2 * TUN - Universal TUN/TAP device driver.
3 * Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
21 * Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
22 * Add TUNSETLINK ioctl to set the link encapsulation
24 * Mark Smith <markzzzsmith@yahoo.com.au>
25 * Use eth_random_addr() for tap MAC address.
27 * Harald Roelle <harald.roelle@ifi.lmu.de> 2004/04/20
28 * Fixes in packet dropping, queue length setting and queue wakeup.
29 * Increased default tx queue length.
33 * Daniel Podlejski <underley@underley.eu.org>
34 * Modifications for 2.3.99-pre5 kernel.
37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
39 #define DRV_NAME "tun"
40 #define DRV_VERSION "1.6"
41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
42 #define DRV_COPYRIGHT "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
44 #include <linux/module.h>
45 #include <linux/errno.h>
46 #include <linux/kernel.h>
47 #include <linux/major.h>
48 #include <linux/slab.h>
49 #include <linux/poll.h>
50 #include <linux/fcntl.h>
51 #include <linux/init.h>
52 #include <linux/skbuff.h>
53 #include <linux/netdevice.h>
54 #include <linux/etherdevice.h>
55 #include <linux/miscdevice.h>
56 #include <linux/ethtool.h>
57 #include <linux/rtnetlink.h>
58 #include <linux/compat.h>
60 #include <linux/if_arp.h>
61 #include <linux/if_ether.h>
62 #include <linux/if_tun.h>
63 #include <linux/if_vlan.h>
64 #include <linux/crc32.h>
65 #include <linux/nsproxy.h>
66 #include <linux/virtio_net.h>
67 #include <linux/rcupdate.h>
68 #include <net/net_namespace.h>
69 #include <net/netns/generic.h>
70 #include <net/rtnetlink.h>
72 #include <linux/seq_file.h>
73 #include <linux/uio.h>
74 #include <linux/skb_array.h>
76 #include <asm/uaccess.h>
78 /* Uncomment to enable debugging */
79 /* #define TUN_DEBUG 1 */
84 #define tun_debug(level, tun, fmt, args...) \
87 netdev_printk(level, tun->dev, fmt, ##args); \
89 #define DBG1(level, fmt, args...) \
92 printk(level fmt, ##args); \
95 #define tun_debug(level, tun, fmt, args...) \
98 netdev_printk(level, tun->dev, fmt, ##args); \
100 #define DBG1(level, fmt, args...) \
103 printk(level fmt, ##args); \
107 /* TUN device flags */
109 /* IFF_ATTACH_QUEUE is never stored in device flags,
110 * overload it to mean fasync when stored there.
112 #define TUN_FASYNC IFF_ATTACH_QUEUE
113 /* High bits in flags field are unused. */
114 #define TUN_VNET_LE 0x80000000
115 #define TUN_VNET_BE 0x40000000
117 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
119 #define GOODCOPY_LEN 128
121 #define FLT_EXACT_COUNT 8
123 unsigned int count; /* Number of addrs. Zero means disabled */
124 u32 mask[2]; /* Mask of the hashed addrs */
125 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
128 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
129 * to max number of VCPUs in guest. */
130 #define MAX_TAP_QUEUES 256
131 #define MAX_TAP_FLOWS 4096
133 #define TUN_FLOW_EXPIRE (3 * HZ)
135 struct tun_pcpu_stats {
140 struct u64_stats_sync syncp;
146 /* A tun_file connects an open character device to a tuntap netdevice. It
147 * also contains all socket related structures (except sock_fprog and tap_filter)
148 * to serve as one transmit queue for tuntap device. The sock_fprog and
149 * tap_filter were kept in tun_struct since they were used for filtering for the
150 * netdevice not for a specific queue (at least I didn't see the requirement for
154 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
155 * other can only be read while rcu_read_lock or rtnl_lock is held.
159 struct socket socket;
161 struct tun_struct __rcu *tun;
162 struct fasync_struct *fasync;
163 /* only used for fasnyc */
167 unsigned int ifindex;
169 struct list_head next;
170 struct tun_struct *detached;
171 struct skb_array tx_array;
174 struct tun_flow_entry {
175 struct hlist_node hash_link;
177 struct tun_struct *tun;
182 unsigned long updated;
185 #define TUN_NUM_FLOW_ENTRIES 1024
187 /* Since the socket were moved to tun_file, to preserve the behavior of persist
188 * device, socket filter, sndbuf and vnet header size were restore when the
189 * file were attached to a persist device.
192 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
193 unsigned int numqueues;
198 struct net_device *dev;
199 netdev_features_t set_features;
200 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
201 NETIF_F_TSO6|NETIF_F_UFO)
206 struct tap_filter txflt;
207 struct sock_fprog fprog;
208 /* protected by rtnl lock */
209 bool filter_attached;
214 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
215 struct timer_list flow_gc_timer;
216 unsigned long ageing_time;
217 unsigned int numdisabled;
218 struct list_head disabled;
221 struct tun_pcpu_stats __percpu *pcpu_stats;
224 #ifdef CONFIG_TUN_VNET_CROSS_LE
225 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
227 return tun->flags & TUN_VNET_BE ? false :
228 virtio_legacy_is_little_endian();
231 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
233 int be = !!(tun->flags & TUN_VNET_BE);
235 if (put_user(be, argp))
241 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
245 if (get_user(be, argp))
249 tun->flags |= TUN_VNET_BE;
251 tun->flags &= ~TUN_VNET_BE;
256 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
258 return virtio_legacy_is_little_endian();
261 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
266 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
270 #endif /* CONFIG_TUN_VNET_CROSS_LE */
272 static inline bool tun_is_little_endian(struct tun_struct *tun)
274 return tun->flags & TUN_VNET_LE ||
275 tun_legacy_is_little_endian(tun);
278 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
280 return __virtio16_to_cpu(tun_is_little_endian(tun), val);
283 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
285 return __cpu_to_virtio16(tun_is_little_endian(tun), val);
288 static inline u32 tun_hashfn(u32 rxhash)
290 return rxhash & 0x3ff;
293 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
295 struct tun_flow_entry *e;
297 hlist_for_each_entry_rcu(e, head, hash_link) {
298 if (e->rxhash == rxhash)
304 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
305 struct hlist_head *head,
306 u32 rxhash, u16 queue_index)
308 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
311 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
312 rxhash, queue_index);
313 e->updated = jiffies;
316 e->queue_index = queue_index;
318 hlist_add_head_rcu(&e->hash_link, head);
324 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
326 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
327 e->rxhash, e->queue_index);
328 hlist_del_rcu(&e->hash_link);
