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/sched/signal.h>
48 #include <linux/major.h>
49 #include <linux/slab.h>
50 #include <linux/poll.h>
51 #include <linux/fcntl.h>
52 #include <linux/init.h>
53 #include <linux/skbuff.h>
54 #include <linux/netdevice.h>
55 #include <linux/etherdevice.h>
56 #include <linux/miscdevice.h>
57 #include <linux/ethtool.h>
58 #include <linux/rtnetlink.h>
59 #include <linux/compat.h>
61 #include <linux/if_arp.h>
62 #include <linux/if_ether.h>
63 #include <linux/if_tun.h>
64 #include <linux/if_vlan.h>
65 #include <linux/crc32.h>
66 #include <linux/nsproxy.h>
67 #include <linux/virtio_net.h>
68 #include <linux/rcupdate.h>
69 #include <net/net_namespace.h>
70 #include <net/netns/generic.h>
71 #include <net/rtnetlink.h>
73 #include <linux/seq_file.h>
74 #include <linux/uio.h>
75 #include <linux/skb_array.h>
77 #include <linux/uaccess.h>
79 /* Uncomment to enable debugging */
80 /* #define TUN_DEBUG 1 */
85 #define tun_debug(level, tun, fmt, args...) \
88 netdev_printk(level, tun->dev, fmt, ##args); \
90 #define DBG1(level, fmt, args...) \
93 printk(level fmt, ##args); \
96 #define tun_debug(level, tun, fmt, args...) \
99 netdev_printk(level, tun->dev, fmt, ##args); \
101 #define DBG1(level, fmt, args...) \
104 printk(level fmt, ##args); \
108 /* TUN device flags */
110 /* IFF_ATTACH_QUEUE is never stored in device flags,
111 * overload it to mean fasync when stored there.
113 #define TUN_FASYNC IFF_ATTACH_QUEUE
114 /* High bits in flags field are unused. */
115 #define TUN_VNET_LE 0x80000000
116 #define TUN_VNET_BE 0x40000000
118 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
120 #define GOODCOPY_LEN 128
122 #define FLT_EXACT_COUNT 8
124 unsigned int count; /* Number of addrs. Zero means disabled */
125 u32 mask[2]; /* Mask of the hashed addrs */
126 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
129 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
130 * to max number of VCPUs in guest. */
131 #define MAX_TAP_QUEUES 256
132 #define MAX_TAP_FLOWS 4096
134 #define TUN_FLOW_EXPIRE (3 * HZ)
136 struct tun_pcpu_stats {
141 struct u64_stats_sync syncp;
147 /* A tun_file connects an open character device to a tuntap netdevice. It
148 * also contains all socket related structures (except sock_fprog and tap_filter)
149 * to serve as one transmit queue for tuntap device. The sock_fprog and
150 * tap_filter were kept in tun_struct since they were used for filtering for the
151 * netdevice not for a specific queue (at least I didn't see the requirement for
155 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
156 * other can only be read while rcu_read_lock or rtnl_lock is held.
160 struct socket socket;
162 struct tun_struct __rcu *tun;
163 struct fasync_struct *fasync;
164 /* only used for fasnyc */
168 unsigned int ifindex;
170 struct list_head next;
171 struct tun_struct *detached;
172 struct skb_array tx_array;
175 struct tun_flow_entry {
176 struct hlist_node hash_link;
178 struct tun_struct *tun;
183 unsigned long updated;
186 #define TUN_NUM_FLOW_ENTRIES 1024
188 /* Since the socket were moved to tun_file, to preserve the behavior of persist
189 * device, socket filter, sndbuf and vnet header size were restore when the
190 * file were attached to a persist device.
193 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
194 unsigned int numqueues;
199 struct net_device *dev;
200 netdev_features_t set_features;
201 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
202 NETIF_F_TSO6|NETIF_F_UFO)
207 struct tap_filter txflt;
208 struct sock_fprog fprog;
209 /* protected by rtnl lock */
210 bool filter_attached;
215 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
216 struct timer_list flow_gc_timer;
217 unsigned long ageing_time;
218 unsigned int numdisabled;
219 struct list_head disabled;
223 struct tun_pcpu_stats __percpu *pcpu_stats;
226 #ifdef CONFIG_TUN_VNET_CROSS_LE
227 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
229 return tun->flags & TUN_VNET_BE ? false :
230 virtio_legacy_is_little_endian();
233 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
235 int be = !!(tun->flags & TUN_VNET_BE);
237 if (put_user(be, argp))
243 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
247 if (get_user(be, argp))
251 tun->flags |= TUN_VNET_BE;
253 tun->flags &= ~TUN_VNET_BE;
258 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
260 return virtio_legacy_is_little_endian();
263 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
268 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
272 #endif /* CONFIG_TUN_VNET_CROSS_LE */
274 static inline bool tun_is_little_endian(struct tun_struct *tun)
276 return tun->flags & TUN_VNET_LE ||
277 tun_legacy_is_little_endian(tun);
280 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
282 return __virtio16_to_cpu(tun_is_little_endian(tun), val);
285 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
287 return __cpu_to_virtio16(tun_is_little_endian(tun), val);
290 static inline u32 tun_hashfn(u32 rxhash)
292 return rxhash & 0x3ff;
295 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
297 struct tun_flow_entry *e;
299 hlist_for_each_entry_rcu(e, head, hash_link) {
300 if (e->rxhash == rxhash)
306 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
307 struct hlist_head *head,
308 u32 rxhash, u16 queue_index)
310 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
313 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
314 rxhash, queue_index);
315 e->updated = jiffies;
318 e->queue_index = queue_index;
320 hlist_add_head_rcu(&e->hash_link, head);
326 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
328 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
329 e->rxhash, e->queue_index);
330 hlist_del_rcu(&e->hash_link);
335 static void tun_flow_flush(struct tun_struct *tun)
339 spin_lock_bh(&tun->lock);
340 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
341 struct tun_flow_entry *e;
342 struct hlist_node *n;
344 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
345 tun_flow_delete(tun, e);
347 spin_unlock_bh(&tun->lock);
350 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
354 spin_lock_bh(&tun->lock);
355 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
356 struct tun_flow_entry *e;
357 struct hlist_node *n;
359 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
360 if (e->queue_index == queue_index)
361 tun_flow_delete(tun, e);
364 spin_unlock_bh(&tun->lock);
367 static void tun_flow_cleanup(unsigned long data)
369 struct tun_struct *tun = (struct tun_struct *)data;
370 unsigned long delay = tun->ageing_time;
371 unsigned long next_timer = jiffies + delay;
372 unsigned long count = 0;
375 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
377 spin_lock_bh(&tun->lock);
378 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
379 struct tun_flow_entry *e;
380 struct hlist_node *n;
382 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
383 unsigned long this_timer;
385 this_timer = e->updated + delay;
386 if (time_before_eq(this_timer, jiffies))
387 tun_flow_delete(tun, e);
388 else if (time_before(this_timer, next_timer))
389 next_timer = this_timer;
394 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
395 spin_unlock_bh(&tun->lock);
398 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
399 struct tun_file *tfile)
401 struct hlist_head *head;
402 struct tun_flow_entry *e;
403 unsigned long delay = tun->ageing_time;
404 u16 queue_index = tfile->queue_index;
409 head = &tun->flows[tun_hashfn(rxhash)];
413 /* We may get a very small possibility of OOO during switching, not
414 * worth to optimize.*/
415 if (tun->numqueues == 1 || tfile->detached)
418 e = tun_flow_find(head, rxhash);
420 /* TODO: keep queueing to old queue until it's empty? */
421 e->queue_index = queue_index;
422 e->updated = jiffies;
423 sock_rps_record_flow_hash(e->rps_rxhash);
425 spin_lock_bh(&tun->lock);
426 if (!tun_flow_find(head, rxhash) &&
427 tun->flow_count < MAX_TAP_FLOWS)
428 tun_flow_create(tun, head, rxhash, queue_index);
430 if (!timer_pending(&tun->flow_gc_timer))
431 mod_timer(&tun->flow_gc_timer,
432 round_jiffies_up(jiffies + delay));
433 spin_unlock_bh(&tun->lock);
441 * Save the hash received in the stack receive path and update the
442 * flow_hash table accordingly.
