct tun_file *tfile;
1405 wait_queue_head_t *wqueue;
1407 if (!sock_writeable(sk))
1410 if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1413 wqueue = sk_sleep(sk);
1414 if (wqueue && waitqueue_active(wqueue))
1415 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1416 POLLWRNORM | POLLWRBAND);
1418 tfile = container_of(sk, struct tun_file, sk);
1419 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1422 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1423 struct msghdr *m, size_t total_len)
1426 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1427 struct tun_struct *tun = __tun_get(tfile);
1431 ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1432 m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1438 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1439 struct msghdr *m, size_t total_len,
1442 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1443 struct tun_struct *tun = __tun_get(tfile);
1449 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1451 ret = tun_do_read(tun, tfile, iocb, m->msg_iov, total_len,
1452 flags & MSG_DONTWAIT);
1453 if (ret > total_len) {
1454 m->msg_flags |= MSG_TRUNC;
1455 ret = flags & MSG_TRUNC ? ret : total_len;
1461 static int tun_release(struct socket *sock)
1468 /* Ops structure to mimic raw sockets with tun */
1469 static const struct proto_ops tun_socket_ops = {
1470 .sendmsg = tun_sendmsg,
1471 .recvmsg = tun_recvmsg,
1472 .release = tun_release,
1475 static struct proto tun_proto = {
1477 .owner = THIS_MODULE,
1478 .obj_size = sizeof(struct tun_file),
1481 static int tun_flags(struct tun_struct *tun)
1485 if (tun->flags & TUN_TUN_DEV)
1490 if (tun->flags & TUN_NO_PI)
1493 /* This flag has no real effect. We track the value for backwards
1496 if (tun->flags & TUN_ONE_QUEUE)
1497 flags |= IFF_ONE_QUEUE;
1499 if (tun->flags & TUN_VNET_HDR)
1500 flags |= IFF_VNET_HDR;
1502 if (tun->flags & TUN_TAP_MQ)
1503 flags |= IFF_MULTI_QUEUE;
1508 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1511 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1512 return sprintf(buf, "0x%x\n", tun_flags(tun));
1515 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1518 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1519 return uid_valid(tun->owner)?
1520 sprintf(buf, "%u\n",
1521 from_kuid_munged(current_user_ns(), tun->owner)):
1522 sprintf(buf, "-1\n");
1525 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1528 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1529 return gid_valid(tun->group) ?
1530 sprintf(buf, "%u\n",
1531 from_kgid_munged(current_user_ns(), tun->group)):
1532 sprintf(buf, "-1\n");
1535 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1536 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1537 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1539 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1541 struct tun_struct *tun;
1542 struct tun_file *tfile = file->private_data;
1543 struct net_device *dev;
1546 dev = __dev_get_by_name(net, ifr->ifr_name);
1548 if (ifr->ifr_flags & IFF_TUN_EXCL)
1550 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1551 tun = netdev_priv(dev);
1552 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1553 tun = netdev_priv(dev);
1557 if (tun_not_capable(tun))
1559 err = security_tun_dev_attach(tfile->socket.sk);
1563 err = tun_attach(tun, file);
1567 if (tun->flags & TUN_TAP_MQ &&
1568 (tun->numqueues + tun->numdisabled > 1))
1573 unsigned long flags = 0;
1575 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1577 err = security_tun_dev_create();
1582 if (ifr->ifr_flags & IFF_TUN) {
1584 flags |= TUN_TUN_DEV;
1586 } else if (ifr->ifr_flags & IFF_TAP) {
1588 flags |= TUN_TAP_DEV;
1594 name = ifr->ifr_name;
1596 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1598 MAX_TAP_QUEUES, MAX_TAP_QUEUES);
1602 dev_net_set(dev, net);
1603 dev->rtnl_link_ops = &tun_link_ops;
1605 tun = netdev_priv(dev);
1608 tun->txflt.count = 0;
1609 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1611 tun->filter_attached = false;
