2 * TUN - Universal TUN/TAP device driver.
3 * Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
21 * Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
22 * Add TUNSETLINK ioctl to set the link encapsulation
24 * Mark Smith <markzzzsmith@yahoo.com.au>
25 * Use eth_random_addr() for tap MAC address.
27 * Harald Roelle <harald.roelle@ifi.lmu.de> 2004/04/20
28 * Fixes in packet dropping, queue length setting and queue wakeup.
29 * Increased default tx queue length.
33 * Daniel Podlejski <underley@underley.eu.org>
34 * Modifications for 2.3.99-pre5 kernel.
37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
39 #define DRV_NAME "tun"
40 #define DRV_VERSION "1.6"
41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
42 #define DRV_COPYRIGHT "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
44 #include <linux/module.h>
45 #include <linux/errno.h>
46 #include <linux/kernel.h>
47 #include <linux/major.h>
48 #include <linux/slab.h>
49 #include <linux/poll.h>
50 #include <linux/fcntl.h>
51 #include <linux/init.h>
52 #include <linux/skbuff.h>
53 #include <linux/netdevice.h>
54 #include <linux/etherdevice.h>
55 #include <linux/miscdevice.h>
56 #include <linux/ethtool.h>
57 #include <linux/rtnetlink.h>
58 #include <linux/compat.h>
60 #include <linux/if_arp.h>
61 #include <linux/if_ether.h>
62 #include <linux/if_tun.h>
63 #include <linux/crc32.h>
64 #include <linux/nsproxy.h>
65 #include <linux/virtio_net.h>
66 #include <linux/rcupdate.h>
67 #include <net/net_namespace.h>
68 #include <net/netns/generic.h>
69 #include <net/rtnetlink.h>
72 #include <asm/uaccess.h>
74 /* Uncomment to enable debugging */
75 /* #define TUN_DEBUG 1 */
80 #define tun_debug(level, tun, fmt, args...) \
83 netdev_printk(level, tun->dev, fmt, ##args); \
85 #define DBG1(level, fmt, args...) \
88 printk(level fmt, ##args); \
91 #define tun_debug(level, tun, fmt, args...) \
94 netdev_printk(level, tun->dev, fmt, ##args); \
96 #define DBG1(level, fmt, args...) \
99 printk(level fmt, ##args); \
103 #define GOODCOPY_LEN 128
105 #define FLT_EXACT_COUNT 8
107 unsigned int count; /* Number of addrs. Zero means disabled */
108 u32 mask[2]; /* Mask of the hashed addrs */
109 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
112 /* 1024 is probably a high enough limit: modern hypervisors seem to support on
113 * the order of 100-200 CPUs so this leaves us some breathing space if we want
114 * to match a queue per guest CPU.
116 #define MAX_TAP_QUEUES 1024
118 #define TUN_FLOW_EXPIRE (3 * HZ)
120 /* A tun_file connects an open character device to a tuntap netdevice. It
121 * also contains all socket related strctures (except sock_fprog and tap_filter)
122 * to serve as one transmit queue for tuntap device. The sock_fprog and
123 * tap_filter were kept in tun_struct since they were used for filtering for the
124 * netdevice not for a specific queue (at least I didn't see the requirement for
128 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
129 * other can only be read while rcu_read_lock or rtnl_lock is held.
133 struct socket socket;
135 struct tun_struct __rcu *tun;
137 struct fasync_struct *fasync;
138 /* only used for fasnyc */
141 struct list_head next;
142 struct tun_struct *detached;
145 struct tun_flow_entry {
146 struct hlist_node hash_link;
148 struct tun_struct *tun;
152 unsigned long updated;
155 #define TUN_NUM_FLOW_ENTRIES 1024
157 /* Since the socket were moved to tun_file, to preserve the behavior of persist
158 * device, socket filter, sndbuf and vnet header size were restore when the
159 * file were attached to a persist device.
162 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
163 unsigned int numqueues;
168 struct net_device *dev;
169 netdev_features_t set_features;
170 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
171 NETIF_F_TSO6|NETIF_F_UFO)
175 struct tap_filter txflt;
176 struct sock_fprog fprog;
177 /* protected by rtnl lock */
178 bool filter_attached;
183 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
184 struct timer_list flow_gc_timer;
185 unsigned long ageing_time;
186 unsigned int numdisabled;
187 struct list_head disabled;
191 static inline u32 tun_hashfn(u32 rxhash)
193 return rxhash & 0x3ff;
196 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
198 struct tun_flow_entry *e;
199 struct hlist_node *n;
201 hlist_for_each_entry_rcu(e, n, head, hash_link) {
202 if (e->rxhash == rxhash)
208 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
209 struct hlist_head *head,
210 u32 rxhash, u16 queue_index)
212 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
215 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
216 rxhash, queue_index);
217 e->updated = jiffies;
219 e->queue_index = queue_index;
221 hlist_add_head_rcu(&e->hash_link, head);
226 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
228 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
229 e->rxhash, e->queue_index);
230 hlist_del_rcu(&e->hash_link);
234 static void tun_flow_flush(struct tun_struct *tun)
238 spin_lock_bh(&tun->lock);
239 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
240 struct tun_flow_entry *e;
241 struct hlist_node *h, *n;
243 hlist_for_each_entry_safe(e, h, n, &tun->flows[i], hash_link)
244 tun_flow_delete(tun, e);
246 spin_unlock_bh(&tun->lock);
249 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
253 spin_lock_bh(&tun->lock);
254 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
255 struct tun_flow_entry *e;
256 struct hlist_node *h, *n;
258 hlist_for_each_entry_safe(e, h, n, &tun->flows[i], hash_link) {
259 if (e->queue_index == queue_index)
260 tun_flow_delete(tun, e);
263 spin_unlock_bh(&tun->lock);
266 static void tun_flow_cleanup(unsigned long data)
268 struct tun_struct *tun = (struct tun_struct *)data;
269 unsigned long delay = tun->ageing_time;
270 unsigned long next_timer = jiffies + delay;
271 unsigned long count = 0;
274 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
276 spin_lock_bh(&tun->lock);
277 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
278 struct tun_flow_entry *e;
279 struct hlist_node *h, *n;
281 hlist_for_each_entry_safe(e, h, n, &tun->flows[i], hash_link) {
282 unsigned long this_timer;
284 this_timer = e->updated + delay;
285 if (time_before_eq(this_timer, jiffies))
286 tun_flow_delete(tun, e);
287 else if (time_before(this_timer, next_timer))
288 next_timer = this_timer;
293 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
294 spin_unlock_bh(&tun->lock);
297 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
300 struct hlist_head *head;
301 struct tun_flow_entry *e;
302 unsigned long delay = tun->ageing_time;
307 head = &tun->flows[tun_hashfn(rxhash)];
311 if (tun->numqueues == 1)
314 e = tun_flow_find(head, rxhash);
316 /* TODO: keep queueing to old queue until it's empty? */
317 e->queue_index = queue_index;
318 e->updated = jiffies;
320 spin_lock_bh(&tun->lock);
321 if (!tun_flow_find(head, rxhash))
322 tun_flow_create(tun, head, rxhash, queue_index);
324 if (!timer_pending(&tun->flow_gc_timer))
325 mod_timer(&tun->flow_gc_timer,
326 round_jiffies_up(jiffies + delay));
327 spin_unlock_bh(&tun->lock);
334 /* We try to identify a flow through its rxhash first. The reason that
335 * we do not check rxq no. is becuase some cards(e.g 82599), chooses
336 * the rxq based on the txq where the last packet of the flow comes. As
337 * the userspace application move between processors, we may get a
338 * different rxq no. here. If we could not get rxhash, then we would
339 * hope the rxq no. may help here.