444 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
446 if (unlikely(e->rps_rxhash != hash))
447 e->rps_rxhash = hash;
450 /* We try to identify a flow through its rxhash first. The reason that
451 * we do not check rxq no. is because some cards(e.g 82599), chooses
452 * the rxq based on the txq where the last packet of the flow comes. As
453 * the userspace application move between processors, we may get a
454 * different rxq no. here. If we could not get rxhash, then we would
455 * hope the rxq no. may help here.
457 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
458 void *accel_priv, select_queue_fallback_t fallback)
460 struct tun_struct *tun = netdev_priv(dev);
461 struct tun_flow_entry *e;
466 numqueues = ACCESS_ONCE(tun->numqueues);
468 txq = skb_get_hash(skb);
470 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
472 tun_flow_save_rps_rxhash(e, txq);
473 txq = e->queue_index;
475 /* use multiply and shift instead of expensive divide */
476 txq = ((u64)txq * numqueues) >> 32;
477 } else if (likely(skb_rx_queue_recorded(skb))) {
478 txq = skb_get_rx_queue(skb);
479 while (unlikely(txq >= numqueues))
487 static inline bool tun_not_capable(struct tun_struct *tun)
489 const struct cred *cred = current_cred();
490 struct net *net = dev_net(tun->dev);
492 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
493 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
494 !ns_capable(net->user_ns, CAP_NET_ADMIN);
497 static void tun_set_real_num_queues(struct tun_struct *tun)
499 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
500 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
503 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
505 tfile->detached = tun;
506 list_add_tail(&tfile->next, &tun->disabled);
510 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
512 struct tun_struct *tun = tfile->detached;
514 tfile->detached = NULL;
515 list_del_init(&tfile->next);
520 static void tun_queue_purge(struct tun_file *tfile)
524 while ((skb = skb_array_consume(&tfile->tx_array)) != NULL)
527 skb_queue_purge(&tfile->sk.sk_write_queue);
528 skb_queue_purge(&tfile->sk.sk_error_queue);
531 static void __tun_detach(struct tun_file *tfile, bool clean)
533 struct tun_file *ntfile;
534 struct tun_struct *tun;
536 tun = rtnl_dereference(tfile->tun);
538 if (tun && !tfile->detached) {
539 u16 index = tfile->queue_index;
540 BUG_ON(index >= tun->numqueues);
542 rcu_assign_pointer(tun->tfiles[index],
543 tun->tfiles[tun->numqueues - 1]);
544 ntfile = rtnl_dereference(tun->tfiles[index]);
545 ntfile->queue_index = index;
549 RCU_INIT_POINTER(tfile->tun, NULL);
550 sock_put(&tfile->sk);
552 tun_disable_queue(tun, tfile);
555 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
556 /* Drop read queue */
557 tun_queue_purge(tfile);
558 tun_set_real_num_queues(tun);
559 } else if (tfile->detached && clean) {
560 tun = tun_enable_queue(tfile);
561 sock_put(&tfile->sk);
565 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
566 netif_carrier_off(tun->dev);
568 if (!(tun->flags & IFF_PERSIST) &&
569 tun->dev->reg_state == NETREG_REGISTERED)
570 unregister_netdevice(tun->dev);
573 skb_array_cleanup(&tfile->tx_array);
574 sock_put(&tfile->sk);
578 static void tun_detach(struct tun_file *tfile, bool clean)
581 __tun_detach(tfile, clean);
585 static void tun_detach_all(struct net_device *dev)
587 struct tun_struct *tun = netdev_priv(dev);
588 struct tun_file *tfile, *tmp;
589 int i, n = tun->numqueues;
591 for (i = 0; i < n; i++) {
592 tfile = rtnl_dereference(tun->tfiles[i]);
594 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
595 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
596 RCU_INIT_POINTER(tfile->tun, NULL);
599 list_for_each_entry(tfile, &tun->disabled, next) {
600 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
601 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
602 RCU_INIT_POINTER(tfile->tun, NULL);
604 BUG_ON(tun->numqueues != 0);
607 for (i = 0; i < n; i++) {
608 tfile = rtnl_dereference(tun->tfiles[i]);
609 /* Drop read queue */
610 tun_queue_purge(tfile);
611 sock_put(&tfile->sk);
613 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
614 tun_enable_queue(tfile);
615 tun_queue_purge(tfile);
616 sock_put(&tfile->sk);
618 BUG_ON(tun->numdisabled != 0);
620 if (tun->flags & IFF_PERSIST)
621 module_put(THIS_MODULE);
624 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
626 struct tun_file *tfile = file->private_data;
627 struct net_device *dev = tun->dev;
630 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
635 if (rtnl_dereference(tfile->tun) && !tfile->detached)
639 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
643 if (!tfile->detached &&
644 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
649 /* Re-attach the filter to persist device */
650 if (!skip_filter && (tun->filter_attached == true)) {
651 lock_sock(tfile->socket.sk);
652 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
653 release_sock(tfile->socket.sk);
658 if (!tfile->detached &&
659 skb_array_init(&tfile->tx_array, dev->tx_queue_len, GFP_KERNEL)) {
664 tfile->queue_index = tun->numqueues;
665 tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
666 rcu_assign_pointer(tfile->tun, tun);
667 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
671 tun_enable_queue(tfile);
673 sock_hold(&tfile->sk);
675 tun_set_real_num_queues(tun);
