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>
76 #include <linux/bpf.h>
77 #include <linux/bpf_trace.h>
78 #include <linux/ieee802154.h>
79 #include <linux/if_ltalk.h>
80 #include <uapi/linux/if_fddi.h>
81 #include <uapi/linux/if_hippi.h>
82 #include <uapi/linux/if_fc.h>
85 #include <net/6lowpan.h>
87 #include <linux/uaccess.h>
89 /* Uncomment to enable debugging */
90 /* #define TUN_DEBUG 1 */
95 #define tun_debug(level, tun, fmt, args...) \
98 netdev_printk(level, tun->dev, fmt, ##args); \
100 #define DBG1(level, fmt, args...) \
103 printk(level fmt, ##args); \
106 #define tun_debug(level, tun, fmt, args...) \
109 netdev_printk(level, tun->dev, fmt, ##args); \
111 #define DBG1(level, fmt, args...) \
114 printk(level fmt, ##args); \
118 #define TUN_HEADROOM 256
119 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
121 /* TUN device flags */
123 /* IFF_ATTACH_QUEUE is never stored in device flags,
124 * overload it to mean fasync when stored there.
126 #define TUN_FASYNC IFF_ATTACH_QUEUE
127 /* High bits in flags field are unused. */
128 #define TUN_VNET_LE 0x80000000
129 #define TUN_VNET_BE 0x40000000
131 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
133 #define GOODCOPY_LEN 128
135 #define FLT_EXACT_COUNT 8
137 unsigned int count; /* Number of addrs. Zero means disabled */
138 u32 mask[2]; /* Mask of the hashed addrs */
139 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
142 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
143 * to max number of VCPUs in guest. */
144 #define MAX_TAP_QUEUES 256
145 #define MAX_TAP_FLOWS 4096
147 #define TUN_FLOW_EXPIRE (3 * HZ)
149 struct tun_pcpu_stats {
154 struct u64_stats_sync syncp;
160 /* A tun_file connects an open character device to a tuntap netdevice. It
161 * also contains all socket related structures (except sock_fprog and tap_filter)
162 * to serve as one transmit queue for tuntap device. The sock_fprog and
163 * tap_filter were kept in tun_struct since they were used for filtering for the
164 * netdevice not for a specific queue (at least I didn't see the requirement for
168 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
169 * other can only be read while rcu_read_lock or rtnl_lock is held.
173 struct socket socket;
175 struct tun_struct __rcu *tun;
176 struct fasync_struct *fasync;
177 /* only used for fasnyc */
181 unsigned int ifindex;
183 struct list_head next;
184 struct tun_struct *detached;
185 struct skb_array tx_array;
188 struct tun_flow_entry {
189 struct hlist_node hash_link;
191 struct tun_struct *tun;
196 unsigned long updated;
199 #define TUN_NUM_FLOW_ENTRIES 1024
201 /* Since the socket were moved to tun_file, to preserve the behavior of persist
202 * device, socket filter, sndbuf and vnet header size were restore when the
203 * file were attached to a persist device.
206 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
207 unsigned int numqueues;
212 struct net_device *dev;
213 netdev_features_t set_features;
214 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
220 struct tap_filter txflt;
221 struct sock_fprog fprog;
222 /* protected by rtnl lock */
223 bool filter_attached;
228 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
229 struct timer_list flow_gc_timer;
230 unsigned long ageing_time;
231 unsigned int numdisabled;
232 struct list_head disabled;
236 struct tun_pcpu_stats __percpu *pcpu_stats;
237 struct bpf_prog __rcu *xdp_prog;
240 #ifdef CONFIG_TUN_VNET_CROSS_LE
241 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
243 return tun->flags & TUN_VNET_BE ? false :
244 virtio_legacy_is_little_endian();
247 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
249 int be = !!(tun->flags & TUN_VNET_BE);
251 if (put_user(be, argp))
257 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
261 if (get_user(be, argp))
265 tun->flags |= TUN_VNET_BE;
267 tun->flags &= ~TUN_VNET_BE;
272 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
274 return virtio_legacy_is_little_endian();
277 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
282 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
286 #endif /* CONFIG_TUN_VNET_CROSS_LE */
288 static inline bool tun_is_little_endian(struct tun_struct *tun)
290 return tun->flags & TUN_VNET_LE ||
291 tun_legacy_is_little_endian(tun);
294 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
296 return __virtio16_to_cpu(tun_is_little_endian(tun), val);
299 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
301 return __cpu_to_virtio16(tun_is_little_endian(tun), val);
304 static inline u32 tun_hashfn(u32 rxhash)
306 return rxhash & 0x3ff;
309 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
311 struct tun_flow_entry *e;
313 hlist_for_each_entry_rcu(e, head, hash_link) {
314 if (e->rxhash == rxhash)
320 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
321 struct hlist_head *head,
322 u32 rxhash, u16 queue_index)
324 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
327 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
328 rxhash, queue_index);
329 e->updated = jiffies;
332 e->queue_index = queue_index;
334 hlist_add_head_rcu(&e->hash_link, head);
340 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
342 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
343 e->rxhash, e->queue_index);
344 hlist_del_rcu(&e->hash_link);
349 static void tun_flow_flush(struct tun_struct *tun)
353 spin_lock_bh(&tun->lock);
354 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
355 struct tun_flow_entry *e;
356 struct hlist_node *n;
358 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
359 tun_flow_delete(tun, e);
361 spin_unlock_bh(&tun->lock);
364 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
368 spin_lock_bh(&tun->lock);
369 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
370 struct tun_flow_entry *e;
371 struct hlist_node *n;
373 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
374 if (e->queue_index == queue_index)
375 tun_flow_delete(tun, e);
378 spin_unlock_bh(&tun->lock);
381 static void tun_flow_cleanup(unsigned long data)
383 struct tun_struct *tun = (struct tun_struct *)data;
384 unsigned long delay = tun->ageing_time;
385 unsigned long next_timer = jiffies + delay;
386 unsigned long count = 0;
389 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
391 spin_lock_bh(&tun->lock);
392 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
393 struct tun_flow_entry *e;
394 struct hlist_node *n;
396 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
397 unsigned long this_timer;
399 this_timer = e->updated + delay;
400 if (time_before_eq(this_timer, jiffies))
401 tun_flow_delete(tun, e);
402 else if (time_before(this_timer, next_timer))
403 next_timer = this_timer;
408 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
409 spin_unlock_bh(&tun->lock);
412 static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
413 struct tun_file *tfile)
415 struct hlist_head *head;
416 struct tun_flow_entry *e;
417 unsigned long delay = tun->ageing_time;
418 u16 queue_index = tfile->queue_index;
423 head = &tun->flows[tun_hashfn(rxhash)];
427 /* We may get a very small possibility of OOO during switching, not
428 * worth to optimize.*/
429 if (tun->numqueues == 1 || tfile->detached)
432 e = tun_flow_find(head, rxhash);
434 /* TODO: keep queueing to old queue until it's empty? */
435 e->queue_index = queue_index;
436 e->updated = jiffies;
437 sock_rps_record_flow_hash(e->rps_rxhash);
439 spin_lock_bh(&tun->lock);
440 if (!tun_flow_find(head, rxhash) &&
441 tun->flow_count < MAX_TAP_FLOWS)
442 tun_flow_create(tun, head, rxhash, queue_index);
444 if (!timer_pending(&tun->flow_gc_timer))
445 mod_timer(&tun->flow_gc_timer,
446 round_jiffies_up(jiffies + delay));
447 spin_unlock_bh(&tun->lock);
455 * Save the hash received in the stack receive path and update the
456 * flow_hash table accordingly.