341 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb)
343 struct tun_struct *tun = netdev_priv(dev);
344 struct tun_flow_entry *e;
349 numqueues = tun->numqueues;
351 txq = skb_get_rxhash(skb);
353 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
355 txq = e->queue_index;
357 /* use multiply and shift instead of expensive divide */
358 txq = ((u64)txq * numqueues) >> 32;
359 } else if (likely(skb_rx_queue_recorded(skb))) {
360 txq = skb_get_rx_queue(skb);
361 while (unlikely(txq >= numqueues))
369 static inline bool tun_not_capable(struct tun_struct *tun)
371 const struct cred *cred = current_cred();
372 struct net *net = dev_net(tun->dev);
374 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
375 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
376 !ns_capable(net->user_ns, CAP_NET_ADMIN);
379 static void tun_set_real_num_queues(struct tun_struct *tun)
381 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
382 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
385 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
387 tfile->detached = tun;
388 list_add_tail(&tfile->next, &tun->disabled);
392 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
394 struct tun_struct *tun = tfile->detached;
396 tfile->detached = NULL;
397 list_del_init(&tfile->next);
402 static void __tun_detach(struct tun_file *tfile, bool clean)
404 struct tun_file *ntfile;
405 struct tun_struct *tun;
406 struct net_device *dev;
408 tun = rtnl_dereference(tfile->tun);
411 u16 index = tfile->queue_index;
412 BUG_ON(index >= tun->numqueues);
415 rcu_assign_pointer(tun->tfiles[index],
416 tun->tfiles[tun->numqueues - 1]);
417 rcu_assign_pointer(tfile->tun, NULL);
418 ntfile = rtnl_dereference(tun->tfiles[index]);
419 ntfile->queue_index = index;
423 sock_put(&tfile->sk);
425 tun_disable_queue(tun, tfile);
428 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
429 /* Drop read queue */
430 skb_queue_purge(&tfile->sk.sk_receive_queue);
431 tun_set_real_num_queues(tun);
432 } else if (tfile->detached && clean) {
433 tun = tun_enable_queue(tfile);
434 sock_put(&tfile->sk);
438 if (tun && tun->numqueues == 0 && tun->numdisabled == 0 &&
439 !(tun->flags & TUN_PERSIST))
440 if (tun->dev->reg_state == NETREG_REGISTERED)
441 unregister_netdevice(tun->dev);
443 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED,
444 &tfile->socket.flags));
445 sk_release_kernel(&tfile->sk);
449 static void tun_detach(struct tun_file *tfile, bool clean)
452 __tun_detach(tfile, clean);
456 static void tun_detach_all(struct net_device *dev)
458 struct tun_struct *tun = netdev_priv(dev);
459 struct tun_file *tfile, *tmp;
460 int i, n = tun->numqueues;
462 for (i = 0; i < n; i++) {
463 tfile = rtnl_dereference(tun->tfiles[i]);
465 wake_up_all(&tfile->wq.wait);
466 rcu_assign_pointer(tfile->tun, NULL);
469 BUG_ON(tun->numqueues != 0);
472 for (i = 0; i < n; i++) {
473 tfile = rtnl_dereference(tun->tfiles[i]);
474 /* Drop read queue */
475 skb_queue_purge(&tfile->sk.sk_receive_queue);
476 sock_put(&tfile->sk);
478 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
479 tun_enable_queue(tfile);
480 skb_queue_purge(&tfile->sk.sk_receive_queue);
481 sock_put(&tfile->sk);
483 BUG_ON(tun->numdisabled != 0);
485 if (tun->flags & TUN_PERSIST)
486 module_put(THIS_MODULE);
489 static int tun_attach(struct tun_struct *tun, struct file *file)
491 struct tun_file *tfile = file->private_data;
494 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
499 if (rtnl_dereference(tfile->tun))
503 if (!(tun->flags & TUN_TAP_MQ) && tun->numqueues == 1)
507 if (!tfile->detached &&
508 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
513 /* Re-attach the filter to presist device */
514 if (tun->filter_attached == true) {
515 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
519 tfile->queue_index = tun->numqueues;
520 rcu_assign_pointer(tfile->tun, tun);
521 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
525 tun_enable_queue(tfile);
527 sock_hold(&tfile->sk);
529 tun_set_real_num_queues(tun);
531 /* device is allowed to go away first, so no need to hold extra
539 static struct tun_struct *__tun_get(struct tun_file *tfile)
541 struct tun_struct *tun;
544 tun = rcu_dereference(tfile->tun);
552 static struct tun_struct *tun_get(struct file *file)
554 return __tun_get(file->private_data);
557 static void tun_put(struct tun_struct *tun)
563 static void addr_hash_set(u32 *mask, const u8 *addr)
565 int n = ether_crc(ETH_ALEN, addr) >> 26;
566 mask[n >> 5] |= (1 << (n & 31));
569 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
571 int n = ether_crc(ETH_ALEN, addr) >> 26;
572 return mask[n >> 5] & (1 << (n & 31));
575 static int update_filter(struct tap_filter *filter, void __user *arg)
577 struct { u8 u[ETH_ALEN]; } *addr;
578 struct tun_filter uf;
579 int err, alen, n, nexact;
581 if (copy_from_user(&uf, arg, sizeof(uf)))
590 alen = ETH_ALEN * uf.count;
591 addr = kmalloc(alen, GFP_KERNEL);
595 if (copy_from_user(addr, arg + sizeof(uf), alen)) {
600 /* The filter is updated without holding any locks. Which is
601 * perfectly safe. We disable it first and in the worst
602 * case we'll accept a few undesired packets. */
606 /* Use first set of addresses as an exact filter */
607 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
608 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
612 /* Remaining multicast addresses are hashed,
613 * unicast will leave the filter disabled. */
614 memset(filter->mask, 0, sizeof(filter->mask));
615 for (; n < uf.count; n++) {
616 if (!is_multicast_ether_addr(addr[n].u)) {
617 err = 0; /* no filter */
620 addr_hash_set(filter->mask, addr[n].u);
623 /* For ALLMULTI just set the mask to all ones.
