2 * TUN - Universal TUN/TAP device driver.
3 * Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
21 * Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
22 * Add TUNSETLINK ioctl to set the link encapsulation
24 * Mark Smith <markzzzsmith@yahoo.com.au>
25 * Use eth_random_addr() for tap MAC address.
27 * Harald Roelle <harald.roelle@ifi.lmu.de> 2004/04/20
28 * Fixes in packet dropping, queue length setting and queue wakeup.
29 * Increased default tx queue length.
33 * Daniel Podlejski <underley@underley.eu.org>
34 * Modifications for 2.3.99-pre5 kernel.
37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
39 #define DRV_NAME "tun"
40 #define DRV_VERSION "1.6"
41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
42 #define DRV_COPYRIGHT "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
44 #include <linux/module.h>
45 #include <linux/errno.h>
46 #include <linux/kernel.h>
47 #include <linux/major.h>
48 #include <linux/slab.h>
49 #include <linux/poll.h>
50 #include <linux/fcntl.h>
51 #include <linux/init.h>
52 #include <linux/skbuff.h>
53 #include <linux/netdevice.h>
54 #include <linux/etherdevice.h>
55 #include <linux/miscdevice.h>
56 #include <linux/ethtool.h>
57 #include <linux/rtnetlink.h>
58 #include <linux/compat.h>
60 #include <linux/if_arp.h>
61 #include <linux/if_ether.h>
62 #include <linux/if_tun.h>
63 #include <linux/if_vlan.h>
64 #include <linux/crc32.h>
65 #include <linux/nsproxy.h>
66 #include <linux/virtio_net.h>
67 #include <linux/rcupdate.h>
68 #include <net/net_namespace.h>
69 #include <net/netns/generic.h>
70 #include <net/rtnetlink.h>
72 #include <linux/seq_file.h>
74 #include <asm/uaccess.h>
76 /* Uncomment to enable debugging */
77 /* #define TUN_DEBUG 1 */
82 #define tun_debug(level, tun, fmt, args...) \
85 netdev_printk(level, tun->dev, fmt, ##args); \
87 #define DBG1(level, fmt, args...) \
90 printk(level fmt, ##args); \
93 #define tun_debug(level, tun, fmt, args...) \
96 netdev_printk(level, tun->dev, fmt, ##args); \
98 #define DBG1(level, fmt, args...) \
101 printk(level fmt, ##args); \
105 #define GOODCOPY_LEN 128
107 #define FLT_EXACT_COUNT 8
109 unsigned int count; /* Number of addrs. Zero means disabled */
110 u32 mask[2]; /* Mask of the hashed addrs */
111 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
114 /* DEFAULT_MAX_NUM_RSS_QUEUES were chosen to let the rx/tx queues allocated for
115 * the netdevice to be fit in one page. So we can make sure the success of
116 * memory allocation. TODO: increase the limit. */
117 #define MAX_TAP_QUEUES DEFAULT_MAX_NUM_RSS_QUEUES
118 #define MAX_TAP_FLOWS 4096
120 #define TUN_FLOW_EXPIRE (3 * HZ)
122 /* A tun_file connects an open character device to a tuntap netdevice. It
123 * also contains all socket related structures (except sock_fprog and tap_filter)
124 * to serve as one transmit queue for tuntap device. The sock_fprog and
125 * tap_filter were kept in tun_struct since they were used for filtering for the
126 * netdevice not for a specific queue (at least I didn't see the requirement for
130 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
131 * other can only be read while rcu_read_lock or rtnl_lock is held.
135 struct socket socket;
137 struct tun_struct __rcu *tun;
139 struct fasync_struct *fasync;
140 /* only used for fasnyc */
144 unsigned int ifindex;
146 struct list_head next;
147 struct tun_struct *detached;
150 struct tun_flow_entry {
151 struct hlist_node hash_link;
153 struct tun_struct *tun;
158 unsigned long updated;
161 #define TUN_NUM_FLOW_ENTRIES 1024
163 /* Since the socket were moved to tun_file, to preserve the behavior of persist
164 * device, socket filter, sndbuf and vnet header size were restore when the
165 * file were attached to a persist device.
168 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
169 unsigned int numqueues;
174 struct net_device *dev;
175 netdev_features_t set_features;
176 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
181 struct tap_filter txflt;
182 struct sock_fprog fprog;
183 /* protected by rtnl lock */
184 bool filter_attached;
189 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
190 struct timer_list flow_gc_timer;
191 unsigned long ageing_time;
192 unsigned int numdisabled;
193 struct list_head disabled;
198 static inline u32 tun_hashfn(u32 rxhash)
200 return rxhash & 0x3ff;
203 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
205 struct tun_flow_entry *e;
207 hlist_for_each_entry_rcu(e, head, hash_link) {
208 if (e->rxhash == rxhash)
214 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
215 struct hlist_head *head,
216 u32 rxhash, u16 queue_index)
218 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
221 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
222 rxhash, queue_index);
223 e->updated = jiffies;
226 e->queue_index = queue_index;
228 hlist_add_head_rcu(&e->hash_link, head);
234 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
236 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
237 e->rxhash, e->queue_index);
238 sock_rps_reset_flow_hash(e->rps_rxhash);
239 hlist_del_rcu(&e->hash_link);
244 static void tun_flow_flush(struct tun_struct *tun)
248 spin_lock_bh(&tun->lock);
249 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
250 struct tun_flow_entry *e;
251 struct hlist_node *n;
253 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
254 tun_flow_delete(tun, e);
256 spin_unlock_bh(&tun->lock);
259 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
263 spin_lock_bh(&tun->lock);
264 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
265 struct tun_flow_entry *e;
266 struct hlist_node *n;
268 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
269 if (e->queue_index == queue_index)
270 tun_flow_delete(tun, e);
273 spin_unlock_bh(&tun->lock);
276 static void tun_flow_cleanup(unsigned long data)
278 struct tun_struct *tun = (struct tun_struct *)data;
279 unsigned long delay = tun->ageing_time;
280 unsigned long next_timer = jiffies + delay;
281 unsigned long count = 0;
284 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
286 spin_lock_bh(&tun->lock);
287 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
288 struct tun_flow_entry *e;
289 struct hlist_node *n;
291 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
292 unsigned long this_timer;
294 this_timer = e->updated + delay;
295 if (time_before_eq(this_timer, jiffies))
296 tun_flow_delete(tun, e);
297 else if (time_before(this_timer, next_timer))
298 next_timer = this_timer;
303 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
304 spin_unlock_bh(&tun->lock);
307 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
308 struct tun_file *tfile)
310 struct hlist_head *head;
311 struct tun_flow_entry *e;
312 unsigned long delay = tun->ageing_time;
313 u16 queue_index = tfile->queue_index;
318 head = &tun->flows[tun_hashfn(rxhash)];
322 /* We may get a very small possibility of OOO during switching, not
323 * worth to optimize.*/
324 if (tun->numqueues == 1 || tfile->detached)
327 e = tun_flow_find(head, rxhash);
329 /* TODO: keep queueing to old queue until it's empty? */
330 e->queue_index = queue_index;
331 e->updated = jiffies;
332 sock_rps_record_flow_hash(e->rps_rxhash);
334 spin_lock_bh(&tun->lock);
335 if (!tun_flow_find(head, rxhash) &&
336 tun->flow_count < MAX_TAP_FLOWS)
337 tun_flow_create(tun, head, rxhash, queue_index);
339 if (!timer_pending(&tun->flow_gc_timer))
340 mod_timer(&tun->flow_gc_timer,
341 round_jiffies_up(jiffies + delay));
342 spin_unlock_bh(&tun->lock);
350 * Save the hash received in the stack receive path and update the
351 * flow_hash table accordingly.