458 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
460 if (unlikely(e->rps_rxhash != hash))
461 e->rps_rxhash = hash;
464 /* We try to identify a flow through its rxhash first. The reason that
465 * we do not check rxq no. is because some cards(e.g 82599), chooses
466 * the rxq based on the txq where the last packet of the flow comes. As
467 * the userspace application move between processors, we may get a
468 * different rxq no. here. If we could not get rxhash, then we would
469 * hope the rxq no. may help here.
471 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
472 void *accel_priv, select_queue_fallback_t fallback)
474 struct tun_struct *tun = netdev_priv(dev);
475 struct tun_flow_entry *e;
480 numqueues = ACCESS_ONCE(tun->numqueues);
482 txq = __skb_get_hash_symmetric(skb);
484 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
486 tun_flow_save_rps_rxhash(e, txq);
487 txq = e->queue_index;
489 /* use multiply and shift instead of expensive divide */
490 txq = ((u64)txq * numqueues) >> 32;
491 } else if (likely(skb_rx_queue_recorded(skb))) {
492 txq = skb_get_rx_queue(skb);
493 while (unlikely(txq >= numqueues))
501 static inline bool tun_not_capable(struct tun_struct *tun)
503 const struct cred *cred = current_cred();
504 struct net *net = dev_net(tun->dev);
506 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
507 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
508 !ns_capable(net->user_ns, CAP_NET_ADMIN);
511 static void tun_set_real_num_queues(struct tun_struct *tun)
513 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
514 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
517 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
519 tfile->detached = tun;
520 list_add_tail(&tfile->next, &tun->disabled);
524 static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
526 struct tun_struct *tun = tfile->detached;
528 tfile->detached = NULL;
529 list_del_init(&tfile->next);
534 static void tun_queue_purge(struct tun_file *tfile)
538 while ((skb = skb_array_consume(&tfile->tx_array)) != NULL)
541 skb_queue_purge(&tfile->sk.sk_write_queue);
542 skb_queue_purge(&tfile->sk.sk_error_queue);
545 static void __tun_detach(struct tun_file *tfile, bool clean)
547 struct tun_file *ntfile;
548 struct tun_struct *tun;
550 tun = rtnl_dereference(tfile->tun);
552 if (tun && !tfile->detached) {
553 u16 index = tfile->queue_index;
554 BUG_ON(index >= tun->numqueues);
556 rcu_assign_pointer(tun->tfiles[index],
557 tun->tfiles[tun->numqueues - 1]);
558 ntfile = rtnl_dereference(tun->tfiles[index]);
559 ntfile->queue_index = index;
563 RCU_INIT_POINTER(tfile->tun, NULL);
564 sock_put(&tfile->sk);
566 tun_disable_queue(tun, tfile);
569 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
570 /* Drop read queue */
571 tun_queue_purge(tfile);
572 tun_set_real_num_queues(tun);
573 } else if (tfile->detached && clean) {
574 tun = tun_enable_queue(tfile);
575 sock_put(&tfile->sk);
579 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
580 netif_carrier_off(tun->dev);
582 if (!(tun->flags & IFF_PERSIST) &&
583 tun->dev->reg_state == NETREG_REGISTERED)
584 unregister_netdevice(tun->dev);
586 skb_array_cleanup(&tfile->tx_array);
587 sock_put(&tfile->sk);
591 static void tun_detach(struct tun_file *tfile, bool clean)
594 __tun_detach(tfile, clean);
598 static void tun_detach_all(struct net_device *dev)
600 struct tun_struct *tun = netdev_priv(dev);
601 struct bpf_prog *xdp_prog = rtnl_dereference(tun->xdp_prog);
602 struct tun_file *tfile, *tmp;
603 int i, n = tun->numqueues;
605 for (i = 0; i < n; i++) {
606 tfile = rtnl_dereference(tun->tfiles[i]);
608 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
609 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
610 RCU_INIT_POINTER(tfile->tun, NULL);
613 list_for_each_entry(tfile, &tun->disabled, next) {
614 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
615 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
616 RCU_INIT_POINTER(tfile->tun, NULL);
618 BUG_ON(tun->numqueues != 0);
621 for (i = 0; i < n; i++) {
622 tfile = rtnl_dereference(tun->tfiles[i]);
623 /* Drop read queue */
624 tun_queue_purge(tfile);
625 sock_put(&tfile->sk);
627 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
628 tun_enable_queue(tfile);
629 tun_queue_purge(tfile);
630 sock_put(&tfile->sk);
632 BUG_ON(tun->numdisabled != 0);
635 bpf_prog_put(xdp_prog);
637 if (tun->flags & IFF_PERSIST)
638 module_put(THIS_MODULE);
641 static int tun_attach(struct tun_struct *tun, struct file *file,
642 bool skip_filter, bool publish_tun)
644 struct tun_file *tfile = file->private_data;
645 struct net_device *dev = tun->dev;
648 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
653 if (rtnl_dereference(tfile->tun) && !tfile->detached)
657 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
661 if (!tfile->detached &&
662 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
667 /* Re-attach the filter to persist device */
668 if (!skip_filter && (tun->filter_attached == true)) {
669 lock_sock(tfile->socket.sk);
670 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
671 release_sock(tfile->socket.sk);
676 if (!tfile->detached &&
677 skb_array_resize(&tfile->tx_array, dev->tx_queue_len, GFP_KERNEL)) {
682 tfile->queue_index = tun->numqueues;
683 tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
685 rcu_assign_pointer(tfile->tun, tun);
686 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
690 tun_enable_queue(tfile);
692 sock_hold(&tfile->sk);
694 tun_set_real_num_queues(tun);
696 /* device is allowed to go away first, so no need to hold extra
704 static struct tun_struct *__tun_get(struct tun_file *tfile)
706 struct tun_struct *tun;
709 tun = rcu_dereference(tfile->tun);
717 static struct tun_struct *tun_get(struct file *file)
719 return __tun_get(file->private_data);
722 static void tun_put(struct tun_struct *tun)
728 static void addr_hash_set(u32 *mask, const u8 *addr)
730 int n = ether_crc(ETH_ALEN, addr) >> 26;
731 mask[n >> 5] |= (1 << (n & 31));
734 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
736 int n = ether_crc(ETH_ALEN, addr) >> 26;
737 return mask[n >> 5] & (1 << (n & 31));
740 static int update_filter(struct tap_filter *filter, void __user *arg)
742 struct { u8 u[ETH_ALEN]; } *addr;
743 struct tun_filter uf;
744 int err, alen, n, nexact;
746 if (copy_from_user(&uf, arg, sizeof(uf)))
755 alen = ETH_ALEN * uf.count;
756 addr = memdup_user(arg + sizeof(uf), alen);
758 return PTR_ERR(addr);
760 /* The filter is updated without holding any locks. Which is
761 * perfectly safe. We disable it first and in the worst
762 * case we'll accept a few undesired packets. */
766 /* Use first set of addresses as an exact filter */
767 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
768 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
772 /* Remaining multicast addresses are hashed,
773 * unicast will leave the filter disabled. */
774 memset(filter->mask, 0, sizeof(filter->mask));
775 for (; n < uf.count; n++) {
776 if (!is_multicast_ether_addr(addr[n].u)) {
777 err = 0; /* no filter */
780 addr_hash_set(filter->mask, addr[n].u);
783 /* For ALLMULTI just set the mask to all ones.