624 * This overrides the mask populated above. */
625 if ((uf.flags & TUN_FLT_ALLMULTI))
626 memset(filter->mask, ~0, sizeof(filter->mask));
628 /* Now enable the filter */
630 filter->count = nexact;
632 /* Return the number of exact filters */
640 /* Returns: 0 - drop, !=0 - accept */
641 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
643 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
645 struct ethhdr *eh = (struct ethhdr *) skb->data;
649 for (i = 0; i < filter->count; i++)
650 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
653 /* Inexact match (multicast only) */
654 if (is_multicast_ether_addr(eh->h_dest))
655 return addr_hash_test(filter->mask, eh->h_dest);
661 * Checks whether the packet is accepted or not.
662 * Returns: 0 - drop, !=0 - accept
664 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
669 return run_filter(filter, skb);
672 /* Network device part of the driver */
674 static const struct ethtool_ops tun_ethtool_ops;
676 /* Net device detach from fd. */
677 static void tun_net_uninit(struct net_device *dev)
682 /* Net device open. */
683 static int tun_net_open(struct net_device *dev)
685 netif_tx_start_all_queues(dev);
689 /* Net device close. */
690 static int tun_net_close(struct net_device *dev)
692 netif_tx_stop_all_queues(dev);
696 /* Net device start xmit */
697 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
699 struct tun_struct *tun = netdev_priv(dev);
700 int txq = skb->queue_mapping;
701 struct tun_file *tfile;
704 tfile = rcu_dereference(tun->tfiles[txq]);
706 /* Drop packet if interface is not attached */
707 if (txq >= tun->numqueues)
710 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
714 /* Drop if the filter does not like it.
715 * This is a noop if the filter is disabled.
716 * Filter can be enabled only for the TAP devices. */
717 if (!check_filter(&tun->txflt, skb))
720 if (tfile->socket.sk->sk_filter &&
721 sk_filter(tfile->socket.sk, skb))
724 /* Limit the number of packets queued by dividing txq length with the
727 if (skb_queue_len(&tfile->socket.sk->sk_receive_queue)
728 >= dev->tx_queue_len / tun->numqueues)
731 /* Orphan the skb - required as we might hang on to it
732 * for indefinite time. */
733 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
738 skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
740 /* Notify and wake up reader process */
741 if (tfile->flags & TUN_FASYNC)
742 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
743 wake_up_interruptible_poll(&tfile->wq.wait, POLLIN |
744 POLLRDNORM | POLLRDBAND);
750 dev->stats.tx_dropped++;
757 static void tun_net_mclist(struct net_device *dev)
760 * This callback is supposed to deal with mc filter in
761 * _rx_ path and has nothing to do with the _tx_ path.
762 * In rx path we always accept everything userspace gives us.
767 #define MAX_MTU 65535
770 tun_net_change_mtu(struct net_device *dev, int new_mtu)
772 if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
778 static netdev_features_t tun_net_fix_features(struct net_device *dev,
779 netdev_features_t features)
781 struct tun_struct *tun = netdev_priv(dev);
783 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
785 #ifdef CONFIG_NET_POLL_CONTROLLER
786 static void tun_poll_controller(struct net_device *dev)
789 * Tun only receives frames when:
790 * 1) the char device endpoint gets data from user space
791 * 2) the tun socket gets a sendmsg call from user space
792 * Since both of those are syncronous operations, we are guaranteed
793 * never to have pending data when we poll for it
794 * so theres nothing to do here but return.