353 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
355 if (unlikely(e->rps_rxhash != hash)) {
356 sock_rps_reset_flow_hash(e->rps_rxhash);
357 e->rps_rxhash = hash;
361 /* We try to identify a flow through its rxhash first. The reason that
362 * we do not check rxq no. is because some cards(e.g 82599), chooses
363 * the rxq based on the txq where the last packet of the flow comes. As
364 * the userspace application move between processors, we may get a
365 * different rxq no. here. If we could not get rxhash, then we would
366 * hope the rxq no. may help here.
368 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
369 void *accel_priv, select_queue_fallback_t fallback)
371 struct tun_struct *tun = netdev_priv(dev);
372 struct tun_flow_entry *e;
377 numqueues = ACCESS_ONCE(tun->numqueues);
379 txq = skb_get_hash(skb);
381 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
383 tun_flow_save_rps_rxhash(e, txq);
384 txq = e->queue_index;
386 /* use multiply and shift instead of expensive divide */
387 txq = ((u64)txq * numqueues) >> 32;
388 } else if (likely(skb_rx_queue_recorded(skb))) {
389 txq = skb_get_rx_queue(skb);
390 while (unlikely(txq >= numqueues))
398 static inline bool tun_not_capable(struct tun_struct *tun)
400 const struct cred *cred = current_cred();
401 struct net *net = dev_net(tun->dev);
403 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
404 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
405 !ns_capable(net->user_ns, CAP_NET_ADMIN);
408 static void tun_set_real_num_queues(struct tun_struct *tun)
410 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
411 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
414 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
416 tfile->detached = tun;
417 list_add_tail(&tfile->next, &tun->disabled);
421 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
423 struct tun_struct *tun = tfile->detached;
425 tfile->detached = NULL;
426 list_del_init(&tfile->next);
431 static void tun_queue_purge(struct tun_file *tfile)
433 skb_queue_purge(&tfile->sk.sk_receive_queue);
434 skb_queue_purge(&tfile->sk.sk_error_queue);
437 static void __tun_detach(struct tun_file *tfile, bool clean)
439 struct tun_file *ntfile;
440 struct tun_struct *tun;
442 tun = rtnl_dereference(tfile->tun);
444 if (tun && !tfile->detached) {
445 u16 index = tfile->queue_index;
446 BUG_ON(index >= tun->numqueues);
448 rcu_assign_pointer(tun->tfiles[index],
449 tun->tfiles[tun->numqueues - 1]);
450 ntfile = rtnl_dereference(tun->tfiles[index]);
451 ntfile->queue_index = index;
455 RCU_INIT_POINTER(tfile->tun, NULL);
456 sock_put(&tfile->sk);
458 tun_disable_queue(tun, tfile);
461 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
462 /* Drop read queue */
463 tun_queue_purge(tfile);
464 tun_set_real_num_queues(tun);
465 } else if (tfile->detached && clean) {
466 tun = tun_enable_queue(tfile);
467 sock_put(&tfile->sk);
471 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
472 netif_carrier_off(tun->dev);
474 if (!(tun->flags & TUN_PERSIST) &&
475 tun->dev->reg_state == NETREG_REGISTERED)
476 unregister_netdevice(tun->dev);
479 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
480 &tfile->socket.flags));
481 sk_release_kernel(&tfile->sk);
485 static void tun_detach(struct tun_file *tfile, bool clean)
488 __tun_detach(tfile, clean);
492 static void tun_detach_all(struct net_device *dev)
494 struct tun_struct *tun = netdev_priv(dev);
495 struct tun_file *tfile, *tmp;
496 int i, n = tun->numqueues;
498 for (i = 0; i < n; i++) {
499 tfile = rtnl_dereference(tun->tfiles[i]);
501 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
502 RCU_INIT_POINTER(tfile->tun, NULL);
505 list_for_each_entry(tfile, &tun->disabled, next) {
506 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
507 RCU_INIT_POINTER(tfile->tun, NULL);
509 BUG_ON(tun->numqueues != 0);
512 for (i = 0; i < n; i++) {
513 tfile = rtnl_dereference(tun->tfiles[i]);
514 /* Drop read queue */
515 tun_queue_purge(tfile);
516 sock_put(&tfile->sk);
518 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
519 tun_enable_queue(tfile);
520 tun_queue_purge(tfile);
521 sock_put(&tfile->sk);
523 BUG_ON(tun->numdisabled != 0);
525 if (tun->flags & TUN_PERSIST)
526 module_put(THIS_MODULE);
529 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
531 struct tun_file *tfile = file->private_data;
534 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
539 if (rtnl_dereference(tfile->tun) && !tfile->detached)
543 if (!(tun->flags & TUN_TAP_MQ) && tun->numqueues == 1)
547 if (!tfile->detached &&
548 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
553 /* Re-attach the filter to persist device */
554 if (!skip_filter && (tun->filter_attached == true)) {
555 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
559 tfile->queue_index = tun->numqueues;
560 rcu_assign_pointer(tfile->tun, tun);
561 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
565 tun_enable_queue(tfile);