784 * This overrides the mask populated above. */
785 if ((uf.flags & TUN_FLT_ALLMULTI))
786 memset(filter->mask, ~0, sizeof(filter->mask));
788 /* Now enable the filter */
790 filter->count = nexact;
792 /* Return the number of exact filters */
799 /* Returns: 0 - drop, !=0 - accept */
800 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
802 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
804 struct ethhdr *eh = (struct ethhdr *) skb->data;
808 for (i = 0; i < filter->count; i++)
809 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
812 /* Inexact match (multicast only) */
813 if (is_multicast_ether_addr(eh->h_dest))
814 return addr_hash_test(filter->mask, eh->h_dest);
820 * Checks whether the packet is accepted or not.
821 * Returns: 0 - drop, !=0 - accept
823 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
828 return run_filter(filter, skb);
831 /* Network device part of the driver */
833 static const struct ethtool_ops tun_ethtool_ops;
835 /* Net device detach from fd. */
836 static void tun_net_uninit(struct net_device *dev)
841 /* Net device open. */
842 static int tun_net_open(struct net_device *dev)
844 netif_tx_start_all_queues(dev);
849 /* Net device close. */
850 static int tun_net_close(struct net_device *dev)
852 netif_tx_stop_all_queues(dev);
856 /* Net device start xmit */
857 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
859 struct tun_struct *tun = netdev_priv(dev);
860 int txq = skb->queue_mapping;
861 struct tun_file *tfile;
865 tfile = rcu_dereference(tun->tfiles[txq]);
866 numqueues = ACCESS_ONCE(tun->numqueues);
868 /* Drop packet if interface is not attached */
869 if (txq >= numqueues)
873 if (numqueues == 1 && static_key_false(&rps_needed)) {
874 /* Select queue was not called for the skbuff, so we extract the
875 * RPS hash and save it into the flow_table here.
879 rxhash = __skb_get_hash_symmetric(skb);
881 struct tun_flow_entry *e;
882 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
885 tun_flow_save_rps_rxhash(e, rxhash);
890 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
894 /* Drop if the filter does not like it.
895 * This is a noop if the filter is disabled.
896 * Filter can be enabled only for the TAP devices. */
897 if (!check_filter(&tun->txflt, skb))
900 if (tfile->socket.sk->sk_filter &&
901 sk_filter(tfile->socket.sk, skb))
904 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
907 skb_tx_timestamp(skb);
909 /* Orphan the skb - required as we might hang on to it
910 * for indefinite time.
916 if (skb_array_produce(&tfile->tx_array, skb))
919 /* Notify and wake up reader process */
920 if (tfile->flags & TUN_FASYNC)
921 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
922 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
928 this_cpu_inc(tun->pcpu_stats->tx_dropped);
932 return NET_XMIT_DROP;
935 static void tun_net_mclist(struct net_device *dev)
938 * This callback is supposed to deal with mc filter in
939 * _rx_ path and has nothing to do with the _tx_ path.
940 * In rx path we always accept everything userspace gives us.
944 static netdev_features_t tun_net_fix_features(struct net_device *dev,
945 netdev_features_t features)
947 struct tun_struct *tun = netdev_priv(dev);
949 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
951 #ifdef CONFIG_NET_POLL_CONTROLLER
952 static void tun_poll_controller(struct net_device *dev)
955 * Tun only receives frames when:
956 * 1) the char device endpoint gets data from user space
957 * 2) the tun socket gets a sendmsg call from user space
958 * Since both of those are synchronous operations, we are guaranteed
959 * never to have pending data when we poll for it
960 * so there is nothing to do here but return.
961 * We need this though so netpoll recognizes us as an interface that
962 * supports polling, which enables bridge devices in virt setups to
963 * still use netconsole
969 static void tun_set_headroom(struct net_device *dev, int new_hr)
971 struct tun_struct *tun = netdev_priv(dev);
973 if (new_hr < NET_SKB_PAD)
974 new_hr = NET_SKB_PAD;
980 tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
982 u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
983 struct tun_struct *tun = netdev_priv(dev);
984 struct tun_pcpu_stats *p;
987 for_each_possible_cpu(i) {
988 u64 rxpackets, rxbytes, txpackets, txbytes;
991 p = per_cpu_ptr(tun->pcpu_stats, i);
993 start = u64_stats_fetch_begin(&p->syncp);
994 rxpackets = p->rx_packets;
995 rxbytes = p->rx_bytes;
996 txpackets = p->tx_packets;
997 txbytes = p->tx_bytes;
998 } while (u64_stats_fetch_retry(&p->syncp, start));
1000 stats->rx_packets += rxpackets;
1001 stats->rx_bytes += rxbytes;
1002 stats->tx_packets += txpackets;
1003 stats->tx_bytes += txbytes;
1006 rx_dropped += p->rx_dropped;
1007 rx_frame_errors += p->rx_frame_errors;
1008 tx_dropped += p->tx_dropped;
1010 stats->rx_dropped = rx_dropped;
1011 stats->rx_frame_errors = rx_frame_errors;
1012 stats->tx_dropped = tx_dropped;
1015 static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1016 struct netlink_ext_ack *extack)
1018 struct tun_struct *tun = netdev_priv(dev);
1019 struct bpf_prog *old_prog;
1021 old_prog = rtnl_dereference(tun->xdp_prog);
1022 rcu_assign_pointer(tun->xdp_prog, prog);
1024 bpf_prog_put(old_prog);
1029 static u32 tun_xdp_query(struct net_device *dev)
1031 struct tun_struct *tun = netdev_priv(dev);
1032 const struct bpf_prog *xdp_prog;
1034 xdp_prog = rtnl_dereference(tun->xdp_prog);
1036 return xdp_prog->aux->id;
1041 static int tun_xdp(struct net_device *dev, struct netdev_xdp *xdp)
1043 switch (xdp->command) {
1044 case XDP_SETUP_PROG:
1045 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1046 case XDP_QUERY_PROG:
1047 xdp->prog_id = tun_xdp_query(dev);
1048 xdp->prog_attached = !!xdp->prog_id;
1055 static const struct net_device_ops tun_netdev_ops = {
1056 .ndo_uninit = tun_net_uninit,
1057 .ndo_open = tun_net_open,
1058 .ndo_stop = tun_net_close,
1059 .ndo_start_xmit = tun_net_xmit,
1060 .ndo_fix_features = tun_net_fix_features,
1061 .ndo_select_queue = tun_select_queue,
1062 #ifdef CONFIG_NET_POLL_CONTROLLER
1063 .ndo_poll_controller = tun_poll_controller,
1065 .ndo_set_rx_headroom = tun_set_headroom,
1066 .ndo_get_stats64 = tun_net_get_stats64,
1069 static const struct net_device_ops tap_netdev_ops = {
1070 .ndo_uninit = tun_net_uninit,
1071 .ndo_open = tun_net_open,
1072 .ndo_stop = tun_net_close,
1073 .ndo_start_xmit = tun_net_xmit,
1074 .ndo_fix_features = tun_net_fix_features,
1075 .ndo_set_rx_mode = tun_net_mclist,
1076 .ndo_set_mac_address = eth_mac_addr,
1077 .ndo_validate_addr = eth_validate_addr,
1078 .ndo_select_queue = tun_select_queue,
1079 #ifdef CONFIG_NET_POLL_CONTROLLER
1080 .ndo_poll_controller = tun_poll_controller,
1082 .ndo_features_check = passthru_features_check,
1083 .ndo_set_rx_headroom = tun_set_headroom,
1084 .ndo_get_stats64 = tun_net_get_stats64,
1088 static void tun_flow_init(struct tun_struct *tun)
1092 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1093 INIT_HLIST_HEAD(&tun->flows[i]);
1095 tun->ageing_time = TUN_FLOW_EXPIRE;
1096 setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun);
1097 mod_timer(&tun->flow_gc_timer,
1098 round_jiffies_up(jiffies + tun->ageing_time));
1101 static void tun_flow_uninit(struct tun_struct *tun)
1103 del_timer_sync(&tun->flow_gc_timer);
1104 tun_flow_flush(tun);
1108 #define MAX_MTU 65535
1110 /* Initialize net device. */
1111 static void tun_net_init(struct net_device *dev)
1113 struct tun_struct *tun = netdev_priv(dev);
1115 switch (tun->flags & TUN_TYPE_MASK) {
1117 dev->netdev_ops = &tun_netdev_ops;
1119 /* Point-to-Point TUN Device */
1120 dev->hard_header_len = 0;
1124 /* Zero header length */
1125 dev->type = ARPHRD_NONE;
1126 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
1130 dev->netdev_ops = &tap_netdev_ops;
1131 /* Ethernet TAP Device */
1133 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1134 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1136 eth_hw_addr_random(dev);
1141 dev->min_mtu = MIN_MTU;
1142 dev->max_mtu = MAX_MTU - dev->hard_header_len;
1145 static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1147 struct sock *sk = tfile->socket.sk;
1149 return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1152 /* Character device part */
1155 static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
1157 struct tun_file *tfile = file->private_data;
1158 struct tun_struct *tun = __tun_get(tfile);
1160 unsigned int mask = 0;
1165 sk = tfile->socket.sk;
1167 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
1169 poll_wait(file, sk_sleep(sk), wait);
1171 if (!skb_array_empty(&tfile->tx_array))
1172 mask |= POLLIN | POLLRDNORM;
1174 /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1175 * guarantee EPOLLOUT to be raised by either here or
1176 * tun_sock_write_space(). Then process could get notification
1177 * after it writes to a down device and meets -EIO.