795 * We need this though so netpoll recognizes us as an interface that
796 * supports polling, which enables bridge devices in virt setups to
797 * still use netconsole
802 static const struct net_device_ops tun_netdev_ops = {
803 .ndo_uninit = tun_net_uninit,
804 .ndo_open = tun_net_open,
805 .ndo_stop = tun_net_close,
806 .ndo_start_xmit = tun_net_xmit,
807 .ndo_change_mtu = tun_net_change_mtu,
808 .ndo_fix_features = tun_net_fix_features,
809 .ndo_select_queue = tun_select_queue,
810 #ifdef CONFIG_NET_POLL_CONTROLLER
811 .ndo_poll_controller = tun_poll_controller,
815 static const struct net_device_ops tap_netdev_ops = {
816 .ndo_uninit = tun_net_uninit,
817 .ndo_open = tun_net_open,
818 .ndo_stop = tun_net_close,
819 .ndo_start_xmit = tun_net_xmit,
820 .ndo_change_mtu = tun_net_change_mtu,
821 .ndo_fix_features = tun_net_fix_features,
822 .ndo_set_rx_mode = tun_net_mclist,
823 .ndo_set_mac_address = eth_mac_addr,
824 .ndo_validate_addr = eth_validate_addr,
825 .ndo_select_queue = tun_select_queue,
826 #ifdef CONFIG_NET_POLL_CONTROLLER
827 .ndo_poll_controller = tun_poll_controller,
831 static int tun_flow_init(struct tun_struct *tun)
835 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
836 INIT_HLIST_HEAD(&tun->flows[i]);
838 tun->ageing_time = TUN_FLOW_EXPIRE;
839 setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
840 mod_timer(&tun->flow_gc_timer,
841 round_jiffies_up(jiffies + tun->ageing_time));
846 static void tun_flow_uninit(struct tun_struct *tun)
848 del_timer_sync(&tun->flow_gc_timer);
852 /* Initialize net device. */
853 static void tun_net_init(struct net_device *dev)
855 struct tun_struct *tun = netdev_priv(dev);
857 switch (tun->flags & TUN_TYPE_MASK) {
859 dev->netdev_ops = &tun_netdev_ops;
861 /* Point-to-Point TUN Device */
862 dev->hard_header_len = 0;
866 /* Zero header length */
867 dev->type = ARPHRD_NONE;
868 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
869 dev->tx_queue_len = TUN_READQ_SIZE; /* We prefer our own queue length */
873 dev->netdev_ops = &tap_netdev_ops;
874 /* Ethernet TAP Device */
876 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
877 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
879 eth_hw_addr_random(dev);
881 dev->tx_queue_len = TUN_READQ_SIZE; /* We prefer our own queue length */
886 /* Character device part */
889 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
891 struct tun_file *tfile = file->private_data;
892 struct tun_struct *tun = __tun_get(tfile);
894 unsigned int mask = 0;
899 sk = tfile->socket.sk;
901 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
903 poll_wait(file, &tfile->wq.wait, wait);
905 if (!skb_queue_empty(&sk->sk_receive_queue))
906 mask |= POLLIN | POLLRDNORM;
908 if (sock_writeable(sk) ||
909 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
911 mask |= POLLOUT | POLLWRNORM;
913 if (tun->dev->reg_state != NETREG_REGISTERED)
920 /* prepad is the amount to reserve at front. len is length after that.
921 * linear is a hint as to how much to copy (usually headers). */
922 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
923 size_t prepad, size_t len,
924 size_t linear, int noblock)
926 struct sock *sk = tfile->socket.sk;
930 /* Under a page? Don't bother with paged skb. */
931 if (prepad + len < PAGE_SIZE || !linear)
934 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
939 skb_reserve(skb, prepad);
940 skb_put(skb, linear);
941 skb->data_len = len - linear;
942 skb->len += len - linear;
947 /* set skb frags from iovec, this can move to core network code for reuse */
948 static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
949 int offset, size_t count)
951 int len = iov_length(from, count) - offset;
952 int copy = skb_headlen(skb);
953 int size, offset1 = 0;
956 /* Skip over from offset */
957 while (count && (offset >= from->iov_len)) {
958 offset -= from->iov_len;
963 /* copy up to skb headlen */
964 while (count && (copy > 0)) {
965 size = min_t(unsigned int, copy, from->iov_len - offset);
966 if (copy_from_user(skb->data + offset1, from->iov_base + offset,
983 struct page *page[MAX_SKB_FRAGS];
986 unsigned long truesize;
988 len = from->iov_len - offset;
994 base = (unsigned long)from->iov_base + offset;
995 size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
996 if (i + size > MAX_SKB_FRAGS)
998 num_pages = get_user_pages_fast(base, size, 0, &page[i]);
999 if (num_pages != size) {
1000 for (i = 0; i < num_pages; i++)
1004 truesize = size * PAGE_SIZE;
1005 skb->data_len += len;
1007 skb->truesize += truesize;
1008 atomic_add(truesize, &skb->sk->sk_wmem_alloc);
1010 int off = base & ~PAGE_MASK;
1011 int size = min_t(int, len, PAGE_SIZE - off);
1012 __skb_fill_page_desc(skb, i, page[i], off, size);
1013 skb_shinfo(skb)->nr_frags++;
1014 /* increase sk_wmem_alloc */
1025 /* Get packet from user space buffer */
1026 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1027 void *msg_control, const struct iovec *iv,
1028 size_t total_len, size_t count, int noblock)
1030 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1031 struct sk_buff *skb;
1032 size_t len = total_len, align = NET_SKB_PAD;
1033 struct virtio_net_hdr gso = { 0 };
1036 bool zerocopy = false;
1040 if (!(tun->flags & TUN_NO_PI)) {
1041 if ((len -= sizeof(pi)) > total_len)
1044 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi)))
1046 offset += sizeof(pi);
1049 if (tun->flags & TUN_VNET_HDR) {
1050 if ((len -= tun->vnet_hdr_sz) > total_len)
1053 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso)))
1056 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1057 gso.csum_start + gso.csum_offset + 2 > gso.hdr_len)
1058 gso.hdr_len = gso.csum_start + gso.csum_offset + 2;
1060 if (gso.hdr_len > len)
1062 offset += tun->vnet_hdr_sz;
1065 if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
1066 align += NET_IP_ALIGN;
1067 if (unlikely(len < ETH_HLEN ||
1068 (gso.hdr_len && gso.hdr_len < ETH_HLEN)))
1076 /* Userspace may produce vectors with count greater than
1077 * MAX_SKB_FRAGS, so we need to linearize parts of the skb
1078 * to let the rest of data to be fit in the frags.
1080 if (count > MAX_SKB_FRAGS) {
1081 copylen = iov_length(iv, count - MAX_SKB_FRAGS);
1082 if (copylen < offset)
1088 /* There are 256 bytes to be copied in skb, so there is enough
1089 * room for skb expand head in case it is used.
1090 * The rest of the buffer is mapped from userspace.