567 sock_hold(&tfile->sk);
569 tun_set_real_num_queues(tun);
571 /* device is allowed to go away first, so no need to hold extra
579 static struct tun_struct *__tun_get(struct tun_file *tfile)
581 struct tun_struct *tun;
584 tun = rcu_dereference(tfile->tun);
592 static struct tun_struct *tun_get(struct file *file)
594 return __tun_get(file->private_data);
597 static void tun_put(struct tun_struct *tun)
603 static void addr_hash_set(u32 *mask, const u8 *addr)
605 int n = ether_crc(ETH_ALEN, addr) >> 26;
606 mask[n >> 5] |= (1 << (n & 31));
609 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
611 int n = ether_crc(ETH_ALEN, addr) >> 26;
612 return mask[n >> 5] & (1 << (n & 31));
615 static int update_filter(struct tap_filter *filter, void __user *arg)
617 struct { u8 u[ETH_ALEN]; } *addr;
618 struct tun_filter uf;
619 int err, alen, n, nexact;
621 if (copy_from_user(&uf, arg, sizeof(uf)))
630 alen = ETH_ALEN * uf.count;
631 addr = kmalloc(alen, GFP_KERNEL);
635 if (copy_from_user(addr, arg + sizeof(uf), alen)) {
640 /* The filter is updated without holding any locks. Which is
641 * perfectly safe. We disable it first and in the worst
642 * case we'll accept a few undesired packets. */
646 /* Use first set of addresses as an exact filter */
647 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
648 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
652 /* Remaining multicast addresses are hashed,
653 * unicast will leave the filter disabled. */
654 memset(filter->mask, 0, sizeof(filter->mask));
655 for (; n < uf.count; n++) {
656 if (!is_multicast_ether_addr(addr[n].u)) {
657 err = 0; /* no filter */
660 addr_hash_set(filter->mask, addr[n].u);
663 /* For ALLMULTI just set the mask to all ones.
664 * This overrides the mask populated above. */
665 if ((uf.flags & TUN_FLT_ALLMULTI))
666 memset(filter->mask, ~0, sizeof(filter->mask));
668 /* Now enable the filter */
670 filter->count = nexact;
672 /* Return the number of exact filters */
680 /* Returns: 0 - drop, !=0 - accept */
681 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
683 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
685 struct ethhdr *eh = (struct ethhdr *) skb->data;
689 for (i = 0; i < filter->count; i++)
690 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
693 /* Inexact match (multicast only) */
694 if (is_multicast_ether_addr(eh->h_dest))
695 return addr_hash_test(filter->mask, eh->h_dest);
701 * Checks whether the packet is accepted or not.
702 * Returns: 0 - drop, !=0 - accept
704 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
709 return run_filter(filter, skb);
712 /* Network device part of the driver */
714 static const struct ethtool_ops tun_ethtool_ops;
716 /* Net device detach from fd. */
717 static void tun_net_uninit(struct net_device *dev)
722 /* Net device open. */
723 static int tun_net_open(struct net_device *dev)
725 netif_tx_start_all_queues(dev);
729 /* Net device close. */
730 static int tun_net_close(struct net_device *dev)
732 netif_tx_stop_all_queues(dev);
736 /* Net device start xmit */
737 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
739 struct tun_struct *tun = netdev_priv(dev);
740 int txq = skb->queue_mapping;
741 struct tun_file *tfile;
745 tfile = rcu_dereference(tun->tfiles[txq]);
746 numqueues = ACCESS_ONCE(tun->numqueues);
748 /* Drop packet if interface is not attached */
749 if (txq >= numqueues)
752 if (numqueues == 1) {
753 /* Select queue was not called for the skbuff, so we extract the
754 * RPS hash and save it into the flow_table here.
758 rxhash = skb_get_hash(skb);
760 struct tun_flow_entry *e;
761 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
764 tun_flow_save_rps_rxhash(e, rxhash);
768 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
772 /* Drop if the filter does not like it.
773 * This is a noop if the filter is disabled.
774 * Filter can be enabled only for the TAP devices. */
775 if (!check_filter(&tun->txflt, skb))
778 if (tfile->socket.sk->sk_filter &&
779 sk_filter(tfile->socket.sk, skb))
782 /* Limit the number of packets queued by dividing txq length with the
785 if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues
786 >= dev->tx_queue_len)
789 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
793 sock_tx_timestamp(skb->sk, &skb_shinfo(skb)->tx_flags);
794 sw_tx_timestamp(skb);
797 /* Orphan the skb - required as we might hang on to it
798 * for indefinite time.
805 skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
807 /* Notify and wake up reader process */
808 if (tfile->flags & TUN_FASYNC)
809 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
810 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
816 dev->stats.tx_dropped++;
823 static void tun_net_mclist(struct net_device *dev)
826 * This callback is supposed to deal with mc filter in
827 * _rx_ path and has nothing to do with the _tx_ path.
828 * In rx path we always accept everything userspace gives us.