1179 if (tun_sock_writeable(tun, tfile) ||
1180 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1181 tun_sock_writeable(tun, tfile)))
1182 mask |= POLLOUT | POLLWRNORM;
1184 if (tun->dev->reg_state != NETREG_REGISTERED)
1191 /* prepad is the amount to reserve at front. len is length after that.
1192 * linear is a hint as to how much to copy (usually headers). */
1193 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
1194 size_t prepad, size_t len,
1195 size_t linear, int noblock)
1197 struct sock *sk = tfile->socket.sk;
1198 struct sk_buff *skb;
1201 /* Under a page? Don't bother with paged skb. */
1202 if (prepad + len < PAGE_SIZE || !linear)
1205 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1208 return ERR_PTR(err);
1210 skb_reserve(skb, prepad);
1211 skb_put(skb, linear);
1212 skb->data_len = len - linear;
1213 skb->len += len - linear;
1218 static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1219 struct sk_buff *skb, int more)
1221 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1222 struct sk_buff_head process_queue;
1223 u32 rx_batched = tun->rx_batched;
1226 if (!rx_batched || (!more && skb_queue_empty(queue))) {
1228 skb_record_rx_queue(skb, tfile->queue_index);
1229 netif_receive_skb(skb);
1234 spin_lock(&queue->lock);
1235 if (!more || skb_queue_len(queue) == rx_batched) {
1236 __skb_queue_head_init(&process_queue);
1237 skb_queue_splice_tail_init(queue, &process_queue);
1240 __skb_queue_tail(queue, skb);
1242 spin_unlock(&queue->lock);
1245 struct sk_buff *nskb;
1248 while ((nskb = __skb_dequeue(&process_queue))) {
1249 skb_record_rx_queue(nskb, tfile->queue_index);
1250 netif_receive_skb(nskb);
1252 skb_record_rx_queue(skb, tfile->queue_index);
1253 netif_receive_skb(skb);
1258 static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1259 int len, int noblock, bool zerocopy)
1261 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1264 if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1273 if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1274 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1280 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1281 struct tun_file *tfile,
1282 struct iov_iter *from,
1283 struct virtio_net_hdr *hdr,
1284 int len, int *skb_xdp)
1286 struct page_frag *alloc_frag = ¤t->task_frag;
1287 struct sk_buff *skb;
1288 struct bpf_prog *xdp_prog;
1289 int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1290 unsigned int delta = 0;
1293 bool xdp_xmit = false;
1294 int err, pad = TUN_RX_PAD;
1297 xdp_prog = rcu_dereference(tun->xdp_prog);
1299 pad += TUN_HEADROOM;
1300 buflen += SKB_DATA_ALIGN(len + pad);
1303 alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
1304 if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1305 return ERR_PTR(-ENOMEM);
1307 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1308 copied = copy_page_from_iter(alloc_frag->page,
1309 alloc_frag->offset + pad,
1312 return ERR_PTR(-EFAULT);
1314 /* There's a small window that XDP may be set after the check
1315 * of xdp_prog above, this should be rare and for simplicity
1316 * we do XDP on skb in case the headroom is not enough.
1318 if (hdr->gso_type || !xdp_prog)
1325 xdp_prog = rcu_dereference(tun->xdp_prog);
1326 if (xdp_prog && !*skb_xdp) {
1327 struct xdp_buff xdp;
1331 xdp.data_hard_start = buf;
1332 xdp.data = buf + pad;
1333 xdp.data_end = xdp.data + len;
1334 orig_data = xdp.data;
1335 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1339 get_page(alloc_frag->page);
1340 alloc_frag->offset += buflen;
1341 err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1352 delta = orig_data - xdp.data;
1355 bpf_warn_invalid_xdp_action(act);
1358 trace_xdp_exception(tun->dev, xdp_prog, act);
1365 skb = build_skb(buf, buflen);
1369 return ERR_PTR(-ENOMEM);
1372 skb_reserve(skb, pad - delta);
1373 skb_put(skb, len + delta);
1374 skb_set_owner_w(skb, tfile->socket.sk);
1375 get_page(alloc_frag->page);
1376 alloc_frag->offset += buflen;
1379 skb->dev = tun->dev;
1380 generic_xdp_tx(skb, xdp_prog);
1392 put_page(alloc_frag->page);
1396 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1400 /* Get packet from user space buffer */
1401 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
1402 void *msg_control, struct iov_iter *from,
1403 int noblock, bool more)
1405 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
1406 struct sk_buff *skb;
1407 size_t total_len = iov_iter_count(from);
1408 size_t len = total_len, align = tun->align, linear;
1409 struct virtio_net_hdr gso = { 0 };
1410 struct tun_pcpu_stats *stats;
1413 bool zerocopy = false;
1418 if (!(tun->flags & IFF_NO_PI)) {
1419 if (len < sizeof(pi))
1423 if (!copy_from_iter_full(&pi, sizeof(pi), from))
1427 if (tun->flags & IFF_VNET_HDR) {
1428 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1430 if (len < vnet_hdr_sz)
1434 if (!copy_from_iter_full(&gso, sizeof(gso), from))
1437 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1438 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1439 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
1441 if (tun16_to_cpu(tun, gso.hdr_len) > len)
1443 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
1446 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
1447 align += NET_IP_ALIGN;
1448 if (unlikely(len < ETH_HLEN ||
1449 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
1453 good_linear = SKB_MAX_HEAD(align);
1456 struct iov_iter i = *from;
1458 /* There are 256 bytes to be copied in skb, so there is
1459 * enough room for skb expand head in case it is used.
1460 * The rest of the buffer is mapped from userspace.
1462 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
1463 if (copylen > good_linear)
1464 copylen = good_linear;
1466 iov_iter_advance(&i, copylen);
1467 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
1471 if (tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
1472 /* For the packet that is not easy to be processed
1473 * (e.g gso or jumbo packet), we will do it at after
1474 * skb was created with generic XDP routine.