333 static void tun_flow_flush(struct tun_struct *tun)
337 spin_lock_bh(&tun->lock);
338 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
339 struct tun_flow_entry *e;
340 struct hlist_node *n;
342 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
343 tun_flow_delete(tun, e);
345 spin_unlock_bh(&tun->lock);
348 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
352 spin_lock_bh(&tun->lock);
353 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
354 struct tun_flow_entry *e;
355 struct hlist_node *n;
357 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
358 if (e->queue_index == queue_index)
359 tun_flow_delete(tun, e);
362 spin_unlock_bh(&tun->lock);
365 static void tun_flow_cleanup(unsigned long data)
367 struct tun_struct *tun = (struct tun_struct *)data;
368 unsigned long delay = tun->ageing_time;
369 unsigned long next_timer = jiffies + delay;
370 unsigned long count = 0;
373 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
375 spin_lock_bh(&tun->lock);
376 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
377 struct tun_flow_entry *e;
378 struct hlist_node *n;
380 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
381 unsigned long this_timer;
383 this_timer = e->updated + delay;
384 if (time_before_eq(this_timer, jiffies))
385 tun_flow_delete(tun, e);
386 else if (time_before(this_timer, next_timer))
387 next_timer = this_timer;
392 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
393 spin_unlock_bh(&tun->lock);
396 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
397 struct tun_file *tfile)
399 struct hlist_head *head;
400 struct tun_flow_entry *e;
401 unsigned long delay = tun->ageing_time;
402 u16 queue_index = tfile->queue_index;
407 head = &tun->flows[tun_hashfn(rxhash)];
411 /* We may get a very small possibility of OOO during switching, not
412 * worth to optimize.*/
413 if (tun->numqueues == 1 || tfile->detached)
416 e = tun_flow_find(head, rxhash);
418 /* TODO: keep queueing to old queue until it's empty? */
419 e->queue_index = queue_index;
420 e->updated = jiffies;
421 sock_rps_record_flow_hash(e->rps_rxhash);
423 spin_lock_bh(&tun->lock);
424 if (!tun_flow_find(head, rxhash) &&
425 tun->flow_count < MAX_TAP_FLOWS)
426 tun_flow_create(tun, head, rxhash, queue_index);
428 if (!timer_pending(&tun->flow_gc_timer))
429 mod_timer(&tun->flow_gc_timer,
430 round_jiffies_up(jiffies + delay));
431 spin_unlock_bh(&tun->lock);
439 * Save the hash received in the stack receive path and update the
440 * flow_hash table accordingly.
442 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
444 if (unlikely(e->rps_rxhash != hash))
445 e->rps_rxhash = hash;
448 /* We try to identify a flow through its rxhash first. The reason that
449 * we do not check rxq no. is because some cards(e.g 82599), chooses
450 * the rxq based on the txq where the last packet of the flow comes. As
451 * the userspace application move between processors, we may get a
452 * different rxq no. here. If we could not get rxhash, then we would
453 * hope the rxq no. may help here.
455 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
456 void *accel_priv, select_queue_fallback_t fallback)
458 struct tun_struct *tun = netdev_priv(dev);
459 struct tun_flow_entry *e;
464 numqueues = ACCESS_ONCE(tun->numqueues);
466 txq = skb_get_hash(skb);
468 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
470 tun_flow_save_rps_rxhash(e, txq);
471 txq = e->queue_index;
473 /* use multiply and shift instead of expensive divide */
474 txq = ((u64)txq * numqueues) >> 32;
475 } else if (likely(skb_rx_queue_recorded(skb))) {
476 txq = skb_get_rx_queue(skb);
477 while (unlikely(txq >= numqueues))
485 static inline bool tun_not_capable(struct tun_struct *tun)
487 const struct cred *cred = current_cred();
488 struct net *net = dev_net(tun->dev);
490 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
491 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
492 !ns_capable(net->user_ns, CAP_NET_ADMIN);
495 static void tun_set_real_num_queues(struct tun_struct *tun)
497 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
498 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
501 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
503 tfile->detached = tun;
504 list_add_tail(&tfile->next, &tun->disabled);
508 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
510 struct tun_struct *tun = tfile->detached;
512 tfile->detached = NULL;
513 list_del_init(&tfile->next);
518 static void tun_queue_purge(struct tun_file *tfile)
522 while ((skb = skb_array_consume(&tfile->tx_array)) != NULL)
525 skb_queue_purge(&tfile->sk.sk_error_queue);
528 static void __tun_detach(struct tun_file *tfile, bool clean)
530 struct tun_file *ntfile;
531 struct tun_struct *tun;
533 tun = rtnl_dereference(tfile->tun);
535 if (tun && !tfile->detached) {
536 u16 index = tfile->queue_index;
537 BUG_ON(index >= tun->numqueues);
539 rcu_assign_pointer(tun->tfiles[index],
540 tun->tfiles[tun->numqueues - 1]);
541 ntfile = rtnl_dereference(tun->tfiles[index]);
542 ntfile->queue_index = index;
546 RCU_INIT_POINTER(tfile->tun, NULL);
547 sock_put(&tfile->sk);
549 tun_disable_queue(tun, tfile);
552 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
553 /* Drop read queue */
554 tun_queue_purge(tfile);
555 tun_set_real_num_queues(tun);
556 } else if (tfile->detached && clean) {
557 tun = tun_enable_queue(tfile);
558 sock_put(&tfile->sk);
562 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
563 netif_carrier_off(tun->dev);
565 if (!(tun->flags & IFF_PERSIST) &&
566 tun->dev->reg_state == NETREG_REGISTERED)
567 unregister_netdevice(tun->dev);
570 skb_array_cleanup(&tfile->tx_array);
571 sock_put(&tfile->sk);
575 static void tun_detach(struct tun_file *tfile, bool clean)
578 __tun_detach(tfile, clean);
582 static void tun_detach_all(struct net_device *dev)
584 struct tun_struct *tun = netdev_priv(dev);
585 struct tun_file *tfile, *tmp;
586 int i, n = tun->numqueues;
588 for (i = 0; i < n; i++) {
589 tfile = rtnl_dereference(tun->tfiles[i]);
591 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
592 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
593 RCU_INIT_POINTER(tfile->tun, NULL);
596 list_for_each_entry(tfile, &tun->disabled, next) {
597 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
598 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
599 RCU_INIT_POINTER(tfile->tun, NULL);
601 BUG_ON(tun->numqueues != 0);
604 for (i = 0; i < n; i++) {
605 tfile = rtnl_dereference(tun->tfiles[i]);
606 /* Drop read queue */
607 tun_queue_purge(tfile);
608 sock_put(&tfile->sk);