833 #define MAX_MTU 65535
836 tun_net_change_mtu(struct net_device *dev, int new_mtu)
838 if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
844 static netdev_features_t tun_net_fix_features(struct net_device *dev,
845 netdev_features_t features)
847 struct tun_struct *tun = netdev_priv(dev);
849 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
851 #ifdef CONFIG_NET_POLL_CONTROLLER
852 static void tun_poll_controller(struct net_device *dev)
855 * Tun only receives frames when:
856 * 1) the char device endpoint gets data from user space
857 * 2) the tun socket gets a sendmsg call from user space
858 * Since both of those are synchronous operations, we are guaranteed
859 * never to have pending data when we poll for it
860 * so there is nothing to do here but return.
861 * We need this though so netpoll recognizes us as an interface that
862 * supports polling, which enables bridge devices in virt setups to
863 * still use netconsole
868 static const struct net_device_ops tun_netdev_ops = {
869 .ndo_uninit = tun_net_uninit,
870 .ndo_open = tun_net_open,
871 .ndo_stop = tun_net_close,
872 .ndo_start_xmit = tun_net_xmit,
873 .ndo_change_mtu = tun_net_change_mtu,
874 .ndo_fix_features = tun_net_fix_features,
875 .ndo_select_queue = tun_select_queue,
876 #ifdef CONFIG_NET_POLL_CONTROLLER
877 .ndo_poll_controller = tun_poll_controller,
881 static const struct net_device_ops tap_netdev_ops = {
882 .ndo_uninit = tun_net_uninit,
883 .ndo_open = tun_net_open,
884 .ndo_stop = tun_net_close,
885 .ndo_start_xmit = tun_net_xmit,
886 .ndo_change_mtu = tun_net_change_mtu,
887 .ndo_fix_features = tun_net_fix_features,
888 .ndo_set_rx_mode = tun_net_mclist,
889 .ndo_set_mac_address = eth_mac_addr,
890 .ndo_validate_addr = eth_validate_addr,
891 .ndo_select_queue = tun_select_queue,
892 #ifdef CONFIG_NET_POLL_CONTROLLER
893 .ndo_poll_controller = tun_poll_controller,
897 static void tun_flow_init(struct tun_struct *tun)
901 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
902 INIT_HLIST_HEAD(&tun->flows[i]);
904 tun->ageing_time = TUN_FLOW_EXPIRE;
905 setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
906 mod_timer(&tun->flow_gc_timer,
907 round_jiffies_up(jiffies + tun->ageing_time));
910 static void tun_flow_uninit(struct tun_struct *tun)
912 del_timer_sync(&tun->flow_gc_timer);
916 /* Initialize net device. */
917 static void tun_net_init(struct net_device *dev)
919 struct tun_struct *tun = netdev_priv(dev);
921 switch (tun->flags & TUN_TYPE_MASK) {
923 dev->netdev_ops = &tun_netdev_ops;
925 /* Point-to-Point TUN Device */
926 dev->hard_header_len = 0;
930 /* Zero header length */
931 dev->type = ARPHRD_NONE;
932 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
933 dev->tx_queue_len = TUN_READQ_SIZE; /* We prefer our own queue length */
937 dev->netdev_ops = &tap_netdev_ops;
938 /* Ethernet TAP Device */
940 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
941 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
943 eth_hw_addr_random(dev);
945 dev->tx_queue_len = TUN_READQ_SIZE; /* We prefer our own queue length */
950 /* Character device part */
953 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
955 struct tun_file *tfile = file->private_data;
956 struct tun_struct *tun = __tun_get(tfile);
958 unsigned int mask = 0;
963 sk = tfile->socket.sk;
965 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
967 poll_wait(file, sk_sleep(sk), wait);
969 if (!skb_queue_empty(&sk->sk_receive_queue))
970 mask |= POLLIN | POLLRDNORM;
972 if (sock_writeable(sk) ||
973 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
975 mask |= POLLOUT | POLLWRNORM;
977 if (tun->dev->reg_state != NETREG_REGISTERED)
984 /* prepad is the amount to reserve at front. len is length after that.
985 * linear is a hint as to how much to copy (usually headers). */
986 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
987 size_t prepad, size_t len,
988 size_t linear, int noblock)
990 struct sock *sk = tfile->socket.sk;
994 /* Under a page? Don't bother with paged skb. */
995 if (prepad + len < PAGE_SIZE || !linear)
998 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1001 return ERR_PTR(err);
1003 skb_reserve(skb, prepad);
1004 skb_put(skb, linear);
1005 skb->data_len = len - linear;
1006 skb->len += len - linear;
1011 /* Get packet from user space buffer */
1012 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1013 void *msg_control, const struct iovec *iv,
1014 size_t total_len, size_t count, int noblock)
1016 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1017 struct sk_buff *skb;
1018 size_t len = total_len, align = NET_SKB_PAD, linear;
1019 struct virtio_net_hdr gso = { 0 };
1023 bool zerocopy = false;
1027 if (!(tun->flags & TUN_NO_PI)) {
1028 if (len < sizeof(pi))
1032 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
1034 offset += sizeof(pi);
1037 if (tun->flags & TUN_VNET_HDR) {
1038 if (len < tun->vnet_hdr_sz)
1040 len -= tun->vnet_hdr_sz;
1042 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1045 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1046 gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1047 gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1049 if (gso.hdr_len > len)
1051 offset += tun->vnet_hdr_sz;
1054 if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1055 align += NET_IP_ALIGN;
1056 if (unlikely(len < ETH_HLEN ||
1057 (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1061 good_linear = SKB_MAX_HEAD(align);
1064 /* There are 256 bytes to be copied in skb, so there is
1065 * enough room for skb expand head in case it is used.
1066 * The rest of the buffer is mapped from userspace.