1476 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
1478 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1479 return PTR_ERR(skb);
1486 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1487 linear = good_linear;
1489 linear = tun16_to_cpu(tun, gso.hdr_len);
1492 skb = tun_alloc_skb(tfile, align, copylen, linear, noblock);
1494 if (PTR_ERR(skb) != -EAGAIN)
1495 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1496 return PTR_ERR(skb);
1500 err = zerocopy_sg_from_iter(skb, from);
1502 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1507 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1513 if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
1514 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1519 switch (tun->flags & TUN_TYPE_MASK) {
1521 if (tun->flags & IFF_NO_PI) {
1522 u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1524 switch (ip_version) {
1526 pi.proto = htons(ETH_P_IP);
1529 pi.proto = htons(ETH_P_IPV6);
1532 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1538 skb_reset_mac_header(skb);
1539 skb->protocol = pi.proto;
1540 skb->dev = tun->dev;
1543 skb->protocol = eth_type_trans(skb, tun->dev);
1547 /* copy skb_ubuf_info for callback when skb has no error */
1549 skb_shinfo(skb)->destructor_arg = msg_control;
1550 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
1551 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
1552 } else if (msg_control) {
1553 struct ubuf_info *uarg = msg_control;
1554 uarg->callback(uarg, false);
1557 skb_reset_network_header(skb);
1558 skb_probe_transport_header(skb, 0);
1561 struct bpf_prog *xdp_prog;
1566 xdp_prog = rcu_dereference(tun->xdp_prog);
1568 ret = do_xdp_generic(xdp_prog, skb);
1569 if (ret != XDP_PASS) {
1579 rxhash = __skb_get_hash_symmetric(skb);
1582 if (unlikely(!(tun->dev->flags & IFF_UP))) {
1588 #ifndef CONFIG_4KSTACKS
1589 tun_rx_batched(tun, tfile, skb, more);
1595 stats = get_cpu_ptr(tun->pcpu_stats);
1596 u64_stats_update_begin(&stats->syncp);
1597 stats->rx_packets++;
1598 stats->rx_bytes += len;
1599 u64_stats_update_end(&stats->syncp);
1602 tun_flow_update(tun, rxhash, tfile);
1606 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
1608 struct file *file = iocb->ki_filp;
1609 struct tun_struct *tun = tun_get(file);
1610 struct tun_file *tfile = file->private_data;
1617 if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
1620 result = tun_get_user(tun, tfile, NULL, from, noblock, false);
1626 /* Put packet to the user space buffer */
1627 static ssize_t tun_put_user(struct tun_struct *tun,
1628 struct tun_file *tfile,
1629 struct sk_buff *skb,
1630 struct iov_iter *iter)
1632 struct tun_pi pi = { 0, skb->protocol };
1633 struct tun_pcpu_stats *stats;
1635 int vlan_offset = 0;
1637 int vnet_hdr_sz = 0;
1639 if (skb_vlan_tag_present(skb))
1640 vlan_hlen = VLAN_HLEN;
1642 if (tun->flags & IFF_VNET_HDR)
1643 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1645 total = skb->len + vlan_hlen + vnet_hdr_sz;
1647 if (!(tun->flags & IFF_NO_PI)) {
1648 if (iov_iter_count(iter) < sizeof(pi))
1651 total += sizeof(pi);
1652 if (iov_iter_count(iter) < total) {
1653 /* Packet will be striped */
1654 pi.flags |= TUN_PKT_STRIP;
1657 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
1662 struct virtio_net_hdr gso;
1664 if (iov_iter_count(iter) < vnet_hdr_sz)
1667 if (virtio_net_hdr_from_skb(skb, &gso,
1668 tun_is_little_endian(tun), true,
1670 struct skb_shared_info *sinfo = skb_shinfo(skb);
1671 pr_err("unexpected GSO type: "
1672 "0x%x, gso_size %d, hdr_len %d\n",
1673 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
1674 tun16_to_cpu(tun, gso.hdr_len));
1675 print_hex_dump(KERN_ERR, "tun: ",
1678 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
1683 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
1686 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1692 __be16 h_vlan_proto;
1696 veth.h_vlan_proto = skb->vlan_proto;
1697 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
1699 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
1701 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
1702 if (ret || !iov_iter_count(iter))
1705 ret = copy_to_iter(&veth, sizeof(veth), iter);
1706 if (ret != sizeof(veth) || !iov_iter_count(iter))
1710 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
1713 /* caller is in process context, */
1714 stats = get_cpu_ptr(tun->pcpu_stats);
1715 u64_stats_update_begin(&stats->syncp);
1716 stats->tx_packets++;
1717 stats->tx_bytes += skb->len + vlan_hlen;
1718 u64_stats_update_end(&stats->syncp);
1719 put_cpu_ptr(tun->pcpu_stats);
1724 static struct sk_buff *tun_ring_recv(struct tun_file *tfile, int noblock,
1727 DECLARE_WAITQUEUE(wait, current);
1728 struct sk_buff *skb = NULL;
1731 skb = skb_array_consume(&tfile->tx_array);
1739 add_wait_queue(&tfile->wq.wait, &wait);
1742 set_current_state(TASK_INTERRUPTIBLE);
1743 skb = skb_array_consume(&tfile->tx_array);
1746 if (signal_pending(current)) {
1747 error = -ERESTARTSYS;
1750 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
1758 __set_current_state(TASK_RUNNING);
1759 remove_wait_queue(&tfile->wq.wait, &wait);
1766 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
1767 struct iov_iter *to,
1768 int noblock, struct sk_buff *skb)
1773 tun_debug(KERN_INFO, tun, "tun_do_read\n");
1775 if (!iov_iter_count(to)) {
1782 /* Read frames from ring */
1783 skb = tun_ring_recv(tfile, noblock, &err);
1788 ret = tun_put_user(tun, tfile, skb, to);
1789 if (unlikely(ret < 0))
1797 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
1799 struct file *file = iocb->ki_filp;
1800 struct tun_file *tfile = file->private_data;
1801 struct tun_struct *tun = __tun_get(tfile);
1802 ssize_t len = iov_iter_count(to), ret;
1808 if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
1811 ret = tun_do_read(tun, tfile, to, noblock, NULL);
1812 ret = min_t(ssize_t, ret, len);
1819 static void tun_free_netdev(struct net_device *dev)
1821 struct tun_struct *tun = netdev_priv(dev);
1823 BUG_ON(!(list_empty(&tun->disabled)));
1824 free_percpu(tun->pcpu_stats);
1825 tun_flow_uninit(tun);
1826 security_tun_dev_free_security(tun->security);
1829 static void tun_setup(struct net_device *dev)
1831 struct tun_struct *tun = netdev_priv(dev);
1833 tun->owner = INVALID_UID;
1834 tun->group = INVALID_GID;
1836 dev->ethtool_ops = &tun_ethtool_ops;
1837 dev->needs_free_netdev = true;
1838 dev->priv_destructor = tun_free_netdev;
1839 /* We prefer our own queue length */
1840 dev->tx_queue_len = TUN_READQ_SIZE;
1843 /* Trivial set of netlink ops to allow deleting tun or tap
1844 * device with netlink.