610 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
611 tun_enable_queue(tfile);
612 tun_queue_purge(tfile);
613 sock_put(&tfile->sk);
615 BUG_ON(tun->numdisabled != 0);
617 if (tun->flags & IFF_PERSIST)
618 module_put(THIS_MODULE);
621 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
623 struct tun_file *tfile = file->private_data;
624 struct net_device *dev = tun->dev;
627 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
632 if (rtnl_dereference(tfile->tun) && !tfile->detached)
636 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
640 if (!tfile->detached &&
641 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
646 /* Re-attach the filter to persist device */
647 if (!skip_filter && (tun->filter_attached == true)) {
648 lock_sock(tfile->socket.sk);
649 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
650 release_sock(tfile->socket.sk);
655 if (!tfile->detached &&
656 skb_array_init(&tfile->tx_array, dev->tx_queue_len, GFP_KERNEL)) {
661 tfile->queue_index = tun->numqueues;
662 tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
663 rcu_assign_pointer(tfile->tun, tun);
664 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
668 tun_enable_queue(tfile);
670 sock_hold(&tfile->sk);
672 tun_set_real_num_queues(tun);
674 /* device is allowed to go away first, so no need to hold extra
682 static struct tun_struct *__tun_get(struct tun_file *tfile)
684 struct tun_struct *tun;
687 tun = rcu_dereference(tfile->tun);
695 static struct tun_struct *tun_get(struct file *file)
697 return __tun_get(file->private_data);
700 static void tun_put(struct tun_struct *tun)
706 static void addr_hash_set(u32 *mask, const u8 *addr)
708 int n = ether_crc(ETH_ALEN, addr) >> 26;
709 mask[n >> 5] |= (1 << (n & 31));
712 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
714 int n = ether_crc(ETH_ALEN, addr) >> 26;
715 return mask[n >> 5] & (1 << (n & 31));
718 static int update_filter(struct tap_filter *filter, void __user *arg)
720 struct { u8 u[ETH_ALEN]; } *addr;
721 struct tun_filter uf;
722 int err, alen, n, nexact;
724 if (copy_from_user(&uf, arg, sizeof(uf)))
733 alen = ETH_ALEN * uf.count;
734 addr = memdup_user(arg + sizeof(uf), alen);
736 return PTR_ERR(addr);
738 /* The filter is updated without holding any locks. Which is
739 * perfectly safe. We disable it first and in the worst
740 * case we'll accept a few undesired packets. */
744 /* Use first set of addresses as an exact filter */
745 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
746 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
750 /* Remaining multicast addresses are hashed,
751 * unicast will leave the filter disabled. */
752 memset(filter->mask, 0, sizeof(filter->mask));
753 for (; n < uf.count; n++) {
754 if (!is_multicast_ether_addr(addr[n].u)) {
755 err = 0; /* no filter */
758 addr_hash_set(filter->mask, addr[n].u);
761 /* For ALLMULTI just set the mask to all ones.
762 * This overrides the mask populated above. */
763 if ((uf.flags & TUN_FLT_ALLMULTI))
764 memset(filter->mask, ~0, sizeof(filter->mask));
766 /* Now enable the filter */
768 filter->count = nexact;
770 /* Return the number of exact filters */
777 /* Returns: 0 - drop, !=0 - accept */
778 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
780 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
782 struct ethhdr *eh = (struct ethhdr *) skb->data;
786 for (i = 0; i < filter->count; i++)
787 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
790 /* Inexact match (multicast only) */
791 if (is_multicast_ether_addr(eh->h_dest))
792 return addr_hash_test(filter->mask, eh->h_dest);
798 * Checks whether the packet is accepted or not.
799 * Returns: 0 - drop, !=0 - accept
801 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
806 return run_filter(filter, skb);
809 /* Network device part of the driver */
811 static const struct ethtool_ops tun_ethtool_ops;
813 /* Net device detach from fd. */
814 static void tun_net_uninit(struct net_device *dev)
819 /* Net device open. */
820 static int tun_net_open(struct net_device *dev)
822 netif_tx_start_all_queues(dev);
826 /* Net device close. */
827 static int tun_net_close(struct net_device *dev)
829 netif_tx_stop_all_queues(dev);
833 /* Net device start xmit */
834 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
836 struct tun_struct *tun = netdev_priv(dev);
837 int txq = skb->queue_mapping;
838 struct tun_file *tfile;
842 tfile = rcu_dereference(tun->tfiles[txq]);
843 numqueues = ACCESS_ONCE(tun->numqueues);
845 /* Drop packet if interface is not attached */
846 if (txq >= numqueues)
850 if (numqueues == 1 && static_key_false(&rps_needed)) {
851 /* Select queue was not called for the skbuff, so we extract the
852 * RPS hash and save it into the flow_table here.
856 rxhash = skb_get_hash(skb);
858 struct tun_flow_entry *e;
859 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
862 tun_flow_save_rps_rxhash(e, rxhash);
867 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
871 /* Drop if the filter does not like it.
872 * This is a noop if the filter is disabled.
873 * Filter can be enabled only for the TAP devices. */
874 if (!check_filter(&tun->txflt, skb))
877 if (tfile->socket.sk->sk_filter &&
878 sk_filter(tfile->socket.sk, skb))
881 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
884 skb_tx_timestamp(skb);
886 /* Orphan the skb - required as we might hang on to it
887 * for indefinite time.
893 if (skb_array_produce(&tfile->tx_array, skb))
896 /* Notify and wake up reader process */
897 if (tfile->flags & TUN_FASYNC)
898 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
899 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
905 this_cpu_inc(tun->pcpu_stats->tx_dropped);
909 return NET_XMIT_DROP;
912 static void tun_net_mclist(struct net_device *dev)
915 * This callback is supposed to deal with mc filter in
916 * _rx_ path and has nothing to do with the _tx_ path.
917 * In rx path we always accept everything userspace gives us.
921 static netdev_features_t tun_net_fix_features(struct net_device *dev,
922 netdev_features_t features)
924 struct tun_struct *tun = netdev_priv(dev);
926 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
928 #ifdef CONFIG_NET_POLL_CONTROLLER
929 static void tun_poll_controller(struct net_device *dev)
932 * Tun only receives frames when:
933 * 1) the char device endpoint gets data from user space
934 * 2) the tun socket gets a sendmsg call from user space
935 * Since both of those are synchronous operations, we are guaranteed
936 * never to have pending data when we poll for it
937 * so there is nothing to do here but return.