1068 copylen = gso.hdr_len ? gso.hdr_len : GOODCOPY_LEN;
1069 if (copylen > good_linear)
1070 copylen = good_linear;
1072 if (iov_pages(iv, offset + copylen, count) <= MAX_SKB_FRAGS)
1078 if (gso.hdr_len > good_linear)
1079 linear = good_linear;
1081 linear = gso.hdr_len;
1084 skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1086 if (PTR_ERR(skb) != -EAGAIN)
1087 tun->dev->stats.rx_dropped++;
1088 return PTR_ERR(skb);
1092 err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1094 err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1095 if (!err && msg_control) {
1096 struct ubuf_info *uarg = msg_control;
1097 uarg->callback(uarg, false);
1102 tun->dev->stats.rx_dropped++;
1107 if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1108 if (!skb_partial_csum_set(skb, gso.csum_start,
1110 tun->dev->stats.rx_frame_errors++;
1116 switch (tun->flags & TUN_TYPE_MASK) {
1118 if (tun->flags & TUN_NO_PI) {
1119 switch (skb->data[0] & 0xf0) {
1121 pi.proto = htons(ETH_P_IP);
1124 pi.proto = htons(ETH_P_IPV6);
1127 tun->dev->stats.rx_dropped++;
1133 skb_reset_mac_header(skb);
1134 skb->protocol = pi.proto;
1135 skb->dev = tun->dev;
1138 skb->protocol = eth_type_trans(skb, tun->dev);
1142 if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1144 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1145 case VIRTIO_NET_HDR_GSO_TCPV4:
1146 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1148 case VIRTIO_NET_HDR_GSO_TCPV6:
1149 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1151 case VIRTIO_NET_HDR_GSO_UDP:
1157 netdev_warn(tun->dev,
1158 "%s: using disabled UFO feature; please fix this program\n",
1161 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1165 tun->dev->stats.rx_frame_errors++;
1170 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1171 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1173 skb_shinfo(skb)->gso_size = gso.gso_size;
1174 if (skb_shinfo(skb)->gso_size == 0) {
1175 tun->dev->stats.rx_frame_errors++;
1180 /* Header must be checked, and gso_segs computed. */
1181 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1182 skb_shinfo(skb)->gso_segs = 0;
1185 /* copy skb_ubuf_info for callback when skb has no error */
1187 skb_shinfo(skb)->destructor_arg = msg_control;
1188 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1189 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1192 skb_reset_network_header(skb);
1193 skb_probe_transport_header(skb, 0);
1195 rxhash = skb_get_hash(skb);
1198 tun->dev->stats.rx_packets++;
1199 tun->dev->stats.rx_bytes += len;
1201 tun_flow_update(tun, rxhash, tfile);
1205 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1206 unsigned long count, loff_t pos)
1208 struct file *file = iocb->ki_filp;
1209 struct tun_struct *tun = tun_get(file);
1210 struct tun_file *tfile = file->private_data;
1216 tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1218 result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1219 count, file->f_flags & O_NONBLOCK);
1225 /* Put packet to the user space buffer */
1226 static ssize_t tun_put_user(struct tun_struct *tun,
1227 struct tun_file *tfile,
1228 struct sk_buff *skb,
1229 const struct iovec *iv, int len)
1231 struct tun_pi pi = { 0, skb->protocol };
1233 int vlan_offset = 0, copied;
1235 if (!(tun->flags & TUN_NO_PI)) {
1236 if ((len -= sizeof(pi)) < 0)
1239 if (len < skb->len) {
1240 /* Packet will be striped */
1241 pi.flags |= TUN_PKT_STRIP;
1244 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1246 total += sizeof(pi);
1249 if (tun->flags & TUN_VNET_HDR) {
1250 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1251 if ((len -= tun->vnet_hdr_sz) < 0)
1254 if (skb_is_gso(skb)) {
1255 struct skb_shared_info *sinfo = skb_shinfo(skb);
1257 /* This is a hint as to how much should be linear. */
1258 gso.hdr_len = skb_headlen(skb);
1259 gso.gso_size = sinfo->gso_size;
1260 if (sinfo->gso_type & SKB_GSO_TCPV4)
1261 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1262 else if (sinfo->gso_type & SKB_GSO_TCPV6)
1263 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1265 pr_err("unexpected GSO type: "
1266 "0x%x, gso_size %d, hdr_len %d\n",
1267 sinfo->gso_type, gso.gso_size,
1269 print_hex_dump(KERN_ERR, "tun: ",
1272 min((int)gso.hdr_len, 64), true);
1276 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1277 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1279 gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1281 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1282 gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1283 gso.csum_start = skb_checksum_start_offset(skb);
1284 gso.csum_offset = skb->csum_offset;
1285 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1286 gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1287 } /* else everything is zero */
1289 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1292 total += tun->vnet_hdr_sz;
1297 if (!vlan_tx_tag_present(skb)) {
1298 len = min_t(int, skb->len, len);
1302 __be16 h_vlan_proto;
1306 veth.h_vlan_proto = skb->vlan_proto;
1307 veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb));
1309 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1310 len = min_t(int, skb->len + VLAN_HLEN, len);
1313 copy = min_t(int, vlan_offset, len);
1314 ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy);
1320 copy = min_t(int, sizeof(veth), len);
1321 ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy);
1328 skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len);
1331 tun->dev->stats.tx_packets++;
1332 tun->dev->stats.tx_bytes += len;
1337 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1338 const struct iovec *iv, ssize_t len, int noblock)
1340 struct sk_buff *skb;
1342 int peeked, err, off = 0;
1344 tun_debug(KERN_INFO, tun, "tun_do_read\n");
1349 if (tun->dev->reg_state != NETREG_REGISTERED)
1352 /* Read frames from queue */
1353 skb = __skb_recv_datagram(tfile->socket.sk, noblock ? MSG_DONTWAIT : 0,
1354 &peeked, &off, &err);