1846 static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
1847 struct netlink_ext_ack *extack)
1849 NL_SET_ERR_MSG(extack,
1850 "tun/tap creation via rtnetlink is not supported.");
1854 static struct rtnl_link_ops tun_link_ops __read_mostly = {
1856 .priv_size = sizeof(struct tun_struct),
1858 .validate = tun_validate,
1861 static void tun_sock_write_space(struct sock *sk)
1863 struct tun_file *tfile;
1864 wait_queue_head_t *wqueue;
1866 if (!sock_writeable(sk))
1869 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
1872 wqueue = sk_sleep(sk);
1873 if (wqueue && waitqueue_active(wqueue))
1874 wake_up_interruptible_sync_poll(wqueue, POLLOUT |
1875 POLLWRNORM | POLLWRBAND);
1877 tfile = container_of(sk, struct tun_file, sk);
1878 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
1881 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
1884 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1885 struct tun_struct *tun = __tun_get(tfile);
1890 ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
1891 m->msg_flags & MSG_DONTWAIT,
1892 m->msg_flags & MSG_MORE);
1897 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
1900 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1901 struct tun_struct *tun = __tun_get(tfile);
1902 struct sk_buff *skb = m->msg_control;
1910 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
1914 if (flags & MSG_ERRQUEUE) {
1915 ret = sock_recv_errqueue(sock->sk, m, total_len,
1916 SOL_PACKET, TUN_TX_TIMESTAMP);
1919 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, skb);
1920 if (ret > (ssize_t)total_len) {
1921 m->msg_flags |= MSG_TRUNC;
1922 ret = flags & MSG_TRUNC ? ret : total_len;
1936 static int tun_peek_len(struct socket *sock)
1938 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
1939 struct tun_struct *tun;
1942 tun = __tun_get(tfile);
1946 ret = skb_array_peek_len(&tfile->tx_array);
1952 /* Ops structure to mimic raw sockets with tun */
1953 static const struct proto_ops tun_socket_ops = {
1954 .peek_len = tun_peek_len,
1955 .sendmsg = tun_sendmsg,
1956 .recvmsg = tun_recvmsg,
1959 static struct proto tun_proto = {
1961 .owner = THIS_MODULE,
1962 .obj_size = sizeof(struct tun_file),
1965 static int tun_flags(struct tun_struct *tun)
1967 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
1970 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
1973 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1974 return sprintf(buf, "0x%x\n", tun_flags(tun));
1977 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
1980 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1981 return uid_valid(tun->owner)?
1982 sprintf(buf, "%u\n",
1983 from_kuid_munged(current_user_ns(), tun->owner)):
1984 sprintf(buf, "-1\n");
1987 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
1990 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
1991 return gid_valid(tun->group) ?
1992 sprintf(buf, "%u\n",
1993 from_kgid_munged(current_user_ns(), tun->group)):
1994 sprintf(buf, "-1\n");
1997 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
1998 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
1999 static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2001 static struct attribute *tun_dev_attrs[] = {
2002 &dev_attr_tun_flags.attr,
2003 &dev_attr_owner.attr,
2004 &dev_attr_group.attr,
2008 static const struct attribute_group tun_attr_group = {
2009 .attrs = tun_dev_attrs
2012 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
2014 struct tun_struct *tun;
2015 struct tun_file *tfile = file->private_data;
2016 struct net_device *dev;
2019 if (tfile->detached)
2022 dev = __dev_get_by_name(net, ifr->ifr_name);
2024 if (ifr->ifr_flags & IFF_TUN_EXCL)
2026 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2027 tun = netdev_priv(dev);
2028 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2029 tun = netdev_priv(dev);
2033 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
2034 !!(tun->flags & IFF_MULTI_QUEUE))
2037 if (tun_not_capable(tun))
2039 err = security_tun_dev_open(tun->security);
2043 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER, true);
2047 if (tun->flags & IFF_MULTI_QUEUE &&
2048 (tun->numqueues + tun->numdisabled > 1)) {
2049 /* One or more queue has already been attached, no need
2050 * to initialize the device again.
2057 unsigned long flags = 0;
2058 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2061 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
2063 err = security_tun_dev_create();
2068 if (ifr->ifr_flags & IFF_TUN) {
2072 } else if (ifr->ifr_flags & IFF_TAP) {
2080 name = ifr->ifr_name;
2082 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
2083 NET_NAME_UNKNOWN, tun_setup, queues,
2088 err = dev_get_valid_name(net, dev, name);
2092 dev_net_set(dev, net);
2093 dev->rtnl_link_ops = &tun_link_ops;
2094 dev->ifindex = tfile->ifindex;
2095 dev->sysfs_groups[0] = &tun_attr_group;
2097 tun = netdev_priv(dev);
2100 tun->txflt.count = 0;
2101 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
2103 tun->align = NET_SKB_PAD;
2104 tun->filter_attached = false;
2105 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
2106 tun->rx_batched = 0;
2108 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2109 if (!tun->pcpu_stats) {
2114 spin_lock_init(&tun->lock);
2116 err = security_tun_dev_alloc_security(&tun->security);
2123 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
2124 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2125 NETIF_F_HW_VLAN_STAG_TX;
2126 dev->features = dev->hw_features | NETIF_F_LLTX;
2127 dev->vlan_features = dev->features &
2128 ~(NETIF_F_HW_VLAN_CTAG_TX |
2129 NETIF_F_HW_VLAN_STAG_TX);
2131 INIT_LIST_HEAD(&tun->disabled);
2132 err = tun_attach(tun, file, false, false);
2136 err = register_netdevice(tun->dev);
2139 /* free_netdev() won't check refcnt, to aovid race
2140 * with dev_put() we need publish tun after registration.
2142 rcu_assign_pointer(tfile->tun, tun);
2145 netif_carrier_on(tun->dev);
2147 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
2149 tun->flags = (tun->flags & ~TUN_FEATURES) |
2150 (ifr->ifr_flags & TUN_FEATURES);
2152 /* Make sure persistent devices do not get stuck in
2155 if (netif_running(tun->dev))
2156 netif_tx_wake_all_queues(tun->dev);
2158 strcpy(ifr->ifr_name, tun->dev->name);
2162 tun_detach_all(dev);
2163 /* register_netdevice() already called tun_free_netdev() */
2167 tun_flow_uninit(tun);
2168 security_tun_dev_free_security(tun->security);
2170 free_percpu(tun->pcpu_stats);
2176 static void tun_get_iff(struct net *net, struct tun_struct *tun,
2179 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
2181 strcpy(ifr->ifr_name, tun->dev->name);
2183 ifr->ifr_flags = tun_flags(tun);
2187 /* This is like a cut-down ethtool ops, except done via tun fd so no
2188 * privs required. */
2189 static int set_offload(struct tun_struct *tun, unsigned long arg)
2191 netdev_features_t features = 0;
2193 if (arg & TUN_F_CSUM) {
2194 features |= NETIF_F_HW_CSUM;
2197 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2198 if (arg & TUN_F_TSO_ECN) {
2199 features |= NETIF_F_TSO_ECN;
2200 arg &= ~TUN_F_TSO_ECN;
2202 if (arg & TUN_F_TSO4)
2203 features |= NETIF_F_TSO;
2204 if (arg & TUN_F_TSO6)
2205 features |= NETIF_F_TSO6;