938 * We need this though so netpoll recognizes us as an interface that
939 * supports polling, which enables bridge devices in virt setups to
940 * still use netconsole
946 static void tun_set_headroom(struct net_device *dev, int new_hr)
948 struct tun_struct *tun = netdev_priv(dev);
950 if (new_hr < NET_SKB_PAD)
951 new_hr = NET_SKB_PAD;
956 static struct rtnl_link_stats64 *
957 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
959 u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
960 struct tun_struct *tun = netdev_priv(dev);
961 struct tun_pcpu_stats *p;
964 for_each_possible_cpu(i) {
965 u64 rxpackets, rxbytes, txpackets, txbytes;
968 p = per_cpu_ptr(tun->pcpu_stats, i);
970 start = u64_stats_fetch_begin(&p->syncp);
971 rxpackets = p->rx_packets;
972 rxbytes = p->rx_bytes;
973 txpackets = p->tx_packets;
974 txbytes = p->tx_bytes;
975 } while (u64_stats_fetch_retry(&p->syncp, start));
977 stats->rx_packets += rxpackets;
978 stats->rx_bytes += rxbytes;
979 stats->tx_packets += txpackets;
980 stats->tx_bytes += txbytes;
983 rx_dropped += p->rx_dropped;
984 rx_frame_errors += p->rx_frame_errors;
985 tx_dropped += p->tx_dropped;
987 stats->rx_dropped = rx_dropped;
988 stats->rx_frame_errors = rx_frame_errors;
989 stats->tx_dropped = tx_dropped;
993 static const struct net_device_ops tun_netdev_ops = {
994 .ndo_uninit = tun_net_uninit,
995 .ndo_open = tun_net_open,
996 .ndo_stop = tun_net_close,
997 .ndo_start_xmit = tun_net_xmit,
998 .ndo_fix_features = tun_net_fix_features,
999 .ndo_select_queue = tun_select_queue,
1000 #ifdef CONFIG_NET_POLL_CONTROLLER
1001 .ndo_poll_controller = tun_poll_controller,
1003 .ndo_set_rx_headroom = tun_set_headroom,
1004 .ndo_get_stats64 = tun_net_get_stats64,
1007 static const struct net_device_ops tap_netdev_ops = {
1008 .ndo_uninit = tun_net_uninit,
1009 .ndo_open = tun_net_open,
1010 .ndo_stop = tun_net_close,
1011 .ndo_start_xmit = tun_net_xmit,
1012 .ndo_fix_features = tun_net_fix_features,
1013 .ndo_set_rx_mode = tun_net_mclist,
1014 .ndo_set_mac_address = eth_mac_addr,
1015 .ndo_validate_addr = eth_validate_addr,
1016 .ndo_select_queue = tun_select_queue,
1017 #ifdef CONFIG_NET_POLL_CONTROLLER
1018 .ndo_poll_controller = tun_poll_controller,
1020 .ndo_features_check = passthru_features_check,
1021 .ndo_set_rx_headroom = tun_set_headroom,
1022 .ndo_get_stats64 = tun_net_get_stats64,
1025 static void tun_flow_init(struct tun_struct *tun)
1029 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1030 INIT_HLIST_HEAD(&tun->flows[i]);
1032 tun->ageing_time = TUN_FLOW_EXPIRE;
1033 setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
1034 mod_timer(&tun->flow_gc_timer,
1035 round_jiffies_up(jiffies + tun->ageing_time));
1038 static void tun_flow_uninit(struct tun_struct *tun)
1040 del_timer_sync(&tun->flow_gc_timer);
1041 tun_flow_flush(tun);
1045 #define MAX_MTU 65535
1047 /* Initialize net device. */
1048 static void tun_net_init(struct net_device *dev)
1050 struct tun_struct *tun = netdev_priv(dev);
1052 switch (tun->flags & TUN_TYPE_MASK) {
1054 dev->netdev_ops = &tun_netdev_ops;
1056 /* Point-to-Point TUN Device */
1057 dev->hard_header_len = 0;
1061 /* Zero header length */
1062 dev->type = ARPHRD_NONE;
1063 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1067 dev->netdev_ops = &tap_netdev_ops;
1068 /* Ethernet TAP Device */
1070 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1071 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1073 eth_hw_addr_random(dev);
1078 dev->min_mtu = MIN_MTU;
1079 dev->max_mtu = MAX_MTU - dev->hard_header_len;
1082 /* Character device part */
1085 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1087 struct tun_file *tfile = file->private_data;
1088 struct tun_struct *tun = __tun_get(tfile);
1090 unsigned int mask = 0;
1095 sk = tfile->socket.sk;
1097 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1099 poll_wait(file, sk_sleep(sk), wait);
1101 if (!skb_array_empty(&tfile->tx_array))
1102 mask |= POLLIN | POLLRDNORM;
1104 if (sock_writeable(sk) ||
1105 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1106 sock_writeable(sk)))
1107 mask |= POLLOUT | POLLWRNORM;
1109 if (tun->dev->reg_state != NETREG_REGISTERED)
1116 /* prepad is the amount to reserve at front. len is length after that.
1117 * linear is a hint as to how much to copy (usually headers). */
1118 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1119 size_t prepad, size_t len,
1120 size_t linear, int noblock)
1122 struct sock *sk = tfile->socket.sk;
1123 struct sk_buff *skb;
1126 /* Under a page? Don't bother with paged skb. */
1127 if (prepad + len < PAGE_SIZE || !linear)
1130 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1133 return ERR_PTR(err);
1135 skb_reserve(skb, prepad);
1136 skb_put(skb, linear);
1137 skb->data_len = len - linear;
1138 skb->len += len - linear;
1143 /* Get packet from user space buffer */
1144 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1145 void *msg_control, struct iov_iter *from,
1148 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1149 struct sk_buff *skb;
1150 size_t total_len = iov_iter_count(from);
1151 size_t len = total_len, align = tun->align, linear;
1152 struct virtio_net_hdr gso = { 0 };
1153 struct tun_pcpu_stats *stats;
1156 bool zerocopy = false;
1161 if (!(tun->dev->flags & IFF_UP))
1164 if (!(tun->flags & IFF_NO_PI)) {
1165 if (len < sizeof(pi))
1169 n = copy_from_iter(&pi, sizeof(pi), from);
1170 if (n != sizeof(pi))
1174 if (tun->flags & IFF_VNET_HDR) {
1175 if (len < tun->vnet_hdr_sz)
1177 len -= tun->vnet_hdr_sz;
1179 n = copy_from_iter(&gso, sizeof(gso), from);
1180 if (n != sizeof(gso))
1183 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1184 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1185 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1187 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1189 iov_iter_advance(from, tun->vnet_hdr_sz - sizeof(gso));
1192 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1193 align += NET_IP_ALIGN;
1194 if (unlikely(len < ETH_HLEN ||
1195 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1199 good_linear = SKB_MAX_HEAD(align);
1202 struct iov_iter i = *from;
1204 /* There are 256 bytes to be copied in skb, so there is
1205 * enough room for skb expand head in case it is used.