677 /* device is allowed to go away first, so no need to hold extra
685 static struct tun_struct *__tun_get(struct tun_file *tfile)
687 struct tun_struct *tun;
690 tun = rcu_dereference(tfile->tun);
698 static struct tun_struct *tun_get(struct file *file)
700 return __tun_get(file->private_data);
703 static void tun_put(struct tun_struct *tun)
709 static void addr_hash_set(u32 *mask, const u8 *addr)
711 int n = ether_crc(ETH_ALEN, addr) >> 26;
712 mask[n >> 5] |= (1 << (n & 31));
715 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
717 int n = ether_crc(ETH_ALEN, addr) >> 26;
718 return mask[n >> 5] & (1 << (n & 31));
721 static int update_filter(struct tap_filter *filter, void __user *arg)
723 struct { u8 u[ETH_ALEN]; } *addr;
724 struct tun_filter uf;
725 int err, alen, n, nexact;
727 if (copy_from_user(&uf, arg, sizeof(uf)))
736 alen = ETH_ALEN * uf.count;
737 addr = memdup_user(arg + sizeof(uf), alen);
739 return PTR_ERR(addr);
741 /* The filter is updated without holding any locks. Which is
742 * perfectly safe. We disable it first and in the worst
743 * case we'll accept a few undesired packets. */
747 /* Use first set of addresses as an exact filter */
748 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
749 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
753 /* Remaining multicast addresses are hashed,
754 * unicast will leave the filter disabled. */
755 memset(filter->mask, 0, sizeof(filter->mask));
756 for (; n < uf.count; n++) {
757 if (!is_multicast_ether_addr(addr[n].u)) {
758 err = 0; /* no filter */
761 addr_hash_set(filter->mask, addr[n].u);
764 /* For ALLMULTI just set the mask to all ones.
765 * This overrides the mask populated above. */
766 if ((uf.flags & TUN_FLT_ALLMULTI))
767 memset(filter->mask, ~0, sizeof(filter->mask));
769 /* Now enable the filter */
771 filter->count = nexact;
773 /* Return the number of exact filters */
780 /* Returns: 0 - drop, !=0 - accept */
781 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
783 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
785 struct ethhdr *eh = (struct ethhdr *) skb->data;
789 for (i = 0; i < filter->count; i++)
790 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
793 /* Inexact match (multicast only) */
794 if (is_multicast_ether_addr(eh->h_dest))
795 return addr_hash_test(filter->mask, eh->h_dest);
801 * Checks whether the packet is accepted or not.
802 * Returns: 0 - drop, !=0 - accept
804 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
809 return run_filter(filter, skb);
812 /* Network device part of the driver */
814 static const struct ethtool_ops tun_ethtool_ops;
816 /* Net device detach from fd. */
817 static void tun_net_uninit(struct net_device *dev)
822 /* Net device open. */
823 static int tun_net_open(struct net_device *dev)
825 netif_tx_start_all_queues(dev);
829 /* Net device close. */
830 static int tun_net_close(struct net_device *dev)
832 netif_tx_stop_all_queues(dev);
836 /* Net device start xmit */
837 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
839 struct tun_struct *tun = netdev_priv(dev);
840 int txq = skb->queue_mapping;
841 struct tun_file *tfile;
845 tfile = rcu_dereference(tun->tfiles[txq]);
846 numqueues = ACCESS_ONCE(tun->numqueues);
848 /* Drop packet if interface is not attached */
849 if (txq >= numqueues)
853 if (numqueues == 1 && static_key_false(&rps_needed)) {
854 /* Select queue was not called for the skbuff, so we extract the
855 * RPS hash and save it into the flow_table here.
859 rxhash = skb_get_hash(skb);
861 struct tun_flow_entry *e;
862 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
865 tun_flow_save_rps_rxhash(e, rxhash);
870 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
874 /* Drop if the filter does not like it.
875 * This is a noop if the filter is disabled.
876 * Filter can be enabled only for the TAP devices. */
877 if (!check_filter(&tun->txflt, skb))
880 if (tfile->socket.sk->sk_filter &&
881 sk_filter(tfile->socket.sk, skb))
884 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
887 skb_tx_timestamp(skb);
889 /* Orphan the skb - required as we might hang on to it
890 * for indefinite time.
896 if (skb_array_produce(&tfile->tx_array, skb))
899 /* Notify and wake up reader process */
900 if (tfile->flags & TUN_FASYNC)
901 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
902 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
908 this_cpu_inc(tun->pcpu_stats->tx_dropped);
912 return NET_XMIT_DROP;
915 static void tun_net_mclist(struct net_device *dev)
918 * This callback is supposed to deal with mc filter in
919 * _rx_ path and has nothing to do with the _tx_ path.
920 * In rx path we always accept everything userspace gives us.
924 static netdev_features_t tun_net_fix_features(struct net_device *dev,
925 netdev_features_t features)
927 struct tun_struct *tun = netdev_priv(dev);
929 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
931 #ifdef CONFIG_NET_POLL_CONTROLLER
932 static void tun_poll_controller(struct net_device *dev)
935 * Tun only receives frames when:
936 * 1) the char device endpoint gets data from user space
937 * 2) the tun socket gets a sendmsg call from user space
938 * Since both of those are synchronous operations, we are guaranteed
939 * never to have pending data when we poll for it
940 * so there is nothing to do here but return.