1612 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1614 spin_lock_init(&tun->lock);
1616 security_tun_dev_post_create(&tfile->sk);
1620 err = tun_flow_init(tun);
1624 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1626 dev->features = dev->hw_features;
1628 INIT_LIST_HEAD(&tun->disabled);
1629 err = tun_attach(tun, file);
1633 err = register_netdevice(tun->dev);
1637 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1638 device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1639 device_create_file(&tun->dev->dev, &dev_attr_group))
1640 pr_err("Failed to create tun sysfs files\n");
1642 netif_carrier_on(tun->dev);
1645 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1647 if (ifr->ifr_flags & IFF_NO_PI)
1648 tun->flags |= TUN_NO_PI;
1650 tun->flags &= ~TUN_NO_PI;
1652 /* This flag has no real effect. We track the value for backwards
1655 if (ifr->ifr_flags & IFF_ONE_QUEUE)
1656 tun->flags |= TUN_ONE_QUEUE;
1658 tun->flags &= ~TUN_ONE_QUEUE;
1660 if (ifr->ifr_flags & IFF_VNET_HDR)
1661 tun->flags |= TUN_VNET_HDR;
1663 tun->flags &= ~TUN_VNET_HDR;
1665 if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1666 tun->flags |= TUN_TAP_MQ;
1668 tun->flags &= ~TUN_TAP_MQ;
1670 /* Make sure persistent devices do not get stuck in
1673 if (netif_running(tun->dev))
1674 netif_tx_wake_all_queues(tun->dev);
1676 strcpy(ifr->ifr_name, tun->dev->name);
1684 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1687 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1689 strcpy(ifr->ifr_name, tun->dev->name);
1691 ifr->ifr_flags = tun_flags(tun);
1695 /* This is like a cut-down ethtool ops, except done via tun fd so no
1696 * privs required. */
1697 static int set_offload(struct tun_struct *tun, unsigned long arg)
1699 netdev_features_t features = 0;
1701 if (arg & TUN_F_CSUM) {
1702 features |= NETIF_F_HW_CSUM;
1705 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1706 if (arg & TUN_F_TSO_ECN) {
1707 features |= NETIF_F_TSO_ECN;
1708 arg &= ~TUN_F_TSO_ECN;
1710 if (arg & TUN_F_TSO4)
1711 features |= NETIF_F_TSO;
1712 if (arg & TUN_F_TSO6)
1713 features |= NETIF_F_TSO6;
1714 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1717 if (arg & TUN_F_UFO) {
1718 features |= NETIF_F_UFO;
1723 /* This gives the user a way to test for new features in future by
1724 * trying to set them. */
1728 tun->set_features = features;
1729 netdev_update_features(tun->dev);
1734 static void tun_detach_filter(struct tun_struct *tun, int n)
1737 struct tun_file *tfile;
1739 for (i = 0; i < n; i++) {
1740 tfile = rtnl_dereference(tun->tfiles[i]);
1741 sk_detach_filter(tfile->socket.sk);
1744 tun->filter_attached = false;
1747 static int tun_attach_filter(struct tun_struct *tun)
1750 struct tun_file *tfile;
1752 for (i = 0; i < tun->numqueues; i++) {
1753 tfile = rtnl_dereference(tun->tfiles[i]);
1754 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1756 tun_detach_filter(tun, i);
1761 tun->filter_attached = true;
1765 static void tun_set_sndbuf(struct tun_struct *tun)
1767 struct tun_file *tfile;
1770 for (i = 0; i < tun->numqueues; i++) {
1771 tfile = rtnl_dereference(tun->tfiles[i]);
1772 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1776 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1778 struct tun_file *tfile = file->private_data;
1779 struct tun_struct *tun;
1784 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1785 tun = tfile->detached;
1789 ret = tun_attach(tun, file);
1790 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1791 tun = rtnl_dereference(tfile->tun);
1792 if (!tun || !(tun->flags & TUN_TAP_MQ))
1795 __tun_detach(tfile, false);
1803 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1804 unsigned long arg, int ifreq_len)
1806 struct tun_file *tfile = file->private_data;
1807 struct tun_struct *tun;
1808 void __user* argp = (void __user*)arg;
1816 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1817 if (copy_from_user(&ifr, argp, ifreq_len))
1820 memset(&ifr, 0, sizeof(ifr));
1822 if (cmd == TUNGETFEATURES) {
1823 /* Currently this just means: "what IFF flags are valid?".