1092 if (copylen < gso.hdr_len)
1093 copylen = gso.hdr_len;
1095 copylen = GOODCOPY_LEN;
1099 skb = tun_alloc_skb(tfile, align, copylen, gso.hdr_len, noblock);
1101 if (PTR_ERR(skb) != -EAGAIN)
1102 tun->dev->stats.rx_dropped++;
1103 return PTR_ERR(skb);
1107 err = zerocopy_sg_from_iovec(skb, iv, offset, count);
1109 err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len);
1112 tun->dev->stats.rx_dropped++;
1117 if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1118 if (!skb_partial_csum_set(skb, gso.csum_start,
1120 tun->dev->stats.rx_frame_errors++;
1126 switch (tun->flags & TUN_TYPE_MASK) {
1128 if (tun->flags & TUN_NO_PI) {
1129 switch (skb->data[0] & 0xf0) {
1131 pi.proto = htons(ETH_P_IP);
1134 pi.proto = htons(ETH_P_IPV6);
1137 tun->dev->stats.rx_dropped++;
1143 skb_reset_mac_header(skb);
1144 skb->protocol = pi.proto;
1145 skb->dev = tun->dev;
1148 skb->protocol = eth_type_trans(skb, tun->dev);
1152 if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1154 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1155 case VIRTIO_NET_HDR_GSO_TCPV4:
1156 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1158 case VIRTIO_NET_HDR_GSO_TCPV6:
1159 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1161 case VIRTIO_NET_HDR_GSO_UDP:
1162 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1165 tun->dev->stats.rx_frame_errors++;
1170 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1171 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
1173 skb_shinfo(skb)->gso_size = gso.gso_size;
1174 if (skb_shinfo(skb)->gso_size == 0) {
1175 tun->dev->stats.rx_frame_errors++;
1180 /* Header must be checked, and gso_segs computed. */
1181 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1182 skb_shinfo(skb)->gso_segs = 0;
1185 /* copy skb_ubuf_info for callback when skb has no error */
1187 skb_shinfo(skb)->destructor_arg = msg_control;
1188 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1191 skb_reset_network_header(skb);
1192 rxhash = skb_get_rxhash(skb);
1195 tun->dev->stats.rx_packets++;
1196 tun->dev->stats.rx_bytes += len;
1198 tun_flow_update(tun, rxhash, tfile->queue_index);
1202 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
1203 unsigned long count, loff_t pos)
1205 struct file *file = iocb->ki_filp;
1206 struct tun_struct *tun = tun_get(file);
1207 struct tun_file *tfile = file->private_data;
1213 tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count);
1215 result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count),
1216 count, file->f_flags & O_NONBLOCK);
1222 /* Put packet to the user space buffer */
1223 static ssize_t tun_put_user(struct tun_struct *tun,
1224 struct tun_file *tfile,
1225 struct sk_buff *skb,
1226 const struct iovec *iv, int len)
1228 struct tun_pi pi = { 0, skb->protocol };
1231 if (!(tun->flags & TUN_NO_PI)) {
1232 if ((len -= sizeof(pi)) < 0)
1235 if (len < skb->len) {
1236 /* Packet will be striped */
1237 pi.flags |= TUN_PKT_STRIP;
1240 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi)))
1242 total += sizeof(pi);
1245 if (tun->flags & TUN_VNET_HDR) {
1246 struct virtio_net_hdr gso = { 0 }; /* no info leak */
1247 if ((len -= tun->vnet_hdr_sz) < 0)
1250 if (skb_is_gso(skb)) {
1251 struct skb_shared_info *sinfo = skb_shinfo(skb);
1253 /* This is a hint as to how much should be linear. */
1254 gso.hdr_len = skb_headlen(skb);
1255 gso.gso_size = sinfo->gso_size;
1256 if (sinfo->gso_type & SKB_GSO_TCPV4)
1257 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1258 else if (sinfo->gso_type & SKB_GSO_TCPV6)
1259 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1260 else if (sinfo->gso_type & SKB_GSO_UDP)
1261 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1263 pr_err("unexpected GSO type: "
1264 "0x%x, gso_size %d, hdr_len %d\n",
1265 sinfo->gso_type, gso.gso_size,
1267 print_hex_dump(KERN_ERR, "tun: ",
1270 min((int)gso.hdr_len, 64), true);
1274 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1275 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1277 gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1279 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1280 gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1281 gso.csum_start = skb_checksum_start_offset(skb);
1282 gso.csum_offset = skb->csum_offset;
1283 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1284 gso.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1285 } /* else everything is zero */
1287 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total,
1290 total += tun->vnet_hdr_sz;
1293 len = min_t(int, skb->len, len);
1295 skb_copy_datagram_const_iovec(skb, 0, iv, total, len);
1298 tun->dev->stats.tx_packets++;
1299 tun->dev->stats.tx_bytes += len;
1304 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1305 struct kiocb *iocb, const struct iovec *iv,
1306 ssize_t len, int noblock)
1308 DECLARE_WAITQUEUE(wait, current);
1309 struct sk_buff *skb;
1312 tun_debug(KERN_INFO, tun, "tun_do_read\n");
1314 if (unlikely(!noblock))
1315 add_wait_queue(&tfile->wq.wait, &wait);
1317 current->state = TASK_INTERRUPTIBLE;
1319 /* Read frames from the queue */
1320 if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) {
1325 if (signal_pending(current)) {
1329 if (tun->dev->reg_state != NETREG_REGISTERED) {
1334 /* Nothing to read, let's sleep */
1339 ret = tun_put_user(tun, tfile, skb, iv, len);
1344 current->state = TASK_RUNNING;
1345 if (unlikely(!noblock))
1346 remove_wait_queue(&tfile->wq.wait, &wait);
1351 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
1352 unsigned long count, loff_t pos)
1354 struct file *file = iocb->ki_filp;
1355 struct tun_file *tfile = file->private_data;
1356 struct tun_struct *tun = __tun_get(tfile);
1361 len = iov_length(iv, count);
1367 ret = tun_do_read(tun, tfile, iocb, iv, len,
1368 file->f_flags & O_NONBLOCK);
1369 ret = min_t(ssize_t, ret, len);
1375 static void tun_free_netdev(struct net_device *dev)
1377 struct tun_struct *tun = netdev_priv(dev);
1379 BUG_ON(!(list_empty(&tun->disabled)));
1380 tun_flow_uninit(tun);
1381 security_tun_dev_free_security(tun->security);
1385 static void tun_setup(struct net_device *dev)
1387 struct tun_struct *tun = netdev_priv(dev);
1389 tun->owner = INVALID_UID;
1390 tun->group = INVALID_GID;
1392 dev->ethtool_ops = &tun_ethtool_ops;
1393 dev->destructor = tun_free_netdev;
1396 /* Trivial set of netlink ops to allow deleting tun or tap
1397 * device with netlink.