941 * We need this though so netpoll recognizes us as an interface that
942 * supports polling, which enables bridge devices in virt setups to
943 * still use netconsole
949 static void tun_set_headroom(struct net_device *dev, int new_hr)
951 struct tun_struct *tun = netdev_priv(dev);
953 if (new_hr < NET_SKB_PAD)
954 new_hr = NET_SKB_PAD;
960 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
962 u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
963 struct tun_struct *tun = netdev_priv(dev);
964 struct tun_pcpu_stats *p;
967 for_each_possible_cpu(i) {
968 u64 rxpackets, rxbytes, txpackets, txbytes;
971 p = per_cpu_ptr(tun->pcpu_stats, i);
973 start = u64_stats_fetch_begin(&p->syncp);
974 rxpackets = p->rx_packets;
975 rxbytes = p->rx_bytes;
976 txpackets = p->tx_packets;
977 txbytes = p->tx_bytes;
978 } while (u64_stats_fetch_retry(&p->syncp, start));
980 stats->rx_packets += rxpackets;
981 stats->rx_bytes += rxbytes;
982 stats->tx_packets += txpackets;
983 stats->tx_bytes += txbytes;
986 rx_dropped += p->rx_dropped;
987 rx_frame_errors += p->rx_frame_errors;
988 tx_dropped += p->tx_dropped;
990 stats->rx_dropped = rx_dropped;
991 stats->rx_frame_errors = rx_frame_errors;
992 stats->tx_dropped = tx_dropped;
995 static const struct net_device_ops tun_netdev_ops = {
996 .ndo_uninit = tun_net_uninit,
997 .ndo_open = tun_net_open,
998 .ndo_stop = tun_net_close,
999 .ndo_start_xmit = tun_net_xmit,
1000 .ndo_fix_features = tun_net_fix_features,
1001 .ndo_select_queue = tun_select_queue,
1002 #ifdef CONFIG_NET_POLL_CONTROLLER
1003 .ndo_poll_controller = tun_poll_controller,
1005 .ndo_set_rx_headroom = tun_set_headroom,
1006 .ndo_get_stats64 = tun_net_get_stats64,
1009 static const struct net_device_ops tap_netdev_ops = {
1010 .ndo_uninit = tun_net_uninit,
1011 .ndo_open = tun_net_open,
1012 .ndo_stop = tun_net_close,
1013 .ndo_start_xmit = tun_net_xmit,
1014 .ndo_fix_features = tun_net_fix_features,
1015 .ndo_set_rx_mode = tun_net_mclist,
1016 .ndo_set_mac_address = eth_mac_addr,
1017 .ndo_validate_addr = eth_validate_addr,
1018 .ndo_select_queue = tun_select_queue,
1019 #ifdef CONFIG_NET_POLL_CONTROLLER
1020 .ndo_poll_controller = tun_poll_controller,
1022 .ndo_features_check = passthru_features_check,
1023 .ndo_set_rx_headroom = tun_set_headroom,
1024 .ndo_get_stats64 = tun_net_get_stats64,
1027 static void tun_flow_init(struct tun_struct *tun)
1031 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1032 INIT_HLIST_HEAD(&tun->flows[i]);
1034 tun->ageing_time = TUN_FLOW_EXPIRE;
1035 setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
1036 mod_timer(&tun->flow_gc_timer,
1037 round_jiffies_up(jiffies + tun->ageing_time));
1040 static void tun_flow_uninit(struct tun_struct *tun)
1042 del_timer_sync(&tun->flow_gc_timer);
1043 tun_flow_flush(tun);
1047 #define MAX_MTU 65535
1049 /* Initialize net device. */
1050 static void tun_net_init(struct net_device *dev)
1052 struct tun_struct *tun = netdev_priv(dev);
1054 switch (tun->flags & TUN_TYPE_MASK) {
1056 dev->netdev_ops = &tun_netdev_ops;
1058 /* Point-to-Point TUN Device */
1059 dev->hard_header_len = 0;
1063 /* Zero header length */
1064 dev->type = ARPHRD_NONE;
1065 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1069 dev->netdev_ops = &tap_netdev_ops;
1070 /* Ethernet TAP Device */
1072 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1073 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1075 eth_hw_addr_random(dev);
1080 dev->min_mtu = MIN_MTU;
1081 dev->max_mtu = MAX_MTU - dev->hard_header_len;
1084 /* Character device part */
1087 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1089 struct tun_file *tfile = file->private_data;
1090 struct tun_struct *tun = __tun_get(tfile);
1092 unsigned int mask = 0;
1097 sk = tfile->socket.sk;
1099 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1101 poll_wait(file, sk_sleep(sk), wait);
1103 if (!skb_array_empty(&tfile->tx_array))
1104 mask |= POLLIN | POLLRDNORM;
1106 if (sock_writeable(sk) ||
1107 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1108 sock_writeable(sk)))
1109 mask |= POLLOUT | POLLWRNORM;
1111 if (tun->dev->reg_state != NETREG_REGISTERED)
1118 /* prepad is the amount to reserve at front. len is length after that.
1119 * linear is a hint as to how much to copy (usually headers). */
1120 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1121 size_t prepad, size_t len,
1122 size_t linear, int noblock)
1124 struct sock *sk = tfile->socket.sk;
1125 struct sk_buff *skb;
1128 /* Under a page? Don't bother with paged skb. */
1129 if (prepad + len < PAGE_SIZE || !linear)
1132 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1135 return ERR_PTR(err);
1137 skb_reserve(skb, prepad);
1138 skb_put(skb, linear);
1139 skb->data_len = len - linear;
1140 skb->len += len - linear;
1145 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1146 struct sk_buff *skb, int more)
1148 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1149 struct sk_buff_head process_queue;
1150 u32 rx_batched = tun->rx_batched;
1153 if (!rx_batched || (!more && skb_queue_empty(queue))) {
1155 netif_receive_skb(skb);
1160 spin_lock(&queue->lock);
1161 if (!more || skb_queue_len(queue) == rx_batched) {
1162 __skb_queue_head_init(&process_queue);
1163 skb_queue_splice_tail_init(queue, &process_queue);
1166 __skb_queue_tail(queue, skb);
1168 spin_unlock(&queue->lock);
1171 struct sk_buff *nskb;
1174 while ((nskb = __skb_dequeue(&process_queue)))
1175 netif_receive_skb(nskb);
1176 netif_receive_skb(skb);
1181 /* Get packet from user space buffer */
1182 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1183 void *msg_control, struct iov_iter *from,
1184 int noblock, bool more)
1186 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1187 struct sk_buff *skb;
1188 size_t total_len = iov_iter_count(from);
1189 size_t len = total_len, align = tun->align, linear;
1190 struct virtio_net_hdr gso = { 0 };
1191 struct tun_pcpu_stats *stats;
1194 bool zerocopy = false;
1198 if (!(tun->dev->flags & IFF_UP))
1201 if (!(tun->flags & IFF_NO_PI)) {
1202 if (len < sizeof(pi))
1206 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1210 if (tun->flags & IFF_VNET_HDR) {
1211 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1213 if (len < vnet_hdr_sz)
1217 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1220 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1221 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1222 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1224 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1226 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1229 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1230 align += NET_IP_ALIGN;
1231 if (unlikely(len < ETH_HLEN ||
1232 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1236 good_linear = SKB_MAX_HEAD(align);
1239 struct iov_iter i = *from;
1241 /* There are 256 bytes to be copied in skb, so there is
1242 * enough room for skb expand head in case it is used.
1243 * The rest of the buffer is mapped from userspace.