1824 * This is needed because we never checked for invalid flags on
1826 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1827 IFF_VNET_HDR | IFF_MULTI_QUEUE,
1828 (unsigned int __user*)argp);
1829 } else if (cmd == TUNSETQUEUE)
1830 return tun_set_queue(file, &ifr);
1835 tun = __tun_get(tfile);
1836 if (cmd == TUNSETIFF && !tun) {
1837 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1839 ret = tun_set_iff(tfile->net, file, &ifr);
1844 if (copy_to_user(argp, &ifr, ifreq_len))
1853 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1858 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1860 if (copy_to_user(argp, &ifr, ifreq_len))
1865 /* Disable/Enable checksum */
1867 /* [unimplemented] */
1868 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1869 arg ? "disabled" : "enabled");
1873 /* Disable/Enable persist mode. Keep an extra reference to the
1874 * module to prevent the module being unprobed.
1877 tun->flags |= TUN_PERSIST;
1878 __module_get(THIS_MODULE);
1880 tun->flags &= ~TUN_PERSIST;
1881 module_put(THIS_MODULE);
1884 tun_debug(KERN_INFO, tun, "persist %s\n",
1885 arg ? "enabled" : "disabled");
1889 /* Set owner of the device */
1890 owner = make_kuid(current_user_ns(), arg);
1891 if (!uid_valid(owner)) {
1896 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1897 from_kuid(&init_user_ns, tun->owner));
1901 /* Set group of the device */
1902 group = make_kgid(current_user_ns(), arg);
1903 if (!gid_valid(group)) {
1908 tun_debug(KERN_INFO, tun, "group set to %u\n",
1909 from_kgid(&init_user_ns, tun->group));
1913 /* Only allow setting the type when the interface is down */
1914 if (tun->dev->flags & IFF_UP) {
1915 tun_debug(KERN_INFO, tun,
1916 "Linktype set failed because interface is up\n");
1919 tun->dev->type = (int) arg;
1920 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1932 ret = set_offload(tun, arg);
1935 case TUNSETTXFILTER:
1936 /* Can be set only for TAPs */
1938 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1940 ret = update_filter(&tun->txflt, (void __user *)arg);
1944 /* Get hw address */
1945 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1946 ifr.ifr_hwaddr.sa_family = tun->dev->type;
1947 if (copy_to_user(argp, &ifr, ifreq_len))
1952 /* Set hw address */
1953 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1954 ifr.ifr_hwaddr.sa_data);
1956 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1960 sndbuf = tfile->socket.sk->sk_sndbuf;
1961 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1966 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1971 tun->sndbuf = sndbuf;
1972 tun_set_sndbuf(tun);
1975 case TUNGETVNETHDRSZ:
1976 vnet_hdr_sz = tun->vnet_hdr_sz;
1977 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
1981 case TUNSETVNETHDRSZ:
1982 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
1986 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
1991 tun->vnet_hdr_sz = vnet_hdr_sz;
1994 case TUNATTACHFILTER:
1995 /* Can be set only for TAPs */
1997 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2000 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2003 ret = tun_attach_filter(tun);
2006 case TUNDETACHFILTER:
2007 /* Can be set only for TAPs */
2009 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2012 tun_detach_filter(tun, tun->numqueues);
2027 static long tun_chr_ioctl(struct file *file,
2028 unsigned int cmd, unsigned long arg)
2030 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2033 #ifdef CONFIG_COMPAT
2034 static long tun_chr_compat_ioctl(struct file *file,
2035 unsigned int cmd, unsigned long arg)
2040 case TUNSETTXFILTER:
2045 arg = (unsigned long)compat_ptr(arg);
2048 arg = (compat_ulong_t)arg;
2053 * compat_ifreq is shorter than ifreq, so we must not access beyond
2054 * the end of that structure. All fields that are used in this
2055 * driver are compatible though, we don't need to convert the
2058 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2060 #endif /* CONFIG_COMPAT */
2062 static int tun_chr_fasync(int fd, struct file *file, int on)
2064 struct tun_file *tfile = file->private_data;