1206 * The rest of the buffer is mapped from userspace.
1208 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1209 if (copylen > good_linear)
1210 copylen = good_linear;
1212 iov_iter_advance(&i, copylen);
1213 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1219 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1220 linear = good_linear;
1222 linear = tun16_to_cpu(tun, gso.hdr_len);
1225 skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1227 if (PTR_ERR(skb) != -EAGAIN)
1228 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1229 return PTR_ERR(skb);
1233 err = zerocopy_sg_from_iter(skb, from);
1235 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1238 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1243 if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1244 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1249 switch (tun->flags & TUN_TYPE_MASK) {
1251 if (tun->flags & IFF_NO_PI) {
1252 switch (skb->data[0] & 0xf0) {
1254 pi.proto = htons(ETH_P_IP);
1257 pi.proto = htons(ETH_P_IPV6);
1260 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1266 skb_reset_mac_header(skb);
1267 skb->protocol = pi.proto;
1268 skb->dev = tun->dev;
1271 skb->protocol = eth_type_trans(skb, tun->dev);
1275 /* copy skb_ubuf_info for callback when skb has no error */
1277 skb_shinfo(skb)->destructor_arg = msg_control;
1278 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1279 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1280 } else if (msg_control) {
1281 struct ubuf_info *uarg = msg_control;
1282 uarg->callback(uarg, false);
1285 skb_reset_network_header(skb);
1286 skb_probe_transport_header(skb, 0);
1288 rxhash = skb_get_hash(skb);
1291 stats = get_cpu_ptr(tun->pcpu_stats);
1292 u64_stats_update_begin(&stats->syncp);
1293 stats->rx_packets++;
1294 stats->rx_bytes += len;
1295 u64_stats_update_end(&stats->syncp);
1298 tun_flow_update(tun, rxhash, tfile);
1302 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1304 struct file *file = iocb->ki_filp;
1305 struct tun_struct *tun = tun_get(file);
1306 struct tun_file *tfile = file->private_data;
1312 result = tun_get_user(tun, tfile, NULL, from, file->f_flags & O_NONBLOCK);
1318 /* Put packet to the user space buffer */
1319 static ssize_t tun_put_user(struct tun_struct *tun,
1320 struct tun_file *tfile,
1321 struct sk_buff *skb,
1322 struct iov_iter *iter)
1324 struct tun_pi pi = { 0, skb->protocol };
1325 struct tun_pcpu_stats *stats;
1327 int vlan_offset = 0;
1329 int vnet_hdr_sz = 0;
1331 if (skb_vlan_tag_present(skb))
1332 vlan_hlen = VLAN_HLEN;
1334 if (tun->flags & IFF_VNET_HDR)
1335 vnet_hdr_sz = tun->vnet_hdr_sz;
1337 total = skb->len + vlan_hlen + vnet_hdr_sz;
1339 if (!(tun->flags & IFF_NO_PI)) {
1340 if (iov_iter_count(iter) < sizeof(pi))
1343 total += sizeof(pi);
1344 if (iov_iter_count(iter) < total) {
1345 /* Packet will be striped */
1346 pi.flags |= TUN_PKT_STRIP;
1349 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1354 struct virtio_net_hdr gso;
1356 if (iov_iter_count(iter) < vnet_hdr_sz)
1359 if (virtio_net_hdr_from_skb(skb, &gso,
1360 tun_is_little_endian(tun))) {
1361 struct skb_shared_info *sinfo = skb_shinfo(skb);
1362 pr_err("unexpected GSO type: "
1363 "0x%x, gso_size %d, hdr_len %d\n",
1364 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1365 tun16_to_cpu(tun, gso.hdr_len));
1366 print_hex_dump(KERN_ERR, "tun: ",
1369 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1374 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1377 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1383 __be16 h_vlan_proto;
1387 veth.h_vlan_proto = skb->vlan_proto;
1388 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1390 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1392 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1393 if (ret || !iov_iter_count(iter))
1396 ret = copy_to_iter(&veth, sizeof(veth), iter);
1397 if (ret != sizeof(veth) || !iov_iter_count(iter))
1401 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1404 /* caller is in process context, */
1405 stats = get_cpu_ptr(tun->pcpu_stats);
1406 u64_stats_update_begin(&stats->syncp);
1407 stats->tx_packets++;
1408 stats->tx_bytes += skb->len + vlan_hlen;
1409 u64_stats_update_end(&stats->syncp);
1410 put_cpu_ptr(tun->pcpu_stats);
1415 static struct sk_buff *tun_ring_recv(struct tun_file *tfile, int noblock,
1418 DECLARE_WAITQUEUE(wait, current);
1419 struct sk_buff *skb = NULL;
1422 skb = skb_array_consume(&tfile->tx_array);
1430 add_wait_queue(&tfile->wq.wait, &wait);
1431 current->state = TASK_INTERRUPTIBLE;
1434 skb = skb_array_consume(&tfile->tx_array);
1437 if (signal_pending(current)) {
1438 error = -ERESTARTSYS;
1441 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
1449 current->state = TASK_RUNNING;
1450 remove_wait_queue(&tfile->wq.wait, &wait);
1457 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1458 struct iov_iter *to,
1461 struct sk_buff *skb;
1465 tun_debug(KERN_INFO, tun, "tun_do_read\n");
1467 if (!iov_iter_count(to))
1470 /* Read frames from ring */
1471 skb = tun_ring_recv(tfile, noblock, &err);
1475 ret = tun_put_user(tun, tfile, skb, to);
1476 if (unlikely(ret < 0))
1484 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1486 struct file *file = iocb->ki_filp;
1487 struct tun_file *tfile = file->private_data;
1488 struct tun_struct *tun = __tun_get(tfile);
1489 ssize_t len = iov_iter_count(to), ret;
1493 ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK);
1494 ret = min_t(ssize_t, ret, len);
1501 static void tun_free_netdev(struct net_device *dev)
1503 struct tun_struct *tun = netdev_priv(dev);
1505 BUG_ON(!(list_empty(&tun->disabled)));
1506 free_percpu(tun->pcpu_stats);
1507 tun_flow_uninit(tun);
1508 security_tun_dev_free_security(tun->security);
1512 static void tun_setup(struct net_device *dev)
1514 struct tun_struct *tun = netdev_priv(dev);
1516 tun->owner = INVALID_UID;
1517 tun->group = INVALID_GID;
1519 dev->ethtool_ops = &tun_ethtool_ops;
1520 dev->destructor = tun_free_netdev;
1521 /* We prefer our own queue length */
1522 dev->tx_queue_len = TUN_READQ_SIZE;
1525 /* Trivial set of netlink ops to allow deleting tun or tap
1526 * device with netlink.