1245 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1246 if (copylen > good_linear)
1247 copylen = good_linear;
1249 iov_iter_advance(&i, copylen);
1250 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1256 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1257 linear = good_linear;
1259 linear = tun16_to_cpu(tun, gso.hdr_len);
1262 skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1264 if (PTR_ERR(skb) != -EAGAIN)
1265 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1266 return PTR_ERR(skb);
1270 err = zerocopy_sg_from_iter(skb, from);
1272 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1275 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1280 if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1281 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1286 switch (tun->flags & TUN_TYPE_MASK) {
1288 if (tun->flags & IFF_NO_PI) {
1289 switch (skb->data[0] & 0xf0) {
1291 pi.proto = htons(ETH_P_IP);
1294 pi.proto = htons(ETH_P_IPV6);
1297 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1303 skb_reset_mac_header(skb);
1304 skb->protocol = pi.proto;
1305 skb->dev = tun->dev;
1308 skb->protocol = eth_type_trans(skb, tun->dev);
1312 /* copy skb_ubuf_info for callback when skb has no error */
1314 skb_shinfo(skb)->destructor_arg = msg_control;
1315 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1316 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1317 } else if (msg_control) {
1318 struct ubuf_info *uarg = msg_control;
1319 uarg->callback(uarg, false);
1322 skb_reset_network_header(skb);
1323 skb_probe_transport_header(skb, 0);
1325 rxhash = skb_get_hash(skb);
1326 #ifndef CONFIG_4KSTACKS
1327 tun_rx_batched(tun, tfile, skb, more);
1332 stats = get_cpu_ptr(tun->pcpu_stats);
1333 u64_stats_update_begin(&stats->syncp);
1334 stats->rx_packets++;
1335 stats->rx_bytes += len;
1336 u64_stats_update_end(&stats->syncp);
1339 tun_flow_update(tun, rxhash, tfile);
1343 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1345 struct file *file = iocb->ki_filp;
1346 struct tun_struct *tun = tun_get(file);
1347 struct tun_file *tfile = file->private_data;
1353 result = tun_get_user(tun, tfile, NULL, from,
1354 file->f_flags & O_NONBLOCK, false);
1360 /* Put packet to the user space buffer */
1361 static ssize_t tun_put_user(struct tun_struct *tun,
1362 struct tun_file *tfile,
1363 struct sk_buff *skb,
1364 struct iov_iter *iter)
1366 struct tun_pi pi = { 0, skb->protocol };
1367 struct tun_pcpu_stats *stats;
1369 int vlan_offset = 0;
1371 int vnet_hdr_sz = 0;
1373 if (skb_vlan_tag_present(skb))
1374 vlan_hlen = VLAN_HLEN;
1376 if (tun->flags & IFF_VNET_HDR)
1377 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1379 total = skb->len + vlan_hlen + vnet_hdr_sz;
1381 if (!(tun->flags & IFF_NO_PI)) {
1382 if (iov_iter_count(iter) < sizeof(pi))
1385 total += sizeof(pi);
1386 if (iov_iter_count(iter) < total) {
1387 /* Packet will be striped */
1388 pi.flags |= TUN_PKT_STRIP;
1391 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1396 struct virtio_net_hdr gso;
1398 if (iov_iter_count(iter) < vnet_hdr_sz)
1401 if (virtio_net_hdr_from_skb(skb, &gso,
1402 tun_is_little_endian(tun), true)) {
1403 struct skb_shared_info *sinfo = skb_shinfo(skb);
1404 pr_err("unexpected GSO type: "
1405 "0x%x, gso_size %d, hdr_len %d\n",
1406 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1407 tun16_to_cpu(tun, gso.hdr_len));
1408 print_hex_dump(KERN_ERR, "tun: ",
1411 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1416 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1419 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1425 __be16 h_vlan_proto;
1429 veth.h_vlan_proto = skb->vlan_proto;
1430 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1432 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1434 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1435 if (ret || !iov_iter_count(iter))
1438 ret = copy_to_iter(&veth, sizeof(veth), iter);
1439 if (ret != sizeof(veth) || !iov_iter_count(iter))
1443 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1446 /* caller is in process context, */
1447 stats = get_cpu_ptr(tun->pcpu_stats);
1448 u64_stats_update_begin(&stats->syncp);
1449 stats->tx_packets++;
1450 stats->tx_bytes += skb->len + vlan_hlen;
1451 u64_stats_update_end(&stats->syncp);
1452 put_cpu_ptr(tun->pcpu_stats);
1457 static struct sk_buff *tun_ring_recv(struct tun_file *tfile, int noblock,
1460 DECLARE_WAITQUEUE(wait, current);
1461 struct sk_buff *skb = NULL;
1464 skb = skb_array_consume(&tfile->tx_array);
1472 add_wait_queue(&tfile->wq.wait, &wait);
1473 current->state = TASK_INTERRUPTIBLE;
1476 skb = skb_array_consume(&tfile->tx_array);
1479 if (signal_pending(current)) {
1480 error = -ERESTARTSYS;
1483 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
1491 current->state = TASK_RUNNING;
1492 remove_wait_queue(&tfile->wq.wait, &wait);
1499 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1500 struct iov_iter *to,
1503 struct sk_buff *skb;
1507 tun_debug(KERN_INFO, tun, "tun_do_read\n");
1509 if (!iov_iter_count(to))
1512 /* Read frames from ring */
1513 skb = tun_ring_recv(tfile, noblock, &err);
1517 ret = tun_put_user(tun, tfile, skb, to);
1518 if (unlikely(ret < 0))
1526 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1528 struct file *file = iocb->ki_filp;
1529 struct tun_file *tfile = file->private_data;
1530 struct tun_struct *tun = __tun_get(tfile);
1531 ssize_t len = iov_iter_count(to), ret;
1535 ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK);
1536 ret = min_t(ssize_t, ret, len);
1543 static void tun_free_netdev(struct net_device *dev)
1545 struct tun_struct *tun = netdev_priv(dev);
1547 BUG_ON(!(list_empty(&tun->disabled)));
1548 free_percpu(tun->pcpu_stats);
1549 tun_flow_uninit(tun);
1550 security_tun_dev_free_security(tun->security);
1554 static void tun_setup(struct net_device *dev)
1556 struct tun_struct *tun = netdev_priv(dev);
1558 tun->owner = INVALID_UID;
1559 tun->group = INVALID_GID;
1561 dev->ethtool_ops = &tun_ethtool_ops;
1562 dev->destructor = tun_free_netdev;
1563 /* We prefer our own queue length */
1564 dev->tx_queue_len = TUN_READQ_SIZE;
1567 /* Trivial set of netlink ops to allow deleting tun or tap
1568 * device with netlink.