1356 ret = tun_put_user(tun, tfile, skb, iv, len);
1364 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1365 unsigned long count, loff_t pos)
1367 struct file *file = iocb->ki_filp;
1368 struct tun_file *tfile = file->private_data;
1369 struct tun_struct *tun = __tun_get(tfile);
1374 len = iov_length(iv, count);
1380 ret = tun_do_read(tun, tfile, iv, len,
1381 file->f_flags & O_NONBLOCK);
1382 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);
1461 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1462 struct msghdr *m, size_t total_len,
1465 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1466 struct tun_struct *tun = __tun_get(tfile);
1472 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1476 if (flags & MSG_ERRQUEUE) {
1477 ret = sock_recv_errqueue(sock->sk, m, total_len,
1478 SOL_PACKET, TUN_TX_TIMESTAMP);
1481 ret = tun_do_read(tun, tfile, m->msg_iov, total_len,
1482 flags & MSG_DONTWAIT);
1483 if (ret > total_len) {
1484 m->msg_flags |= MSG_TRUNC;
1485 ret = flags & MSG_TRUNC ? ret : total_len;
1492 static int tun_release(struct socket *sock)
1499 /* Ops structure to mimic raw sockets with tun */
1500 static const struct proto_ops tun_socket_ops = {
1501 .sendmsg = tun_sendmsg,
1502 .recvmsg = tun_recvmsg,
1503 .release = tun_release,
1506 static struct proto tun_proto = {
1508 .owner = THIS_MODULE,
1509 .obj_size = sizeof(struct tun_file),
1512 static int tun_flags(struct tun_struct *tun)
1516 if (tun->flags & TUN_TUN_DEV)
1521 if (tun->flags & TUN_NO_PI)
1524 /* This flag has no real effect. We track the value for backwards
1527 if (tun->flags & TUN_ONE_QUEUE)
1528 flags |= IFF_ONE_QUEUE;
1530 if (tun->flags & TUN_VNET_HDR)
1531 flags |= IFF_VNET_HDR;
1533 if (tun->flags & TUN_TAP_MQ)
1534 flags |= IFF_MULTI_QUEUE;
1536 if (tun->flags & TUN_PERSIST)
1537 flags |= IFF_PERSIST;
1542 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1545 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1546 return sprintf(buf, "0x%x\n", tun_flags(tun));
1549 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1552 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1553 return uid_valid(tun->owner)?
1554 sprintf(buf, "%u\n",
1555 from_kuid_munged(current_user_ns(), tun->owner)):
1556 sprintf(buf, "-1\n");
1559 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1562 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1563 return gid_valid(tun->group) ?
1564 sprintf(buf, "%u\n",
1565 from_kgid_munged(current_user_ns(), tun->group)):
1566 sprintf(buf, "-1\n");
1569 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1570 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1571 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1573 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1575 struct tun_struct *tun;
1576 struct tun_file *tfile = file->private_data;
1577 struct net_device *dev;
1580 if (tfile->detached)
1583 dev = __dev_get_by_name(net, ifr->ifr_name);
1585 if (ifr->ifr_flags & IFF_TUN_EXCL)
1587 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1588 tun = netdev_priv(dev);
1589 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1590 tun = netdev_priv(dev);
1594 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1595 !!(tun->flags & TUN_TAP_MQ))
1598 if (tun_not_capable(tun))
1600 err = security_tun_dev_open(tun->security);
1604 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1608 if (tun->flags & TUN_TAP_MQ &&
1609 (tun->numqueues + tun->numdisabled > 1)) {
1610 /* One or more queue has already been attached, no need
1611 * to initialize the device again.
1618 unsigned long flags = 0;
1619 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1622 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1624 err = security_tun_dev_create();
1629 if (ifr->ifr_flags & IFF_TUN) {
1631 flags |= TUN_TUN_DEV;
1633 } else if (ifr->ifr_flags & IFF_TAP) {
1635 flags |= TUN_TAP_DEV;
1641 name = ifr->ifr_name;
1643 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1644 NET_NAME_UNKNOWN, tun_setup, queues,
1650 dev_net_set(dev, net);
1651 dev->rtnl_link_ops = &tun_link_ops;
1652 dev->ifindex = tfile->ifindex;
1654 tun = netdev_priv(dev);
1657 tun->txflt.count = 0;
1658 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1660 tun->filter_attached = false;
1661 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1663 spin_lock_init(&tun->lock);
1665 err = security_tun_dev_alloc_security(&tun->security);
1672 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1673 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1674 NETIF_F_HW_VLAN_STAG_TX;
1675 dev->features = dev->hw_features;
1676 dev->vlan_features = dev->features &
1677 ~(NETIF_F_HW_VLAN_CTAG_TX |
1678 NETIF_F_HW_VLAN_STAG_TX);
1680 INIT_LIST_HEAD(&tun->disabled);
1681 err = tun_attach(tun, file, false);
1685 err = register_netdevice(tun->dev);
1689 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1690 device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1691 device_create_file(&tun->dev->dev, &dev_attr_group))
1692 pr_err("Failed to create tun sysfs files\n");
1695 netif_carrier_on(tun->dev);
1697 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1699 if (ifr->ifr_flags & IFF_NO_PI)
1700 tun->flags |= TUN_NO_PI;
1702 tun->flags &= ~TUN_NO_PI;
1704 /* This flag has no real effect. We track the value for backwards
1707 if (ifr->ifr_flags & IFF_ONE_QUEUE)
1708 tun->flags |= TUN_ONE_QUEUE;
1710 tun->flags &= ~TUN_ONE_QUEUE;
1712 if (ifr->ifr_flags & IFF_VNET_HDR)
1713 tun->flags |= TUN_VNET_HDR;
1715 tun->flags &= ~TUN_VNET_HDR;
1717 if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1718 tun->flags |= TUN_TAP_MQ;
1720 tun->flags &= ~TUN_TAP_MQ;
1722 /* Make sure persistent devices do not get stuck in
1725 if (netif_running(tun->dev))
1726 netif_tx_wake_all_queues(tun->dev);
1728 strcpy(ifr->ifr_name, tun->dev->name);
1732 tun_detach_all(dev);
1734 tun_flow_uninit(tun);
1735 security_tun_dev_free_security(tun->security);