1528 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1533 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1535 .priv_size = sizeof(struct tun_struct),
1537 .validate = tun_validate,
1540 static void tun_sock_write_space(struct sock *sk)
1542 struct tun_file *tfile;
1543 wait_queue_head_t *wqueue;
1545 if (!sock_writeable(sk))
1548 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
1551 wqueue = sk_sleep(sk);
1552 if (wqueue && waitqueue_active(wqueue))
1553 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1554 POLLWRNORM | POLLWRBAND);
1556 tfile = container_of(sk, struct tun_file, sk);
1557 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1560 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1563 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1564 struct tun_struct *tun = __tun_get(tfile);
1569 ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1570 m->msg_flags & MSG_DONTWAIT);
1575 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1578 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1579 struct tun_struct *tun = __tun_get(tfile);
1585 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1589 if (flags & MSG_ERRQUEUE) {
1590 ret = sock_recv_errqueue(sock->sk, m, total_len,
1591 SOL_PACKET, TUN_TX_TIMESTAMP);
1594 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT);
1595 if (ret > (ssize_t)total_len) {
1596 m->msg_flags |= MSG_TRUNC;
1597 ret = flags & MSG_TRUNC ? ret : total_len;
1604 static int tun_peek_len(struct socket *sock)
1606 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1607 struct tun_struct *tun;
1610 tun = __tun_get(tfile);
1614 ret = skb_array_peek_len(&tfile->tx_array);
1620 /* Ops structure to mimic raw sockets with tun */
1621 static const struct proto_ops tun_socket_ops = {
1622 .peek_len = tun_peek_len,
1623 .sendmsg = tun_sendmsg,
1624 .recvmsg = tun_recvmsg,
1627 static struct proto tun_proto = {
1629 .owner = THIS_MODULE,
1630 .obj_size = sizeof(struct tun_file),
1633 static int tun_flags(struct tun_struct *tun)
1635 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1638 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1641 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1642 return sprintf(buf, "0x%x\n", tun_flags(tun));
1645 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1648 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1649 return uid_valid(tun->owner)?
1650 sprintf(buf, "%u\n",
1651 from_kuid_munged(current_user_ns(), tun->owner)):
1652 sprintf(buf, "-1\n");
1655 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1658 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1659 return gid_valid(tun->group) ?
1660 sprintf(buf, "%u\n",
1661 from_kgid_munged(current_user_ns(), tun->group)):
1662 sprintf(buf, "-1\n");
1665 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1666 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1667 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1669 static struct attribute *tun_dev_attrs[] = {
1670 &dev_attr_tun_flags.attr,
1671 &dev_attr_owner.attr,
1672 &dev_attr_group.attr,
1676 static const struct attribute_group tun_attr_group = {
1677 .attrs = tun_dev_attrs
1680 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1682 struct tun_struct *tun;
1683 struct tun_file *tfile = file->private_data;
1684 struct net_device *dev;
1687 if (tfile->detached)
1690 dev = __dev_get_by_name(net, ifr->ifr_name);
1692 if (ifr->ifr_flags & IFF_TUN_EXCL)
1694 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1695 tun = netdev_priv(dev);
1696 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1697 tun = netdev_priv(dev);
1701 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1702 !!(tun->flags & IFF_MULTI_QUEUE))
1705 if (tun_not_capable(tun))
1707 err = security_tun_dev_open(tun->security);
1711 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1715 if (tun->flags & IFF_MULTI_QUEUE &&
1716 (tun->numqueues + tun->numdisabled > 1)) {
1717 /* One or more queue has already been attached, no need
1718 * to initialize the device again.
1725 unsigned long flags = 0;
1726 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1729 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1731 err = security_tun_dev_create();
1736 if (ifr->ifr_flags & IFF_TUN) {
1740 } else if (ifr->ifr_flags & IFF_TAP) {
1748 name = ifr->ifr_name;
1750 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1751 NET_NAME_UNKNOWN, tun_setup, queues,
1757 dev_net_set(dev, net);
1758 dev->rtnl_link_ops = &tun_link_ops;
1759 dev->ifindex = tfile->ifindex;
1760 dev->sysfs_groups[0] = &tun_attr_group;
1762 tun = netdev_priv(dev);
1765 tun->txflt.count = 0;
1766 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1768 tun->align = NET_SKB_PAD;
1769 tun->filter_attached = false;
1770 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1772 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
1773 if (!tun->pcpu_stats) {
1778 spin_lock_init(&tun->lock);
1780 err = security_tun_dev_alloc_security(&tun->security);
1787 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1788 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1789 NETIF_F_HW_VLAN_STAG_TX;
1790 dev->features = dev->hw_features | NETIF_F_LLTX;
1791 dev->vlan_features = dev->features &
1792 ~(NETIF_F_HW_VLAN_CTAG_TX |
1793 NETIF_F_HW_VLAN_STAG_TX);
1795 INIT_LIST_HEAD(&tun->disabled);
1796 err = tun_attach(tun, file, false);
1800 err = register_netdevice(tun->dev);
1805 netif_carrier_on(tun->dev);
1807 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1809 tun->flags = (tun->flags & ~TUN_FEATURES) |
1810 (ifr->ifr_flags & TUN_FEATURES);
1812 /* Make sure persistent devices do not get stuck in
1815 if (netif_running(tun->dev))
1816 netif_tx_wake_all_queues(tun->dev);
1818 strcpy(ifr->ifr_name, tun->dev->name);
1822 tun_detach_all(dev);
1824 tun_flow_uninit(tun);
1825 security_tun_dev_free_security(tun->security);
1827 free_percpu(tun->pcpu_stats);
1833 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1836 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1838 strcpy(ifr->ifr_name, tun->dev->name);
1840 ifr->ifr_flags = tun_flags(tun);
1844 /* This is like a cut-down ethtool ops, except done via tun fd so no
1845 * privs required. */
1846 static int set_offload(struct tun_struct *tun, unsigned long arg)
1848 netdev_features_t features = 0;
1850 if (arg & TUN_F_CSUM) {
1851 features |= NETIF_F_HW_CSUM;