1399 static int tun_validate(struct nlattr *tb[], struct nlattr *data[])
1404 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1406 .priv_size = sizeof(struct tun_struct),
1408 .validate = tun_validate,
1411 static void tun_sock_write_space(struct sock *sk)
1413 struct tun_file *tfile;
1414 wait_queue_head_t *wqueue;
1416 if (!sock_writeable(sk))
1419 if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
1422 wqueue = sk_sleep(sk);
1423 if (wqueue && waitqueue_active(wqueue))
1424 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1425 POLLWRNORM | POLLWRBAND);
1427 tfile = container_of(sk, struct tun_file, sk);
1428 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1431 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock,
1432 struct msghdr *m, size_t total_len)
1435 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1436 struct tun_struct *tun = __tun_get(tfile);
1440 ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len,
1441 m->msg_iovlen, m->msg_flags & MSG_DONTWAIT);
1447 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock,
1448 struct msghdr *m, size_t total_len,
1451 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1452 struct tun_struct *tun = __tun_get(tfile);
1458 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1460 ret = tun_do_read(tun, tfile, iocb, m->msg_iov, total_len,
1461 flags & MSG_DONTWAIT);
1462 if (ret > total_len) {
1463 m->msg_flags |= MSG_TRUNC;
1464 ret = flags & MSG_TRUNC ? ret : total_len;
1470 static int tun_release(struct socket *sock)
1477 /* Ops structure to mimic raw sockets with tun */
1478 static const struct proto_ops tun_socket_ops = {
1479 .sendmsg = tun_sendmsg,
1480 .recvmsg = tun_recvmsg,
1481 .release = tun_release,
1484 static struct proto tun_proto = {
1486 .owner = THIS_MODULE,
1487 .obj_size = sizeof(struct tun_file),
1490 static int tun_flags(struct tun_struct *tun)
1494 if (tun->flags & TUN_TUN_DEV)
1499 if (tun->flags & TUN_NO_PI)
1502 /* This flag has no real effect. We track the value for backwards
1505 if (tun->flags & TUN_ONE_QUEUE)
1506 flags |= IFF_ONE_QUEUE;
1508 if (tun->flags & TUN_VNET_HDR)
1509 flags |= IFF_VNET_HDR;
1511 if (tun->flags & TUN_TAP_MQ)
1512 flags |= IFF_MULTI_QUEUE;
1517 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1520 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1521 return sprintf(buf, "0x%x\n", tun_flags(tun));
1524 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1527 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1528 return uid_valid(tun->owner)?
1529 sprintf(buf, "%u\n",
1530 from_kuid_munged(current_user_ns(), tun->owner)):
1531 sprintf(buf, "-1\n");
1534 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1537 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1538 return gid_valid(tun->group) ?
1539 sprintf(buf, "%u\n",
1540 from_kgid_munged(current_user_ns(), tun->group)):
1541 sprintf(buf, "-1\n");
1544 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1545 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1546 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
1548 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
1550 struct tun_struct *tun;
1551 struct tun_file *tfile = file->private_data;
1552 struct net_device *dev;
1555 if (tfile->detached)
1558 dev = __dev_get_by_name(net, ifr->ifr_name);
1560 if (ifr->ifr_flags & IFF_TUN_EXCL)
1562 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
1563 tun = netdev_priv(dev);
1564 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
1565 tun = netdev_priv(dev);
1569 if (tun_not_capable(tun))
1571 err = security_tun_dev_open(tun->security);
1575 err = tun_attach(tun, file);
1579 if (tun->flags & TUN_TAP_MQ &&
1580 (tun->numqueues + tun->numdisabled > 1))
1585 unsigned long flags = 0;
1587 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1589 err = security_tun_dev_create();
1594 if (ifr->ifr_flags & IFF_TUN) {
1596 flags |= TUN_TUN_DEV;
1598 } else if (ifr->ifr_flags & IFF_TAP) {
1600 flags |= TUN_TAP_DEV;
1606 name = ifr->ifr_name;
1608 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
1610 MAX_TAP_QUEUES, MAX_TAP_QUEUES);
1614 dev_net_set(dev, net);
1615 dev->rtnl_link_ops = &tun_link_ops;
1617 tun = netdev_priv(dev);
1620 tun->txflt.count = 0;
1621 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
1623 tun->filter_attached = false;
1624 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
1626 spin_lock_init(&tun->lock);
1628 err = security_tun_dev_alloc_security(&tun->security);
1634 err = tun_flow_init(tun);
1638 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
1640 dev->features = dev->hw_features;
1642 INIT_LIST_HEAD(&tun->disabled);
1643 err = tun_attach(tun, file);
1647 err = register_netdevice(tun->dev);