2206 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2212 /* This gives the user a way to test for new features in future by
2213 * trying to set them. */
2217 tun->set_features = features;
2218 tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2219 tun->dev->wanted_features |= features;
2220 netdev_update_features(tun->dev);
2225 static void tun_detach_filter(struct tun_struct *tun, int n)
2228 struct tun_file *tfile;
2230 for (i = 0; i < n; i++) {
2231 tfile = rtnl_dereference(tun->tfiles[i]);
2232 lock_sock(tfile->socket.sk);
2233 sk_detach_filter(tfile->socket.sk);
2234 release_sock(tfile->socket.sk);
2237 tun->filter_attached = false;
2240 static int tun_attach_filter(struct tun_struct *tun)
2243 struct tun_file *tfile;
2245 for (i = 0; i < tun->numqueues; i++) {
2246 tfile = rtnl_dereference(tun->tfiles[i]);
2247 lock_sock(tfile->socket.sk);
2248 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2249 release_sock(tfile->socket.sk);
2251 tun_detach_filter(tun, i);
2256 tun->filter_attached = true;
2260 static void tun_set_sndbuf(struct tun_struct *tun)
2262 struct tun_file *tfile;
2265 for (i = 0; i < tun->numqueues; i++) {
2266 tfile = rtnl_dereference(tun->tfiles[i]);
2267 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2271 static int tun_set_queue(struct file *file, struct ifreq *ifr)
2273 struct tun_file *tfile = file->private_data;
2274 struct tun_struct *tun;
2279 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
2280 tun = tfile->detached;
2285 ret = security_tun_dev_attach_queue(tun->security);
2288 ret = tun_attach(tun, file, false, true);
2289 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
2290 tun = rtnl_dereference(tfile->tun);
2291 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
2294 __tun_detach(tfile, false);
2303 /* Return correct value for tun->dev->addr_len based on tun->dev->type. */
2304 static unsigned char tun_get_addr_len(unsigned short type)
2308 case ARPHRD_TUNNEL6:
2309 return sizeof(struct in6_addr);
2316 case ARPHRD_IEEE802154:
2317 case ARPHRD_IEEE802154_MONITOR:
2318 return IEEE802154_EXTENDED_ADDR_LEN;
2319 case ARPHRD_PHONET_PIPE:
2323 case ARPHRD_6LOWPAN:
2324 return EUI64_ADDR_LEN;
2329 case ARPHRD_IEEE802:
2332 return ROSE_ADDR_LEN;
2334 return AX25_ADDR_LEN;
2335 case ARPHRD_LOCALTLK:
2342 static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2343 unsigned long arg, int ifreq_len)
2345 struct tun_file *tfile = file->private_data;
2346 struct tun_struct *tun;
2347 void __user* argp = (void __user*)arg;
2353 unsigned int ifindex;
2357 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == SOCK_IOC_TYPE) {
2358 if (copy_from_user(&ifr, argp, ifreq_len))
2361 memset(&ifr, 0, sizeof(ifr));
2363 if (cmd == TUNGETFEATURES) {
2364 /* Currently this just means: "what IFF flags are valid?".
2365 * This is needed because we never checked for invalid flags on
2368 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
2369 (unsigned int __user*)argp);
2370 } else if (cmd == TUNSETQUEUE)
2371 return tun_set_queue(file, &ifr);
2376 tun = __tun_get(tfile);
2377 if (cmd == TUNSETIFF) {
2382 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2384 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
2389 if (copy_to_user(argp, &ifr, ifreq_len))
2393 if (cmd == TUNSETIFINDEX) {
2399 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2403 tfile->ifindex = ifindex;
2411 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
2416 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2418 if (tfile->detached)
2419 ifr.ifr_flags |= IFF_DETACH_QUEUE;
2420 if (!tfile->socket.sk->sk_filter)
2421 ifr.ifr_flags |= IFF_NOFILTER;
2423 if (copy_to_user(argp, &ifr, ifreq_len))
2428 /* Disable/Enable checksum */
2430 /* [unimplemented] */
2431 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
2432 arg ? "disabled" : "enabled");
2436 /* Disable/Enable persist mode. Keep an extra reference to the
2437 * module to prevent the module being unprobed.
2439 if (arg && !(tun->flags & IFF_PERSIST)) {
2440 tun->flags |= IFF_PERSIST;
2441 __module_get(THIS_MODULE);
2443 if (!arg && (tun->flags & IFF_PERSIST)) {
2444 tun->flags &= ~IFF_PERSIST;
2445 module_put(THIS_MODULE);
2448 tun_debug(KERN_INFO, tun, "persist %s\n",
2449 arg ? "enabled" : "disabled");
2453 /* Set owner of the device */
2454 owner = make_kuid(current_user_ns(), arg);
2455 if (!uid_valid(owner)) {
2460 tun_debug(KERN_INFO, tun, "owner set to %u\n",
2461 from_kuid(&init_user_ns, tun->owner));
2465 /* Set group of the device */
2466 group = make_kgid(current_user_ns(), arg);
2467 if (!gid_valid(group)) {
2472 tun_debug(KERN_INFO, tun, "group set to %u\n",
2473 from_kgid(&init_user_ns, tun->group));
2477 /* Only allow setting the type when the interface is down */
2478 if (tun->dev->flags & IFF_UP) {
2479 tun_debug(KERN_INFO, tun,
2480 "Linktype set failed because interface is up\n");
2483 tun->dev->type = (int) arg;
2484 tun->dev->addr_len = tun_get_addr_len(tun->dev->type);
2485 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2497 ret = set_offload(tun, arg);
2500 case TUNSETTXFILTER:
2501 /* Can be set only for TAPs */
2503 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2505 ret = update_filter(&tun->txflt, (void __user *)arg);
2509 /* Get hw address */
2510 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2511 ifr.ifr_hwaddr.sa_family = tun->dev->type;
2512 if (copy_to_user(argp, &ifr, ifreq_len))
2517 /* Set hw address */
2518 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2519 ifr.ifr_hwaddr.sa_data);
2521 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
2525 sndbuf = tfile->socket.sk->sk_sndbuf;
2526 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2531 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2540 tun->sndbuf = sndbuf;
2541 tun_set_sndbuf(tun);
2544 case TUNGETVNETHDRSZ:
2545 vnet_hdr_sz = tun->vnet_hdr_sz;
2546 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2550 case TUNSETVNETHDRSZ:
2551 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2555 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2560 tun->vnet_hdr_sz = vnet_hdr_sz;
2564 le = !!(tun->flags & TUN_VNET_LE);
2565 if (put_user(le, (int __user *)argp))
2570 if (get_user(le, (int __user *)argp)) {
2575 tun->flags |= TUN_VNET_LE;
2577 tun->flags &= ~TUN_VNET_LE;
2581 ret = tun_get_vnet_be(tun, argp);
2585 ret = tun_set_vnet_be(tun, argp);
2588 case TUNATTACHFILTER:
2589 /* Can be set only for TAPs */
2591 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2594 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
2597 ret = tun_attach_filter(tun);
2600 case TUNDETACHFILTER:
2601 /* Can be set only for TAPs */
2603 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2606 tun_detach_filter(tun, tun->numqueues);
2611 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
2614 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
2631 static long tun_chr_ioctl(struct file *file,
2632 unsigned int cmd, unsigned long arg)
2634 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
2637 #ifdef CONFIG_COMPAT
2638 static long tun_chr_compat_ioctl(struct file *file,
2639 unsigned int cmd, unsigned long arg)
2644 case TUNSETTXFILTER:
2649 arg = (unsigned long)compat_ptr(arg);
2652 arg = (compat_ulong_t)arg;