1741 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1744 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1746 strcpy(ifr->ifr_name, tun->dev->name);
1748 ifr->ifr_flags = tun_flags(tun);
1752 /* This is like a cut-down ethtool ops, except done via tun fd so no
1753 * privs required. */
1754 static int set_offload(struct tun_struct *tun, unsigned long arg)
1756 netdev_features_t features = 0;
1758 if (arg & TUN_F_CSUM) {
1759 features |= NETIF_F_HW_CSUM;
1762 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1763 if (arg & TUN_F_TSO_ECN) {
1764 features |= NETIF_F_TSO_ECN;
1765 arg &= ~TUN_F_TSO_ECN;
1767 if (arg & TUN_F_TSO4)
1768 features |= NETIF_F_TSO;
1769 if (arg & TUN_F_TSO6)
1770 features |= NETIF_F_TSO6;
1771 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1775 /* This gives the user a way to test for new features in future by
1776 * trying to set them. */
1780 tun->set_features = features;
1781 netdev_update_features(tun->dev);
1786 static void tun_detach_filter(struct tun_struct *tun, int n)
1789 struct tun_file *tfile;
1791 for (i = 0; i < n; i++) {
1792 tfile = rtnl_dereference(tun->tfiles[i]);
1793 sk_detach_filter(tfile->socket.sk);
1796 tun->filter_attached = false;
1799 static int tun_attach_filter(struct tun_struct *tun)
1802 struct tun_file *tfile;
1804 for (i = 0; i < tun->numqueues; i++) {
1805 tfile = rtnl_dereference(tun->tfiles[i]);
1806 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1808 tun_detach_filter(tun, i);
1813 tun->filter_attached = true;
1817 static void tun_set_sndbuf(struct tun_struct *tun)
1819 struct tun_file *tfile;
1822 for (i = 0; i < tun->numqueues; i++) {
1823 tfile = rtnl_dereference(tun->tfiles[i]);
1824 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1828 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1830 struct tun_file *tfile = file->private_data;
1831 struct tun_struct *tun;
1836 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1837 tun = tfile->detached;
1842 ret = security_tun_dev_attach_queue(tun->security);
1845 ret = tun_attach(tun, file, false);
1846 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1847 tun = rtnl_dereference(tfile->tun);
1848 if (!tun || !(tun->flags & TUN_TAP_MQ) || tfile->detached)
1851 __tun_detach(tfile, false);
1860 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1861 unsigned long arg, int ifreq_len)
1863 struct tun_file *tfile = file->private_data;
1864 struct tun_struct *tun;
1865 void __user* argp = (void __user*)arg;
1871 unsigned int ifindex;
1874 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1875 if (copy_from_user(&ifr, argp, ifreq_len))
1878 memset(&ifr, 0, sizeof(ifr));
1880 if (cmd == TUNGETFEATURES) {
1881 /* Currently this just means: "what IFF flags are valid?".
1882 * This is needed because we never checked for invalid flags on
1884 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1885 IFF_VNET_HDR | IFF_MULTI_QUEUE,
1886 (unsigned int __user*)argp);
1887 } else if (cmd == TUNSETQUEUE)
1888 return tun_set_queue(file, &ifr);
1893 tun = __tun_get(tfile);
1894 if (cmd == TUNSETIFF && !tun) {
1895 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1897 ret = tun_set_iff(tfile->net, file, &ifr);
1902 if (copy_to_user(argp, &ifr, ifreq_len))
1906 if (cmd == TUNSETIFINDEX) {
1912 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
1916 tfile->ifindex = ifindex;
1924 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1929 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1931 if (tfile->detached)
1932 ifr.ifr_flags |= IFF_DETACH_QUEUE;
1933 if (!tfile->socket.sk->sk_filter)
1934 ifr.ifr_flags |= IFF_NOFILTER;
1936 if (copy_to_user(argp, &ifr, ifreq_len))
1941 /* Disable/Enable checksum */
1943 /* [unimplemented] */
1944 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1945 arg ? "disabled" : "enabled");
1949 /* Disable/Enable persist mode. Keep an extra reference to the
1950 * module to prevent the module being unprobed.
1952 if (arg && !(tun->flags & TUN_PERSIST)) {
1953 tun->flags |= TUN_PERSIST;
1954 __module_get(THIS_MODULE);
1956 if (!arg && (tun->flags & TUN_PERSIST)) {
1957 tun->flags &= ~TUN_PERSIST;
1958 module_put(THIS_MODULE);
1961 tun_debug(KERN_INFO, tun, "persist %s\n",
1962 arg ? "enabled" : "disabled");
1966 /* Set owner of the device */
1967 owner = make_kuid(current_user_ns(), arg);
1968 if (!uid_valid(owner)) {
1973 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1974 from_kuid(&init_user_ns, tun->owner));
1978 /* Set group of the device */
1979 group = make_kgid(current_user_ns(), arg);
1980 if (!gid_valid(group)) {
1985 tun_debug(KERN_INFO, tun, "group set to %u\n",
1986 from_kgid(&init_user_ns, tun->group));
1990 /* Only allow setting the type when the interface is down */
1991 if (tun->dev->flags & IFF_UP) {
1992 tun_debug(KERN_INFO, tun,
1993 "Linktype set failed because interface is up\n");
1996 tun->dev->type = (int) arg;
1997 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2009 ret = set_offload(tun, arg);
2012 case TUNSETTXFILTER:
2013 /* Can be set only for TAPs */
2015 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2017 ret = update_filter(&tun->txflt, (void __user *)arg);
2021 /* Get hw address */
2022 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2023 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2024 if (copy_to_user(argp, &ifr, ifreq_len))
2029 /* Set hw address */
2030 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2031 ifr.ifr_hwaddr.sa_data);
2033 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2037 sndbuf = tfile->socket.sk->sk_sndbuf;
2038 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2043 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2048 tun->sndbuf = sndbuf;
2049 tun_set_sndbuf(tun);
2052 case TUNGETVNETHDRSZ:
2053 vnet_hdr_sz = tun->vnet_hdr_sz;
2054 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2058 case TUNSETVNETHDRSZ:
2059 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2063 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2068 tun->vnet_hdr_sz = vnet_hdr_sz;