1651 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) ||
1652 device_create_file(&tun->dev->dev, &dev_attr_owner) ||
1653 device_create_file(&tun->dev->dev, &dev_attr_group))
1654 pr_err("Failed to create tun sysfs files\n");
1656 netif_carrier_on(tun->dev);
1659 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
1661 if (ifr->ifr_flags & IFF_NO_PI)
1662 tun->flags |= TUN_NO_PI;
1664 tun->flags &= ~TUN_NO_PI;
1666 /* This flag has no real effect. We track the value for backwards
1669 if (ifr->ifr_flags & IFF_ONE_QUEUE)
1670 tun->flags |= TUN_ONE_QUEUE;
1672 tun->flags &= ~TUN_ONE_QUEUE;
1674 if (ifr->ifr_flags & IFF_VNET_HDR)
1675 tun->flags |= TUN_VNET_HDR;
1677 tun->flags &= ~TUN_VNET_HDR;
1679 if (ifr->ifr_flags & IFF_MULTI_QUEUE)
1680 tun->flags |= TUN_TAP_MQ;
1682 tun->flags &= ~TUN_TAP_MQ;
1684 /* Make sure persistent devices do not get stuck in
1687 if (netif_running(tun->dev))
1688 netif_tx_wake_all_queues(tun->dev);
1690 strcpy(ifr->ifr_name, tun->dev->name);
1698 static void tun_get_iff(struct net *net, struct tun_struct *tun,
1701 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
1703 strcpy(ifr->ifr_name, tun->dev->name);
1705 ifr->ifr_flags = tun_flags(tun);
1709 /* This is like a cut-down ethtool ops, except done via tun fd so no
1710 * privs required. */
1711 static int set_offload(struct tun_struct *tun, unsigned long arg)
1713 netdev_features_t features = 0;
1715 if (arg & TUN_F_CSUM) {
1716 features |= NETIF_F_HW_CSUM;
1719 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
1720 if (arg & TUN_F_TSO_ECN) {
1721 features |= NETIF_F_TSO_ECN;
1722 arg &= ~TUN_F_TSO_ECN;
1724 if (arg & TUN_F_TSO4)
1725 features |= NETIF_F_TSO;
1726 if (arg & TUN_F_TSO6)
1727 features |= NETIF_F_TSO6;
1728 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
1731 if (arg & TUN_F_UFO) {
1732 features |= NETIF_F_UFO;
1737 /* This gives the user a way to test for new features in future by
1738 * trying to set them. */
1742 tun->set_features = features;
1743 netdev_update_features(tun->dev);
1748 static void tun_detach_filter(struct tun_struct *tun, int n)
1751 struct tun_file *tfile;
1753 for (i = 0; i < n; i++) {
1754 tfile = rtnl_dereference(tun->tfiles[i]);
1755 sk_detach_filter(tfile->socket.sk);
1758 tun->filter_attached = false;
1761 static int tun_attach_filter(struct tun_struct *tun)
1764 struct tun_file *tfile;
1766 for (i = 0; i < tun->numqueues; i++) {
1767 tfile = rtnl_dereference(tun->tfiles[i]);
1768 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
1770 tun_detach_filter(tun, i);
1775 tun->filter_attached = true;
1779 static void tun_set_sndbuf(struct tun_struct *tun)
1781 struct tun_file *tfile;
1784 for (i = 0; i < tun->numqueues; i++) {
1785 tfile = rtnl_dereference(tun->tfiles[i]);
1786 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
1790 static int tun_set_queue(struct file *file, struct ifreq *ifr)
1792 struct tun_file *tfile = file->private_data;
1793 struct tun_struct *tun;
1798 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
1799 tun = tfile->detached;
1804 ret = security_tun_dev_attach_queue(tun->security);
1807 ret = tun_attach(tun, file);
1808 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
1809 tun = rtnl_dereference(tfile->tun);
1810 if (!tun || !(tun->flags & TUN_TAP_MQ))
1813 __tun_detach(tfile, false);
1822 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
1823 unsigned long arg, int ifreq_len)
1825 struct tun_file *tfile = file->private_data;
1826 struct tun_struct *tun;
1827 void __user* argp = (void __user*)arg;
1835 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) {
1836 if (copy_from_user(&ifr, argp, ifreq_len))
1839 memset(&ifr, 0, sizeof(ifr));
1841 if (cmd == TUNGETFEATURES) {
1842 /* Currently this just means: "what IFF flags are valid?".
1843 * This is needed because we never checked for invalid flags on
1845 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
1846 IFF_VNET_HDR | IFF_MULTI_QUEUE,
1847 (unsigned int __user*)argp);
1848 } else if (cmd == TUNSETQUEUE)
1849 return tun_set_queue(file, &ifr);
1854 tun = __tun_get(tfile);
1855 if (cmd == TUNSETIFF && !tun) {
1856 ifr.ifr_name[IFNAMSIZ-1] = '\0';
1858 ret = tun_set_iff(tfile->net, file, &ifr);
1863 if (copy_to_user(argp, &ifr, ifreq_len))
1872 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
1877 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
1879 if (copy_to_user(argp, &ifr, ifreq_len))
1884 /* Disable/Enable checksum */
1886 /* [unimplemented] */
1887 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
1888 arg ? "disabled" : "enabled");
1892 /* Disable/Enable persist mode. Keep an extra reference to the
1893 * module to prevent the module being unprobed.