1570 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1575 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1577 .priv_size = sizeof(struct tun_struct),
1579 .validate = tun_validate,
1582 static void tun_sock_write_space(struct sock *sk)
1584 struct tun_file *tfile;
1585 wait_queue_head_t *wqueue;
1587 if (!sock_writeable(sk))
1590 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
1593 wqueue = sk_sleep(sk);
1594 if (wqueue && waitqueue_active(wqueue))
1595 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1596 POLLWRNORM | POLLWRBAND);
1598 tfile = container_of(sk, struct tun_file, sk);
1599 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1602 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1605 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1606 struct tun_struct *tun = __tun_get(tfile);
1611 ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1612 m->msg_flags & MSG_DONTWAIT,
1613 m->msg_flags & MSG_MORE);
1618 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1621 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1622 struct tun_struct *tun = __tun_get(tfile);
1628 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1632 if (flags & MSG_ERRQUEUE) {
1633 ret = sock_recv_errqueue(sock->sk, m, total_len,
1634 SOL_PACKET, TUN_TX_TIMESTAMP);
1637 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT);
1638 if (ret > (ssize_t)total_len) {
1639 m->msg_flags |= MSG_TRUNC;
1640 ret = flags & MSG_TRUNC ? ret : total_len;
1647 static int tun_peek_len(struct socket *sock)
1649 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1650 struct tun_struct *tun;
1653 tun = __tun_get(tfile);
1657 ret = skb_array_peek_len(&tfile->tx_array);
1663 /* Ops structure to mimic raw sockets with tun */
1664 static const struct proto_ops tun_socket_ops = {
1665 .peek_len = tun_peek_len,
1666 .sendmsg = tun_sendmsg,
1667 .recvmsg = tun_recvmsg,
1670 static struct proto tun_proto = {
1672 .owner = THIS_MODULE,
1673 .obj_size = sizeof(struct tun_file),
1676 static int tun_flags(struct tun_struct *tun)
1678 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1681 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1684 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1685 return sprintf(buf, "0x%x\n", tun_flags(tun));
1688 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1691 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1692 return uid_valid(tun->owner)?
1693 sprintf(buf, "%u\n",
1694 from_kuid_munged(current_user_ns(), tun->owner)):
1695 sprintf(buf, "-1\n");
1698 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1701 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1702 return gid_valid(tun->group) ?
1703 sprintf(buf, "%u\n",
1704 from_kgid_munged(current_user_ns(), tun->group)):
1705 sprintf(buf, "-1\n");
1708 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1709 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1710 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1712 static struct attribute *tun_dev_attrs[] = {
1713 &dev_attr_tun_flags.attr,
1714 &dev_attr_owner.attr,
1715 &dev_attr_group.attr,
1719 static const struct attribute_group tun_attr_group = {
1720 .attrs = tun_dev_attrs
1723 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1725 struct tun_struct *tun;
1726 struct tun_file *tfile = file->private_data;
1727 struct net_device *dev;
1730 if (tfile->detached)
1733 dev = __dev_get_by_name(net, ifr->ifr_name);
1735 if (ifr->ifr_flags & IFF_TUN_EXCL)
1737 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1738 tun = netdev_priv(dev);
1739 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1740 tun = netdev_priv(dev);
1744 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
1745 !!(tun->flags & IFF_MULTI_QUEUE))
1748 if (tun_not_capable(tun))
1750 err = security_tun_dev_open(tun->security);
1754 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER);
1758 if (tun->flags & IFF_MULTI_QUEUE &&
1759 (tun->numqueues + tun->numdisabled > 1)) {
1760 /* One or more queue has already been attached, no need
1761 * to initialize the device again.
1768 unsigned long flags = 0;
1769 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
1772 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1774 err = security_tun_dev_create();
1779 if (ifr->ifr_flags & IFF_TUN) {
1783 } else if (ifr->ifr_flags & IFF_TAP) {
1791 name = ifr->ifr_name;
1793 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1794 NET_NAME_UNKNOWN, tun_setup, queues,
1800 dev_net_set(dev, net);
1801 dev->rtnl_link_ops = &tun_link_ops;
1802 dev->ifindex = tfile->ifindex;
1803 dev->sysfs_groups[0] = &tun_attr_group;
1805 tun = netdev_priv(dev);
1808 tun->txflt.count = 0;
1809 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1811 tun->align = NET_SKB_PAD;
1812 tun->filter_attached = false;
1813 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1814 tun->rx_batched = 0;
1816 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
1817 if (!tun->pcpu_stats) {
1822 spin_lock_init(&tun->lock);
1824 err = security_tun_dev_alloc_security(&tun->security);
1831 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1832 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
1833 NETIF_F_HW_VLAN_STAG_TX;
1834 dev->features = dev->hw_features | NETIF_F_LLTX;
1835 dev->vlan_features = dev->features &
1836 ~(NETIF_F_HW_VLAN_CTAG_TX |
1837 NETIF_F_HW_VLAN_STAG_TX);
1839 INIT_LIST_HEAD(&tun->disabled);
1840 err = tun_attach(tun, file, false);
1844 err = register_netdevice(tun->dev);
1849 netif_carrier_on(tun->dev);
1851 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1853 tun->flags = (tun->flags & ~TUN_FEATURES) |
1854 (ifr->ifr_flags & TUN_FEATURES);
1856 /* Make sure persistent devices do not get stuck in
1859 if (netif_running(tun->dev))
1860 netif_tx_wake_all_queues(tun->dev);
1862 strcpy(ifr->ifr_name, tun->dev->name);
1866 tun_detach_all(dev);
1868 tun_flow_uninit(tun);
1869 security_tun_dev_free_security(tun->security);
1871 free_percpu(tun->pcpu_stats);
1877 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1880 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1882 strcpy(ifr->ifr_name, tun->dev->name);
1884 ifr->ifr_flags = tun_flags(tun);
1888 /* This is like a cut-down ethtool ops, except done via tun fd so no
1889 * privs required. */
1890 static int set_offload(struct tun_struct *tun, unsigned long arg)
1892 netdev_features_t features = 0;
1894 if (arg & TUN_F_CSUM) {
1895 features |= NETIF_F_HW_CSUM;
1898 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1899 if (arg & TUN_F_TSO_ECN) {
1900 features |= NETIF_F_TSO_ECN;
1901 arg &= ~TUN_F_TSO_ECN;
1903 if (arg & TUN_F_TSO4)
1904 features |= NETIF_F_TSO;
1905 if (arg & TUN_F_TSO6)
1906 features |= NETIF_F_TSO6;
1907 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1910 if (arg & TUN_F_UFO) {
1911 features |= NETIF_F_UFO;
1916 /* This gives the user a way to test for new features in future by
1917 * trying to set them. */
1921 tun->set_features = features;
1922 netdev_update_features(tun->dev);
1927 static void tun_detach_filter(struct tun_struct *tun, int n)
1930 struct tun_file *tfile;
1932 for (i = 0; i < n; i++) {
1933 tfile = rtnl_dereference(tun->tfiles[i]);
1934 lock_sock(tfile->socket.sk);
1935 sk_detach_filter(tfile->socket.sk);
1936 release_sock(tfile->socket.sk);
1939 tun->filter_attached = false;
1942 static int tun_attach_filter(struct tun_struct *tun)
1945 struct tun_file *tfile;
1947 for (i = 0; i < tun->numqueues; i++) {
1948 tfile = rtnl_dereference(tun->tfiles[i]);
1949 lock_sock(tfile->socket.sk);
1950 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1951 release_sock(tfile->socket.sk);
1953 tun_detach_filter(tun, i);
1958 tun->filter_attached = true;
1962 static void tun_set_sndbuf(struct tun_struct *tun)
1964 struct tun_file *tfile;
1967 for (i = 0; i < tun->numqueues; i++) {
1968 tfile = rtnl_dereference(tun->tfiles[i]);
1969 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1973 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1975 struct tun_file *tfile = file->private_data;
1976 struct tun_struct *tun;
1981 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1982 tun = tfile->detached;
1987 ret = security_tun_dev_attach_queue(tun->security);
1990 ret = tun_attach(tun, file, false);
1991 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1992 tun = rtnl_dereference(tfile->tun);
1993 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
1996 __tun_detach(tfile, false);
2005 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2006 unsigned long arg, int ifreq_len)
2008 struct tun_file *tfile = file->private_data;
2009 struct tun_struct *tun;
2010 void __user* argp = (void __user*)arg;
2016 unsigned int ifindex;
2020 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == SOCK_IOC_TYPE) {
2021 if (copy_from_user(&ifr, argp, ifreq_len))
2024 memset(&ifr, 0, sizeof(ifr));
2026 if (cmd == TUNGETFEATURES) {
2027 /* Currently this just means: "what IFF flags are valid?".