2067 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2071 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2074 tfile->flags |= TUN_FASYNC;
2076 tfile->flags &= ~TUN_FASYNC;
2082 static int tun_chr_open(struct inode *inode, struct file * file)
2084 struct tun_file *tfile;
2086 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2088 tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2092 rcu_assign_pointer(tfile->tun, NULL);
2093 tfile->net = get_net(current->nsproxy->net_ns);
2096 rcu_assign_pointer(tfile->socket.wq, &tfile->wq);
2097 init_waitqueue_head(&tfile->wq.wait);
2099 tfile->socket.file = file;
2100 tfile->socket.ops = &tun_socket_ops;
2102 sock_init_data(&tfile->socket, &tfile->sk);
2103 sk_change_net(&tfile->sk, tfile->net);
2105 tfile->sk.sk_write_space = tun_sock_write_space;
2106 tfile->sk.sk_sndbuf = INT_MAX;
2108 file->private_data = tfile;
2109 set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2110 INIT_LIST_HEAD(&tfile->next);
2115 static int tun_chr_close(struct inode *inode, struct file *file)
2117 struct tun_file *tfile = file->private_data;
2118 struct net *net = tfile->net;
2120 tun_detach(tfile, true);
2126 static const struct file_operations tun_fops = {
2127 .owner = THIS_MODULE,
2128 .llseek = no_llseek,
2129 .read = do_sync_read,
2130 .aio_read = tun_chr_aio_read,
2131 .write = do_sync_write,
2132 .aio_write = tun_chr_aio_write,
2133 .poll = tun_chr_poll,
2134 .unlocked_ioctl = tun_chr_ioctl,
2135 #ifdef CONFIG_COMPAT
2136 .compat_ioctl = tun_chr_compat_ioctl,
2138 .open = tun_chr_open,
2139 .release = tun_chr_close,
2140 .fasync = tun_chr_fasync
2143 static struct miscdevice tun_miscdev = {
2146 .nodename = "net/tun",
2150 /* ethtool interface */
2152 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2155 cmd->advertising = 0;
2156 ethtool_cmd_speed_set(cmd, SPEED_10);
2157 cmd->duplex = DUPLEX_FULL;
2158 cmd->port = PORT_TP;
2159 cmd->phy_address = 0;
2160 cmd->transceiver = XCVR_INTERNAL;
2161 cmd->autoneg = AUTONEG_DISABLE;
2167 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2169 struct tun_struct *tun = netdev_priv(dev);
2171 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2172 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2174 switch (tun->flags & TUN_TYPE_MASK) {
2176 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2179 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2184 static u32 tun_get_msglevel(struct net_device *dev)
2187 struct tun_struct *tun = netdev_priv(dev);
2194 static void tun_set_msglevel(struct net_device *dev, u32 value)
2197 struct tun_struct *tun = netdev_priv(dev);
2202 static const struct ethtool_ops tun_ethtool_ops = {
2203 .get_settings = tun_get_settings,
2204 .get_drvinfo = tun_get_drvinfo,
2205 .get_msglevel = tun_get_msglevel,
2206 .set_msglevel = tun_set_msglevel,
2207 .get_link = ethtool_op_get_link,
2211 static int __init tun_init(void)
2215 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2216 pr_info("%s\n", DRV_COPYRIGHT);
2218 ret = rtnl_link_register(&tun_link_ops);
2220 pr_err("Can't register link_ops\n");
2224 ret = misc_register(&tun_miscdev);
2226 pr_err("Can't register misc device %d\n", TUN_MINOR);
2231 rtnl_link_unregister(&tun_link_ops);
2236 static void tun_cleanup(void)
2238 misc_deregister(&tun_miscdev);
2239 rtnl_link_unregister(&tun_link_ops);
2242 /* Get an underlying socket object from tun file. Returns error unless file is
2243 * attached to a device. The returned object works like a packet socket, it
2244 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
2245 * holding a reference to the file for as long as the socket is in use. */
2246 struct socket *tun_get_socket(struct file *file)
2248 struct tun_file *tfile;
2249 if (file->f_op != &tun_fops)
2250 return ERR_PTR(-EINVAL);
2251 tfile = file->private_data;
2253 return ERR_PTR(-EBADFD);
2254 return &tfile->socket;
2256 EXPORT_SYMBOL_GPL(tun_get_socket);
2258 module_init(tun_init);
2259 module_exit(tun_cleanup);
2260 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2261 MODULE_AUTHOR(DRV_COPYRIGHT);
2262 MODULE_LICENSE("GPL");
2263 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2264 MODULE_ALIAS("devname:net/tun");