1895 if (arg && !(tun->flags & TUN_PERSIST)) {
1896 tun->flags |= TUN_PERSIST;
1897 __module_get(THIS_MODULE);
1899 if (!arg && (tun->flags & TUN_PERSIST)) {
1900 tun->flags &= ~TUN_PERSIST;
1901 module_put(THIS_MODULE);
1904 tun_debug(KERN_INFO, tun, "persist %s\n",
1905 arg ? "enabled" : "disabled");
1909 /* Set owner of the device */
1910 owner = make_kuid(current_user_ns(), arg);
1911 if (!uid_valid(owner)) {
1916 tun_debug(KERN_INFO, tun, "owner set to %u\n",
1917 from_kuid(&init_user_ns, tun->owner));
1921 /* Set group of the device */
1922 group = make_kgid(current_user_ns(), arg);
1923 if (!gid_valid(group)) {
1928 tun_debug(KERN_INFO, tun, "group set to %u\n",
1929 from_kgid(&init_user_ns, tun->group));
1933 /* Only allow setting the type when the interface is down */
1934 if (tun->dev->flags & IFF_UP) {
1935 tun_debug(KERN_INFO, tun,
1936 "Linktype set failed because interface is up\n");
1939 tun->dev->type = (int) arg;
1940 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
1952 ret = set_offload(tun, arg);
1955 case TUNSETTXFILTER:
1956 /* Can be set only for TAPs */
1958 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1960 ret = update_filter(&tun->txflt, (void __user *)arg);
1964 /* Get hw address */
1965 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1966 ifr.ifr_hwaddr.sa_family = tun->dev->type;
1967 if (copy_to_user(argp, &ifr, ifreq_len))
1972 /* Set hw address */
1973 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
1974 ifr.ifr_hwaddr.sa_data);
1976 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1980 sndbuf = tfile->socket.sk->sk_sndbuf;
1981 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
1986 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
1991 tun->sndbuf = sndbuf;
1992 tun_set_sndbuf(tun);
1995 case TUNGETVNETHDRSZ:
1996 vnet_hdr_sz = tun->vnet_hdr_sz;
1997 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2001 case TUNSETVNETHDRSZ:
2002 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2006 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2011 tun->vnet_hdr_sz = vnet_hdr_sz;
2014 case TUNATTACHFILTER:
2015 /* Can be set only for TAPs */
2017 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2020 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2023 ret = tun_attach_filter(tun);
2026 case TUNDETACHFILTER:
2027 /* Can be set only for TAPs */
2029 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
2032 tun_detach_filter(tun, tun->numqueues);
2047 static long tun_chr_ioctl(struct file *file,
2048 unsigned int cmd, unsigned long arg)
2050 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2053 #ifdef CONFIG_COMPAT
2054 static long tun_chr_compat_ioctl(struct file *file,
2055 unsigned int cmd, unsigned long arg)
2060 case TUNSETTXFILTER:
2065 arg = (unsigned long)compat_ptr(arg);
2068 arg = (compat_ulong_t)arg;
2073 * compat_ifreq is shorter than ifreq, so we must not access beyond
2074 * the end of that structure. All fields that are used in this
2075 * driver are compatible though, we don't need to convert the
2078 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2080 #endif /* CONFIG_COMPAT */
2082 static int tun_chr_fasync(int fd, struct file *file, int on)
2084 struct tun_file *tfile = file->private_data;
2087 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2091 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2094 tfile->flags |= TUN_FASYNC;
2096 tfile->flags &= ~TUN_FASYNC;
2102 static int tun_chr_open(struct inode *inode, struct file * file)
2104 struct tun_file *tfile;
2106 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2108 tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL,
2112 rcu_assign_pointer(tfile->tun, NULL);
2113 tfile->net = get_net(current->nsproxy->net_ns);
2116 rcu_assign_pointer(tfile->socket.wq, &tfile->wq);
2117 init_waitqueue_head(&tfile->wq.wait);
2119 tfile->socket.file = file;
2120 tfile->socket.ops = &tun_socket_ops;
2122 sock_init_data(&tfile->socket, &tfile->sk);
2123 sk_change_net(&tfile->sk, tfile->net);
2125 tfile->sk.sk_write_space = tun_sock_write_space;
2126 tfile->sk.sk_sndbuf = INT_MAX;
2128 file->private_data = tfile;
2129 set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags);
2130 INIT_LIST_HEAD(&tfile->next);
2135 static int tun_chr_close(struct inode *inode, struct file *file)
2137 struct tun_file *tfile = file->private_data;
2138 struct net *net = tfile->net;
2140 tun_detach(tfile, true);
2146 static const struct file_operations tun_fops = {
2147 .owner = THIS_MODULE,
2148 .llseek = no_llseek,
2149 .read = do_sync_read,
2150 .aio_read = tun_chr_aio_read,
2151 .write = do_sync_write,
2152 .aio_write = tun_chr_aio_write,
2153 .poll = tun_chr_poll,
2154 .unlocked_ioctl = tun_chr_ioctl,
2155 #ifdef CONFIG_COMPAT
2156 .compat_ioctl = tun_chr_compat_ioctl,
2158 .open = tun_chr_open,
2159 .release = tun_chr_close,
2160 .fasync = tun_chr_fasync
2163 static struct miscdevice tun_miscdev = {
2166 .nodename = "net/tun",
2170 /* ethtool interface */
2172 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2175 cmd->advertising = 0;
2176 ethtool_cmd_speed_set(cmd, SPEED_10);
2177 cmd->duplex = DUPLEX_FULL;
2178 cmd->port = PORT_TP;
2179 cmd->phy_address = 0;
2180 cmd->transceiver = XCVR_INTERNAL;
2181 cmd->autoneg = AUTONEG_DISABLE;
2187 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2189 struct tun_struct *tun = netdev_priv(dev);
2191 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2192 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2194 switch (tun->flags & TUN_TYPE_MASK) {
2196 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2199 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2204 static u32 tun_get_msglevel(struct net_device *dev)
2207 struct tun_struct *tun = netdev_priv(dev);
2214 static void tun_set_msglevel(struct net_device *dev, u32 value)
2217 struct tun_struct *tun = netdev_priv(dev);
2222 static const struct ethtool_ops tun_ethtool_ops = {
2223 .get_settings = tun_get_settings,
2224 .get_drvinfo = tun_get_drvinfo,
2225 .get_msglevel = tun_get_msglevel,
2226 .set_msglevel = tun_set_msglevel,
2227 .get_link = ethtool_op_get_link,
2231 static int __init tun_init(void)
2235 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2236 pr_info("%s\n", DRV_COPYRIGHT);
2238 ret = rtnl_link_register(&tun_link_ops);
2240 pr_err("Can't register link_ops\n");
2244 ret = misc_register(&tun_miscdev);
2246 pr_err("Can't register misc device %d\n", TUN_MINOR);
2251 rtnl_link_unregister(&tun_link_ops);
2256 static void tun_cleanup(void)
2258 misc_deregister(&tun_miscdev);
2259 rtnl_link_unregister(&tun_link_ops);
2262 /* Get an underlying socket object from tun file. Returns error unless file is
2263 * attached to a device. The returned object works like a packet socket, it
2264 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
2265 * holding a reference to the file for as long as the socket is in use. */
2266 struct socket *tun_get_socket(struct file *file)
2268 struct tun_file *tfile;
2269 if (file->f_op != &tun_fops)
2270 return ERR_PTR(-EINVAL);
2271 tfile = file->private_data;
2273 return ERR_PTR(-EBADFD);
2274 return &tfile->socket;
2276 EXPORT_SYMBOL_GPL(tun_get_socket);
2278 module_init(tun_init);
2279 module_exit(tun_cleanup);
2280 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2281 MODULE_AUTHOR(DRV_COPYRIGHT);
2282 MODULE_LICENSE("GPL");
2283 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2284 MODULE_ALIAS("devname:net/tun");