2028 * This is needed because we never checked for invalid flags on
2031 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2032 (unsigned int __user*)argp);
2033 } else if (cmd == TUNSETQUEUE)
2034 return tun_set_queue(file, &ifr);
2039 tun = __tun_get(tfile);
2040 if (cmd == TUNSETIFF) {
2045 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2047 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
2052 if (copy_to_user(argp, &ifr, ifreq_len))
2056 if (cmd == TUNSETIFINDEX) {
2062 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2066 tfile->ifindex = ifindex;
2074 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2079 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2081 if (tfile->detached)
2082 ifr.ifr_flags |= IFF_DETACH_QUEUE;
2083 if (!tfile->socket.sk->sk_filter)
2084 ifr.ifr_flags |= IFF_NOFILTER;
2086 if (copy_to_user(argp, &ifr, ifreq_len))
2091 /* Disable/Enable checksum */
2093 /* [unimplemented] */
2094 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2095 arg ? "disabled" : "enabled");
2099 /* Disable/Enable persist mode. Keep an extra reference to the
2100 * module to prevent the module being unprobed.
2102 if (arg && !(tun->flags & IFF_PERSIST)) {
2103 tun->flags |= IFF_PERSIST;
2104 __module_get(THIS_MODULE);
2106 if (!arg && (tun->flags & IFF_PERSIST)) {
2107 tun->flags &= ~IFF_PERSIST;
2108 module_put(THIS_MODULE);
2111 tun_debug(KERN_INFO, tun, "persist %s\n",
2112 arg ? "enabled" : "disabled");
2116 /* Set owner of the device */
2117 owner = make_kuid(current_user_ns(), arg);
2118 if (!uid_valid(owner)) {
2123 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2124 from_kuid(&init_user_ns, tun->owner));
2128 /* Set group of the device */
2129 group = make_kgid(current_user_ns(), arg);
2130 if (!gid_valid(group)) {
2135 tun_debug(KERN_INFO, tun, "group set to %u\n",
2136 from_kgid(&init_user_ns, tun->group));
2140 /* Only allow setting the type when the interface is down */
2141 if (tun->dev->flags & IFF_UP) {
2142 tun_debug(KERN_INFO, tun,
2143 "Linktype set failed because interface is up\n");
2146 tun->dev->type = (int) arg;
2147 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2159 ret = set_offload(tun, arg);
2162 case TUNSETTXFILTER:
2163 /* Can be set only for TAPs */
2165 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2167 ret = update_filter(&tun->txflt, (void __user *)arg);
2171 /* Get hw address */
2172 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2173 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2174 if (copy_to_user(argp, &ifr, ifreq_len))
2179 /* Set hw address */
2180 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2181 ifr.ifr_hwaddr.sa_data);
2183 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2187 sndbuf = tfile->socket.sk->sk_sndbuf;
2188 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2193 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2198 tun->sndbuf = sndbuf;
2199 tun_set_sndbuf(tun);
2202 case TUNGETVNETHDRSZ:
2203 vnet_hdr_sz = tun->vnet_hdr_sz;
2204 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2208 case TUNSETVNETHDRSZ:
2209 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2213 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2218 tun->vnet_hdr_sz = vnet_hdr_sz;
2222 le = !!(tun->flags & TUN_VNET_LE);
2223 if (put_user(le, (int __user *)argp))
2228 if (get_user(le, (int __user *)argp)) {
2233 tun->flags |= TUN_VNET_LE;
2235 tun->flags &= ~TUN_VNET_LE;
2239 ret = tun_get_vnet_be(tun, argp);
2243 ret = tun_set_vnet_be(tun, argp);
2246 case TUNATTACHFILTER:
2247 /* Can be set only for TAPs */
2249 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2252 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2255 ret = tun_attach_filter(tun);
2258 case TUNDETACHFILTER:
2259 /* Can be set only for TAPs */
2261 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2264 tun_detach_filter(tun, tun->numqueues);
2269 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2272 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2289 static long tun_chr_ioctl(struct file *file,
2290 unsigned int cmd, unsigned long arg)
2292 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2295 #ifdef CONFIG_COMPAT
2296 static long tun_chr_compat_ioctl(struct file *file,
2297 unsigned int cmd, unsigned long arg)
2302 case TUNSETTXFILTER:
2307 arg = (unsigned long)compat_ptr(arg);
2310 arg = (compat_ulong_t)arg;
2315 * compat_ifreq is shorter than ifreq, so we must not access beyond
2316 * the end of that structure. All fields that are used in this
2317 * driver are compatible though, we don't need to convert the
2320 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2322 #endif /* CONFIG_COMPAT */
2324 static int tun_chr_fasync(int fd, struct file *file, int on)
2326 struct tun_file *tfile = file->private_data;
2329 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2333 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2334 tfile->flags |= TUN_FASYNC;
2336 tfile->flags &= ~TUN_FASYNC;
2342 static int tun_chr_open(struct inode *inode, struct file * file)
2344 struct net *net = current->nsproxy->net_ns;
2345 struct tun_file *tfile;
2347 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2349 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2353 RCU_INIT_POINTER(tfile->tun, NULL);
2357 init_waitqueue_head(&tfile->wq.wait);
2358 RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2360 tfile->socket.file = file;
2361 tfile->socket.ops = &tun_socket_ops;