1854 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1855 if (arg & TUN_F_TSO_ECN) {
1856 features |= NETIF_F_TSO_ECN;
1857 arg &= ~TUN_F_TSO_ECN;
1859 if (arg & TUN_F_TSO4)
1860 features |= NETIF_F_TSO;
1861 if (arg & TUN_F_TSO6)
1862 features |= NETIF_F_TSO6;
1863 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1866 if (arg & TUN_F_UFO) {
1867 features |= NETIF_F_UFO;
1872 /* This gives the user a way to test for new features in future by
1873 * trying to set them. */
1877 tun->set_features = features;
1878 netdev_update_features(tun->dev);
1883 static void tun_detach_filter(struct tun_struct *tun, int n)
1886 struct tun_file *tfile;
1888 for (i = 0; i < n; i++) {
1889 tfile = rtnl_dereference(tun->tfiles[i]);
1890 lock_sock(tfile->socket.sk);
1891 sk_detach_filter(tfile->socket.sk);
1892 release_sock(tfile->socket.sk);
1895 tun->filter_attached = false;
1898 static int tun_attach_filter(struct tun_struct *tun)
1901 struct tun_file *tfile;
1903 for (i = 0; i < tun->numqueues; i++) {
1904 tfile = rtnl_dereference(tun->tfiles[i]);
1905 lock_sock(tfile->socket.sk);
1906 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1907 release_sock(tfile->socket.sk);
1909 tun_detach_filter(tun, i);
1914 tun->filter_attached = true;
1918 static void tun_set_sndbuf(struct tun_struct *tun)
1920 struct tun_file *tfile;
1923 for (i = 0; i < tun->numqueues; i++) {
1924 tfile = rtnl_dereference(tun->tfiles[i]);
1925 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1929 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1931 struct tun_file *tfile = file->private_data;
1932 struct tun_struct *tun;
1937 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1938 tun = tfile->detached;
1943 ret = security_tun_dev_attach_queue(tun->security);
1946 ret = tun_attach(tun, file, false);
1947 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1948 tun = rtnl_dereference(tfile->tun);
1949 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
1952 __tun_detach(tfile, false);
1961 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1962 unsigned long arg, int ifreq_len)
1964 struct tun_file *tfile = file->private_data;
1965 struct tun_struct *tun;
1966 void __user* argp = (void __user*)arg;
1972 unsigned int ifindex;
1976 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == SOCK_IOC_TYPE) {
1977 if (copy_from_user(&ifr, argp, ifreq_len))
1980 memset(&ifr, 0, sizeof(ifr));
1982 if (cmd == TUNGETFEATURES) {
1983 /* Currently this just means: "what IFF flags are valid?".
1984 * This is needed because we never checked for invalid flags on
1987 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
1988 (unsigned int __user*)argp);
1989 } else if (cmd == TUNSETQUEUE)
1990 return tun_set_queue(file, &ifr);
1995 tun = __tun_get(tfile);
1996 if (cmd == TUNSETIFF) {
2001 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2003 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
2008 if (copy_to_user(argp, &ifr, ifreq_len))
2012 if (cmd == TUNSETIFINDEX) {
2018 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2022 tfile->ifindex = ifindex;
2030 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2035 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2037 if (tfile->detached)
2038 ifr.ifr_flags |= IFF_DETACH_QUEUE;
2039 if (!tfile->socket.sk->sk_filter)
2040 ifr.ifr_flags |= IFF_NOFILTER;
2042 if (copy_to_user(argp, &ifr, ifreq_len))
2047 /* Disable/Enable checksum */
2049 /* [unimplemented] */
2050 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2051 arg ? "disabled" : "enabled");
2055 /* Disable/Enable persist mode. Keep an extra reference to the
2056 * module to prevent the module being unprobed.
2058 if (arg && !(tun->flags & IFF_PERSIST)) {
2059 tun->flags |= IFF_PERSIST;
2060 __module_get(THIS_MODULE);
2062 if (!arg && (tun->flags & IFF_PERSIST)) {
2063 tun->flags &= ~IFF_PERSIST;
2064 module_put(THIS_MODULE);
2067 tun_debug(KERN_INFO, tun, "persist %s\n",
2068 arg ? "enabled" : "disabled");
2072 /* Set owner of the device */
2073 owner = make_kuid(current_user_ns(), arg);
2074 if (!uid_valid(owner)) {
2079 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2080 from_kuid(&init_user_ns, tun->owner));
2084 /* Set group of the device */
2085 group = make_kgid(current_user_ns(), arg);
2086 if (!gid_valid(group)) {
2091 tun_debug(KERN_INFO, tun, "group set to %u\n",
2092 from_kgid(&init_user_ns, tun->group));
2096 /* Only allow setting the type when the interface is down */
2097 if (tun->dev->flags & IFF_UP) {
2098 tun_debug(KERN_INFO, tun,
2099 "Linktype set failed because interface is up\n");
2102 tun->dev->type = (int) arg;
2103 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2115 ret = set_offload(tun, arg);
2118 case TUNSETTXFILTER:
2119 /* Can be set only for TAPs */
2121 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2123 ret = update_filter(&tun->txflt, (void __user *)arg);
2127 /* Get hw address */
2128 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2129 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2130 if (copy_to_user(argp, &ifr, ifreq_len))
2135 /* Set hw address */
2136 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2137 ifr.ifr_hwaddr.sa_data);
2139 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2143 sndbuf = tfile->socket.sk->sk_sndbuf;
2144 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2149 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2154 tun->sndbuf = sndbuf;
2155 tun_set_sndbuf(tun);
2158 case TUNGETVNETHDRSZ:
2159 vnet_hdr_sz = tun->vnet_hdr_sz;
2160 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2164 case TUNSETVNETHDRSZ:
2165 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2169 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2174 tun->vnet_hdr_sz = vnet_hdr_sz;
2178 le = !!(tun->flags & TUN_VNET_LE);
2179 if (put_user(le, (int __user *)argp))
2184 if (get_user(le, (int __user *)argp)) {
2189 tun->flags |= TUN_VNET_LE;
2191 tun->flags &= ~TUN_VNET_LE;
2195 ret = tun_get_vnet_be(tun, argp);
2199 ret = tun_set_vnet_be(tun, argp);
2202 case TUNATTACHFILTER:
2203 /* Can be set only for TAPs */
2205 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2208 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2211 ret = tun_attach_filter(tun);
2214 case TUNDETACHFILTER:
2215 /* Can be set only for TAPs */
2217 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2220 tun_detach_filter(tun, tun->numqueues);