2071 case TUNATTACHFILTER:
2072 /* Can be set only for TAPs */
2074 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2077 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2080 ret = tun_attach_filter(tun);
2083 case TUNDETACHFILTER:
2084 /* Can be set only for TAPs */
2086 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2089 tun_detach_filter(tun, tun->numqueues);
2094 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2097 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2114 static long tun_chr_ioctl(struct file *file,
2115 unsigned int cmd, unsigned long arg)
2117 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2120 #ifdef CONFIG_COMPAT
2121 static long tun_chr_compat_ioctl(struct file *file,
2122 unsigned int cmd, unsigned long arg)
2127 case TUNSETTXFILTER:
2132 arg = (unsigned long)compat_ptr(arg);
2135 arg = (compat_ulong_t)arg;
2140 * compat_ifreq is shorter than ifreq, so we must not access beyond
2141 * the end of that structure. All fields that are used in this
2142 * driver are compatible though, we don't need to convert the
2145 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2147 #endif /* CONFIG_COMPAT */
2149 static int tun_chr_fasync(int fd, struct file *file, int on)
2151 struct tun_file *tfile = file->private_data;
2154 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2158 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2159 tfile->flags |= TUN_FASYNC;
2161 tfile->flags &= ~TUN_FASYNC;
2167 static int tun_chr_open(struct inode *inode, struct file * file)
2169 struct tun_file *tfile;
2171 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2173 tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2177 RCU_INIT_POINTER(tfile->tun, NULL);
2178 tfile->net = get_net(current->nsproxy->net_ns);
2182 init_waitqueue_head(&tfile->wq.wait);
2183 RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2185 tfile->socket.file = file;
2186 tfile->socket.ops = &tun_socket_ops;
2188 sock_init_data(&tfile->socket, &tfile->sk);
2189 sk_change_net(&tfile->sk, tfile->net);
2191 tfile->sk.sk_write_space = tun_sock_write_space;
2192 tfile->sk.sk_sndbuf = INT_MAX;
2194 file->private_data = tfile;
2195 set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2196 INIT_LIST_HEAD(&tfile->next);
2198 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2203 static int tun_chr_close(struct inode *inode, struct file *file)
2205 struct tun_file *tfile = file->private_data;
2206 struct net *net = tfile->net;
2208 tun_detach(tfile, true);
2214 #ifdef CONFIG_PROC_FS
2215 static int tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2217 struct tun_struct *tun;
2220 memset(&ifr, 0, sizeof(ifr));
2225 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2231 return seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2235 static const struct file_operations tun_fops = {
2236 .owner = THIS_MODULE,
2237 .llseek = no_llseek,
2238 .read = do_sync_read,
2239 .aio_read = tun_chr_aio_read,
2240 .write = do_sync_write,
2241 .aio_write = tun_chr_aio_write,
2242 .poll = tun_chr_poll,
2243 .unlocked_ioctl = tun_chr_ioctl,
2244 #ifdef CONFIG_COMPAT
2245 .compat_ioctl = tun_chr_compat_ioctl,
2247 .open = tun_chr_open,
2248 .release = tun_chr_close,
2249 .fasync = tun_chr_fasync,
2250 #ifdef CONFIG_PROC_FS
2251 .show_fdinfo = tun_chr_show_fdinfo,
2255 static struct miscdevice tun_miscdev = {
2258 .nodename = "net/tun",
2262 /* ethtool interface */
2264 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2267 cmd->advertising = 0;
2268 ethtool_cmd_speed_set(cmd, SPEED_10);
2269 cmd->duplex = DUPLEX_FULL;
2270 cmd->port = PORT_TP;
2271 cmd->phy_address = 0;
2272 cmd->transceiver = XCVR_INTERNAL;
2273 cmd->autoneg = AUTONEG_DISABLE;
2279 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2281 struct tun_struct *tun = netdev_priv(dev);
2283 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2284 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2286 switch (tun->flags & TUN_TYPE_MASK) {
2288 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2291 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2296 static u32 tun_get_msglevel(struct net_device *dev)
2299 struct tun_struct *tun = netdev_priv(dev);
2306 static void tun_set_msglevel(struct net_device *dev, u32 value)
2309 struct tun_struct *tun = netdev_priv(dev);
2314 static const struct ethtool_ops tun_ethtool_ops = {
2315 .get_settings = tun_get_settings,
2316 .get_drvinfo = tun_get_drvinfo,
2317 .get_msglevel = tun_get_msglevel,
2318 .set_msglevel = tun_set_msglevel,
2319 .get_link = ethtool_op_get_link,
2320 .get_ts_info = ethtool_op_get_ts_info,
2324 static int __init tun_init(void)
2328 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2329 pr_info("%s\n", DRV_COPYRIGHT);
2331 ret = rtnl_link_register(&tun_link_ops);
2333 pr_err("Can't register link_ops\n");
2337 ret = misc_register(&tun_miscdev);
2339 pr_err("Can't register misc device %d\n", TUN_MINOR);
2344 rtnl_link_unregister(&tun_link_ops);
2349 static void tun_cleanup(void)
2351 misc_deregister(&tun_miscdev);
2352 rtnl_link_unregister(&tun_link_ops);
2355 /* Get an underlying socket object from tun file. Returns error unless file is
2356 * attached to a device. The returned object works like a packet socket, it
2357 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
2358 * holding a reference to the file for as long as the socket is in use. */
2359 struct socket *tun_get_socket(struct file *file)
2361 struct tun_file *tfile;
2362 if (file->f_op != &tun_fops)
2363 return ERR_PTR(-EINVAL);
2364 tfile = file->private_data;
2366 return ERR_PTR(-EBADFD);
2367 return &tfile->socket;
2369 EXPORT_SYMBOL_GPL(tun_get_socket);
2371 module_init(tun_init);
2372 module_exit(tun_cleanup);
2373 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2374 MODULE_AUTHOR(DRV_COPYRIGHT);
2375 MODULE_LICENSE("GPL");
2376 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2377 MODULE_ALIAS("devname:net/tun");