2363 sock_init_data(&tfile->socket, &tfile->sk);
2365 tfile->sk.sk_write_space = tun_sock_write_space;
2366 tfile->sk.sk_sndbuf = INT_MAX;
2368 file->private_data = tfile;
2369 INIT_LIST_HEAD(&tfile->next);
2371 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2376 static int tun_chr_close(struct inode *inode, struct file *file)
2378 struct tun_file *tfile = file->private_data;
2380 tun_detach(tfile, true);
2385 #ifdef CONFIG_PROC_FS
2386 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2388 struct tun_struct *tun;
2391 memset(&ifr, 0, sizeof(ifr));
2396 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2402 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2406 static const struct file_operations tun_fops = {
2407 .owner = THIS_MODULE,
2408 .llseek = no_llseek,
2409 .read_iter = tun_chr_read_iter,
2410 .write_iter = tun_chr_write_iter,
2411 .poll = tun_chr_poll,
2412 .unlocked_ioctl = tun_chr_ioctl,
2413 #ifdef CONFIG_COMPAT
2414 .compat_ioctl = tun_chr_compat_ioctl,
2416 .open = tun_chr_open,
2417 .release = tun_chr_close,
2418 .fasync = tun_chr_fasync,
2419 #ifdef CONFIG_PROC_FS
2420 .show_fdinfo = tun_chr_show_fdinfo,
2424 static struct miscdevice tun_miscdev = {
2427 .nodename = "net/tun",
2431 /* ethtool interface */
2433 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2436 cmd->advertising = 0;
2437 ethtool_cmd_speed_set(cmd, SPEED_10);
2438 cmd->duplex = DUPLEX_FULL;
2439 cmd->port = PORT_TP;
2440 cmd->phy_address = 0;
2441 cmd->transceiver = XCVR_INTERNAL;
2442 cmd->autoneg = AUTONEG_DISABLE;
2448 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2450 struct tun_struct *tun = netdev_priv(dev);
2452 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2453 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2455 switch (tun->flags & TUN_TYPE_MASK) {
2457 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2460 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2465 static u32 tun_get_msglevel(struct net_device *dev)
2468 struct tun_struct *tun = netdev_priv(dev);
2475 static void tun_set_msglevel(struct net_device *dev, u32 value)
2478 struct tun_struct *tun = netdev_priv(dev);
2483 static int tun_get_coalesce(struct net_device *dev,
2484 struct ethtool_coalesce *ec)
2486 struct tun_struct *tun = netdev_priv(dev);
2488 ec->rx_max_coalesced_frames = tun->rx_batched;
2493 static int tun_set_coalesce(struct net_device *dev,
2494 struct ethtool_coalesce *ec)
2496 struct tun_struct *tun = netdev_priv(dev);
2498 if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
2499 tun->rx_batched = NAPI_POLL_WEIGHT;
2501 tun->rx_batched = ec->rx_max_coalesced_frames;
2506 static const struct ethtool_ops tun_ethtool_ops = {
2507 .get_settings = tun_get_settings,
2508 .get_drvinfo = tun_get_drvinfo,
2509 .get_msglevel = tun_get_msglevel,
2510 .set_msglevel = tun_set_msglevel,
2511 .get_link = ethtool_op_get_link,
2512 .get_ts_info = ethtool_op_get_ts_info,
2513 .get_coalesce = tun_get_coalesce,
2514 .set_coalesce = tun_set_coalesce,
2517 static int tun_queue_resize(struct tun_struct *tun)
2519 struct net_device *dev = tun->dev;
2520 struct tun_file *tfile;
2521 struct skb_array **arrays;
2522 int n = tun->numqueues + tun->numdisabled;
2525 arrays = kmalloc(sizeof *arrays * n, GFP_KERNEL);
2529 for (i = 0; i < tun->numqueues; i++) {
2530 tfile = rtnl_dereference(tun->tfiles[i]);
2531 arrays[i] = &tfile->tx_array;
2533 list_for_each_entry(tfile, &tun->disabled, next)
2534 arrays[i++] = &tfile->tx_array;
2536 ret = skb_array_resize_multiple(arrays, n,
2537 dev->tx_queue_len, GFP_KERNEL);
2543 static int tun_device_event(struct notifier_block *unused,
2544 unsigned long event, void *ptr)
2546 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2547 struct tun_struct *tun = netdev_priv(dev);
2549 if (dev->rtnl_link_ops != &tun_link_ops)
2553 case NETDEV_CHANGE_TX_QUEUE_LEN:
2554 if (tun_queue_resize(tun))
2564 static struct notifier_block tun_notifier_block __read_mostly = {
2565 .notifier_call = tun_device_event,
2568 static int __init tun_init(void)
2572 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2573 pr_info("%s\n", DRV_COPYRIGHT);
2575 ret = rtnl_link_register(&tun_link_ops);
2577 pr_err("Can't register link_ops\n");
2581 ret = misc_register(&tun_miscdev);
2583 pr_err("Can't register misc device %d\n", TUN_MINOR);
2587 register_netdevice_notifier(&tun_notifier_block);
2590 rtnl_link_unregister(&tun_link_ops);
2595 static void tun_cleanup(void)
2597 misc_deregister(&tun_miscdev);
2598 rtnl_link_unregister(&tun_link_ops);
2599 unregister_netdevice_notifier(&tun_notifier_block);
2602 /* Get an underlying socket object from tun file. Returns error unless file is
2603 * attached to a device. The returned object works like a packet socket, it
2604 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
2605 * holding a reference to the file for as long as the socket is in use. */
2606 struct socket *tun_get_socket(struct file *file)
2608 struct tun_file *tfile;
2609 if (file->f_op != &tun_fops)
2610 return ERR_PTR(-EINVAL);
2611 tfile = file->private_data;
2613 return ERR_PTR(-EBADFD);
2614 return &tfile->socket;
2616 EXPORT_SYMBOL_GPL(tun_get_socket);
2618 module_init(tun_init);
2619 module_exit(tun_cleanup);
2620 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2621 MODULE_AUTHOR(DRV_COPYRIGHT);
2622 MODULE_LICENSE("GPL");
2623 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2624 MODULE_ALIAS("devname:net/tun");