2225 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2228 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2245 static long tun_chr_ioctl(struct file *file,
2246 unsigned int cmd, unsigned long arg)
2248 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2251 #ifdef CONFIG_COMPAT
2252 static long tun_chr_compat_ioctl(struct file *file,
2253 unsigned int cmd, unsigned long arg)
2258 case TUNSETTXFILTER:
2263 arg = (unsigned long)compat_ptr(arg);
2266 arg = (compat_ulong_t)arg;
2271 * compat_ifreq is shorter than ifreq, so we must not access beyond
2272 * the end of that structure. All fields that are used in this
2273 * driver are compatible though, we don't need to convert the
2276 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2278 #endif /* CONFIG_COMPAT */
2280 static int tun_chr_fasync(int fd, struct file *file, int on)
2282 struct tun_file *tfile = file->private_data;
2285 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2289 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2290 tfile->flags |= TUN_FASYNC;
2292 tfile->flags &= ~TUN_FASYNC;
2298 static int tun_chr_open(struct inode *inode, struct file * file)
2300 struct net *net = current->nsproxy->net_ns;
2301 struct tun_file *tfile;
2303 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2305 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2309 RCU_INIT_POINTER(tfile->tun, NULL);
2313 init_waitqueue_head(&tfile->wq.wait);
2314 RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2316 tfile->socket.file = file;
2317 tfile->socket.ops = &tun_socket_ops;
2319 sock_init_data(&tfile->socket, &tfile->sk);
2321 tfile->sk.sk_write_space = tun_sock_write_space;
2322 tfile->sk.sk_sndbuf = INT_MAX;
2324 file->private_data = tfile;
2325 INIT_LIST_HEAD(&tfile->next);
2327 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2332 static int tun_chr_close(struct inode *inode, struct file *file)
2334 struct tun_file *tfile = file->private_data;
2336 tun_detach(tfile, true);
2341 #ifdef CONFIG_PROC_FS
2342 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2344 struct tun_struct *tun;
2347 memset(&ifr, 0, sizeof(ifr));
2352 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2358 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2362 static const struct file_operations tun_fops = {
2363 .owner = THIS_MODULE,
2364 .llseek = no_llseek,
2365 .read_iter = tun_chr_read_iter,
2366 .write_iter = tun_chr_write_iter,
2367 .poll = tun_chr_poll,
2368 .unlocked_ioctl = tun_chr_ioctl,
2369 #ifdef CONFIG_COMPAT
2370 .compat_ioctl = tun_chr_compat_ioctl,
2372 .open = tun_chr_open,
2373 .release = tun_chr_close,
2374 .fasync = tun_chr_fasync,
2375 #ifdef CONFIG_PROC_FS
2376 .show_fdinfo = tun_chr_show_fdinfo,
2380 static struct miscdevice tun_miscdev = {
2383 .nodename = "net/tun",
2387 /* ethtool interface */
2389 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2392 cmd->advertising = 0;
2393 ethtool_cmd_speed_set(cmd, SPEED_10);
2394 cmd->duplex = DUPLEX_FULL;
2395 cmd->port = PORT_TP;
2396 cmd->phy_address = 0;
2397 cmd->transceiver = XCVR_INTERNAL;
2398 cmd->autoneg = AUTONEG_DISABLE;
2404 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2406 struct tun_struct *tun = netdev_priv(dev);
2408 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2409 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2411 switch (tun->flags & TUN_TYPE_MASK) {
2413 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2416 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2421 static u32 tun_get_msglevel(struct net_device *dev)
2424 struct tun_struct *tun = netdev_priv(dev);
2431 static void tun_set_msglevel(struct net_device *dev, u32 value)
2434 struct tun_struct *tun = netdev_priv(dev);
2439 static const struct ethtool_ops tun_ethtool_ops = {
2440 .get_settings = tun_get_settings,
2441 .get_drvinfo = tun_get_drvinfo,
2442 .get_msglevel = tun_get_msglevel,
2443 .set_msglevel = tun_set_msglevel,
2444 .get_link = ethtool_op_get_link,
2445 .get_ts_info = ethtool_op_get_ts_info,
2448 static int tun_queue_resize(struct tun_struct *tun)
2450 struct net_device *dev = tun->dev;
2451 struct tun_file *tfile;
2452 struct skb_array **arrays;
2453 int n = tun->numqueues + tun->numdisabled;
2456 arrays = kmalloc(sizeof *arrays * n, GFP_KERNEL);
2460 for (i = 0; i < tun->numqueues; i++) {
2461 tfile = rtnl_dereference(tun->tfiles[i]);
2462 arrays[i] = &tfile->tx_array;
2464 list_for_each_entry(tfile, &tun->disabled, next)
2465 arrays[i++] = &tfile->tx_array;
2467 ret = skb_array_resize_multiple(arrays, n,
2468 dev->tx_queue_len, GFP_KERNEL);
2474 static int tun_device_event(struct notifier_block *unused,
2475 unsigned long event, void *ptr)
2477 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2478 struct tun_struct *tun = netdev_priv(dev);
2480 if (dev->rtnl_link_ops != &tun_link_ops)
2484 case NETDEV_CHANGE_TX_QUEUE_LEN:
2485 if (tun_queue_resize(tun))
2495 static struct notifier_block tun_notifier_block __read_mostly = {
2496 .notifier_call = tun_device_event,
2499 static int __init tun_init(void)
2503 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2504 pr_info("%s\n", DRV_COPYRIGHT);
2506 ret = rtnl_link_register(&tun_link_ops);
2508 pr_err("Can't register link_ops\n");
2512 ret = misc_register(&tun_miscdev);
2514 pr_err("Can't register misc device %d\n", TUN_MINOR);
2518 register_netdevice_notifier(&tun_notifier_block);
2521 rtnl_link_unregister(&tun_link_ops);
2526 static void tun_cleanup(void)
2528 misc_deregister(&tun_miscdev);
2529 rtnl_link_unregister(&tun_link_ops);
2530 unregister_netdevice_notifier(&tun_notifier_block);
2533 /* Get an underlying socket object from tun file. Returns error unless file is
2534 * attached to a device. The returned object works like a packet socket, it
2535 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
2536 * holding a reference to the file for as long as the socket is in use. */
2537 struct socket *tun_get_socket(struct file *file)
2539 struct tun_file *tfile;
2540 if (file->f_op != &tun_fops)
2541 return ERR_PTR(-EINVAL);
2542 tfile = file->private_data;
2544 return ERR_PTR(-EBADFD);
2545 return &tfile->socket;
2547 EXPORT_SYMBOL_GPL(tun_get_socket);
2549 module_init(tun_init);
2550 module_exit(tun_cleanup);
2551 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2552 MODULE_AUTHOR(DRV_COPYRIGHT);
2553 MODULE_LICENSE("GPL");
2554 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2555 MODULE_ALIAS("devname:net/tun");