2657 * compat_ifreq is shorter than ifreq, so we must not access beyond
2658 * the end of that structure. All fields that are used in this
2659 * driver are compatible though, we don't need to convert the
2662 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
2664 #endif /* CONFIG_COMPAT */
2666 static int tun_chr_fasync(int fd, struct file *file, int on)
2668 struct tun_file *tfile = file->private_data;
2671 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
2675 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
2676 tfile->flags |= TUN_FASYNC;
2678 tfile->flags &= ~TUN_FASYNC;
2684 static int tun_chr_open(struct inode *inode, struct file * file)
2686 struct net *net = current->nsproxy->net_ns;
2687 struct tun_file *tfile;
2689 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
2691 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
2695 if (skb_array_init(&tfile->tx_array, 0, GFP_KERNEL)) {
2696 sk_free(&tfile->sk);
2700 RCU_INIT_POINTER(tfile->tun, NULL);
2704 init_waitqueue_head(&tfile->wq.wait);
2705 RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
2707 tfile->socket.file = file;
2708 tfile->socket.ops = &tun_socket_ops;
2710 sock_init_data(&tfile->socket, &tfile->sk);
2712 tfile->sk.sk_write_space = tun_sock_write_space;
2713 tfile->sk.sk_sndbuf = INT_MAX;
2715 file->private_data = tfile;
2716 INIT_LIST_HEAD(&tfile->next);
2718 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
2723 static int tun_chr_close(struct inode *inode, struct file *file)
2725 struct tun_file *tfile = file->private_data;
2727 tun_detach(tfile, true);
2732 #ifdef CONFIG_PROC_FS
2733 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f)
2735 struct tun_struct *tun;
2738 memset(&ifr, 0, sizeof(ifr));
2743 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
2749 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
2753 static const struct file_operations tun_fops = {
2754 .owner = THIS_MODULE,
2755 .llseek = no_llseek,
2756 .read_iter = tun_chr_read_iter,
2757 .write_iter = tun_chr_write_iter,
2758 .poll = tun_chr_poll,
2759 .unlocked_ioctl = tun_chr_ioctl,
2760 #ifdef CONFIG_COMPAT
2761 .compat_ioctl = tun_chr_compat_ioctl,
2763 .open = tun_chr_open,
2764 .release = tun_chr_close,
2765 .fasync = tun_chr_fasync,
2766 #ifdef CONFIG_PROC_FS
2767 .show_fdinfo = tun_chr_show_fdinfo,
2771 static struct miscdevice tun_miscdev = {
2774 .nodename = "net/tun",
2778 /* ethtool interface */
2780 static int tun_get_link_ksettings(struct net_device *dev,
2781 struct ethtool_link_ksettings *cmd)
2783 ethtool_link_ksettings_zero_link_mode(cmd, supported);
2784 ethtool_link_ksettings_zero_link_mode(cmd, advertising);
2785 cmd->base.speed = SPEED_10;
2786 cmd->base.duplex = DUPLEX_FULL;
2787 cmd->base.port = PORT_TP;
2788 cmd->base.phy_address = 0;
2789 cmd->base.autoneg = AUTONEG_DISABLE;
2793 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2795 struct tun_struct *tun = netdev_priv(dev);
2797 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
2798 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
2800 switch (tun->flags & TUN_TYPE_MASK) {
2802 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
2805 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
2810 static u32 tun_get_msglevel(struct net_device *dev)
2813 struct tun_struct *tun = netdev_priv(dev);
2820 static void tun_set_msglevel(struct net_device *dev, u32 value)
2823 struct tun_struct *tun = netdev_priv(dev);
2828 static int tun_get_coalesce(struct net_device *dev,
2829 struct ethtool_coalesce *ec)
2831 struct tun_struct *tun = netdev_priv(dev);
2833 ec->rx_max_coalesced_frames = tun->rx_batched;
2838 static int tun_set_coalesce(struct net_device *dev,
2839 struct ethtool_coalesce *ec)
2841 struct tun_struct *tun = netdev_priv(dev);
2843 if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
2844 tun->rx_batched = NAPI_POLL_WEIGHT;
2846 tun->rx_batched = ec->rx_max_coalesced_frames;
2851 static const struct ethtool_ops tun_ethtool_ops = {
2852 .get_drvinfo = tun_get_drvinfo,
2853 .get_msglevel = tun_get_msglevel,
2854 .set_msglevel = tun_set_msglevel,
2855 .get_link = ethtool_op_get_link,
2856 .get_ts_info = ethtool_op_get_ts_info,
2857 .get_coalesce = tun_get_coalesce,
2858 .set_coalesce = tun_set_coalesce,
2859 .get_link_ksettings = tun_get_link_ksettings,
2862 static int tun_queue_resize(struct tun_struct *tun)
2864 struct net_device *dev = tun->dev;
2865 struct tun_file *tfile;
2866 struct skb_array **arrays;
2867 int n = tun->numqueues + tun->numdisabled;
2870 arrays = kmalloc_array(n, sizeof(*arrays), GFP_KERNEL);
2874 for (i = 0; i < tun->numqueues; i++) {
2875 tfile = rtnl_dereference(tun->tfiles[i]);
2876 arrays[i] = &tfile->tx_array;
2878 list_for_each_entry(tfile, &tun->disabled, next)
2879 arrays[i++] = &tfile->tx_array;
2881 ret = skb_array_resize_multiple(arrays, n,
2882 dev->tx_queue_len, GFP_KERNEL);
2888 static int tun_device_event(struct notifier_block *unused,
2889 unsigned long event, void *ptr)
2891 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2892 struct tun_struct *tun = netdev_priv(dev);
2895 if (dev->rtnl_link_ops != &tun_link_ops)
2899 case NETDEV_CHANGE_TX_QUEUE_LEN:
2900 if (tun_queue_resize(tun))
2904 for (i = 0; i < tun->numqueues; i++) {
2905 struct tun_file *tfile;
2907 tfile = rtnl_dereference(tun->tfiles[i]);
2908 tfile->socket.sk->sk_write_space(tfile->socket.sk);
2918 static struct notifier_block tun_notifier_block __read_mostly = {
2919 .notifier_call = tun_device_event,
2922 static int __init tun_init(void)
2926 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
2928 ret = rtnl_link_register(&tun_link_ops);
2930 pr_err("Can't register link_ops\n");
2934 ret = misc_register(&tun_miscdev);
2936 pr_err("Can't register misc device %d\n", TUN_MINOR);
2940 ret = register_netdevice_notifier(&tun_notifier_block);
2942 pr_err("Can't register netdevice notifier\n");
2949 misc_deregister(&tun_miscdev);
2951 rtnl_link_unregister(&tun_link_ops);
2956 static void tun_cleanup(void)
2958 misc_deregister(&tun_miscdev);
2959 rtnl_link_unregister(&tun_link_ops);
2960 unregister_netdevice_notifier(&tun_notifier_block);
2963 /* Get an underlying socket object from tun file. Returns error unless file is
2964 * attached to a device. The returned object works like a packet socket, it
2965 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
2966 * holding a reference to the file for as long as the socket is in use. */
2967 struct socket *tun_get_socket(struct file *file)
2969 struct tun_file *tfile;
2970 if (file->f_op != &tun_fops)
2971 return ERR_PTR(-EINVAL);
2972 tfile = file->private_data;
2974 return ERR_PTR(-EBADFD);
2975 return &tfile->socket;
2977 EXPORT_SYMBOL_GPL(tun_get_socket);
2979 struct skb_array *tun_get_skb_array(struct file *file)
2981 struct tun_file *tfile;
2983 if (file->f_op != &tun_fops)
2984 return ERR_PTR(-EINVAL);
2985 tfile = file->private_data;
2987 return ERR_PTR(-EBADFD);
2988 return &tfile->tx_array;
2990 EXPORT_SYMBOL_GPL(tun_get_skb_array);
2992 module_init(tun_init);
2993 module_exit(tun_cleanup);
2994 MODULE_DESCRIPTION(DRV_DESCRIPTION);
2995 MODULE_AUTHOR(DRV_COPYRIGHT);
2996 MODULE_LICENSE("GPL");
2997 MODULE_ALIAS_MISCDEV(TUN_MINOR);
2998 MODULE_ALIAS("devname:net/tun");